Manual of

forest inventory

With special reference

to mixed tropical forests
 Reprinted 1992

The designations employed and the presentation of material in this

publication do not imply the expression of any opinion whatsoever on

the part of the Food and Agriculture Organization of the United
Nations concerning the legal status of any country, territory, city or
area or of its authorities, or concerning the delimitation of its frontiers
or boundaries

M-35
ISBN 92-5-1 01 132-X

All rightsreserved No part of this publication may be reproduced, stored in a

retrievalsystem, or transmitted in any form or by any means, electronic, mechani-
cal, photocopying or otherwise, without the prior permission of the copyright owner

Applications for such permission, with a statement of the purpose and extent of the
reproduction, should be addressed to the Director, Publications Division, Food and
Agriculture Organization of the United Nations. Viale delle Terme di Caracalla,
00100 Rome. Italy

FAO 1981
 - 1 11 -

FOREWORD

The Forestry Department (formerly Forestry and Forest Products Division) of the Food
and Agriculture Organization of the United Nation* has been involved since this
Organisation s inception in the definition and implementation of forest resource evaluation
1

programmes, at all levels, from world and regional forest appraisals to local management
inventories. It hae performed a number of forest resource surveys in many countries of
the world, has carried out a series ol world and regional studies - such ae the World
Forest Inventory and the regional timber trends and prospects studies - and hac produced
a few publications on the methodology eide such as "Planning a Forest Inventory" by
Dr. B. Husch, and the "Manual for forest inventory operations executed by FAO M (1^6/j.

To take the experience gained by FAO in the last few years into account and to iulfil
in thia field FAO'n role concerning dissemination of knowledge, Mr. J . P. Lanly, Foreet
Resource Surveys Officer, was asked to write a new manual of forest inventory. This manual
is intended to be of use mainly to professionals dealing with the evaluation and management
of mixed tropical f create, since it is restricted to inventory methods and practices which
have been found feasible in these areas.

At the beginning of 197? about thirty specialists all over the world were asked to
give their comments on a draft of the outline and of the main contents. Most of their
ftuggeetiona have been taken into consideration. They must all be thanked here, with a
particular mention of Dr. P.G. de Vries from Wageningen University in the Netherlands who
made the mopt substantial proposals. The sections of Chapter V devoted to measurement
and volume estimation techniques formed the basis of lectures delivered by Mr. J.P. Lanly
at the training course on forest inventory organized in August and September 1973 by the
Royal College of Forestry in Stockholm in cooperation with FAO, and include information
on recovery studies and accessibility problems, two topics which need to be given mor*
consideration in future inventory work. The section on accessibility problems wae reviewed
by Prof. U. Sundberg, Chief of the Forest Logging and Transport Branch of the Forest
Resources Division, and Chapter VI on data recording and processing wao drafted by
Dr. H. K. Marech of the Forest Management Branch. Thanks are due also to Messrs. R. Bo ton,
I

J.W. Eastwood, J. Jackson and I). A. Harcharik for their contribution to the editing of the
English version and to Mrs. R.S. Borelli for her secretarial help.

R.G. Fontaine
Director
Forest Resources Division
 TABLE OF CONTENTS

CHAPTER I - INTRODUCTION 2

1 Historical background 2

? Main feature^ of thiB new edition 3

CHAPTE/r II - PURPOSE AND PLANNING OF A FOREST INVENTORY 5

1
Purpose of the inventory 5

11 Introduction 5

12 Definition of the objectives 6

2 Outline for preparing inventory plane 10

CHAPTER III - BASIC SAMPLING TECHNIQUES 13

1 Introduction 13
11 Sampling in forest inventory 13
1? Outline of the chapter 14

Statistical concepts 14
21 Population 14

22 Distribution 16

221 Different kinds of values oi parameters in one unit of a

population 16

222 Distribution of the values of a parameter over a whole

population 16

?23 Characteristics of central value and dispersion of the

distribution 18

224 Value of a parameter per area unit in one unit of the

population 1 V

23 Sampling 20

231 Size of sample 20

232 Precision and sampling error 21

233 Other concepts 22

24 Bias and measurement errors 23

241 Bias 23
242 Measurement errors 24

3 Basic mathematical and statistical techniques 25

31 Principles of sampling error estimation 25
311 Introduction 25
s*
312 Estimation of the sampling error on Mj from its variance 28

32 Variance of compound values 30

321 Introduction 30

322 Variances of some functions 31

 33 Ratio estimates 3S

34 Optimization in design 36
341 Optimization of a sampling design 3o
342 Optimization of an inventory design 40
4 Classical sampling designs 40
41 Classification of aaropling designs 40
411 Characteristics of sampling designs 40
412 Clusters and record units 43
4? Classical sampling designs used in forestry 43
421 Introduction 43
4^2 Random sampling designb 44

423 Systematic sampling designs S6

CHAPTffl IV - RO40TE SENSING AM MAPPING FOR ARKA ESTIMATION IN FOREST INVENTORY b?

1 Introduction 62

Forest and lana use classi i i cations 63

21 Various kinds oi classifications M

22 Classifications based on vegetation/enviro/iment relationships 64

*M "hlxiBtiii^ land use" classification used by FAO inventory operation*} 6 r)

1

3 Interpretation of conventional aerial photographs in forest inventory {0

31 Introduction 70
Area estimation with or without forest mapping 7
311

"Compulsory classifications and classifications developed

11
312
within the inventory /O

32 Some information on aerial photographs and aerial coverages ( 1

321 Characteristics of aerial photographs f 1

322 CharacteristicB of aerial coveragea 74

323 Some problems related to aerial surveying 1b

324 Mosaics U
33 Photointerpretation (
f
l

v
331 Qualities of good photointerpretation /
-

332 Stereoscopic interpretation Jb

333 Assessment of photointerpretation keys 1}

334 Photointerpretation of plots and photointerpretation

with delineation
1

I )
U
4 Forest mapping from conventional aerial photographs 1

'

41 Introduction 1

42 Transfer from single photographs i1

43 Transfer from stereoscopic pairs d1

J

Area estimation from aerial photographs and maps B'

5

51 Introductory remarks &-'

 - VI -

Direct measurements by planimetenng on maps 83

$?.

Estimation methods based on sampling techniques 83

53
Area estimation from maps 83
531
Area estimation from photographs 84
532
ContinuouB area estimation 84
54
6 Recent developments in remote sensing ajid mapping techniques 85

61 Brief presentation of recent techniques 85

611 New forms of remote sensing 85
612 New media for information storage and reproduction 88

613 New procedures for information analysis 88

614 Or tho photography 89

62 Current operational applications for forest inventory 89
621 Use of radiation outside the visible spectrum 89
62^ Use of space platforms 90
CHAPTER V - MEASUREMENT CONSIDERATIONS 9?

1 Introduction 93
2 Tree measurements 94
21 Definition of terms 94
22 Enumeration 97
221 Enumeration in sampling with units of a given area 9f
222 Enumeration in point or line sampling 97
23 Species identification 98
24 Measurements 99
c'41 Measurement units 100
242 Measurement classes 100
243 Measurement procedures and instruments 102
3 Volume estimation 105
31 Definition of volumes 105
3*- Volume units 106
33 Classification of volume estimation techniques 107
34 Volume estimation on a tree basis 108
341 Geometric formulas applied to standing or felled trees 108
342 Volume equations 109
343 Volume estimation by taper functions 116
344 Selection of the most suitable volume estimation technique 117
4 Quality assessment 117
41 Preliminary remarks on quality assessment 117
411 Definition of quality assessment in a forest inventory 117
412 Assessment of "net volumes" and usefulness of this concept 118
(with special reference to forest inventory of mixed
tropical hardwoods)
413 Other applications of quality assessment in forest inventory 118
 - Vll -

42 Methods of quality assessment

421 Assessment of external characteristics and defects
422 Assessment of internal defects 125
5 Recovery studies 125
51 Principle 125
52 Related problems 126
53 General procedure 126

531 Main steps of a recovery study 126

532 Implementation 12 /

533 Example 12 f

6 Accessibility studies 129

61 Introduction 12}
62 Selection ot accessibility parameters 130
63 Quantification and/or claseifi cation of accessibility parameters 132

631 Parameters relevant to felling 132

632 Parameters related to transport and road construction 133

CHAPTER VI - DATA RECORDING AND PROCESS I NO IN FOREST INVENTORY 136

1 Introduction 137

2 Data recording 137

21 General requirements 137

22 Specific requirements 133

221 With relation to the type of data 138

222 With relation to data processing 140

23 Main kinds of data recording 140

24 Some practical aspects of data recording 141

241 Organisation in the field 141

242 Preparation for further processing 141

3 Data processing 1 42

31 Steps of data processing

311 Data capture 1 42

312 Editing of data

1 43

313 Data generation 144

314 Presentation of the inventory results 1

45

315 System design

1 55

32 Selection of type of data preceding 1 57

321 Manual data processing 1 5/

322 Electronic data processing (EDP) 158

16
323 Combined types of data processing
 - "Ill -

33 Some practical aspects of EDP 160

331 Project-integrated data processing 160

332 Sub-contracted data processing 160

333 borne views of the use of standard programmes 163

CHAPM* VII - COHbl DERATIONS ON IIWENTORY DESIGNS 164

1 Introduction 165
? Combinations of photointerpretation and field sampling procedures 166

21 Preliminary remarks 166

/ ? /Vr^aa of the strata exactly or almost exactly known 166

?3 Areas of the strata estimated through sampling 167

T31 Area estimates from one sample only 16 /

J32 Area estimates with correction in the field 171

f'4 Other uses of double sampling designs 1 /3

3 Considerations on field sampling designs 176

31 iJistributiori of the sample 1/6
311 Unrestricted versus stratified sampling 176
31'' Random versus systematic sampling 17 1

313 One stage versus multi-stage sampling 17 f

314 tiqual or unequal probability in sampling 178

315 Use of an auxiliary parameter 178

3*
J

Characteristics of the sampling units 178

321 Plot sampling versus polyareal sampling 178
322 Size of the sampling units 179
3^3 Shape of the sampling units 180
4 Continuous forest inventory 182
41 Definition and utilization 182
42 Description of design 182
4?1 Different types of continuous forest inventory 182
422 Sampling with partial replacement (SPR) 183
5 Sequential sampling 186

ANNKX I - EXAMPLE OF TECHNICAL SPEC I FT CATIONS for inclusion in a contract of 188

aerial surveying

II - SELECTED ANNOTATED BIBLIOGRAPHY

194
CRPT2R I
 CHAPTER I

INTHODUCTICW

1 Historical background

During the period 11-22 September 1967, a meeting of forest inventory experts
attached to OVDP/SP projects was held at FAG Headquarters in Rome 0). The purpose of
tola nesting was the improvement of FAO's inventory operations with the following four
main objectives in view:

1 ) In the light of experience, to clarify the basic forest resource information

needed for potential forest industry investors and to use this knowledge in
guiding present and future projects.

2) To achieve greater uniformity and standardization in the form of inventory

results being obtained, while still leaving latitude for specialised information
that night be required to meet local conditions . This would facilitate the
comparing and combining of the resource data obtained on different inventories.

3) To improve the efficiency of inventory operations so that better and more

reliable forest resource information can be obtained at lower cost.

4) To intensify the cooperation and collaboration between forest inventory

specialists and the users of inventory results so that more pertinent inventory
information can be obtained*

As a result of this meeting, a recommendation was made that the present manual be
prepared to assist inventory experts in the planning and execution of project inventory
work*

The first edition of the Manual was prepared by B. Rusch, the former Chief of the
Forest Resources Survey Section of the FAO Forestry Department, and incorporated the
recommendations of the meeting*

In that first edition it was stated that "a manual of this type should be a working
document subject to periodic modification or revision to incorporate improvements which
will become evident in the course of its use". A revision of the manual was undertaken
during 1972. A questionnaire was sent to a number of inventory specialists, together
with a general outline and the main contents of the proposed revision, asking for review
and suggestions. Most of them replied and gave us many valuable suggestions regarding
information to be included in the new edition. The main emphasis was on the definition of
the purposes of the inventory, the accuracy of the measurements and data processing
procedures. We wish to thank them here for their kind ool labor at ion, without which this
edition would have omitted certain important items and its preparation would have been
much more difficult*

(1) A report of this meeting is given in document FO:SF/7, D< 1? dated 9 October
and entitled "Report of the Headquarters Meeting of Forest Inventory Experts on
UFDP/SF Projects-.
?. Main features of this new edition

a) This document is a manual. Its purpose is not to be a comprehensive textbook

on forest inventory but to list and describe briefly the main tools which are used in this
important field of forest activity and to give advice on their use. It does not seek to
provide detailed practical instructions for use by technicians and workers, as
each inventory has its own requirements in this respect and needs its own special
instructions.

This manual obviously cannot suffice for all purposes and a number of questions
remain unanswered. *rhe information given should be supplemented by information taken
from other sources such as statistical and forestry textbooks and periodicals, inventory
reports ae well as from individual research and reasoning. Each inventory operation, with
2 to own purposes and requirements, must have its own specifications. One cannot imagine
a single book which provides an answer for all cases.

b) The manual, in the first edition, was entitled "Manual for forest inventory
operations executed by PAO". It cannot be claimed, however, that inventories executed by
F'AO are any different from those executed by public agencies or private firms, and this
manual takes into account some forest inventory techniques used by specialists outside FAO.
Moreover, one has to consider the role that PAO plays in training and dissemination of
information; this manual may be of use not only within the framework of FAO inventory
operations but also in many developing countries where there is a lack of relerrant
expertise, ?'or these reasons, reference to PAO inventory operations has been suppressed
in the new title.

c) Inventories carried out by FAO or developing countries are most oiten in

tropical regions. Although this manual deals with forest inventory in general, most
consideration IR given to inventories of broadleaved tropical forests. These are of
special interest for most of the developing countries with forest areas. As they are cf
less use for tropical foresta some interesting techniques us*d in temperate tones, such
as photogramnetric measurements, photo volume tables and point sampling methods, are only
briefly described. However, these techniques may assume increasing importance for
plantations and homogeneous natural stands in many developing countries.

d) Almost all forest inventories in developing countries are carried out for the
evaluation of wood resources with emphasis on estimation of gross volumes, quality
assessment and utilisation studies. However one must not forget that there may be other
data to collect and other j^ara^etero to estimate according to the purpoeos of the inventory.
 CHiPTER II

HJRP06K AMD PUHNCTQ OP A FOREST IKVIHTORY

 CHAPTER II

PURPOSE AND PLANNING OF A FOREST INVENTORY

The main components of a forest inventory and the programming depend upon the aims
of the operation. Purpose and planning are closely related; purpose must be clearly
defined and planning designed to achieve that purpose. For this reason they are put
together in this chapter* Further comments regarding these matters will be found in
"Planning a Forest Inventory", by B. Huech (FAO Forestry and Forest Products Studies
No. 1 I).

1 Purpose of the inventory

1 1 Introduction

It is very important to define clearly the various objectives of th* proposed

inventory. The relative importance of each must be considered, in order to design and
implement an operation which best solves the problem. Account must also be taken of the
unavoidable constraints and limitations such as available time and funds and ability of
the staff.

A usual criticism made to the people responsible for designing and executing
inventory operations is that they undertake such work without a clear idea of the
objectives to be met and thus provide forestry officers, economists, luggers and
industrialists with inadequate or even useless information*

Sometimes a thorough study of the problem may indicate that inventory will not
provide the correct answer. A cost benefit analysis may a} BO conclude that a forest
inventory is not the most efficient tool for providing the information required due to
existing constraints and limitations. Compilation of information already available,
comparison with other similar stands already inventoried and use of research results, may
meet the required degree of precision at less expenoc.

Tfcere may be f at the same time and in the same country, a need for different kinds
of inventory, for instance inventory at a country-wide level ( "national forest inventory" ),
inventories of big units of forest area (for instance 100,000 ha of forest) or inventories
of stands for the preparation of working plans. But, as an example, it cannot be
expected that information obtained from a national forest inventory will be adequate to
form the basis of a detailed local management plan. This has to be pointed out to the
decision-makers who sometimes believe that a single type of forest inventory will provide
them with all the information they need at different levels- Generally, for lack of
resources, priority has to be given to that type of inventory which wilJ solve the more
urgent probl

Sometimes a careful study may demonstrate that the most useful operation to be
carried out ia a combination of partial inventories at the various levelu* Recently a
request to UKDP/SF for a national forest inventory wa converted into a combination of the
following operations:

reconnaissance by photointerpretaticn and some field plote of the forested area

of the country, for an estimation of the area* Covered by each vegetation and
forest type;
 - 6 -

-
vegetation mapping of a selection of forest reserves, with complementary fieJd
plots for rough estimation of the growing stock of each forest type;

- intensive inventory, with vegetation mapping, of th* most valuable forest area*

This example shows that a forest inventory programme may include different types of
inventory in order to meet different objectives *

12 Definition of the objectives

121 The objectives must be defined jointly by the people who will make use of
the inventory results (e.g. decision-makers, forest managers) and by the
inventory specialist, not by the latter alone. The inventory specialist should design
an inventory which will provide the users with the information they need in a suitable
form and with the required precision. This cooperation with the potential users is
necessary from the time that the inventory is prepared until the delivery of the final
results.

Regarding thie cooperation two difficulties may be encountered:

a) In certain cases the inventory specialist has to prepare the inventory at a

time when the users are either not present or do not have a clear and definitive
idea of the information needed. For instance, in a forestry development
project, management, logging and economics advisers may arrive only after the
inventory has started or has even been completed, because they cannot function
effectively until the information provided by the inventory IB available.
Usually they do not participate in the preparation of the inventory, and this may
explain why the information given by the inventory iB inadequate in some oases.
Poeaibili ties of avoiding this drawback are to seek advice from the greatest
poBBible number of eventual users, to compare with other similar inventories
already completed and if necessary to request conaultancies from forest
management, logging, or forest economics specialists at the time when the
inventory IB being designed.

b) Another difficulty comes from the evolution of the purpose of the inventory*
The auns defined during the preparation of the inventory may change during the
course ol the operation. This occurs, for instance, when the unit sige of the
blocks, for which the results are to be provided with a given precision, changee.
This IB also true for long programmes of forest inventory at a national level.
There is no general courue of action to overcome this difficulty; the only
observation to be made is that the more flexible the initial design, the easier
the trans f unaation thereof. Moreover, every effort should be made to foresee
some of thse eventual modifications when designing the inventory.

122 Priority of objectives. Not all the objectives have the same importance.
Same ar** very fundamental and can in themselves justify the whole inventory.
The corresponding information has to be given in the required form or the operation will
fail. The degree of precision of the information provided 10 also a most important
requirement. On the other hand, it may be acceptable to fulfill a secondary objective
only approximately ( f or instance, by accepting a lower precision in the corresponding
information).

The priority of the objectives to bo met has to be clearly assessed before designing
an inventory. For instance, if the estimation of the area of a foreet IB more important
than the estimation of its volume, the inventory design will strengthen the work of
interpretation of remote sensing imagery and mapping and give less importance to tree
aaurements on imagery or in the field. Likewise, priority can be assessed among the
 - 7 -

zones or the blocks of the region inventoried. As far as volume estimation is concerned,
species do not have the same economic value, BO the inventory will be designed in order
to provide the results with a specific precision for the most important species:
volumes of individual secondary species may be estimated less precisely, especially if
they have a very low stocking density and are unevenly distributed*

1?"} Additional requirements Forest inventories generally include a substantial

amount of field work, which implies high expenditure and more or less
difficult logistics. In particular the ratio:cost of accessibility to the sampling
plots/cost of data recording is sometimes very high. The additional cost incurred by
measuring and recording other parameters, not directly related to the purpose of the
inventory, may prove insignificant. Under such conditions it may be desirable to take
the opportunity offered by the inventory logistics to collect data of value to specialists
not concerned with the primary purpose of the inventory (soil scientists, dendrologiets,
!iy+ ci 'logiRtfc;, etc.). This is all the more justified as a forest inventory often
the most objective and exhaustive way of penetrating unknown and remote areas.

IB no general answer to this question, arid each inventory operation IB a

special case. Many things have to be considered, amori/* which we can quote (a) the cost
of the collection of these additional data, and (b) the qualifications and training of
the field staff for this additional work and the corresponding reliability and precision
of the data collected.

HVen if not explicitly required for the purpose of the in/entory, some data have to
be systematically collected, because they are known to b* useful in any case. In
tropical forest inventories these data are:

- logging
parameters, i.e. slopes, soil bearing capacity, terrain obetacleB,
undergrowth, occurrence of swampy areas, etc.;
-
complete enumeration oi trees by species arid diameter classes above a given
minimum diameter (say 10 cm) in a sub-sample of the sampling plots, if only
certain commercial species are to be inventoried in the main sample;
- enumeration of
seedlings, saplings and poles of the most important species
in a sub sample of the sampling plots, for further regeneration and
majiagernent studies.

Other data, although not of direct relevance to the purpose of a forest inventory,
can easily be recorded either in the office (such as climatic data) or in the field (such
ae seed collection aspects for individual species f degree of dominance of the crowns, dates
and periodicity of seed crops, etc.).

In any case the attitude of the inventory officer, when he is faced with requests
concerning additional data, should be positive. Probably the most advantageous solution
is to ask the respective specialists and researchers or some of their trained staff to
use the inventory infrastructure and join the inventory staff in order to collect their
own data. This solution would also be the most efficient as the data collected would be
more reliable and the cost of the infrastructure shared between the two parties.

V4 Most important specifications for the purpose of a forest

inventory

i) Eract limits and size of the area to be inventoried (the existence of

good recent topographic and land-use maps and/or remote sensing imagery
will facilitate decisions at this stage);
 - ft -

ii) Divisions to he made within the area: this question is important ae the
intensity of the inventory depends on the size of the ultimate forest sub-
division for which results are requested with a specific precision
(these classification exclude the stratif ication(s ) performed to improve
the precision of the results for the above units); these classifications
may be:
baaed on bioclimatic relationshipR ( as for instance land capability
class i fi cat ion) ;

related to *xiPtin land-use and vef^etation;

- related to ^or* r 't
maruvreroent criteria such as:

ownership and
administration
physiography and accessibility
protection (watershed catchment area)
other manAfement critera, e.g. losing compartments
combination ol two or several of the above claasif ications .

lii ) Nature of the information required: Information may be pictorial (maps,

mosaics, graphs and chartR, etc.). descriptive (qualitative description
of the forest types, for instance), or quantitative.

Regarding pictorial information some characteristics have to be defined

such as scale and resolution (what the dimensions of the smallest patch
of forest type to be shown on the map will be). This last characteristic
pertaine also to the precision of th<* required information.
As for quantitative results one may consider the following questions:
do they correspond to a static and/or dynairic appraisal ? (i.e. at the
time of the inventory only, or also concerning the evolution of the forest);
- are they means per unit area, e.^. per hectare (or, for some stands, per
tree) or totals ?
- are the final results areas, numbers of trees, volumes, weights (for
instance, in forest inventories for pulp production), prices (taking
into account the unit prices of the products) ?

iv) Presentation of the information required: Once the type of information to

be provided is know, as well as the desired precision, the method of eventual
preoentation of the results can be decided. TY>e format of the final

tables, for instance, will be drafted and shown to the users in order to get
their agreement. This must be considered as an important item because
clarity and reduction of the "access time to the results are two important
11

qualities of an inventory report. Moreover a dialogue between the users

and the makers of the inventory at this stage, regarding the eventual
presentation of the results, sometimes facilitates a clearer definition
of the objectives by the users.
A set of table outlines for quantitative results concerning areas and volumes,
considered as the basic minimum information to be presented by all PAO forest
inventories, was designed during the meeting of forest inventory experts
attached to UNDP/SF projects held in 196? in Rome. Tke purpose of this
exercise was to harmonize the presentation of the results of PAO operations
(and thus facilitate in particular the periodic assessment of forest resources
at national, regional or world vide levels). Tftiese tables are presented in
Chapter 6 of this Manual which deals with data recording and processing
problems in forest inventory.
 Complete standardization of the tables of results given by all forest
inventories is probably not foreseeable in the near future. However, it
would help considerably if the table outlines in Chapter 6 could be used
as far an possible, with additional tables produced whenever

Precision of the information required: the precision of the results

corresponding to the moot important parameters must be determined prior to
the inventory. For some other parameters t the precision required may nut
be determined exactly, but must not exceed a oertain order of magnitude.
Regarding precision^ three important considerations have to be taken into
account :

a) the total error of a sampling estimate has two components:

- one is the sampling error calculated from the values measured in iho
sampling units, which is related to "precision" in its statistically
restricted sensej
- the second is the bias which may originate either from the sampling
procedure, from the estimation procedure or from the measurement
errors (for analysis of bias in sampling, see paragraph 24 of /

Chapter 3) Sometimes the bias may far exceed the sampling error/
which is, of tan and wrongly, the only one taken into account. When
we speak about precision of an estimate in general, we must refer to
the total error and not only to the sampling error* One must try
to estimate the total error using in particular objective checking
procedures, and must design the inventory to ensure that this total
error is no more than the admissible error. This point is on** of
the most tricky problems and, unfortunately, one of the least studied
in textbooks and inventory reports.

b) Required precision in all sampling designs must be given at A certain

probability level* The meaning and the choice of a probability level is
not always well conceived by the potential users of inventory results,
although this has a considerable impact on the intensity of the inventory
work* Whenever necessary, this point will have to be well clarified
before designing the inventory.

c) Required precision must be referred to a given population, which may be

the whole area inventoried or only subdivisions of it
units, vegetation types, compartments, watershed catchments,
blocks). The mean size of these divisions greatly influences tn*
intensity of the inventory work.
It is desirable that each figure be given at least with its corresponding
sampling error. Precision for many secondary results is frequently not
estimated in order to reduce the cost of data processing. However,
precision of certain figures may be very low f due to the high variability
of the corresponding parameter. It is necessary to point out the
expected low precision of these figures or to omit them altogether (for
instance if the result in related to the volume of a species of very rar<
occurrence, it may be combined with the corresponding figure for rther
species so that the combined figure is reasonaly precise).

I/ Especially in complete (100^ sample) inventories in which, by definition,

sampling error is zero.
 - 10 -

2 Outline for preparing inventory plane

The following outline ia presented as an example of a format which can be used in

preparing plans for a forest inventory* Thin example is given with the knowledge that
there is no single outline which should be used at all times 9 since the outline will
necessarily vary to fit the inventory under consideration. The important point is that
a written plan should be prepared and all the topics shown below should be considered.
Items mentioned in the previous sub-chapter are summarized under the heading "Purpose
of the inventory".

I. Purpose of the inventory

a) General definition of the objectives in collaboration with the potential

users of the results of the inventory.

b) Priority of objectives.

c) Additional requirements (to be discussed with interested specialists:

eoil scientists, eoologists f botanists, etc.).

d) Detailed specif i cations of the objectives:

- exact limits and size of the area to be inventoried
- divisions to be made within the area
- nature of the information required
- presentation of the information required
- precision of the information required

II. General information

a) Authority responsible for the inventory and other agencies collaborating.

b) Available information and data on the area to be inventoried from past

surveys, reports, maps or r-imote sensing imagery on:
- general description of forest
~
variability of parameters to be measured
- condition of terrain, accessibility, transport facilities

c) Resources available for carrying out the inventory.

III. Inventory design

a) Outline of inventory design to be used.

b) General description of the various phases.

i aerial surveys, interpretation of remote sensing imagery

ii mapping and area estimation procedures
iii complete tally or sampling methods for recording of forest characteristics
iv relationships to be used for expressing estimated quantitative data of
stands, e.g. volume tables
 - 11 -

IV. Measurement procedures

a) Description of design for both office and field work; in particular, size, shape,
number and distribution of sampling unite to meet required precision.

h) Procedures of interpretation of remote sensing imagery:

detailed instructions on all techniques;

11 staffing and description of duties;
iii instruments;
iv) forme and recording of observations

c) Field organisation:

crew organization and description of duties;

transportation procedures and directives;
camping instructions;
provisions for logistical support

d) Field procedures including detailed procedures on:

i sampling unit location;

11 establishment of sampling unit;
111 measurements on sample unit;
iv instruments and directives for use;
v) tree and other plot measurements;
vi ) other field measurements such as growth, insect damage, mortality, soil
and topographic conditions, seed collection aspects and information
on non-productive roles of the forests;
vil) design of forms and recording of observations

V. Compilation procedures

a) Detailed instructions on processing of data from imagery interpretation

and field measurements:

i) formulae for estimates of means totals and their sampling errors;

ii) relationships to be used for converting imagery or field measurements
to desired expressions of quantity; e.g. photo-volume tables,
individual tree volume tables, etc.

b) Calculation and compilation methods:

i) description of procedure, e.g. desk calculation, electronic computers, etc.

ii) detailed description of all phases of calculation from raw data on
original forms to final results (for electronic computation, description
of inputs, programmes, and outputs).

VI. Final Report

a) Outline (note that the with some modifications, can serve

inventory plan,
aa a baeie for the final report).
b) Estimated tine for preparation.
c) Reeponftibilitiee for preparation.
d) Method of reproduction.
e) Number of copies
f) Distribution.
 - 12 -

CHAPTER III

BASIC SAMPLING TBC PIQUES

 - 13 -

CHAPTER III

BASIC SAMPLING TBCWIQUBS

1. Introduction

1 1 Sampling in forest inventory

111 Sampling is a necessary technique used in most forest inventories for

economic reasons. Populations to be inventoried t e.g. population of
forest plot units f or population of trees for the assessment of volume tables or of defect,
or for the estimation of mensuration parameters in plantations, are usually too large to
be fully enumerated.

Several sampling procedures may be used in the same forest inventory in different
parts of the operation. Their main, but not exclusive, use IB for the estimation of the
forest areas and of the mensuration parameters. In speaking about the sampling design
of a forest inventory one generally refers to the disposition of the field samples.

Although other statistical techniques may be used in a forest inventory, an example

of which is multivariable analysis - regression - for assessment of volume tables, this
Manual will deal only with sampling techniques. Information on these other relevant
statistical techniques will be found in statistical textbooks, some of which are lasted
in the bibliography.

112 Objectivity in sampling. Sampling must be objective in order not to

introduce a subjective bias in the sampling estimates. A forest
inventory using subjectively selected plots (for instance by selection baaed on the
"
"experience and a knowledge of the forest area of the designer) cannot give valid
estimates as it is impossible to know the importance and the sign of bias and to determine
the sampling error of them. Objectivity IB not synonymous with unbiased estimates of
the parameters. One can deliberately use a biased estimation provided that the bias IB
lower than a given limit. "Ratio estimation" IB a biased type of estimation which proves
very useful in many cases,

Nor is objectivity synonymous with randomness. Msjiy systematic designs are

objective and also give unbiased estimates, although in most cases a systematic layout
cannot be assimilated to a random device. The only trouble with samples of this type is
that their non-randomness prevents the user from applying the statistical sampling theory
and getting unbiased estimates of the sampling errors*

113 Selection of the sampling design. Theoretically the most efficient

sampling design is the one which provides the most precise estimates
for a given cost, or which costs the least for a given precision of the estimates. Under
the term "precision" one must consider not only sampling errors but also constant and
variable biases (see paragraph 24 of this chapter)* The evaluation of cost should also
be the most comprehensive one. In many cases it is very difficult to perform a complete
and precise study of efficiency and to find out the most efficient sampling design.
Approximative calculations taking into account only sampling errors and basic cost figures
point out the best sampling design according to this simplified procedure.

But the selection of the sampling design must not rely only on this partial
calculation based on the sampling theory and available cost figures. The incidence of
the measurement errors, which generally cannot be entered in the efficiency calculation,
must be reduced to a minimum. This means that the tasks in which junior staff are
involved - especially field work - must be easy and simple, as must be their control.
 - 14 -

These requirements are very important and must be kept in mind when selecting the sampling
designs(s) of the inventory. All possible measures must be taken to improve the reliability
of the data, even if this leads to a slight increase of the sampling error and/or *~
total coats.

114 Relative importance of the sampling techniques in forest inventory*

The last paragraph shows that the sampling methodology has to be
considered together with practical matters in order to make the inventory results precise
and reliable.

Furthermore, photointerpretation and forest mensuration techniques which are not -

or only partly - related to the sampling theory are fundamental in forest inventory f for
instance forest classification, estimation of volumes of standing trees, defect assessment,
etc. Thus sampling techniques can be considered as only one among other tools of forest
inventory f and their importance must not be overestimated. Forestry officers involved in
inventory work must have a basic knowledge in this field - and that is why one complete
chapter in this Manual is devoted to sampling techniques. Reliability and validity of
the inventory results are not only a question of sampling procedure; forest inventories
using sophisticated sampling designs may give unreliable and useless results while simple
sampling designs may be a characteristic of good and effective forest inventory operations.

12 Outline of the chapter

The first priority is to define the most useful statistical concepts and describe
how they have to be used in the field of forest sampling. Then the basic mathematical
and statistical techniques will be given which are commonly used in sampling designs, such
as normal distribution hypothesis, variance of compound values, ratio estimation and
optimization of a design. Finally some of the most common classical sampling designs used
in forest inventory will be presented with the corresponding formulas for the estimation
of the mean value of a given parameter per unit and its sampling error.

'
2 Statistical concepts^

21 Population

A population IB an aggregate of units (or elements) of the same nature, the

definition of which has to be clearly expressed. For instance, a forest is considered
as the aggregate of a finite number of contiguous plots of equal or unequal size, or as
the aggregate of all the living trees within the forest (this definition being useful in
the inventory of plantations); a mapped zone is considered aa the aggregate of an infinite
number of points, when sampling by dot counts is used for estimating its area.

It is important to consider the two following points:

- the population may have a finite or infinite number of units and in both cases
the definition of a unit of the population must be sufficiently precise to know
without ambiguity if a given unit belongs or not to the population;

- the units are of the same nature but are not necessarily the same size: for
instance each unit may be a forest compartment, the whole population being a
large forested area; one unit may also be a variable part (or subplot) of a
plot area, if the plot areas are distributed in a partially forested area
(in this latter case, the population is the forested part of the area).

(1) Hie term parameter will be used in the manual to indicate any variate of
a sampled population which is to be estimated by the sampling procedure.
The following remarks can be made regarding the concept of population:

(i) The term "population" ha0 a statistical meaning which is more definite than
in the common language. For instance, there is only one population of
Italian people living in Italy; but it is useful in demographic studies of
this population, to define various statistical populations which have the
same overall size f but the units of which have not the same attributes: the
unit of one of these statistical populations may be the family, another
statistical population could be made up of groupo of people living in a
specific building, etc.

(ii) Sometimes the term population is referred to the aggregate of the values of
one of the parameters to be estimated by the inventory over all the units
and not to the aggregate of the units themselves; we think that, unless
otherwise specified, population must be understood in the sense of aggregate
of the units as sampling is done among the units themselves and not among
the values of the parameter*

(iii) It may happen that the aggregate of all the units cannot correspond to the
whole population. For instance, if the units are circular plots, there is
no way of aggregating them in order to form the whole population. Although
the use of circular plots is common, this problem does not seem to have been
dealt with. For small sampling intensities and a relatively small size of
the plots, this problem can be considered of minor importance and formulas
will be used as if there were a possible aggregation of the units.

(iv) The population to be sampled must be clearly defined, before any sampling
procedure IB designed* For instance, if it IB an area, the limits of the
area must be known. Two considerations are related to thie statement:

(a) Sampling theory does not provide any way to estimate a parameter over
a population if one has obtained an estimate of the same parameter by
sampling only in a part of the population, or in a different population.

(b) A distinction must be made between the "overall population" and the
"population of reference". Let us take an example to clarify this
point* Being given a high tropical forest of 100 000 ha which is the
"overall population" to be inventoried, one may expect a precise
estimate of the mean parameters per area unit for this whole population
or for subdivisions of it ("blocks", "compartments", "management units",
etc*) which we will call "populations of reference". Tnus there will
be various inventory options for the same "overall population 11 and for
the same precision on a given parameter, according to whether this
knowledge is requested over the whole population or over sub-divisions
of it ("population of reference"), Tbe definition of the "population
of reference" IB of course the one which is important from a statistical
point of view. This remark points out the need to state clearly at
which level are the results requested in a forest inventory. Sometime*
decision makers ask for the survey of a given forest area without
defining the sise of the reference unit areas for which they need a
precise estimate of the important parameters. Tney have to define it,
because the range of the sampling intensities of the forest inventory
is wide, from light intensity inventories if they are interested in an
overall knowledge, to very high intensities if the useful sub-divisions
are very small.
 - 16 -

2.? Distribution

221 Different kinds of values' of parameters in one unit of a population

A given parameter has one value in each unit of a population. For instance, let
us assume that the parameter is the "number of stems of Shorea albida of more than 10 cm
reference diameter per hectare", that the population is a forest area, the units of which
are 0.1 ha plots. If in one unit there are two Shorea albida with a reference diameter
of more than 10 cm, the corresponding value of the parameter for this unit is 20. This
parameter has discrete values, because the corresponding it^ms (stems) are not divisible.

If, in this population, we consider the parameter "gross volume of the bolea of
Shorea albida with a reference diameter more than 10 cm per hectare" - the volume of a
tree being related to its reference diameter and height through a volume table - the
values of this parameter corresponding to the various units will be continuous, because tht
values which can he taken by the parameter are close to one another according to the many
possible combinations of numbers of trees in a plot with various possible reference
diameters and heights.

The value in a unit of the first parameter is determined through a count while that
of the second is obtained from a count and measurements of specific characteristics of the
trees which are used to estimate volumes through a regression equation (volume table).

Very often a parameter of a population is estimated by assigning each sampling unit

to a given class. Let us suppose we want to estimate from a certain number of photo-
interpretation plots of equal size the proportion (in area) of a given forest type within
a forest area (population). To each plot we will assign a variable, the value of which
for the plot is:

if the plot is not, or only for a minor part, in the forest type;
1 if the plot is entirely, or for a major part, in the forest type.

The mean value of this variable over all the sampling plots (which will be a
positive value between and 1 ) is an estimate of the proportion in area of this forest
type within the whole forested area.

If we consider all the forest types of this forest area, it is easy to verify that
the sum of the related estimated proportions IB equal to 1 (and that the variance of this
sum is equal to 0, which is an expected result as, whatever the sample, the sum of these
estimated means is constant and equal to 1).

Most of the following considerations in this chapter apply equally to sampling for
estimation of parameters resulting from measurement, count or assignment. The way of
calculating the variance is the same for continuous parameters and variates (0,1) but the
resulting formulae are different, for they can be simplified in the case of the variables
(0,1) used for estimation of proportions.

222 Distribution of the values of a parameter over a whole population

The value of the parameter to estimate in all the units of the population are
distributed in a certain way. Let us assume the parameter can take only discrete values
in each unit of the population, e.g. when it is the number of stems of a given special.
If we consider all the unit* of the population, we can represent the population by a chart
of points (tig. 1) the coordinate* of which aret

- on the x~axis the number of stems of thia apeoiea in one unit;

- on the y-axia the number of unite of the population having a given
number of stems of thia apeciee (or frequency)

number number A
of unite of unite

3 4
number number
of atems of atema
per unit per unit

Very often auch a graph is presented in the form of a histogram: for each discrete
value of the parameter there will be a rectangle baaed on the corresponding ordinate, the
height of which ia given by the frecruency related to this ordinate.

In the caae of a continuous parameter, we can gather the information by classes of

parameter value. In this case we will also get a histogram} the smaller the classes, the
narrower and more numerous will be the rectangles*

We can imagine that if the width of the classes (and of the corresponding
rectangles) decreases, the representative points will be closer to each other and their
y-ordinates will also decrease* The distribution can then be represented by a curve
which Joins the points (distribution curve)*

One distribution for continuous variables is very useful in sampling techniques.

It is the normal distribution with a curve symmetrical in relation to the axis of
x-ordinate equal to the mean value of the parameter. We will see that if the estimates
of the values of a mean obtained from all samples of the same type are distributed
"normally" around the expeoted value of this mean, it is possible to give the sampling
error of this estimate at a given probability level*
 - 18 -

223 Characteristics of central value and dispersion of the distribution

Two characteristics of the distribution of the values of a parameter over a whole

population are of particular interest for its estimation* They are (assuming the
population is finite):

a) the mean value of the parameter per unit* If:

x is the parameter

th
XA its value in the i unit

V the total number of units in the population

the mean X of the parameter per unit will be equal to:

We can also write X - JL *

p.jc.

where p, * is the probability for the parameter x to be equal to x. * t the

* >
*
.. .
, being extended to all the possible values x. of the units of th
whole population;
J. is also called the expected value of the parameter x in the population
which is written B(x);
the total of the parameter over the whole population is the sum of the values
of the parameter for all the units:
x -

b) the variance of the value of the parameter in a given unit is

This characteristic is a measure of the dispersion of the values of the

parameter around the mean value of the parameter*
2 2
We can write alsoj (f x . ,
- X) 2
PI< (x^ -JE^

(p,* t x. and ~> having the same meaning as above)

The variaooe is also the expected

expect value of th^ sguaxe. of ths dsviations
between the values of the parameter inn the
ibt _population
_ and the mean ^t and is written:

- 2
Bifx B(x)7
 - 19 -

The square root of the variance is the standard deviation which iss

The standard deviation 1m not a pure number and depends on the system of
measuring unit*.
<Tx
The ooeffioient of variation C -
is a pure number and is very useful
v
x
to characterize the variability of a parameter over the whole population,

If two parameters are to be estimated in the same population x and y, we

can define a cross measure of variation, called oovariance i

v U,y)

y^ and Y being respectively the value of the parameter y in the i unit and the
mean value of y over the whole population*

correlation ooeffioient o between y and x in the population defined

by the equation!

will characterise the cross variation of the two parameters y and x in the population

( f in a pure number whereas u^y depends on the units choeen for estimating the

parameters x and y).

224 Value of a parameter per area unit in one unit of the population

As we have seen from the example in paragraph 221 there is one value of the
parameter "number of trees of a given species per hectare" in each unit of a forest area*
This parameter has a mean value over all the units of the whole population, which is the
mean number of stems per hectare of this species* In order to avoid confusion we
recommend using the complete expression "mean value per area unit" to denote the mean
value of the parameter per unit (of the population) referred to the area unit. If all
the units of the population have the same area (size), the mean value per area unit is
equal also to the mean of the values per area unit in all the units of the population*
 ~ 20 -

23 Sampling

As in most oases the population to be studied is too large to be fully enumerated

and/or measured, it is feaaible to have this work dona only on a aalaction of units from
tha whole population* Such a aalaotion ia oallad aampling, tha aalaotad alamanta are
tha aampling units, and the whole aat of sampling unite ia tha e ample. Provided given
prooedurea of aalaotion are followed y sampling techniques are useful:

i) to get an eatmate of the true values of the mean and of the total, over the
whole population, for a given parameter, from the valuea of thie parameter
in the sampling unite;

ii) to get an estimate of the aampling rror (or of the precision or of the
confidanoe limits! at a given probability level, for the eetimated mean or
total given by the sample.

Estimating the aampling error ie sometimes very difficult and, in oertain caaea 9
the formulae used are only approximate* This calls for the use of simple sampling
designs whenever possible.

Some concepts have to be stated clearly in relation to sampling*

231 Sice of sample

Given a sise unit which may be an element or an individual (such as a tree) y or a

measurement unit (such as a hectare) the size of a given population and any of its units
must be measurable* The unite of the population may be of different sizes this is the
case where a forest area has curvilinear limits y the units of which are stripe of equal
width but of unequal length, the size being expressed, as in many forest invent oriee,
in tents of area*

The sise of sample is the sum of the sizes of the sampling units*

The sampling intensity (or aampling fraction) is equal to the ratio of the size
of the sample to the total sise of the population. If the population has N units of
equal sise and if the sample baa been made up by selection of n different sampling
unite | then the aampling fraction f ia equal to:

In this oaae (equal units) n can be aaid to be the eize of sample.

A difficulty ariaaa when the unit ia a point and therefore has no dimensions. This
oaae occurs when one eetimatee the proportion of area in a given foraat type over the
whole inventoried area, using a dot grid laid on a foraat map or whan the forester is
using a point sampling method (aae paragraph 422*2). The sampling fraction in the first
oaae will be aaid to be negligible, whilst in the second oaae tha sampling intensity will
vary according to tha relevant characteristic of the traaa (baaal area, diameter, square
of height, height) i if it is tha horizontal point sampling method, tha sampling
intewity of traaa cf 56 am diameter will be exactly aqual to four times tha eempling
intensity of traaa of 28 OB diameter* Generally, even for the biggtst or higheet traaa
this sampling fraction ia avail and can be considered negligible in the sampling error
calculation*
 - 21 -

232Precision and sampling error* The estimate of the mean or the total of a
parameter given by a sample is generally different from the true oorree ponding
value over the whole population* The sample will be more valuable am the estimate become*
more accurate f which means that the estimate is closer to the true value. This accuracy
cannot be expressed in an absolute way as, in the following sentence, "the accuracy of the
estimate of the mean of this parameter over the whole population is + 3^* It can,
however, be expressed in a probable way, and is called precision or sampling error* In
the former example precision is stated as being equal to +3# aT~a given probability level.
If we say "at 0.95 probability level", this will indicate that for 95% of similar samples
drawn from the same population the true mean will be within the +3% interval fro*
estimated mean given by every sample* It is also expressed by saying that the probability
of the true mean being within the +yf> confidence interval is 0*95, or, which is equivalent,
that the probability of the true mean being outside this interval is 0.05. In this oase f
the 0.05 measures the risk we are prepared to accept of bein wrong when we say that the
sampling error is +3J& at 0.93 probability level. Whatever the sampling, the sasmlJM
error would be meaningless and infinite (or the precision will be null) if one attempted
to express it at a 100% probability level.

Itoe selection of a probability level to which the sampling error of an estimate

corresponds mist be made by the user(s) of the forest inventory. It is not enough to
request a given precision; the probability level must be specified. A precision equal
to 10#
-i- at 0.95 probability level ("19 out of 20") is better than th* same precision at
0,6$ probability level ("2 out of 3"). We will see (paragraph 31) that in many oases
the seoond precision is equivalent to + 20$ at 0,95 probability level* Any iaaplipj:
precision given without a reference to a probability level is meaningless.

The most frequently used probability levels in sampling techniques are 0.95 *&d
0.68 (the reason will be given in paragraph 31 ) These two probability levels are
sometimes referred to respectively as 0.05 *** 0.3^*

Sampling errors are expressed not only in percentage of the oa timated result but
also in the corresponding measuring units. In this oa*e it is also possible to express
the "confidence limits" at a yiven probability level in these unit** For instance, if
the sampling estimate of a mean volume per hectare is 40.0 nH and the sampling error
+ 5.0 at 0.95 probability level, the confidence limits will bet
3 - 0.05 x 40. Osi 3 - 38.0m-5 - lower confidence limit at 0.95 probability level
40.0si

40.0r + 0.05 x 40. Os^ - 42. Om^ - upper confidence li*it at 0.95 probability level

the confidence interval being (38.0 - 42.0).

the probability for the true value to be less than the lower confidence limit is
equal to the probability for it to be more than the upper confidence limit. In oaae of
a 0.95 probability level, this means that the true value ha* a 0.025 probability to be
less than the lower confidence limit and a 0.025 probability to be store than the mpper
confidence limit.

II may be useful to refer the estimate to only one of the confidence limits.
tfhere theestimate is a mean exploitable volume per hectare in a forest, we are acre
interested in the lowest volume to be expected at a given probability level. In this
case the lower confidence limit at the probability level is called the "reliable
If the probability level corresponding to this confidence limit it
ee1JBat" (HM1).
0.95 (or 0.05), the reference probability level for this HMJ, will be 0.975 (or 0.025).
fhis means that this HUE will h*ve 0.975 probability to be exceeded by the true value,
or 0.025 probability to be higher than this true value*
 - 22 -

In a given forest inventory, several sampling designs may be used, one for the
interpretation of the remote sensing imagery (area estimation), one for the field work
(estimation of number of stems), a third for the calculation of volume (using a sample of
completely measured trees for elaboration of volume tables), etc., and each will have a
different sampling error which may be related to the others. The total sampling error
is a combination of these partial sampling errors and is not the mere addition of then.
In most oases, it is difficult if not impossible to estimate it properly. This calls,
once more, for the use of simple inventory designs which provide the \iser with reliable
results with the requested precision.

233 Other concepts. The selection of the units of the sample (or sampling in
the narrow sense of this word) must be made at random* this is a fundamental
requirement for the application of the sampling theory. For instance, there is no
completely valid method of assessing the sampling error of an estimate given by a
systematic sampling design, because the basic requirement of randomness is generally
lacking. This does not mean that the estimate itself is not valid (see paragraph 422 of
this chapter).

Probability of selection in sampling

Sampling theory can be applied in principle only if the composition and probability
of selection of all possible samples of a given sampling design from the whole population
is known. In practice it is sufficient to know the probability of inclusion for the
units themselves. 7ne probability of selection of a unit is the chance that it has of
being drawn for inclusion in a sample during the constitution of the sample. This
probability is expressed as a figure of less than one; one, or certainty, being the sum
of the sampling probabilities of all the units of the population. The probability of
selecting a given type of sample of n units can be assessed from the probabilities of
inclusion of each sampling unit in the sample.

Examples (sampling with replacement):

a) In a simple random sampling of a population of N units, in which all

units are selected with equal probability, the probability of selection
of the unit i is equal to
N

P -
$
w ith
^P! -Nx
b) In a sampling with probability proportional to the size M^ of each
unit (called PPS sampling) we will have i
N
P1 *
Mi
^th M -
y being the size of the
ii f-^Hj
"
whole population.

Replacement

When the first unit of a sample has been selected, two alternatives are possible:

to draw the second unit from the whole population including the first
sampling unit (sampling with replacement);
or to draw the second unit from the population which doee not include the
first sampling unit (sampling without replacement)

(the procedure used being repeated after each drawing of a sampling unit, until the full
pie is constituted).
 In sampling with replacement the same units may be drawn and included in the sample
more than once*

Most of the formulae used in practice are valid only for sampling with replacement.
However, almost all samples are drawn without replacement. Thin approximation is
acceptable provided that the number of sampling units is relatively small in proportion to
the total number of units in the population (or the stratum if it is a stratified sampling
see paragraph 4114)

24 Bias and measurement errors

241 Bias. Uie conotpt of eip+oted value (see paragraph 223 of this chapter) can
be applied to the estimate given by a sample. If^R is the estimate of a
parameter given by a certain type of sample, and >j the estimate given by the sample J
from this type of sample, the expected value of is:
yU

where 7f
j
is the probability of the sample j (%7Tj 1) *nd the sum ~ is extended
J J J
to all tne samples of the same type.
A A
y
The estimate 4 from a sample of this type is unbiased if its expected value
^
E(/*)
is e(rual to the actual value u of the parameter over the whole population! E(A) M .

