RemembeRINGs
The Development and Application of Local and Regional Tree-Ring
Chronologies of Oak for the Purposes of Archaeological and Historical
Research in the Netherlands
Esther Jansma
COLOFON
NAR 19 is een ongewijzigde uitgave van het academisch proefschrift van E. Jansma.
/ This volume of the NAR presents the original version of the Ph.D. thesis of Ms E. Jansma.
Uitgever / Publisher Rijksdienst voor het Oudheidkundig Bodemonderzoek, A. Haytsma
Kerkstraat 1, 3811 CV Amersfoort / Postbus 1600, 3800 BP Amersfoort, the Netherlands
Redactie / Editors nr. 19 Prof.Dr. W. Groenman-van Waateringe, Prof.Dr. H. Kars (eindred.),
Prof.Dr. J. Strackee, Dr. A. Voorrips
Pre-press & layout ROB, Amersfoort
Omslagontwerp / Coverdesign Ingrid Jansen, Amersfoort; foto / photograph Ernst Voller, Berlin
Druk / Printed by Casparie Heerhugowaard B.V.
Besteladres / Distribution U kunt dit boek bestellen door fl. 25,-- plus fl. 7,50 verzendkosten
over te maken op postbanknummer 4792702 ten name van de Joan Willems Stichting te
Amersfoort onder vermelding van NAR 19. / You can order this volume from the Joan
Willems Stichting, Kerkstraat 1, 3811 CV Amersfoort, the Netherlands,
tel. +31 033 - 4634233, fax (31) 033 - 4653235, E-mail info@archis.nl
Subscription fl. 25,--, postage and packing are not included.
CIP-GEGEVENS KONINKLIJKE BIBLIOTHEEK, DEN HAAG
Jansma, Esther
RemembeRINGs : the development and application of local and regional tree-ring chronologies
of oak for the purposes of archaeological and historical research in the Netherlands / Esther
Jansma. - Amersfoort : Rijksdienst voor het Oudheidkundig Bodemonderzoek. - Ill. (Nederlandse archeologische rapporten, ISSN 0169-3859 ; nr. 19) Ook verschenen als
proefschrift Universiteit van Amsterdam, 1995 - Met lit. opg.
ISBN 90-73104-26-2
Trefw.: archeologie ; Nederland / dendrochronologie.
All rights reserved
© 1995 E. Jansma & ROB, Amsterdam & Amersfoort
ISBN 90-73104-26-2
ISSN 0169-3859
RemembeRINGs
The Development and Application of Local and Regional Tree-Ring
Chronologies of Oak for the Purposes of Archaeological and Historical
Research in the Netherlands
ACADEMISCH PROEFSCHRIFT
ter verkrijging van de graad van doctor
aan de Universiteit van Amsterdam,
op gezag van de Rector Magnificus
Prof. Dr. P.W.M. de Meijer
ten overstaan van een door het college van dekanen ingestelde
commissie in het openbaar te verdedigen in de Aula der Universiteit
op maandag 4 december 1995 te 12.00 uur
door Esther Jansma
geboren te Amsterdam
Promotores:
Prof. Dr. J. Strackee
Prof. Dr. W. Groenman-van Waateringe
Co-promotor:
Dr. A. Voorrips
CONTENTS
11
PREFACE
1
1.1
1.2
1.3
1.4
2
2.1
2.2
2.2.1
2.2.2
2.2.3
2.3
2.4
3
3.1
3.2
3.3
3.4
3.5
3.6
INTRODUCTION
Dating in archaeology and historical research
A brief history of dendrochronology
Dendrochronology in the Netherlands
Research outline
THE STATISTICAL PROPERTIES OF ‘MEAN SENSITIVITY’ - A REAPPRAISAL
(Corrected version of Strackee, J. and E. Jansma, 1992. Dendrochronologia
10, 121-135)
Abstract
Introduction
Statistical aspects
General
The bivariate Gamma density function
The bivariate Lognormal density function
Experimental assessment
Discussion
15
15
17
18
19
23
23
23
25
25
26
28
30
30
DENDROCHRONOLOGICAL METHODS OF DETERMINING THE ORIGIN OF OAK
TIMBER: A CASE STUDY ON WOOD FROM ’S-HERTOGENBOSCH
(Corrected version of Jansma, E., 1992. Helinium 32, 195-214)
Abstract
Introduction
Material
Statistical method
Results
Discussion
Conclusion
33
33
33
36
40
42
44
45
4
AN 1100 YEAR TREE-RING CHRONOLOGY OF OAK FOR THE DUTCH COASTAL REGION
(2258 - 1141 BC) (Jansma, E., 1995. In: J. S. Dean, D. M. Meko and
T. W. Swetnam (eds.), Tree-Rings, Environment and Humanity - Proceedings
of the International Tree-Ring Conference 1995, Univ. of Arizona (provisional title).
Radiocarbon, Tucson (in print))
47
Abstract
Introduction
Material
Methods
Results
Discussion
Conclusion
4.1
4.2
4.3
4.4
4.5
4.6
5
OAK TREE-RING CHRONOLOGIES FOR THE NETHERLANDS BETWEEN
AND AD
5.6
5.7
325 BC
57
57
57
58
60
60
60
61
62
62
62
65
65
65
65
66
67
MEDIEVAL TREE-RING CHRONOLOGIES OF OAK FROM DUTCH
ARCHAEOLOGICAL AND HISTORICAL SITES (AD
AD
6.1
6.2
6.3
6.3.1
6.3.2
6.3.3
6.3.4
6.3.5
6.3.6
6.4
6.4.1
6.4.2
6.4.3
6.4.4
563 (Jansma, E., 1994. Helinium 34 (in print))
Abstract
Introduction
Terminology
Material
Methodology
Methods of dating
Chronology development
Estimating the chronology signal
Results
General
The local eastern chronology NLRom_E (AD 190-395)
Local western chronology NLRom_W1 (84 BC - AD 50)
Local western chronology NLRom_W2 (140 BC - AD 87)
The regional low altitude chronology NLRom_R
(325 BC - AD 563)
Interpretation and discussion
Conclusion
5.1
5.2
5.3
5.4
5.4.1
5.4.2
5.4.3
5.5
5.5.1
5.5.2
5.5.3
5.5.4
5.5.5
6
47
47
49
50
51
54
55
427 - 1752 ;
1023 - 1666 ; AD 1041 - 1346)
Abstract
Introduction
Material
Methodology
Quality control
The detrending of tree-ring patterns
Methods of dating
The clustering of tree-ring index series
Assessment of the reliability of tree-ring chronologies
The environmental signal of the Dutch cluster chronologies
Results
General
The signal within the chronologies
Growth regions
Crossdating with existing master chronologies
69
69
69
69
69
69
70
70
70
71
71
73
73
73
75
76
6.5
6.5.1
6.5.2
6.5.3
6.5.4
Interpretation and discussion
Chronology NLHist_1 (AD 427 - 1752)
Chronology NLHist_2 (AD 1023 - 1666)
Chronology NLHist_3 (AD 1041 - 1346)
The use of the cluster chronologies for dating Dutch
sites and objects
Conclusions
6.6
7
79
79
THE MEAN CORRELATION TECHNIQUE: THE ‘EFFECTIVE CHRONOLOGY
SIGNAL’ AS AN ESTIMATOR OF THE SIGNAL IN TREE-RING CHRONOLOGIES
Abstract
Introduction
The Mean Correlation Technique
The behaviour of r̄eff
Discussion
The domain of r̄eff and r̄bt
Estimating the mean of correlation coefficients
The hyperbolical relationship between r̄wt and r̄eff
Conclusion
7.1
7.2
7.3
7.4
7.4.1
7.4.2
7.4.3
7.5
8
77
77
78
79
SYNTHESIS: DISCUSSION AND CONCLUSIONS
8.1
8.2
Introduction
Methodological aspects: the estimation of the signal in tree-ring
chronologies
The application of chronologies from Dutch contexts
Prehistory
The Iron Age/Roman period
The Merovingian and Carolingian periods
The Late Middle Ages
8.3
8.3.1
8.3.2
8.3.3
8.3.4
81
81
81
82
84
85
85
85
86
87
89
89
89
93
93
95
96
97
SAMENVATTING
99
REFERENCES
107
APPENDIX
A-
Absolutely dated tree-ring series of oak from anthropogenic and
natural sites in the Netherlands
115
APPENDIX
B-
Tree-ring chronologies of oak used for dating
139
APPENDIX
C-
Average growth-index values of some Dutch chronologies
NLPre_ZH (2258 - 1141 BC)
NLRom_E (AD 190-395)
NLRom_W1 (84 BC - AD 50)
NLRom_W2 (140 BC - AD 87)
NLRom_R (325 BC - AD 563)
143
Perhaps old mayflies sit around complaining how life this minute
isn’t a patch on the good old minutes of long ago. Whereas the trees,
which are not famous for their quick reactions, may just have time
to notice the way the sky keeps flickering before the dry rot and
woodworms set in.
Terry Pratchett
PREFACE
In this volume absolutely dated tree-ring chronologies are presented from oak
series derived from Dutch anthropogenic and natural sites from the past. This
study has been funded by the ‘Albert Egges van Giffen’ Institute for Pre- and
Protohistoric Archaeology (IPP, Univ. of Amsterdam (NL); 1985 - 1987), the
archaeological section of the Netherlands Organization for Scientific Research
(ARCHON/NWO, Doss. No. 280-151-040; 1987-1991); the Dutch State
Service for Archaeological Soil Research (ROB; 1991 - present); and the many
archaeological and historical organizations that from 1985 onwards funded the
dating of archaeological structures and historic buildings conducted by the
former tree-ring laboratory of the IPP and the current Dutch Centre for
Dendrochronology (RING Foundation, ROB). The verification of the datings
on which this study is based was funded by the EC as part of the project
‘Temperature Change over Northern Eurasia during the last 2500 Years’ (Contract
no. CV5V CT94 0500; 1994-1996; under supervision of the Climatic
Research Unit, Univ. of East Anglia (GB)). The tree-ring data discussed in
this thesis that date after 500 BC were incorporated in this project in January
1995; the remaining tree-ring series will be incorporated in 1996 as part of the
continuation of this project, entitled ‘Analysis of Dendrochronological Variability
and Associated Natural Climates in Eurasia - the last 10,000 years’ (1996-1998,
forthcoming).
Parts of this study have appeared previously in the journals
Dendrochronologia (Chapter 2) and Helinium (Chapter 3; Chapter 5 (in print));
Chapter 4 is published in the proceedings of the 1994 International Tree-Ring
Conference (Tucson, Univ. of Arizona (US)).
Apart from my promoters Dr. J. Strackee and Dr. W. Groenman-van
Waateringe and my co-promoter Dr. A. Voorrips, many people have
contributed to the separate chapters in this volume: Dr. A. van Oosterom
(Dept. of Medical Physics, Univ. of Nijmegen (NL)) was consulted during the
analyses described in Chapter 2; A. van Drunen and E. Vink (Bouwhistorische
Dienst ’s-Hertogenbosch (NL)) provided part of the historical facts mentioned
in Chapter 3; H. van Haaster and Dr. J.P Pals (IPP) made valuable
suggestions on the content of Chapter 3; the bog oak samples discussed in
Chapter 4 were collected by V. van Amerongen (IPP), P. van Rijn (RING)
and members of ARCHEON Archaeological Theme Park (Alphen a/d Rijn
(NL)); V. van Amerongen also sampled part of the bog oaks discussed in
Chapter 5; and P.P.Th.M. Maessen (Holtland Dendroconsult, Veenendaal
(NL)) gave valuable suggestions on the classification of ecological growth
regions in the Netherlands used in Chapter 6.
This study would have been impossible without the earlier research of
Dr. J. Bauch and Dr. D. Eckstein (Ordinariat für Holzbiologie, Univ. of
Hamburg (D)), and Dr. J.A. Brongers (ROB); they were the first to apply
dendrochronology in a systematic way as a dating technique on tree-ring series
from Dutch contexts and objects, and their work has been of paramount
importance in convincing archaeologists and historians in the Netherlands that
Preface
Dutch timbers from the past can be dated dendrochronologically. I am grateful
to Dr. B. Schmidt (Labor für Dendrochronologie, Institut für Ur- und Frühgeschichte, Univ. of Köln (D)) for the dendrochronological training I received
in 1984; to Dr. H. Kamermans (IPP) for helping me to develop software in
1984 for crossdating and printing tree-ring curves; and to Dr. M.A.R. Munro
(Paleoecology Centre, Queens Univ. of Belfast (Ireland)) for supplying his
crossdating program for the Apple IIe computer in 1986.
This study could not have taken place without the support of an
international group of dendrochronologists for the development of tree-ring
research in the Netherlands as a dating technique and as an independent
analytical field with applications in archaeological and historical studies,
geology, forestry and climatology. I especially thank Dr. F.H. Schweingruber
(Eidgenössische Forschungsanstalt für Wald, Schnee und Landschaft,
Birmensdorf (CH)) for introducing me to dendrochronology, creating an
European network of young dendrochronologists through the annual Swiss
fieldweek, and continuously reminding the statisticians and archaeologists
among us that a tree is a living organism. I am very grateful for the biological,
analytical and practical background provided by R.K. Adams, Dr. H.C. Fritts,
Dr. L.J. Graumlich, T.P. Harlan, R.L. Holmes, Dr. M.K. Hughes,
G.R. Lofgren, Dr. T.W. Swetnam and Dr. F.W. Telewski of the Laboratory of
Tree-Ring Research (LTRR, Univ. of Arizona (US)), where I spent the
academic year 1989/1990 as a visiting scholar and turned into a
dendrochronologist. The analytical programs distributed by the International
Tree-Ring Data Bank (ITRDB; NOAA Palaeoclimatology Program/World Data
Centre A; Boulder, Colorado (US)), which have been written by, or after
suggestions of R.K. Adams (LTRR), Dr. E.R. Cook (Lamont-Doherty
Geological Observatory, Univ. of Columbia (US)), Dr. H.C. Fritts (LTRR),
Dr. H.D. Grissino-Mayer (LTRR) and R.L. Holmes (LTRR) have been of
great value during all parts of the analysis. Dendrochronological dating requires
access to absolutely dated tree-ring chronologies, and I thank all
dendrochronologists who have made their chronologies accessible through
publications or personal communication. Most helpful were chronologies
provided by: Dr. D. Eckstein (Ordinariat für Holzbiologie, Univ. of Hamburg
(D)); Dr. A. Delorme (Institut für Forstbenutzung, Univ. of Göttingen (D));
Dr. P. Hoffsummer (Laboratoire de Dendrochronologie, Univ. of Liège (B));
Dr. H.-H. Leuschner (Institut für Palynology, Univ. of Göttingen (D));
J. Hillam (Archaeology Research School, Univ. of Sheffield (GB)); H. Tisje
(Neu-Isenburg (D)); I. Tyers (Museum of London (GB)); and Dr. T. Wazny
(Academy of Fine Arts, Conservation Faculty, Warschau (P)).
The support of the Dutch archaeological and historical community is greatly
appreciated. Dr. W.A. Casparie (Biological Archaeological Institute, Univ. of
Groningen (NL)), Dr. H.A. Heidinga (IPP), Dr. H. Kars (RING/ROB) and
Dr. D.J. de Vries (State Service for the Preservation of Historic Buildings and
Monuments (RDMZ), Zeist (NL)) have helped me considerably throughout
the years.
My deepest gratitude goes to Dr. J. Strackee (Laboratory of Medical Physics
and Informatics, Univ. of Amsterdam) who has guided me through many
statistical labyrinths and eliminated my fear of formula’s by his elegant and
practical use of blackboard and chalk; Dr. W. Groenman-van Waateringe (IPP),
who never stopped believing that the current volume would be produced and
always picked the right time to remind me of this; Dr. A. Voorrips and
S. Loving (IPP), who during the past decade have garnished their theoretical,
practical and moral support with marvellous dinners and conversations about
poetry; Dr. H. Kars who for years has been supporting dendrochronology as a
science; Dr. H.-H. Leuschner who has welcomed me into his laboratory and
home, has introduced me to ‘big science’ (the more exciting elements of current
dendrochronological research in Northwest Europe), and as long as I know him
has been such a loyal and inspiring colleague; P.P.Th.M. Maessen, whom I can
talk to endlessly about all kinds of forest research; Dr. H.D. Grissino-Mayer,
who has always reminded me that science and fun are not mutually exclusive
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
and in the process has provided me with excellent computer programs, his
extensive bibliography and all reprints I could wish for; and my co-workers
Pauline van Rijn and Elsemieke Hanraets (RING), who have run the tree-ring
lab of RING so excellently during the periods I worked on this thesis and
without whom the dating business would have been a lone venture indeed.
I thank Sue McDonnell for the correction of the English text, Ingrid Jansen
(ROB) for designing the current volume and Bert van As (ROB) for typing all
the formula’s and general advice.
Mireille, Maaike and Barbara Jansma have patted my back during all phases
of this research. My friends Marion Oskamp, Rita Horst and Inger Kolff have
hauled me through the more difficult times, as has my friend and colleague in
poetry Nydia Ecury. This study could not have been completed without
Casper le Fèvre, who stuck with me for better and worse during the past
decade, never tired of discussing tree-rings with me, and has given me all
support a female doctor-to-be asks for in this liberated era - which is a lot.
Thank you all.
1
INTRODUCTION
1.1
1. The French physicists Antoine Henri
Becquerel (1852 - 1908), Marie Curie
(1867 - 1934) and Pierre Curie (1859 1906).
2. Varve analysis was developed by Swedish
scientists in the early twentieth century. A
varve is a sediment layer or a sequence of
sediment layers deposited within a single year
in stagnant water. Geologists can establish
the age of a geological event in years from the
number of varve units deposited after this
event (Dalrymple 1991).
3. Obsidian hydration dating calculates the
age of volcanic glass from 200 to 200 000
years old by determining the thickness of
hydration rinds produced by water vapour
diffusing into their freshly chipped surfaces
(Dalrymple 1991).
4. Thermoluminescence dating is based on
the phenomenon of natural ionizing radiation
inducing free electrons in a mineral that can
be trapped in defects of the mineral’s
structure. These trapped electrons escape as
thermoluminescence when heated to a
certain temperature. By recording the
thermoluminescence of a mineral the last
drainage of the trapped electrons can be
dated back to several hundred thousand years
(assuming a constant natural radiation level;
O’Reilly 1984).
DATING IN ARCHAEOLOGY AND HISTORICAL RESEARCH
The dating of structures and objects from the past is a prerequisite for
most archaeological and historical research. Different methods have been
devised to date the past. These approaches can be classified into a number of
categories (Michels 1973): (a) periodization: the delineation of synchronic
segments, (b) relative dating: establishing the correct order of events, and
(c) absolute dating: relating events to an absolute time scale.
The earliest dating method, developed in the nineteenth century and often
used in geology and archaeology, was based on relative time scales determined
on the basis of stratigraphy. The main principle of stratigraphy is the ‘law of
superposition’, which states that in an undisturbed succession of strata, the
youngest deposits are on top and the oldest ones at the bottom. Dating by
means of stratigraphy can give only relative dates. The development of
radiometric methods, after the discovery of radioactivity at the end of the
nineteenth century1, led to the first breakthroughs in establishing an absolute
time scale. In addition to radiometric techniques, such as the radiocarbon
method, in the twentieth century other absolute methods (with varying
chronological precision) were devised, among which are varve analysis2,
obsidian hydration dating3, thermo-luminescence dating4, and
dendrochronology.
In research of the past, the intended precision of dating depends on the
investigated period and region (the research topic), whereas the feasible
precision is determined by the material remains that are available and the
dating methods that are used. In the Netherlands, for instance, the intended
chronological precision with regard to the tenth century is high, because a
major cultural shift took place (the transition from the Carolingian period to
the Late Middle Ages) and it is not clear when exactly this happened. The
feasible precision in the tenth century is, however, low, because ceramics
associated with the remains of settlements from this century cannot be placed
on a refined time scale and, due to the spread of Christianity, finds associated
with burials are scarce.