If it is not so, the estimate is biased and the bias is equal to the difference
between the expected value of the estimate and the actual value of the parameter of the
whole population.

In this caset 1(U) - m / /* with the bias: B - m -/l

Hie concept of bias relates also to other estimates given by a sample and in
particular to the estimate of the sampling error.

(1) Let us suppose that the characteristic ^ia the mean basal area per hectare g
of all trees of more than 10 cm diameter in a given plantation, and consider that we
estimate g from a certain type of a sample of trees. Let us assume that the section at
breast height of all the tress is circular, but that the tape used for measuring the
diameters has shrunk and says 21 cm when the diameter is actually 20 cm and that there
is no other possible measurement error.
2
21
Efcch tree will be given a basal area *' equal to(-rr)g 0? 5 gt being its V
actual basal area. The expected value E(J; of the mean basal area per hectare over all
the samples of that type will bet g* - V025 g.

Here the bias is equal tot

B ? - - g (1,1025-1) - O t 1025 g
g
 (2) Let us assume that the characteristic/* is still the mean basal area per
hectare "g of al trees more than 10 cm diameter in a given forest area*
1
The sample is
constituted of n sampling ujiits of unequal sjge sampled with equal probability* There
is no measurement error*

A commonly used estimate for g is s

r -

where I i is the reference number of a sampling unit

is the sun extended to the n sampling units of the sample

gl is the total basal area in the sampling unit i

l is the area of the sampling unit i

It is called a ratio estimate and is sometimes used in forest sampling because it is not
always possible to have sampling units of the same sise,

Inis estimate
111 is biased and the bias has an order of magnitude
**
of * *
n

(3) As ha* already been stated (paragraph 112 of this chapter), a bias ay result
from an incorrect sampling design* For instance this occurs when the sample is constituted
of the so~oslled "representative" s sapling units chosen by the forester on the basis of his
experience" and knowledge* Another case of bias due to the sampling may arise when some
perts of the population to be sampled are not taken into account in the sampling design:
this occurs in forest inventory when, for instance, sampling units falling in less
accessible areas are systematically replaced by more convenient units*

These three examples show that bias may have three origins t

- measurement errors ;

- estimation of the parameter from the sample;

- sampling procedure.

Bias due to the sampling procedure is not always reoognieed and practically
impossible tc evaluate* AJI estimate of the bias can be provided in some biased estimation
procedures

Measures] en t errors* Th* measurement errors can be split into three

components t

a oonstant bias over the whole population (as for instance in the case of
eaeurenents bj a device in which the cero graduation does not correspond to
the sero me Mure);
a variable component, in relation to the sampling unit which nay be correlated
tc the exact value of the Measured parameter in the corresponding unit (for
instance, as in the case described in the first example of the former
pmrsfimphs the variable oosiponent of the bias on the diameter Measurement
is in this csse erpMil to 5* of the disaster );
 a "fluctuating" component in a given sampling unit of mean (its variance
could be estimated if several measurements of the sane parameter were taken
in the given sampling unit),

The measurement error affects the reliability both of the estimate of the mean (the
expected value of which is different from the true value of the mean) and of the estimate
of the sampling error* Assuming that the sampling design is correct and that the exact
sampling error calculation is done, the resulting estimate cf the sampling error is a
biased eatimate of the true sampling error.

3 Basic mathematical and statistical techniques

31 Principles of sampling error estimation

311 Introduction* Let us call/^the true value of a mean of a parameter over th*
^whole population, the estimate of this value given by a certain type of
sample an<LMj the estimate given by the sample j of this tyj*. If we takt all th
samples of this type , the corresponding estimates will be distributed in a certain way.
The distribution of the estimates jl will have the following characteristics!
A
(a) characteristic of central value (mean of the estimate

whores M j
stands for the probability of drawing the sample j (
iTTj 1)
j

E stands for "expected value" (see paragraph 223)

and *g extended to all the samples of this type.

j
1
If all the samples have the same probability of being drawn (TT, TT. -
where M is the total number possible of samples of a same type), we will have:

We have seen (paragraph 241) that iff* is unbiased wt will hmvej m

(b) Characteristic of dispersion (variance of the estimate )

(with th same annotations)

If all th* sanplts bars tht probability of bing dram (TTj

- 1T
|)
wt will have i T A o
 - ?6 -

Attention is called to the fact that the distribution which is dealt with here is
the distribution of the estimates of the mean value of a parameter over the ufaoje
>*j
population given by all the samples of a certain type, but is not the distribution of the
values of this parameter over this population.

Sample i Let us consider a forest stand (population) of N equal units of 0.25

hectare* is a sample f we select at random with equal probability and without replacement
n (n ^ N) units ("sampling units") where we enumerate the number y of stems of a given
species with a reference diameter more than 10 cm (this is the parameter).

The total number of possible samples of that type is equal tot

M N n ""TPTTir^"TT
nl (N-n);

which is the total number of distinct combinations of n different units from the total
number N of the population units. All these samples constitute a population of
element*. ffcch different sample has the same probability IT 1 of
TTj
selected (they have an equal chance of being drawn, which can be
calculated from the probabilities of selection of their n successive units).

estimate from the cample j of the mean value (per unit) of y is:

where the sum ~ is extended to the n units i of the sample j.

(a/ There are M possible estimate* y and the expected valuf of y ist
J J

ly.
-
8(7,) -fWj Jj
"
V
the nim ~ bein^ extender* to the K samples j (all the Bwnples of that type).
j

If there IB no bias due to enumeration or measurement errors, it is easy to

prove that for this type of sampling, the expected value of ^jiTi ** IB!* 1
to the trqe value T of the mean value per unit of t^e ptfl^w y (see
paragraph 241).

Thw y, is an unbiased estimate of Y and this type of sampling, which is

called simple (or unrestricted) random sampling, is an unbiased sampling design.

T Ny is likewise an unbiased estimate of the total T of the pavmster

^ J
over the whole population.

PI B**ui the product of the first N mwbersi Ni 1 x 2 x 3 x 4 x ... T (N-1) x M.

 (b) The variance of the estimate y. (i.e. the characteristic _>r
of the values y for the M possible samples of that type) ist
j

V (v
J
) - r
E ^ y
j
E vy
(v

the Bum ~- being extended to the M possible samples.

J

Sampling theory shows that V(y ) for simple random sampling IB equal to:

'
J n N

where Q^ stands for the variance of the values oi th<- parameter over the
whole population of N units.

The sampling error at probability level Qg5 on y ia equal to:

- 1,96
. ,., $
.
(y.,)

provided that n is large enough.

""
5 c *llad the standard error and corresponds, in case the
Y 1 *

normal distribution hypothesis is verified, to the sampling

^rT"
error at probability level 0,68 (see paragraph 31?1)

It is shown that an unbiased estimate of this sampling error from the

^t probability level 0,95) is

.
(?j>
. ,,96-t. /L-
where s' - - a
'

is an unbiased estimate of (j from thr

n i

(the sum ^ being extended to the n units i of the sample).

Similarly an unbiased estimate of the standard error (at probability

0,68) ist

This sxampls show* that sampling theory giv< for a given sampling design* "the

following estimates!

estimates of the true mean and total of the parameter for the
whole
ole ppopulation ;

estimate of the sampling error on these estimates at a given

probability
protabili level.

ren if there is no bias due to the sampling design or to measurement operations,

biteed estimates may De preferred for the sake of convenience or simplicity in the
o*loultion In these cases oare should be taken to keep the bias as low as possible.
 - 23 -

It should be kept in mi ul as has already been pointed out, that we cannot have the
f

true values, but only estimate^, of both the mean (or totpl) of the parameter and the
sampling error. Generally speaking, ood (unbiaeed or biased with a very en\e.ll bias)
estimates of means and totals are not difficult to determine. Acceptable sampling error
estimates (on these estimates) are more difficult to calculate. The following paragraphs
give some basic techniques used for the estimation of the sampling error.
A
31? Estimation of the sampling error on
distribution hypothesis.
^
fro itc variance. Normal
As has been seen in the case of simple random ^
sampling, sampling theory gives an unbiased estimate of the variance of the estimate^,
provided that some requirements are observed (random selection, sampling over the whole
population and not only over a part of it, etc.).

problem now is to estimate the sampling error at a given probability level from
the estimate of this variance, which is^again the characteristic of dispersion of the
distribution of all possible estimates A
(the characteristic of central value being m,
equal to the true value
in the estimation).
^
of the parameter over the whole population, if there is no bias

31?. 1 Simple random sanpliup

A
In this type of sampling we will assume that the distribution of all possible^,
corresponding to all samples of the same size n (which is not the distribution of the
parameter over the population) is a normal distribution. This assumption is approximately
verified when the numbej n of sampling unita is lTg enough: it is usually agreed that
normal distribution of/S i obtained when 3oO ) B whatever the distribution of the
n^
sampled population. The sampling error on
to the square root of its variance:
^the probability level u is proportional

Uie square root of the Variance V( ..

) is called the atancUird error.

(1) Ihis agreement of course is somewhat arbitrary. For instance, if the sampling
units are equal fixed-siae plots of forest area and the parameter is^the "number
of stems of a givn species more than 10 cm diameter", the estimate^ of the
mean value par unit of this parameter may have a more or less normal distribution
when evaluated from only fiftwn 0.4 ha sampling units, while the distribution
of/Ij cannot be considered normal
when the sample is constituted of sixty 0,1 ha
plots. With this reservation in mind, the figure 30 may be considered as a
good order of magnitude of the minimum siie of a sample in ascertaining whether
the distribution of A. is a normal distribution*
 - 29 -

average value* of t corresponding to different probability levels for samples

with n ** 30 airs given in the following tablet

For other t value* oorree ponding to other probability levels and epeoifio valuee of n,
U
one ha* to r*f*r to the Student * table and the table of normal distrbution.
1

The two mo*t u*eful re*ult* 9 when 30, aret

- at a probability level 0,68 ('one ohanoe in three of beirur wrong" ) the ampling
or in^j i* *qual to th* *quar* root of it* variano* j/VCy&j) f i.e.
error
tandard frrori

- at a probability level 0,95 ("one ohanoe in twenty of being wrong") the sampling
error in A. i* approximately twice the square root of the variance.

Wfren n<
30 (numberA of emmpling unit* in th* *mjpl* le** than 30), the aa*u*ption
Ihere ij generally no way of
of normal aisTricwvion */*j in no longer acceptable.
*tribution of/x,
obtaining th* sampling *rrof from th* variance of the due to the effect of
estimate/^,
the inaoouraoy of the estimate of th* variance from mall *ampl*t. Howsv*r, if w* can
consider that th* valu** of th* parameter are more or les* normally di*tribut*d i^tjj*
population, then we will *till have th* following relation between * u (yu ) and

the t value* being taken free) th* Student's table.

Ihi* table i* a two-entry table, one correeponding to the probability level, the
M
other to th* number of "flMTte* of fredosi . Thi* la*t oono*pt is not dealt with in
this manual. It is *nou^i to know that in the simple random sampling, and in many
other classical designs, th* timber of degree* of freedom i* equal to n-1, where n
is th* number of sampling unit* in th* whole population (or in a stratum in stratified
sampling)*
 - 30 -

The following table is an excerpt from the table of Student's t for selected
probability levels and sizes of sample:

These t -values can be used f in the case of small samples (n < 30), only if the distribution
of the parameter values over the population is roughly normal.

312.2 Other sampling designs

Simple random sampling is only one of a very large number of possible designs. In
many forest inventories these other sampling designs have to be used and there is no simple
answer as to whether the distribution of /A can be considered as a normal distribution or
not. 'Hie "normal approximation" will be ^raore verified as the number n of sampling
unite becomes larger, the desirable minimum value of n depending on the distribution of
the parameter over the whole population, arid of course on the type of the estimate Jt ,

itself a function of the type of sanpling* The "normal approximation" is generally

accepted, but one must always check if this is reasonable. The table of normal
distribution will be used as indicated in the first part of paragraph 3121. It may be
desirable in some cases to stratify the population into several sub-populations ("strata"),
within each of which the distribution of the estimates u. may be considered more "regular"
^
and closer to the normal distribution.

32 Variance of compound values

321 Introduction. Let us consider a random value y, which may be the value of
a parameter in one unit of a given population, or the estimate of the mean
value per unit of a parameter over a population from a sample j of a certain type*
IMc random value may be estimated, not directly from measurements within the sampling units,
but through other random values which are themselves directly measured within the sample
or already estimated from the sample.

Examples > An unbiased estimate of the mean value per unit of a measurable parameter
(e.g. basal area per unit, taking into account all trees more than 10 om diameter over
bark) in a simple random sample is:

(1)

where i refers to all the n units of the sample.

The estimate J is a function of the random values y which are directly measured
in the sample.
 - 31 -

~An unbiased estimate of the mean value per unit of the sane para/neter y for the
whole (population) of two stands (oub-populationa or strata) sanpled independently, is:

S ~ S ~
(2)

where 3 and y correspond respectively to the total area of the i stand and the
estimate of the mean per unit in this stand (i - 1 or 2), while 5 is the Hum of th*
S

The estimate y here is a function of the two estimates y. and y ,

The unbiased estimate of the variance of the value y. of the parameter in one unit
is easy to obtain. So to estimate the variances of the estimates y in (1) and (?), we
need to know the relations between:

- the variance of a mean of values (formula (1)) and the vpriance 01 an

individual value;

* the variance of a weighted mean (fomula (?)) from the vanancer of

the varioun means.

More generally the definitive estimates of a mean of n t)irancter from Rfiwplea f

other than the unrestricted random ones are functions of other partial entimateu.
F

variances of these definitive estimates (and then of their sanplin^ errors) can be
estimated from the variances of the partial ones through the use oi the following
relationships.

32? Variances of some functions

322.1 Variance of a product of a random variate and a constant

We have V(ky) k^(y) where y ic th random vnnat? -I'K!

k IB the couctant.

For instance, if y is an estimate of the mean value of a parameter per unit it. n
simple random sample and N the total number of units of the population (which ir
supposed to be known exactly), the corresponding estimate of the total of the parameter
over the whole population isi

T -
Ny

and the variance of this total volume will be:

2
V(T) - N V(y)

322.2 Variance of a sum of independent .random values

a) Independence of two random values

Two random values are independent if their covariance is zero (see paragraph Pir'3

for the definition of covariance).

 b) Kxamples of dependent random values

(i) Given a line (population) of 1,000 trees (unitn) rQon^ a roadside we want to have
an estimate of the mean reference diameter (D3H) per tree by neane of a systematic
sample of 50 trees, one tree every 20 trees. The rank number of the first sanple
tree will be selected at random amorxg the numbers from 1 to ?0, for instance 13,
and the following sample trees will be the 33rd, the 53rd ... until the 993rd.
The "stinate of the mean DBH will be actual to;

(DBH) .
-
(DBH)-, +
*
(DBH) M
-**
+ ... +

More generally, if k - - (N being the total number of units of the population

and. n the number of units of the sample), the estimate of the mean of a
parameter y from a systematic sampling can be expected to be as follows:

y - ~ r 1 "-/* wi th T J
^ k

There are k pairs of valueo (y^, yi+fc) from i - 1 to i - k. The values y^

and y\+% will not be statistically independent if their covariance is not zero,
that is to say if:
k
'

i-1

where

and

Likewise the values y^ and 7i+2k m V

*lso not be statistically independent if
GOT (y^ 9 yi+2fc) ^ 0* whioh means that even the double interval (2k) between the
trees is not enough* More generally, the same reasoning can be applied to any
pair of the random values y of the numerator of y*

In the example of the roadside trees, this independence may not be fulfilled if,
for example, the distance between two consecutive sample trees is too small (possible
interaction between trees), or if this interval is more or lese the same ae the "waveleng
of certain eoil characteristics.

o) Qitlmates from samples are not independent if the samples themselves are depend

Let us again consider the two stands from the example given in paragraph 321, and
let us suppose that saw pi ing is not independent in these tuo strata! this means that the
number and/or the composition of all the contemplated samples in the second stratum is
dependent on the selected sanple in the first Stratum- (or vice versa).0) It is eas

(1) l*i is could be the oast if, in order to avoid too expensive a forest inventory, it
were decided for the second stratum to select only from, those with a small proportion
of eattple plots with difficult aooese when the sejnple from the first stratum already
has a relatively large number of such plot*. It is obvious in this oase that the
number and the composition of all possible samples in the second stratum will depend
on vne sample selected in the first stratum.
 - 33 -

to demonstrate that in thip cape and y2 (estimates of the mean of a parameter y

y<j
reepertively in the first and second strata) are statistically dependent estimates, which
means that their covarinnce is not zero:

*Ll
(y, yJ -

where Y. and Y ae the triie value* of the means per unit in respectively the first and
second strata," y-j^ and are respectively the estimates of YI from the sample j
y^v
in the first stratum and of Yp from the sample k in the second stratum.

MI and Kp are the numbers of possible samples of the given tvpes in the strata 1 and
?, ~
the sum ^" bein/' extended to all the M< x Mp^ cross products
* (y
X 7 4
Y. /x
) (y^. - Y).
'
/ *
1i1 *'2k2

d) It can be demonstrated that the variance of the sum of independent random

values is the sum of the variances of these values.

If y 2 v .i
1
9
<the
y^X being independent,
i

V(y) -
2v(y )
i

3?2.3

We have y - --
Variance of the estimate of a mean from a simple random
sample for an infinite population

y1 + y2 +

By application of the two former theorems we will have:

+ y
n
and the y are independent

V (y)
-J? /V (y,) + V (y 2 ) + ... + V (y n )_7

All the n variances between brackets are equal to the variance Q (y) of the
values of the parameter over the whole population and we will have the very important
result:

The variance of the estimate of the mean in simple random sampling is eqrual to the
variance of the value of the parameter over the whole population divided by the number "of
sampling units in the sample.

Consequently the sampling error on the estimate of the mean (which is proportional
to the square root of the variance when n/N is small), will be inversely proportional
to the square root of the number of the sampling units.
 Case of finite populationn

The f^rm-r results air* v^lid for infinite populations or for finite population* when
ajnplm fraction (or ea'nplinf intensity) f rr is relatively small (nay 5?b).

In the ra*e of finite populations, the sampling theory shows that in fact we have:

and

The tern- 1-f and ^/T-f are called the "finite population corrections'* and may be
ted if f is snail*

itroduction of theae terms can be justified a posteriori; if the sampling

complete censue, standard error (or sampling error) must be equal to
a
result can be obtained from the above formula as, in this case, f 1,

2 ? 4 Variance of a weighted linear expression of random values

If we hav-: y =
a^ +
?y ?
+ ... 4 a y
p ?
.
^ a.y.

(a. f
SL^
... a being positive or negative constants)

the application of theorems from 3^2.1 and 3?22 gives:

2
V(y) - a^ V( yi ) +
a/ V(y ? ) + . . . + a V(y )
p p
In the case of the second example of paragraph 3^1 we will have:

s s
S,
V(y) - ~- V(y J +
c
4-V(y
c ' 2
s s

This theorem is sonetimes called "theorem of propagation of errors 11

and is used for
stratified random sampling.

322.3 Variance of products and ratios of two independent random values

Let DC and y be two independent random values, the variances of which are small
in comparison with the squares of their respective values* We will have the approximate
formulae:

V(y)
 3^2.6 Variance of functions oi' dependent random valuen

If i yoi v are d c P ndcn<fc random value* and a., a ... a positive or negative
y-j

constants, the variance of y:

p
y -
-^ +
a^ + ... + a y
p p
- a.y.

where the sum . is extended to all the different possible combinations of i and j,

j being different from i f

and V(y , y ) IB the covariance between y, and y .

If a, x ap = a and the y are the values of the same paramptrr in the

p consecutive equal size sampling units of a systematic cample, y IB, in this cn^r, tlie

estimate of the mean per unit from thia sample and can be \mttrri y. Its vprianre

T a/ V(y ), for the covarianc^s V(y . y

^r i

which mpJcr up the additional tern will, in most eases, not b* equal to 0.

More generally, it can be said that the application of the formula of random
sampling for the estimate of the variance of the estimated mean IB not valid in many
systematic sampling designs for many parameterc. In a systematic forest inventory, the
gquidistance between neighbouring jiamplinf: unita will have to be nuf ficirntly lon: in
order to avoid any positive C r negative) covariance (or "correlatior.' ) between the 1

respective values of a qiven parameter f if one wants to apply the relation

px ^ V(y)
Product and ratio of two dependent random values

In this case, provided that the variances are anall in comparison with the
of the respective random values, the formulae of paragraph 3^?^ become:

" ft

V(x) -f x V(y) 4 Jry V (x, y)

. v (^. y
?
xy

V (x, y) being the covariance between x and y.

33 Ratio estimates

ttie precision of the estimate of the mean of a parameter y is increased if we can

relate it to another parameter x f which is known exactly for the population, or is known
more precisely from a large sample* On the other hand, one may be interested in knowing
not the single parameter v but rather a combination of it with one or more other
parameters* In forest inventory if we have sampling units of different size, we will
have to know two random values in each sampling unit:

- the value y. found in the unit for the parameter y itself;

- the area z. of this unit.

The requested estimate will be essentially an estimate of the mean value of the parameter
y per area unit (not per sampling unit), which in fact corresponds to the ratio
Z of the two parameters,
x
Because in many forest inventories, sampling unitn will not have the same size, the
use of an auxiliary variate x related in ^eiieral to the size of the sampling urqt appearg
to be necessary i n. many cages . Hie model of the relationship between y and :: used i*,
essentially, a linear Ones

y a -f bx ("linear regression eatimates")

and often y bx ("ratio estimatea")

The first model is acceptable if the relationship between the two paranetern ic
strong and may be satisfactorily represented by a straight line. It is equivalent to
say that the correlation coefficient must be as close to 1 as possible. For efficient
use of the second nodel it io necessary, in addition, for th^ line to go through the
origin; in other words when x and y tend to together. Tins is particularly th<-
case of the variaten "area of a sampling unit" and "any usual forest parameter": when
the area tends to be 0, the forest parameters tend also to be G.

That is why ratio estimates are used in forest sampling.

TWo types of ratio estimates are commonly used in einp]e random sampling and they are:
n
* x
R
v*-
- -=i1 "ratio of the means"
n
x
x.
i
i-1

and r .^i
~ * '
"mean of the ratios"

Paragraph 322.6 indicates how to obtain an acceptable estimate of the varianrr of ff,

provided that the variances are small in comparison with the squares of the respective
estimates x and y.

34 Optimisation^ in design

341 Optimisation of a sampling design. The most efficient sampling design is the
one that for a specific cost gives the smallest error for the parameter to be
estimated or for an accepted error is the least expensive.

(1) Optimization is used in the sense of setting out and selection

of the most efficient or optimum design.
 34 1 1 Problems related to optimization.

Definition 01 efficiency is easy, however to find the most efficient design is much
more complicated because of a number of problems, some of which are:

a) For practical reasons all the possible sampling designs are not considered when
selecting the desigr- to be used. Some characteristics of the contemplated design are
taken for granted. In the case of field sampling the area of the unite at each stage,
the number of stages and the stratification are fixed prior to optimization. This is
usually based on previous experience. It must be emphasized that thic leads to partial
rather than absolute optimization. There is such an "optimum" design for each set of
predetermined characteristics. In many textbooks and manuals the calculation of
optimization ie restricted to the estimation of the optimum number of sampling units at
each stage based on a simple cost fomulation.

b) A cornpletp estimation o^ the error should take into consideration not only the
sampling err^r but also biasee and measurement errors (see paragraph 24). These latter
are generally difficult if not impossible to determine. In most cases efficiency studies
deal or.iy with sampling: errors and they are valid insofar as the measurement errors are
reduced to a minimum. Elimination or reduction of measurement errors is often more
important than the exact optimization of the sampling design.

c) A third problem occurring in the determination of the optimal sampling design

is related to parameters. A sampling design it optimal for the estimation of a given
parameter (say, gross volume over 60 cm diameter of a commercial species) but not for any
other one (say, p^ross volume over 60 cm diameter of a group of commercial species). When
designing the inventory it is therefore essential to select the most important parameter
and to look for the m^st efficient -leBigri for the estimation of this parameter. Very
often it is difficult to single out the most important parameter from those it is
necessary to estimate. Rven when one species only is very important, as in the case of
Qkoume' in Gabon or Pine in tropical pine forests, there may be various other useful
parameters. In addition to the volume of the exploitable trees of the particular species,
it is important to know precisely, for management purposes, the number of stems in the
lower diameter clasnes. In any caee the inventory officer must avoid a sampling design
based on irrelevant parameters. One such cae is where the selection of "the gross
volume of all species over 10 cm diameter" has been used as a basic parameter for
designing an inventory in a tropical forest where few species are utilized. TMs is
generally not valid but it has been done frequently.

d) In a sampling design, the sampling error on a given parameter depends on,

among other things, the variability of this parameter in the units of the sampled
population. For optimizing the design, one must have a prior estimate of this variability.
This knowledge can be obtained from former inventories, however in many cases there is
no possible source of information and a small preliminary sampling is needed to have an
estimate of the variances and coefficients of variation which are included in the
sampling error formula. This small pilot inventory is also necessary for estimating
cost figures which will be used for the optimization of the sampling design.
 34 1 ? Mathematical formulation of an optimization

Let ue erpress the precision of a given sampling by the standard error of the
estimate, SE, of the parameter which has been selected aa the moat important, and the cost
of it by C. SE and C are functions of various characteristics of the design, such
as the size of a sanpling unit and the number of sampling units or the sampling intensity
in the case of a simple random sampling. Let us call these characteristics x.. f Xp x

We have: SE SE ... x )
p
C C (x 1f x 2 , ... x )
p

Let us suppose that the total cost of the inventory is already fixed and equal to Co.
Ue will have in this case to minimize the standard error SE. So we must have:

SE - minimum

with C ( C

It is demonstrated that the suitable values f

(x.K, 2>0
' (
Vo
characteristics x ,
x ol , x will be determined by resolving the following

system ol p + 1
equations with the p + 1 unknowns x

<

C(x 1f . . . x ) - Co

and etandirig for the partial derivatives of SE and C with respect to the
characteristic x .

Let us suppose that a given precision of the estimate of the parameter at a given
probability level IB wanted. In this case the standard error will be fixed and
equal to a given value (SE) In this oase we have to minimize the cost G and
this is expressed by:

minimum

with ... x ) (SB), (1)

p

The suitable values wil1 be *** X determined by resolving the sane

(X'^Q (^'p^n *P^O ^

system of equations f the last equation of the system being replaced by the equation (1).
It can be noted that the first p equations are the sane and that the solutions
(X.J, ( X ^n *"* ^ T ^ arc ^^ niced tov * ne Siunc (P"l) equations.
 - 39 -

cj Numerical application.

Let us consider all the possible simple random field campling designs (usin^ area
elements as sampling units) for the estimation of a given parameter over a forested
area to be inventoried* Let us assume that we have found empirically from a pilot
inventory and/or former inventories that the standard error SE of the estimate
of this parameter can be expressed approximately by the following relation:

SE =

^ no
s being the area of the sampling unit of a sample
log s being the napierian logarithm of s

n being the number of the sampling unite of the sample

a and b being coefficients

Let us assume that the total cost C of this sampling IB satisfactorily expressed
by the following formula:

C o( + n + ns
fS
</, being the fixed cost independent of the size of the sampling
A being the cost of access to one sampling unit
y being the cost of enumeration per area unit
If the total cost of the inventory is known and equal to C OI the optimal area of a
sampling unit s and the optimal number of sampling units n^ will be the solution
of the following system of equations:

ds ^
T /^ T
9 n On

OC + ^n + ns = Co

Eliminating from the two first equations we have

v
as
w

0n
^
On
*
~>r
os
"

If we replace these derivations by their right expressions in formula (?) and cimplify,'
we find out that the solution S Q (optimal value of the size of the sampling unit) is
such that it fills the following equation*

log - (

s can be determined by a graphical way: it is the s-ordinate of the point where

the representative curve of the function y - log s crosses the straight line
m r $
representative of the function * b
y A '^"""i
 The corresponding optimal value of n, n
Ql
is easily calculated from the relation

n s a GO or n
o o o

34? Optimization of an inventory design. The definition of the optimum inventory

design is identical to that of a sampling design given in paragraph 341. The
problems of optimization of an inventory design are similar to those quoted for a sampling
design but are more difficult to solve because an inventory design often consists of a
series of several sampling designs, e.g. one for the estimation of the areas, another for
the field estimation of the mean number of stems per diameter class and a third for the
assessment of the volumes of trees through volume table*. The moat efficient inventory
design might not be a combination of variouo optimal sampling designs. It can be
imagined that it mi/;ht be better to calculate the appropriate volumec from measurements
taken on ctandinr trees within th<* sampling units rather than build up volume tables from
a selection of sample trees. Thic would eliminate the need for the third sampling design.
In the sane example it night be nore efficient to estimate areas from the field samnle
rather than estimate them through the interpretation of aerial photographs. In most
cases it is too difficult to make an exhaustive efficiency study prior to selecting an
outline for the inventory design. Thi* is often done on the banin of a rough estination
of the general cost and precision involved but also considering such factors as
documentation available - aerial photographs, maps, volume tables, etc.*- the skill of the
workers, the transportation facilities and others. Once the general structure of the
inventory design has been decided, efficiency studies are made for each of the sampling
designs. Though this method is not theoretically the best, it cannot be avoided in
most cases.

4 Classical sampling designs

41 Classification QI'
sampling designs

411 Characteristics of sampling designs. A sampling design is defined by a

combination of characteristics which correspond to the following item*:

411.1 Nature of the units of the population

The same forested area can be considered as a population of either trees, plots (of
the same or different areas), point* or lines. It is very important to specify from the
beginning what it to be considered as a unit of the population, since the ultimate sampling
unit are selected from these units. In particular, if the forested area is cons dered
as a population of units of ^yial size, the unit area is to be specified. When one
wants to characterise th* "' lability of a given parameter, one should always give the
coefficient of varioricr ^sco paragraph 223 of this chapter) with the mention of the area of'
the unit to which this r< *ff .cient refers* Hie variability of two parameters cannot
bs compared if their oof lic.ents of variation refer to two different areas of the units.
:

4112 Use of an auxiliary parameter

Undsr certain circumstances, as in the case of units of different sice, and to

improve the precision of the estimates, the parameters are combined with another one,
oallsd auxiliary variate. The estimation of this auxiliary variate may not be an
objective of the inventory but its use as an intermediate parsmeter is necessary end
 y advantageous. Two main casee nan be considered?

an additional Raroplinp is performed in order to obtain an estimate of the

auxiliary variate: thie IE the cape of the sampling design called double sampling
(or tw

- there is no special (or additional) sampling: in the framework of the normal

sampling design the auxiliary variate is recorded together with the parameters to
be estimated. Thi* IB the case of sampling designs for ratio (or regression)
estimates,
the total of the auxiliary parameter over the whole parameter being
known*

411.3 Number of 3tages

If the sampling is done directly among the units of the population, the sampling is
railed a one-stage sampling.

It may happen that for the sake of convenience, the population is connidered at the
first stage as a population of groups of units; a sampling is then done among these groups
(called primary unite). At the second stage, a sampling of units (secondary units) is
done within each group selected during the first stags* The whole sampling operation is
called a two-stage sampling. This design must not be confuted with cluftttr sampling (see
paragraph

Other intermediate stages may be included, and more generally these samplings are
called multi-stage sampling designs. The size of an intermediate unit will be defined by
the number of units of the next stage it contains.

In forestry, there are many examples of two-stage sampling designs, some of three-
ptage and probably very few with a larger number of stages.

411.4 Other charaoteriitioB

The following characteristics are relative to any stage of a sampling design.

Stratification

In order to reduce the variability of a parameter within the whole population and
consequently the sampling error of the estimate, it is generally most appropriate to divide
the population into more homogeneous (with respect to this parameter) sub- populations or
strata and to make a separate sampling within each stratum. In this case the corresponding
stage of the design is said to be stratified. Stratification is sometimes effected by
taking into consideration an existing and useful sub-division of the population: this is
the case when the population is a forested area already subdivided in geographical natural
units (catchment areas, basins),

The stratification may be done prior to selection of the sample (this is the true
stratification or "stratification a priori") or after selection of the sample (in this
case it is called "stratification a posteriori *).
1

In a two-stage sampling, stratification may occur in the first stage, where group*
of units called "primary units" are distributed among different strata, or in the second
stage, or in both stages.

Random or systematic selection

Sampling theory is applicable in principle, only when sampling is- made at random by
using one of the numerous possible devices. For instance, in a plantation with a constant
spacing of trees, a simple random sampling design may be obtained through the selection of
 ~ 42 -

numbers from a table of random numbers after having numbered all the trees of the
plantation. (1 )

Sometimes, for practical reasons or for reasons directly related to the estimation
of the parameters (for the mapping of forest types in the case of some forest inventories)
it is more convenient to use systematic sampling at one or all the stages of the sampling
design. If we have to inventory a line of trees (units), a systematic sampling will
consist of the selection of one tree every p trees (the sampling intensity being f / );
if we have to inventory a forested area we may distribute the sample according to a
grid laid on the map each point of the grid being the centre of a sampling unit.
f

Although sampling theory does not provide users with a completely satisfactory
estimate of the sampling error in systematic sampling designs f it is possible to get
acceptable ones by using certain devices, some of which ere listed in paragraph 423 of
this chapter*

Selection with or without replacement

As already said in paragraph 233, most of the sampling designs are considered "with
replacement" although they are in fact "without replacement" as the same unit is not
considered twice ao a sampling unit (or in other words, as the sampling units in a given
sample are all different units). All sampling designs listed in paragraph 42 are
considered as if they were with replacement. This approximation is more acceptable as
the eanpling fraction (number of sampling unite with respect to total number of units)
gets smaller.

Selection with egual or unequal probabilities

At a given stage of e sampling design, units may be selected with equal or unequal
probabilities (see paragraph 233). In some ssmpling designs units are selected with
probabilities proportional to their sjge (PFS). Thin sometimes proves to be very
efficient. Thin is the case of two-stage forest sampling designs in which the units at
the first stage ("primary units") are selected with probabilities proportional to their
area of forest* (2)

""
%rual or unequal site of the units

Let us ftitume that a rectangular forested area to be inventoried is divided

arbitrarily in equal square blocks of, say, 1 km 2 * Let us consider that the population
is the whole area, including the non-forested parts. In this case the units oan be the
entire blocks and will be of equal size. However, if we consider that the population is
restricted to the forested parts of the area, the units will be the forested part of
each block and will not be the same siee

In both caees these blocks oan be selected with either equa,l or unequal probabilities
(for instance in the case of unequal sizjed blooks probabilities might be proportional to
the sizes). Blocks may be the ultimate units (in which case the sampling design will be

(1) If there are less than 1,000 row* and 1,000 oolwn* in the whole population, each
tree might be numbered by a 6-digit number, 3 digits for the number of its row and
3 digits for the number of its line.

(2) The use of the expression "PPS sampling" for "point sampling" in which trees are
selected at each point by the prism or the Bitterlich Relaskop (i.e. proportionally
to their basal area) is not correct as the sampling units are the points
themselves (or eventually clusters of points) and not the trees.
a one-stage sampling design) or they may be the primary units of a
two-stage sampling
design with secondary unite being for example plots of equal area.

Equal or unequal number of sampling units per unit of the prior stage
(multi-stage sampling designs)

may be a different number of secondary units selected per primary unit in a

two-stage sampling design. When the primary units are of equal size, the sampling
intensity in the second stage (i.e. the ratio of the number of secondary units selected
in a primary unit to the total number of secondary unite per primary unit) may vary from
unit to unit. When the prinary units are of unequal oize, the number of secondary units
selected may be proportional to the size of the primary unite (and in this case the
sampling intensity at the second stage remains constant )

41- Clusters, and record units. Generally the term cluster is used to define a
sampling unit which is in fact a group of smaller units. The cluster is
the statistical unit whereas the smaller ones are only record units. Information is
collected separately in each record unit and is then merged with information from the
other record units to constitute the information related to the sampling unit (cluster )
The record units are not the statistical units. In no case should cluster sampling be
understood as being synonymous with two stage sampling. More generally stages of a
sampling design and clustering are two different concepts which can co-exist as it is
easy to build up a multi-stag* sampling design wherein the sampling units at different
stages are clusters.

4? Classical sampling designs used In forestry

421 Introduction. Below are listed some classical samplin/r deei/TiB used in foreatry.
They are not inventory designs. These will be dealt with in Chapter ? As
we have already seen, a forest inventory design is generally n. combination of sampling
designs.

Ail the sampling designs listed below, like any sampling design, may or nay not use
cluster sampling. As we have seen in the last paragraph, clustering is a sub-division
of the sampling unit into smaller units. It can be applied et any stage of the sampling
and it does not influence the respective formulas.

For each sampling desi/^i, two formulae will be given:

- the expression of the best estimate of the mjafl value of a parameter over the
whole population per sampling unit of the last stage ("ultimate sampling unit")
In case of equal size of these units or per size unit In case of unequal slze|

- the expression of the beet estimate of the variance of this estimate of the mean:
as we have seen in paragraph 3^2 the corresponding sampling error at a given
probability level Is proportional to the square root of the variance provided
that the assumption of normal distribution Is acceptable.

Estimation of a pro portion t Ike proportion of units of a population having a given

characteristic (for instance proportion of units of a forested area belonging to a given
forest type) can be considered as the mean value per unit of a particular parameter, the
only values of which are 1 if the unit has the characteristic and if it has not.
The general formulae apply for this parameter but can be expressed more simply. ifie

corresponding simplified formulas will be given for the most usual designs* For the
other design* they will be easily established by keeping in mind that the oorreaponding
parameter oan have only the two values or 1
 TV Randorr sampling designs (not systematic). The following table includes the
renHon Campling designs considered in this paragraph. The figures refer to
the relevant sub- paragraph ft.
 /1?2.1 Units m area elements or trees (or points or lines for estimation
of proportions )

Sanpling using angl gaugeo, wedge prisms or Spiegel Relaskopo is excluded from
this class of design (for these designs see paragraph 4232 "Point or line sampling
designs").

/|22.11 One stage sampling designs

We will consider only the designs in which unitB are selected with equal probability*
The symbols used are lipted below:

f sampling intensity (or sampling fraction): (f 5)

-
f sampling intensity (or sampling fraction) in the stratum h (f, rp)
h index of a stratum
i index of a unit

(hi index of a unit (i) in a stratum (h))

L total number of otrata (h - 1 to L)

n number of units in the sample

n, number of units in the sample of stratum h
N total nunber of unitB in the population
II, total nunber of unitB in the stratum h
h
p entimate of * proportion P from the sample

R estimate of the nean vp.lue per size unit of the parameter y ("ratio of
the meanc estimate) with a measure of size x as an auxiliary parameter
11

e^ estimated variance of the parameter y in the stratum h

n n,
T ~ 2
* v N

2 jTi^yhi h^
=
B, 3

V exact value of a variance of a given estimate

v estimated value from a sample of a variance of a given estimate
x auxiliary parameter (measure of size) in sampling with ratio estimate
x. value of x in the i unit
th
x^ value of x in the i unit of the stratum h n
nl _ -1
9
x estimate of the mean value per unit of x i x
ili*i

Z estimate of the mean value per unit of x in the stratum ht ^ ~

X -
9
X total value of x over the whole populations

hi -
JL total value of x over the whole stratum X^
N
-
1 T x.
i -
X
I exact mean per unit of x over the whole pojnilationt
 - 46 -

y parameter to be estimated
y1 value of y in the i unit
-th
y,. value of y in the i unit of the stratum h
n

y estimate of the mean value per unit of y in the population! y n i-1

y estimate of the mean of y in ths stratum hi y, - - **,

n

422.111 Units of egual

42?. 11 1.1 Unstratified random sampling (or "unrestricted random sampling" or

"simple random sampling "}
1

Thifl sampling design has already been explained in paragraph 311. Let us recall
the corresponding formulae i

estimate of the mean j>er unit; (1)

estimate of the variance of y:

n-1 (2)

Two remarks i

-\2 2 - -2
ny -

If f is small (for instanoe less than 0*01 1

percent) 1-f is approximately
equal to 1 and we hares

of > troportion P (of unit. (trn obar*ot>i.ti6).

IB thl. <** bwr. 7-4
- 1 or ooordinc io h*th*r th. unit i ha. th glrm
or not.

of th. iroportjon Ft

(3)

where a is the number of units in the ssjiple which hare the given characteristic.
 Sfctimate of the variance of pt

V(P )

If i-rr is small we et: v( P ) . (4')

HJ^EI

Remark ; In the case of the estimation ->T a proportion related to a ch^r^rtori- tic \.kich
be attributed to a point itself (and not to a plot around the point) tht unite may be the
points. This is the case where the estimation of the area of a given foreet type ie
carried out on photographs or mape. In this oaae N ie Infinite and f ia practically
equal to zero and formula (4 ) is the right one*
1

4^2.111..- Stratified random sampling

4 ^.111,'M Stratification prior to *amplin

The sizec
N^ of the various strata ore exactly known and the sampling is made
independently in each stratum*

Estimate of the mean j>er unit

't (5)

Estimate of the variance of y_

Thic formula can be written in different forme and may be simplified ii the sampling

fraction 1' in constant whatever the stratum.

^
422,111.22 Stratification after sampling

The sizes of the various strata oan be determined exaotly or fairly accurately,
N^
but there is only one sampling for the whole population (and not an independent sampling
in eaoh stratum) and the sampling units are classified into otrata after they have been
sampled ftiis is the oase of a foreet inventory in which s

stratum sices are known by photo interpretation)

only one sampling is done for the whole area to be inventoried}

the fact that a unit belong* to a given stratun is not known or not
investigated before data recording in the field.
 - /i 8 -

of the mean per unit (7)

Estimate of the variance of y_

(fl)

n
~
(where n - <2E n, IB the total number of sampling units and f * whatever h IB,
h

as in this case there ia only one sampling arid thun only one nampling fraction),
*"

II the ^ntimated variances B, within the diiferent strata are not too different
h
ami if the sample IP large the second tern of V (y1
) is nmall, 'Hiic estimate IR then
S "t

not very different from the one for stratification prior to stripling given by fcrnula (6).

Formula (3) that there ifl no systematic error in the estimation of N. .

HP Units of unequal size (ratio estimate)

If the statistical units of a forested area to be inventori^i BTP strips or line

plots, they will be in most cases of different size for one or several of the following
reasons :

irregular p.hape of the forested area;

- irregular contour of the different strata within the nroa;
- different sizes of the horizontal projections of the units (the^e
projections being in fact the units) due to different terrain features from
one unit to another.

In such cases a judicious approach is to consider the size (area) of ^ach unit as an
auxiliary parameter and to estimate the mean value of the parameter per size uni't as the
ratio of the parameter itself over this auxiliary parameter (see paragraphs 32?. 6 and 33).

422.112.1 Without stratification

Ultimate of the mean value per size unit: . (9)

("ratio of the means")

Estimate of the variance of H, s

I/ ^
n (10)
-2
 - 49 -

- i
If the sample estimate per sampling unit is x - then (I0)*y be written as
,
follows:

Important remark; The ratio estimate R. is a biased estimate but the bias will be
the more negligible asr

- the number n of sampling units is larger;

- the regression between y and T is better represented by a straight line
through the origin (thie is generally well verified if y is a volume or
number of stems and if x is the area of the unit).0)

422.112.2 With stratification

422.112.21 Stratification prior to saaplin*

If the numbers n, of sampling units in the strata are large enough and if the
ratios R in the various strata are different enough, it is demonstrated that the best
estimates of the ratio and of the variance of the estimate are the following:

Estimate of the mean per unit

("separate ratio estimate )i ,
1s r *i h-1 T =~ (11)
11

Estimate of the variance of R ^

1
(12)

42?. 112. 2? Stratification after sampling

A
If the variances of the R... are approximately the same in all strata, and if the

sample is large enough, formula (1?) may be applied j if not, formula (8) has to be

adapted to the case of this ratio estimate*

(1) In any case we havei bias R

| V
* * *
V(R 1 ) i the exaet variance of R (etiste4 by v( 1
)
<1

! the exact standard error of "x (standard deviation of x

divided by /n ).
 422.12 Two-stage sampling designs

We will consider only the designs in which the same number of secondary units are*
selected per primary unit*

All the estimated means in the following formulae are means per secondary unit.

The symbols used are liated below:

f
1
sampling intensity at the first stage t f
1
5 (equal primary units)

fp sampling intensity at the second stage t

f v (equal primary units and same
sampling intensity at the
second stage in all primary
sampling units)
.
sampling intensity at the second stage in the stratum ki of the primary
fp X1
unit i
h index of a stratum (stratification of primary units)
i index of a primary unit

j index of a secondary unit

(ij index of a secondary unit in a primary unit)

ki index of a stratum (stratification of secondary units) within the primary
unit i
L total number of strata for primary units (h 1 to L)
L. '
total number of strata for secondary units within the primary unit i

(ki 1 to L f
)
1

m number of secondary units in the sample of each primary unit

rtL . number of secondary units in the sample of stratum ki in the primary unit i

M total number of secondary units in one primary unit

R. total number of secondary units in stratum ki in the primary unit i

n number of primary units in the sample

n, number of primary units in the sample of the stratum h

N total number of primary units in the population

N, total number of primary units in the stratum h

p estimate of a proportion P from the whole sample

p, estimate of a proportion P from the sample of primary unit i
2
B estimate of the variance of the parameter y in the stratum ki in the
ki
primary unit i

2
'
ki

x auxiliary parameter (measure of site) in sampling with ratio estimation

x, . value of x in the j secondary unit of the i primary unit
x. estimate of the mean value per secondary unit of x in the primary unit it

X exact mean value per secondary unit of x over the whole population

y parameter to be estimated
y. . value of y in the j secondary unit of the i primary unit

y, . . value of y in the j secondary unit of the i primary unit of the stratum h

y, . . value of y in the j secondary unit of the stratum ki in the i

J
primary unit
yi ectimate of the mean value per secondary unit of y in the primary unit i:

y. . mean value per secondary unit of

estimate of the me y in stratum ki
(primary unit i):

ki

y ~ estimate of the mean value per secondary unit of y in primary unit i

1
(with stratification within primary units):

yNSi " ki-1 TT yki

y, estimate of the mean value per secondary unit of y in stratum hi

yh " _L
rm
t
i-1 j^

estimate of the mean value per oondmry unit of y over the primary units
of stratum h (with stratification within the primary units )t

" .
n i-1 yNSi
 422.121 Primary units of equal size

In the designs listed below f the primary units are selected with .equal probabilities.

422.121.1 Secondary units of e<jual size

422,121.11 No stratification of primary units

422.121.111 No stratification of secondary units within primary unite.

Bitiate of the mean value per secondary unit t

iti y j
y - 03)
mn

Estimate of the variance of y;

nm 0/0

flitimation of a proportion

In this case the values of the parameter y in the secondary unite are equal to
or 1
(y *
. .1 or 0). The above formulae may be (simplified.
J

Estimate of the proportion P

(mean value of y per secondary unit):

(where p^ is the estimate of the proportion for the primary unit i: pi , a., being
the number of units in the sample of the primary unit i for which y. .
- 1).

Estimate of the variance of p

v(p) - (14')
n-1 nm

If the aeoondary units are points, then the number M of secondary unit a per primary
unit is infinite and f - o

422,121,112 Stratification of secondary units within primary units (prior to

of the mean value per aeoondary unit

ki (15)
 Estimate of the variance of y._ _ _
f

n-1 ki (16)

Fomvuln (16) is similar to (1/1)and reduces to this latter when there IB no stratification
within th* primary units. Th* nature and the number of the strata may not be the ame in
the various primary unitn.

42?.1?1.1P Stratification of primary unite (prior to sampling).

4??.1?1.1?1 No fltratification of secondary units within primary unite.

Qetimate of the mean value per eeoondary unitt

(17)

Estimate of the variance of y

S

L R
(18)

v(y ) being obtained by formula (14) wherein the primary unite taken into consideration
are those which belong to stratum h.

121.1?? Stratification of secondary units within unite

(prior to sampling)

Let us refer the index h to the stratification of the primary units and the index k
to the stratification of the secondary unite within the primary units. These two
stratifications must not overlap. In a forestry sampling design the stratification of
primary units may be a geographical stratification (by catchment area) or a broad
vegetation classification, while the stratification of the secondary units within each
primary unit may be a stratification by density and height of the dominant trees
("condition classes
11
).

Estimate of the mean value per secondary unitt

ysa -
h*i r (19)

stratum h.
y._ being calculated by formula (15) applied to the units of the
^h
 h<* variance of y r

(20)

v("-vr by formula (16) applied to the unite of stratum h.

S **rondary unite of unequal size

thiR '"as?, mtio estimation is used as in onestage sampling designs f the cize
1
>

>' a ".*<: unit bMiic the axixilipry parameter.

ori'lary ^e will coneider only the cae where
here ir no triti irHtion either of primary urits or of secondary unite
P. f

Us t mate of value per elze unit

("ratio of t

(21)

01 t,h e of R

(22)
2 2
2 cir 4>
!

"
4 2 . 1 r
Primary units of unequal size

will noncid^r only the case of the primary units being selected with a
V/e

Ity proportional to their aizes, without any prior strati fioetlon of the primary
uiitg or of the eecondary unite and with secondary unite of Lat us reoall
eqyal ig
thRt we ?m8ume that there is the same number of secondary anpling unita per primary unit

of the mean value per secondary unit:

PPS (23)

The formula is the eame as (13) which gives the estimate j for a two stage sampling with
equal primary unltf) selected with equal probabilities (unweighted sample an pr
secondary unit)*
 - 55 -

Estimate of the variance of

(24)

The simplicity of formulae (?3) and (2/l) and the usual efficiency of this design
make it particularly interesting.

4? .? Point (or line) sampling designs

This type of sampling is use^ in th^ field work of v forest inventory. The ultimate
units are points (or lines) and are not characterized by a given area or by a tree. At
each point the treee (trees are not the sampling unito) are selected with a probability
proportional to on? characteristic which is:

- to banal area in "horizontal

a point sampling"
- its diameter in "horizontal line sampling"

the so^iar^ of its height in 'Vertical point sampling"

- its height in Vertical lina sampling".

In other words for each tree there is a corresponding area of plot proportional to thin
characteristic and the bigger this characteristic, the larger the plot. It ip for this
reason that this type of design is sometimes called "polyareal plot sampling".

These decigns - in particular the "horizontal point sampling" - have developed very
fast in the last twenty years in forestry, first in North America (using angle gauges or
prisms), and then in other temperate zones (especially in Europe with the Spiegel
Relaskop). Their use in tropical forests is hampered by limiting practical factors, such
as the opacity of the undergrowth and the various heights of buttresses. In addition,
these designs do not give directly a representative picture of the forest at each point because
the trees are not selected with the same probability: this shortcoming is more serious in
mixed tropical forests where it may be interesting to know the distribution of the species
and diameter classes at each sampling location. However, it has been successfully used
in some casec and be recommended in the tropics in homogeneous stands (pine forestc
may
or plantations).

Except for thic difference in the nature of the sampling units, all the designs
described above are applicable. In particular, if the lines have different lengths, it
is worthwhile using a ratio estimate with the length of the line as an auxiliary parameter.
As in the case of area elements cr trees, clustering may be used with any type of design.

Therefore it is not necessary to resume the designs listed above, and we will only
mention what is to be changed in the formulae.

- Value of the parameter in one point (or in one line)

Let us take the case of a horizontal point sampling design. The point i being the
this unit Is
ultimate upling unit, the value y< of the pftrmeter y Pr area unit in
equal tot
 -56-

where: pi is the total nunber of trees selected in point i

y if the vnlue of the parameter y for the k selected tree of the point i
1
(if y is a nunber of stems, y will be ecjunl to 1 whatever Y is).
1,K
BA. , is the basal area of the k selected tree of the point i
i f
K
2 Q
F is the basal area factor (er^ial to P = 10,000 ^in TJ,
in square metres

(basal are?) per hectare, wherein ic the ^pugo r.n^le).

Similar formulae ar^ applicable to the three other designs in. which the banal arers of the
trees are replaced by the relevant characteristics, dianoter, n^uare of height or height,
and corresponding v^luo or the i>ctor P.

of units - sampli ifc i

Strictly cpeakin, as the yonulation (and/or the ntrat^, and/or the intermediate
unitn) in P forester] area, the sisee of the units are to bo measured in areas. But as a
unit has no defined area (there are PR many areas as valuer; of the characteristic) the
total number of ultimate unito i'\ tho population (or in ?ny division of it) cannot be
defined precisely. Provided that the total nunber of trees relecteH in all the units is
relatively small compared to the total number of treec of the population (or of the
related subdivision of it) t

- the sampling intensity at the ultimate stage will be considered

sufficiently low to be made equal to 0;
- the mmiber of ultimate units at the last stage per intermediate unit
of the former eta^e in eo large thr.t its inverse c?n be made equal to 0.

A very thorough analysis of the point and line sampling principles ir in lf

Forest
^iven
Mensuration' by B. Rusch, Ch. I. Miller pnd Th. M. Beerr. (pare ^5/1-291).
1

4?3 Systematic sampling designs

1 General considerations

By systematic sampling designs we mean all nanplin^ designs r,t one, several or all
stages at which a selection of sampling unite ie made pccordin^j to a systematic pattern,
i.e. by selecting only a first unit at random, the location of the other sanplin^ units
being automatically 'lerfuced fron this first selection.

Any design which includes a systematic selection of the units in not strictly
according to sejnpling theory for the following reasonn:

- Only one unit is selected at random, the other units are not independently
selected (in terms of statistics it is said that each one does not correspond
to a "degree of freedom"); in this case the variance cannot be estimated.
This can be understood also if we compare the whole systematic sample as a
cluster. We have seen that the cluster can only be considered as a sampling
unit (not the constitutive unite) and no variance can be calculated fron the
vmluft of the parameter in only one unit*
 - 57 -

- Onoe the first unit IB selected the other unitp, which do not belong to the
future sample, have a zero-probability of being elected and the other unite
of the sample have a probability of being selected equal to 1, In other
words moat of the units of the population are definitely excluded from the
selection because of the prearranged pattern of thifl systematic design*
This in contrary to a basic principle of the sampling theory.

In addition to these considerations, it must be pointed out that any calculation of

the variance of the mean ic complicated by the fact that there may be dependence of the
valuec of the parameter in couples of neighbouring units. In this case the estimate of
the variance of the estimated nean in not simply related to the variance of the value of
the parameter in one unit (cee paragraph 322.3) and the calculation of an estimated
variance of the nean becomes practically impossible.

All random designs listed in paragraph 422, with the exception of the two-stage
sampling design with unequal probabilities f have one or more corresponding systematic ones:
in particular, several noRsibilities of systematic designs exist, corresponding to each
two-stage random design, whether the systematic selection is made at the first stage, at
the second etage or at both stages.

Amendments to the formulae for random designs

423.21 QBtimation of the means

The estimate of the mean value of a parameter per unit (or per ize unit) in, a
systematic design is, in most cases, given by the earne expression as for the corresponding
random design. So the formulae given in paragraph 422 are generally applicable.

However, caution is necessary in the estimation of the mean value of some parameters.
If there ia a periodic trend in the values of a parameter and if the systematic design has
the same "wave length" f the estimate of the mean might have a rather important bias. This
may happen, for instance, if the topography is roughly a succession of parallel ridges and
valleys arid if the systematic layout of the sampling units is such that units appear mainly
on the ridges (or in the valleys), the mean value of the parameter will be overestimated (or
underestimated ).

In order to avoid ouch troublesome coincidences, one would have to check very
carefully that the distances between sampling units are not espial ( or a multiple) of the
'

_ ;

"wavelength" of any periodic trend in the population.

4232? Estimation of the variances of the estimated means

423.221 "Random para/jet era"

distribution of the value* of a given paraneter in the unite of a population nay

be "at random 11 ! which mean* that there is a very little or zero correlation between the
values of a parameter in two different units whatever the distance between these two units
(the corresponding covarianoes are equal to - see paragraph 322.2). This is the case
of some parameters related to species with a very little occurrence in some mixed tropical
forests (tho distribution of which is comparable to Poisson's distribution).

For such parameters the variance* of the estimated means can be determined by the
formula for the corresponding random design*
 -58-

/)"'% ,""'? All other parameters

Many statisticians have otudied the problem and although there la no completely
f

factory estimation of the variances given by the sampling theory, some calculations
give reasonably reliable estimates. We will give belou some of the more usual ones.
For the sake of simplicity we will limit ourselves to the one-stage staple systematic
design. Extension of this case can be made easily for the stratified and/ or two-stage
systematic designs.

423.222.1 Units of equal size

First method: Stratification with overlapping strata Kith two sampling units

a) Let us suppose we have only one line of sampling units (plots or trees along a line)
or that the sample consists of eo^ial parallel and eciuidistant sampling units
(strips or lines of plot-record units). An estimation of the variance is:

n-1

(24)

where y is the sample nean (estimate of the mean value per unit of the parameter

n is the number of sampling units

y . ,
f y. are the values of the parameter respectively in the (i+1)
and i*k sampling units

f 77 is the sampling intensity (N total nunber of units in th* population)

This formula is established by considering that the whole population is divided in

(n-1 ) ntrata f each containing two neighbouring sampling units and each sampling unit
belonging to tv/o overlapping ntrat?,, with the exception of the first and last
eanpling unite.

b) If we have several parallel and equidistant linos of sampling units it ic worth

considering another stratification of the lines: each line will belong to a strata
the size of which is proportional to the number of sampling units along the line.

The formulae will be*

n N m N
h
..
lh ~ i (25)
h-1 N"

where m is the number of lines

N, is the total number of units in the stratum h (corresDonding to line h)

N is the total number of units in the population (area inventoried)

N
 N 'Hi
J\ may be ostimated by n^
and n are respectively the number of
(v/here
units "lo-'- the line h and the total number of snplinr units:
n
n
and are the th
y ih yh respectively value of y in the i sampling unit of
line h and the estimated mean value of y per unit in line h:

\r
i=1 y ih

In each ctratui the variance v(y h ) crji be estimated by the formula (2/1) where C :i
f

le diffo
nd the differences (y.,+1 - y ) are restricted to the stratum and the estimated variance
of y will be:

Second method - stratification with non-overlapping strata with 4 sampling units

(for two-dimensional samples of areae only with a equare or
rertangnlar pattern).

Instead of considering strata containing two sampling units, we consider strata

Containing four sampling units (two units on a given line and the two corresponding ones
on the next parallel line). If we take no overlapping strata we will have:

where y. is the mean value of the parameter in the stratum j

n 1
is the number of strata: n 1
# -r

the other symbols having the same meaning as above.

423.222.? Units of unequal eize

We will limit ouraelve* to the cae of one line of unagual sampling unit* or of
equidistant parallel gampling unit* of unequal jge (for inatanoe trip of me width
but of unequal length). We will apply the first method of trtifiotion with
overlapping trat* of two unit t nd will coruilder th are* x of a unit a the
auxiliary parameter.
 - 60 -

J&timate of the mean value of the parameter per size (area) unit;

Intimate of the variance of this estimate:

n-1
1 1-f
,
72 2n(n-i; 1
f sys

n-1
- 2R. t
i1 (
v

(where the symbols are similar to the one used in paragraph /' 2.11 ;.
 - 61 -

CHAPTER IV

REMOTE SENSING AND HAPPING FOR ARS&

ESTIMATION IN FOREST INVENTORY
 - 62 -

CHAPTER IV

RfllOTE SENSING AND KAPPIKQ FOR ARHL

IN FOREST INVENTORY

1 Introduction

Most forest inventories aim at providing satisfactory estimates of total values of

parameters of the forests (mainly volumes of wood) over tke whole inventoried area and/or
over parts of it. These total values are obtained through estimation of the
corresponding
area and estimation of the mean values per area unit of these parameters.

Both areas and mean values of the parameters per area unit may be estimated
through
use of aerial photographs (and nape) and/or field measurements and
observations, these
estimations being made by complete census or by sampling.

We will not consider in this edition the case of the estimation of the mean values
per area unit of the parameters made completely or partially through use of aerial
photographs (or any other remote sensing data), since such techniques are mostly restricted
to some fairly uniform temperate forests and plantations and are for the time
being of
little relevance for tropical countries. Difficulty of species identification from
aerial photographs in the tropics, loose correlation between crown characteristics and bole
dimensions, and the impossibility of defect assessment from aerial photographs severely
limits the applicability of photogrammetric measurements in surveys of tropical forests.
However, some literature is quoted in Appendix ITwhich should be consulted in case such
methods appear feasible in the tropics (for instance in plantations).

As indicated by the title, this chapter is limited to the use of remote

sensing
and map data for the estimation of area, although not all area information is obtained
from these data. It IB sometimes partially or even entirely provided by field
measurements and observations.

Conventional aerial photographs have in the paet been the only retnote sensing tool
used in forest inventory, and will certainly remain the most important one for a
long time.
One first part of this chapter will reetrict the study of remote
sensing in forest inventory
to that of conventional aerial photographs, but sub-chapter 6 will be devoted to a
description and present applications to forest inventory of the new remote sensing tools.

We will assume in this chapter that good topographic (or only planimetric) mapping
over the whole area exists at a suitable^ '7 scale, permitting the reduction to a
negligible value of the error on the estimate of the total area to be inventoried. If
this is not the case, and if the inventory to be made is not a broad reconnaissance
survey or does not apply to a relatively small area which can be topographically surveyed
on the ground, the first objective of the inventory should be to have this mapping done
from the existing remote sensing imagery by a cartographic institute
using plotters of the
first order. We will not study this technique nor the topographical
survey methods which
are well described in relevant manuals and textbooks.

In case this topographic mapping is not possible for financial or other reasons it
must be pointed out that the resulting error over the total area will increase the errors
on the estimates of the total volurr.es (or totals of other
parameters).

(1) "Suitable" in relation to the sice of the area to be inventoried and

the intensity of the
inventory work.
 - 63 -

2 Forest and land-use classifications

21 Various kinds of classifications

Given a forested area to be inventoried, results of the inventory usually have to

be given not only for the whole area, but also for parts of it. When dividing the
inventoried area several simultaneous criteria may be used. Broadly speaking they are
the following:

- Criteria of vege tat ion/ environment relationships taking into account

environmental factors such as climate, altitude ana soils. The corresponding
classification does not generally indicate the existing land-use, but is
useful for land and forest management: decisions on whether to maintain the
forest or clear it for agriculture, on location and species for reforestation,
on silvicultural treatments, etc. can be made with greater confidence from
information gathered through this classification.

- Criterion of present land use; this criterion defines the most important
classification as it separates the forests from other land ueee and vegetation
types. In this classification the forest areas are broken down into very
broad and universally accepted classes.

- Criteria of forest management: under this heading we include all the factors
which are of direct relevance for forest management such as:

- ownership and tenure; for instance, if there are publicly and privately
owned forests in the inventoried area it is almost certain that it will
be necessary to give separate results for each type of forest. The same
occurs if there are forests under concessions which need to be separated
from the other forests;

- administration: results may have to be given by administrative units

(districts, counties, regions, departments, provinces and states) if the
inventoried area is spread over several such unite;

- physiography and accessibility: results will have to be given by isolated

forest unit, by watershed, by type of relief, etc.... the corresponding
units having to be managed separately^

- management units: in addition to the foregoing there may be another

management and/or logging classification (sustained yield unit, logging
compartment, etc.). For instance, the classification into unexploited
forests and logged-over forests is of primary importance in many tropical
zones for the preparation of harvesting regulations. Results have to be
given separately for each class.

Criteria for statistical stratification; if in the same management and/or

"present land use" and/or "potential land capability" unit forests are
it
significantly different with regard to the parameter* to be estimated,
is useful to stratify the forests accordingly in order to decrease the
sampling error on the estimates and make the inventory or* efficient.
Stratification will be more efficient if it is done prior to sampling
(* paragraph 411.4 of Chapter III). This stratification ie generally done
through photo interpretation and is based on criteria which can be appraised
from aerial photographs, such as dominant species or heinfrt and density of
the dominant trsss. Classes based on the latter criteria are bften called
"condition classes". Rssults by condition class nay not be useful since
condition classes are in principle only strata used for sampling. But they
ars useful if this stratification is thsn ussd as a basis for management purposes.
 - 64 -

22 Classifications based on vegetation/environment relationships

In most cases this type of classification is not used for assessing the present
land use pattern as this pattern differs generally from the one of the potential or
climax vegetation types which are considered in these classifications* However, they
may be used in forest inventory for a primary broad stratification of the forests to be
inventoried 9 especially when original vegetation has been more or less untouched*

Many systems of classification and mapping based on vegetation/environmental

relationships have been evolved, some on a world basis, others on a more restricted
regional basis, e.g. for Africa, Southeast Asia, and South America. They generally use
various methods of analysis of the complex of factors making up the environment, with
emphasis chiefly on one or more climatic factors, e.g. rainfall, temperature, evapo-
transpiration, etc. For the purpose of the forest inventory, it is necessary for the
expert first to familiarize himself with a national classification system adopted for
local usage, and relate this as far as possible to a suitable (1) regional and (2) world
type classification.

It is not intended in this manual to consider the merits or limitations of the

various world and regional classifications. It is advisable first to find out for each
inventory project what vegetation/climatic maps already exist on a country and regional
basis* Several maps have been published on world and regional classification systems,
and some of the more recent may be mentioned:

1. Vegetation map of Africa, published under the auspices of the "Association

pour 1'itude taxonoraique de la flore d'Afrique tropicale with the
11

assistance of UNESCO, by the Oxford University Press, 1958-

2. Atlas of maps of the "Crop Bcologic Survey in West Africa", by J. Papadakis

under the auspices of FAO in 1966 (which includes also a classification of
climatic tones correlated with vegetation types).

3. A series of maps for individual countries in Southeast Asia, Africa and

Europe, published by Oaussen and his co-workers, e.g. publications of the
Institut Fran9ais de Pondioh^ry covering India, Ceylon and Madagascar
among others*

4* World climatic maps on the systems of Thornthwaite, Swain and KSppen*

5* Climatic vegetation maps for Latin America by Hbldridge.

6t World vegetation map by SchmithOsen at 1/25 f OOO ? 000.

UHB5CO, the World Meteorological Organization (WMO) and FAO have been specially
interested in f and have sponsored, the preparation of world or regional maps on
specialised ecological aspects (including soils ) f e.g* the Crop Ecological Map of
Papadakis referred to above, a study of the "Agroclimatology of the Semi-Arid Areas
South of the Sahara in West Africa", by J. Cochem* and P. Franquin, jointly published
in 1967 by FAO, UNESCO and WMO, and a World Soils Map by FAO to be completed in 1974.
Though MO has not yet specifically recommended any one system for a classification of
vegetation on a world or regional basis, the Holdridge System appears to be of speoial
interest both for its simplicity and easy adoption for forest inventory purposes. A
description of this system is given by Holdridge (1967). IMs sytem has been applied
to the compilation of ecological maps in several countries of Latin America.
 -65-

?3 "Eristin^ land use" classification uged by FAQ inventory operations*

231 An area classification scheme was devised at PAO Headquarters at a Beet ing of
forest inventory experts in September 1967 and PAO inventory projects hay* been
asked to use it (with eventual introduction of sub-divisions tc satisfy individual project
requirements). In the preparation of this classification efforts were made to conform to
the categories and definitions used by PAO in its World Forest Inventory compilation.
ftie elaboration of world, regional and national statistics on forest resources will be nore
easily obtained as more agencies accept this classification.

The step* in the classification and the definitions and explanations required are
ae follows:

Classification steps Definition or explanation, if required

Classify the area into: The basis for this division should be defined and
I. Land area the date given, e.g. according to existing carto-
II* titter area graphy, aerial photography, etc. Mangrove and
coastal palm forests are to be assigned to "land".

Classify "Land Area" into: Consider as forests: all lands with a "forest
A. Forest Area caver" (including natural bamboo and palm); i.e.,
3. Other Wooded Area with trees whose crowns cover more than 20jt of
Non-Pores t Area the area, and not used primarily for purposes
other than forestry. Por the definition of a
tree use the following one given by the
"Terminology of forest science, technology,
practice and products" (edited by P.C. Pord
Robertson and authorized by the joint PAO/IUPRO
Conmiittee on Forestry Bibliography and Terminology )i

tree i "a woody perennial plant typically large

and with a single well-defined stem
carrying a more or less definite crown"
Include: a) Public and private forests;
b; All plantations, including one-
rotation plantation, primarily
used for forestry purposes;
(o) Areas temporarily unstooked as well
as young natural stands and all
plantations established for forestry
purposes, which have not yet reached
a crown density of more than 20jf;
(d) Porest roads and streams and other
small open areas, as well as forest
nurseries, that constitute an
integral part of the forest and
which cannot be readily excluded
by the survey system used.
Exclude:
(a) Isolated tree groups smaller than
0.5 *a;
(b) City parks, private gardens and
pastures }
(o) Areas of windbreak and shelterbelt
trees with an area exceeding 0.5 ha,
but too narrow to be managed as
forests.
 - 66 -

Classification steps Definition of explanation, If required

2. (contd. ) Consider as other wooded **** l*&d with trees

whose crowns oover less than 20}C of the area or
with shrubs or stunted trees covering more than
20 of the area, not primarily used for agri-
cultural or other non-forestry purposes (suoh as
gracing of domestic animals). Also include areas
occupied by trees in lines (along roads y canals and
streams, etc., converted to area by 0*8 ha per
1,000 m) as well as windbreaks and shelterbelts
which are not included in "Forests .
11

Classify the "Forest ire* Tbe differentiation between natural and man-made
11

into: forests should be made on the basis of the

1 . Natural Forests classification given in paragraph 233.
2. Han-Made Forests

Classify "Natural Forests Categories (a) and (b) will be defined by a

11
4.
intoj composition of 80^ or more of the species groups*
(a) Broadleaved species ex- Mixed forests with less than 80)f of (a) or (b)
cluding mangrove forests will be classified as (c). Where bamboo occurs
(b) Coniferous species in the above types as understorey, this should be
(o) Nixed broad leaved and recorded to estimate mixed bamboo area. It may
coniferous species be useful to subdivide (a) into mixed broad leaved
Pure bamboo areas forests and pure (or almost pure) broadleaved
e) Mangrove forests in certain tropical areas where pure or
f ) Coastal and riverine
id) almost pure stands of a broadleaved species occur
palm fonet (case of Oilbertiodendron dewevrei in Congo basin).
(s) Temporarily unstocked
areas
Shifting cultivation areas already re-stocked
with forest vegetation should be classified as
sub-divisions of categories (a), (b) or (c).

Classify "Man-Made Forests

11

into those applicable divisions

shown under Step 4

Classify "Other Wooded Area" Savannahs should be considered as areas of

into: scattered trees or scrub over graminaceous or
1. Savannah, open woodlands herbaceous layer. Crown oover of the woody
2. Heath, stunted and scrub vegetation
can exceed 20jf. Categories (2) and
forest (3) are defined under Step 2.
3. Trees in lines, wind-
breaks and shelterbelts
4- Other areas

Classify the "Non-Forest Area* (a) includes shifting cultivation areas wnere
1

intot land is under preparation for agricultural

1* Agricultural land cropping, or is planted with agricultural crop
a. Crops and improved or is not renitooked yet with forest vegetation.
pastures
b. Plantations Plantations include orchards, rubber, palms, etc.
 Classification ftepe Definition or explanation, If required

(contd. )

2. Other lands
a. Barren land rock, sand, ice, etc.
b. Natural range lands E.g. prairies, pampas, steppes. If a scattered
and tfraPslandB layer of woody vegetation exists, the area should
be classified in "other wooded areas *
11

c. Swamps This includes swampy area* without a tree cover.

d. Heath without trees E.g. tundra in northern zones.
e. Urban, industrial Includes rights of way for roads, railroads,
and communication areas power lines , etc.
f. Other areas

23? FAO'o proposed classification is, thus, the following:

Gristing Land-Use and Forest Classes

I. Total Land Area

A. Forest area

1. Natural forests
a. Broadleaved excluding mangroves
b. Coniferous
c. Mixed broadleaved and coniferous
d. Pure bamboo
e. Mangrove
f. Coastal and riverine palms
g. Temporarily unstocked

2. Man-made fdrests

(those above divisions of a. to g. which are applicable)

B. Other wooded area

1. Savannah, open woodlands

2. Heath, stunted and ecrub f crest
3* Tree* in line*, windbreaks and shelterbelta
4. Other areas

C. Mon- format area

1 . Agricultural land
a. Crop* and lmpcrov*d paaturee
b. Plantations

2. Other land*
a. Barren
b. Natural rang and grassland*

d. Hath, tundra
e. Urban, industrial and oowwunioaticn
f. Other
 - 68 -

? 33 Be fin i tjon and interpretation of man-^ade forests

The following teirt is issued from the note "Actual and potential role of man-made
forosts in the changing world pattern of wood consumption which was delivered by the
1 11

Secretariat of the "World Symposium on man-made forests and their industrial importance"
(Canberra, 14-?5 April

"The phrase sounds simple enough but has caused difficulty in definition and
differences in interpretation. In fact, certain of the natural distinctions
between types are blurred and some degree of arbitrary definition is needed. Any
final, authoritative definition must await the findings of the current Multilingual
Forestry Terminology Project, which is working under the guidance of the joint
FAO/IUFRO Committee on Bibliography and Terminology and with the comprehensive
support of the U.S. Forest Service, the Department of Forestry in Canada, and the
Society of American Foresters. Meanwhile, some guidance is available from the
definitions adopted by the second session of the European Forestry Commission's
Working Party on Afforestation and Reforestation (1953) f as amended by its third
session (1954), as well as existing terminologies such as "British Commonwealth
Forest Terminology, Part I" and the Society of American Foresters' "Forestry
Terminology".

It seeas simplest to equate the definition of a man-fnade forest with that

given for a plantation in the BC Forest Terminology, i.e.: "A forest crop raised
artificially, either by sowing or planting". This could be interpreted to include
all forms of artificial regeneration but no natural regeneration. "To regenerate"
in Bi^lish is normally defined as "to cause to be born again, to re-create", which
implies the renewal of something pre-existing rather than its replacement by
something different. In this sense, a forest formed by artificial regeneration can
be said to be re made by man rather than made by man.

The different types of forest, according to their means of origin, are:

1. Those established artlfically by afforestation on land which previously did

not carry forest. This is the most clear-cut example of a man-made forest
and invariably involves the extension of the area of the forest. A clear
definition of the period of time for which the land previously carried no
forest is needed. "Within living memory" is suitable for areas where there
are no records, but within 50 years" is suggested as an alternative for
fl

areas where records exist.

2. Those established artificially by reforestation on land which carried forest

within the previous 50 years or within living memory, and involving the
replacement of the previous crop by a new and essentially different crop.
The change most frequently involved is species conversion, but the use of
seed known to be genetically different from the previous crop, e.g. from seed
orcharun consisting of superior genotypes demonstrated by progeny trials,
would also qualify. Inasmuch as the forest established artifically by man
is essentially different from its predecessor, this too is a clear-cut example
of a mao-made forest, though it does not involve any change in forest areas.
The term "reforestation", it is suggested, should be confined to this type,
to distinguish it from the following.

3. Those -etalrlished by artifioal regeneration on land which carried forest

within the previous 50 years or within living memory, and involving the
of what is essentially the same crop aa before. Inasmuch as the
new OT-^P is essentially the same as its predecessor, this is forest re-made.
rather than made, by man.
 - 69 -

*U Those established by natural regeneration, with deliberate silvi cultural

assistance from man* In the past, such assistance has sometimes cost more
in time, effort and money than certain of the cheaper forme of artifical
regeneration. Nevertheless, inasmuch as the source of seed or vegetative
reproduction ie natural, it seems logical to consider this as natural (but
man assisted ) forest.

% Those eetablinhed by natural regeneration without deliberate assistance

from man. They would include so-called "Virgin Forests , as well as those
11

regenerated by wholly natural means. They are the most clear cut examples
of a Natural Forest.

The definition included in the questionnaire distributed to countries

attempted to draw the line between number (2) and (3) of the above types, in order
to include as man-made forests all those which involved the creation of something
essentially new but to exclude those which are formed by renewal of the same type
of forest as before. This seemed a logical distinction and was in line with that
made by the European Forestry Commission's Working Party on Afforestation and
Reforestation, when it defined artificial regeneration as: "Restoration of forest
cover by planting or sowing in the normal course of management", and reforestation
as: "Restoration of forest cover by planting or sowing f when it has not been
possible to effect this restoration in the normal course of management". Some
difficult; "H, hc-wever, have arisen in interpretation. Borderline cases may arise
in whi<~h it in difficult to determine whether the specific composition of the new
forest is or is not essentially the same as that of the previous crop, or whether
the management methods in use are normal or not. It seems preferable, therefore,
to draw the line between 3 and 4 and to include within the term "man-made" all
forme of artificial regeneration. This agrees exactly with the existing simple
definition of a "Plantation" in the British Commonwealth Forestry Terminology -
"A forest crop raised artificially either by sowing or planting."

Mixed regeneration systems. - Difficulty ariees when both natural and

artifical regeneration are carried out in the same area. In such cases it is
proposed that the deciding factor should be the intended composition of the final
crop. If over 50^ of the intended final crop has been regenerated artificially,
the forest should be considered ae man-made.

bhape> - "Forest implies width ac well ae length and can scarcely be applied
to row plantations. Likewise "forest crops" implies that a high proportion of the
trees are growing in competition with each other in the "crop" rather than with
other forms of vegetation outside it and are thus capable of forming a true forest
environment. Row plantations, avenues, etc., in which a high proportion of the
treep are subject to edge effect, do not conform to this description. Though
unqueotionably man-fnade, they are not strictly forests. Wide shelterbelte of a
kilometre or so wide, like the "green belt" at Khartoum, on the other hand, equally
rlelinitely are. It in thought that a 100 m width should be the minimum to
constitute a forest. In practice, the importance of row plantations and shelter-
bolts in many countries often makes it essential that they be considered together
with more orthodox shapes of man-fnade forests, as will be done in the present
nympoeium, but they ehould be mentioned explicitly. It should be noted that the
European Forestry Commission s Working Party on Afforestation and Reforestation
1

defined "Plantations outside the Forest" as "Row plantations (road-side planting,

windbreaks, etc.) and plantations in stands associated with a permanent
a /spri cultural revenue on the same Bite".

Stocking. - "r'or^st" implies a closed canopy, at least when the trees are
old enough to form one, and hence a certain minimum stocking. This needs to be
specified strictly in order to avoid the fallacious inflation of figures for areas
"at forested" by the inclusion of "plantations" only 1o stocked and largely
 incapable of making full use of the productive capacity of the site. It is
proposed that for young crops not yet thinned, full stocking tfhould Mean a minimum
of 1,000 stem/ha or 75$ of the trees planted, whichever ie the less, with a
reasonably uniform distribution. Plantations with 25-75# survival or 300 to 1 f OOO
stens/ha should be considered as partially stocked and those with less than 2$%
survival or less than 300 stems/ha as poorly stocked* The latter should, in many
cases, be considered for writing off and complete replanting*

Naturalisation, - Plantations of exotics are, ipso facto, marwade during the

first rotation* If subsequent rotations are regenerated naturally, it is debatable
whether the forests so formed should be called natural on account of their method
of regeneration or man-made because they could never exist had it not been for man's
active intervention through the initial introduction. In such oases it is necessary
to have recourse to a purely arbitrary definition; it is suggested that naturally
regenerated crope of exotics should be considered as "man-made forests up to 2^0
91

years from the date of their original introduction into that area, but that after
250 years the species should be considered as naturalized, when only artificially
raised crops could be considered as man-made.

Agricultural v. forestry crops. - The logic of the distinction between

agricultural tree crops and forestry tree crops is often obscure. There seems no
good reason, for example, why plantations of rubber trees are thought of as an
agricultural crop, while plantations of tan-bark acacia trees are classed as a forest
crop. It is pointless to try to change distinctions which are now generally
accepted by tradition, but it is important to ensure either conformity between
countries in the species included in man-made forests for which area figures are
cited, and those excluded as being "agricultural crops", or at least a knowledge
of the differences. As an example, Ivory Coast has included over 5,000 hectares
of JLnacardiura in its man-made forests, whereas in other countries this may be
considered an agricultural crop."

3 Interpretation of conventional aerial photographs in forest inventory

31 Introduction

311 Area estimation with or without forest mapping.. The use of all the
classifications listed in paragraph 21 leads to a distribution of the total
area in sub-divisions or strata. One objective of a forest inventory is to get
satisfactory stimates of these areas (or of the proportions of these areas with respect
to the total area). In many oases an additional objective of a forest inventory is to
know the exact location of these areas by a delineation on a map.

312
The classifications listed above are of two sorts:

(a) Some existing classifications are already given and cannot be avoided or
amended by the inventory expert. TMs is the case, for instance, of the ownership and
administrative classifications. Estimation and mapping of the areas of the corresponding
classes do not generally require interpretation of aerial photographs: they consist
mainly in transferring onto the available topographic (or planimetrio) maps information
obtained from existing documentation (such as laws creating forest reserves, concession
agreements, eto.), and in measuring areas on these maps. Aerial photographs may be used
occasionally to add to the maps some missing details referred to in the relevant
documentation (planimetric or topographic features such as a small river, the ridge of a
range of hills, etc.).
 (b) 3ome other classifications are set up partly or entirely by the
Inventory
officer, e.g. classifications of present land-use, of accessibility, of "condition
classes". But the inventory officer must
try as far as possible to fit his olaJiifjoations
with existing and well-accepted ones, so as to allow for comparison and addition of
results given by different inventories designed by different foresters. Unfortunately
this is not very often the case and too many inventory operations develop their own
classifications, disregarding the existing ones. If former classifications are not
deemed satisfactory one should first try to refine or to condense them in order that
classes (or groups of classes) of his own classification are compatible with classes (or
groups of classes] of an exiting one. Only if this attempt proves unsuccessful will a
new classification be made.

Estimation and delineation of the areas corresponding to these classifications are

made generally through interpretation of aerial photographs.

32 Some information on aerial photographs and aerial coverages

(Many text books and manuals exist on aerial photography and photointerpretation.
Some of them are listed in the bibliography, and it is not intended to review or even to
summarize in the following paragraphs all relevant information, but more simply to point
out the most important elements for inventory officers using conventional aerial
photography. )

321 Characteristics of aerial photographs

321.1 Scale

321.11 Definition

The scale (or representative fraction) is the relationship between a distance on

the photographs and the corresponding distance on the ground expressed as a fraction
(1/25,000; or ae a ratio (1:<?5,000): 1 cm on aerial photograph is equivalent to
- ?50 m on ground.
250,000 cm
. ., focal length of the camera f
T . . . .
"
It is equal to the ratio: H
a l ti tude of the camera above the ground

H is exactly the distance between the lens of the camera and its vertical
projection on the ground*

If the terrain is not completely flat the scale is not uniform all over a given
photograph. This is particularly important when the scale is relatively large
differences
(relatively low average altitude) and the terrain is mountainous (significant
between altitudes of various points of the terrain).

3?1.1? Suitable soalee for forest inventory (scale requirements for forest
j nventory }

A classification nf scales of aerial photographs can be the following:

1/100 to 1/3,000 very large-scale aerial photographs

1/3,000 to 1/10,000 large-scale aerial photographs

1/10,000 to 1/?5,000 medium-scale aerial photographs

1/P5 t OOO and smaller

small-scale aerial photographs
 The large or very large aerial photographs are used in forest inventories in which
estimation of the otand parameters is carried out mainly through measurements on the
photographs (photogrammetry). This is the case in some temperate forests with a limited
number of recognizable species, parameters measured being the height of dominant and
co-dominant trees, the density of the crown cover, the diameters of the individual crowns,
etc. One of the problems to be solved is the precise estimation of the elevation of the
camera above the ground (useful in its turn for obtaining a precise estimate of the scale
of the photographs).

The medium and small scales are generally used only for stratifying, mapping and
estimating the forest areas in forest inventories where all or most of the parameters are
estimated through field work. The most suitable scale in each forest inventory depends
much on the refinement of the stratification to be made and on the criteria used for this
stratification. For instance, if individual species must be recognized, a medium scale
may be necessary. But if the stratification does not imply the recognition of individual
species, a smaller scale may suffice. This happens very often in mixed tropical forests
in which identification of individual species is generally not feasible at present, at
least at scales smaller than 1/10,000 and therefore is not used as a criterion for
stratification.

In many tropical countries the only photographs available have a small scale since
aerial surveys are primarily designed for topographic mapping at a scale generally not
larger than 1 /SO, 000.

The following table roughly summarizes the use of the various scales of aerial
photographs for forest inventory:

Scale Uses

Very large scale - Photogrammetric measurements (stand

1/200 1/3,000
- parameters estimation) or refined
Large scale 1/3,000 1/10,000 Gratification baaed on quantitatively
measured criteria

Medium scale - Broader stratification based on

1/10,000 1/25,000
qualitative criteria including species
occurrence or on estimated quantitative
criteria (such as crown density or
height of dominant trees)

Small scale < 1/25,000 Broad stratification based on qualitative

criteria excluding in most cases species
occurrence and on broad classes of
quantitative criteria

321.? Tfrpe of emulsion

Conventional aerial photographs are images of the reflexion by the objects of the
electromagnetic radiations of the visible spectrum (O^/* - 0.8/^ ) with the possible
addition of those of the near infra-red (up to 1 JUL ), In the latter case the photographs
are callsd infra-red photographs.
 32 1 . 2 1
(Conventional photographs witnout the near infra-red

Panchromatic black and white photographs are the moot common and moet used up to now,
In fact most of the oxtenfuve aerial nurveys for topographic mapping use panchromatic
films. They are also the ones which most people are more accustomed to. The spectral
sensitivity of moet of the serial mnchromatir films ranges between 0.36/*- and 0.72^
In many cases they are uped with a minus blue filter to cut off blue light below G.5yU
Resolution (in terms of lines per millimetre recorded by the film) f speed and grain vary
considerably from one film to another,

Colour aerial photo^raphn ere much lees oomnon and are generally used for special
purposes on limi ted area*. They are now only a little more expensive than the corresponding
panchromatic photographtu It io said that "the human eye can separate more than 100 times
more colour combinations (huep, values and chroma* ) than gray scale values (ratio of 20 f OOO
to 200)". Species identification as well as detection of diseased stands is easier on
colour aerial photographs than on panchromatic ones so that a more refined stratification
is generally feasible*.

321.22 "Infra-red" photographs

The sensitivity of infra-red black and white films extends to 0.9yU-or LO^U .
Filters generally cut out the blue and blue-green wavelengths, so that the range of
sensitivity is approximately from 0.6,U to 0.9^ . The main differences for forestry
interpretation between the black and white film and the infra-red one are the following:

- easier distinction on infra-red photographs between angiosperms (broadleaved

species) and gymnosperms (coniferous);

- possible distinction between healthy and diseased broadleaved trees on

infra-red photographs;

- easy differentiation on infra-red photographs between water areas and land

areas, and bntweon wt arid dry soils: this is particularly useful for the
delineation between forest on dry soils arid swamp forests;

- better individual izat ion of the trees on the infra-red photographs than on the
panchromatic onee consequently it is easier to delineate the stand* on
;

panchroraatic photograph*. This is especially true on large scales.

The colour infra-red films ("false-col our" film or Russian "spectro zonal" film)
allow for the reproduction in colour of the same spectral range as the infra-red black
and white film. It combines the advantages of the colour reproduction (large number
of combination of hues, values and chroma*) and these of sensing the near infra-red
(detection of humidity and of atreus of vegetation). Identification of species, and
more generally refined et ratifications baaed on soil moisture, stand composition and
vegetation health in eam*r or, thir. type of film than on the three former ones.

Colour additive viewing maxea possible a large number of different falsecolour

pictures from the same set of one-band photographs (see paragraph 613 of this chapter).

321.3 Printing

Prints (on opaque or transparent mediums) are made from negatives by contact
printing or automatic dodging printers. The latter is more economic and "haa the
/^
advantage of accommodatl n# on one print the range of densities occurring on the negative".

(1) from J.A. Howard, 'Serial photo-ecology", pages 36-37.

"Glossy prints iri th A high contrast have the advantage of a greater density rang* but matt
prints can easily be written on and the reflected glare from the surface is not so trying
to the eyes. Nowadays a semi-matt or semi-glossy or gloe*y~fnatt surface is generally
preferred as a compromise between the two types'1 * (1) Die possible defects of aerial
printc are numerous and provision must be made in the contracts of aerial surveying in
order to avoid them as much ae possible* They are listed in the book referred to in ()):
blurred areas, fingerprints, abrasion streaks and scratches, air belts, fog, streaka,
irregular white spots, irregular dark spots, flat prints, excessive contrast (see below),
density too high or too low, blisters, brown spots, bellow stain, fading, double image,
drying marks, pressure-bar marks, fork-like or finger patterns y dark streaks , small over-
exposed circular zone,

321.4 Basic image quality factors

The interpretation of the aerial photographs depends on the ability of the

interpreter, on the equipment used and also of oourse on the qualities of the photographs*
I'hene qualities can be estimated from the following three main factors i

contrast (or colour contrast for colour photography) which oan be defined as
"the actual difference in photographic tone or brightness between a particular
feature that is to be interpreted and the background against which it is imaged";

"sharpness" which corresponds to "the abruptness with which a change appears to

occur on the photograph": the sharper the photograph, the easier the inter-
pretation;

"stereoscopic parallax" which means "the difference in the apparent position of

one feature - such as the top of a tree - with respect to another feature - such
afi the base of that tree - caused by a shift in the point of observation". It
is measured by the parallax difference Dp, and is related to the difference in
elevation h of these two features by the following formula:

where H is the height of the camera above the ground and P is the "air base",
vis, the horizontal distance between the two points of observation; Dp being
expressed as the sum of the projections parallel to the flight line on each
photograph of the distances between the two features. This formula is used
in photogrametry for measurement on the photographs of such things as tree
height* A measuring instrument based on this formula is the parallax bar; a
simpler device is the parallax wedge*

Overlaps

Along the same flight line, the areas covered by two successive photographs ar*
overlapping* This overlap can be expressed as a percentage of the area covered by one
photograph and is called "forward lap" (or endlap)* The "side lap" corresponds to the
overlap between photographs in adjacent parallel flights*

Forward lap is usually between *)$% to 65^. and side lap is more variable, from
to 4Q< or nore. Both overlaps are necessary for good stsrtosooplo interpretation and
plotting for mapping of the surveyed area*

(1) from J.A. Howard, "Aerial photo-ecology", pe$*s 36-37.

 - 75 ~

The number N of photographs rorrBpondin$ to the aerial oorvraf* of a (ir*a ara 8

ie approximately equal to:

where e is the average eoale of the aerial survey

1 IB the aide of an aerial photograph

R is the forward lap (usually 0.55<R 1 <0.65)

1

R ie the side lap (uaually R >0.10)

2

(if S is expressed in ha, 1 must be expressed in hft)

For inotance, let UB suppose that:

S - 1,000,000 ha

3 cm - 0,0023 hjn (Q inches)

R - 0.60
I

R - 0.20
?

WH will have:

0.0023 z 0.40 T O.o

CiraphB oan be constructed on which it is pottible to read N for a (Iven eoale 9

assuming given overlaps and aiz-e of the photograph*.

Due to significant variationn of ecalee in mountainous area* it ie advisable to

request larger endlaps (up to 8<#) and eidelaps (up to 40jt)*

322^ Flight lirett

Plight lines should be theoretically parallel and e<juidistant Due to man/ factors
rf^h ae crabbing and drift of the flight, tilt and tip of the aircraft f the overlaps are
not ronatant,

Flight indices on stable transparent mediupi, where the flight lines are drawn with
the indication of the centre of the successive photographs, are very useful and ust be
requested in any aerial survey contract*

W.3 Tilt and rectification

Tilt and tip of the aircraft result in the osjiexa tilt which is equal to the
inclination of oajnera axis with respect to the vertical, ttms the plane of the photograph
is not exactly horisontal. Bie tilte4 photographs oan be reotiflfd. i.e. projects* onto
a horizontal reference plane, the anfle between the photograph plans and the horisontal
being det Drained by ground reoonnal seance ur frosi filgfo'i d^ta.
 - 76-

322.4 Other specifications

Annex I refers to other specifications of aerial surveys such as junction of stripe

(important when the aerial coverage if obtained from several flights or when it is to be
uped in combination with other surveys covering neighbouring areas ) f cloud cover (which
must be less than a given percentage), time of day - shadows must not reduce the value of
the photographs for interpretation.

323 Some problems related to aerial surveying. In many forest inventories it is

desirable to get new aerial photographs because they are missing on a part or
on the whole area to be inventoried, or because they are too old (in case there have been
rapid changes in land use since the date of the former coverage) or because they do not have
the suitable scale or characteristics. If money is available and if the inventory design
includes the use of aerial photographs f aerial surveying will have to be contemplated.

In addition to the technical specifications whioh are listed in Annex I ,

tome other
points have to be considered when preparing a contract of aerial surveying,

(a) Government restrictions:

Many governments exercise varying degrees of restriction over the execution of

aerial photography. It is necessary to investigate whether the aerial surveying is
permitted over the inventoried area (in particular by foreign contractors), if the
processing is permitted outside the country (when the local processing facilities do not
meet with the technical specifications), etc.

(b) Preliminary information to be collected:

In order to enable an invitation to bid to be issued, investigation has to be made

on such it ens as:

-
flying season, number of likely photographio days and general meteorological
information;
-
topography, terrain and altitudec in area to be photographed;
-
existing documentation (mape) and groirid control;
-
location of airports with servicing facilities
-
permits required, etc,..

(c) Cost and payment:

The cost of an aerial survey is not only determined by the size of the area to oe
inventoried for given scale and characteristics of the photographs. Positioning
(mobilization and demobilization) and stand-by may have an important bearing, especially
in tropical countries where the cloud cover is often a serious inconvenience. The basis
for payment may be photography only; photography plus positioning; photography, positioning
and standby; photography, positioning, standby and flying hours, or various combinations.
It is a matter of weighing the various components, when bids are received, in an attempt
to ensure that photography will be obtained at a reasonable price.
 - 77 -

The example below shows the range of costs per item according to the relative
importance given to them by the tenderers. It is related to a survey made in 1972
various scales within an area of 5,000 km^ in a tropical country*

Bids (in US S)

1. Mobili7,a1.ion 1,200 to 6,150

2. Demobilization 1.200 to 6.150
Sub-total (positioning) 2 f 400 to 12,300

3. Rate per km'" for 1/15,000 panchromatic photography 4.73 to 7.98

->

4. Rate per km' for 1/30,000 infrared photography 3.42 to 6.77

P
5* Rate per km for 1/5,000 panchromatic photography 24.47 to 37-31
9
6. Rate per km" for semi-fnatt contact prints from 3 above
first set 0.17 to 0.75
second set 0.13 to 0.35
subsequent sete up to 6 0.13 to 0.30
2
7. Rate per km for semi-matt contact prints from 4 above
first set 0.09 to 0.26
b) second set 0.07 to 0.15
c)
!a) subsequent sets up to 6 0.05 to 0.15
n
8. Rate per kj/ for semi-matt contact prints from 5 above
a) first set 3.00 to 5.71
b) second set 1.60 to 3.00
c) subsequent seta up to 6 1.53 to 3.00

9. Rate for map photo indices of the photography in 3

above (indication of the indices on existing
1/50,000 topographic maps) 0.10 to 26.64

10. Rate for map photo indices of the photography in

4 above 0.10 to 1.21

11. Rate for map photo indices in the photography in

) above 1.50 to 61.35

12. Rate per km" for the production of an uncontrolled

mosaic at a scale of 1/15,000 from the photography
in 3 above including a screen positive of each set 1.41 to 3.66

324 Mosaics. It is sometimes useful to use an assemblage of the aerial

inventoried area or of
photographs in order to get a clear picture of the
This assemblage may be of unoorreoted prints, the detail of which
parts of it. "composed
matched from to print without ground control or other orientation" and the
being print
mosaic Is said to be uncontrolled. A controlled mosaic is made up of prints which have
been rectified and ratiood (in order that all the photographs have the same average scale).
Semi -con trolled mosaics are made up of corrected or unoorrected photographs arranged
control".
through "a common basis of orientation other than ground
for lens aberrations
Controlled mosaics are not yet comparable to a map as variations
Ifce completely correct ed mosaics
and displacement due to topography are not corrected.
614 of this chapter).
are the orthophoto^nosai cs called also "orthophotomaps" (see paragraph
terrain controlled mosaics can be used as planimetric
However, in flat or gently undulating
maps.
 "*
3

in forest inventory consists in the identification on the aerial

photographs and eventually i r J
f
h delineation of the different clause* correspond ing to
th<* classifications adopted.

Some clauses within the inventoried area are relatively easy to identify and
delineate, aa they are defined by geographical limit*, such as classes corresponding to
ownership, tenure, administration or physiography* Die nost important and most
interesting part of the photointarpretation work in forest inventory is related to the
identification (and eventually delineation) of the various olassss of land use t vagetation,
forest type and accessibility,

331 Qualities of flood photointerpretation* A good photoint*rpretation must be aa

objective ac possible. Although the analysis of the aerial photographs is
made through direct observation and interpretation by a human being, it should always be
baaed on a set of precise criteria and definitions or keys. This requirement is necessary
for the following two main reasons:

-
photointerpretation must be aa uniform as possible over time from the beginning
to the end of the work; the keys thus nerve as a permanent reference to the
photoint*rprter;

-
photointerprotation must be aa conaietent aa possible regardl^'* of the photo-
interpreters and the keys will serve to reduce the discrepancies between them,
These keys also serve as an aid for the field teame in on-the-spot recognition of
the different photointerpretation classes and thua make it easier to match the
photointerpretation and field classifications.

The criteria for the classifications used in photointerpretetion muat be easily

identifiable on the ground in order that they oan also toe used in the field inventory.
In other words, classifications designed on the basis of photography parameters such as
crown cover, height of dominant trees, soil Moisture, topographic features, should be
Meaningful and acceptable for a person on the ground; otherwise they are useless for the
inventory. In Many inventories the photointerpretation work and the field inventory have
been done irriependently, with the result that the photointerpretation work used for forest
mapping has not served to reduoe the sampling error of the inventory results and conversely
these results oould not be applied to the individual forest classes delineated on the maps.
It may happen that changee in the forest cover since the taking of the aerial photographs,
or errors in the photointerpretation work, introduce SOBS discrepancies between the results
of the photointerpretation work and the field work* These discrepancies oan be taken into
rocoont in the inventory design (see chapter 7) provided the classifications of the forsst
areas used in the photointerpretation work and in the field are basically the same,

As a consequence of the above requirements (objectivity and feasibility for use in

the field), the pbotointerpretation classifications mist net be too refined* If they art,
the risk* of errors, of inconsistencies over tiae and between interpreters, and ^f
discrepancies with the field observations, art increased and consequently result in a peer
performance.

332 Stereoscopic interpretation* Interpretation of ocaventional aerial photographs

should always be isade in order to profit fro* ths perception
siefecspoyically
cf height which, according tc many specialists, should b* considered >n* of the most
fftotivt ooapontnts of the interpretation. Observation of single photographs and of
noialos s*j, of course, provide useful information, but to a lesser d*fr<M
 ''
>> ok open car bi iaralfieu in three categories* len steraosoope*, mirror
*.t*jr"~ 'pe* r
and special application stereoscopes.