Besides stratigraphy, modern archaeology uses dating methods such as
typochronology (the application of more or less precisely dated calendars of
artefact types), the radiocarbon method and dendrochronology. Building history
uses typochronology, dendrochronology, written sources and additional data.
For certain periods and regions, dating by means of typochronology results in
chronological information that is exact to a few decades; applied to imported
ceramics, for example, it results in satisfactory dates for Roman and Iron Age
structures throughout the Netherlands (1st to 4th/5th century AD).
Archaeological sites from this period that do not contain imported ceramics as
well as sites from the following centuries are more difficult to date. In the
northern Netherlands the period between the fifth and eighth century and the
tenth century are problematic in terms of dating; in the southern Netherlands
this applies to the sixth, early seventh and tenth century; the province of
Drenthe is difficult throughout the Early Middle Ages (AD 500 - 1000). The
chapter 1
| Introduction
5. Conifers sometimes form one or more
‘false’ rings during a single year. Using a
microscope these can, however, be
distinguished from annual rings.
6. Sometimes a dendrochronological date
is subject to a certain margin, in
consequence of the fact that the outer, last
formed, rings are not present in a wood
sample. In this case the year of death is
estimated by adding an estimated number
of (missing) growth rings to the last ring in
the sample. The rings that are present in
the sample are, however, dated to the year.
7. Oak settled in Europe after the last ice
age ended (around 8000 BC). Only a few
Northwest European chronologies of oak
extend back this far; the majority covers the
last thousand years only.
fact that it is sometimes difficult to date archaeological sites from the Late
Middle Ages (AD 1000 - 1500) by, for instance, ceramics and brick types, is
illustrated by the discrepancy between the typochronological and dendrochronological dates of the recently excavated ‘Castle of the Lords of Amstel’
(Amsterdam; Jansma and Kars 1995). To summarize, we can say that for many
historical periods typochronology does not result in the precise dates that are
required within the research setting. In order to refine dates established through
typochronology, and for independent verification, physical dating techniques
are required.
The radiocarbon method, first developed by the American chemist W. F. Libby
at the University of Chicago in 1947, estimates the age of organic material on
the basis of the fraction of radioactive carbon it contains. During their life,
organisms take up carbon from the atmosphere (13C and 14C). A fraction of this
assimilated carbon is radioactive (14C). After the death of an organism, this
fraction decreases according to a known rate; 14C has a half-life of 5730 ± 40
years. The approximate moment of death can be calculated from the remaining
fraction of 14C in organic material. The radiocarbon method can be used to
date material up to 50 000 years old, although it is sometimes extended to
70 000 years.
Radiocarbon estimates of dendrochronologically dated tree rings have shown
that the fraction of 14C in organisms at the time of death is variable, i.e.,
atmospheric 14C is not constant. In the curve of 14C measured in tree rings
dated by means of dendrochronology, the ‘radiocarbon calibration curve’,
changes of atmospheric 14C show up as wiggles and horizontal intervals
(‘plateaus’); the precise age of organisms that lived during these periods cannot
be deduced from their radiocarbon content. Furthermore, a radiocarbon date is
often based on a single measurement of 14C, which has an inherent uncertainty.
Another basic problem is postdepositional contamination of the material
(incorporation of younger and older carbon) by, for example, percolating
groundwater and by contamination during and after sampling. Wiggles and
plateaus in the calibration curve and statistical fluctuation mean that
radiocarbon dates often have a broad chronological margin. The method does
not improve datings of material from periods that are well-known
chronologically, such as the Roman period (AD 1 - 450) and the Late Middle
Ages (AD 1000 - 1500), and periods that have anomalous values of 14C (e.g.,
800 - 400 BC, the 14C ‘Hallstatt Plateau’).
Dendrochronology makes use of the growth rings in trees. In tree species from
the temperate climatic zone the growth rings represent single years, i.e., they
are ‘annual’ rings.5 Because trees grow very old, their trunks contain many
rings: from between 100 - 300 (oak) to 2000 - 5000 (Giant sequoia, Bristlecone
pine). As a result, single tree-ring measurement series consist of many
observations and are more precise than a single observation of 14C.
In Europe, oak is most commonly used for dendrochronological dating
(Quercus robur L. and Quercus petraea (Mattuschka) Liebl.). The growth patterns
of oak reflect climatic conditions that operated over relatively large areas, with
unfavourable conditions expressed as narrow rings and favourable conditions as
wide ones. Because of their similar response to annual weather conditions, the
patterns in oaks that grow in these areas can be matched, i.e. ‘crossdated’. Long
chronologies of oak have been established through crossdating for different
regions in Europe; the patterns in living oaks were matched with the patterns in
timbers used in buildings from earlier times, and these were matched with even
older material, etc. If a measurement series of an undated growth pattern of oak
is matched with a dated oak chronology, a calendar date can be determined;
each ring width in the undated pattern is matched with an annual value in the
dated chronology, including the last ring in the pattern, which is closest to, or
formed during, the year in which the tree died.6 Dendrochronology has a
chronological restriction in comparison to the radiocarbon method: it cannot be
used to date material older than 10 000 years, because absolutely dated oak
chronologies do not extend back further.7 For the last 10 000 years it has an
advantage over all other methods, because it is exact to the year.
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
1.2
8. The authorized text of this lecture was
printed in the newspaper Pasadena Star (19
December 1908, 11-12; Eckstein et al. 1975).
9. Since the mid 1960’s European
dendroclimatology has mostly used
radiodensitometric data, i.e. wood densities
determined by X-ray procedures, since these
are better suited for climate reconstructions
than total ring-width. Radiodensitometry is
only suitable for the research of coniferous
tree species. Since the wood must not be
decomposed, recent material is used and
most density chronologies do not extend back
more than a few centuries. Climate
reconstructions derived from these
chronologies are necessarily short.
A BRIEF HISTORY OF DENDROCHRONOLOGY
The earliest application in Europe of dendrochronology as a climateanalysing method was by the Dutch astronomer J. C. Kapteyn (1851-1922).
He measured the ring widths in groups of Dutch and German oak, produced
chronologies that extended back for several centuries, and compared these to
existing meteorological records (Kapteyn 1914). As early as 1908, Kapteyn
gave a lecture in Pasadena (US) entitled ‘Tree growth and meteorological
factors’.8 It was A. E. Douglass (Flagstaff, Arizona) who formulated the
principles and techniques of dendrochronology as a tool to study climate and
by crossdating built the first long chronologies of tree rings (Douglass 1909;
Douglass 1914; Douglass 1919). The subsequent earliest applications of
dendrochronology mainly involved living trees, concentrating on questions
regarding forestry, biology and climate.
In the 1940’s, B. Huber introduced dendrochronology to Europe in a
systematic way. From tree-ring patterns that reflect more or less continental
growth conditions, long chronologies of oak were developed in Germany (e.g.,
Huber 1941; Huber and Holdheide 1942; Huber et al. 1949). Dendrochronology was considered unsuitable for dating oaks from regions where
oceanic conditions prevail. Botanists and palaeobotanists in England and
Ireland in the 1960’s were still convinced that climate and the multiplicity of
site conditions affecting tree growth would make dendrochronology
unworkable in their countries (Baillie 1982). The relatively high amount of
rainfall was considered an impediment to crossdating; the absence of droughts
was believed to result in tree-ring patterns with a low ‘mean sensitivity’, i.e.
little variation of the annual widths.
The focus of European dendrochronology broadened from the late 1960’s
onwards. Researchers found that the narrow rings in tree-ring patterns of oak,
which allow these patterns to be crossdated over relatively large regions, also
occur in oak growing in wet surroundings and that the causes are also climatic;
for example, the trees respond negatively to spring frost and drastic changes in
the water table. From this time onwards, dendrochronological dating was
more and more routinely applied to oak from wet, oceanic sites and long
‘oceanic’ oak chronologies began to be developed (e.g., Delorme 1974;
Delorme 1976; Baillie and Pilcher 1976). In addition, dendrochronological
methods of studying the European climate were devised (Schweingruber et al.
1978; Schweingruber et al. 1979).
In Europe, palaeodendrochronology concentrates on oak and is focused on
the dating of cultural objects, whereas research on living trees involves a
variety of coniferous species and is focused on climatology (e.g., Hughes 1987;
Hughes et al. 1984; Briffa et al. 1988, Schweingruber et al. 1987;
Schweingruber et al. 1991; Briffa and Schweingruber 1992).9 This approach is
in marked contrast to the US, where most sub-fields of dendrochronology,
including palaeodendrochronology, have always concentrated on climate
studies. There are several reasons for this difference. First, palaeodendrochronologists in northwestern Europe work with relatively short series (mainly
based on oak) in regions where natural old-growth forests no longer exist; if
long chronologies are to be produced, series from historical and archaeological
structures have to be used. Dendrochronologists in the US, on the other hand,
mainly use long-lived coniferous species from natural forests, and the dead
trees in these forests are used to obtain tree-ring series from earlier periods.
Second, the anthropogenic past that can be studied dendrochronologically
extends back further in Europe than in the US, and the demand for
dendrochronological dating of cultural objects in Europe is large. Third,
climate studies using series of oak in northwestern Europe are complicated
and do not result in unambiguous data due to (a) anthropogenic factors such
as forest management, pollution and managed water tables, and (b) the fact
that oak in many regions is not very climate sensitive because it does not grow
at the limits of its natural distribution. Despite these complicating factors the
focus in European palaeodendrochronology is currently shifting from
chronology development to climatology, among other reasons because data
chapter 1
| Introduction
sets of absolutely dated oak tree-ring series are now available up to 8000 BC
(Irish oak: 5289 BC - present (Pilcher et al. 1984); North German oak:
6255 BC - present (Leuschner and Delorme 1988); Central German oak:
8021 BC - present (Becker 1993)). In Europe, climate studies oak have been
undertaken by, among others, Briffa et al. 1986, Hughes et al. 1978 and Kelly
et al. 1989. Climate studies on a large temporal and spatial scale, based on
combined data sets of coniferous species and oak, are currently undertaken by
collaborating dendroclimatological and palaeodendrochronological
laboratories.10
1.3
10. ‘Temperature Change over Northern
Eurasia during the last 2,500 Years’ (EC,
Contract No. CV5V CT94 0500); ‘Analysis
of Dendrochronological Variability and
Associated Natural Climates in Eurasia - the
last 10,000 years’ (EC, forthcoming).