Lns stereoscopes are generally amall instruments with magnifying (2 to 4) lenses

separated by a distance equal to the spacing of the eyes. Tfcey can he ueed eaaily in tne
field, but have drawbacks such as limited magnifying power, the impossibility of viewing
the entire BtereoBccpic area in the line of flight without raiting the edge of one of the
photographs, and the impossibility of viewing the entire stereoscopic area aoross the line
of flight without shifting the stereoscope or photographs*

Mirror stereoscopes use a combination of prisms, mirrors and lenses to avoid the
abcve-wentioned defects of the lenc stereoscopes. They are the basic instruments of
photointerpretation an the office. Different types of frame wake it possible to soan ths
whole area vnth high ma^nifi cation, either by moving the plate on which the photographs are
put, or by moving the whole optical net, or oijy the mirrore ("0*d Delft" scanning

Of the opecial application etereoscoper, one ha* been designed to allow for the
f

stereoRcopic viewing of seveial ucc*RBive photographs of the &oue flight strip; another
has been designed fcr training purposes and permits simultaneous viewing of the sane
photographs by two interpreters.

333 Assessment of photointerpretation keys. The preparation of phoV;in%erpretation

keys should always be considered "an important part of the photo interpretation
work, and sufficient time nhould be spent on it. Ae it is e*ential that the photo-
interpretation classification be easily utiliwtble in the field, many ground oheoke have to
be carried cut on th* different ve#etxion and forest typee. A key should elwuya be
supported t\y c oet of test tereogi wnmeB each class being illustrated by one or more
t

atereo gramme. If denoity cf the <^rown cover is used as criterion for stratification,
"density ecalen" reproducing crown covers with different percentages of crown oioture can
be used.

~-s34 D'otointerpr^tatton cl ^lotfi _aiid_ jghojointerjpretatJlQn with delineation

S34 1 I n tr oduc t. j o ri

In a larg-Boale forest inventory, forest mapping may not be an objective of the

operation and it may be sufficient to estimate the areas of the different classes of forest
by a aa/npling of plot* on the aerial photograph* without delineating the classes on the
photographs aivi without transferring the Jimitu onto topographic or planimetrio maps*

n of plotn v ^fnerally mere satisfaotory than photointerpretation

with delineation. As a matter of fact it im often difficult to draw an exact limit
between classes; some subjectivity is unavoidable and there are transition soaes between
vegetation types or forest tvpen, Photointerpretation by plots is less liable to
subjectivity. Estimate* of the area of the same forest type by the same photo interpreter
may be significantly different under each of the two methods, and this difference will not
oe entirely due to the sampling design used for the photointerpretation by plots. It is
preferable in many cases to aooept a known sampling error on valid basic data (photo-
interpretAtiou by plote) than to obtain i a values not subject to sampling errors but
subject to unknown subjective bia** (phc to interpretation with delineation)*

334.2 Photointerpretation of plots

- In photoint*rpr*tetion without d*lineation r plots of equal sise should always be
used rather than points, no that the assessment of the vegetation or fo**t typ* i always
the assessment
done on the same reference area. (In phoUi interpretation with delineation
of the areas of the different oleses oar be made tjy dot ocute on the photographs, or on
the map* once information from the photographs ha* been *raaf erred)* 13>e plots are
 - 80 -

generally circular with a radius of two or more millimetres on a photograph. These

circular plots are generally printed on a stable transparent material laid on each photo-
If the aerial coverage in an inventory is based on more than one scale, it is
graph.
better to retain the plots at a constant size; thus, the circles printed on the
transparent overlay will have to vary in size in accordance with the scale.

A plot is assigned to a given class if more than 5&f> of its area belongs to this
clafle. The corresponding parameter for this class has the value 1 in this plot; in
a plot not assigned to this class. The proportion of the total area covered by a given
photointefrpretation class and the standard error are estimated by formulas given in
paragraph 53 of this chapter.

- Sampling design.

In principle, the sampling of the plots to be interpreted should be made on a base

map or on a mosaic, and the plots thus selected transferred onto the photographs. The
layout may be random or systematic over the whole area to be inventoried, or over each
stratum already delineated on the map (such as administrative or physiographic units).

For the sake of convenience the layout of the sampling plots is often made directly
on the photographs. The effective area of one photograph is assumed to be a rectangle in
the middle part, the sides of which are determined from the average endlap and sidelap of
the photographic coverage. Plots are selected in each rectangle according to a systematic
or random pattern. If the topography is even and the overlaps nearly constant, then a
systematic distribution of the plots on each photograph will result approximately in a
systematic distribution over the inventoried are*.

Due to variations in scale and in overlaps, the sampling intensity - which is the sane
in each rectangle - varies from one photograph to the next. A correction factor haa to be
applied to all the plots of a given photograph (or group of photographs of approximately
the same scale and having the same overlaps) to take into account these variations in
sampling intensity when they are significant. 0)

A systematic or random sample of aerial photographs may be selected for interpretation

instead of having all photographs interpreted. In this case the layout corresponds to a
two-stage sampling design, the rectangular effective areas of the photograph* being the
primary units, and the plots the secondary unite (estimation procedure may be the one
described in paragraph 422.1^1.111 "estimation of a proportion", if the plots are not
systematically distributed on each selected photograph).

334-3 Photointerpretation with delineation

When forest mapping is required, the various classes must be delineated on the
photographs. The effective area of each photograph has first to be delineated and the
delineation on this photograph restricted to the effective area. Detailed instructions
have to specify the dimensions of the minimum patch to be delineated (in relation to the
scales of both the photography and the final map), the precision of the delineation, and
other items such as degree of illumination of the photographs under the stereoscope, type
of pen or pencil to be used, rubbing out of the wrong lines, etc.

(1) Other more sophisticated methods of correction exist: cne of them consists of
determining the effective area of each photograph and using a correction factor
related to the scale of the photograph.
 - 81 -

t\ Forest mapping from conventional aerial photographs

41 Introduction

as generally done once the interpretation of the aerial photographs arid

delineation on th* photographs of the different forest and vegetation classes have been
completed. This is necessarily the case wnen the mapping is done uaing non-stereoscopic
plotters (see paragraph 42). Vfhen stereoplotters are used, it way be possible to dc the
nhotointerpretatiori and the transfer simultaneously since the operator has a stereoscopic
view of the area to be mapped* Stereoplotters of the third order can be bought and used
within the framework cf an inventory operation, whereas stereoplotters of the first and
cjecond order are generally the property of photogrammetric companies and institutes.

As already stated in sub-chapter 1 f we will assume that an acceptable planimetric or

'
jpographic mapping exists at a suitable scale over the whole inventoried area.

42 Transfer from single photographs

In this case the operator does not have a stereoscopic view of the area to be mapped
during the transfer, since only one photograph is used at a given time. The stereoscopic
interpretation is assumed to have been done before the mapping operation and cannot be
checked and eventually corrected during the mapping; this is a serious shortcoming but on
the other hand the equipment used is simpler and less expensive* The instruments must
allow for adaptation cf the scale of the photograph to that of the map to be drawn.
Adjustment of the photograph must be possible around its centre in order to rectify the
photograph, if necessary, to take into account the relief displacements and thus t9 put in
optical coincidence the same terrain features on the photograph and on the map. To avoid
too much manipulation of the photographs the transfer of the interpretation lines should
be restricted to the central effective area of each photograph (see paragraph 3343)

The usual instruments are based on the principle of the camera lucida and most of them
are called sketchmasters (Zeiss Aerosketchmaster f Aero Service Universal and Vertical
Sketchmasters, Abrams Oblique Sketchmaster). The following description of this type of
instrument is extracted from the "Manual of Photographic Interpretation" of the American
M
Society of Photogrammetry: The observer perceives two superimposed images, one from the
photograph and the other from the manuscript (map). This result is attained by means of
a serai-transparent mirror which both reflects and transmits light. The eye receives the
image of the manuscript by transmitted light. The operator can adjust the instrument so
that selected images on the photograph coincide wj,th their true positions on the manuscript.
Most of the camera lucida instruments can be raised or lowered to change the scale and
tilted to compensate for tilts in the photographs". The ratio of the scales of the
photograph and of the map can generally be down to 1*3 or 1i4

This type of instrument is particularly recommended for transfer from single

photographs if the relief is not too broken and the strata not too small and not too
intricate. If this is not the oase it may be more efficient to make a visual transfer
without the use of any instrument.

43 'fransfer from stereoscopic pairs

The main advantage of this type of transfer over the use of single photographs is
that it allows for a simultaneous photointerpretation f or for checks and possible corrections
of the photointerpretation work if this has already been performed.

The simplest and most u*ual instruments belong to two groups.

- The Radial Line Stereoscopic Plotters (third order), such as the Kail Plotter or
the Hilger and Watte Plotter, whose principle is the following: each point of the
 terrain correflponds to the intersection of two linea, each one passing through the
centrtof one of the two photographs ("radial lines"): the two lines are moved in
order to intersect at the points along a limit between forest classes, and a
mechanism links their intersection, possibly through a pantograph, to a drawing
pencil marking the reference map or manuscript* These instruments to not correct
for tilt and are difficult to use for transfer of detail near the flight line.

- The Multiscope and the Hilger and Watts Stereoskstch consist basically of a mirror
stereoscope combined with a camera luoida. Itoe photograph tables of the first
instrument (or the drawing table of the second) can be tilted to rectify ths
photographic image and take into account the relief displacement. In addition,
scales can be adjusted by inserting different lenses or moving up and down the
drawing table (Stereoaketch).

- More sophisticated stereoplotters using the principle of the fused floating dots
(for instance a parallax bar) can be used, such as the Zeiss Stereopret or some
stereoplotterR of third order (Zeiss Stereotope or S.O.M. Stereoflei).

5 Area estimation from aerial photographs and maps

51 Introductory remarks

As already mentioned in paragraph 334 f estimation of the areas of ths different forest
or vegetation classes does not require mapping of these areas; an objective estimation
can be performed by allocating, through photointerpretation, every photoplot of a sample to
its forest or vegetation class, provided the sampling design is sound and that correction
factors are applied to take into account the possible variations in scale and in overlapping
of the photographs.

Accuracy of the area results should be of the same order of magnitude as the total
error (sampling and measurement errors) of these estimates; there is no point providing
area results to the nearest hectare if the total error is expected to be around 100 hectares.

Area results should be given in the metric system whenever possible. If the British
system is used, both British and metric units should be given, as is required in all
inventories carried out by FAO.

Whatever method is used for estimating areas on maps, the precision of the estimation
will be higher with larger scales. Thie shows the importance of transferring details of
photographs onto maps at a scale which is not too much smaller than that of the photographs.

On a forest map, patches with dimensions smaller than a certain minimum are not shown*
This provision is generally necessary to avoid difficulties in reading the map, but it may
lead to biased estimations of some forest classes. The most obvious case of such bias is
when a classification of forests is made according to ths individual site of forest patches,
the area of the forest class corresponding to the smallest patents being underestimated*
This problem must be kept in mind, particularly when estimating forest areas on small-scale
maps.

A fundamental prerequisite for estimating aras on maps is that ths maps must be
drawn on a pap*r with a high dimensional stability opsffioisnt* This is particularly tru<
for nap* drawn on tracing paper ordinary
j oossisrcial tracing j*per oan expand up to 15jt.
Many stable-based materials are readily available (with polycstsr bass or estar base, ths
lcttr b*iog less stable), and it is highly recommend d that t\?c* b* u**d for mapping in
ordsr to avoid considerable and unknown bis*** in ths arsn estimation*
 - 83 -

There are several ways of measuring area* on maps. The indi cations given below refer
only to planimetering and to methods baaed on sampling techniques, sinoe the other ones are
deemed less practical in most cases (such as the one using geometric formulas and
coordinates or the one baaed on weighting)*

52 Direct measurements by planisie taring on nape

Hand planimeters have been used for some time and now there exist the much more
accurate and rapid electronic planlmeters such as the Stanley Cintel Electronic Planimeter
or the Kimoto Electronic Scanning Planimeter. In addition to the need for dimensionally
stable paper , mentioned above, other precautions have to be taken, among which can be
quoted the following:

(a) correct the scale setting given with the instrument against a master area
generally provided with the instrument;

(b) check that the measuring wheel is parallel to the tracer arm;

(c) avoid any slipping of the measuring wheel during the measurement operation;

(d) perform a sufficient number of measurements in order to reduce the measurement

error.

Unless an electronic plani meter is available, it is advisable to use statistical

methods (especially the dot-grid system) as they are leas liable to measurement errors and
allow for a computation of the sampling error

53 totimat ion methods based on sampling techniques

531 Within an exactly known total area (overall area

Area estimation from maps.
of the map, for instance), the principle of these methods is to determine the
proportion of the total ure* occupied by a given forest class* The estimate of this
proportion in given through a Hvnt*mAtin Trjoi\t ^f either:

dt, to each of which ie attached the value 1 if it is inside the forest class
or if it IP ^-itflide (dot grid wy*tem);

- or parallel lines (or "transects") to each of which is attached the part of its
length within the forest class the transect system is less used than the dot
;

grid system since the Je/igthfl have to be measured.

The areas of the forest clafls have to be estimated by multiplying the number of dots
within the class (dot grid system), or the mum of the lengths of the parts of transects
within the class, by an area extension factor (area of the unit square of the dot grid or
unit length multiplied by distance between two neighbouring transects).

The error formula given in paragraph 422 of Chapter 3 for estimation of a Droportion
is not applicable in the case of the dot grid method as the dots (sampling units) are
systematically distributed. Several authors have worked on the problem of error
estimatior in dot grid measurements f and error formulas hare been developed. An
of the standard error is given by the following formula Rafter
acceptable approximation
Chevrou - 97 1 ) 1

(*) - 56.5

where s
^^
n
the arrop ln p^.nt^
is the number of dots found
O n the estimated area of a forest class;
within this forest class j

and increases as the

k is a factor depending on the shape of this area,
for more regular shapes the value of this
area >>eooee more irregular;
coefficient is *id to be between 5 to ?
 - 84 -

Taking a conservative value for k equal to 7 we will have approximately

e ($) * corresponding to the following values for n:

\JA

dot count must be carefully performed and repeated, preferably by another operator.
If the two counts differ significantly, at least one other count must be carried out* Use
of small hand counters is recommended in order to avoid mistakes in enumeration* Counting
can be restricted to marginal parts of the area, the central part being divided into
rectangles or squares of known area* If a slight discrepancy is found between the sum of
the areas of the forest classes estimated by dot count and the exact known value of this
total, each individual area has to be corrected by the ratio of this exact value to the
estimated total.

Estimation by dot countr. and by planimetering can be combined in certain cases* For
instance the areas of the inventory units can be estimated by planimetering, the estimates
being corrected by the ratio of the exact value of the total inventoried area to the sum
of the estimated areas of the inventory units, and the areas of the strata within the
inventory units can be calculated from dot counts and corrected according to the same
principle.

532 Area estimation from photographs

Sampling designs in which the dots are replaced by "photoplots" are generally used*
The delineation of the forest classes on the photographs is not necessary unless the number
of plots per photograph is large* Examples of one- or two-stage sampling designs have
already been described in paragraph 334.2. Other designs are possible: for instance one
can imagine a three-stage sampling design wherein the primary units are parallel, but non-
adjacent strips, the secondary units are photographs within these strips and the tertiary
units are the photo plots on the effective area of each selected photograph* Multi-stage
sampling designs can also be foreseen using different photographic coverage, for instance
one space satellite coverage plus two aerial coverages, one of the latter at email-scale
and the other at a larger scale (see paragraph 622),

54 Continuous area estimation

By "continuous area estimation is meant the estimation of areas at different times

11

using successive photographic coverage*. Repeated estimations of forest areas are of

utmost importance in the tropics as the forest cover IB endangered in many places and it
is necessary to monitor these changes in order to achieve better control and to develop
appropriate land-use policy* The statistical basis for such studies is regression analysis
and some methods used are Identical to those used for continuous forest inventory in the
field (BO Chapter ?) Th* different coverages can be either complete or partial. If
the first one IB complete, the estimation of change can be assessed through a new partial
coverage using Bimple regression estimation (such a study ua* performed in the Ivory Coast
using a complete coverage of 1956 *t a scale of 1/50,000 and a partial coverage of 1966 at
a scale of 1/40,000),

If only one old complete or partial coverage is available, and if no significant

interpretation error is foreseen, the estimation of change can be obtained by regression
analysis from a ground check of photo plots precisely located in the field and interpreted
previously on the photographs*
 A part of the photopiots previously interpreted on the old coverage ran be
interpreted again on the subs^uent coverage, as in continuous forest imrentory, after
carefully transferring then from the old photographs onto the new onen, tnd or each further
coverage a partial sample of new photoplots can be interpreted.

It is thought that Kith the rapid development of remote sensing techniques and the
increasing concern for the maintenance of the forest cover these studies on forest
monitoring will develop considerably in the near future at the local and nationaJ levels
tLB well as at the rei^ional and world level*. Sampling theory offers a Jot ^f very useful
and efficient techniques provided they are utilized carefully and on * nound b*in*

6 Recent developments in remote sensing and mapping technicLuep

61 Brief presentation of recent techniques

611 New forms of remote sensing * The principle of coTventlonal panchromatic aerial
photographs can be stated as follows!

By means of a film covered with a silver salt emulsion they reproduce the re}atlvs
intensities of the natural electromagnetic radiations of all the bands of the visible
spectrum (the bands may be limited by a filter), the camera be i rig" situated aboard an
aircraft* Innovations in remote sensing with respect to normal panchromatic aerial
photography relate simultaneously to one or mor* of the characteristics mentioned in this
definition.

611.1 New platforms

a) The orbiting of manned or unmanned artificial satellites has now become a comnonplaoe
operation. When such satellites are fitted with remote sensors (cameras or scanners) and
with devices for storing or transmitting the images or signals collected (-television ey*t f
reproduction of signals on magnetic tape), one can obtain actual or televised photographic
Images, or data recorded on magnetic tape or disc, corresponding to the radiations received
front the overflown areas. The images obtained have the following advantages i

- they oan cover a very large area in a single exposure (approximately 3 million ha
for 10 x 70 mm negatives at a scale of 1/2,500,000) ;

- distortions due to relief are negligible and the picture has the planimetric
value of a map*
-'
The images suffer from one major drawback, namely their poor ground resolution
whioh corresponds to a dimension generally exceeding 80 m. Ground resolution ! limited
primarily by the height of the spacecraft, but it also depend* on the nature of the imog*
obtained. If the photograph is taken directly on a sensitive film, ground resolution
will also depen^ on the fineness of the emulsion grain. When the picture is televised,
i.e. from an unmanned satellite, ground resolution will depend on the scanning intensity
of th television system*

b) Krperls^nts have been conducted in Canada with vsry large-oale photographs

Such r,c<verage is designed tc
(1/1.500 and over) taken from helicopters or light aircraft*
certain species (e.g. Pioa glauoa and Abies balsmea), to quantify the damage
identify
caused by epidemics and insect attacks and to assess the characteristics of forest
means of trio measurements and "aerial" volume tables. Experiments
potential by photogravure
on species identification from very large-scale aerial photographs hv* also ben performed
in tropical American forests*

a specified contrast.
J/ i*e. the siae of the smallest object detectable on the image for
 - 86 -

The Most difficult problem ha* been to develop a precise height-finding system, an
accurate measurement of the camera's height from the ground being essential for a proper
estimate of scale and, consequently, sufficiently preoiee photograrmetrio data; the
first altimeters used worked on radar which did not penetrate all plant cover.

611.2 Other electromagnetic raoiation

The following table taken from the book "Remote Sensing, with special reference to
Agriculture and Forestry" (U.S. National Academy of Science) indicates the wavelength and
frequency intervals of the different electromagnetic radiations as well as the
corresponding sensors used to study natural resources.

Ftnohromatio photographs employ the reflection by objects of electromagnetic

radiations of the visible spectrum with a wavelength ranging from 0.36 Ato 0.78/x . The
real innovation in the field of remote sensing, admittedly less spectacular than the use
of space platforms, has been the use of radiation intervals other than those of the
visjble spectrum.

Hie use of part of the near infra-red (fron 0. 78/4- to O.JOyu ) in association with
visible radiations of 0.51* (or 0.6 M J to 0.78iAhas already been dealt with in paragraph
321.22.

Spectral region Wavelength Currently used imaging

sensors

Microwave (radar) Decimetre 10-100 cm Scanning antennas

Centimetre 1-10 OBI
Millimetre 0.1-1 cm

Par IR Scanners with IR

8-1,000^
Intermediate detectors
IR
Infra-red radiation
Near IR Photographic film to
0.780-3/*
approximately \jj*
Scanners with IR detectors

mtra-rlolet Newr UV Photographic film (quartz

radiation lens).
Middle UV 0.280-0. Scanners with photo-
electric sensors.

flbs use of "tfrsrMl* infra-red - the radiations emitted in

(3-14/0 greater quantity
fagr hotter bed! as - is especially suitable for the detection cf latent fires and of diseases
and. insect attacks affecting the forest (there result in a slight heating of the vegetation)*

i at ion
(between 0.5 cm and 1 m and especially between 0.86 am and 3.3 cm)
possesses the great advantage of being able to penetrate cloud formations and of being
relatively little attenuated by rain. Its use is proving very interesting in tropical
regions where conventional photographic coverage is a time-consuming procedure because
of the Marly permanent presence of a rather low cloud base.
 - 87 -

611.3 Scanners

The use of new radiations and requirements for continuous automatic re- transmission
(especially from unmanned satellites) have led to the increased use of scanners. A staple
scanner consists of an optical device (generally a rotating siirror ooupled to a parabolic
mirror) and a sensor which converts into electric signals the variations in the intensity
of radiation in a certain band of the spectrum (like a photo-electric cell)* Through the
movement of its optical system and of the aircraft, the scanner examines the soene to be
observed in parallel contiguous stripe. Hie electric signals are then received in a
cathode ray tube for display on a televioion screen or for printing on a sensitive film.
They can also be transcribed onto magnetic tape and then processed by computer.

611.4 Artificial radiation

The radiations used in conventional photography are natural radiations reflected by

the object photographed and which come mainly from the sun, either directly or indirectly
via other objects that have re)ayed these radiations by reflection or transmission.
Certain remote sensing systems (active radar devices) employ specially emitted (artificial)
radiations which are reflected back by the objects photographed* Such systems can also
perform sensing at night without difficulty.

611.5 Band selection

Another important element in the field of remote sensing is the separate reception of
the radiation received on each waveband. The advantage of isolating certain spectrum
bands is obvious. For example y two objects to be differentiated may reflect with the
same overall intensity the total radiation of the visible spectrum whilst reflecting with
very different intensities a specified band of the visible spectrum* In other words, their
"signature", or response with reaptct to that band, will be different and distinction
between them will be clearer*

Application of the principle of band selection is not new in faot| the yellow filter
Intended to stop radiation with a wavelength of under O.JO/*, is an exam pie. of selection in
that it amounts to selecting the 0*50,4* to Q.'lQ/*. band of the visible spectrum.
Conventional colour photography also involves such selection, as in these* emulsions a
y*llcw layer receivea 0.38 A*. to 0.48/A radiation, a magenta layer O.JC^ to O.Jo^ji radiation,
a eye*) layer 0.6o/^to 0. InJA f
the three images being superimposed (unlike black-and-white
panchromatic film uji which a single image is formed). An interesting example of colour
fill* is the spectrozonal film used in the U.S.S.k.j its emulsion possesses only two layers
and it has proved very useful in forestry.

Separate recording of the images in the different bands is effected essentially in

two new ways:

by combinations of several cameras (up to 2? in one oast) or by cameras with

-
several lenses where each camera (or each lens) corresponds to a specific
of the earn*
emulsion/filter combination? thus different simultaneous images
scene are produced, each corresponding to a particular band of the spectrum;

by an equivalent system in which the GmMrator cameras, are replaced Igr ee*or|
for example, the "multi spectral line scanner eyatem, in which several scanners
19

in a
operate together, each one reproducing the radiation emitted by the object
band of the it has a optical device beaming the total
speoifio spectrum | single
radiation onto a prism which scatters it according to wavelength! the scattered
radiation passes through a nwiber of sensors each of whioh is sensitive to a
r

then oowrwted either

given band of the spectrum, and the resulting signals are
into a televised image or into a photographic image (in both oases through a
oathode ray tube), or into data stored on a macn*tio base*
yet again
 'ip New media for information storage and reproduction* The information collected
nas so far been assembled in the form of black-and-white or colour photographs
on an opaque or transparent base, in negative or positive form, produced directly by simple
chemical reaction development on the sensitized film. For a long time to come the forester
wilJ continue to use this type of information base for much of his work*

The major innovation in the field of reproduction is the use of the cathode ray tuba*
The electrical data transmitted by a sensor are converted by the tube into visible
information* It is the principle of television applied to the recording of radiations
which are not limited to the visible spectrum. The picture obtained may therefore be a
black-and-white or colour image on a television screen, or it may be obtained directly by
sensitizing a film at the output of the tube* In this way black-and-white cliches are
obtained from radiation in the thermal infra-red or radar radiation. These same cliches
can be converted into pictures in standard coded colours where each shade of grey is
represented by a colour and a nhade in that colour, thus permitting naked-eye differentiations
which would be impossible on the corresponding black-and-white clichfc. Electric impulses
at the scanner output can also be stored on magnetic tapes or discs*

M New procedures for information analysis.

\
Analysis of conventional aerial
photo-interpretation - employs the human eye and brain aided by
-
photographs
optical devices (magnifying equipment, stereoscopes). Despite the great weaknesses of
human interpretation, this type of analysis will continue to be done both on conventional
pictures and on black-and-white and colour pictures corresponding to radiations outside
the visible ppectrunu

A imple devire whir.h can improve the human interpretation of the cliches obtained is
\

additive viewing* This Consists of the projection onto a single screen, through particular
colour filters and with varying intensities, of positive black-and-white transparencies -
each corresponding to a given waveband. The result obtained is a "false colour" picture.
The /alue of this device lien in the fact that the different shades of grey are converted
into a much greater number >f colours and colour shades (chromae, hues, and values)* By
altenn,', the filter^ ne can obtain the false colour image that best displays the difference
netween two object* that won id otherwise be undetectable on black-and-white cliches.

Micro dens it pane trie anal^Bio is a technique whereby a light spot scans a photographic trann-
parenoy and the variations in luminous intensity transmitted through the photograph are
transcribed onto a graph. The system operates by means of a sensor which converts the
variations in luminous intensity into electric impulses, the impulses being amplified and
transmitter through a ncnbing arm to the graph. A type of crown can thus be reproduced
ae a certain curve shape. One can imagine the possibilities of this method* For
example, if the type of ourve corresponding to a given species is fed into a computer
together with the tolerated fluctuations, it becomes possible, with the mi erode nsi tome trie
device linked to the computer, to count (and perform calculations on) the number of
"f>rreBp:> riding orownfi encountered.

Mere generally, tne introduction into a computer of the magnetic media storing all
tn* data relating to the electric signals produced by scanners sensing the radiation from
the observed acene (or fro a photographic image of it) permits automat i c ( and objective )
processing of the data* It it is likewibe given the data on the aircraft's path the
oonputer can thus make It possible for instance to locate the hot points observed by the
thermal lufr-r*d sensor. The advantages in many cases of such automatic processing
over huiai) interpretation are clear, especially for radiation outside the visibl* ptotrup
where the cliches obtained have a poor rtswiblance to the visual imago to which we are
ecu* toned*
 -89-

614 Orthophotography* A new technique has been developed for plotting from
aerial photographs. It consists of reproducing photographically and without
geographical distortion the portion of land common to the two photographs of a stereogra*.
Orthophotography is therefore photographic plotting as opposed to the conventional
cartographic plotting which results in topographic maps. All the orthophotographs for
a particular region can therefore be brought together to form what is called an
orthophotoplan which has the same plan! me trie value as a map. Ihe lines traced on the
stereograms - such as boundaries of forest types and, of course , contour lines - can be
automatically reproduced on the orthophotoplan. Plotting of forest boundaries can also
be done under a stereoscope from the stereogram composed of the worked up photograph and
the corresponding or the photograph.

The chief interest of this device is that it allows more thorough and concrete
mapping than normal mapping. It is also about ten times quicker. Its price remains
high; for example f plotting by Orthophotography of the useful part of a 23 x 23 on
negative costs between US$ 80 and 140, the price depending on the scale and on the quantity
of additional information to be plotted (e.g. boundaries of forest types). If contour
plotting is added to this, the cost ranges between about US $ 160 and 250.

62 Current operational applications for forest inventory

The possible combinations of these different innovations are, of course, extremely

numerous and an immense field of application lies open in the field of natural resources
evaluation. Applications in forestry and in vegetation studies are at the present time
largely in the research and experiment phase. In general, their use, even when.no
spatial observations are involved ^ assumes a technological infrastructure and financial
resources which are not available to all institutes or even to all governments* The
following paragraphs simply indicate the accessible and /or operational procedures in the
field of forest resources evaluation.

621 Use of radiation outside the visible spectrum

621.1 Use of radiation in the thermal infra-red

Systems for the early detection of forest fires are worth mentioning although they
are not directly relevant to forest inventory. On<t employs an aircraft flying at a

height of 7,000 m deploying a scanner recording IR radiations of 3 to 6 /A and 8 to 14/A

Electric signals corresponding to radiations of 8 to 14 }*- are transmitted to a cathode ray
tube and continuously act on a film which is developed very rapidly. The 3 to 6/A
radiation band is used to indicate hot points at the moment when they are sensed. A
computer which integrates the data on the aircraft's path (ground speed, altitude, bearing,
drift) determines the film speed and makes a mark on the border of the film corresponding
to unit distances covered. If a hot spot is overflown its coordinate along the line of
flight is also indicated on the border of the film, thus making it easy to pinpoint and
analyse it. Action can then be taken before the fire develops.

621.2 Use of radar radiation

An "active" radar device (i.e. one recording the bea reflected by ground objects
of the radiations emitted by itself) has been successfully used in one of the dense tropical
forest fones of Latin America (Nicaragua, southeast Panama and northwest Colombia) and has
permitted mapping of eone* permanently covered by a fairly low cloud base. Another very
important "active" radar mapping operation was carried out in the northwestern part of
Brasilia* Anaconia.

fciebasic principle is still the saMf namely, a scanner (in this oae an antenna)
covering the observed terrain in strip* transverse to the direction of flight, electric
signals being introduced into a cathode ray tube which continuously senaitiaaa a film.
 - 90 -

One feature of thefle devices which should be noted is that the photographed strip is not
situated vertically underneath the aircraft but to one side (side^-looking radar); this
allows better determination of the distance to the ground from chronometric measurements*

The original scale of the pictures obtained in Panama was around 1/200,000. In the
southeast region (southern part of the province of Darien) they have allowed satisfactory
planimetric mapping at a scale of 1/250,000 and mapping of vegetation by major classes.
The control pointsused were small metal pyramids cleared of all vegetation f possessing
known coordinates and easily located on the radar pictures.

622 Use of opace platforms

622.1 Edirth Resource Technology Satellite (ERTB) Programme

Several thousands of satellites have already been launched for various purposes
(weather observation, intercontinental broadcasts, radiation Btudies f etc.). In the
field of natural resources evaluation, data obtained from flights such as Gemini and
Apollo have opened the way to special studies. A great step forward was taken in July
1972 with the launching in the United States of the first Earth Resources Technology
Satellite (ERTS-A).

The characteristics of this flight were the followingi

- lifetime: 1
year;
- altitude: y2<j km on a sun synchronous orbit;
- repeated coverage of the same zone every 18 days;
- earth distance between two passes 2 160 km;
- satellite-borne sensors: - a television system (RBV) recording images in three
bands of the visible spectrum and of the near
i nf ra-red ;

- a multiple scanner (MSS) recording images in four

bands of the visible spectrum and infra-red; tho
electric signals are coded and recorded on tape.

The scale of the original pictures obtained (70 mm) - for each band and also for
the composite colour pictures^- is about 1/2,500 f OOO. Each negative corresponds to a
180 km square, i.e. 33 f 000 km* Ground resolution of the negatives varies between 60
1

and 150 m depending on the contrast of the scene examined and the sensor concerned.
Longitudinal and lateral overlap is low, around 15$. At the time this manual is written
it is too early to indicate what is the real value of this imagery for forest inventory
in the tropics, but it can be predicted that it may be very useful for broad vegetation
and forest typing in large-scale forest surveys f especially if dombined with other
photographic coverages as indicated in the following paragraph.

622.2 Multi-stage designs using apace photographs

A statistical design, simple in its principle though more complex in its

mathematical formulation, has been conceived for inventories of vast areas J/. It uses
spatial photographic coverage on which a square grid is superimposed* A number of
squares are selected proportionally to the forest area they contain, the latter being
determined by interpretation of the space photographs. Small-Hioale (l/30 f OOO to 1^0,000)
aerial photographic coverage is then carried out on these selected squares. This

\J See "The benefits of multi-stage variable probability sampling u*ing space and
aircraft i^^ery" by Philip 0. Langley in "Application of remote senior* in forestry",
joint report by working group on Remote Sensing of former Section 25 of IUFRO.
 - 91 -

coverage is, in turn, divided into a grid from which squares are Belected on the same
basis as previously, and theee square* are photographed on a larger scale (1/5,000 to
1/?5,000). Finally, field sampling plots are selected within the latter squares, and
the reaults of the field work are applied to the whole zone. This method will probably
prove profitable when spatial coverage i readily available, but it is limited by its
nature to national or regional inventory operations.
CELPTER V

c OKI DERI TIOKS

 - 93 -

CHJLFEER V

MEaSUHEHENT COlEIDERATTOg

1 Introduction

In addition to the areas, there are many characteristics of the forest stands which
it is useful to know for their management and that inventory is aimed at estimating. The
most common and generally the most important character is tics are related to the volume of
wood; gross or net or extract able volumes, by species, groups of species, by diameter
classes or groups of diameter classes, by quantity classes, down to a minimum diameter,
estimated at the time of the inventory or subsequently (through the estimation of volume
increment), etc. But often other characteristics are just as, if not more, important.
Humbeaof stems by area unit, by species and diameter classes are basic parameters which
are generally easy to determine from the basic inventory data and are necessary in forest
management. Other parameters related to the volume of wood may be more interesting to
know than the volume itself: for instance in many forest inventories it would be more
important to estimate the wood potential in terms of value, taking into consideration the
different species and types of the standing volume. Assessment of quantities of other
forest products, such as cork, is sometimes the main objective of a forest inventory. In
almost all forest inventories additional parameters have to be estimated, such as those
related to site and accessibility. The contents of this chapter will be restricted to
the problems of volume estimation, quality appraisal and accessibility assessment.

In most forest inventories volume information is obtained from the field inventory,
although stratification by photointerpretation may be based on items in relation with the
total volume of the stands (such as density and height of the dominant trees). However,
in BOB* temperate countries, when species identification is feasible on aerial photographs,
mott of the volume information is taken from the photographs through pho to gramme tri o
measurements, the remaining part of this information being obtained from a few field
samples. This method which has proved efficient for some temperate forests is not
applicable to the mixed tropical forests. For this reason photograometrie measurements
for volume estimation will not be considered in this chapter.

Volume estimation is based on measurements of tree or stand characteristics (diameter,

height, basal area ...) and on quantity relationships between those measured characteristics
and the volumes to be estimated, whereas the assessment of quality of wood and to a lesser
extent the evaluation of accessibility is based, at least partly, on personal Judgement
and is consequently less objective. Although this cannot be avoided, the extent of
subjectivity should be restricted to a minimum in order to avoid discrepancies in
estimation between different taxators and even for the same taxator throughout the
inventory. For instance, quotations of quality (allocation of a given tree or parts of
a tree to a given quality class) must correspond to the occurrence of one or a given number
of precisely defined defects* The quality class has to be narrow enough to make the
exercise worthwhile but at the same tine wide enough to make it less difficult and in the
end more reliable. The problem is similar for the assessment of some aooessibility and
logging parameters such as soil-bearing capacity or irregularities of terrain which are
not measured, strictly speaking, but are only qualified by reference to a given class.

A broad classification of inventory measurements not directly related to area

determination is the following!

a) tree and log measurements on standing or felled trees

- enumeration and species identifioationi the assignment of a tree to the sample

is preliminary to further measurement and is sometimes done through a measuring
instrument (such as the Bitterlioh relasoope in horizontal point sampling).
 Speciee identification, although not strictly speaking a measxurement, is a
fundamental operation and sometimes difficult and time-consuming (especially in
inventories of mixed tropical hardwoods);

- measurements of diameter (over bark or under bark) at breast height, at the

stump, at various levels on the upper stem, of diameter increment, height and
length (total, of the bole, of the merchantable bole, from the ground or from
the top of the buttresses, up to a given diameter), of bark thickness and of
characteristics related to minor products (such as cork thickness);

o) other measurements

- regeneration counts by species and by density classes of seedlings, by height

and/or diameter classes of saplings and poles;

- measurements for site quality assessment I in addition to the measurement of some

tree and stand characteristics, site quality can be assessed by counts of ground
and scrub vegetation and measurements of parameters related to soils and to
topography;

- measurement of accessibility parameters;

- various measurements such as seed collection parameters for individual species*

c) quality appraisal and other evaluation without true measurements, mainly for stand
description, site quality and accessibility*

2 tree measurements

21 Definition of^tej

The following definitions have served for forest inventories carried out by FAO.

a) Stem: for trees of deliquescent form, the length of the trunk between ground
level and the crown point (eee below for the definition of the crown point);
for trees of exourrent form the length of the trunk between ground level and
the top of the tree.

Hemarki Deliquescent formed trees, especially broadleaved species, hare a stem which
is strongly evident in the lower portion but, due to branching, becomes less distinguishable
in the upper crown (many trees of tropical broadleaved species have however a clear bole up
to the crown point and the stem is easily distinguishable). Exourrent-formed trees
exemplified by numerous coniferous species have a definite central stem which extends from
ground to top* When there is a fork, the number of sterna to be recorded depends on the
location of the fork with regard to the reference height for diameter measurement (breast
height for non-buttressed trees): if at the reference height the main stem ia already
divided in two or more stems, these latter hare to be recorded instead of the main stem.

b) Crown point: the crown point is located at the origin of the lowest crown-
forming branches, living or dead; at this point in many broadleaved species
the stem starts to disperse into the crown; isolated single branch** below
this point, if they are at a distance of more than half a specified log length
from other branches, should net be used to determine the crown point*

o) D.bh t the diameter at 1*30 metres (4*3 feet) above ground levti (for trees
standing on slopes, the point of meamsgmsmnt wmmt be determined ca the upidll
side).
 Case of buttressed trees: if buttresses exist and are higher than the breast
height level, d.b.h. measurements are useless: buttresses are often irregular
in cross section, are difficult to measure at standard breast height with
accuracy and their dimensions have a loose relationship with the volume of
the tree; a satisfactory standard procedure for measuring the diameter of
buttressed trees has still not been developed and different procedures have
been used up to now; measurement just above the termination of swelling or
irregularity or at a given distance above that point (30 centimetres was
reooemiended for FAO inventory operations in the former edition of this manual).

Remarks: Clear definitions or instructions have to be given to inventory crews as to how

and where to measure the d.b.h. (definition of "ground level", of d.b.h. for buttressed
trees, of d.b.h. for stems of irregular form, etc...). This la a fundamental prerequisite
in obtaining basic data a* homogeneous as possible. The definitions used should be as
identical as possible to those commonly used elsewhere. Comparability of inventory
results is sometimes impossible due to differences in the definitions between inventories.
4 feet 3 inches should be taken as equivalent to 1.30 m and not 4 feet 6 inohes as in North
America. For the sake of comparability and consistence (see paragraph 24) diameter
measurements and diameter classes should be used rather than girth measurements.

d) Height and length measurements; Ihe following classification of height and

length measurements on standing trees (adapted from "Forest Mensuration** by
Husoh, Miller and Beers) and illustrated in figure V-1 has served for
inventories carried out by FAO.

- Total height: the vertical distance between ground level and top of the tree.

- Bole height: the distance between ground level and crown point: it
expresses the height of the clean, main stem of a tree.

- Merchantable height: the distance between ground level and the terminal
position of the last usable portion of a tree.

There are several criteria which can define this upper terminal and the
exact location is, to a large extent, subjective and made more problematical
due to the difficulty of sighting the upper part of a stem in a tree crown
under forest conditions* The upper position may be defined by a chosen
minimum top diameter or by branching, irregular form, defect, etc., which
limits what is considered the utllisable wood in a stem* The merchantable
height may be up to a minimum top diameter or below but never above it.
The minimum top diameter chosen will depend on the intended use of the wood
in the stem. The definition of the utilisable wood and of the corresponding
eliminating defects must be as precise as possible in order to reduce the
personal component to a minimum (and preferably to sero).

-
Stump height: the distance between ground level and the basal position
of the main stem where a tree is out.

this length depends on cutting practices. For the buttressed trees in the
tropi oe, stump height is generally considered just above the buttresses.

- Merchantable length i the aum of the lengths of the portions of a tree which
are out and utilised: this includes material suoh as trim allowance which
may be wasted in the manufacturing process .

Ifeere are threemajor difficulties in the assessment of merchantable length

oa standing tress. Ibe evaluation of the external defects on the upper parts
of the stem by an observer at ground level may not be precise emough unless
 - 96 -

Figure V- 1 Tn* height Md ti length olaMifiotl(*m

** Excurrent__form

crown
point
1
}J[miatopdian
Ld
*
util.limit
1
1
H Hm
2

T
Hm
-d \
Lm'l

b Deliquescent fora

crown
point mm top du
frown
point
H
Lm
'
/ Hm T
L Hb
Hr

r
H
y \
(note no mm. top dia.)
 binocular* are used. Secondly the log* deemed merchantable 67 the inventory
people May not be identical to those out by the logger* &ie is particularly
true for mixed tropical hardwoods due to the changing condition* in local and
international market*, in accessibility of the inventoried mone and in logging
practice*. Finally the assessment of Merchantable length on tattling tree*
take* only the external defect* into account in most case* and not inner
defect* which are often vjore decisive factors ef merchantibility*

- Defective length: the SUM of the lengths of the portion of the stem which
diameter is larger than the minimum acceptable but which cannot be utilised
because of some kind of defect*

The sasje remark* a* fcr the assessment of merchantable lengths on standing

trees apply in the case of defective lengths.

- Crown lengths the distance between crown point and the tip of the tree*

Height measurements of standing trees are vertical distances while length

Measurements may be made on sections whose axis departs from the vertical*
A.dditionally y a merchantable height may include some defective lengths below
the point defining the upper limit of merchantability* Consequently it is
possible that the totals of length measurements may not agree with height
measurements e*g the total of merchantable leng-fch* may not agree with the
,

merchantable height for a tree.

22 Enumeration

Before any tree is measured, It must be decided whether the tree b* longs to the
pie or not* This is the principle of enrameratioc and its importance must not be under-
estimated* Enumeration is not the samt in sampling units of * givn area and in point or
line sampling*

221 Enumeration in sampling with units of a given area* Enumeration consist* of

two check*}

a) whether the tree i* within the sampling unit: the distance of the axis of the tree
from the centre of the plot (circular plots) or from a side (or an axis) of the plot
(square or rectangular plots) must be smaller than a given length; precise
instructions hare to be given to the inventory crews as to how to measure this
distance, and whether the distance is to be measured horisontally or along the
terrain, as well as particular indications on borderline trees;

b) whether the characteristics of the tree make it enumerable: these characteristics

are its species (see paragraph 23 for problems of species identification), since in
some inventories, especially in the tropics, several species are not recorded, or it*
dimensions; generally the dimensional criterion for enumerable trees is a minimum
diameter and more measurement* must be made than the one* which are recorded (see
paragraph 24 for measurement problems)*

222 toomeratiop. in point or line sampling* The selection of the tree* to be

recorded is made by mean* of a measuring instrument* In horizontal point
sampling which is the meet oommon sampling design of that typ*, a tree is selected if it*
horizontal distance H from the sampling point is smaller than its liametey D divided by a
certain factor k equal in the metric system to

being called the basal ar*a factor of the instrument*

 9 ! the horizontal gauge angle of the instrwent (angle of eight from the
sampling point of a borderline tree); and
2
BAP - 10,000 in | (in the metric system)

Every tree in the surrounding* of the point which satisfies the species and minisnmi Bite
requirements and with R ^r belongs to the eejiple.

Onoe the tree is known to belong to the Maple, then other measurements on this tree
nay be made. Sometimes point sampling ie used only for estimating the ba*al area of the
stand: in each sampling point the basal area per hectare ie equal to:

p x BiP

where p is the number of trees around the sampling point for which &<r
H and are measured in the horizontal plan of the observer's eye* In inclined
terrain a correction has to be introduced for every tree in relation to the slope of the
sight line of the tree. Simple angle gauges and prisms do not permit automatic
correction whereas the Bitteriioh relascopee make provision for it.

23 Species identification

Species identification poses sosie problems in inventories of tropical mixed forests*

The relatively large number of tree species t the restriction of botanical knowledge to a
few individuals, the similar appearance of trees of different species, make species
identification particularly difficult* Nevertheless good identification of the speoies
is a fundamental prerequisite of any forest inventory: it is much more seriou* in most
cases to make a mistake concerning the species of a tree than to make it about one or
several of its dimensions*

In many tropical forest inventories it is hardly feasible to combine perfect

botanical work and satisfactory efficiency* Indeed it is generally too time-consuming
to assure a completely accurate species identification* The following indications have
to be kept in mind when looking for a compromise betweec the conflicting requirements of
botanical exactitude tuxl of efficiency.

a) It is often not neoesaary to identify all the trees botanioally within the
whole inventory sample. In view of this, several devices can be adopted
such an:

- enumeration of a limited number of commercial and oommerciali table speoies

("desirable" species) in the entire inventory sample and enumeration of all
species in a subsample in order to reduce the costs of th* inventory and at
the sane time to obtain an acceptable knowledge of the floristio composition
of the inventoried forests;

~ enumeration of all trees with a diameter larger than the minimum exploitable
diameter and ot the trees of the "desirable" speoies below this diameter i
this device saves quite a lot of time as a large nonber of small trees do
not have to be measured and recorded;

- species with very little ooourreaoe vhioh have very little ofcaaoe of being
us*d need oot be identified with certainty and o*& be merged under one or
several groups of "undetetrvined", possibly hy botanical fmmilie**

The selection of one ox the other of the two first device* assomeo that the
*deeirabl* M epeoiee o*n be perfectly id*trtifie^..
 b) In order to record less trees, and taking into account the fact that the
coefficients of variation of parameters related to mailer diameter classes are
often lower than those for large diameter claasee, it is useful to adopt
different ies of sampling unite (or of the sample) according to diameter
classes t for instance each sampling unit or recording unit nay consist of two
or three concentric oiroulax plots, the smaller circle being the sampling unit
or the recording unit of the smaller dl matter classes.

c) Once the different characteristics of the enumeration work has been decided,
efforts should be aimed at securing species identification bj every possible
Beans, among which can be quoted the following:

- thorough and intensive training of the treeepfctters, preferably in different

parts of the inventoried one;

- limited number and permanence of the tree spotters in order to obtain the most
homogeneous data;

- establishment of a corresponding list of local (vernacular) names and

scientific names: this is generally a long and difficult taak since the
criteria of classification used by the botanists on the one hand and the bush
people on the other are different, reeultlrg in many discrepancies, e.g.
several local nax*s for the same species depending on the age and sex of the
tree, or, on the other hand, one local name only for several species and
sometimes genera, the traditional use of which is at the same time identical
and limited!

- assessment of simple and practical field identification keys based on a

limited number of characteristics such as bark, slash of bark, leaves and
fruit;

- preliminary collection of wood, leaf and fruit samples for reference purposes
throughout the inventory;

- control checks by a botanist or tb* beet treeepotter in randomly selected

sampling units, immediate analysis of the results and consequent further
instructions' to the crews;

- use of systematic botanical check procedures such aa the following which was
used in Sarawak: a sample of leaves was collected acd put in a separate oag
for every tree of the sample, the bags being further despatched to a botanist
who crosschecked the local name given by the trespotter and his own
identification.

If rectifications] in the exoneration work related to species identification are to be

made in ths course of the inventory, attention must be paid to the way they are introduced.
In order to get homogeneous data in all inventory units, it is better to avoid introducing
these corrections in the inventory units wherein the enumeration work has already started,
unless enumeration is entirely taken up in these units,

24 Measurements

Estimation of the volumes of the trees of the field sample and of the stands is made
- at breast height and
through measurement* of characteristics of these trees: diametsr
at aay other l*rel of the stem or possibly of the branches (on felled treee); heigfrt up
-
to a firwn lenrel of the stem, or leajrth along the stm or the branches, and bark thickness
generally at toremst height only on standing trees. Measurements are made either on standing
treee or on felled treee, especially for the assessment of volume relationships*
 - 100 -

241 Measurement units. *Pie use of the metric system IB highly recommended as it is
the most practical and as most countries have adopted it or intend to adopt it
in the near future. Conversion to volume and weight units through appropriate relation-
ship* is also easier with the DO trie system. Diameter is generally expressed in centimetres,
or sometimes in metres. Height and length are practically always expressed in metres and
bark thickness is often given in Millimetres. Diameter increment in determined in
centimetres or millimetres.

In countries where the Britiflh System IB presently used, it is advisable to convert

the linear results directly obtained from the measurements into metric units as for the
volume results*

242 Measurement classes

242.1 It could be imagined that all measurements are taken to the smallest possible
discernible unit or part of unit. This would be unrealistic, mainly because
measurement errors in a forest inventory are often greater than the nearest unit. As this
might also be more expensive, many measurements - with the exception of bark and diameter
increment measurements - are made by classes.

Die amplitude of the classes used for diameter, height and length measurement is
determined in the light of such factors as:

- instruments and devices used for measuring and their accuracy;

- environmental conditions such as visibility and tree fcna;
- skill and training of the crews;
- homogeneity and comparability of data from different inventories.