11. At about the same time, Munaut
investigated sub-fossil pine from Terneuzen
(province of Zeeland; Munaut 1966) and
Emmen (Province of Drenthe; Munaut and
Casparie 1971).
12. Of the hundreds of dates obtained for
medieval wood samples from Dutch
contexts, only five were made through
matching with Chronology 1 (Molengat:
1 sample; Genhoes: 2 samples; Yerseke:
2 samples (Appendix A)).
DENDROCHRONOLOGY IN THE NETHERLANDS
In the Dutch archaeological and historical fields there was little interest
in dendrochronology until the dating of oak panel paintings by 17th century
Dutch artists began in 1965 (Bauch 1968; Bauch and Eckstein 1970; Bauch et
al. 1972). In the early 1970’s, J. A. Brongers of the State Service for
Archaeological Research (ROB; Amersfoort) produced chronologies of living
oaks growing on various habitats on the Pleistocene soils, and crossdated these
with German master chronologies (Brongers 1973). His collaboration with
J. Bauch and D. Eckstein (Ordinariat für Holzbiologie, Univ. of Hamburg)
lead to the establishment of Chronology 1 (AD 1385 - 1973) representing living
trees, timbers from windmills and panel paintings dating from after AD 1650.
This chronology represents oak from the eastern higher parts of the
Netherlands and from Germany. Chronology 2 (AD 1140 - 1623), from panel
paintings dated before 1650, was presumed to represent oak from coastal sites
in the Netherlands and England (Bauch et al. 1972; Eckstein et al. 1975; Bauch
1978).11
Efforts to establish a ‘Dutch’ dendrochronology were complicated by
crossdating problems whose causes were not well understood. Unfortunately,
Chronologies 1 and 2 did not result in many new datings of timbers from Dutch
buildings and archaeological structures. We now know that the early interval of
Chronology 1 probably contains an amorphous signal12, and therefore most likely
represents oak from a variety of environmentally different regions in Germany
(and possibly the Netherlands), and that Chronology 2 represents oak that grew
in the Baltic region. This could not be established at the time, for a number of
reasons. First, the extent of Medieval importing of oak timber in the
Netherlands was underestimated, in part because few regional chronologies
from adjacent countries existed that could be used as a reference. Second, the
lack of crossdating among ring-width patterns of ‘Dutch’ timbers from
archaeological and historical contexts was still interpreted as the result of the
oceanic conditions in the Netherlands. Third, computers were not widely
available; tree-ring patterns were measured and plotted by hand, and the
analysing techniques were simple. Fourth, no dendrochronological
infrastructure existed nationally within which these and other research problems
could be solved.
In the early 1980’s, the fact that oak in oceanic regions can be crossdated had
become widely recognized throughout northwestern Europe. (e.g., Baillie and
Pilcher 1976; Baillie 1982; Delorme 1974; Delorme 1976; Delorme et al. 1981;
Leuschner et al. 1987). Furthermore, E. Hollstein (Rheinisches Landesmuseum
Trier) had published his Central German master chronology and the
constituent regional chronologies (690 BC - present), which proved to be well
suited to date oak from Dutch archaeological and historical contexts (Hollstein
1980). As a result, dendrochronology in the Netherlands began to develop
more quickly. Using a home-made computer program, D. J. de Vries of the
State Service for the Preservation of Monuments and Historic Buildings
(RDMZ, Zeist) dated eleven oak staves from Voorst Castle (province of
Overijssel) using the Central German master chronology (De Vries 1983).
I dated oak posts from the Roman fortress Velsen 1 (province of NoordHolland; Jansma 1985) and bog oaks from the vicinity of Abcoude (province of
Utrecht; Jansma 1987). The latter datings, which were also made using the
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
chronologies of Hollstein, led to the establishment of a small
dendrochronological laboratory at the ‘Albert Egges van Giffen’ Institute for
Pre- and Protohistoric Archaeology (IPP, Univ. of Amsterdam) in 1985.13 The
work by J. Bauch, J. A. Brongers and D. Eckstein was followed-up in 1986,
when a dendrochronological laboratory was created at the ROB.14
1.4
13. The programs used for crossdating at the
IPP were written by Munro (1984), and,
after Baillie and Pilcher (1973) and Hollstein
(1980), by Jansma and Kamermans. An
automated measuring table was designed by
Jansma and built at the instrument workshop
of the Medical Faculty of the University of
Amsterdam.
14. The crossdating and plotting programs
used at the ROB were written by Aniol
(1983), who also designed the automated
measuring equipment (Aniol 1987). The
dendrochronological datings between 1986
and 1991 were carried out by P. Schut
(1986-1987); M. van Veen (1987); G.
Vervoort (1987-1988); A. Runhardt (1988);
and E. Hanraets (1988-1991), under the
supervision of J. A. Brongers (ROB).
RESEARCH OUTLINE
The main objective of this study was to provide archaeology and
building history in the Netherlands with a dendrochronological yardstick
based on oak, sufficiently refined to date Dutch archaeological and historical
contexts which otherwise could not be dated, or only roughly. In the
archaeological field, in the first millennium BC, these are contexts from the
period between 800 and 400 BC, for which the radiocarbon method does not
result in precise enough dates. In the first millennium AD, these include
contexts from the Iron Age/Roman period that do not contain imported
ceramics, as well as the period between the fifth and eighth centuries in the
northern Netherlands, the sixth and early seventh century in the southern
Netherlands, and the tenth century throughout the country. Building history
requires precise datings for any historical period from which buildings exist,
i.e. the twelfth century and later, in order to link construction dates of (phases
of) buildings to events described in written sources.
The Dutch Chronology 1 could be matched with German chronologies,
whereas Chronology 2 could not (Eckstein et al. 1975). This indicated that in
the Netherlands distinct, and unique, groups of indigenous oak exist.
Therefore, methods had to be selected, or new ones developed, to distinguish
between these groups.
The first efforts to analyse the different signals in tree-ring series were based
on the dendrochronological parameter ‘Mean Sensitivity’ (); see Chapter 2).
This parameter, developed by A. E. Douglass (1928), was commonly used in
early dendrochronology to estimate the variability and climate sensitivity in
tree-ring series (e.g. Fritts 1976). In Europe this parameter played a role in
the argument that oak from oceanic regions is less suited for dendrochronological analyses (Fürst 1963; Fürst 1978). I used to compare the signal in
dated growth patterns of oaks from the Iron Age/Roman period that in all
likelihood grew in surroundings with different hydrological characteristics (wet
versus dry). The purpose was to establish whether marked differences occur in
their values for over time, and whether these differences reflect the growth
response of oak to different types of environment. Although different signals
were indeed found (Jansma, unpublished data), this approach was abandoned;
in collaboration with Prof. Dr. J. Strackee (Department of Medical Physics
and Informatics, Univ. of Amsterdam) it was established that is an
ambiguous parameter that is dependent on other statistical parameters (the
auto-correlation and standard deviation of a time series), and that it does not
give an accurate estimate of the variability present in a time series (Chapter 2).
In the early phases of this study two assumptions were common in
European dendrochronology as regards the provenance of timber: (a) the
location where timbers were used was in general close to the location of the
forest where the oaks were felled, and (b) the strength of a
dendrochronological match between tree-ring series from different trees is a
function of the absolute distance between the locations where the trees grew,
i.e., the provenance of imported timber is in the region represented by the
chronology that produces the best match (Hollstein 1980). It took some years
of dating research and (efforts towards) developing Dutch chronologies before
I realized that these assumptions are invalid for the Netherlands. The first
assumption does not hold for the Roman and Medieval periods because
timber was often brought here from elsewhere; the second one does not hold
for any period because coastal regions such as the Netherlands are
characterized by a variety of micro-environments, in which case the strength of
a match is related to the similarity between the environmental conditions of
chapter 1
| Introduction
15. An example is the dating of the Roman
Meuse bridge at Cuyk (province of
Limburg; Chapter 5), an archaeological
under-water site from the fourth century
that could not be dated accurately by either
the contextually related material finds or
the radiocarbon method. The Cuyk
chronology, which at the time consisted of
six series only, could not be dated with
established chronologies. It could, however,
be matched with the chronologies of two
fourth-century water wells in Gennep.
Cuyk and Gennep are situated only 15
kilometres apart. The Gennep chronologies
were well replicated and had been matched
with the Central German chronology of
Hollstein (1980).
the sites where the trees grew, not to geographical distance.
I therefore adopted the assumption that data sets of tree-ring series from
Dutch archaeological and historical structures do not contain a homogeneous
environmental signal. Given heterogeneous data sets, the following is feasible:
(a) using correlation techniques it might be possible to discern sub-groups with
a similar signal; (b) chronologies of indigenous oak can be produced from trees
that died naturally and are preserved in former bogs (‘bog oaks’); if these
chronologies can be dated by matching with chronologies from coastal regions
in the neighbouring countries it is precisely because distance is less important for
crossdating in the coastal region of Northwest Europe; (c) given an
international dendrochronological data set that is large enough, and given
correlation techniques that are suited to distinguish sub-groups in this set, in
the future it might be possible to assess in detail the provenance of groups of
tree-ring series representing imported oak.
At this point (1990) I concentrated, with regard to the Middle Ages, on oak
from the Dutch town of ’s-Hertogenbosch (province of Noord-Brabant), for
which written sources referring to local oak plantations and a regional wood
trade are available. The aim was to determine whether it is possible, using
correlation techniques, to distinguish between locally grown and imported oak,
and whether it is possible to assess in a general way from which region imported
timber was derived. For this analysis I applied the ‘Mean Correlation
Technique’ (Wigley et al. 1984), which is used in dendroclimatological studies
to determine whether chronologies from trees that grow at the same forest site
are suitable for climate reconstructions (Chapter 3).
In 1992, in order to collect a data set of oak that unquestionably grew in the
Netherlands, I started the Sub-Fossil Forests (SFF) Project, which is dedicated to
the research of bog oaks (RING/ROB). The methodological purpose of the
study was to determine the strength of the ‘Expressed Population Signal’ (EPS)
in tree-ring patterns of bog oaks from single locations using the Mean
Correlation Technique, to be used later as a criterion for clustering tree-ring
series of unknown provenance. For the analysis a bog oak chronology was
produced that runs from 2258 to 1141 BC (NLPre_ZH; Chapter 4).
Another reason for analysing oak trunks from former forests was that dated
growth patterns of oak from natural contexts could possibly cover chronological
gaps in the existing archaeo-dendrochronological data set. The dating by
dendrochronology of cultural structures in the Netherlands had become more
successful. By developing and applying local Dutch chronologies derived from
archaeological material it had become possible to date structures that were
undatable with the established, mainly German, chronologies.15 However, in
the historical part of the Dutch data set chronological gaps still existed in the
second century AD, and in the fourth up to the sixth century. These could be
finally covered by tree-ring series of bog oaks collected in the SSF project. An
Iron Age/Roman period chronology for the central Netherlands was developed
which runs from 325 BC to AD 563 (NLRom_R; archaeological material and
bog oaks). Three shorter chronologies that exclusively represent archaeological
timbers run from 84 BC to AD 50 (NLRom_W1), 140 BC to AD 87
(NLRom_W2), and AD 190 to 395 (NLRom_E; Chapter 5).
The clustering of medieval oak tree-ring series in the Netherlands is
complicated compared to clustering series from prehistoric times and the
Roman period. Few Medieval oak finds date from before AD 1300 and those
that do were often derived from objects that are relocatable, such as barrels,
whose origin is necessarily uncertain. Furthermore, after AD 1300 in the
Netherlands all tree-ring series that can be dated through dendrochronology are
derived from timber that may have been imported. Also, no sub-fossil remains
of natural forests have been found in the Netherlands that date from after the
sixth century AD, i.e., there is no set of indigenous oak samples dating from
after AD 600 that can be used as a starting point for a Dutch chronology. The
analysis of Medieval oak was therefore postponed until analytical methods to
discern groups of tree-ring data that differ in terms of their signal had been
tested, and until the dendrochronological data set was sufficiently large to be
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
reliable. This point was reached in 1995, when the data set contained about
six-hundred dated Medieval tree-ring series. Using correlation techniques,
80% of the series could be clustered into distinct groups (Chapter 6). Three
historic chronologies resulted that differ in terms of length, sample size, and
geographical distribution of the locations where the timbers were used. The
longest chronology, NLHist_1, runs from AD 427 to 1752 and mainly contains
series from timber used in the southern Netherlands (the provinces of
Limburg and Noord-Brabant). Chronology NLHist_2 runs from AD 1023 to
1666 and represents timber used in the central and northern Netherlands.
Chronology NLHist_3 runs from AD 1041 to 1346 and represents timber
from the coastal region and the IJssel and Vecht Valley.
The ‘Mean Correlation Technique’ is applied in this study to assess the
strength of the signal shared by tree-ring series in a data set. With this
technique, the ‘Expressed Population Signal’ (EPS) is calculated; this
parameter estimates how well a tree-ring chronology resembles the
hypothetical perfect chronology. The higher the value for EPS, the more the
series in the data set resemble each other. In dendroclimatology this is taken as
an indication of the trees’ climate sensitivity (Wigley et al. 1984), and in the
current study as an indication that the trees reacted to similar environmental
influences, i.e., grew in approximately the same region, under approximately
the same hydrological, pedological and climatic regimes. During the first
analyses based on this technique (Chapters 3 and 4), I used an adapted
version published by Briffa and Jones (1990), which distinguishes correlation
coefficients between series representing samples from the same tree (the
‘within-tree signal’) and correlation coefficients between series representing
samples from different trees (the ‘between-tree signal’). During these and
subsequent analyses I stumbled on some problems regarding the statistical
relationship between these estimators and EPS, the most important one being
that the within-tree-correlation is hyperbolically related to EPS, i.e., as the
former increases the latter decreases. These and other problems associated
with the Mean Correlation Technique are described in Chapter 7.
Chapter 8 contains a summary of this study as well as a critical discussion
of the results and some considerations about the direction in which
dendrochronology in the Netherlands is currently developing or might be
developed in order to extend its application in archaeological and historical
research and palaeo-environmental studies.
2
THE STATISTICAL PROPERTIES OF ‘MEAN SENSITIVITY’ A REAPPRAISAL 1
ABSTRACT - This paper investigates some statistical properties of the
dendrochronological parameter ‘Mean Sensitivity’, s, referred to in this chapter by
. It is demonstrated that if the underlying time series obeys a Lognormal
distribution, is directly related to the variance and the first order autocorrelation
coefficient of the series. A model of this relationship is developed and applied to
experimental tree-ring data. The main finding is that is an ambiguous parameter
and that, when characterizing time series, the combination of variance and first
order autocorrelation is to be preferred.
2.1
INTRODUCTION
A variety of parameters and descriptive measures are routinely used to
describe the characteristics of tree-ring chronologies. Of the basic measures,
we mention average, variance and first order autocorrelation coefficient (e.g.
Box and Jenkins 1971).
In this paper we investigate the Mean Sensitivity () of tree-ring series. This
measure is often cited in dendrochronology as an indicator of the climate
sensitivity of tree growth at different types of site. Mean Sensitivity is usually
presented as a descriptive parameter comparable to mean, standard deviation
and first order autocorrelation (e.g., Fritts 1976, 300-311; Cook and Briffa
1990, 157). It was introduced by Douglass to assess the usefulness of
particular tree-ring series for absolute dating purposes (Douglass 1928). He
defined it as
with xn the ring-width in year n and N the number of observations. Douglass
added that ‘the practical application of handling mean sensitivity is to take the sum
of all changes in 10 years without regard to the sign and divide them by the sum of
the 10 years growth’ (Douglass 1928, 30). This is expressed as:
1. Corrected version of Strackee, J. and
E. Jansma, 1992. Dendrochronologia 10,
121-135.
chapter 2 | The statistical properties of ‘mean sensitivity’ - a reappraisal
Huber and Holdheide (1942) restricted the domain of xn to xN, such that
Fritts (1976) nearly returned to Douglass’ original definition, proposing
Notwithstanding the slight differences, these all measure the local deviations
of a time series.
However, during the last decades some problems have arisen regarding the
interpretation of :
i)
measures year-to-year variations of the ring widths. Although high
values of have been shown to reflect the influence of growth-limiting
phenomena (e.g. drought in semi-arid regions; Fritts 1976), the exact
nature of these phenomena cannot be deduced from the value of .
However, in studies of undetrended ring-width series of oak, fir, silver fir and
spruce from western and central European sites, the value of has been used to
deduce the degree of continentality of the central european climate during the
Neolithic and Bronze Ages (Fürst 1963; Fürst 1978). Interpreting the value of
in terms of the continental climate only ignores the possibly growth-limiting
effect of (a) local endogenous and exogenous environmental factors and (b)
growth-limiting phenomena related to climate in oceanic regions.
ii) Modifying the original time-series, e.g., by removing the slope and/or
low frequencies, may result in a decrease of . However, the value of is often
treated as being relatively independent of the method of standardization. The
exact stage at which is calculated during the multistage process of chronology
building generally remains implicit in the literature; the same is true of the
detrending method that is used.
iii) The value of is dependent on the variance of the variable represented
by a time series. In living matter the magnitude of the variance is nearly always
related to the nature of the variable itself. Whether or not standardization is
performed, a series representing cell-wall thickness shows less variability, and
therefore lower values for , than a series comprised of ring width. This often
remains implicit when the values of for different tree-ring variables are
compared.
Because methods to assess the magnitude of the climate signal within ring series
have been refined or altered, the emphasis placed on as a parameter for
characterizing tree-ring time series has diminished. Currently, analysis of
variance (ANOVA) and signal-to-noise ratio (SNR) are among the methods
commonly used (Fritts 1976; Hughes et al. 1982; Cook and Kairiukstis 1990).
However, to our knowledge, the assumption that is a reliable parameter for
characterizing time series has never been formally questioned.
We have therefore looked in more depth at the statistical properties of . The
original ring series we used were detrended by division with values from an
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
estimated growth curve, rather than by subtraction. Standardizing by
subtraction results in data with zero mean. This renders meaningless; the
summation in the denominator of a positive value and a negative one can
result in a small value, thereby inflating to infinity.
2.2
STATISTICAL ASPECTS
2.2.1 General
To investigate the statistical properties of , we introduce a bivariate
density - with positive variates - for the dendrochonological indices xn and xn+1.
In the sequel the subscript n turns out to be redundant and is therefore
dropped. Denoting this density by f(x,y), we compute the density, f(), of a
variate defined by = (x-y)/(x+y). Both the factor 2 (factor ½ in the
denominator of Douglass’ original formula), and the fact that the of interest
involves the absolute value of (x-y), 兩 x-y 兩, is dealt with afterwards.
For the computation, let t be an auxiliary variable such that
(x-y)/(x+y) =
x=t
Solving these equations for x and y and denoting the solution as xo and yo, we
first create a bivariate density f(,t)
f(,t)ddt = f(xo,yo)dxodyo / J;
-1 ≤ ≤ +1, t ≥ 0
with xo and yo the solution of the above equations and J the Jacobian of the
transformation, i.e., the determinant of the matrix
␦
␦
␦x
␦y
␦t
␦t
␦x
␦y
For our transformation, J reduces to J = - ␦ / ␦y. Now xo and yo solve as
xo = t and
yo = t (1-)/(1+),
and one has
J = - (1+)2/(2t).