Regarding the last consideration the following remarks can be made!

a) 1he amplitude of the diameter classes used in the metric system is generally
5 or 10 om f the minimum diameters being a multiple of 2*5 cm* However, due
75
to the different minimum diameter* adopted (5 om t cm, 10 cm, 15 cm) it is
not always feasible to make easy comparisons between inventories;

b) when using British measurements, one should try to use class limits
approximately equal to clans limits in the metric system, which means diameter
classes equal to 2 or 4 inches, and length or height classes equal to 5 or
10 feet|
the attached table show* the minimum standard class limits which are
recommended for FiO inventories in the metric and British systems;

o) when the inventory uers are interested only in volume estimates, the use of
classes of equal basal area amplitude (basal area classes) instead of diameter
classes may be recommended; the volume estimates obtained through volume
relationships are indeed statistically more valid when they are derived from
enumeration with basal area classes than when derived from enumeration with
diameter classes; basal area classes can be used for instance in an inventory
of tropical mixed hardwoods for tress above the minimum exploitability diameter
(volume estimates above this diameter) whereas diameter clasnss are used for
tress below this diameter, sinoe estimates of numbers by diameter classes of
these stems are more useful for management purposes than volume estimates.
Moreover, suoh a system can be applied insofar as comparability of inventories
is possible, i.e. when one is interested only in the comparison of volume
estimates above the minimum exploit ability diameter.
 - 101 -

Minimum Diameter and Height Class Limits

for Inventory Calculations and Results

Diameter
Metric System Approx. equivalent Actual equivalents
cm. in British system in cm.
ins .

- 5 0-2 0.00 - 5.08

5 - 10 2 -
4 5.08 - 10.16
10 -
15 4 - 6 10.16 - 15.24
15-20 6 - 8 15.24 - 20.32
20 - 25 8-10 20.32 - 25-40
- 10 - 12 25.40 - 30-48
25 30
etc. etc . etc .

Height
Metric System Approx. equivalent Actual equivalents
cm. in British system in cm.
ins.

0- 3
 - 102 -

242.2 Understandably the precision of the inventory results is affected by the

class grouping of the basio seasurement data and that ths larger the classes
the more significant the corresponding errors. i thorough analysis of this type of error
is made in "Forest Inventory", vol. II by Loetsoh, ZBhrer and Bailor (pages 85 to X)) 9
which can be presented concisely as follows:

- there are two components of this error;

- a systematic component (bias) originates from the difference between the actual
moan diameter of the trees within a diameter class and ths aide lass di erne tor,
and/or from ths difference between the mean basal area of the trees of this
diameter class and the basal area corresponding to ths midclass diameter; these
differences come from the distribution of ths diameters within the diameter class;

- a random component comes from the fact that, in a forest inventory, only a sample
of trees of a given diameter class is measured and the estimate of the mean
diameter of this class (and of the mean basal area) has a sampling error (in
general the estimate of the mean diameter of the class from ths sample is not
equal to the actual mean diameter of this class, itself different from the mid-
class diameter in many oases - see above).

It is complicated to have to take this type of error into account in the error
calculation of the final results. The best solution is to adopt small classes since the
magnitude of these errors increases with the smplitude of ths classes, and to consider that
the corresponding errors are negligible .

243 Measurement procedures and instruments. It is not intended to describe and

comment on the various instruments used for measuring diameter, height, length
and bark thickness, but rather to give some information and advice on Measurement
procedures which are more directly relevant to inventories in nixed tropical hardwoods.
More complete information is available in forest mensuration and forest inventory
publications as well as in advertising leaflets published by the manufacturers.

243*1 Diameter measurements

243.11 Diameter at breast height

For reasons mainly of practicability calipers are little used in tropical forest
inventories} the fairly large sise of the trees, the occurrence of high buttresses, of
aerial roots, ths difficult working conditions maks ths use of oaliper* little adapted
to thess forests.

dirth tapes are used on felled trees and on standing trees with buttresses less than
approximately 2 metres high* Regarding the use of tapes in tropical forest inventories,
ths following indications are worth mentioning:

- fibreglass tapes often prove to be the most suitable;

- a hook at the sere cf the tape permits the measurement of large trees by
one person;
- risk of tilt from the horizontal plan of measurement and of looseness of the
tape is relatively high with large trees and great cars must b* sxsrcised in
having the tape well stretched in the horizontal plan of measurement;
oreepore along the bole are very common and they must be cut at the height of the
measurement or the tape must bu pat below them if this is feasible;
 -
graduations of diameters in length unite (centimetres or inches) or diameter
class limits mu*t be indicated on one face of the tape; purchase of tapes with
a blank face where limits of the adopted diameter classes are further marked with
special ink is recommended*

For trees with "diameter at breast height " higher than 2 metres, procedures and
instruments for measurement of upper stem diameters have to be used*

243 1 2 Upper stem diameters

Many fairly sophisticated instraents of varying degrees of accuracy have been

devised for the measurement of diameters at various heights of the bole. The simplest
ones are not necessarily the least accurate and much depends on the way they are handled
and consequently on the training of the inventory staff*

When the reference height is relatively email (measurements of dlmseters above

buttresses) a simple graduated rule at the top of a metallic or wooden pole held olose to
the bole and facing the observer, with the aero in coincidence with one side of the bole,
can be considered a suitable instrument in the case of measurement by diameter classes.
As the observer in a tropical forest cannot stand very far from the tree (generally not
more than 15 metres), parallax error is not negligible: graduations of the rule have to
be corrected to take the parallax error for a given horizontal distance between the
observer and the tree into account* This simple device has been used in many forest
inventories in West Africa* The Finnish parabolic oaliper and other simple instruments
based on the principle of the Blltmore stick oan also be used for measurement of
diameters above buttresses or at relatively small heights*

Instruments for measuring diameters at any height are more sophisticated and more
expensive* In order of increasing sophistication the more interesting ones are:

- the Wheeler pentaprism which consists of a metallic rail with one fixed and one
sliding prism, the distance between the two prisms being equal to the measured
diameter; the only drawback of this simple and precise instrument is that the
rail must be as long as the maximum diameter to be measured, which does not
permit its use for the biggest trees in tropical inventories;

- instruments such as the "Dlatromb" wherein two indices are pat in optical
coincidence with the edges of the stem and are at a fixed horizontal distance
from the eye of the observer (the indices are fitted on a bar sliding on a rod,
the distance between the eye at the end of the rod and this bar being such that
its horizontal projection is constant);

- the multi-purpose Bltterlioh relasoepe which permits simultaneous measurement

of the height and of the diameter of the stem at this height 4 the wide-scale
relasoope has proved to be useful for measurement of upper stem diaaeter in
tropical forests, although visibility in these forests is not always sufficient
to permit accurate measurements;

- the Barr and Stroud dendrometer has magnifying optics, uses split-image
coincidence and Is a very precis* but rather expensive instrwent*

243*' Height measurements

Height measurements in a forest inventory are madet

a) on all trees (or a fraction of tarns.) of the sample in connection with measurements
of tipper stem diameters when the volimie of the standing trees of the sample is
estimated by geometric formulas using these measurement data;
 - 10/1

b) on all trees (or a fraction of them) of the sample in addition to the d.b.h.
when the volume of the trees of the sample is estimated through volume
equations tuning diameter and height ao independent parameters;

c) on a relatively small subsample of trees in connection with Measurements of

upper stem diameters, the volume of these trees being estimated by geometric
formulas, and the volume, d.b.h*, and height data being utilized for the
essment of volume equations by recession analysis

Height measurements y like measurementr of upper stem diameters, are indirect

measurements made by optical instruments (contrary to d.b*h* which is generally a direct
and rapid measurement) and are consequently t impconsuming. When selecting the method of
volume estimation in a forest inventory, it should be carefully checked whether these
additional measurements on all the trees of the sample (or on a significant part of them)
In many inventories of mixed it has been found that it
are justified* tropical hardwoods
is more efficient to use "local volume tables" ( U by species with measurement of d.b.h.
only on all the trees of the sample than to use volume equations with d.b.h. and height
a independent parameters with measurement of d.b.h. and height on all the trees of the
sample: the increase in precision ie small in relation to the consequent increase in the
enumeration cost (see below paragraph 34?.?).

As fcr the measurement of upper stem diameters f many instruments exist and the less
expensive and sophisticated ones may be particularly useful in certain conditions and
especially in some tropical forest inventories* Direct measurement with telescopic poles
is possible only fcr small heights - for greater heights (total height, bole height f
merchantable height of usual trees) indirect measurement by hypsometers has to be used*
The Christen faypscmeter is a very cheap and handy instrument which is recommended for
tropical forest inventories when the precision required is not very high* Other well-
known hypsometers much as the Blume-Leiss or Haga hypeometers are more precise but
measurements are mere time-consuming and sometimes require too great a distance between
the observer and the tree in tropical forests with a thick undergrowth* Clinometer*,
such as the Suunto clinometer t can be used also but the heights cannot be read directly
and have to be calculated from the elopes measured with the instrument. The Bitterlich
relascope ie used also for height measurements generally in connection with upper stem
diameter measurements.

243*3 Bark measurements

111 diameter*, at breast height and on the upper stem, are measurements over bark on
the standing trees, but merchantable volume* do not include the volume of the bark. The
problem is to relate the volumes under bark of the tree with the diameters over bark and
possibly also with measurement* of the bark*

If the volum* of the standing trees of the sample is estimated without the help of
volume equations, volume* under bark have to be estimated from volumes over bark using a
conversion factor calculated from bark measurement at breast height*

If volume equations are used and established from a sample of felled trees, the
best solution is to estimate the volumes under bark of these sample trees and to relate
these volumes through regression analysis to the d.b.h* over bark (and possible height
and other upper stem diameter* over bark)* In this csss bark thickness is measured with
a rule on the face of the logs.

The most common bark gmuges for measuring bark thickness at breast height on standing
trees have been designed in Sweden. Risks of underestimation and over estimation in bark
thickness measurements are mwerous find much care and training is necessary.

(1) Volume equations whereii? volume is a function of d.b.h. only*

3 Volume estimation

31 Definition of volumes

It is of the utmost importance to define the volumes referred to in an inventory

clearly and objectively* There are unfortunately too many inventory documents wherein
it ie not clearly stated which is the minimum db.h* of the corresponding trees f which
portions of the trees are considered (are branches included ? what is the minimum top
diameter ?), whether or not the volume of bark is included, whether the volume* are gross
volumes or exclude defective parts, which the criteria are for excluding parts as defective ,
whether the "net" or "utilizable" volumes correspond to what is likely to be extracted or
do not exclude the logging losses, etc. It is easily understandable that the definition
by an adjective such as "gross", "net" or "industrial" is generally rot sufficient, and
must be completed by a clear explanation of the adjective itself*

The drawbacks of a lack of definition or of an incomplete or imprecise definition

are serious . For instance when the inventory results are used for a feasibility study,
misinterpretation of the concept of volume jeopardizes the whole study* In particular,
this happens when the incompletely defined inventory volumes are considered as extractable
volumes although they are "gross" cr "net" volumes including logging losses.

The following definitions were included in the first edition of this Manual as
standard definitions for all FAO forest inventories:

Pro SB volume; the volume of a specified portion of a tree without bark^

or deduction for defects; when used, the term should be qualified
by a word or statement specifying the portion of the tree to which
it refers, e.g. total tree gross volume (2).

Net volume; the volume of a specified portion of a tree without bark and with
deductions made for defects or unusable material; the term should alro be
qualified according to the portion of the tree to which it refers,

total volume: the volume included in the main stem cf & tree; for deliquescent-
formed trees, up to the crown point; for excurrent-forwed trees up to the
tip of the tree,

Branch volume; for excurrent-formed trees, the volume of all branches; for
deliquescent-formed trees, the volume above the crown point (and any
branches which may occur below).

Industrial volume; the potentially usable net volwe of round wood, without
deduction for losses due to utili station standards of logging and manufacturing
processes; it equals the sum of log volumes plus other usable volume.

Log volume; the net volume of a tree considered suitable for veneer logs, sawlogs,
sleeper logs, piling and poles; this volume may also be used for pulpwood,
chipboard or other industrial use*

Other usable voluge; the net volume of a tree not suitable for purposes listed
under log volume but usable for posts, pulpwood, chipboard and for other
industrial

(1) Qross volumes include bark volume in many inventories.

(2) Gross volumes as well as all other volumes refer to a minimum d.b.h. of the
relevant trees and also to a minimum diameter at the small end of the stew
branches .
 - 106 -

In addition to the above definitions the following remarks can be made*

a) It ia suggested that the adjective "commercial" when added to the terms "industrial",
"log" or "other usable volume" or to specified portions of the volumes, distinguishes
volume which can be economically removed under given conditions.

b) To estimate commercial or merchantable volumes it is necessary to know the

nerohantability specifications for a given species or group of species at the tame of the
inventory (i.e. for known and specific wood products and situation of the wood market), for
a given situation of logging and for the inventoried area or for a neighbouring and similar
area* To assess the merchantable volumes it is generally necessary to complete the
inventory operation itself by measurements in the logging units of the extracted logs v and
of the losses in order to determine the ratio of the extracted merchantable volumes to the
inventory volumes* This is particularly true in inventories of mixed tropical forests for
which no tradition of exploitation exists and changes in market and infrastructure are
rapid and significant. "Commercial" volumes are inventory results which can be given only
when such "utilisation studies" (or "recovery studies" or "harvestings-intensity studies")
are thoroughly carried out. If this is not the case the adjective "commercial" is
inappropriate and must not be used.

c) In no oase should the expression "net volumes" be used or understood as synomymous

with "commercial volumes". The only quality assessment (or grading) of the standing trees
of the sample cannot provide a satisfactory estimate of the merchantable volumes as these
latter are determined also by the internal defects, the logging damage (splits, broken
trees | etc....) arid other factors which cannot be precisely predicted from observation of
the trees of the sample. Drilling of trees at breast height for rot determination (see
paragraph 42 below) ie useful for the assessment of decay occurrence iibich provides
additional information on the quality of the standing trees, but still other parameters
must be ascertained for a valid assessment of commercial volumes.

d) 7*ieactual usefulness of the assessment of "net volumes" thus does not appear
fundamental, as these are generally different from the commercial volumes which are in most
inventories among the most important results to obtain. This is all the more true as some
subjectivity and personal bias is almost unavoidable in the assessment of net volumes. In
many tropical forest inventories procedures such as the following can be adopted!

1. use the inventory measurements themselves excluding basic quality data, to

determine gross standing volumes objectively;

2. use the basic quality data to classify (or "stratify") these gross standing
volumes by quality classes (or grades), the basic quality data being obtained
by observation of external defects, by decay information at breast height on
standing trees and possibly also by detailed quality analysis of a subsample
of felled trees;

3. perform a recovery study by making a survey of the output in a sample of

logging units, a partial recovery factor being determined for each of the
former graded or quality classes of the standing volumes.

32 Volume units

Volume estimates can be expressed either in cubic unit 4 showing the total contents of
a tree (or portion specified), or in terms of the quantity of the ultimate products which
can be processed from the tree or section. The North American board foot unit is such an
etad product volume. The use of an estimated end product volume has the advantage of a
direct assessment of final products expected, and thus facilitates evaluation. However,
 - 107 -

this type of measurement unit has significant shortcoming in that it shows the estimated
output in terms of only one product, sawn board* or lumber, and for sawlogs this volume
depends on the amount of defects in the log, the skill of the sawyer, the thickness of the
saws used f the thickness of the lumber sawn and the amount of taper in the log. There is
indeed a general implicit agreement in tropical forest inventories to limit the volume
estimation at the exit of the logging unit or sometimes to the yard of the wood mills. In
view of the above and other limitations it seems logical to follow this latter custom and
to abandon the UBS of end product volume units. Additional results in board foot units
will be given only when it is considered essential.

As for linear measurements (see above paragraph 241) the use of the metric system
(cubic metres as volume units) is highly recommended. In countries where the British
system is still in use, it may be advisable to produce results in cubic feet and also in
cubic metres: many of these countries intend to shift to the metric system in the future,
and international statistics on wood resources could be facilitated. The use of the metric
system and of both systems in countries using the British system has been recommended as a
standard procedure in all forest inventories carried out by PAO.

33 Classification of volume estimation techniques

values of the known parameters measured and recorded in the sample are used to
estimate the means and totals of these parameters or of other related characteristics in
the inventoried forest area and in parts of it. Mean and total volumes are among the
main estimates to obtain from the inventory.

All observation made in the sampling units for volume estimation are observations on
trees. In the firt*t step of the volume estimation process these observations are used
either for assessing the volumes of the trees in the sampling units (and consequently the
corresponding volumes of the stand in the sampling units) or for directly estimating the
stand characteristics in the sampling unite. A Rimplr example of this latter case is the
sweep with a Bitterlich relaacope made at a point in a plantation, the sampling unit being
the point and th stand characteristic being its basal area at this point; the elementary
observations are made on the trees around this point, but the individual tree volumes are
not (arri cannot) be *Rt:lroat^ from this p"mp}

The estimation of the individual tree volumes in the first case or the direct
*otimation of the stand volumes in *aoh warpHng unit in the second case can be made

- either ty formulas (such aa geometric formulae for volumes of simple solids) and
graphic procedure*, provided that miffici^nt detailed measurements are made;
- or by quantitative relationships between the fe</
'
measured parameters and the
volumes, such as volume equations established by regression analysis.

Iii view of the above two considerations a general classification of volume estimation

techniques could be the following:

1. volume estimation techniques on a tree basis

11. without "quantitative relationships"
1?. with "quantitative relationships"

2. volume estimation techniques on a stand basis

Z1. without "quantitative relationships"
?2. with "quantitative relationships"

the expression "quantitative relati one hips" being synonymous with relations derived from

(1) In most cases for estimation of individual tree volumes the measured parameters are
limited to diameter at breast height and height (total, or merchantable, etc.).
 - 108 -

other trees or stands^ ' and of restricted application: the volume equations, for instance,
are generally valid for a given region, for a given Bite quality, for a given species or
group of species, sometimes for a given range of d.b.h. y eto. (The principles of this
classification are taken from "Planning a forest inventory" by B. Husch, but the
classification is presented in a different order.)

The following development will be restricted to the problems of volume estimation on

a tree basis as this type of volume estimation is more frequently used in forest inventory,
especially in the tropics where most of the stands are mixed and uneven.

34 Volume estimation on a tree basis

341Geometric formulas applied to standing or felled trees. The total volume of

a stem (or its volume up to a minimum diameter) is expressed by the well-
known formula:

V - f.g.h

where f f g and h stand respectively for the form factor, the basal area at breast height
(or above buttresses) and the total height of the stem (from the stump or from the
buttresses) or up to minimum diameter.

If it ie generally possible to determine the basal area at breast height and the
height effectively and with reasonable accuracy, this is not the case for the form factor,
since the form of the stem is not easy to characterise and may not be uniform all along the
stem. In addition the volumes to be estimated may include the volume of the branches, and
in this case the ratio of the volume to the product basal area x height is even more
difficult to estimate.

The most obvious way of computing the volume(8) of a single tree is to divide it
(virtually with an optical device if it is standing) in sections of equal or unequal length
(logs or "frustums" of the stem, branches in certain cases), to estimate by geometric
formulas the volume of these individual parts and then to add the volumes so obtained* The
ends
geometric formulas giving the volume of a section from its length and diameters at the
and/or at mid-length cannot give completely accurate results as a log is never identical
to one of the simplest corresponding geometric solids. But their accuracy is generally
sufficient with respect to the measurement errors and it is all the more true as the number
of sections is higher.

The usual geometric formulas are:

- Smalian's formula: v - -+g-

&b u T
L

- Huber's formula:

- Newton's formula: v -

where: v is the volume of the log

TT 2
"
g. is the cross-sectional area at base
&b 4 ^
g is the cross-sectional area at middle: g^ - T
d^
is the cross-sectional area at top: g.
- d
jf
-j u
L is the length of the log

(1) However these trees or stands must belong in principle to the inventoried population.
 - 10* -

(The volumes considered can be volumes over bark or volumes under bark
with, in thie latter oaae f the diameters d, , d , d being measured
under bark.) b m u

These three formulas are valid if the log can be assimilated to a frustum of
paraboloid of revolution. Newton's formula is also applicable if the log is approximat'ily
a frustum of cone or of neloid; for these two latter oases, Huber's formula gives an
underestimation of the volume of the log whereas Smalian's formula gives an over-en timat ion
which is twice as big as the underestimation of Huber s formula.
f

The use of Newton B formula is highly recommended when the logs are long, i*e when
1

the number of logs per tree is small, and of course when the whole stem is not divided into
sections. Thin means that to estimate the volume of a stem up to a given minjmum top
diameter there must be at least two diameter measurements - at stump height or above the
but tresses | and at mid-point between stump and the minimum top diameter - and the
measurement of the height from the stump to the minimum top diameter.

There are three usual wayB of dividing a stem:

- division in sections of equal length from the stump, the last section often being
assimilated to a simple fraction of this unit length; this procedure is the most
common in forest inventory;

- division in sections such as the difference between their diameters at large and
small ends is constant: the lengths of the sections are generally different unless
the tree form is conic;

- division in sections of unequal length using the measurements at breast, height:

in a European national forest inventory the following diameters are measured on
all trees of the sample: stump diameter y diameter at breast height (1.30 m),
diameter at 2.60 m, diameter at merchantable height (or minimum diameter) and
diameter at mid-point between 2.60 m and the merchantable height; then the tree
volume is computed by applying Newton's formula to each of the two log*, the
lower log with a length of approximately 2.6 metres and the upper one from 2.60
metres to the minimum top diameter.

instruments used for measuring upper-stem diameter and height for estimation of
volumes of standing trees were presented briefly in paragraph ?43 Th* Bitterlich
relascopes used with a tripod prove to be very useful dendrometers for the simultaneous
measurement of diameters and lengths of the various sections. Volumes of standing sample
trees for the assessment of volume equations in tropical forest inventories have ofter been
computed from relascope measurements on standing trees, when felling of trees for thie
purpose was not possible,

342 Volume equations

342.1 Introduction

The expression "volume equations" is used rather than the more oomaion one "volume
tables" in order to indicate that only equations (or formulas) giving the volume of a tre*
("dependent variate") as a function of the characteristics (maicly diameter at breast
height and height) and derived from statistical regression analysis, are dealt with in
this paragraph.

The expression Volume tables" includes indeed not only the volume tables drawn from
thl type of equation! but also those which wre (and sometimes still are) established by
graphical or "sewd-graphJcal" methods (like Keen and Page's method) which should no longer
be used, as they include some personal bias and do not allow for a sound estimation of the
statist* oa) error.
 ~ 110

The advantage of the use of volume equations in forest inventory it evident; they
permit 9 fro* detailed measurements on a lift! ted number of trees judiciously selected within
the forested area "sample trees"), the objeotive estimation of the volume of a much larger
1

number of tree* in the sampling units and finally the estimation of the total and mean
volumes within the Inrentoried area*

The construction of volume equations is a difficult, time-consuming and expensive

task. Once it has been deoidad in a forest inventory to estimate the volume of the trees
of the sample by use of volume aquations (and not by additional measurements on the trees)
there is a tendency to reduce the work related to the assessment of volume equations
without valid reasons. For instance, it may happen that volume equations are already
available and that the need for their adjustment or for new ones is not felt* It is very
important, however, to oheok oare fully whether the equations are valid, in general, and
applicable in the particular case of the inventory to be oarried out* In paragraph 342*4
some indications will be given on the statistical validity of volume equations* is for
their application to a given inventory many questions have to be answered positively: is
the fora of the trees in the inventoried eone similar to that of the trees of another region
used for the volume equations (in the oase of volume equations with diameter and height as
independent variates) ? Is the average height per diameter class of a given species the
in the two regions (in the case of "local* volume tables for one speoies with only d*b.h* as
independent variate) ? Is the species composition in the inventoried zone similar to that
of the region of the sample trees used for the volume tables, when the volume equation is
used for the trees whatever the speoies ? If the available "local" volume tables are for
groups of species, is the representation of each species in the group identical in the
inventoried scne and in the area for which the volume tables were initially constructed ?
etc***

To oheok the applicability of the available volume equations objectively, a field

test is strongly recommended* This may be done by directly determining the volume of a
number of Judiciously selected trees from the forested area to be inventoried using the
same measurement standards and methods as employed for these volume equations* The volumes
of these trees are also read from the volume equations being tested* The percentage of
divergence of the individual actual volumes from those of the equations is then computed*
These percentages may be tabulated and averaged for the same species or group of species
by diameter (and possibly also height) classes. If they reveal a significant difference
in volume, the equations may be considered not applicable or requiring adjustment before
use*

in inappropriate set of volume equations or* more generally, bad volume equations,
whether constructed for this inventory or not, may significantly reduce the reliability
of the results. This is all the more true as the sampling error due to the inventory
sampling itself is low: in this oase a large component of the total error (including the
Measurement errors) may unexpectedly be due to the use of the volume equations*

The methods used for volume estimation by volume equation are statistical methods*
Some of them - method of least squares, principally, but also some "non-parametric" multi-
variate analysis methods and automatic classification - consist purely in mathematical
computations and are used for the assessment of the equations* Regression analysis on
the contrary serves for the application of the equation to the trees in the sampling units
and oannot be applied without restrictive conditions. In other words it ie always possible
to establish by a mathematical method a relation of a statistical nature between the volume
and some characteristics of sample trees, but if we want to know tbe statistical error
corresponding to the application of this equation to a tree or group of treea in th*
population, some conditions must be fulfilled which we shall deal with in paragraph M?4
 - 111 -

342.2 Basic types of volume equations

The most significant measured characteristic to which the volume(s) of a tree is (are)
related ie its d.b.h. Therefore all volume equation will have d.b.h. and possibly
exponents and functions of d.b.h. as independent variates. Very often one height - either
total height or bole height or any other specified height - and exponents or functions of
this height are added to d.b.h. Finally other characteristics are introduced in the Boat
elaborated ones, such as bark thickness or some form quotients (i.e. ratios between two
diameters at different heights of the stem).

Thus volume equations can bs grouped into the tree following categories:

a) "local volume tables which relate the volume(e) of a tree only to d.b.h. or exponents
11

and functions of d.b.h. (such as basal area). The two most common ones are:

v -

log v - b + Klog d or v - B d

b b
(with B }0 ore whether the logarithms are decimal or naperian)

with: d - diameter at breaat height (d.b.h.)

a f
a f
b ,
b. constants

b) "standard" volume equations which include as independent variates d.b.h. and a given
height, and functions of these two character in tics. The most common "standard" volume
equation is:
o
v - c * c^d^h
o 1

with d - diameter at breast height (d.b.h.)

h - total height, or bole haiffht, or any merchantable or other height

c and o, - oonstanta
o 1

c) more elaborated volume equations specially developed for research purposes or for
national forest surveys and whioh include d.b.h., one or several heights and other
characteristics*

Whioh of these types of volume equations to choose for a given forest inventory is a
difficult question. In many tropical forest inventories the choice is often between
"local" volume equations for every epecies or group of species and possibly by part of the
inventoried area and "standard" volume equations for all species or for groups of species.
There is no general solution, and each case must be studied carefully taking into
consideration the total precision required, the costs involved in both solutions and
logistic problems. However, regarding this problem, two remarks are worth mentioning:

a) It has been found in some mixed tropical hardwoods (e.g. semi-deciduous and
evergreen forests in west Africa) that, for a given species or possibly group of species,
the avsrag* bole height (from top of buttress to crown point) in every diameter class
above a certain diameter (say 50 centimetres) was nearly constant. Since the most
important volumes are generally those of trees above this diameter (which is, in many
the minimal diameter of exploitability) the inclusion of bole height in the
countries,
volume equation does not appear essential, provided that the number of trees in the sampling
 - 112 -

units of the corresponding species in each diameter class is large enough and that the
variability within each diameter class around this constant mean is relatively small*

b) The use of a "standard" table implies the measurement of a height on every tree
within the sampling units, or at least on a part of them* The additional cost is
significant since the time required for enumeration may be double or even longer* On the
other hand the reduction of the sampling error due to the use of a "standard" table instead
of a "local" table may be insignificant in relation to the total error which includes the
sampling error of the sampling design and the measurement errors. In such cases the use
of local volume tables proves to be more efficient since the increase of the sampling error
is more than compensated by the reduction of the cost*

342*3 Combined types of volume equations

In some recent inventories, the volume equation approach is not as simple as those
described above and consists in a more or less complicated combination of "standard" and
"local" volume equations* The main purpose of this type of procedure is to avoid the
measurement of the second characteristic (height) on every tree of the inventory sample
while trying at the same time to reduce the sampling error due to the use of volume
equations. Many procedures of this type can be contemplated, and it is impossible to
list and describe them all* The following two examples give a good illustration of this
type of approach*

1st example* Volume equation of the FAO/UMDP nationwide forest inventory in West
Malaysia.

a) Description of the measurements made:

- d.b.h. over bark (D ) and height class on all the 41 * 200 trees of the inventory
sample;
- d.b.h. (D,) 9 diameter at 16 feet (Di6)t diameter at 32 feet (1*32), diameter at
crown point (D c ) (all diameters over bark) and bole height (H) on 16,600 trees
out of 41,200 of the inventory sample (subsanple) for determination by geometric
formulas of an estimate V g of the volume of the standing tree;
- detailed measurements for accurate determination of the volume of the bole (Vjp)
on 720 felled trees out of 16,600*

b) Procedure followed:

- assessment of one equation (by method of least squares) between Vy and Vif Y fl

being calculated from the measurements DS, Dl6i &32t D c *nd H made on the standing
trees, from the 720 felled trees:
2
Yp a + b V B + c Vg ;

- application of this latter equation to the 16,600 trees for determination of their
(estimated) YF;
- assessment of local volume equations (by method of least squares) by species and
height class from the 16,600 trees of the type:
2
YF - a '
+ D -I- D (i - 1 to total number of species x height classes);
bj t oj g
-
application orf these local volume equations to the 41 f 200 trees of the inventory
sample for the estimation of the volumes in each sampling unit*
 2nd example. Volume equations of the PAO/UNDP inventory in the Auree Mountains in
Algeria (by the SpanJ ab consulting firm 0*T.lO*

a) Description of the measurement? made:

- in the sampling unite of each compartment: measurement of diameter at br**st

height over bark, D , on all tre**i and m*anurwnt of total beigfct K on u
sample of trees;
~ detailed measurements for ace-viral"* J^ l;*r;n >.atj on i* The voluvne (V) of n
Bubea/nple of trees (336 PJ^TTUB hAlegfrnai^ tr*es and l Cedrug trees). '

b) Procedure followed:

- assessment of one "standard" volume equation (by the method of least squares) for
each of the two species from the small subseirple of trees: y

2~ 2
V - a 4 bH + cD + dD H for Pi"-HUB
- -
hale*Fpa OF is
S 6 ~
'

?
V - a 1
4 b 1
D H for Cedrus
R

- application of these "standard" volume e,piationa to the subsampl** of trees of each

compartment for the estimation of their volume V 3 ;

assessment of five equations (by method of least squares) between II and D ,

each
equation correspo nding to one stratum (oite quality):
n n tt
x
- 1 to c
H - a. 4 b 11 4 c. P vi
V - ^)f
i 1 B 1 B

- allocation of *n^h compartment to one of the five ntrata with the UB* of the
height measxirem^ntB in the Bubsa^pJe of the
- aseessment of one "local volume equaticr* per species ri
ar>d jver Btra*cuw
V B and D B from the subsample of trees of e,oh *7>eeie ftnd of all aorrpftrtmentB of
the stratum considered:

IM Ml D 4c llt t
mwber
V . a
J
4 b
J J
D (j^lto total of r<peoies i
Rtratum rlasaes)

-
applioation of these "local" volume mqia*.t:on tr. 1L the trees of tbe inventory
sample for the fRtimation of the vc2umes jn tbe uairpHn/r

type of approach illustrated bj the above two samples cue be oonsidered generally
as more efficient than a simple conventional approach using only one type of volosie
equations. How*ver f their relative efficiency oanrtot he. aB***ed precisely as the ocwplei
procedure makes the computation of the sampling error too difficult if not impossible. It
is of the vtost importance, as for the simple approach, to a*o*rtair fulfilment of the
conditions required for the use of statistical methods at each step oT tV prooedtire.

142.4 Statistical aspects of the volume equjiti on approach

In manr forest inventories using the volume equation approach for the estimation of
the volue(s) of the trees, insufficient consideration is given t:> thn otmtisti^al aspects p
ajxi more precisely to the statistical reqnair*ments f which mist be fulfilled to get reMnble
estimates of the volumes and of the sampling error* of these eetimntet* Up to now there
is no evidence in forest literature that all these aepeots have been treated, jtnd there Is
an urgent need for a clarification o^ all the related probliws. %* following
oonni derations are no more than general indications sn<3
 - 11/1
-

Selection of the sample trees

Geographic distribution of the plots fron fhich the sample trees are selected should
preferably bs based on an objective sampling design either at random or systematic, or on
a stratified random or stratified systematic design (by forest type, for instance).
Concentration of sample felled trees in a very limited number of locations within the
forest area is often decided upon for evident logistic and economic reasons. However, it
must be realised, even if the location of these large plots is determined objectively -
which should always be the case - that the more concentrated the sample, the larger is
likely to be the sampling error* Again for logistic and economic reasons the sample trees
are often selected close to roads or to openings; in such cases, a bias in volume
estimation may occur as the growing conditions of the trees are different.

Regarding the distribution per species and/or per diameter class the problem is more
complicated, A representative distribution of the sample trees (i.e. proportional to the
ties occurrence) among the species and/or the diameter class appears intuitively the surest
method, if not the best. However, some considerations may preclude the use of a
representative sample; for instance, such a sample often results in a large uncertainty in
the volume estimates of the biggest trees since these are little represented in the Copulation.
It may be more efficient to solve this problem by proportioning the number of sample trees
per diameter class to the relative volumes of each class in the whole population (which is
in many oases close to a Heyman'e allocation). In each location, the sample trees can be
selected by relascope in such a way as to approach more closely this optimum allocation.
The rejection of trees from the sample because such characteristics as crooks, leaning
trees, forked trees, etc, must be done on a sound basis. A general principle is that no
sample tree should be rejected if the corresponding volume equation is applicable to any
tree of the same standards within the forested area. The only exception to this rule
should bs when the tree cannot be accurately measured (for instance when creepers and
foliaceous epiphytic vegetation make diameter measurements unreliable).

There is no universal answer as to the number of sample trees to be selected for any
one volume equation* The larger the msaber the more precise will be the estimate, but
it also depends on many other factors such as the diameter and the height range of ths
trees, the sise of the area, the number of forest types, the variation of site factors,
etc. "Local" volume equations in limited areas for a given epeoiee or group of species
have been constructed with 100, or even less, sample treea.

Problems of regression analysis

a) Once the sample trees have been selected and measured, and their volumes computed,
the soatter diagram with the volumes, V, on the y-axis and the main independent variates,
d.b,h., or (d.b.h. )?i H, on the x-axis is drawn, A first visual observation will give
an idea of the strength and form of the correlation, and will show any need for more
sample trees in certain olasses of the independent variate and the possible abnormalities
which may result from measurement or computation errors,

The observation of the soatter diagram will also show if it has to be split in two
or more parts corresponding to different portions of the range of values of the dependent
variates. If this is the case there will be a need for two or more volume equations.

b) If the variation of the volume within a class of the independent variate innrefcsee
with this variate, it will be necessary:

- either to transform the dependent and independent variates (for instance by the
use of a logarithmic transformation) in order to get a more satisfactory scatter
diagram;
- or to weight all the variates in each class of the dependent variates by a quantity
proportional to the inverse of the standard deviation of th* volumes in this class ;

- or to uo* both prooed'jres together*

 The Constance of the variance of the dependent variate within each class of the
independent variates is a prerequisite of regression analysis. If this condition is not
fulfilled, the strength of the relationship between the volume and the independent variates
cannot be evaluated and the sampling error cannot be determined. Generally, the variance
of the volume in each class of the independent variateo can be said to be proportional to
2 (or to a
(dbh) larger exponent of dbh) or to (dbh)^ x H (or to a larger exponent of this
product). The regression between V and dbh for instance ("local" volisis equation) of the
model: V - a + b(dbh) ? , with a variance proportional to (dbh)4 is transformed in the
following regression:

V
2
f.lbh)

(wherein the variates have been weighted by ^

which is proportional to the
2
(dbh)
inverse of the standard deviation of the volume).

c) Then a model of volume equation has to be chosen. The choice of the most suitable
model is facilitated by observation of the scatter diagram of the weighted and/or
transformed variates if the curve :

/- ._V 1_ 7
^
?
(dbh) fonh/

of th* above example Phows * parabolic tendency toward the x-axie, it may b* convenient to
fi model;

-I- b +
(dbh)' (dbh)

j
lrk.h w 4
1. 1 result finally In th* following volume equation:
2
V - a -f c' v
dbh) + b(dhh)

Another way of constructing th< model is to draw up a list of the most significant
(or ^ritre forced) *rpre*slons of the independent variatep, for instance:

1 -
t
-i- , oonatam, TTT t ** (expressions, respectively weighted
2 dbh Hbh
(dbh) (dbh)

2
,
ofi constant, dbh, (dbh)', (dbh)H, (dbh) H, H)

to applj the ao-called multiple stepwise regression analysis. Computer programmes

for this method exist which proTlde set of volume equations, the first one including
only one eTprfission of the independent variate

(for instance - + b)
?
(dbh) (dbh)

and saoh of the subsequsnt onss differing from the former by ths inclusion of an additional
variats. Tb* last one is the most complete and pored ss f but ths fmin in precision from
the prsedin# on* maj not b signifioant. Interpretation f ths oomputsr outputs and
lection of *.h inputs in tb^se prourrmsmies re<iuire the assistance of a statistician*
" OM o<njtrjun^n rar> b* added to the model in these programmes, such as a fixed value of
conntant In the
 - 116 -

With the development of automatic data processing, the use of sighting (including
the research of the beet weight) and of stepwise regression analysis should be practised
to an increasing extent* These techniques allow for a sound estimation of the
statistical errors in the use of voluae equations .

Statistical errors in the use of volume equations

The validity of the adjustment by regression analysis can be tested by the value of
the multiple correlation coefficient which is equal to the correlation coefficient when
the dependent variate (volume) is a function of only one expression of the independent
2
variates, for instance: V - a + b(dbh) . This adjustment will become more valid as this
coefficient approaches 1 (maximum value of the coefficient in the case of a perfect
adjustment),

Estimation of the variance and standard error of the mean value of the volume given
by the equation for given values of the independent variates, and estimation of the standard
error of the application of this value to the trees of the inventory sample with the same
characteristics, are given by rather sophisticated formulas, and advice from a statistician
would have to be sought for detailed information. The problem posed by the combination of
the error with the error due to the inventory sampling is a difficult question which does
not seem to have been very much dealt with in the forestry literature. Here too, the
advice of a specialist must be sought as it is necessary to know what is the incidence of
the volume equations on the total error of the volume estimates*

Grouping of the species

In certain cases, and especially in inventories of mixed tropical hardwoods, it is

not possible to assess separate volume equations for each different species. Various
solutions are possible:

- assess separate volume equations for the most important species and apply a
common volume equation for the remaining ones;

- assess separate volume equations for the most important species and group each
of the remaining species with one of these species, by a comparative study either
of the fonr factor (in the rase of "standard" volume tables) or of the diameter-
height diagram (in the case of "local" volume tables);

-
group the species by homogeneous classes in order to get more sample trees for
each volume equation*

This last grouping can be done more or less objective!/ by comparison of scatter
diagrams corresponding to different species, by a oovariance analysis by groups of two
species (when the number of species ie small), or by a statistical method of automatic
classification (multi variate analysis plus cluster analysis). The two latter methods are
the best and the most comprehensive and objective ones, but they require a sufficient
expertise and the availability of appropriate computer programmes. Here too, the
assistance of a statistician is necessary.

Volume estimation by taper functions. "Paper functions are assessed by

regression analysis and generally give, for a species or a group of species in
a given area, the ratio of the diameter of the stem at a given height to the diameter at
breast height (or abovb the buttresses), expressed as * function of thie given height.
Within the sane species or group of species there may be different taper functions according
to the diameter clauses.
 - 117 -

Once the taper functions are determined, the volume of a tree in a

sampling unit,
whose d.b.h. and bole height (or other height) has been measured can be
easily calculated
by dividing ths whole stem into frustums of equal length and adding their volumes. ftie
volumes of the individual frustums are calculated by geometric formulas
using the diameters
given by the taper function. The rather lengthy computation of th volume of each tree of
the sample is not a problem when an electronic computer is available*

Hie remarks made in paragraph 342.4 on the statistical

aspects of volume equations
are equally valid here since regression analysis is used in both cases. In particular,
the problem of the grouping of species (and/or of diameter classes within the same
species)
can be solved in the same way. The solution adopted should be the most objective
possible, taking into account costs and other limiting factors.

344 Selection of the most suitable volume estimation technique. There is no

general answer as to the best volume estimation technique to be used in a
forest inventory. Volume estimations by additional meaeureme.itu on the trees of the
sample or by use of regression analysis (volume equations taper functions) both have
,

advantages and shortcomings which are summarized below.

The main advantage of volume estimation from additional measurements on standing

trees is that there is no sampling error on the estimation of the volume, the only sampling
error being that arising from the sampling design of the inventory. There are some draw-
backs; the main one is the additional cost due to a longer enumeration work, another one
being the additional measurement errors which may be important and difficult to assess
precisely.

Methods based on volume equations and taper functions introduce an additional

component to the sampling error which may be significant with respect to the other
components (other sampling and measurement errors). There may also be some bias at one or
several of the steps of the construction of volume equations or of taper functions. Ttoe

advantages are Important: in comparison with the direct method of volume estimation it
generally costs considerably lees; the volume equations can be used for further
inventories and other purposes and will remain as a tool for the f ores ttr which can be
refined later on; the measurement data collected for the construction of volume equations,
which are generally more detailed than in the other method, may serve as a basic material
for further mensuration studies. These advantages are of particular interest in many
developing countries of the tropical world where forest mensuration is a new research field and
where mensuration data and volume tables are lacking.

4 Quality assessment

41 Preliminary remarks on quality assessment

411 Definition of o^iality assessment in a forest inventory. The previous Bub-

chapter dealt with the methods used in a forest inventory to timate
objectively gross volumes of the trees within the sample (total gross volume, or gross
volume of the bole, or up to a minimum diameter), these volumes being afterwards expanded
to the whole inventoried stands according to the sampling design used. Each tree or part
of a tree has, in addition to its dimension*, other characteristics such as shape, aspect,
defect and decay, which make its wood more or less useful and valuable for given purposes.
The classification, quotation or quantification of these characteristics and the recording
- are thus
and processing of the corresponding data - which constitute quality assessment
results with more meaningful and detailed
necessary to provide the users of the inventory
information.
 412 Assessment of "net volumes" and usefulness of this concept (with special
reference to forest inventory of mixed tropical hardwoods)* Very often
external defects of the standing trees are classified in two main categories - those which
are acceptable with regard to a given type of utilization of the wood and those which make
the corresponding portion of the tree unusable for this utilization* The volume of the
of the tree with this class of defect - "defective
parts"
- and sometimes of the
parts
whole tree itself are subtracted from the gross volume and the reduced volume is called the
"net volume" (see paragraph 31 ) In many inventory reports estimates of "net sawnwood
'

volumeV or "LJt pl~*o v d vcl-jm-s A*-e thob obtained L^ uedu^tion of portions claoo-.Tj.ed as
1

defective for sawnwood or plywood production from the corresponding gross volumes.
Although this is a common procedure, the validity and usefulness of the concept of "net
volvwe" are doubtful in the case of inventory of mixed tropical hardwoods, for the reasons
given below*

There in always a subjective component in quality assessment f which must be reduced to

a minimum by a good classification and quotation of the defects* This is particularly
important in tropical forestry since the inventory people are most often on the management
or ownership side and not on the wood utilization side* Net sawnwood volumes given by the
inventory may not correspond to the volumes effectively used for aawnwood*

Even if we consider as negligible the bias coming from the viewpoint of the inventory
people, there are other reasons why the so-called net volumes are significantly different
from the volume of usable material, among which can be quoted:

- quality assessment of standing trees most often does not take into consideration
the inner defects which are not visible and which cannot be safely predicted and
precisely estimated from external observation; even for the external defects
quality assessment may be invalidated by difficulty in evaluating the defects on the
upper part of the tree;

- in most of the mixed tropical hardwoods there is an incomplete knowledge of site

quality and environmental factors which may have a bearing on wood quality; this
situation is different from that of many European forests which have long been
known in detail oy the foresters and the loggers: assessment of a precise
percentage of logging losses and rejected wood IB much easier in these stands;

- there are rapid and significant changes in the conditions of utilization of

tropical hardwoods (changes in the domestic and international markets with lower
qualities being accepted when demand is high, modifications in the local wood
processing facilities, mechanization and changes in the sizes and practices of
the logging units, etc.): thus the specifications of the defective volumes used
for assessing the net volumes may not be applicable one or two years afterwards,
and differences between "net volumes" and usable material may increase even more.

For all these reasons - and as already mentioned in paragraph 31 - assessment of

"net volumes" does not appear to be of very great help in mixed tropical hardwoods, at
least in economic terms* Nevertheless this must not be considered as the only purpose
of quality assessment which has other useful applications in forest inventory*

413 Other applications of quality assessment in forest inventory*

Comparison of the qualities of the same stand at intervale

Quality assessment ie useful for controlling (or "monitoring") the growing stock.
In repeated inventories of a given forested area, quality assessment makes possible a study
of the evolution of the stands with regard to the quality characteristics and the defect
occurrence. For such a comparative study, and provided that the quality classification
of the standing volumes is meaningful, the equivalence between the volume of the
 - 11Q -

"non-defective" classes and the usable material is not required; what is important is
that, for instance, a better "score" in the second inventory as compared to the first
inventory means a real improvement with regard to quality characteristic* aoi defect
occurrence*

Comparison of qualities for several stands

Quality assessment is also useful for comparing two or more inventoried areas* Here
too, there is no need for an equivalence between the volumes of the "non-defective" classes
and the usable material, but the quality indicators must be relevant*

Stratification of the standing volume for further studies

The stratification of the standingv>lumes into different quality classes may be

useful in estimations related to quality* Biis is particularly so in the case of the
assessment of a "recovery factor" (or "conversion factor" or "utilization factor"), i*e.
the ratio of the usable (or "ertractable", or "commercial") volumes to the corresponding
inventory volumes (gross or possibly net volumes). The conversion factors of the various
classes are in general significantly different, if the quality classification is meaningful,
and the estimation of a global conversion factor is thus improved by the quality
classification*

Some indications of the assessment of recovery factors are given below in subchapter
5 "Recovery studies".

42 Methods of quality assessment

For the sake of presentation, these methods can be divided into methods of assessment
of external characteristics and defects and methods of assessment of inner defects*
External characteristics and defects can be recorded on standing trees as well as on felled
trees whereas the estimation of inner defects on standing trees can be made only through
partial and somewhat imprecise observations.

In a particular forest inventory there may be quality assessment (external and

possibly internal characteristics and defects) on all trees of the sample (or of a sub-
sample) and quality assessment on a small subsample of felled trees which may be the sample
trees for the volume equations. Regression analysis between the two corresponding
categories of data can then be used to improve the quality data obtained from the
observation of the standing trees. Ttoere may be additional quality indicators recorded
on the felled trees, but for the common ones the system of quotation or quantification and
recording must be the same to permit the comparison of the data from standing trees and
from felled trees.

421 Assessment of external characteristics and defects

421*1 Recording unit of assessment

For assessment of external charact eristics and defects on standing trees as well as
on felled trees two basic approaches may be adopted:

1) the section concept: the stem is divided into a number of sections, each of an
absolute, relative or variable length, the quality of each section being assessed
separately;

2) the tree concept: the stem is classified according to a series of selected

quality or defect classes.
 - 120 -

421*11 The oil on concept

a) Sections of absolute length

The length of the sections in a stem remains constant according to chosen

specifications depending on local requirements (see Figure V-2) A standard log length
of five Metre*, or about 16 feet, ie often used ae the length of a section* The gross
volume of each section is assessed either by means of taper functions, or on the basis of
a percentage of the gross volvsie. of the total stem derived from the sample trees used for
the formulation of the volume equations, or possibly by measuring the diameter at the mid-
point of the section* Care is necessary to allocate defects to the appropriate sections*

V -2

Quality Con at ant Sections

Log No

@ B

J A*.
 - 121 -

b) Sections of relative length

Quality assessment by dividing the stem into sections of relative length is

illustrated in Figure V-3 In this method the number of sections in the stem remains
constant, while the length of eaoh section varies in accordance with the length of the
stem. Section limits are determined by using a section ruler. This instrument may be
used independently of the distance between the tree and the observer* With increasing
number of sections, quality assessment under this concept becomes correspondingly more
detailed and time consuming and also less reliable as the allocation of defects to the
appropriate sections becomes more difficult. Figure V-3 shows a division into three
sections, a number which might permit good progress of work without losing too much of the
indicative nature of the results. The gross volume of eaoh stem section is estimated
from taper functions or from the data of the sample trees used for the volume equations or
by direct measurement of length and mid-diameter of each section, this last solution being
time cons umi ng.

Figure V-3

Quality Assessment - Relative Sections

cog

B
 - 122 -

c) Section* of variable length

Tha diviaion of the stem la governad by tha location of tha important da facts, tha
purpoaa being to separate tha logs of matarial deemed usable from tha dafaotiva portions t
This mathod ia illustratad in Figure V-4 Tha limit of each aaction would have to ba
datarminad by aya using judgamsnt. To datarmina volumes, tha length! of tha section*
would have to ba datarminad and tapar function* used* If these functions are not
available it would ba necessary, in addition, to measure and- or mid-diametars of tha
section*.

This procedure is accurate and practical for quality studies based on measurements
of felled traaa but ita practicality ia doubtful if appliad to quality assessment of
standing traaa; it is time-consuming, difficult and entails a graat element of subjectivity.
Moreover assessment of defective parts and usable parts cannot generally ba dona with
precision, at least in tha oasa of mixed tropical hardwoods as has bean shown in paragraph
412.

Figure V-4

Assessment ~ Variable Sections.

Duality

n
t t

t t
 - 123 -

421.12 The tree concept

Figure V-5 illustrates the manner in which the tree concept of quality assessment
applies the quality of only a specified lower portion of the stem to classify the whole
stem. The specified length of this chosen portion of the lower stem does not usually
exceed six to eight metres* Usually, only one specified length is used for each inventory
though f in some instances, it may be necessary to specify lengths according to species.
In buttressed trees the specified length is applied to the trunk above the buttress*
Volume is presented as total volume of the tree. In general, the lower portion of the
stem contains the greater part of the total volume and is the part of the tree of greatest
potential value.