The density of follows from integrating f(,t) over t from 0 to ⬁. Since as
defined is symmetric on [-1,+1], the change to 兩 x-y 兩 and the factor 2 are
introduced by mapping f() onto [0,+2].
Possible candidates for f(x,y) were restricted to those that complied with the
following two conditions:
chapter 2 | The statistical properties of ‘mean sensitivity’ - a reappraisal
a)
b)
f(x,y) should be meaningful with respect to the data, i.e., x and y
should have a positive domain;
to facilitate further analysis, f() should have a closed analytical
form.
The following two densities meet these conditions:
a)
b)
the bivariate Gamma density (as product of two independent
Gamma densities);
the bivariate Lognormal density.
2.2.2 The bivariate Gamma density function
The bivariate Gamma density can be used to describe series of treering data in which the autocorrelation is insignificant. In practice, this implies
that the series have been pre-whitened. It can be represented as
with ⌫(x) the gamma function (Abramowitz and Stegun 1965). This function
contains the parameters and k and arises as the product of two identical
independent Gamma densities.
The marginal expectations and variances of x and y are
E[x] = E[y] = k/
Var[x] = Var[y] = k/2
The density of = ½兩 x-y 兩 /(x+y) turns out as
0 ≤ ≤ 2.
Figure 2.1 depicts f() for some values of k.
FIGURE 2.1 - Density f() for a bivariate
Gamma density, k being a descriptive
parameter of the latter density
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
The expectation and variance of equal
;
see Table 2.1.
With respect to this outcome, we note the following:
a) For k = 1, f(x,y) reduces to the product of two simple Exponential
densities
f(x,y 兩 k=1)dxdy = 2e-(x+y)dxdy,
and f() becomes the uniform density with
f()d = ½d, 0 ≤ ≤ 2;
E[] = 1 and Var[] = ⅓.
However, for biological material the value of k is generally larger than 1.
b) For large values of k (k > 8), one has
E[] ⬇ 1.13/公k,
Var[] ⬇ 0.73/k;
(1.13 ⬇ 2/公, 0.73 ⬇ 2(-2)/).
c) For the interpretation of k, note that for large values of k (k > 8), Gamma
distributions approximate Normal distributions (mean and variance 2).
Equating first and second order moments from both densities shows that
= /2 and k = (/)2,
the approximation holding well for - 3 > 0.
Because the coefficient of variation, CV[x], has CV[x] = / = 1/公k, for large
values of k we have the useful relation
Eappr[] ⬇ 1.13CV[x]
(1.13 ⬇ 2/公)
(1)
As k increases, E[] decreases while SD[], due to the finite domain of , first
increases and then decreases (Table 2.1).
As an example, let a series of xn be bivariate normally distributed with
= 12 and = 3. Approximating these data with a bivariate Gamma density
function yields k = 16, and according to Table 2.1 E[] = 0.280 and SD[] =
0.207. Since CV[x] = 3/12 = 0.25, we derive from equation (1) Eappr[] =
0.282, demonstrating that the approximation works rather well. However, the
large value for SD[] indicates weak reliability of the estimation for .
chapter 2 | The statistical properties of ‘mean sensitivity’ - a reappraisal
k
E[]
SD[]
k
E[]
SD[]
0.0625
1.845
0.390
2.0
0.750
0.487
0.125
1.719
0.496
4.0
0.547
0.381
0.25
1.526
0.583
8.0
0.393
0.285
0.5
1.273
0.616
16.0
0.280
0.207
1.0
1.000
0.577
32.0
0.199
0.149
TABLE 2.1 - E[] and SD[]
(= 公Var() for different values of k)
2.2.3 The bivariate Lognormal density function
The second density for f(x,y) we considered, was the standard bivariate
Lognormal density function. This density function introduces a mutual
dependence between x, y, i.e. between xn and xn+1. It can be used to describe
autocorrelated tree-ring series.
Let Q = [{ln(x)-␣}2 -2r{ln(x)-␣}{ln(y)-␣}+{ln(y)-␣}2], then
containing the three parameters ␣,  and r.
For the marginal expectations and variances of ln(x) and ln(y) one has
E[ln(x)] = E[ln(y)] = ␣
Var[ln(x)] = Var[ln(y)] = 2,
and for the covariance
Cov[ln(x),ln(y)] = r2
From this we see that ␣ and 2 represent the mean and variance of the
logarithmically transformed data, while r is the correlation coefficient between
ln(x) and ln(y). In our situation r also equals r1.
The density of can be computed as
; 0≤ ≤ 2
with D = 2(1-r). Figure 2.2 gives f() for a series of values for 公D.
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
FIGURE 2.2 - Density f() for a
bivariate Lognormal density, 公D being
a descriptive parameter of the latter
density
E[] and Var[] are expressed by the following integrals
(2)
(3)
TABLE 2.2 - Expectation (E[]) and
standard deviation (SD[]) of the
Mean Sensitivity as function of a
series of values of 公(1-r), computed
with equations 2 and 3
We failed to solve these integrals analytically. Table 2.2 lists E[] and SD[]
(= 公Var[]) for a series of values of 公D = 公(1-r), with  the standard
deviation and r the first order autocorrelation of the logarithmically
transformed data.
Note that E[] and SD[] depend solely on the value of D, i.e. 2(1-r). This
implies that differences between experimental values of can arise from
changes in , in r or in a combination of both.
公(1-r)
E[]
SD[]
公(1-r)
E[]
SD[]
0.0625
0.070
0.053
4.0
1.629
0.537
0.125
0.140
0.105
8.0
1.806
0.424
0.25
0.276
0.203
16.0
1.902
0.315
0.5
0.525
0.362
32.0
1.951
0.228
1.0
0.901
0.532
64.0
1.976
0.163
2.0
1.312
0.599
chapter 2 | The statistical properties of ‘mean sensitivity’ - a reappraisal
FIGURE 2.3 - Histogram and fitted
frequency distribution of series 5288
(left)
FIGURE 2.4 - Histogram and fitted
frequency distribution of series 5291
(right)
2.3
EXPERIMENTAL ASSESSMENT
2.4
DISCUSSION
To test our theory, we randomly selected seven index series of oak from
a Dutch archaeological context (the Roman fortress Velsen 1; Jansma 1985).
The indices had been derived by dividing raw ring widths by predicted values,
the latter obtained by fitting an exponential or linear growth curve on the data,
using least squares.
Each series was first tested against a Lognormal density by comparing the
frequency histograms of the log-transformed data with their theoretical
counterpart based on estimates of and 2. None of the series deviated
(p < 0.05) from a Lognormal density (Figs. 2.3 and 2.4).
Table 2.3 summarizes the results for the seven series. For each
logarithmically transformed series we calculated the mean (␣est), the standard
deviation (est) and the first order serial correlation coefficient (rest). est is an
estimator for using the non-log-transformed data. 公Dest represents the
estimated 公D = 公(1-r).
It is apparent that while for some series (numbers 5288, 5291, 5292 and
5293) est hardly differs, the corresponding est and rest are markedly different.
When substituting 公Dest into equations (2) and (3) to obtain E[] and SD[],
we find little difference between E[] and est, thus confirming that the standard
deviation and first order autocorrelation of a series can indeed be successfully
used for predicting .
Like estimates of the mean and standard deviation, is also a stochastic
variable. Assuming different bivariate densities for the tree-ring data xn, we
obtained two closed analytical expressions for the density of . We actually
found a third expression, using the dependent bivariate Exponential density
(Johnson and Kotz 1969); this latter function is not, however, meaningful for
biological data.
The bivariate Gamma density appears useful because it approximates
bivariate Normal densities. However, detrended tree-ring time series may show
small values for the coefficient of variation with corresponding small values for
. In pre-whitened tree-ring series the dependence between xn and xn+1 is weak,
such that mainly reflects the standard deviation of the series. As previously
noted, the large magnitude of SD[] indicates weak reliability in the estimation
of .
As was shown for the indexed tree-ring series, the bivariate Lognormal
density represents the statistical aspects rather well. Mean Sensitivity may
therefore be redundant. Since E[] depends solely on 公(1-r1),  being the
estimated standard deviation of the series, the value of can be derived from
and r1 of the series. Any change in  can be compensated for by a change in r1
and vice versa. A large value for may therefore result from a large value of
2, a small value for r1, or a combination of the two. This indicates that the use
of will yield more ambiguous information than the use of r1 and 2.
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Sample No. # Rings
␣est
est
rest
est
公Dest
E[]
SD[]
5269
61
-0.051
0.316
0.659
0.189
0.184
0.206
0.153
5271
75
-0.067
0.364
0.571
0.274
0.238
0.264
0.195
5288
79
-0.041
0.296
0.580
0.202
0.192
0.214
0.159
5289
101
-0.072
0.382
0.628
0.248
0.233
0.259
0.191
5291
106
-0.084
0.385
0.770
0.209
0.190
0.211
0.157
5292
74
-0.052
0.337
0.686
0.199
0.189
0.210
0.156
5293
73
-0.045
0.306
0.629
0.207
0.186
0.208
0.155
TABLE 2.3 - Results for seven index
series of Dutch oak. # Rings: the
number of growth rings measured in
each sample; ␣est, est and rest: estimates
of the mean, standard deviation and
autocorrelation coefficient of the
logarithmically transformed data;
est: estimated Mean Sensitivity (no
logarithmical transformation);
公Dest = est公(1-rest); E[] and SD[]
again from equation (2) and (3),
respectively.
To summarize, Mean Sensitivity is simply related to the variance and first
order autocorrelation of a time series. The estimator for is not consistent
since its variance does not converge to zero for increasing N. In addition, its
variance is large over the domain of interest of , making even less attractive
and its interpretation more problematic. Because of these questionable assets,
data are better assessed solely by their mean, variance and first order
autocorrelation.
3
DENDROCHRONOLOGICAL METHODS OF DETERMINING THE
ORIGIN OF OAK TIMBER: A CASE STUDY ON WOOD FROM
’S-HERTOGENBOSCH 1
ABSTRACT - Dendrochronological methods to determine the origin of wood are few
and unreliable. In the study presented here, dendroclimatological correlation
techniques are used that have been developed for assessing the homogeneity of the
growth patterns in living trees. The data set consists of fifteenth century oak timber
applied in the Dutch town of ’s-Hertogenbosch. The tree-ring series, with felling
dates between AD 1463 and 1465, are compared to regional oak chronologies from
Belgium and Germany. They are found to crossdate best with a chronology from
the eastern Belgian Meuse Basin. The hypothesis of a Belgian origin for the timber
is examined in the light of historical information on local trade and wood
management. The correlation techniques are applied to the series from
’s-Hertogenbosch and the Meuse Basin. The results of the analysis confirm that the
timber came from eastern Belgium.
3.1
1. Corrected version of Jansma, E., 1992.
Helinium 32, 195-214.
INTRODUCTION
Current dendrochronological research in the Netherlands focuses on
the construction of average tree-ring chronologies of oak for dating purposes.
In order to generate meaningful tree-ring chronologies of timber from
archaeological and historical contexts, the origin of this timber should be
known. In this chapter, methods are discussed with which the origin of oak
timber can be established dendrochronologically.
Dendrochronology is a relatively young discipline in the Netherlands, and
valid regional chronologies do not yet exist. Dated short chronologies of oak
from Dutch locations range from 325 BC to AD 150, from AD 250 to 400
and from AD 825 to about 1800. Most of this data set has been derived from
beams, planks and posts from archaeological and historical structures. The
scarcity of quality construction wood in the Netherlands dates at least from
the fourteenth century, when the expansion of urban centres must have
involved the use of large quantities of timber. In historical terms, the local
scarcity of tree species suitable for construction purposes is evidenced by the
importance of centres of the wood trade like Dordrecht and Maastricht in the
fourteenth and fifteenth century (Fig. 3.1), and, in the seventeenth century
and later, by trade connections with areas as remote as the Baltic region. In
dendrochronological terms, the application of timber from trees that grew
outside the current Dutch borders is evidenced by the ease with which oak
tree-ring patterns from Dutch Medieval contexts can be absolutely dated by
means of regional chronologies from the surrounding countries. The regional
chronologies best suited for dating timber from Dutch Medieval contexts have
been derived from forested areas in Germany and Belgium (Table 3.1,
Fig. 3.2).
When discussing tree-ring chronologies, we distinguish between site
chronologies and regional chronologies. A site chronology represents the growth
of trees at a single, well-defined site. A dendrochronological site is
characterized by homogeneous growth conditions, such as altitude and slope,
chapter 3 | The origin of oak timber (’s-Hertogenbosch)
2. Climatological signal: the pattern or
variation in a series of ring widths that can
be contributed to climate (Keannel 1992).
soil type and hydrology, distribution of light and shadow, and rain and
temperature. A regional chronology represents the growth of trees at a large
number of sites within a broadly defined geographical and climatological region.
When analysing samples from living trees, the location of the site(s) and the
nature of the growth conditions are known, but they are unknown for
archaeological and historical samples.
The environmental influences that cause trees to grow as they do cannot be
deduced from tree-ring patterns in any straightforward manner, because tree
growth is influenced by a combination of factors. These factors can be divided
into five categories (Cook 1990): (1) the age of the tree (the older the tree, the
narrower its annual rings); (2) tree-specific, or endogenous, conditions such as
the social position of a tree compared to the surrounding trees (a dominant tree
has access to more light and food than its subdominant neighbours, and as a
result forms wider rings); (3) exogenous non-climatological influences operating
upon all trees growing at the same site (pollution, insect outbreaks, fires,
flooding etc.); (4) climatological conditions such as rainfall and temperature;
and (5) a random component, i.e., growth that cannot be explained in any of
the above terms. When tree-ring patterns from relatively large areas are found
to crossdate, this is mainly due to the influence of the climatic conditions (4).
In a chronology that is constructed for purposes of dating, the climate signal2
should therefore be strong.
Because climatological and exogenous factors both influence all trees that
grow at the same site, good dendrochronological crossdating between the trees
from a single site can be explained in both climatological and nonclimatological terms. Whenever chronologies from different sites are found to
crossdate, however, it is most likely caused by the fact that the climatological
signal in the chronologies agrees well. The reason is that exogenous influences
vary from site to site, whereas climatological influences are less variable. When
site chronologies that represent the same type of site and climatological region
are averaged into a regional chronology, the effects of exogenous influences tend
to cancel each other out while the climatological signal is strengthened. This
means that a regional chronology should be better suited for the purposes of
dating than the separate site chronologies that were used to generate the
regional chronology.
Which kind of climatological signal a chronology contains (e.g., a
precipitation signal, a temperature signal, or a combination of the two), is
dependent on the type(s) of site where the trees grew. Several statistical
methods exist with which one can establish the nature of the climatic variables
to which living trees from a given site or region respond (e.g., Response
Function Analysis; Fritts 1976). These methods assume that the location of the
site(s) is known, and involve the use of average monthly meteorological data.
It is impossible to establish the exact nature of the climatological signal
contained in a chronology that is based on archaeological or historical material.
First of all, no monthly meteorological data exist for Medieval and earlier times.
Second, the location of a wooden structure does not necessarily coincide with
the place where the timber came from. The exact location of the site(s) where
such trees grew is always unknown, which means that the site conditions that
influenced the growth of the trees (altitude, slope, soil type, etc.), cannot be
known.
When the absolutely dated tree-ring series from a single wooden structure are
averaged, the result is not necessarily a meaningful chronology; the inclusion of
trees from different types of site could (and often does) result in a chronology
that does not represent specific climatologic and exogenous influences, but an
amorphous mixture of growth conditions. A chronology that, for example,
would include oaks from both extremely wet and extremely dry sites, the former
yielding trees that are mainly stressed by moisture, the latter yielding trees that
are mainly stressed by drought, would not contain a clear signal. In years in
which the former trees would show a narrow ring, the latter ones would show a
wide ring (and vice versa). When these ring widths were averaged, the values
would cancel each other out. In the resulting chronology, the growth response
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
3. Within a climatologically homogeneous
region oak a, which grows on a site of type x,
should crossdate well with oak b which grows
kilometres away but also on a type x site,
whereas it should show less similarity with
oak c, which grows only a few hundred
meters away but on a site of type y. This
phenomenon occurs when the environment is
characterized by a diversity of microenvironments (e.g., in mountains and coastal
regions).
to precipitation and ground water levels would therefore be absent. The
quality of such a chronology would be low, because it would be difficult to
date and not well suited for the absolute dating of as yet undated tree-ring
patterns.
For Medieval times, the Dutch data set of absolutely dated tree-ring
patterns in part consists of locally grown oak, and in part of oak that was
brought here from what is currently Germany and Belgium. Before these treering patterns are averaged into chronologies, we suggest that statistical
techniques be applied in order to establish which of these patterns represent
the same general (unknown) growth conditions. Based on the outcome, the
available patterns could then be clustered into homogeneous groups, each
group yielding an average chronology with as strong a signal as possible. As
long as the nature of such a signal remained unexplained, the signal could be
described in statistical terms. Comparison with available regional chronologies
from the neighbouring countries and with the ring-patterns of sub-fossil oaks
from Dutch contexts, i.e. indigenous oak found in situ, could provide
information about the type(s) of site such a chronology would represent and
the nature of its signal. This would allow for a discussion of the signal in more
geographical, environmental and climatological terms.
Trees growing at the same site often show remarkably similar ring-width
patterns. This observation has prompted efforts to express the degree of
similarity between any two tree-ring patterns as a function of the distance
between the two trees from which the samples were taken (Hollstein 1980,
18-24). Determinations of the origin of oak timber have accordingly been
based on the degree of crossdating between patterns of unknown origin and
regional tree-ring chronologies representing the growth of oak in well-defined
regions (Hollstein 1980; Weiss, verbal communication 1992). According to
this approach, the regional chronology that most resembles the tree-ring
patterns is taken to represent the area where the trees came from.
Some objections can be raised against this approach. First of all, one may
not have access to the regional chronology that represents the area where the
studied timber came from, in which case a region of origin might be deduced
that does not coincide with the region the wood actually came from. Second,
the region formally represented by a published chronology may differ from the
region it represents in reality. Hollstein (1980) assumes that the region where
the wooden structures he dated were located is the region where the trees
actually grew (although he makes an exception for relocatable objects like
barrels and ships). As will be shown below, this assumption is wrong for at
least part of one of the regional chronologies he published. Third, the majority
of the regional chronologies that are available, like the chronologies from
Germany (Table 3.1, Fig. 3.2), have not been tested for homogeneity.
Whenever these chronologies are used for determining the origin of timber, it
is only assumed that they do not include tree-ring patterns from multiple
climatological and geographical regions. Fourth, within a climatologically
homogeneous region the climatic variables are constant, and only the
exogenous factors and site characteristics vary. When the average rainfall and
temperature are the same for all trees, oaks that grow at similar sites must
broadly respond in the same manner. This means that the degree of similarity
between oaks that grow in a climatologically homogeneous region can not be a
function of distance.3 Finally, the similarity between tree-ring patterns is in
part influenced by the severity of the climatic conditions in the studied period
and region. When a severe summer drought occurs in Northwest Europe, for
instance, oaks growing in its more southern parts, and on well drained soils,
tend to form a narrow ring (Kelly et al. 1989). When summer droughts occur
during several years, the section of the ring pattern formed by these oaks
during this period should crossdate well and show strong similarity. During
years in which weather conditions are less growth-limiting for oak, the growth
of the trees would in part be determined by weather conditions, and in part by
site-related (exogenous) or even tree-specific (endogenous) factors. During
such a period, their patterns should not crossdate as well and their degree of
chapter 3 | The origin of oak timber (’s-Hertogenbosch)
similarity should be lower. In other words, both the severity of the weather
conditions and the size of the region where these conditions occur vary through
time. The relationship of the distance between trees to their degree of
crossdating varies accordingly. By generalizing observed values of statistical
agreement and distance between trees into one single formula, assumed to be
valid for all times and regions, this variability is ignored.