Ibe advantages of assessing quality by the tree concept, as compared with the
sectional concept, are:

- quality class specifications may be determined most easily on the lower

portion of the stem;
- the possibilities for subjective bias are reduced;
- the method is more rapid;
- the results of the quality assessment studies can also be presented in the form
of stand tables;
- volume estimation is simpler.

Figure V-5

ABpccamnt - Tree Aporaleal

,
Th quality of Lof No. 1 nay d fro *
of pacified quality claa*. The trt mm whol*
takva it quality from th quality classification of
Lof No. 1.
 421.2 Quotations

Once the gross volume of a "recording unit" (tree f or section of a tree) bar
-^ *-.
determined, it has to be classified in one of the various quality classes. Th<* i

number of quality classes must be decided carefully, taking into account the purport of
the quality assessment exercise, the increased risks of subjectivity inherent in a -i-t^i i-
classification and the relatively sparse information obtained from a very broad
classification*

The allocation of the volume of a recording unit to a given quality class can be
done in a global way, direotly from a global appraisal of the various characteristics and
defects, or in an analytical way by evaluating separately the different types of defect
and subsequently regrouping the corresponding quotations for the final allocation of the
unit to one of the quality classes.

An example of the first type of assessment is often ueed in simple classifications

on a tree concept, such as "trees rejected" ajid "trees not rejected", or "sound",
"defective" or "dubious". The second type has been used in West Africa irith classification
on the basis of sections of relative length, the defects being regrouped in three
categories: those related to the shape of the section (bends, crooks, oval cross-section,
to), its healthiness (broken branches, rotten knots, etc.) and the aspect of its wood
(frain, scars, twist, sound knoti, etc.); each section is ranked from one to five within
each of the three defect categories and then these rankings are combined to form a range
of five global quotation* for the whole section.

1 thorough analytical appraisal is, however, to be preferred to the global apfproach,

which ie more liable to subjective bias. The number of rankings of the quotation must
be large enough to allow for detailed evaluation and at the same time small enough to make
the exercise as reliable as possible. Five steps appear to be an acceptable compromise,

42U3 Some recommendations for the assessment of external characteristics and

defects in forest invento^r

If quality assessment is to be made on a subsample of trees from the whole inventory

sample, the selection of these trees must be made on a purely objective basis. Any
yitematic procedure such as "one subplot in every sampling unit" or "one sampling unit
every n^ eenpling unit", or "each i tree within each sampling unit" is reconunended
insofar a* it does not entail any other bias.

Subjective bias in quality assessment can be reduced by several means, such as a

very clear and comprehensive "quotation key in which every defect is mentioned with
11

poeaible degrees of gravity and corresponding quotations. It is preferable to ask the

field crewe to record single defects and their degrees of gravity separately and to apply
the corresponding quotation* only in the office, especially when these quotations result
from a combination of several defects* To make the quotations more reliable and more
hoogeneou0 it it important to restrict the number of people in charge of the quality
aeeeenent work and to train them very thoroughly* The work should be done by one mar
per ore*, preferably the orew leader*

The use of binoculars is reoomnended for the observation of the upper sections of
the standing sterna when the section procedure it ueed* Ihe observer must turn all around
the trees, and at a certain distance, in order to make a full inspection; this latter
requirement is fundamental but is not always fulfilled in inventories of mixed tropical
hardwood* where access ie often difficult*
 - 12 5 -

422 Assessment of internal defects

422.1 Assessment of internal defecta on felled trees

The internal defecta are evaluated on stump cross-sections or breast height cross-
sections and possibly also on other cross-sections. Volume of rotten parts and
consequently of sound volume can thus be estimated precisely* !Bae most important problem
is how to quote, combine and enter in the classification other internal defects such as
insect damage, eccentricity, B tains, splits, etc. Here, too, a satisfactory quotation
key has to be devised whereby the various defects and their different degrees of gravity
can be combined. Applicability of existing grading scales is worth testing.

422.2 Assessment of decay in standing trees

In recent years, electrical or mechanical drilling instruments have been used to

determine and measure decay at breast height in standing trees of mixed tropical forests.
Results of these observations have shown that decay occurrence at breast height can be
detected with sufficient precision (relatively few trees with a very small decay or an
eccentric decay are wrongly classified as trees without decay at breast height).
to be
However, precise estimation of the extent of the decay at breast height appears not
eccentric and irregular shape of the trees and of
feasible, at least in certain forests;
the decay, prevent this. Moreover, in some cases there is only a very loose correlation
between diameter of the decay at breast height and its length, so that a direct and
reliable estimate of the volume of the decay and consequently of the sound volumes appears
of
impossible. Finally, problems are caused by the occurrence of high buttresses and
upper decay.
are very
However, observations showing the mere presence of decay at breast weight
useful. an indicator of decay occurrence in a given forested area, which
They provide
allows for comparison with other forests. They also permit a stratification on a tree
basis of the standing volume into quality classes (such ae volumes "without decay at
breast height", "with small decay at breast height" and "heavily decayed at breast height")
which can be used for further studies and especially for the assessment of recovery
factors.

*) Recovery studies

51 Principle

forest inventory are gross volumes up to

Regarding volumes, the usual results of a
a given diameter, expressed per diameter claaa and group of diaaeter claaaea, per apeciea,
unit.
and per quality olaaa, for eaoh inventory unit or atratun within thia inventory
For aoe important end-ueee, auch aa aawnwood and plywood, theae grosa voluwea repreaent
have to
an overeatiaation of the actual ueable voluwea, i.e. of economic value and which
and aubnational economic
be coMidered in prei treatment atudiea and alao in national
and planning prograasjee. Therefore it ia neceaaary to obtain an eatiaate par apeciea
volumea which are
or group of apeoiea and par inventory unit of the ratio between the
the inventory , taking into con-
likely to be extracted and the groaa voluaea given by of accessibility
aideration the prevailing oonditiona of foreat management, of logging,
utiliaation and of domestic and international marketa. Too
and infrastructure, of wood
oountriee becauae the
often there ia atrong oritioia. of inventory documents in tropical
ia not of direct use for the
information produced by theae expenaive operations
oonosjio and planning purpoaea which ahould be one of their
main objectivee. la it la
an eetimate of the imable wood, eoonomieta and induatrialiete are
neoeeeaxy to get "net volumee") which
the ao-called
applying an arbitrary ratio to the groaa volumee (or
ia aeldovj valid.
 - 126 -

Recovery factors are also useful for the owner and the manager of the inventoried
forests. Fro* the Market prices of the usable wood he can determine, by the use of the
results of the recovery study the price at which he can "sell" the standing volumes,
taking into consideration the logging and transport costs* More generally, the informa-
tion provided toy a recovery study is useful for the owner or the manager of the forest
when dealing with the logging contractors*

The question has been raised as to whether the estimation of these ratios should
be provided by the inventory people or whether it should be left to others auoh as
economists or logging specialists.

It seems obvious, however, that the people responsible for the inventory work
are the beet acquainted with the figures they produce, with their validity and their
applicability, and that they should therefore be involved in the estimation of these
ratios, either by themselves or preferably in cooperation with logging and other
specialists.

Studies related to the estimation of these ratios are called here "recovery studies , 11

but their principle is the same as for those called in other documents "forest utilization"
studies or "harvesting intensity" studies,

52 Related pro b leas

In addition to the estimation of the e it rac table volumes (or of the parts of the
gross volumes which are left in forests), it is also interesting to know the percentage
of each commercial grade of the extractable volumes, as the respective value and
possible utilisation of the various grades are different. A recovery study should aim
at estimating these percentages, although in some oases grading is not as objective as
it could be,

The eztractable volumes to be considered are those which come to the mill yards if
the wood is to be processed locally, or to the harbours if it is to be exported. In the
case of domestic processing the conversion factor of the mill is applied to the extract--
able volumes given by the inventory in order to relate the output of the mill to the
existing volumes in a given forest area.

Due to the changing conditions in wood utilisation, in markets, in infrastructure

and possibly also in forest management and logging, the results of a recovery study are
valid for a relatively short period and must be up-dated when significant changes occur..
Even within a given region or country +,he recovery factors vary according to accessibi-
lity and to the possible utilisation of the forests; it is of utmost importance to
distinguish for Instance, between the case of a forest used for pulp production and that
of a forest used for sawnwood and plywood production, or between the case of local
processing and the case of log export.

In mixed tropical hardwood forests, recovery studies must be performed by epecies

or groups of species of the same utilisation destination* In most oases a recovery
factor covering all species or very large groups of speoiee is meaningless as there may
be tremendous differences in recovery and in value per volume unit within these groups.

53 deueral procedure

531 Main steps of a recovery study

In mixed tropical hardooods the recovery studies can be split in three main parts:

~ estimation of the selection of the standing trees; if for certain epeciee all
trees which can be logged according to the management regulations are effectively f*ile*} ?
for many o there thers is a selection of the standing trees, for instance those which
 - 12? -

assumed to be rotten inside after tapping the bole with a humor are generally not felled;

- estimation of the
percentage of treee felled but not used due to splitting,
breaking, significant decay not detected previously, bad logging conditions, etc.;

-estimation per inventory quality class of the percentage of rejected volumes and
possibly of percentages of the different commercial grades and further determination of
the global recovery factors and grade percentages for the whole gross stock.

532 Implementation

If logging units do not exist in the inventoried area or in a similar neighbouring

area, an experimental logging operation may be carried out. This solution is expensive
but allows for an easier planning of the study. In this case, however, the logging is
not necessarily included within the commercial process and the output of the study may be
optimum recovery factors.

If a survey of logging units is carried out, the sample of these units will have
to be carefully constituted and stratified according tc the main utilization destination
of the extracted wood (e.g. local processing or log experts, pulp or other industrial
uses) and possibly also according to the logging system used and to the accessibility if
this varies significantly from one part of the inventoried area to the other. Die most
evident procedure within a logging unit seems to be:
- before logging, the full enumeration and numbering of all trees of the studied
species above the minimum acceptable diameter, with estimation of their gross volumes
and with quality assessment, in an area which is soon to be logged;
- after logging, the counting of abandoned standing and felled treer and the
measurement of the extracted logs and logging losses and wast*.

Recording and processing of the data (with possible use of multiple regression
analysis between inventory quality classes and commercial grades) has to be* designed
carefully. The total cost of a consistent recovery study is not negligible, and its
programming and budgeting must be included in the planning stage of the inventory.

533 Example

The following graph illustrates the principle and the procedure for a recovery
study. It is adapted from a similar graph drawn for a recovery study performed within
the framework of an PAO/UNDP inventory In hill dipterooarp forests of Sarawak.
 - 128 -

Reasons for not

felling trees
oaw Log Length
considered too short
Diameter considered
too small
No demand for this
species
bi nicer

Too much decay

expected
Dangerous to fell
Damaged during the
felling oi other trees
Likely to get damaged
during felling
Likely to fall in
unmanageable position
Poisonous
Other reasons

Reasons for leaving

felled material in
the Forest

Log not found I

Log irretrievable I

Log considered buttress ]

Log considered form

defect

Log considered too

heavy fluting
Log considered too
heavy internal decay
VOLUME J ELLED A LEFT
Log damaged during
felling
Log considered too
short

Log diameter considered

too email

Misjudged by felling

Other reasons
 - 129 -

6 Accessibility studies

61 Introduction

Subchaptere 1 to 5 deal with the production of volume estimates in a forest inventory

from data collected in the field. In addition to the quantity and quality of the standing
and extractable volumes, knowledge of the physical and socio-economic parameters which have
a bearing on the forest management and harvesting costs is necessary. These parameters
define the accessibility of the standing volumes and future yields of the inventoried
forests.

Assessment of the accessibility parameters should be carried out within the

framework of the inventory for many reasons:

- the inventory results must be as complete and meaningful as possible and must
therefore include data on accessibility in order to give information to the users
of the inventory which can be directly and immediately utilized by them;

- the inventory crews are collecting data in a large number of units objectively
located over the whole inventoried area, and are thus in a better position than
most of the users to record also accessibility data, particularly as it may be
appropriate for further computations to have accessibility data and volume data
recorded entirely or partly in the same sampling units;

- it is much more practical and economical to have accessibility data recorded by

the inventory crews than to perform another survey designed only for collecting
this additional information.

Information on accessibility is, like that on volume estimates, needed at different

scales and levels of precision depending on the type of results required* The method
used for assessing accessibility will vary accordingly. However a standardized approach,
which unfortunately does not exist at present, would greatly facilitate the comparative
use and interpretation of accessibility results

For a given level of the forest inventory (world, regional, national, subnational,
preinvestment or local level) and a given level of precision, the accessibility study can
be defined as providing an answer to the following question (Nilsson,

"How much wood specified by Eipecies, dimensions and qualities can be made available
at tentative markets (mill sites) within alternative cost limits per unit volume ?"

Given this definition, poesibly complemented by the introduction of the time factor,
the method used for assessing accessibility in a given forest inventory nust solve the
following problems:

1) Which parameters need to be collected ?

2) How can these parameters be quantified or classified ?

3) Which quantitative relationships have to be used to estimate the management and

extraction costs from these accessibility parameters ?

It is clear from the above considerations that a permanent dialogue between the
inventory and the logging specialists must be maintained in order to design a suitable
method of accessibility assessment within a given forest inventory. This is particularly
true for the problems of the selection and quantification of the accessibility parameters,
whereas the third problem lies essentially within the competence of the logging specialist.
In the following paragraphs we will deal only with the selection and quantification of
accessibility
 - 130 -

62 Selection of accessibility parameters

621 The cost of logging, transport and forest roads varies according to an almost
infinite number of factors. Some of these have a strong influence, whereas
other are of only minor importance and can therefore be disregarded for the purpose of
assessing accessibility* The practical approach is, then, to establish the correlation
between the selected factors and the logging and transport costs. This can now be made
by logging specialists in various ways, the most practical being to set up mathematical
models for each work operation quantifying the influence on the cost of the selected
factors. In such model building the logging specialist should endeavour to make use of
parameters which are already measured in the inventory end should be rather cautious in
introducing additional parameters which might be difficult and costly to measure by the
inventory crews. The advantage of the use of mathematical models is that they can be
included in the computer programmes of the inventory, thus enabling the accessibility
of the inventoried forests to be assessed simultaneously with the inventory results on
the quantity and quality of the resource.

The accessibility parameters can be classified mainly into two groups, namely the
forest condition parameters and the socio-economic parameters.

The values of these parameters for a given inventoried forest or for a part of it
have to be combined with production data in the mathematical models for the assessment of
the management and harvesting costs. These production data have to be determined for the
various forest work operations (mainly logging, transport, road construction and maintenance)
from work studies and production and cost control schemes.

Other classification can also be applied to accessibility parameters. It IB

important, for instance, to distinguish between those of a stochastic nature (e.g. volumes
by diameter classes) which are estimated by field sampling and those of deterministic
nature (e.g. the distance of an inventoried stand from mill site or market). The latter
are usually not measured in the field. Whether a parameter is to be assessed by a sampling
procedure or by a deterministic method depends on practical factors.

Like other parameters (see paragraph 221 of Chapter 3) accessibility parameters can
be given continuous or discrete values through direct measurements or can be assigned to
a given class (especially in the case of descriptive or qualitative parameters).

622 The forest condition parameters. The forest condition parameters are
physical parameters related to the trees, to the stands and to the land, to
the climate and to the location (with regard to existing or possible access to the forest)*

Most of the tree and stand parameters are generally estimated by field sampling
procedures and many physical data on the land can also be collected during the forest
inventory. It is important that the forest inventory data be collected in such a way as
to enable the results to be applied to exploitation area units. Their precision is
dependent on the sampling intensity of the forest inventory in relation to the size of
the exploitation unit. Data obtained from low sampling intensity over a large area can
give accurate estimates for the whole area, but do not generally give sufficiently
accurate data for the assessment of logging costs for parts of that area. Apart from
the sampling intensity the distribution of the sample has also to be considered* A
systematic one-stage sampling design is often considered useful as it allows for the
possibility of changing the limits of the exploitation units.

622.1 Tree and stand parameters

These are estimated in all forest inventories and are mainly numbers of stems and
the corresponding gross or net volumes per diameter (and possibly height) classes, and per
species or groups of species. Rie mean and total estimates must be given with sufficient
precision per exploitation units if they are to be used for assessing accessibility.
 - 131 -

They include also parameters of quality (oee eubchapter 4) which are to be estimated to
determine the percentage per species (or groups of speciee) and per size classes of the
trees which vail not be felled due to their visible defects and which also serve for
estimating the extractable volumes by meanc of a recovery study.

The results of the recovery study (see subchapter

5) aleo provide physical data
essential for the accessibility study. Since they are collected in logging units, it is
possible to find out which volumes are to be considered in each logging operation (felling
skidding and truck hauling).

Other tree and stand parameters are important with regard to

accessibility but are
difficult to quantify, such as the thickness of the undergrowth or the branchiness of the
trees.

6??.^ Terrain and soil parameters

The main characteristics of terrain and soil which have an influence on forest
accessibility are considered in the "IUFRO proposal for international system of terrain
classification and in the PAD note "Tentative checklist for describing and quantifying
11

g transport and reading conditions . These aret

11

a) terrain parameters
- terrain pattern which can be described by geomorpho logical features such as
relief amplitudes, regular or irregular drainage pattern, stream frequencies,
rock outcrops, etc.;
- area of a single physiographic feature ("terrain unit") defined by its relief
amplitude, its length and width;
"~
width and depth of rivers and creeks ;

- slopes (longer than 50 metres);

- ground roughness defined by the frequency of obstacles with a relief amplitude
exceeding 0.3 metres;
- microtopography defined by the frequency of gullies and slopes shorter than
50 metres;
" area flooded with an indication of the duration and frequency of the floods;

b) soil parameters
- bearing capacity in moist state for off-the-road vehicles;
- suitability for road construction and maintenance which takes into account the
depth and the texture of the soil and the stoniness of the surface (1);
- susceptibility to erosion which is important when considering in particular the
road construction and maintenance costs.

622.3 Climatic parameters

The climatic parameters useful for assessing logging costs are related to the
following itemst
- rainfall, the corresponding characteristics being the amount and the distribution
of annual rainfall and the maximum rainfall per hour and per day;

(1) Deposit of gravel and quarry sites should also be recorded by the inventory
crews outside the sampling unite, especially when such deposits and sites are
rare.
 - 132 -

- water discharge and its variation;

- temperatures ;

- insolation, with the important characteristic of the length of sunny periods

during the rainy season.

622.4 Location and access parameters

The main location and access parameters useful for accessibility assenprnent ar:
distance and transport routes to wood-using centres (mills, cities, densely
populated areas, export ports);
- data concerning the existing and potential transport network inside the forests,
such as the network density and the distance of cross-country transport per typen,
standards and capacity classes (estimation of the road length can be made through
the use of a rectangular transect system and the application of the "needle
problem" method).

623 The cocioeconomic parameters. Hies* parameters are absolute and are
determined by the national economy and laws. They are needed to establish copt
unit time of production factors in the cost formulas. Three groups of data can be
distinguished here: labour, equipment, and other data.

Labour: Information on labour availability and skill, payments, fringe benefits,

transport or camp facilities, number of working day* per year, number of effective
working hours per day, etc.

Equipment: Operating costs based on purchase prices, customs duties, cost of fuel,
facilities for and cost of servicing, etc.

Other data: Such as laws on transport, railway tariffs, exploitation restrictions,

composition of the enterprise, etc.

63 Quantification and/or classification of accessibility parameters

631 Parameters relevant to felling. For felling, in addition to the most

important tree character is tics, other parameters such as crown length or any
other indicator on branch! nees and on de limbing difficulty is useful. Slope and other
terrain conditions are also important. A combined quantification of the form and quality
of the trees and of the terrain and vegetation on the felling site is shown in the
following table. The coefficient should reflect how production is influenced in relation
to "normal conditions.
11

Correction factor reflecting working conditions (other than normal)

 - 133 -

632 Parameters related to transport and road construction

632.1 following classifications of slope and ground roughness are extracted

TTic
from the IUPRO "Proposal for International System of Terrain Classification"
and are given as examples but not as models to be adopted in all cases.

632.11 Slo

Slope is described by the gradient (a) f

the length of the slope (b) and the aspect
of the slope (c).

Slopes with lengths of more than 50 metres are defined ae major slopes. Slopes with
lengths of 10-5 metres are defined as minor elopes, for example lesser valleys and
hillocks. Variations of less than 10 metres in length are defined as "ground roughness".

(a) Gradient

The average gradient in recorded in per cent. The following classes are adopted:

1 . o - 5 $ .

Intermediate classes may be used if necessary.

The slope IB stated in terms of its relation to the access road as

1. positive (uphill to reach the access road)
ff M "
2. negative (downhill )

The description may be completed by recording the maximum gradient.

(b) Length of the slope

With regard to major slopes the length of slope is recorded in metres in the
following classes:
Minor slopes
1. 10 - 50 m

Major slopes
2. 50 - 100 m
3. 100 - 200 m
4. 200 - 400 m
5. 400 - 600 m
6. 600 - 800 m
7. 800 - 100C m
8. 10OO - 1200 m

3. 1200 - I^OO m
10. > 15OO m
Minor slopes may
Classes with wider intervals may be obtained by amalgamation.
occur also as parts of a major slop*.
 - 134 -

(c) Aspect of the slope

The aspect of the slope is recorded as the compass bearing in the following way:

632.12 Ground roughness

The roughness of the ground IB described independently of the slope on the basis of
the occurrence of obstacles, i.e. local surface
variations, boulders, rocks, stumps, holes,
hollows, etc. of more than 30 cm height or depth.

The roughnesc is described by the general occurrence of obstacles When

(a).
required a description of the occurrence of obstacles by types and size is undertaken (b).

(a) General occurrence of obstacles

The following classes are adopted:

1. very smooth ground; average distance between obstacles > 5.0 m; minimum
distance between obstacles >
3.0;
2. smooth ground; average distance between obstacles > 5.0 m; minimum distance
between obstacles <T 30;
3 uneven ground; average distance between obstacles 5.0 - 3.0 m;
4. very uneven ground; average distance between obstacles < 3.0 m;
5 ground with boulders and scree;
6. ground with precipices and clefts.

3y distance between obstacles the unencumbered distance between the limits of the
obstacles in understood.

(b) Occurrence of obstacles by typs and sizes

 - 135 -

The occurrence of types of obstacles should be recorded in absolute

figures. With
to the class "precipices and clefts" the
height/depth class may be Bub-divided.
Problems related to slope assessment.

The slopes considered the most important with regard to

logging costs are slopes on
distances longer than 50 metres ("major" slopes). In the field inventory the slopes can
be measured generally only on shorter distances (10 to 20 metres in Mean
many cases).
slopes longer than 50 metres can be estimated on a sampling basis on aerial photographs or
on good large-scale (smaller than 1/25 t OOO) topographic maps. The comparison between
measurement of slopes on the ground and on the photographs or maps shows that the occurrence
of steep slopes is overestimated on the ground, since a certain proportion of steep
slopes
in the terrain are on Bhort distances.

Whatever the gradient classification adopted, 50$ (practical maximum slope for a
tractor at present) and possibly 70$ (occurrence of difficult road construction problems)
should be used ae limits in the gradient classification.

A recommended classification or lengths of slopes with a

gradient larger than 50$ is:
SO metres - 300 metres

300 metres - 700 metres

^> JOO metres

633 Another example of ground roughness classification

A further method of recording ground roughness together with some other factors
difficult to measure IB shown in the following table. Each parameter is subjectively
assessed in R. difficu 11^ class ranging from 1 to 5 where difficulty class 3 could be regarded
as "normal". Ihe points of the difficulty parameters are added and the sum is then an
erprescion on their aggregate influence on extraction to be ueed in the cost formula for
this operation.

The sum of points out of table gives E through the following conversion table:
 - 136 -

CHtPOR VI

DiTi RKORDIK AMD PROCSSIK IM rOHPT

 - 137 -

CHAPTER VI

DATA RECORDING AND PROCESSING IN FOREST INVENTORY

Introduction

Data recording and data processing can be viewed as major linka between the planning
and completion of a forest inventory. Tfce basic inventory data, either gathered in the
field, from photo interpretation or from other sources, cannot be processed without having
first been recorded, edited and condensed. In addition, the data processing prooedur*
mu*t be adapted to the specific requirements and design of the inventory itself. It is,
therefore, essential that the treatment of the data be considered as an integral part of
forest inventory from the very beginning planning stages .

Data recording and processing should be given careful consideration, particularly

during the initial planning stage of a forest inventory, since the means of data
processing (e.g. available facilities and personnel for computation) or the cost may have
considerable impact on the design, intensity, and timing of the entire inventory. Within
the framework of information required and money and time available, the treatment of the
data has to be considered as a limiting factor which directly influences the choice of
inventory method. Broadly speaking, the simpler the inventory design, the lower the cost
of data handling and the less time consuming the job. Although data processing serves
more as a tool of forest inventory than as a determining factor, its influence on the
realisation of the inventory should not be underestimated.

Since the whole field of data recording and processing is very complex and
particularly fast-developing, the contents of this chapter can only be considered as
general guidelines for choosing and implementing the appropriate data recording and
processing methods for a given inventory. The sub-chapter "Data Recording* is restricted
1

to general requirements for data recording and the means and methods of recording forest
inventory information from various sources . In the sub-chapter "Data Processing" the
different steps and types of data processing are discussed and some practical aspectn vrith
respect to tropical zones are given*

2 Data recording

21 General requirement a

Before undertaking the design of any "recording document"^ \ attention should be

drawn to the specific data to be recorded. Keeping in mind the simple rule that results
to be obtained by means of data processing oan never be better or more reliable than the
basic input data itself, a determined effort should be made to improve the quality of the
data to be recorded* One of the main requirements for reliable data is its obj+otivity
and comparability. Since inventory data will genarally be recorded by different people
urrier different conditions, any possibility of human influence on the data itself should
of the people
be omitted. Recording instructions should not allow any parsonal judgement
of data should be preferred whenever possible
concerned. Objective quantitative assessment
In cases where coding cannot be avoided (e.g. classification of crown cover into density
the coding criteria.
classes) the recording instructions should define very clearly
Although a refined stratification for a given variable will permit precise claasifi cation,
th* reoultB may be lepo reliable than those obtained from measured data, because coding
*ntail personal judgement.

will be
M) A "recording document" is any prepared form on which original data
or automatically by
e.g. ttJLly sheet*. Recording can be done either manually,
special devices (see para. ?3)
 - 138 -

To minimise human bias, whenever possible the parameter* to be reoorded should be

urable parameter* (continuous variable)* All discrete variable* ahould be reoorded
using a broad classification (or coding) *ytem, permitting allocation* in a given claee
a* objectively a* possible. The aim should be to get homogeneous data which could be
compared between one "record unif'O) and another.

The data should be reoorded aa they are measured and no pro oee sing should be done
during the recording stage. If, for example, mean diameter of individual tree* 1* needed
for future calculation, and is assessed by two or more direct measurements, only direct
measurement* *hould be reoorded and the mean diameter calculated later. Similarly, when
measuring diameter* it is better to record actual measurement* rather than the corresponding
diameter classes. This eliminates the possibility of errors occurring in transferring from
measured diameters to classes. Broadly speaking, the simpler the parameters and the
method of recording, the more accurate the corresponding results .

One of the most important tasks in this respect is the design of simple and clearly
arranged recording documents, easy to handle in the field and easy to fill in. The design
of the recording document should aim primarily at facilitating the recording work, and to
a lesser extent the later use of this document for process ing. This is particularly *o
since recording work is usually much more expensive and difficult than processing work*
Thus, practicability appears to be another requirement of general nature to be considered
in data recording.

Before any data recording begins, it is necessary to become acquainted with the
condition* under which the work ie to be executed and with the actual data itself First
of all 9 it is very useful to adapt recording to the personnel in charge of it* For
instance, a very low level of training or experience may call for a simple way of recording*
An example would be use of special tapes for diameter measurement, on which only the
diameter class can be read* Environmental conditions, such as weather, terrain or thick
undergrowth, can also affect recording in the field* In wet weather, for example, it may
be necessary to uee waterproof paper for the field sheets.

22 Specific requirement*

221 With relation to the type oof data. A practical ola**ification of type* of
type
forest inventory data 1*
is as follows:

- information on the area by interpretation of remote sensing imagery;

- information on the field plot* including site, soil and accessibility data;

- tree tally by species and diameter classes;

- measurements on standing or felled trees;

- information on quality of standing timber*

(1) For definition of a record unit (RU) see paragraph 412 of Chapter III*
 - 139 -

Experience from many inventories in tropical zones has led to the general acceptance
of four basic type* of recordbin each RU in the field:

For the Record Unit:

i: this form include* the identification of the HU and gives all information
on the plot itself as opposed to tree tally, as for instance:

For the tree tally: generally only one of the three following types is need in
connection with type 1 in a given RU:

fope 2; the number of trees IB indicated in a matrix^ in which the rowe

correspond to the species and the columns to the diameter classes:

Type 3: the tree parameters are recorded within a matrix, in which every tree
forms a row and the tree parsmetere the columns.

Tree No. Tree parameters

Tfrp* 4 the trees are enumerated by species and diameter continuously:

Species code dbh Species code dbh Species dbh etc.

Col. (1) (2) (3) (4) (5) (6)

In a given inventory, different combinations of recording may occur. For example,

in a cluster sampling, tree-by-tree recording (method 3) nay be required in only every
fourth HU, whereas in the remaining 3 RU's the trees will only be recorded either by method
2 or 4 above, depending on the procedure selected for data processing*

as for data recording in the office, more sophisticated methods can be used and the
recording system can be aimed more at facilitating data processing since the risk of errors
is very much reduced.

(1) The word matrix stands for any type of two or several entry table
 - uo -

222 With relation to data processing. The method of data processing to be used
will affect the recording only to a relatively *mall extent. In any case, the
data fhould be written simply arrl clearly. If th data are to he key punched for electron <-

processing (EDP), it IP more convenient to arrange the data in the punching seruence although
this is not essential ainoe it i generally easy for the krey punch operator to adapt to a
punching plan. In the caae of manual processing, Hnta ehould be arranged in *uch a w*y
"< re-
that computing can start immediately without waiting for time consuming sorting
arranging of the data. In this case the fourth method of tree tally should be avoided
BI nee a continuous recording of species code and diameter requiree a time consumi ^
re-
arrangement of the trees into species/diameter classes and complicates the work. In the
oaae of qualitative data, the coding key for all parameters should be determined prior to
record! ng

P3 Main kinds of data recording

of data recording differ also with the methods of data processing. The most
common ype of recording, which is applicable to manual data processing or EDP, uses
1

handwritten field documents. The plot information sheet (i.e. T^rpe 1) should alwaye
ccrtair. tn following sections:

1
. ption of work:
Deiieri date, field crew, starting and finishing date
of pl^t survey.

Identification of plot: inventory unit, sampling unit, plot or point

within the sampling unit, etc,

3. Plot criteria: stand description, environment, accessibility, road network.

These groups of data should be clearly defined when designing the sheet, to facilitate
control and identification in the office. In the case of further punching for EDP f
numbers can be indicated under the field of the corresponding item on the field
form. The total amount of information recorded in the field form may exceed the 80
columns of a punch card (see also paragraph 311 )

transfer of data from field documents onto punch cards is a significant component
in the total cost of data processing. Therefore, there is an increasing use of other than
documents which avoid this transfer.

(a; Port-a-punoh cards

Data are recorded on pre- punched 40 column cards with a special portable device.
They are later read directly by the computer and stored on disks or tapes. This method,
usd successfully in the Swedish National Forest Inventory, requires skilled personnel in
the field. Some trials of this method in tropical forests under difficult conditions were
not very successful.

(b) Mark-* en ing documents and cards

The basic procedure of this method is to mark with a special derioe (e.g. pencil or
narking ink) within predetermined positions on the cards or sheets the data to be recorded
which Mill be interpreted later by an electronic optical reader. The adrantage of wing
sheets rather than cards is that more data can be recorded on sheets 9 whereas most cards
are were or lass limited to a certain amount of data* Trial of this method in one tropical
forest inventory was nut vary satisfactory duo to the large sise of the HU's and the
consequently largo number of species and troes which called for eereral mark- sens ing
documents par RU. However, in some cases, such as inventories of plantations with easy
environmental conditions, these method* should be successful.
 - 141 -

( c ) Handwritten documents for optical reader

A relatively new method allows handwritten docninientB to be read

by a multifunction
optical reader. Reading is performed by an electronic luminous spot scanning system,
through which each character ie investigated and, in most oases, identified by the "reading
ray tracing in a spiral motion the contours of each character for identification*
41

This method, used very successfully in forest inventories in West Germany, requires
a computer centre equipped with an optical reader, to read the data and store it on disks
or tapes. On the other hand, the characters recorded in the field must be written very
carefully, following certain standard rules of writing, and must be placed exactly in
their appropriate location on the recording sheet. Since misinterpretation occurs
relatively often, the stored data must be scrutinized thoroughly by special editing
computer routines, specifically designed to locate such errors. Thus the main drawback
of this method for use in tropical forest inventory is its sophistication.

Other methods which may be applicable for recording forest inventory data are:
machine-typed documents for data occurring in the office (photo-interpretation) or serai-
automatic devices such aa auto-recording paper tape calipers described by Badan, which
automatically record the diameter measurements. In special oases, where increment has to
be determined by boring instruments (e.g. plantation inventories in temperate oneB) f tbe
"Increment core measuring device" after Eklund may be used. With a special set of machines
the annual ring information obtained under a microscope is transferred automatically onto
punched cards, or onto paper tapes, since this device can be linked to a card puncher.

All these methods mentioned are applicable only when skilled personnel and
appropriate equipment are available. In most tropical forest inventories for which these
conditions have yet to be fulfilled use of handwritten field documents appears to be the
most appropriate way of recording, regardless of the method used for data processing.

24 Some practical aspects of data recording

Organisation in the field.

241 Once the recording document him been provisionally
designed, its usefulness and practicability ehould be tted during the
preparatory phaee of the inventory.

field instructions must be tested ae w*ll. They oheuld ^Ive details about:

the composition of the field inventory orew;

the functions of the crew with regard to data recording}
the data to be recorded ;

the manner the data are to be recorded on the prepared recording documents*

The field instructions should also ffiva details or how to progress within the RU r on
the sequence of tallying tree species and dbh, on the checking procedure during the tally,
and any further information which will facilitate the data collection.

Preparation for further processing*

24? Before the recorded data can be processed
or, oards in tn*"~oiuie of EDP or manually processed), two
(i.eT punched
preparatory actions should be taken:

(a) The field documents should be sorted manually Into logical order, which is
A check can also be made to verify
sUiuHanecuely a first check on consistency.
that no document of a sampling unit is missing.

(b) A visual check of the data itself is advisable to detect inconsistencies which could
be cleared with the field crews if they cannot be corrected in the office.
 - 142 -

3 Data processing

31 Steps of data processing

311 Data capture . One of the main problems in data processing, especially from
the economical point of view, is the arrangement of basic data into a feasible
form for further processing. EDP in particular demands practical and efficient Methods
of data capture, that is the arrangement of the basic data in a computer readable fora*
Any manual transfer of data between the recording phase and the processing phase should be
avoided as much as possible. In addition to the methods described above (see para* 23) f
there are two main types of data capture used for the transfer of the original data from
field sheets to the computer:

(a) Perforated (or punched) paper tape

Although of decreasing importance in forest inventory, paper tape (as used, for
instance, in teletype technique) is especially appropriate for the transfer of large
amount** of homogeneous data, euoh as long enumeration lists* Paper tapes adapt well also
to automatic data capture, aa in the case of continuous recording of climatic measurements
at field stations* A drawback of the paper tape method is, however, the relatively
troublesome transfer of the paper tape data onto magnetic tape. Special hardware devices
and "free format facility" 0) are needed which will be available only at very few computer
centres.

(b) 8Q-oolumn punch cards

The standard sice punch card consists of columns and rows, in which the perforations
are made at predetermined places by use of special card-punching machines, the keyboard of
which resembles a typewriter.

Punching errors during punching operations can be reduced, if not totally avoided,
by the use of "card-verifier*' machines, on which the punching of the original data is
repeated and checked against the punched card*

Although changes in temperature and humidity could affect dimensions and weight of
cards, causing warping (relative humidity in card-storage rooms should not exceed 65$), the
punched cards are, under present conditions in tropical countries, the most appropriate
way of data capture and transfer, since at almost all computer centres punohed-card readers
are available*

In addition the etandard-sieed punched cards are adaptable to different makes of

computer* As mentioned above, the data is punched on the cards at predetermined places
which are specified by the punch card design* Some basic rules for the punch card design
which will help to avoid possible errors are the following:

Always use only one card for the storage of a "logical record* . In EDP a
9
(i)
logical record is the smallest part of a data file which is determined by the
logical structure of the recorded data* As an example the logical records
for the 4 basic types of records given in paragraph 221 are respectively!
1* the information on a single plot;
2. the information on a single tree;

3* the tree tally for one species (number of trees per diameter class);

4. epeoiee code and diameter of all trees, tallied in an RU.

(1) The use of froe-foniat requires special machine-oriented software, which does not need
the definition of the format of each reoord*
 - 143 -

If, especially in the latter case, the logical record cannot be stored on a
single card, extension cards for this logical record may be used, provided
that an extension code \s given on the first card.

(ii) The amount of columns to be reserved for on* parameter ("itsa" or "attribute"
of the logical record) is predetermined by the largest figure, which can occur
for this item. Thus in the planning stage of the punch card design, careful
analysis of each item to be stored is necessary*

(iii) The "data fields" (i.e. the number of columns to be reserved for the storage
of one item) can follow on the data card one after the other without blank
columns in between. All items, regardless of whether there are continuous
or discrete variables, should be punched "right justified^ 1 ) in their
respective data fields, decimal points not being punched* All items of
different logical records stored in a given data field require exactly the
same amount of decimals. The identification of the decimals is done later
on by a "format specification" within the computer programme.

312Editing of data. Editing aims at producing from the basic data file a clean
data file free of punching errors and other inconsistencies. The different
steps of editing are the following:

(a) Sorting of the data

Sorting is necessary because many checks require that data sets be in chronological
order* It can be done manually, mechanically or electronically. Mechanical sorting of
80-oolumn punch cards by the use of special sorting machines requires that one logical
record be punched on one single card, the hierarchy of sorting being given by the inventory
design. First f coding errors within the sequence codes will be detected and corrected.
Larger amounts of data require electro nioal sorting, making use of special SORTVroutines,
usually provided by the computer manufacturer.

(b) Error detection by checks

We can consider a given data file a* a two dime visional data matrix of the form:

^n X - item of a logical record

n I - index for the poeitior of th* item X within
a logical record

X (I f j) J - index for the position of a logical record

within a whole data file
*~V-s- n - number of items of a logical record (variables)

m - number of logical records of the whole data

file

Checks can be made:

- on every individual item X (I, J) separately!

- horizontally on the relation between two or nore different items of a logical
record, e.g. the relation between X(1,J) and X(?,j) or between X(l f J) f X(2,J)
and X(5,J);
- vertically on the relation between given items of different logical records,
e.g. the relation between X(I f 2) and X(I,3)

(1) For example, the value 355 is stored right- justified in a data field of 6 digits if
it is punched in places 4 to 6 of the data field, places 1 to 3 not being used*
 - 144 ~

The different data checks common to all inventory data-processing procedures can be
grouped generally under the following types of consistency checks:

- control of completeness; aims at checking the completeness of the whole data

file at each level of data accumulation* First, it should be checked that no
logical record within a data set (record unit, sampling unit, etc.) and no item
within a logical record (e.g. diameter for counted trees f heights for measured
trees y etc.) are missing. The completeness of data on different unit levels is
usually checked by comparing the accumulated number of sub-units (e.g. trees in
a plot v or plots in a sampling unit) with the given totals on the unit level,
which are either assessed in the field (in the case of tree totals) or given by
the inventory design (number of plots per sampling unit, number of sampling units
per block, etc.). To avoid under* or over-estimation, completeness checks should
be made prior to any later inventory data processing*

- logical and likelihood control; includes all checks on individual items and on
the above-mentioned relations of different items, which can be defined logically
in view of the record and sampling design* Such checks are:

- checks of alphanumeric and numeric punches column by column;

- range of continuous variables by determining minimum and maximum values,
between which the value of the variables can vary;
- possible range of discrete variables, e.g. for established species codes
or for codes of vegetation types, etc,;
- check of horizontal and vertical inter- relationships in the whole data
matrix mentioned above, using logical comparisons of the values of the
parameters* In the case of EDP, such comparisons can be very efficiently
done using "logical operators , available in advanced computer languages
11

(for details see any FORTRAN programmer e guide)*

1

(c) Listing of inconsistencies and correction

In the case of EDP, a list of the detected inconsistencies is printed automatically

as well as all the items of the corresponding cards. Corrections of the basic data should
be made in close collaboration with the responsible field officer* The cards are punched
again with the corrected data and then added to the original data file, replacing the
wrong records on the tape or disk files.

(d) Production of clean data file

This data file, which should contain only clean and reliable data organized in the
hierarchy of the inventory, serves as input ior all generation operations*

313 Data generation* This step of an inventory data processing system includes
all operations aiming at preparing the basic data for further computations,
and at computing from this data intermediate and definitive results*

The data must be sorted depending on the inventory design, on the type of data
gathered in the field (see para* 221) and on the various classifications used (strata and
inventory units). At any level of study the following estimates have to be computed and
presented inmulti entry tables, related to species or group* of species and to diameter
classes:
- means per area unit (number of trees per ha, volumes per ha, etc*)
- totals (mainly volumes)

pi ing errors
 The position of every logical data record,
therefore, has to be identified within
the hierarchy of the inventory and the oorr re ponding dmta must be weighted
accordingly and
aggregated within subtotals and to tale for the level of study concerned. The final output
of the calculations at the differed levelc of study are

(a) Stand tables: mean numbers of stems per area unit and possibly corresponding totals
per strata and inventory unit*.

(b) Volume tables (derived from voluwft ecru*tinnei): indi1<H*l tree volumes by dbh
(local volume taVlee) or more ofton bv dbh R^ height, p*r species or group of
species,

(c) Stock tables: mean volumes per area u.nit arri corresponding totals, computed from
stand tables and volume equations, or from individual tree volumes in the sample.

(d) Corresponding standard errors or sampling rror* for given probability levels
according to ths particular sampling design.

While stand tables can be developed directly from the basic data, according to the
sampling design used, special trials have to be carried out to determine a valid set of
volume equations. Prior to the calculation of the final regression analysis, many other
computations have to be performed on the basic data (which are most often Measurement data
on felled trees) which we can call "pre~r egression studies' .1
They include the computation
of the individual volumes of the sample trees, the possible grouping of the data into
different groups of species, the drawing of scatter diagrams, the transformation or
weighting of variables, the testing of provisional regression models and the comparison of
the corresponding regression lines with the basic data. Onoe these various trials are
completed, the final volume equations and the corresponding volume tables can be developed
(for more information on volume equations see paragraph 342 of Chapter V).

"Pre- regression" and regression studies can be carried out on electronic desk
computers with limited storage facilities. It is more appropriate of course to use large
computers, since statistical trials can be carried out much more easily on these machines,
especially if a large amount of basic data has to be treated.

Besides ths basic calculations mentioned above, there are in every forest inventory
some special investigation* which have to be carried out according to specific requirements.
Examples of such additional r^miits required are:

- specific volumes such as "net" or "industrial or extract able volumes taking

1*

into account information on the quality of the trees;

- accessibility results computed from assessment of accessibility parameters (such

as breakdown of the areas by slope classes, soil bearing capacity olamses, ate.);

- cost studies, including evaluation of logging and transportation costs. These

additional investigations require several rearrangements of the basic data and
the computation of intermediate parameters, such as decay indicators.

314 Presentation of the inventory results

3U.1 BMJO table formats

The rsemlts of a forest inventory, regsrding both areas and parameters, are regrouped
for presentation in tables. Following the i?67 meetlcg of inventory experts at FAO
headquarters, a minimum set of standard tables WM recommended which were already
rprodoed in the first edition of this manual. These mininmm table* required by all
inventories are listed below..
 - 146 -

Imble Title

VI- 1 Summary of areas by present land-use and forest classes.

YI-2 Areas by (life sones), inventory classes and administrative unit* or other
specific classes.

VI-3 Present forest areas according to ownership classes, administrative units or

other specific inventory classes.

VI -4 Stand table for each inventory class*

VI -5 Total areas f volumes per area unit and total volumes with precision
estimates according to inventory units and inventory classes

VI -6 Volumes according to use classes and species or species groups by

inventory classes.

VI-7 Stock tables for each administrative unit and inventory class according
to dbh classes.

YI-8 Net annual volume increments for inventory classes by administrative unit*.

The formats and some explanation of these tables are given in the following pages.
As stated in Chapter V, paragraph 32, results should be given also in metric system if
the British System is in use in the country.
 - 147 -
table TI-1

of by pre*ent and forest olasses i/

For detail* on thi* ola*ifioatioa eee Chapter IY 9 paragraph 23

If a olaeelfloatlon ba*ed on Tefetation/enrirooBent relationship* * *uoh i
olaeeifioatlon in life *on*e after Boldrid^ - 1* not u*ed f then only one
ooluan nill bt hown.
 - 14B -

Ifcble VI-

Areas by (Life Zones - * ) Inventory Classes

C and Administrative
Units or" Other Specific Classes
ler Specific

Date of inventory:

\J This table is to be presented for the whole inventoried area and possibly
for each inventory unit.

2/ Life tones are recognised according to the Holdridge Systesi if adopted}

if another type of class! float ion based on vegetation/environnent
relationships is used, specify the corresponding classes; if such classi-
fication is not used owit it*

jj Inventory classes are mainly classes corresponding to existing land use 9

physiography, accessibility, operability f or aiqr other stratification
used in the inventory (see Chapter 17, paragraph 21). If two or more
inventory olassioations are used simultaneously , the table mist be split
up accordingly*
 - 149 -

Tbbls I- 11

to
Ownership Classes.
Present forest Arta^Accordiqg
AdministratiTe Unite or Other Specific Inretrtcrr Classes

Date of inventory-!

This tabU to b
! pr*nt*d for the hol inventoried araa and poeeibly for aoh
I/
imrantory unit.

2/ Porect area* an the** oonaldwad undar itw I .A of table 1. Similar table* can
poMibly be prevented for other wooded areaa (itea IB of table 1) if deeded
M 19

neoeeeary, ae well at tables oorrespoadinf to the nm of the two oateforiee.

y Classification of ownership

Ua* Inolude forest owned by national , state v and oantonal governsientt, corenaenV*
owned corporations, and Crown forests*

1,b. forests belonginf to towm f Tillsfes and ooswnuiss and other local authorities.
Include any other publicly owned forests not elsewhere specified*

2,b. ill forests owned by indiri duals, f sallies 9 or corporations engaged In

afrioalture as well*
2.0* ill priTetely owned forests not included Is where, ooaprisinf forests owned

by institution (religious, educational, to.).

3. Forests for which ownership status is in doubt or has not yet bean stabliihed,

If necessary, the abore ownership classes aey be further sub-dlrided to acre

specifically described local conditions*

/ If desired, a separate table can be prepared for each administrative unit.

 - 150-

Stand Table for Eaoh Inventory Claaa

Adainiatrative Unit (or other specific olasa)s Date of Inventory!

Inventory Claaei * Artm ia H+otartit

SPECIES D.B.H. CUSS CM. 2/

OR
SPECIES
OROUPB

TOT1L

j/ Thl table ia to be presented for the whole inventoried area and

poeeibly for each inventory unit

Z/ Claaa limitt abould be ehown for eaoh dbh olaae in afreenent with lifcite
indicated in table of paragraph 2421 9 Chapter V.
 - 151 -

T*bl> VI-5

Total Yolm * I/.

v*r ra unit agjjtetal
Pr+oiitou IrtjjMtt+i for Inrtatory toit> and Igy^fltorr CI

I/ Toluve to b defined in a footnote

2/ See footnote y Table 71-2

 - 152 -

Tfcble TI-6

7ol< U* ClaBee and Speoiea or Speoio Orouca

Aooording to U
by Innventory Cla0ee*

Adainirtrative unit or other specific cla**e;

Pate of inventory

I/ Yoluve to be defined in footnote.

2/ See footnote $/ 9 Tkble VI-2

H
^/ JboMiple of "U*e Claaeee t Presently marketable epeciee
1x
Specie* not jet Marketable but of potential value
2s
31 Specie* of no preaent value and of unknown
potential value
fpeoi* vill occur in only one of the three u*e olaaaee. Aete olaeeee
be further eub-<iivided according to local need*
 - 153-

frbla VI-7

Stock Tfcblas for Baoh AdiiinUtratiTa Unit and

Inrantory Claai
^
According to Dbh Claaaaa

Unit (or othar epaoifio olaaa)s ......... Pstt of Inrtntoryt

Inrtntory Class i in ...... ,

Saa footnota y, Tfcbla VI-2

I/

2/ ClM0 liaita abould ba ahown for aach dbh olaa in with limits
indioatad in table of paragraph 242*1 in Chaptar V*
 Table YI-6

1/ 2/
Net Annual Volume Increments -* for Invtntory Cla*ee -*

by Administrative Units

\J (a) 9p+oify how TO 1 urn* inormtnt figurtt have b*tn obtained and whthr
are I

for all foraeta oonBidarad ae productive; or

for fortits pr^atntly undtr tzploltation.

(b) State wbetner all apaoiei are included (noi>~o<MBeroial f etc*)

If no, which epeoiee are included and what peroectafe of total

volume do the species included represent*

(o) Specify if all a^es and diaaetert are included; if no. which
and/ or diaaetere are included.

2/ See footnote ^, Table H-

 - 155 -

314*2 Significance and accuracy

The final tables should contain some indication of the reliability of the results.
Besides the standard errors whloh should be given for every level of stratification for
which final result tables are computed, careful consideration should be given to the
slirnlfloanoe of the results. If t for example, on a certain level of stratification the
number of sampling units representing a particular stratusi is too small, the corresponding
result tables for this stratum should be omittod. If for some reason, however, it is
deemed useful to produce these results, their relative reliability should always be
indicated by giving the area of the corresponding simple and the number of sampling units.
The mention of such indications is therefore highly recoomeided.

The accuracy with which the different results are given depends on the total error
of these results (including sampling and measurement errors). Accuracy of the results
should be of the same order of magnitude as the estimated total error.

The accuracy of frequency figures - numbers of trees per area unit of given species
or species group aod of given diameter classes - must be compatible with their expected
total error: a very small number of trees/ha has often a high total error and consequently
has only an indicative value. A sensible disposition would be to indicate the figure by
a special sign instead of presenting it in the same way as the other figures. A similar
presentation to that indicated by Guin&udeau (1973) could be applied:

blank (no figure) : no data in the stratum cell concerned

dash - ) : no trees for the stratum cell concerned

(

asterisk * : at least one tree within the stratum cell, the average
( )

number of trees/ha however being less than a given figure

(say 0.1/ha)

figure : the number of trees /ha of the stratum cell with an accuracy
of 0,1 tree /ha

If it is intended also to produce the very small values, it is better to present a

stand table with the actual numbers of trees found in the sample by species or groups of
species and by diameter classes.

Presentation of inventory results with an accuracy much higher than their expected
total errors is illusive and misleading, the more so when they are obtained by EDP, since
people trust EDP more than manual data processing.

of the results: if f for

Efcphasls must also be placed on consistency in accuracy if,
***
oe f the
instance, "*" volume
mean vr>inn* r*r h* is
per ha in iriven accuracv of
given with an accuracy u.i m^
or 0.1 m-' ana
and me
the total
^otal area of
a block with an accuracy of 100 ha, the total volume should not be given to the nearest
cubic metre*

315 System design. A system design for data processing as a whole should
indicate the logical sequence of the various "activities to be considered in
11

the planning and implementation phases. Figure H-1 illustrates this in f loir-chart form
in the case of processing of field intentory data including sample tree measurements for
assessment of volwe equation, the area data being derived separately by photo
interpretation*
 156 -

ACTTVITEES

ItfUT/OUTPUT INPUT/OUTPUT
ST1RT

INVEOTORY SOURCE
AND DESIGN OP DOCUMENTS
SAMPLING SOURCE DOCUM. PLOT (I'
DESIGN

INVENTORY
DAT* FILE

UPDATED
DAT* FILE
 - 157 -

A system design in useful for the following reasons:

(a) The diffsrsnt steps foreseen in a provisional checklist are placed in logical order;

(b) the system design clarifies the way the results are produced, from the basic data
by the computational procedure* corresponding to the inventory design;

(c) the system design is essential to decide upon the type of data processing to be
employed;

(d) the system desigti is a necessary tool for all time and cost estimates related to
data processing;

(e) in the case of EDP, the system design is an essential source of information for the
analyst/programmer to develop the appropriate computer routines; the system design
is finally of great help in deciding whether the data processing is to be carried
out by the inventory project itself or whether it must be subcontracted to
specialised data processing firms (see details in paragraph 332).

32 Selection of type of data processing

321 Manual data processing* There are a number of circumstances under which
manual proces si ug by means of mechanical or non-programmable electronic desk
calculators may be considered the most appropriate way of processing inventory data* When
deciding upon the type of data processing to be used the factors to be considered are:

- the amount and complexity of data (input)

- scope of the inventory
- the time and cost factors
- available expertise and machine capacity for more sophisticated processing
procedures, such as EDP.

For small-scale inventories, especially in relatively simple forest conditions (e.g.

few species and little variation in forest types), manual data processing may be more
appropriate than EDP since the latter requires greater initial investment and investigation
prior to the actual processing. In addition, manual processing may be more generally
preferable when a relatively small amount of data are to be treated and results are required
fairly rapidly. Manual processing cannot of course be avoided, even in large-eoale
inventories, if no computers or competent staff are available for EDP.

If, on the other hand, DP is adopted, manual processing serves as a procedure for
determining the various computing steps and to analyse the computing procedure to be used
as a basis for the EDP programming. Furthermore, manual calculations are essential for
testing KDP* This is particularly relevant when EDP is being subcontracted to outside
institutions or firms. The programmes are checked by calculating a small portion of the
required results manually and by comparing the results obtained with those of the computer
using a set of teat cards. Thin is often the only means of detecting errors in ths

To avoid human error as much as possible when manually process ing data, it is
essential to give clear instructions for all calculations, for instance by means of detailed
system designe. In addition it is of great help if all computations are done on special
predesigned forms, in which all step* of computation are described and ths user has to
follow ths instructions given on the form. Suoh forms have been developed by Dawkins
(1966) at the Commonwealth Forestry Institute, Oxford, for the calculation in classical
statistical designs including basic sampling designs (unrestricted random and stratified
random). Analogous forms can be designed for any type of manual processing for forest
inventory*
 - 158 -

322 Electronic data processing (EDP). EDP by means of digital computers requires
from the inventory expert a basic knowledge of the fundamental concepts of a
computer, its advantages and restrictions (for details see Loetsoh, ZShrer and Haller 1973
and cited literature). Ibe basic elements of EDP as compared with those of manual data
processing are:

Items Manual Processing EDP

1. Instructions for basic mathematical computation programmes in problem-

computation rules oriented languages (e.g.
FORTRAN, ALGOL, COBOL, PL1 )

2. Data to be treated handwritten or typed data on data input in machine-

field recording documents readable form (e.g. punched
cards)

3. Computation by man by control unit, also called

Central Processing Unit
(CPU)

4. Auxiliary devices desk calculators, slide-rules, arithmetic and logic unit

calculation tables (e.g. within the computer
trigonometry or logarithm
tables)

5. Intermediate calculation and composition of automatic storage by storage

results tables etc. facility in the CPU (computer
memory) or on tapes /disks/
drums during operation

6. Results written or typed result tables, automatically printed

etc. results (print-out sheets)

Hie advantages of EDP compared with manual processing are evident:

- manpower is only necessary during the phase of setting the rules for computing
(i.e. the programming phase). Once the programmes have been developed and
tested by the analyst/programmers in close collaboration with the inventory
expert, the computer carries out all calculations automatically, so that human
error is avoided;
- arithmetic and logic operations which are carried out in manual processing by
man or auxiliary devices, are preprogrammed and operational during one "run" I/
in the computer;
- tables and intermediate results (. volume, stand or stock tables) can be
stored in the CPU and every "cell" / can be addressed automatically at any and
every moment during the processing operations;
- all results can be printed in definitive form to be included directly in the
final report.

jy
~ "Run" or "job" signifies the whole operation carried out by the computer under the
command of one programme or one set of programmes between the start and stop signal
of the system control*

2/ A "cell" is the smallest unit in the computer which can be addressed and occupied
directly by the programme oooaand.
 - 159 -

However, the disadvantages of EDP are as follows:

- specialists are required for writing the programmes (analyst /programmers) and
for running the machine (operators);
- the inventory project must have access to a suitable
computer centre;
- the basic data has to be transferred to machine readable documents
prior to
processing, the transfer being one of the main cost components of EDP.

It is evident then that data processing for a given

inventory has to be evaluated
carefully before deciding whether EDP or manual processing is the appropriate method.
However, in most forest inventories a huge amount of data (e.g. 50,000 - 100,000 logical
records) has to be treated and various types of output (result tables) are required.
Therefore, in most cases, EDP will be chosen, since it allows, after having captured the
data in machine readable form and prepared the programmes, for the processing of the
complete results very quickly in their final form. In addition, results can be reprocessed
very easily and at a small additional expense if some changes are required in the computation.
This is the case, for instance, when a new breakdown of areas by inventory units is necessary
or when a new grouping of species or a new set of volume specifications is adopted.

The basic data stored by EDP on tapes, disks or drums can be considered as a data
bank. Whenever additional information on the inventoried area or parts of it are required
they can be processed using the data bank as input. The facility of data storage is of
particular importance in continuous forest inventory (CFI) for updating purposes or for
timber forecasts made from data recorded at different periods. Finally, data banks are
also useful for the elaboration of countrywide or region-wide statistics using the basic
data of various forest inventories.

If electronic processing of the data of a forest inventory is to be used, it must

be decided at an early stage during the planning of the inventory. The following steps
have to be taken into consideration by the inventory expert:

(a) During preparatory phase

1. Definition of the final results to be reached by the inventory in relation

to its objectives.
2. Listing of the basic parameters to be recorded either in the field or from
other sources (e.g. aerial photographs and mape).
3. Elaboration of the system design (see paragraph 315).
4. Survey of the available computer facilities with the aim of selecting those
which can cope with a part or the whole of the processing work.
5 Design of recording documents and punched cards (or other machine-readable
documents).
6. Development of flow charxs for the input and editing programmes by the
analyst/programmer according to inventory and system designs in close
collaboration with the inventory specialist.

(b) During the implementation

7. Elaboration and testing of the input and editing programmes by the

analyst/ pro grander.
8. Punching of basic data and editing of the data*
9. Correction of incorrect data in close collaboration with the field crew leaders
and continuous storage of correct data on tapes, disks or drums.
10. Flow charts for calculation and output prograjBoes, and for programming and
testing according to the inventory and system design.
11. Documentation of the data processing system.
 - 160 -

(o) After completion of data collection

12. Production of final results (output).

13. Storage of basic data and intermediate result B (data bank) on tapes.

The main workload for the inventory specialist in relation to HDP occurs before and
during the first part of the implementation stage of the inventory. Data processing has
to be carefully analysed from the very beginning of the inventory operation and must be
taken into account in the related cost and time studies.

323 Combined types of data processing. In certain cases (small inventories ,

working plans, etc.) manual and electronical data processing may be combined.
While editing and minor calculations are carried out manually, more complicated procedures
such as regression analysis or sampling error calculations are performed on pre-programmed
desk computers (such as the HA NO 700 series or Olivetti 600 series) or larger digital
computers at computer centres. One disadvantage of this method is that editing is not
carried out automatically! but the main drawback comes perhaps from the fact that the data
are manipulated and transferred several times which may result in additional errors during
processing. Combined methods of data processing are thus different from totally
integrated data processing procedures which can be seen as "closed systems", and in which
the basic data remain untouched, once they have been edited. Therefore, they should be
recommended only if EDP is not feasible.

33 Some practical aspects of EDP

331 Project-integrated data processing.

ing, Prom the practical point of view it is
important to decide which parts of IEDP will be carried out in the project.
is mentioned above, the ye tec design has to be prepared in any case by the inventory expert
himself. Punching of data oouid be done in the project, if a punching machine could be
hired for a determined period and secretarial/clerical staff could be trained in the use of
card punchers and verifiers. Programing could be incorporated in the project, if the
Inventory expert has experience and time for this activity, or if other staff members
could be trained in computer programming. Since programming in FORTRAN or any other
problm-oriented computer language can be learnt easily within a two or three week
programmer's course, the field crew leaders may, for instance, carry out the programming
during periods when not engaged in field work. This method has been practised with
success in the Swedish Batiorml Forest Inventory for many years.

totally project-integrated data processing has the advantage of having continuously

close oontact between the inventory staff and the data processing staff, which is of
particular importance during the implementation phase (editir^ of basic data) and later
stages, such as pre- regression studies for volume estimation* As for
punching t experience
has shown that these activities in many case* can be carried out more cheaply and
effectively outside the project at specialised firms, which normally have qualified staff
tad long experience at their disposal, which reduce punching errors to a considerable
extent.

332 Sub-oontraoted data processing. Those parts of data processing not being
carried out by the project itself will normally be sub-contracted to
adequately staffed computer centres, data processing consulting firms, or data processing
departments of universities, etc. A few forms of sub-contracts merit consideration, the
mopt appropriate to be chosen according to the special aseds of the inventory project*
 - 161 -

332.1 Sub-contracting data processing as a whole

All data processing will be subcontracted, including oard punching, the elaboration
of the system design (in close collaboration with the inventory expert), the treatment of
the data on computer* following the agreed eye tea design and the preparation of the final
results required by the inventory project.

A totally sub-contracted EDP eye tea requires from the inventory project a very
careful appraisal and statement of the work to be done under the contract, defined as
follows:

! The technical specif i cations.

(a) Definition of the computational procedures, such ast

-
regression models for volume equations or taper functions;
-
significance tests;
- qxtality assessment (calculation of different types of volumes);
-
sampling error calculation on the different levels of study*

(b) Specifications of the output required, suoh ass

- stand and stock tables for - species
- groups of species
- forest
types
-
inventory units
containing figures per area unit and/or totals;
- volume tables (tabulation of the volume equations used for volume estimation);
- plotted scatter diagrams t including the calculated regression line;
- hlstogrammes of diameter distribution for main species or species groups,
forest types, logging units, etc,;
- breakdown of total inventoried area by classes of accessibility, preferably
for blocks or smaller units.

II t Obligations of the contractor

-
Timing of the EDP work;
- delivery of provisional results for checking and final approval by the project;
- documentation of the EDP programmes, including flow charts of computation
procedures, programme lists, programme decks on cards and/or magnetic tapes,
detailed description of programme operation (preparation of system control
cards, parameter cards, etc.);
- delivery of corrected and sorted data tapes for storage purposes and future
use Iff the project;
-
delivery of monthly progress reports on status of work done and planned for
the forthcoming month.
 - 162 -

Cost breakdown intoz

- man hours, unit costs and total coats for:

- supervisory time

- -
analyst/ programmer: programming/ tee ting
- documentation
- production
-
key punching verifying
- computer time (hours and unit costs, description of computer model,
type, series):
- testing
-
production
- mi see 1 laneous t
- rent or purchase of tapes
- travel costs of the contractor
- expenditure costs, etc.

332.2 Sub-contracting parts of the data processing

Many forms of sub-contracting parts of the data processing are feasible, such ass

- project-integrated punching and verification of the data, editing on small

computers y to which the project might have access; sub-contracting of data
generation and production of final results to large computer centres;
- the use of the computer and other installation* of a computer centre only*

When data processing totally integrated vdthin the inventory project is not possible,
partly sub-contracted ED? is in many cases the most appropriate alternative*
Technical and other specifications to be defined in the contract will be less detailed*
If the contract between the inventory project and the contractor concerns only the use of a
computer and related facilities, the following specifications should be clearly stated!

- use of the computer itself (reservation of computer time, staffing of the

computer with operators);
- use of peripheral hardware (card- puncher, sorter, doubler and translator);
- use of computer hardware (tapes, discs v drum files for predetermined periods);
- use of software (statistical routines, etc* - see paragraph 233 of
computer
Chapter VI);
- system and programming advice from the computer centre in system-oriented
questions of the computer language and the system control.

In most oases the computer centre should charge only for the u0e of the computer
itself and perhaps access to the computer software, on a time or per-oase basis y respectively.
Other services should be granted free of charge. These points should be fully agreed upon
whenever an inventory project contracts EDP work to a computer centre. Only in very few
oases will the use of the computer be totally free to the project*
 - 163 -

233 Some vie we of the use of standard programmes. To our knowledge there have
not yet been developed generalised EDP systems for forest inventory
applicable to various types of inventory (sampling) designs and flexible enough for
producing different types of result tables* The FIN5YS system developed in the United
States several years ago (see Prayer et al 1968), although highly flexible, covers only a
few used in that country and requires fairly large computers (minimum
sampling designs
32-K memory). In addition, the programme control by parameter cards appears to be
particularly sophisticated.

PAD is therefore at present developing a generalized EDP system for tropical forest
inventory with the following main features:
- full flexibility for the INPUT and EDITING parts;
- restrictions of the generation and error calculation phase to the most
common sampling designs;
- certain flexibility in the production of OUTOJT (result tables), to generate
tables in addition to the standard ones recommended by PAO (see paragraph 314.1).

There are, however, quite a few other pre-programmed EDP routines available at
almost all computer centres which are very useful and can be included in forest inventory
EDP systems* All regression analysis, variance and co-variance trials, scatter diagrams
and histograms, stratifications and functional descriptions of distribution can be
calculated by means of standard programmes, provided that the computational procedures
incorporated in the programmes are appropriate to the problems to be solved. Collections
of standard programmes to be recommended are, for instance:

- BUD (Biomedioal computer programmes), developed at the University of California

(see Dixon 1968), originally for IBM 7094 computers but applicable to other
computer makes with a minimum of approximately 32 K-words memory.

- Statistical programmes of the German Calculation Centre (DRZ - Deutsohes

Rftohen-Zentrum - see DRZ 1969), which are designed to complement the BCD series.

In addition, almost all computer manufacturers supply their computer centres with
since these
pre-programmed statistical routines, to which the user can have direct access,
"STSATPACKS" are normally stored on disks or drums.

Standard programmes should be used whenever possible during forest inventory data
processing operations, since time and a considerable amount of money can be saved.
 - 164 -

CHAPTER VII

C01BIDERATION5 ON INVENTORY DESIGNS

 - 165 -

CHAPTER VII

CONSIDERATIONS ON I_NVENTORT DESIGNS

1 Introduction

The former chapters have been devoted to the study of the principal techniques that
are useful in forest inventory, namely sampling techniques f remote sensing techniques,
forest mensuration techniques and data processing. Planning and designing a forest
inventory consist mainly in developing the most efficient combination of these various
techniques to fulfil the objectives of the operation, taking into account the prevailing
human and environmental conditions. In this respect even data pro oes sing problems have
to be contemplated from the very beginning since available means in manpower and
computing facilities also have a bearing on the type of inventory methodology used, as
has already been mentioned in chapter VI.

There is no point trying to cover all the situations and objectives assigned to
forest resource surveys and the corresponding combinations of techniques which are likely
to be the most appropriate in each case. This would be an endless and illusory task.
A more modest and also more realistic approach is used in this chapter where some problems
arising from the combination of these techniques will be dealt with and some recommendations
will be made on the suitability of these techniques to actual working conditions.

Techniques of volume estimation and quality assessment have already been compared
in chapter V. Their effects on the precision and on the usefulness of the inventory
results are far from negligible; but, especially in mixed tropical forests, the largest
part of the total cost involved comes from the field enumeration work* The most
important questions to be considered are therefore related to the latter part of the
inventory work and are mainly the following:

- to what extent interpretation of remote sensing imagery can be combined with

the field enumeration work so ae to decrease the effort spent on the latter,
and thus reduce the total cost of the inventory (for a given precision of the
final estimates of the parameters over the whole inventoried area) ?

- the importance of interpretation of remote sensing imagery being decided upon,

what are the most suitable characteristics of the field sampling design ?

Some indications useful in solving these problems are given in the two main
following sections of this chapter. The formulas corresponding to some classical
combinations of photointerpretation and field sampling procedures are indicated, together
with the oases to which they apply (section 2). General formulas corresponding to the
most classical field sampling designs have already been given in chapter III and the
contents of section 3 sxe restricted to some guidelines on the lection of the most
appropriate field sampling design with special reference to mixed tropical forests.

Most forest inventories aim at estimating the characteristics of forest stands at

a given time There exist, however, some permanent inventory designs which consist in
the combination of different samples selected on successive occasions, which oan be
grouped under the generic denomination of "Continuous Forest Inventory", and which art
briefly commented on in section 4* Another type of sampling designs follows a stspwis*
procedure f in which the decision whether or not to undertake further sampling depends on
the results already obtained from the sample; this is called sequential sampling and is
dealt with in section 5
 - 166 -

2 Combinations of photointerpretation and field sampling procedures

21 Preliminary remarks

It is assumed in this section that some basic conditions which art those already
indicated in the introduction of Chapter IV are fulfilled! the total area of the
inventoried zone is supposed to be exactly known and mapped at a suitable scale (which
allows for the definition of frames for sampling designs ) y and the stand characteristics
are estimated through field sampling (the case of photogranmetric measurements of stand
characteristics being excluded as it is generally not applicable to inventories of mixed
tropical forests).

The main use of interpretation of remote sensing imagery in forest inventory is to

stratify the area to be inventoried into more homogeneous parts or strata which are sampled
separately in the field, in order to get more precise estimates of the total values of the
forest characteristics over the whole area* For reasons mentioned in the following
paragraph it is relatively rare that areas of the strata are known exactly or almost
exactly, and generally they have to be estimated through a sound sampling design.

22 Areas of the strata exactly or almost exactly known

Stratum areas can be said to be exactly or almost exactly known when the actual and
present limits of the strata are drawn on a reliable and stable planimetrio map and their
areas carefully planimetered or estimated very precisely by use of very dense dot grids
(see section 5 of chaptBrlV)* The actual and present limits of the strata can be located
on the photographs if:

- very recent aerial photographs at a suitable scale are available, or, if they are
not very recent (say if they are two or three years old, but not more) if no
significant changes are likely to have occurred between the date of the aerial
coverage and the date of the field inventory (significant, that is, in relation
to the accepted accuracy of the area figures);

- no systematic photointerpretation error is expected, and consequently none - or a

- of the sampling units of the field
negligible part inventory will have to be
transferred from the strata to which, through misinterpretation, they have been
assigned by photointerpretation to different strata to which they actually belongt

It can be easily understood from the above considerations that such cases very seldom
happen* Even if the aerial coverage has been taken in the same year as the field work,
inconsistencies and errors in the photointerinterpretation work can always be expected
since conventional interpretation of remote sensing imagery is not a purely objective
exercise* Inconsistencies and errors over time by the same interpreter or between
interpreters can be avoided only if the stratification is simple and easy and if there are
sharp limits between strata (as has already been mentioned in paragraph 334-1 of chapter IV
it is generally difficult to draw objectively a limit between strata as there may be more
or less wide transition zones)*

In the rare oases when an exact (or almost exact) evaluation of the areas of the
strata can be secured, i.e. when it is assumed that there is no misinterpretation in the
photointerpretation, the sampling frame for each stratum is well defined and an independent
selection of the field sample can be made within each stratum* The mean value* and
estimated variances of the forest characteristics per ultimate unit over the whole
population estimated from the field sample can be derived fro* the corresponding formula*
given in chapter III for sampling design* with stratification prior to sampling (formula*
5, 6 of 11, 12 in case of one-stage sampling designs), in which the total sites ft (or
of the strata and N (or X) of the whole population can be replaced by the total area*
S n and S respectively.
 - 167 -

Ihe total values of the forest characteristics for the whole inventoried area is
obtained by multiplying the mean value per ultimate unit
by the total number (which is
exactly known) of ultimate units in the whole area. The percentage standard error is the
same aa that of the mean value.

It may happen that an unstratified sample is first selected for the whole area and
that stratification ie made by photointerpretation after the sampling on the basis of
criteria recognised in the field and identifiable on the photographs. If it is possible
to assume that this stratification is fully valid, then the exact sizes of these strata
can be known and the estimated means and their estimated variances are derived from
formulae (?) and (8) of chapter III (in the case of one-stage sampling with equal units),
which are somewhat different from those corresponding to stratification prior to sampling.
Total values are obtained in the same way as in the case of stratification prior to sampling
and the estimated relative standard errors are the same as for the corresponding mean values.

23 Areas of the strata estimated through sampling

In this case the estimation of the strata through a sampling design increases the
sampling error of the estimates of the total values of stand characteristic! over the whole
inventoried area or over individual strata.

In forest inventory there are various ways of estimating the areas of the strata
using sampling* The main alternatives are:

(a) sampling on the photographs (cr on the maps), the photointerpretation being supposed
to be unbiased (same case as in the above paragraph, planimetering being replaced by
sampling);

(b) sampling in the field from the same (or a larger) sample used for the estimation of
the stand characteristics: in this method the area estimates are supposed to be
unbiased and do not need to be corrected since identification of the strata is made
on the spot;

(c) sampling from the photographs (or even possibly planimetering on the maps), the area
estimates obtained being afterwards corrected through a sampling in the field, using
the sample used for estimation of the stand characteristics or a larger sample.

When estimating areas through sampling it is important also to distinguish the case
when the sampling units are points or plots (the associated parameter having the values
1 or according to whether the plot belongs to the relevant stratum or not) from the
case when the sampling units are lines or strips (in which case the associated parameter
is a length or a ratio of lengths, both continuous variates).

1
231 Area estimates from one sample only!
'
Stratification and estimation of the
areas of the strata oaji be made by sampling on photographs or in the field,
if photographs are considered unsuitable (too old or too bad to allow for a useful
stratification). The sample used for estimation of the areas of the strata may be the
same as the one used in the field for estimation of the stand characteristics or may be a
of the stand characteristics).
larger one (including the sampling units used for the estimation

to designate the whole set of

(1) Sample is used hsre as in the rest of this manual
sampling units selected according to a given sampling design.
 - 168 -

2311 Area estimation using plots as sampling units

In this case the associated parameter can take the values 1 or according to
whether the plot belongs to the particular stratum considered or not. The area of a
stratum is obtained by multiplying the total area of the inventoried zone (which is
supposed to be known exactly) by the proportion (estimate) of plots falling in this stratum*

231*11 Area and ataiid characteristics estimation from the same sample

If area estimation is made by photointerpretation, the sampling units in the field

used for the estimation of the stand characteristics are centred on the plots interpreted
on the photographs (see paragraph 334*2 of chapter IV ) They may be single plots or
clusters of plots, a bit smaller or larger than the photoplots but each of them is supposed
to belong entirely to the stratum of the corresponding plot on the photographs (no
correction of the photointerpretation is supposed to be necessary).

If the plots are randomly distributed the proportions Pu of the strata and their
variances are estimated by formulas (3) and (4) of chapter III if it is a one stage
sampling design, or by formulas (13*)and 14*)of the same chapter if a two-stage design has
been used, in which the primary units are the effective areas of the photographs (supposed
equal) and are selected at random, the same number of plots (secondary units) being sampled
on each selected photograph. Other sampling designs can be contemplated as has already
been mentioned in paragraph 532 of chapter IV. The percentage standard errors of the
estimates of the areas of the strata are equal to those of the corresponding proportions
since the total area is supposed to be known exactly.

If the plots are systematically distributed in a one-stage design, it is recommended

to use the formula given in paragraph 531 of chapter IV for the estimation of the variances,
in which the constant k is given a value according to the shape of the corresponding
stratum.

The estimates of the mean values per ultimate sampling unit of the stand
characteristics and their variances are obtained by the formulas of the corresponding
unstratified sampling design. The corresponding estimates of the tptal values over the
whole population are obtained by multiplying the estimated means by the total number of
plots in the whole inventoried zone (or by its total area if they are means per area unit)
and their percentage standard errors are the same as those of the respective means.

Estimates of the means per ultimate unit in each stratum and of their variances are
obtained from the individual values in the sampling units belonging to the corresponding
stratum* However oare must be exorcised since the sampling is done within the population
aa a whole and the same formulas as for the whole population do not apply necessarily to
each stratum* Estimates of the totals per stratum are obtained by multiplying the
corresponding estimated means by the estimated area of the respective stratum.

If y, is the estimated mean per area unit of a stand characteristic y in stratum h,

tfh

P. is the estimated proportion of stratum h

S is the total area of the inventoried zone,

A * *
an estimate T. of the total value of this characteristic in stratum h is: s
Y^ ^h^h
and an estimate of its variance is:

v(Th ) -
*
S
2
/-Ph
2
v(yh ) + V v(?h )7
~
where v(y.) and v(P.) stand respectively for the estimated variances of yfa and P^* This
2
formula is acceptable if v(y) and v(P) are small relative to y and ? respectively.
h
 - 169 -

Attention is drawn to the fact that these deigns are definitely different from
designs using stratification after sampling. In these latter it is assumed that the sixes
of the strata are known exactly or almost exactly, which is not the case with the design*
studied in this paragraph where only an estimate of the sices of the strata is obtained.

In the above paragraphs, as well as in the rest of section 23. we do not consider
the case of correction or weighting of the values found in the individual plots interpreted
on the photographs which are necessary for taking into account the variation in scale and
in overlaps of the photographic coverage used (see paragraph 334.2 of chapter IV). This
correction corresponds to coefficients which appear in the formulas of the estimated means
and totals and adds some complication to the estimation of the corresponding variances.
In order to avoid it f it is recommended that the sample to be used for area estimation by
photointerpretation ia laid out on an existing map or on the corresponding uncontrolled
mosaic provided this can be accepted as a reasonable approximation to a map. In case
correction is deemed preferable, the reader will study with profit the examples given
in paragraph 25.32 of "Forest Inventory" (Volume 1, pages 235-244) by P. Loetsoh and
K. Haller.

231.12 Area estimation from a larger sample than the field sample used fer
estimation of stand characteristics

The field sample for the estimation of the stand characteristics is a subs ample of
the sample used in photointerpretation or in the field for the estimation of the areas of
the strata. If area estimation is made by photointerpretation the field sampling units
of the subsample are centred on the plots interpreted on the photographs and each of them
is supposed to belong entirely to the stratum of the corresponding plot. We will assume
that the selection of the large sample and of the subsample from the large sample are
both random one- at age samples. This type of combined sampling is called two- phase
sampling or double sampling.

Areas are estimated from the large sample of plots t in the same way as described in
the preceding paragraph.

If S, f P, and S stand respectively for the estimated area of the stratum h (L strata
in total), the estimated proportion from the large sample and the total area (exactly known)
of the whole inventoried zone, we have*

\ '
V s

The estimated means y . of the stand characteristics over the whole inventoried zone

per sampling unit are given by:

where the y. are the estimated mean of the stand oharaot eristic per sampling unit in each
stratum obtained fro* the field subsanple v the summation being extended to all the strata.

the calculation of the variances of the estimates y Bt depends on the way the
subsample has been selected from the sample.
 - 170 -

If the numbers of
(a) sampling units per stratum in the subsample it do not depend on the
estimated proportion f^ f * ne corresponding stratum, then ah estimate of the variance
of y Bt is:

<*.> 4
which oan aleo be written ae:

where: n' IB the number cf sampling unite of the

large sample
?
B IB the estimated variance of the stand characteristic y in stratum h:
h

2
8
h - 1
n,

(i being the index of a sampling unit in stratum h of the field subsample)

(b) If the nh depend on the Pftl the formula giving the estimated variance of
yet is
somewhat different. In case of a proportional allocation of the sampling units in
the field subsample, i.e. if - where n - n, is the size of the subsample,
r^ nP^ jfc
an estimate of the variance is:

y
2
- (* iJ (4)

Formulae (?) f (3) f (3') and (4) are known as formulas of double sampling with
stratification, Totals over tne whole inventoried area are obtained by
multiplying the
estimated means by the total area exactly known and their relative standard error is the
same as those of the corresponding means. Estimates of the totals per strata are:

(5)

231.2 Area gtimation using parallel lines as sampling units

In this case the line will be considered as the sampling unit and the associated
parameter for a given stratum is a continuous variable since the length of the portion of
a line which is found (on the photographs or mape or
possibly in the field) within a given
stratum oan take in principle all values between and the total length of the line .
The oae applies aleo to stripe and to linee of plots, the occurrence of the etretum
being
checked only at the plots in the latter case, and even also to clusters of plots >
provided
that the number of plots per line or per cluster ie large enough,

We will confine ourselves to the cases when the same sample is used for
estimating
the areas of the strata and the stand characteristics.
 - 171 -

If all the lines have the same total length 1 within the whole inventoried zone
(e.g. if the inventoried zone is square or rectangular and the lines parallel to one side
of the araa) the areas of the strata are estimated (in a
one-stage design) by:

where 3 is the total area of the inventoried aone

ie the estimated mean length of a line (sampling unit)

1^

in the stratum h: 1 (i being the index of a line)

If the parallel lines are randomly distributed formula (2) of chapter III is
applicable to the estimation of the variance of T n the percentage standard error of S,
,

being equal to that of 1 . If they are systematically distributed formulas suggested

in section 423 of chapter III have to be adapted.

In most cases the lines have unequal total lengths and then ratio estimation is
necessary. If we assume a one-stage random design of n parallel lines, and if 1. and
l^ h stand respectively for the total length of line i and the length of the portion(s)
of line i within stratum h, an estimate of the area of stratum h will be:

(7)

T, and 1 being respectively the means per line of lengths within stratum h and of
total length.

The standard error of S^ can be estimated by using formulas (10) or (10') of chapter
III related to the variance of a ratio estimate.

The estimates of the means per area unit of the stand characteristics over the
whole population are ratio estimate* in the ca.ee when the lines have different total
lengths, with total length of a line ae the auxiliary parameter. Means for each stratum
oan be estimated also and will be ratio estimates. But, as has already been said in
paragraph 231.11, the sampling design must be considered as unstratified, since the exact
8ie of each stratum ie not known. Indications given in the above mentioned paragraph
concerning the estimation of the means and totals per stratum are valid also in this oaae.

The oaae of continuous lines oan be extended to the caae of continuous stripe, of
linee of plots and of duet era.

232 Area estimates with correction in the field* In many oases there are
unavoidable mietakee biases in the photointerpretation work due to the
ar*d

interpret ere, to the Gratification adopted which may be too refined, to the bad
characteristic* of the photograph* and, most often, to change* in vegetation which have
occurred between the aerial survey and the field inventory. Estimation of the areas
made by photointerpretation need then to be corrected by field checks made in a subsample
of the photointerpretation sample.
 - 172 -

We will confine ourselves to the design described below which is very much used in
forest inventory in temperate zones and which needs to be adapted in an efficient way in
mixed tropical forests, which consists of:

(a) selection of an unstratified sample of plots to be interpreted on the photographs

for an assessment of the strata h as identified by photointerpretation, e,g, by a
systematic grid put on the effective area of each photograph, or by a systematic grid
put on a mosaic of these photographs, or by a random selection on a map of points
transferred afterwards onto the corresponding photographs, etc.);

(b) selection (in one stage) of .a subsaaple of plots among the interpreted plots, the
number of these plots in each stratum h (as interpreted on the photographs) being
dependent or independent of the proportion of this stratum (as found by photointerpretation);

(c) identification in the field of the actual strata k to which the plots of the
subsample belong and measurements for estimation of the etand characteristics y in these
plots.

If P. is the proportion of plots of the large sample found in stratum h by

photointerpretation

fl, is the proportion of plots of the subsanple found to be in stratum h by

photointerpretation and in stratum k in the field (whenever k / h there
is misinterpretation)

an estimate R of the actual proportion of stratum k in the whole inventoried zone is

given by:

An estimate yet f t*16 mean value per unit (or per area unit) over the whole
inventoried zone of the stand characteristic y IP given by:

- -
k. hihk (9)

where y_^ is the estimate of the mean value per sampling unit (or per area unit)
of the stand characteristic y in the part of actual stratum |c belonging to the

photointerpretation stratum h.

The estimate of the mean value per unit (or per area unit) in stratum k of y is
equal to:

p
1 h' p hk
dk *

The estimated totals T over the whole inventoried zone and T^ over the stratum k
are obtained by multiplying the expressions in (9) and (10) respective by S and
L - -
S P p
^li5'1 h" hk^
f ( S b * in* the arca of the inventoried zone),
y^ in formulas (9) and (10)
being the mean value per area unit.
 - 173 -

An estimation of the variance of jxia given by applying the formula of double

sampling with stratification to the variable (1,0) indicating whether a plot interpreted
in stratum h on the photographs belongs to actual stratum k or^not. Tranefoimation
of formula (3*) gives thufl (in case the are independent of the P.):
n^

n* and IL standing as in formula (3) for the sice of the large photo interpretation
sample and the number of units of the subsample selected in stratum h. In most
cases the term ^h can be neglected.
7T*

If a proportional allocation of the eubeample among the photointerprctation strata h

is made,(i.e. if n, nP. ,
with n- n, ), then the formula (11) becomes:
iy-

r - (
v,,) J

Although the estimates of the variances of the means and totals of the stand
characteristics per stratum and for the whole inventoried zone are somewhat complicated,
such a design is very useful as it permits a reduction of the error by use of photo-
interpretation, even when the photographs are not completely recent as is often the case.
(Assistance of a statistician will be looked for to determine an estimate of the variances
of the means and totals of the stand characteristics.) However it must be realised that
if the discrepancies between photointerpretation and ground checks are likely to be large
and if the size of the subsample is relatively small y the areas of the strata will be
estimated with a very low accuracy.

24 Other uses of double sampling designs

The double sampling designs indicated above are used to improve the precision of
the estimates of the stand characteristics through a better estimation of the siee (area)
of the strata. They are called double sampling methods for stratification.

However, double sampling designs can be used in forest inventory for other purpose*.
Double sampling for regression is also used in some cases, for instance when photogrammetric
measurements of a stand character is tic (e.g. gross volume of all species) are made on a
large sample of plots on the photographs and are corrected by regression on the field
subeaiiple of these plots. Assuming that each ample is an unst ratified random sample,
the corrected mean y tie
n per sampling unit is estimated by:

y + b (x'-Z) (13)

where) y is the estimate of the mean per sampling unit of the oharaoteriatio
obtained from the field subsample
 - 174 -

is the estimate of the mean per sampling unit of the photogrammetric

measurements of this characteristic on the plots of the field subsample

is the estimate of the mean per sampling unit of the photogrammetric measurements
of this characteristic on the plots of the large sample

n'

ft*

and b IB estimated by:

wliere x^ and y< stand for the value of x (photogrammetric measurement) and

y (characteristic measured in the field) in the plot i of the field subsample,

An estimate of the variance of y^ e when the size n of the subsample is not too
email is:

with: n' being the size of the large sample

2
n -
4
y 1

Double sampling for regression can be imagined for the estimation of a stand
characteristic y with any other auxiliary variate x estimated from a larger sample on
photographs or in the field and which is linear! ly correlated with y.

Double sampling for regression is also useful with estimates of areas of strata
obtained by reconnaissance flights along parallel transects used, for instance f to correct
estimates obtained by photointerpretation* However such methods must be used carefully
since it is generally difficult to locate precisely on a map a point overflown because of
the irregular speed and orientation of the plane,
 - 175 -

When the straight line

representing the relation between y and the auxiliary
parameter x goes through the origin - i.e. when y tends to zero with x - double
sampling is used for ratio estimation. In this case the corrected ratio estimate y
will be equal to: R

'
-
yR -t- i - fct*
(15)

where y ie the estimate of the mean per sampling unit of the characteristic
y
obtained from the subsample
x is the estimate of the mean per sampling unit of the auxiliary parameter
obtained from the field subsample

*
is the estimate of the mean per sampling unit of the auxiliary parameter
obtained from the large sample

An estimate of the variance of yp - the sma] ler sample (size n ) being a subsample
of the larger one (size n*)
- is equal TO:
p
**
**p p * *^ p
B - 2K 8 + R B 2R B - R s

2
with n, n ,
R and s having the same meaning as above and

a
xy
- -- 2
n-l
,

(estimate of the covariance of x and y)

An example of double sampling with ratio estimation is the one where an estimate of
timber volume is performed in a quick way on a field sample (this quick estimate being the
auxiliary parameter x) and a more accurate assessment of the volume (y) being made from
detailed tree measurements on a subsanple.

Double sampling designs are a very powerful tool in forest inventory either for
stratification or with the use of an auxiliary parameter (regression or ratio estimation)
but leads in many cases to difficult and rather complicated estimation of the variances of
the result*. In case of double sampling with ratio or regression estimation ths
relationship between the parameter to be estimated and the auxiliary parameter has to be
assessed and ths design must be conceived in order to reduce the unavoidable biases of
ths estimates. Assistance of a statistician proves to be particularly useful in this case.
 - 176 -

3 Considerations on field sampling designs

The former section dealt with Borne aspects of the combination of information obtained
from remote sensing imagery and information collected in the field* It is important indeed
when designing a forest inventory to reduce the field work as much as possible by making
the greatest possible use of interpretation of remote sensing imagery since this involves
less manpower f equipment and operating expenses than field work* However, field work
cannot be avoided in most forest inventories, especially when stand characteristics
including gross and extractable volumes have to be estimated precisely* Even in the case
of surveys of homogeneous stands using large-scale aerial photography and photo volume
tables, ground checks are necessary to correct, through a double sampling procedure, the
estimates obtained from the photographs* In mixed tropical forests field sampling is
generally the most important and expensive part of forest inventory operations, due to the
limitations of interpretation of aerial photographs in these areas. As has already been
mentioned in chapter IV the first constraint on the use of aerial photographs for
estimation of stand characteristics in mixed tropical forests is the difficulty of
identifying tree species. But even if species could be identified satisfactorily, other
difficulties in the estimation of stand characteristics from remote sensing imagery remain,
such as the relatively loose correlation between crown characteristics and stem dimensions
in these natural forests* Furthermore there is no way to assess, from remote sensing
imagery, characteristics of regeneration, of quality and of occurrence of decay or of
accessibility such as soil bearing capacity or ground roughness.

Therefore the choice of the field sampling design is particularly important. Some
indications are given below of the suitability and relative advantages of various types of
distribution of the sample (sampling design strictly speaking) and of the possible nature,
eize and shape of the sampling units. These two topics are considered separately although
they are in fact very closely linkedi for instance the decision whether to use a one-stage
or two-stage sampling design depends partly on the size of the sampling units; if the
latter are relatively large a two-stage sampling design may not bring a significant
increase in efficiency even in an inventory of a vast forested area*

31 Distribution of the sample

311 Unrestricted versus stratified sampliog. As said in Cochran's "Sampling

techniques (2nd edition, page 99), stratification "if intelligently used
11

nearly always results in a smaller variance for the estimated mean or total than is given
by a comparable simple random sample*"! Stratification in forest inventory is generally
performed through interpretation of remote sensing imagery prior to the field sampling (or
after sampling if stratification criteria are assessed after the field sampling). It must
be emphasized again that the field sampling is actually stratified only if the size of the
strata can be exactly (or almost exactly) known or if, as in the case of double sampling
for stratification, their size is estimated from a larger sample* Thus formulas giving
the estimated variances obtained from stratified samples (such as formulas (6) or (8) of
chapter III) should not be applied when limits of "strata" are drawn around a set of
sampling unite, and there is no further interpretation to ascertain whether all the units
of these "strata" correspond to the criteria defined for this stratification (e.g* slope
greater than 50J&? height of the dominant trees larger than 15 metres, eto...)*

The criteria for stratification Bust be defined in a clear and under* tandable fora*
Vary refined stratification by photo interpretation is generally illusory since the possible
gain in precision by comparison with a more simple classification may be more than counter-
balanced by subjective biases, misinterpretations and discrepancies between photoi interpreters
and over time* Even in the case of a simple and easy classification, misinterpretations
are possible due to low quality and the age of the remote sensing imagery; it is therefore
necessary to correct the areas of the strata by a sampling procedure such as the one
described in paragraph 232 of this chapter i the precision of the estimated means of the
stand characteristics over the whole inventoried area decreases as the proportion of
misinterpretations increases and the gain by stratification may become insignificant
 - 17.'
-

compared with the total coBt of the stratification work. It must be realized aloo that
when stratification has to be corrected by a sampling procedure an exact assessment of the
actual location of the ntrata IB not possible, th* estimates of their real areas becoming
less precise as the intensity of the fielr* namplr i^ lower.

312 Random versus systematic sampling. In the case of one-stage sampling designs
there is no doubt that the practical advantages of the systematic distribution
of the sampling units greatly exceed its main theoretical shortcoming, that is the difficulty
of estimating the variances of the results. Most of the practical advantages of systematic
sampling in temperate forests are still more evident in tropical mixed forests where
environmental conditions hamper field work. AmDng these advantages may be quoted
reduction of access cost for ari area unit of sample, greater certainty of objectivity in
the selection of the sample (the systematic distribution of the sample leaves less room for
possible modification of the location of the sampling units by the field crews) and more
uniform distribution of the sample (and consequently of information) over the inventoried
area (this latter advantage being more significant in areas which are surveyed for the first
time). Moreover research ie being pursued on the estimation of statistical error in
systematic sampling and it is expected that methods based on the theory of stochastic
processes will soon become available in practice. For all these reasons it is highly
recommended that a systematic distribution of the sampling units should be adopted whenever
one stage sampling is feasible.

In multi-stage sampling designs, the choice between a systematic and a random

distribution of the sampling units has to be made at each stage of the sampling procedure.
Tlie advantages of a systematic distribution are not equally important in the various stages.
Regular distribution of information within the penultimate units (within the primary units
in a two stage design) irc generally not e&czeritial, while the systematic layout of the
sampling units of the first stages may be particularly useful. For instance, in a two-
stage sampling design where the primary units are squares of 2 kilometre sides, and
secondary units strips of 2 kilometres in length and, say, 10 metres in width, the random
allocation of the strips within each primary unit will not bring a significant increase in
access cost but, on the contrary, the systematic distribution of the oquares over the whole
inventoried area may be of great value, especially if the ar<-a IB surveyed for the first
time.

313 One-stage versus multi-stage sampling. The main advantage of a multi-stage

design in comparison with a one-stage design of the same overall sampling intensity end with
the same size and Bhape of the (ultimate) sampling units, is that the component of the cost
allocated to the access of the (ultimate) sampling unite is greatly reduced. This is
true in mixed tropical forests where penetration is difficult. Against this
particularly
the concentration of the sample resulting from a multi-stage procedure increases the
variance of the estimates, and the greater the variability is between the units of the first
stage, the larger is this increase in variance.

These considerations can be illustrated in a very simple and sketchy way in the case
of two-etage designs (aee Desabie - 1966)t

(a) the variance of the estimate can very often be expressed aa :

v -
A
n
+ -B
nm

where t n is the number of primary sampling units

m is the mean number of secondary sampling unite per primary sampling

unit
A is a measure of the variability between primary unita

B in a measure of the average variability between secondary units within

a giv^n primary unit.
 - 1/8 -

(b) the cost of the sampling can often be expressed approximately as:

C nC- + nmC o

where i C. is the cost of access to and reconnaissance of one primary unit

is the cost of access to a secondary unit (when the primary unit hatf
Cp
been reached) and of recording inventory data in this unit.

It can be understood from these two formulas that v depends very much on A and n
and ic likely to be larger than the variance corresponding to a one-stage design with nm
sampling units. On the other hand the second formula explains why C will be smaller
than the cost of reaching and recording nm secondary units distributed in a one-stage design.

If acceptable estimates of A, B, C-\ and 02 are available the two -stage design can be
optimized under certain constraints using the procedure indicated in paragraph 341 of
chapter III.

This simple formulation should be kept in mind when deciding between a one-stage or
two-stage design., An already said the size of the ultimate sampling units is an important
factor and some one-stage cluster sampling designs do not differ much in cost although they
are fundamentally different as far as variance estimation is concerned. The larger and the
more inaccessible the inventoried area, the more suitable a multi-stage design, but other
factors are important such as the need for information uniformly distributed over the whole
inventoried area and also the size of inventory units for which estimates have to be
provided.

314 Equal or unequal probability in sampling. Most field sampling designs used in
forest inventory consist of sampling units selected with equal probabilities
(and without replacement). But there exist some efficient designs for which the probability
of selection of the sampling units are proportional to their size as the one indicated in
paragraph 422.122 of chapter III. When selecting such a design one has to remember that
the sizes of all the units of the population considered (population of the primary units in
the above-mentioned example) have to be known and listed: in certain cases the cost of the
corresponding work may be too high compared with the expected gain in precision.

315 Use of an auxiliary parameter. Whenever an auxiliary parameter which is

linearily correlated with the parameters to be estimated by the field sampling
can be known exactly or estimated cheaply from a large sample, its use is recommended*
Examples of ratio and regression estimates in double sampling have already been given in
section 2 of this chapter. A classical example in forest inventory is also the use of the
size of the units as an auxiliary parameter for ratio estimation of the stand
sampling
characteristics. it is very common in inventories of mixed tropical forests to have
ultimate units of different size; for instance, if parallel strips are used as sampling units
their area may vary due to the irregular shape of the inventoried area and of the relevant
stratum or both, and also with the steepness of the terrain (if the dimensions of the
stripe are measured along the terrain and not horizontally). However when using this type
of estimation, it must not be forgotten that the "ratio of the means " estimates are biased
and that this bias has to be reduced to a minimum (see footnote, page 49 )

32 Characteristics of the sampling units

321 Plot sampling versus polyareal sampling. Plot sampling consists of designs
using area elements as sampling units or record units (with the same sampling
or reoord unit possibly composed of two or three plots of different size for the recording
of different parameters - see for an example paragraph 23 of chapter V), while polyareal
sampling corresponds to point (or line) sampling designs in which the size of the recording
area in each unit is a continuous function of a characteristic of the tree (e.g. its basal
area in horizontal point sampling).
 - 179 -

In this latter case there is no sampling or record unit in the

physical sense and
the whole population to be inventoried cannot be
considered, strictly speaking, as the
collection of the points or lines used as sampling "units". The practicality of
polyareal sampling designs in mixed tropical forests has already been briefly discussed
in paragraph 422.2 of chapter III. Cost precision studies in temperate forests have
shown that horizontal point sampling is generally more efficient but there is no evidence
for the time being that it is the same in the tropics. Furthermore a mere efficiency study
is not sufficient and other factors have to be considered such as the reliability of the
data recorded - selection of the trees to be recorded is more difficult in point sampling
than in plot sampling - and the advantages of obtaining in each unit a representative picture
of the forest, which is not provided by point sampling. There exist some combined point
sampling designs where all trees below a given diameter (say 30 cw) are recorded provided
they fall within the circular plot the radius of which is determined by this diameter and
the basal area factor used in the sampling, the larger trees being of course recorded
1n the normal manner uaed in point sampling.

3^2 Size of the sampling units. It is commonly accepted that the coefficient of
variation C y of a given stand characteristic (parameter ), (say number of trees
more than a given diameter per sampling unit) is linked with the area of the sampling unit
by the following empirical relation:

where j (JT
is the standard deviation of the individual values of the stand
characteristic y in the unite of the population

y is the mean of the stand characteristic per sampling unit

a is the individual area of the sampling units

k and c are positive constants independent of a*

"Kris relation can also be written in logarithmic form:

log Cy - log k - c log a - K - c log a

c in equal to 0.5 when the distribution of the values per unit of the stand
characteristic is a random distribution, such ae the Poisaon distribution. This iB
approximately true of parameters related to the occurrence of trees of species with a very
low density in mixed tropical forests (e.g. numbers of stems and correoponding gross
volumes of the "mahoganies" in West Africa forests). For many other parameters in tropical
forests c is found to be rathr lower than 0.5.

It is interesting to compare different sizes cf sampling unit in unatrttified random

sampling design for the same sampling intensity. In this case we have:

na constant

where n and a are respectively the number of the sampling unite and the area of one
aapling unit. The percentage standard error of y (mean value per unit of a stand
characteristic y) is equal tot

L.C?.\
_ M
Is-M*
* 0.5-c
**

th lttr rprion being obtained fro* th pirio&l fomul*

 - 180 -

For most of the parameters in mixed tropical forests we have C <, O.J). It oan be
concluded from (18) that, for a given sampling intensity^ the smaller the sampling units
the better the precision. However it is useful to have in each sampling unit a fairly
representative image of the forest and this can only be obtained if the sampling units
have a reasonable size: a sampling unit of 0.01 ha in a mixed tropical forest for
estimation of the volume of exploitable size would not be useful in this respect* In
addition the total number of borderline trees in the whole sample (all sampling units) is
higher for a sample consisting of a large number of small sampling units than for an
equivalent bample (same total area) consisting of a smaller number of larger sampling
units of the same shape. The selected size of the sampling units is thus a compromise
between the conflicting requirements of the sampling precision and of the important
practical aspects of re press rrtevtivftneos of the sampling units and reliability of the basic
data. An area of the sampling unit equal to 1 acre (0.4 hal or to 03 ha 10 often
conniderod as a suitable compromise in inventories of mixed tropical forests.

3^3 Shape of the sampling units

3^3.1 Circular versus cquare or rectangular plots

The main advantages of circular plots are:

- the minimum perimeter for a given area of the circle compared to other simple
geometric shapes, which in turn implies the minimum number of borderline trees;

- the isotropic image of the forest around the centre given by a circular sampling
unit.

Its use in temperate areab ia increasing although it must be realized that, for
practical reasons, the form of these plots is in fact elliptic whenever there is a slope.
(Interesting devices usinc a ran^e-finder and a stadia rod with adjustable sighting marks
for assessment of this type of plot are mentioned in "Forest Inventory" by Loetsch-Zb'hrer-
Haller, Volume II, pages 324-325 and in "Dendrometrie" by Parde", pages 190-199.)

However the difficult environmental conditions and the need for a larger size of the
sampling units prevent the use of circular sampling units ri mixed tropical forests (but
i

not that of circular recording units: see below, paragraph 324. 3 ) Square or rectangular
sampling unito (and also record units) are often preferred in mixed tropical forests. They
may be stripe of a given width (generally from 10 metres to 25 metres) along parallel lines
of penetration, cut through the undergrowth, and going through all the inventoried area or
through a part or a stratum of it. The width should not be larger than 30 metres - i.e.
15 metres on each side of the transect line - in order to allow for a good control of the
recording operation, and the width can be measured either horizontally or along the terrain;
in the first case no correction has to be made for the determination of the area of ths
strip, but the recording is more time-cone urn ing and possibly less reliable because of the
borderline trees; the second method may be more reliable but involves measurement of the
transverse elopes and more computation. The sampling units can also be either
rectangular plots, or lines of rectangular plots, the plots being, in the second case, the
record units and not the sampling units. The plots cannot be used as the sampling units
as the distance between two consecutive plots along the line is not sufficient to secure
statistical independence with regard to the parameters to be estimated (see paragraph 322.2
of chapter III).

3? 3- 2 Form of the rectangular plots

In a study made in Cameroon it has been found that the more elongated the shape of a
rectangular sampling plot of a given site, the better the precision, although this effect
on the precision of the shape of the sampling units was found to be less important and less
significant than the effect of their size. However this was not true for very long
Btrips (unit areas of more than 5 hectares) and very wide strips (100 metres and more in
 - 181 -

width) were found to give better precision. Of course the use of ouch very wide plots
would not be possible in practice ir, the inventory of mixed tropical forest.

323.3 Clusters

In inventories of mixed tropical forests, sampling units are often groups or clusters
of circular plots, in order to profit from the advantage* of circular plots while having
at the same time sufficiently large sampling unite. In this case the circular plots are
the recording units, and are often arranged along a straight line or a squared or rectangular
line ("tracts of some European national forest inventories).
11
However for the same sice
of sampling unit, a cluster of circular plots may have a longer total perimeter (and
consequently more borderline trees) than the equivalent rectangular sampling unit: for
instance a cluster of five circular plots of 0.1 ha has a longer total perimeter than a
rectangle of 200 metres long on 25 metres width (ro\ighly $60 metres against 450 metres).

Onoe a given size of circular plot (recording unit) is chosen the unit sise of a
sampling unit must be ascertained, i.e. what number M of plots each sampling unit must
contain. This is an optimisation problem with M as one of the characteristics of the
sampling design to be determined. The following paragraph gives an example of such an
optimization procedure and is partly extracted from "Sampling Techniques" by Cochran
(2nd edition, pages 244-247).

Let us consider an unrestricted random sampling with n equal clusters (sampling

units) of M circular plots each. The variance of the mean y per circular plot (record
unit) of a given stand characteristic is equal to:

where Sfc ie the variance between clusters (variance among the total values of y in
the clusters on a circular plot basis)

The first problem is to estimate S, from the variances among the values of y in the
circular plots f i.e. S variance among the values of y in the M plots within a cluster
and S^ variance among Vhe values of y in the circular plots in the whole inventoried area.
We have approximately!
2
# MS - (M-1) S^
S^ (20)

(this result being obtained by an analysis of the variance of y for the whole
population).
2
Tt has been found that, in many surveys S can be expressed by the following
empirical formula!

with A and g positive constants independent of M.

If we have a cost function of the seme type as indicated in paragraph 313 of this
chapter for twostage sampling designs, i.e.:

C * nC^ + nMC<2

(where C, and C ? have the same meaning a in paragraph 3^3, the olunter standing
for the primary unit and the circular plot for the secondary unit)
 - 182 -

then the optimization problem amounts finally to find out the values of M (and also n)
which minimi zee

for a given total cost: C nC + nMC Co (22)

Applying the procedure indicated in paragraph 341 of chapter JII f it can be easily
found that the optimal value Mo of M is given by the following equation:

2
AM c + C M - G M + - c s? '
o^ (
^o I O i o S) r

The corresponding value n of the number of clusters is determined by replacing

M by H in the equation (22).

4 Continuous f crest inventory

41 Definition and utilization

Continuous forest inventory comprises all forest inventory designs in which sampling
is used on successive occafiions. This definition is much broader in scope than tjie one of
tne north American CFI in which all the successive inventories use the saroe sample (all
the sampling units are said to be "permanent" unite).

Sampling on successive occasions should be considered in designing a forest inventory

when, in addition tc an estimate of present forest conditions, accurate determination of
past growth is required and the users of this information are willing to wait the necessary
and often lengthy period of time for its accumulation.

Assembling growth information in this manner presupposes that forest management will
be carried out 0,1 a continuing basis. Although forest management on a continuing basis is
at a very early stage in many tropical countries, and has not even been started in some
casts y inventory officers should always keep in mind the need of such inventories for forest
management purposes and should initiate continuous forest inventory programmes whenever the
concern for forest management and the probability of making use of the results of such
inventories are deemed sufficiently high. In this respect the concept of forest
management must be understood in a broad sense; the monitoring of the forest cover through
the use of permanent plots on remote sensing imagery represents a large field of
application of continuous forest inventory which does not relate only to forest management
but also to land use policy and environmental concern.

42 Description of design

421 Different types of continuous forest inventory. The objectives of repeated

sampling in forest inventory are threefold:

1 to estimate characteristics of the forest present at the first inventory;

2, to do the same on the occasion of the second inventory;

3. to estimate the changes in the forest during the period between inventories,

(Not* that the repetitive process can be continued and on ths occasion of all
subsequent inventories the previous inventory is referred to as the "first
inventory". )
 - 183 -

There are four basic ways in which the above information can be obtained:

1. A completely new sample can be drawn from the forest at the time of each inventory.
The sampling units on occasion 2 are different from those taken on occasion 1.

2. The sampling points taken at the first inventory are remeaaured at the second and
all succeeding inventories. This is the concept of permanent sample plots and
the basis of the Continuous Forest Inventory (CFI) developed in North America.

3. At the second inventory a portion of the initial sampling units are remeasured and
new ones are taken. This is often called successive sampling with partial
replacement (SPR).

4. At the second inventory a portion of the sampling units taken at the first inventory
are remeasured.

422 Sampling with partial replacement (SPR). Of the four approaches the most
efficient is the third, successive sampling with partial replacement. If
repeated inventories are planned f inventory officers should design their procedures on
this basis.

Only a concise summary of the design and analysis of this one method is attempted
here and for more details reference should be made to Ware and Cunia (1962) as shown in
the liet of references. A good description of this and the other three kinds of repeated
sampling are also covered in F. Loetsch and K. Haller "Forest Inventory" (volume 1 pages t

259 to 277).

At the initial inventory there are two kinds of sampling units; plots measured only
on the first occasion (unmatched) and plots measured at the first inventory and to be re-
measured at the second (matched). At the second inventory there will be the plots taken
at the first inventory and now to be remeasured (matched). In addition there will be
new plots to be taken which did not appear at the first inventory (new). The following
symbols for the number of sampling units and the observations are needed:

Firs t i nve nt ory

_

u number of unmatched sampling units taken at the first inventory

x parameter (stand charact eristic) measured on unmatched sampling units at
first inventory
m - number of matched sampling unite taken at the first inventory
x - parameter measured on matched sampling units at first inventory
mj
u + m m n m total number of sampling units at the first inventory

Second inventory

m number of matched sampling units taken at the second inventory (same as

m of first inventory)

y parameter (same as x) measured on matched sampling unite at second

^
inventory
n - new sampling units taken at second inventory

y , parameter (same as x) measured on new sampling units at second inventory

m + n - n - total number of sampling unite at the second inventory.

 - 134 -

Then: x
u
*- --
u
mean of values of the parameter per sampling
unit measured on first occasion from unmatched
units

mean of values of the parameter per sampling

unit on first occasion from matched units

mean of values of the parameter per sampling

unit on second occasion from matched units

y - mean of values of the parameter per sampling

unit on second occasion from unmatched units

a) Estimation of the means per sampling unit at first and second inventories*

1. The estimate z of the mean per sampling unit of the parameter at the
first inventory is;
m u

n
1

2, The best estimate y of the mean per sampling unit of the parameter at the
second inventory is given by:

wh.r.:
4
1

in
c -

2 * -
, .
-
"i

eP 2
"
.2

j j
Iy 2 being obtained similarly to 'x and x
2
 - 185-

and r -

w
J-i 0-1

An estimate of the variance of y, v(y) f is given byi

r. .
x jr

which can b xpr*d or iply a*i

c
m i_|n i r

(b) Estimation of difference between the nean value* per sampling unit of the parameter
at fir t and eeoond inventories.

In the oae when the parameter indicated by x and y ia a volume, this differanoa
will express the growth of the stand corresponding to this volume over the period bstw+sn
the two inventories.

Die bast estimate of this mean growth per sampling unit is given by the formulas

where i m

-
>,

An estimate of the variance of g is given byi

- ZABr. .
 - 186 -

which can also be written as:

9 *- O
" 8 " 8 B
v(g) \ ^
nT rT '
x nT x y -'

5 Sequential sampling

Sequential sampling, like continuous forest inventory, a? so involves a series of

samples but each of these samples includes all the sampling units of the former sample
(u o whatever the sample) and in addition the time span between two successive samples is
negligible so that from a forestry viewpoint all the samples can be considered as
simultaneous.

Uie purpose of sequential sampling is to permit the taking of a relatively secure

decision about a population (forest stand) from a limited number of units of this population.
Let us consider the example of a planted area where it has to be decided whether a cleaning
of the plants is necessary to free them from weed vegetation. Let us suppose also that
this planted area oan be divided into equal lines (units) of twenty plants. Lines will
be selected at random to constitute the successive samples and in each sample the total
number of freed plants is recorded. Each sample drawn (the second sample including all
the lines of the first and some new ones, the third including all the lines of the second
and new ones, and so on) will be represented by a point on a chart, the x-coordinate of
this point being the size (number of lines) of the sample, the ^coordinate being the total
number of freed plants of this sample. Moreover on this "sequential sampling chart" two
parallel lines are drawn which divide the chart into three regions: "no cleaning",
"continue sampling" and "cleaning necessary".

total number of
freed plants in
each sample

NO CLEANING
240

180

CLEANING
60 - NECESSARY

number of
observed trees
100 200 300 400 500 600

If the representative points of the last samples remain in the region "no cleaning"
(as in figure 1; the sampling procedure can be stopped and the decision is taken not to
carry out any cleaning operation. If the points stay in the region "cleaning necessary"
further sampling is unnecessary and the decision is taken to begin cleaning. If the
representative points are in the region "continue sampling", no decision can be taken with
 - 18? -

sufficient security and the sampling has to be pursued*

The slopes and zero ordi nates of the two parallel lines which are the basic elements
of this sampling procedure are a function oft

- the distribution of the "decision parameter' in the studied population (in the
1

above example number of freed trees per line), which is to be assimilated for the
sake of simplicity to a classic distribution ouch as binomial or Poisson
distribution;

- the minimum proportion of freed trees in a line for considering that this line
does not need any cleaning treatment (0ay 6&f> or 12 trees); ("acceptable"
proportion of freed trees);
- the maximum proportion of freed trees in a line for considering that this line
does not need a cleaning treatment (say 5&fo or 10 trees); ("unacceptable"
proportion of freed trees);
- the two accepted risks expressed in percentages of probabilityi

- of cleaning the planted area although it has in fact a sufficient number

of freed trees ("rejection risk or "producer's risk");
11

- of making no cleaning at all in the planted area although it has in fact

an insufficient proportion of freed trees ("acceptance" risk or "consumer's
risk").

The distribution being known and the four quantities above being decided upon, it
is possible to draw the lines of the chart which will help in taking the decision* A
detailed description of the design and the corresponding formulas is given in "Forest
Inventory" by P. Loetsoh and K. Haller (Volume I 9 pages 2?8 to 289).

Although this procedure is very attractive, it has found relatively Xittle application
in forestry, mainly for the reason that little is known on the distribution of forest
parameters* In mixed tropical forests there is another drawback which results from the
fact that all sampling units must be selected at random, which increases the total aooess
cost of the sampling procedure. It has been applied in forestry for regeneration surveys
and for disease and insect surveys.
 - 188 -

ANNEX I

EXAMPLE OP TECHNICAL SPECIFICATIONS

for inclusion in a contract of aerial surveying

1. CAMERA SPECIFICATIONS

(a) Vertioal photography is required to be made with single lens precision aerial camera
of 21 cm focal length and 18 cm x 18 cm format. The camera must be a modern high-
precision aerial camera of the Wild RC 5, RC 8 f or Zeies RMK 15/23 type, equipped
with a high resolution lens capable of producing the highest quality of photography
with panchromatic or infrared film* Details of the camera, lens and filters which
the Contractor uses for this Contract shall be supplied to the Organisation^ ) and
stated in the contract.

(b) Where the Contractor wishes to interpose a window or hatch cover of transparent
material between the camera and the ground, he will ensure that the said window or
hatch cover has been fully tested within six months of the proposed date of
commencement cf photography to establish that it falls within accepted tolerances
for homogeneity, resolution, and freedom from distortion. A certificate of this
test will be submitted to the Organisation for approval with the Calibration
Reports described in Section 3 of the Specifications. The contractor shall ensure
that such window or hatch cover is perfectly olean and free from blemishes at all
times .

(0) *Rie overall focal plane surface of the platen of the camera shall be flat, under
operating conditions, to within plus or minus 0.005 mm. Hie film shall be held
flat in the fooal plane, at the instant of exposure, to within plus or minus
0.005 mjfu ?h* camera platen shall be tested with all the operating stresses present
at the instant of exposure duplicated. The platen tested shall be positively
identified by having the camera cone, or magazine number of the unit, permanently
and irremovably marked thereon. This identifying number shall be noted on the
report of the test.

(d) Between-the-lens shutters, such that light is transmitted simultaneously to all

parts of the emulsion plane when the shutter is open, shall be used. The efficiency
of the shutter shall be at least 75 per cent of the marked value at the fastest
speed. Hie speeds shall be accurate to within 10 per oent of the marked value when
tested at room temperature. ftie results of the efficiency test and the date of the
test shall be recorded.

(s) When a camera is equipped with a pressure altimeter, the altimeter shall be connected
to the static system of the aircraft. The altimeter automatically recorded bjr the
camera must be adjusted to give the sane reading as the aircraft altimeter prior to
the commencement of each photographic flight.

(f) When a easier a is equipped with a cloak for recording times of exposures, the clook
hall be set to correct local tine prior to each photographic flight.

2. CAMERA CALIBRATION

(a) Each omera optical unit to be used in the performance of the Contract shall be
calibrated before the flying of the photography by a competent authority to be
approved by the Organisation and calibration report* rendered as required in Section
3 below* The interval between calibration and photography shall be as short as
possible and in no case shall ezoeed one year.

(1) 6ranisation is ued hereinafter to designate the party which lets the contract.
 - 189 -

(b) Calibration Khali be carried out with the optical unit in the same condition as
when used for the contract photography and with the filter fitted in the name
position. After calibration, no adjustment or repair which could in any way
affect the calibration shal] be wade. If the optical unit should be accidentally
disturbed, no further photography may be taken with that camera, which must be
replaced by another that has been calibrated and approved by the Organization.

3. CALIBRATION REPORTS

A calibration report for each camera used shall be submitted to the Organization and
shall contain the following information!

(a) A certificate of calibration showing the name of the approved authority, date and
place of calibration, the maker's serial number of the camera optical unit, the
serial number of the lens and of the platen.

(b) The coordinates of the principal point of auto-colliation relative to the fiducial
marks,

(c) The radial distortions of the image with reference to the principal point of auto-
collimation at zero, m^axurer outwards at intervals of not more than 7j* toward
1

the fiducial ma^ks or reseau cronses in the format corners. (A statement of the
arithmetic mean of these distortions will not be accepted by itself.) Asymmetry
of the distortions may not exceed 0.03 **

(d) The principal distance at which th*ne distortions apply.

(e) TTie distance between all fiducial mark*.

(f ) The mean resolution of the image as determined across two diagonals at an interval
of not less than 7^ by the standard method of the approved authority.

(g) All measurements shall be recorded to the nearest one-hundredth of a millimetre,

(h) These reports will be retained by the Organization.

4. SCALE

The photography vn 11 be t^ken from a flying height such that the mean oontaot scale
of any exposure it and shall not deviate from the required height above
mean sea level by more than plus or minus 5 P*? cent.

5* OVERLAP

The fore and aft overlap between successive photographs in each strip shall be 60
per cent with tolerances of plus 10 percent or minus 5 percent. The lateral overlap
between photographs of every adjacent strip shall be 30 per cent with toleranoee of
plus 20 percent or minus 15 percent. However, the Organization may agree to aooept,
in exceptional cases, photographs where the maximum overlaps (70 per cent and 50 per
oent respectively) are exceeded for reasons of terrain. Wherever variation of
ground level causes a significant change in the contact soale of the photography,
an increase Must be made in the fore and aft overlap which must in no case be less
than 55 percent, to accommodate the enlarged scale of a part or the whole of any
strip. The corresponding increase in the lateral overlap, which mist in no oase be
less than 10 percent, must be made to the whole of every strip which is so affected
by height distortion.
6. CRABBING

Crab shall not exceed 5 percent or be such tbat it causes gaps in the stereoscopic
cover of the contract area.

CAMERA TILT

not exceed 3
Tilt shalll n for any exposure. The average tilt for nny unction uf 10
eh&ll not exceed 2 and it shall not exceed 1 for the entire photography
produced.

3. COMBINED EFFECTS OF OVERUP. CRAB, DRIFT AHD TILT

(Applicable for mapping photography.) Any point at one-tenth of the width of a

photograph from the lateral ed#e muct ip;*ar in thrwe succe,:.iv photographs of
the sir-e flight line and in three photographs of the adjacent flight line.

9. FLIGHT LINES

The area shall be covered by parallel lines flown in a specified direction and these
flight lines should not diverge from the prescribed directions by more than five
degrees.

10. JUNCTION OF STRIPS

Where the end of a atrip of photography joins the end of another strip flown in the
same ^-eneral direction, the overlap of the fir^t utrip over the second will be 4
photographs.

11. CLOUD

Cloud shall not lie over the principal point of any photograph nor shall it cbscure
any ground control point. The aggregate area of cloud and/or cloud shadow on any
single photograph shall not exceed 5 percent of the area of the photograph.

12. TIME OF DAY

Photographic detail shall not be obscured by the shadows of topographic features

due to th low angle of the '\irj nor by the presence of hot-M;<>tB M .
r
f
lf

13. FILM AJO FTLTJi

This will be determined as appropriate for each contract.

1* . PHOTOGRAPHIC VJ/LITT AND

(a) T*\o photographs are required for detailed forest type napping and photointerpretation
studies and should, therefore, be of the highest photographic quality, according to
the following specifications:

(b) Die film emulsion and film base will be determined to suit each case. The film
base shall have the minimum differential distortion and the negatives aho.ll be
free from stain* y scratches, bar marku, dirt and blemishes of all kinds, and finger
or static marks.

(c) Precautions must be taken to avoid distortion of the film during processing.
 - 191 -

(d) All relevant fiducial marks must be distinct on every photograph.

(0) All automatically recorded data such as flying height, time of photography, and
calibrated focal length must be clearly visible on every print.

(f ) The negatives are required for contact printing and the density, contrast, and
freedom from fog on the negatives is to be such that "normal" grades of paper
will be suitable for the majority of the negatives without excessive shading and
with reasonable times of exposure.

(g) The definition and contrast of the negatives shall be such that prints made as in
the preceding paragraph and the prints supplied shall ohow ample detail throughout
the full range of tones ever the whole photo^riiph and ouch that identification of
detail from one photograph to another shall be possible with certainty.

(h) Prints shall be made using an electroric printer.

(1) Residual hyposulphite in any print shall not exceed 0.023 mgm. per square centimeter,
while residual silver shall not exceed 0.01 gm. per square meter.

(j) Prints shall b trimmed to leave a rebate of up to 6 mm. on 3 sides. On the side
where the images of instruments automatically recorded in the camera are located,
the rebate should be sufficient to include these images.

15. MARKING OF FILMS AN1) NEGATIVES

(a) Each exposure in *ach flight line shall be numbered in consecutive order. Tnin
nMbering shall be in the northeast or south-west corner of the negative. The
numbering should not be le** than 4 * nor more than 6 msi. in height. The
consecutive numbering of exposures should be repeated for each flight line (e.g.
the first three exposure* in flight line 5 would be narked 5-1, 5-2 and 5-3)* the
numbering should include all negative* whether falling inside or outside the
speoified areas and whether or not conforming to the specifications, excluding
only obviously useless negatives such as those completely obscured by oloud.

(b) A film shall be in one continuous length without joins, except for the leader mad
trailer which shall be at least 1 metre long. A film and its leader may include
unrequired negative* and will be given a fila roll mber to be indicated by the
Organisation which will be shown at each end of the film.

(o) The margin of the first and last negative for each flight line shall be clearly
marked with:

(i) The indication of the Organisation, mm* of locality and the nusiber of
contract.

(ii) taber of the film roll.

(iii) Date or datee on which expoeed; the month to be given by neae not number.

(iv) Time of exposure in terms of local time.

(T) taiber of oeera optical wit art lea* and the principal distance
corresponding with the calibration particular* supplied under thi*
peoifi cation*

(vi) Heifht above mean eea level at which expoeed, in metre*.

(d) Each print ehall be titled in the rebate along its northern or southern margin.

(e) The titling required on each print is to be clear blook lettering (between 3 mm. and
5 mm. in height for items (i) to (v) below f and between 4 mm. and 6 mm, for item
(vi ) below) and is to read and print in the following order:

(i) The indication of the Organization (an appropriate project designation)

(ii) Flying height above mean sea level in metres.

(iii) Camera focal length (in mm.) if not automatically reproduced on negative.

(iv) Date of photography.

(v) Number of corresponding map sheet.

(vi) Line and photograph number.

An example of complete marking is:

F*0 NIC/68/509 - 3230M - 153.4mm - 6 SEPT 72 - 12 * 1-18

16 FILM REPORTS

(a) A report containing the following details and any others that are considered to be
relevant shall be provided with each film:

The number of the contract

Name of Contractor
Number of film
Name of territory
Time of first and last exposure (local time)
Date exposed
Die serial number of the camera optical unit, the magazine and lens.
The principal distance as given in the calibration report supplied under this
specification
Lens aperture and negative numbers at which change in the aperture setting was made.
Filter. Shutter speed
Make and trade name of film t batch number and date of coating (if known)
Aircraft registration marks
Height above mean sea level and negative numbers between which changes of height
took place
Weather conditions.

(b) A list of all the numbered negatives on the film in numerical sequence and in two
columns 9 with remarks on the following lines:

PHOTOGRAPHY REMARKS

Offered Additional

1-1 to 1-20 To specification

1-21 to 1-?5 Some cloud
1-26 to 1-36 Heavy cloud
1-37 to 1-40 To specification
1-41 to 1-60 Insufficient lateral overlap;
re flown.
 - 193 -

(o) Although the information in this paragraph maj not be complete on the original
report* compiled in the field, it must be completed in the two copies of the final
film report to be supplied.

(d) A general statement on the photographic quality.

17. MAHPM1 OP FILM COMCAIMSR

On the side of each tin there shall be firnly fixed a label shoving clearly:

1* Number of the contract

2. Kane of the country or of the region (BLOCK CAPIBLLS)
3. Nane of locality
4. Name of the Contractor
5* Date(s) of exposure
6. Film roll number
7* First and last negative numbers of each line contained on roll
8* Serial number of camera optical unit and lens*

18. FLIQHT INDICES

Flight line indices of the photography are required and may vary for each contract*

19. ADDITIQMAL SPECIFICATION

The following clauses may be included in the body of the contract:

(a) Inspection and acceptance

(i) The Organisation reserves the right to reject within six weeks of the delivery
of the contact prints any photography which fails to meet the specifications*

(ii) Subject to other pertinent provisions of the contract, the contractor will
have the option either to forego payment for rejected photography or to
produce acceptable substitute photography*

(iii) If the Organisation determines that certain photography is not in accordance

with the specification*, or that there are gaps in the photographic coverage,
the contractor shall make good such shortcomings prior to extending the
photographic coverage, if so required by the Organisation*

(b) Additional photography

Additional photography, if any, taken in the course of the contract and lying
adjacent to the areas to be photographed in accordance with the indications given to
the Contractor by the Organization may be offered to the Organisation* It shall be
at the Organisation * discretion whether or not to accept any or all of the
1

additional photography offered*

 - 194 -

ANNEX II

SELECTED ANNOTATED BIBLIOGRAPHY

Only some of the most useful documents dealing with one or several techniques used
in forest inventory are quoted in this bibliography. It mainly includes textbooks or
manuals; articles or communications have been selected only when they were considered
essential . With very few exceptions, all references are in English* When they also
exist in French and/or in Spanish, this is indicated in the margin by (P) and/or (S).
More French and Spanish references will be indicated in the bibliography of the
corresponding versions of this manual*

(a) Forest inventory in general

Forest inventory is generally dealt with in specific chapter(s) in forest

mensuration textbooks and manuals, but is the specific subject of some publi cations 9

among which the following can be quoted:

Bonnor G.M. - 1972. Foreet sampling and inventories. A bibliography.
Forest Management Institute, Ottawa, Ontario. Internal Report PMR-24,
27 pp.
Selected references from English language publications excluding
articles on point sampling, on air photo measurements and theses.

FAO - 196/. Report of the Headquarters Meeting of Forest Inventory Experts

on UNDP/SF Projects (held in Rome 11-22 September 196?). FO:SF/67 - IM 17,
?59 PP.
These proceedings include communications of the participants to this
meeting and the reports and recommendations of its various working
committees. The most useful information given by this document has
been taken up again in the "Manual for forest inventory operations
executed by FAO" and in this manual.

Loetsoh F. and Haller K.E. - 1964. Forest inventory. Volume 1, 436 pp.
Loetsch F. Ztfhrer F. and Haller K.E. - 1973-
f
Forest inventory. Volume 2,
469 pp. BLV Verlagsgesellschaft Munchen - Basel - Wien.
This is strictly speaking the only recent textbook on forest inventory and
it covers the whole range of relevant techniques. Tropical aspects of
forest inventory work have not been neglected (as in former textbooks on
forest inventory) and the bibliographies of the two volumes are very
comprehensive.

Nyyeso'nen A.
- 1961. Survey methods of tropical forests* FAO publication
no. 13407.

Comparative study of the inventory methods used in the tropics in the late
fifties.

Proceedings of the 1st FAO/SIDA training oourse on forest inventory

(to be published in early 1974 by the Swedish Royal College of Forestry)
 (b) Purpose and planning of a forest inventory (Chapter II)

Husch B. - 1971. Planning a forest inventory. PAD forestry and forest

products studies No.17 121 pp.
f

As stated in its preface this concise study "outlines the principal

problems ... suggests a logical sequence for considering them ... also
discusses briefly some of the most modern inventory techniques and their
merits and limitations" but does not "cover technical methods".

(c) Basic sampling techniques in forest inventory (Chapter III)

Chacko V.J. - 1965. A manual on sampling techniques for forest surveys.

172 pp. Delhi. Manager of publications.

(S) Ccchran W.O. - 1963 (2nd edition). Sampling techniques. 413 pp. John
Wiley and Sons Inc., New York
A very clear textbook on sampling techniques, describing and demonstrating
most of the sampling designs which may be useful in forest inventory.

(P only) Desabie J. - 1966. The'one et pratique des Bondages. Dunod, Paris.

481 pp.

Easy, clear book in French on sampling techniques with emphasis on

demographic surveys.

Preese P. - 196?. Elementary forest sampling. Agriculture Handbook No. 232.

U.S. Department of Agriculture, Porest Service, 91 pp.

Very useful and practical handbook on sampling techniques for forest

inventory with many numerical examples. Recommended by most of the
specialists consulted when compiling this manual.

Hansen M.H., Hurwitz W.N. and Madow W.G. - 1953. Sample survey methods and
theory. Volume I - Methods and applications. 638 pp. Volume II - Theory.
332 pp. John Wiley and Sons Inc., New York.
Very comprehensive book with emphasiB on demographic surveys.

(P) Schumacher P.X. iind Chapman R.A. - 1954. Sampling methods in forestry and
range management. Durham, North Carolina. Duke University, School of
Forestry. Bulletin No. 7, revised. 222 pp.

Useful textbook on sampling techniques used in forestry although new

methods such as point sampling and SPR are not covered.

Sukhatwe P.V. - 1954. Sampling theory of surveys with applications. lowm

State College Press, Ames, Iowa. 491 PP*
Wsll-known textbook with emphasis on agricultural surveys

(P) Yates F. - 1960 (3rd edition). Sampling methods for censuses and surveys.
440 pp. Oriffin, London.
Tfeisbook is Isss comprehensive than Coohran's as far as the number of
sampling methods and demonstrations are oonosrned, but gives more
explanations and more numerical examples.
 - 196 -

On point or line sampling

Grosenbaugh L.R. - 195^* Point sampling and line sampling: probability

theory, geometric implications, synthesis. USDA Southern Forest Experiment
Station, Occasional Paper No. 160.

Labau V,J. - 1967 Literature on the Bitterlich method of forest cruising.

USDA Pacific Northwent Forest and Range hbrperiment Station Research Paper
PNW- 19.

(d) Remote sensing and mapping (Chapter IV)

(i ) Photogrammetric measurements in forest inventory

Most of the relerences to photogrammetric measurements for forest inventory are

articles and communications. An account of these references is made in the bibliography
of the following documents and also of the textbooks on photointerpretation (see (iii);.

American Society of Photogrammetry - 1966 (3rd edition). Manual of

Photograiwietry. George Banta Co., Menasha, Wisconsin, 1220 pp. (2 volumes).

Nielsen U. - 1971 Tree and stand measurements from aerial photographs: an

annotated bibliography. Forest Management Institute, Ottawa, Ontario.
Internal report FMR-X-29. 111 pp.

Spurr S.H. - 1960. (?nd edition) Photogrammetry and Photointerpretation

(with a oertion on application to forestry). New York, The Ronald Press Co.
47? pp.

(ii ) Vegetation and ecological classjfi cations

In addition to the documents quoted in section ?? of Chapter IV the following

should also be considered:

Holdridge L.R. - 19^7. Life zone ecology (revised edition). Tropical

Science Center, San Jose*, Costa Rica. ?05 pp.

(F)(S) Unesco - 1973. International classification and mapping of vegetation.

Ecology and conservation no. 6. Paris. 93 PP
A very recent classification worked out under the auspices of Unesco.

(iii) Photointerpretation in forest inventory

American Society of Photogrammetry - 1960. Manual of Photographic

Interpretation. George Bant a Co., Menasha, Wisconsin. 868 pp.
In addition to two chapters on, respectively, the procurement of aerial
photography and the fundamentals of photointerpretation, a chapter of 64
pages is included on photointerpretation in forestry.

Avtry T.E. - 1966. Forester's guide to aerial photointarpretation.

USDA Efcndbook no. 308 t 40 pp.

Hildebrandt 0. - 1968. Bibliographic des Schrifttuws auf dem (febiet der

forstliohen Luftbildauswertung, 1887 - 1968.
 - 197 -

Howard J.A. - 1970. Aerial photo-ecology. Paber and Paber f London.

325 pp.
Very clear and useful book on the utilization of aerial photography with
special reference to the study of vegetation, and a fairly comprehensive
bibliography (35 pages).

Stellingwerf D.A. - 1966. Practical applications of aerial photographs in

forestry and other vegetation studies. International Training Centre for
Aerial Survey. UT.C. Publications, Series B. No. 36-37-38-46-47-48.
This series of booklets contains a large number of stereograsimes showing
vegetation and forest types in the tropics and in temperate zones, and is
therefore very useful for training purposes.

(iv) New remote sensing techniques

IUFRO, Section 25 - 1971* Application of remote sensors in Forestry.

Joint Report by Working Group "Application of Remote Sensors in Forestry",
189 PP.
This report comprises 13 communications of foresters who are specialists
in remote sensing techniques and gives a good account of the application
to forestry of the new techniques in 1970.

American National Research Council - 1970. Remote sensing with special

reference to Agriculture and Forestry. National Academy of Sciences,
Washington D.C. 424 pp
f

Thie book, to which many American specialists contributed, provides the

user with the basic technical background concerning remote sensing in
general and in relation to vegetation, soil and water resources.

Krumpe P.F. - 197? Remote sensing of terrestrial vegetation: a

comprehensive bibliography. The University of Tennessee, Knoxville,
Tennessee. 69 pp*

"850 references dealing with the utilization and application of remote

sensing in forestry, agriculture and plant ecology, as well as closely
allied fields such ae land-use planning, resource inventory and management,
and soils and terrain analysis."

Wilson R.C. - 1970. Remote sensing application in forestry (A report of

research performed under the auspices of the forestry remote sensing
laboratory, School of Forestry and Conservation, University of California f
Berkeley, for National Aeronautics and Space Administration). 199 PP.*
The sub-title of this report gives the following precision: "potentially
efficient forest and range application of remote sensing using earth
orbital spacecraft - circa 1980".

() Measurement considerations (Chapter V)

(i) Forest mensuration

Rusoh B. , Miller C.I. and Beers T.W. - 1971. Forest mensuration. The
Ronald Press Co., New York. 410 pp.
A reoent textbook on forest Mensuration with four chapters on forest
inventory and 16 pages of references to relevant literature (mainly English)^
(P only) Parde" J. - 1961. DendromeHrie. Editions de 1'Ecole Nationala deB Eaux
et Forits, Nancy. 3^0 pp.

Very useful and practical textbook on forest mensuration (but with very
little emphasis on tropical forestry).

(11 ) Volume equations

Cunia T. - 1964. Weighted least square method and construction of volume

tables. Forest Science, Volume 10, No. 2. 12 pp.

First main document on the use of weighted regression for volume table
construction.

Draper N.R. and Smith H. - 1966. Applied regression analysis. John

Wiley and Sons Inc.

Freese F. - 1964. Linear Regression Methods for Forest Research. USD!

Foreot Service Research Paper, FPL 17, 138 pp.

Very clear manual on regression techniques with numerical application.

Prodan M. - 1968. Forest Biometrics (English translation by S. Gardiner

of "Forstliche Bioetrie")

(iii) Quality assessment and recovery studies

IUFRO, Section 25 - 1969. Proceedings of the meeting held in Reinbek,

Germany, of the Working Group on "Mensurational Problems of Forest Inventory
in Tropical Countries". Mitt. Bundeeforsch. anst. f. Forst- und Hblzw.
Komm. verl. Max niedebusch, Hamburg.
A record of the discussions held and background paper? submitted at this
meeting which focused principally on quality appraisal and recovery studies,

(iv) Accessibility

Von Se^ebadan G. - 1969* Studies on the accessibility of forest and

forest land in Sweden. Studia Forestalia Tecnica No. 76. 64 pp

(f) Data recording and processing in forest inventory (Chapter VI)

Anderson, D.M. - 1^66. Computer Programming FORTRAN IV. Appelton Century

Crafts, New York. 430 pp. including appendices.
A manual written for beginners on computer programming in FORTRAN IV.
Very instructive, suitable for home tuition and teaching purposes, with
many graphs, exercises and comparison of FORTRAN compilers of different
makes (IBM series, Uni vac- Burroughs, General Electric, Honeywell, etc.)

Dawkins H.C. - 1968. STATFORMS - Formats for elementary statistical

calculation. Commonwealth Forestry Institute, Oxford. Institute Paper
No. 41

26 predesigned forms for hand calculations in statistical techniques

such as analysis of variance (Latin square) and co-variance, linear
regression, stratified random sample, etc., indicating the different
computational steps. Very useful for small manual calculations,
instructive examples.
 - 199 -

Dixon, W.J. - 1968. BHD - Biomedical Computer Programs. University

of California Press, Berkeley and Los Angeles. 600 pp.

A collection of standard programmes (FORTRAN IV) f originally written for

IBM f094 (approx. 32 K-words). Routines useful in forest inventory,
especially BMP 05D (plotting scatter diagrams and histograms), BMD 06D
(description of strata), BMD 03** (factor analysis), BMD 03R (multiple
regression). Detailed description of computational procedures,
preparation of parameter cards and output. Software available at many
computer centres.

(0 only) DRZ - 1969* Statistische Programme des Deutschen Rechenzentrums (statistical

programmes of the German Calculation Centre), parts A and B.
Deutsches Rechenzentrum Darmstadt (Fed. Rep. of Germany), Programme
information PI 3? and 33. 120 pp.
A collection of statistical standard programmes (FORTRAN IV) as a
complement to the I3MD series. To be noted in particular NRMP (test on
normal distribution), REV (comparison of regression of several strata),
LIPR (check of linearity of equations). Documentation and software from
the German Calculation Centre, Darmstadt, upon request.

Frayer E., Wilson W., Peters R. and Bickford C.A. - 1968. "FINSYS, an
efficient data processing system for large forest inventories". Journal
of Forestry, Vol. 66, No.1?. 4 pp.

General description of this data processing system written in FORTRAN IV

for IHM 7094/7040 computer system (3? K-words of CPU capacity required)
and covers the following designs: 1. Complete enumeration (100 percent);
2, Simple random or systematic sampling; 3 Stratified random sampling;
4. Double sampling for stratification.

(F only) Quinaudeau F. - 1973* Experience acquise par le CTFT en matiere de

traitement automatique des donntfes d'inventaire (Experience gained by
Centre Technique Fores tier Tropical - Nogent sur I^arne, France - in
automatic processing of inventory data). Meeting of IUFRO Subject Group
54*02 (Forest Inventory) in Nancy (France), June 19H Proceedings to
be published. 10 pp.

The paper describes the data processing method developed at CTFT for the
processing of field inventory data in tropical high forests. Examples of
field sheet and output attached.

Haller K.E. - 1968. - A computer

Inventory of national tropical forests
programme for the processing of data. Unasylva, Vol.22, No. 89. 7 pp

The paper gives primarily a thorough problem analysis of a data processing

system for forest inventory. The system, designed for an IBM 1401, is of
less importance because it is written in Autocoder language and specifically
designed for an inventory in Liberia.

MoCraoken D.D. - 1?63. Digital computer programming. J. Wiley and Son*

Inc. f New York, London. 240 pp.

Textbook for advanced hone tuition of how to solve problems on modern

digital computer*. Recommended for additional reference. Useful index
of subject*.
 - 200 -

(0 only) Muller K,H. and Streker I. - 1970. FORTRAN Programmierauleitung (FORTRAN

Programming Manual;. Hochschultaschenbucher, Vol. 804. Bibliographiechce
Institut Mannheim, Wien, Zurich. 140 pp.

Very handy and comprehensive manual for self tuition with application of
programming in FORTRAN IV (IBM) and FORTRAN V (UNTVAC). Some basic
knowledge of FORTRAN' recommended.

Nilsaon, N.E. - 1j67. Some views on data processing problems in forest

inventories. Report on FAO Headquartrre meeting of forest inventory
experts, Rome, FO:3F/6f/W Vf.
In the paper the role of data processing in forest inventory is treated
some recommendations including flow-charts for the organization of data
processing are given.

(g) Considerations on inventory designs (Chapter VII )

(i ) Continuous forest inventory

Ware K.D. and Cunia T. - 196?. Continuous forest inventory with partial
replacement of samples. Forest Science Monograph 3 40 pp.
f

Comprehensive study on SPR including mathematical formulation and cost

analysis.

(ii ) Sequential sampling

Chacko V.J. - 1966. Sequential sampling in forest insect surveys and

diseases. The Indian Forester, Volume 92, No. 4 (pp ^33-?39).

Theory and use of sequential sampling analysis in the case of binomial

distribution (absence or presence of insects or of damage) or negative
binomial distribution (contagious type of distribution applying to insect
counts ).
FAO TECHNICAL PAPERS

FAO FORESTRY PAPERS

1 Forest utilization contracts on public land, 1977 31 Appropriate technology in forestry, 1982 (E)
(E F S) 32 Classification and definitions of forest products,
2 Planning forest roads and harvesting systems, 1982 (Ar/E/F/S)
1977 (EFS) 33 Logging of mountain forests, 1982 (E F S)
3 World hat of forestry schools, 1977 (E/F/S) 34 Fruit-bearing forest trees, 1982 (E F S)
3 Rev. 1 World list of forestry schools. 1981 (E/F/S)
. 35 Forestry in China, 1982 (C E)
3 Rev. 2. World list of forestry schools, 1986 (E/F/S) 36 Basic technology in forest operations, 1982 (E F S)
4/1 World pulp and paper demand, supply and trade 37 Conservation and development of tropical forest
- Vol. 1. 1977 (E F S) resources, 1982 (E F S)
4/2 World pulp and paper demand, supply and trade 38 Forest products prices 1962-1981, 1982 (E/F/S)
- Vol. 2. 1977 (E F S) 39 Frame saw manual, 1982 (E)
5 The marketing wood, 1976 (E S)
of tropical 40 Circular saw manual, 1983 (E)
6 National parks planning, 1976 (EPS") 41 Simple technologies for charcoal making, 1983
7 Forestry for local community development, 1978 (EFS)
(Ar EFS) 42 Fuelwood supplies in the developing countries,
8 Establishment techniques for forest plantations. 1983 (Ar EFS)
1978 (ArC E" F S) 43 Forest revenue systems m developing countries,
9 Wood chips - production, handling, transport. 1983 (E F S)
1976 (C E S) 44/1 Food and fruit bearing forest species
1O/1 Assessment of logging costs from forest - 1. Examples from eastern Africa, 1983 (E F S)
inventories in the tropics 44/2 Food and fruit-bearing forest species
- 1 Principles and methodology, 1978 F S) - 2. from southeastern
. (E Examples
10/2 Assessment of logging costs from forest Asia. 1984 (E F S)
inventories in the tropics 44/3 Food and fruit-bearing forest species - 3. Examples
- 2 Data collection and calculations, 1978 (E F S) from Latin America, 1986(ES)
1 1 Savanna afforestation in Africa, 1977 (E F) 45 Establishing pulp and paper mills, 1983 (E)
12 China: forestry support for agriculture, 1978 (E) 46 Forest products prices 1963 1982. 1983 (E/F/S)
13 Forest products prices 1960 1977. 1979 (E/F/S) 47 Technical forestry education - design and
14 Mountain forest roads and harvesting, 1979 (E) implementation, 1984 (E F S)
14 Rev. 1 Logging and transport in steep terrain. 1985 (E)
. 48 Land evaluation for forestry, 1984 (C E F S)
15 AGRIS forestry
- world catalogue of information 49 Wood extraction with oxen and agricultural
and documentation services, 1979 (E/F/S) tractors, 1986 (E F S)
16 China: integrated wood processing industries, 1979 50 Changes in shifting cultivation in Africa, 1984 (E F)

(E F S) 50/1 Changes in shifting cultivation in Africa

17 Economic analysis of forestry projects, 1979 -
seven case-studies, 1985 (E)
(EFS) 51/1 Studies on the volume and yield of tropical forest
17 Sup. 1. Economic analysis of forestry projects: stands - 1. Dry forest formations. 1989 (E F)
case studies. 1979 (E S) 52/1 Cost estimating in sawmilling industries: guidelines,
17 Sup. 2. Economic analysis of forestry projects: readings, 1984 (E)

1980(CE) 52/2 Field manual on cost estimation in sawmilling

18 Forest products prices 1960 1978, 1980 (E/F/S) industries, 1985 (E)

19/1 Pulping and paper making properties of 53 Intensive multiple-use forest management m Kerala,
fast-growing plantation wood species 1984 (E F S)
- Vol. 1. 1980 (E) 54 Planificaci6n del desarrollo forestal. 1984 (S)
19/2 Pulping and paper-making properties of 55 Intensive multiple-use forest management in the
fast-growing plantation wood species tropics, 1985 (E F S)
- Vol. 2, 1980 (E) 56 Breeding poplars for disease resistance. 1985 (E)

20 Forest trae improvement, 1 985 (C E F S) 57 Coconut wood - processing and use. 1985 (E S)

20/2 A guide to forest seed handling, 1 985 (E S ) 58 Sawdoctoring manual, 1985 (E S)

21 Impact on soils of fast-growing species in lowland 59 The ecological effects of eucalyptus. 1985
humid tropics, 1 980 (E F S) (C E F S)
22/1 Forest volume estimation and yield prediction 60 Monitoring and evaluation of participatory forestry
- Vol. 1. Volume estimation, 198O (C E F S) projects, 1985 (E F S)
22/2 Forest volume estimation and yield prediction 61 Forest products prices 1965-1984, 1985 (E/F/S)
- Vol. 2. Yield prediction. 1980 (C E F S) 62 World list of institutions engaged in forestry and
23 Forest products prices 1961 1980, 1981 (E/F/S) forest products research. 1985 (E/F/S)
24 Cable logging systems, 1981 (C E) 63 Industrial charcoal making, 1985 (E)
25 Public forestry administrations in Latin America, 64 Tree growing by rural people, 1985 (Ar EFS)
1981 (E) 65 Forest legislation m selected African countries,
26 Forestry and rural development, 1981 (EFS) 1986 (E F)
27 Manual of forest inventory. 1981 (E F) 66 Forestry extension organization. 1986 (C E S)
28 Small and medium sawmills in developing countries, 67 Some medicinal forest plants of Africa and Latin
1981 (ES) America, 1986 (E)

29 World forest products, demand and supply 1990 68 Appropriate forest industries, 1986 (E)
and 2000, 1982 (E F S) 69 Management of forest industries. 1986 (E)

30 Tropical forest resources. 1982 (E F S)

 - an
70 Wildland fire management terminology. 1986 92 Forestry policies in Europe analysis, 1989 (E)

(E/F/S) 93 Energy conservation in the mechanical forest

71 World compendium of forestry and forest products industries, 1990 (E S)

research institutions, 1986 (E/F/S) 94 Manual on sawmill operational maintenance,

72 Wood gas as engine fuel, 1986 (E) 1990 (E)

73 Forest products: world outlook projections 95 Forest products prices 1969-1988, 1990 (E/F/S)
1985-20OO, 1986 (E/F/S) 96 Planning and managing forestry research: guideline!
74 Guidelines for forestry information processing, for managers, 1 990 (E)

1986 (E) 97 Non-wood forest products: the way ahead,

75 An operational guide to the monitoring and 1991 (ES)
evaluation of social forestry in India, 1986 (E) 98 Les plantations a vocation de bois d'oouvre en
76 Wood preservation manual. 1986 (E) Afrique mtertropicale humide, 1991 (F)
77 Databook on endangered tree and shrub species 99 Cost control in forest harvesting and road
and provenances, 1 986 (E) construction, 1992 (E)
78 Appropnate wood harvesting in plantation forests, 100 Introduction to ergonomics in forestry in developing
1987 (E) countries, 1992 (E)
79 Small-scale forest based processing enterprises, 101 Amenagement et conservation des fordts denses or
1987 (E F S) Amenque tropicale, 1992 (F)
80 Forestry extension methods, 1987 (E) 102 Research management in forestry, 1991 (E)
81 Guidelines for forest policy formulation, 1987 (C E) 103 Mixed and pure forest plantations in the tropics and
82 Forest products prices 1967 1986. 1988 (E/F/S) subtropics, 1992 (E)
83 Trade in forest products* a study of the barriers 104 Forest products prices 1971-1990, 1992 (E)
faced by the developing countries, 1988 (E) 105 Compendium of pulp and paper training and
84 Forest products: world outlook projections research institutions, 1992 (E)

1987-200O - product and country tables. 1988

(E/F/S)
85 Forestry extension curricula, 1988 (E/F/S)
86 Forestry policies in Europe, 1988 (E) Availability December 1992
87 Small-scale harvesting operations of wood and
non-wood forest products involving rural people, Ar Arabic Murtil - Multilingual
1988 (E FS) C Chinese Out of print
88 Management of tropical moist forests in Africa, E English In preparation
1989 (EFP) F French
89 Review of forest management systems of tropical P Portuguese
Asia. 1989 (E) S Spanish
90 Forestry and food security, 1989 (Ar E S)
91 Design manual on basic wood harvesting The FAO Technical Papers are available through the authorized
technology. 1989 (E F S) FAO Sales Agents or directly from Distribution and Sales Section,
(Published only as FAO Training Series, No 18) FAO, Vnle delle Terme di Caracalla, OO1OO Rome, Italy.