This study uses a more careful approach. The dendrochronological data set
is derived from fifteenth century buildings in the Dutch town of ’s-Hertogenbosch (Van Drunen and Glaudemans 1995). The bulk of the oak timber
applied in Medieval ’s-Hertogenbosch and studied dendrochronologically
contains wide ring widths and can neither be dated relatively (by means of other
undated tree-ring patterns from ’s-Hertogenbosch and other locations), nor
absolutely (by means of absolutely dated chronologies). The samples that could
be dated, on the other hand, contain small ring widths and crossdate well
among each other. This means that this subset of trees must have come from a
region and/or type of site that differs from the region(s) and/or type(s) of site
where the majority of the wood applied in ’s-Hertogenbosch came from.
The absolutely dated subset of timber mainly consists of beams from oaks
that were felled between AD 1463 and 1465. The degree of crossdating
between their average chronology (CsH) and the available regional chronologies
from the surrounding countries is used to develop a hypothesis regarding the
origin of the timber. The regional chronology that most resembles CsH is taken
to represent the area where the timber came from. Dendroclimatological
methods are then used to estimate the strength of the climatological/exogenous
signal common to (1) the individual tree-ring series contained in the average
’s-Hertogenbosch chronology (CsH), (2) the individual series included in the
selected regional chronology, and (3) the combination of the individual series
contained in CsH and in the selected regional chronology. The underlying
assumption is that if the tree-ring series contained in CsH do not reflect the same
growth conditions as the tree-ring series contained in the selected regional
chronology, the combination of these series should result in an average
chronology in which the signal is weaker than the signal that is present in the
original two chronologies. The results of the analysis are compared to
historical evidence of transactions in timber by merchants and institutions in
fifteenth century ’s-Hertogenbosch.
3.2
4. Radius: a stretch of wood which runs
from the pith (or oldest ring present) to the
bark (or youngest ring present). When
more radii are measured on a piece of
wood, the approximation of the average
growth pattern of the tree becomes more
accurate.
5. The number of separate trees in this
data set is smaller than the number of
sampled beams (table 3.2), because
occasionally several beams had been cut out
of a single tree.
MATERIAL
The wood samples that provided the tree-ring patterns used in this
study were taken from Medieval houses in the Dutch town of ’s-Hertogenbosch
(Van Drunen and Glaudemans 1995). This town, founded in the twelfth century
by the Duke of Brabant, is situated near the river Meuse in the southern part of
the Netherlands. Over the next few centuries it developed into one of the major
towns in the Duchy of Brabant, which stretched southward into Belgium and
included towns like Leuven and Brussels (Fig. 3.1). In AD 1419 and again in
AD 1463, part of ’s-Hertogenbosch was destroyed by fire. The consequent
building activity and expansion lasted well into the sixteenth century.
The wood samples mainly consist of complete cross-sections of oak beams.
Two radii were measured on each of these cross-sections.4 When cores had
been taken with an increment borer, only one radius was available. Of the wood
that could be dated, the timber with felling dates in and just after AD 1463
should, in combination with incomplete samples that crossdate well with this
timber, provide a data set suitable for determining the origin (Table 3.2, Fig.
3.3).5 In and outside the recorded area of the fire of AD 1463, felling dates
between AD 1463 and 1465 are common, which means that timber must have
been bought in large quantities during this period. This improves the chance
that the wood originated from a restricted number of areas. The ring patterns
in this data set crossdate well with each other and cover the period between AD
1299 and 1465. They overlap optimally between 1360 and 1460.
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
FIGURE 3.1 - The Duchy of Brabant in
the fifteenth century.
1 = ’s-Hertogenbosch;
2 = Brussels;
3 = Leuven;
4 = Maastricht;
5 = Schijndel;
6 = Dordrecht;
7 = Liège
FIGURE 3.2 - Geographical origin of
the regional oak chronologies
(assumed). Abbreviations: see
Table 3.1
chapter 3 | The origin of oak timber (’s-Hertogenbosch)
Region
Author
First year
Last year
Abbrev.
Ardennes-Eiffel
Hollstein (1980)
AD 94
AD 1756
DLAE
Central and eastern Netherlands
De Vries (unpubl.)
AD 1272
AD 1578
NLCE
Ems-Weser
Hollstein (1980)
AD 1314
AD 1618
DLEW
Lower Rhine
Hollstein (1980)
AD 1327
AD 1631
DLLR
Lower Saxony
Leuschner (unpubl.)
AD 915
AD 1873
DLNS
Meuse Basin
Hoffsummer (1989)
AD 672
AD 1986
BMBA
Ostfriesland
Leuschner (unpubl.)
AD 18
AD 1873
DLOF
Rhine-Main
Hollstein (1980)
AD 440
AD 1787
DLRM
Schleswig-Holstein
Eckstein et al. 1970
AD 436
AD 1968
DLSH
Saar-Mosel
Hollstein (1980)
AD 730
AD 1975
DLSM
Weserbergland
Delorme (1972)
AD 1004
AD 1970
DLWB
Westphalia
Tisje (unpubl.)
AD 1260
AD 1669
DLWF
Westerwald-Sauerland
Hollstein (1980)
AD 1369
AD 1773
DLWS
TABLE 3.1 - Regional chronologies
that are relevant to the Netherlands
Most of the regional oak chronologies used in this study (Table 3.1, Fig. 3.2)
consist of averaged yearly growth values. They represent areas in Germany and
Belgium. A preliminary chronology compiled from oak timber applied in the
central and eastern parts of the Netherlands was used (De Vries, unpublished
data). Both the average yearly growth values and the individual tree-ring series
were available of the Meuse Basin (Hoffsummer 1989). No individual
measurement series were available for the other chronologies; the
dendrochronological quality of these chronologies could therefore not be
established.
Documentary evidence of the trade in wood in the Duchy of Brabant is
available from various sources. Written agreements dating from the end of the
fourteenth and the beginning of the fifteenth century show that in this period
hundreds of living oaks were bought and sold in the region around Schijndel,
situated southeast of ’s-Hertogenbosch (Fig. 3.1; Leenders 1991). Evidently, in
this part of the Duchy of Brabant, oak was a valuable but common commodity
during the fifteenth century. The trade in locally grown oak lasted well into the
sixteenth century, and for the northern part of the Duchy does not seem to
have been restricted to any specific area (Vink 1990). The written sources that
are relevant to this study consist of financial accounts kept by two ecclesiastical
charitable institutions in ’s-Hertogenbosch, De Tafel van de Heilige Geest and
Het Gasthuis. They date from AD 1453 to 1515 and from AD 1471 to 1502
respectively, and contain records of the trade in timber in which the institutions
engaged (Vink 1993). This information is not necessarily meaningful for this
study, because it concerns wood used in buildings that at the time were owned
by the institutions themselves, whereas most houses in ’s-Hertogenbosch that
were investigated dendrochronologically had private owners. However, civilians
may have bought wood from these institutions from time to time. Furthermore,
the accounts sometimes mention the origin of the timber that was bought, and
in this manner provide information about the trade routes that were commonly
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
used, and the nearby and more distant regions where wood was abundant
enough to be exploited on a large scale.
The fifteenth century accounts from De Tafel van de Heilige Geest en Het
Gasthuis have no bearing on forest management policies. Sixteenth century
sources from ’s-Hertogenbosch refer to oak being planted, protected and
harvested, and explicitly mention disasters befalling the forests, like flooding
and the effects of war (Vink 1993). This means that oak, although available,
must have been a valuable commodity during the sixteenth century in
’s-Hertogenbosch. The account from De Tafel van de Heilige Geest affirms that
during the fifteenth century part of the demand for wood could be met using
trees that grew on the properties of local land-owning institutions. Between
AD 1460 and 1480, for instance, De Tafel sold large quantities of wood from
sick and dead oak trees (Vink 1993).
FIGURE 3.3 - Map of ’s-Hertogenbosch:
the houses that provided the timber
used in the ’s-Hertogenbosch
chronology CsH;
A = timber felled between AD 1463
and 1465; B = timber dated terminus
post quem;
1 = Hinthamerstraat 36/38;
2 = Hinthamerstraat 85/87;
3 = Hinthamerstraat 89/91;
4 = Hinthamerstraat 113;
5 = Orthenstraat 23/25;
6 = Orthenstraat 41;
7 = Verwerstraat 78;
8 = Visstraat 23
TABLE 3.2 - Tree-ring series used in
chronology CsH
Building
No. of Timbers
No. of Trees
Hinthamerstraat 36/38
1
1
Hinthamerstraat 85/87
2
2
Hinthamerstraat 89/91
7
4
Hinthamerstraat 113
5
4
Orthenstraat 23/25
4
4
Orthenstraat 41
4
4
Verwerstraat 78
1
1
Visstraat 23
1
1
chapter 3 | The origin of oak timber (’s-Hertogenbosch)
Maps dating from the eighteenth century show that the southern part of the
former Duchy of Brabant was largely deforested at that time (Tulippe 1942).
Regardless of the degree of forestation in this part of the Duchy during the
fifteenth century, it is questionable whether it was worth exploiting the area.
No water routes connected ’s-Hertogenbosch to this part of the Duchy, which
means that the utilization of any trees growing here would have involved their
transport over land. In periods of wood scarcity, it may well have been more
profitable to buy wood that could be rafted most of the way to ’s-Hertogenbosch. During the fifteenth century Maastricht (province of Limburg) and
Liège (Belgium), both situated along the Meuse, were important regional
centres of the wood trade (Hoffsummer 1989). The Dieze connected
’s-Hertogenbosch with the Meuse. This means that wood could easily be rafted
from Maastricht and Liège to ’s-Hertogenbosch. The accounts of De Tafel van
de Heilige Geest and Het Gasthuis confirm that some of the wood that was
bought was transported along the Meuse and Rhine rivers, the former being the
route most often mentioned in sources dating from the fifteenth century, the
latter route appearing more often in sources from the sixteenth century (Vink
1993).
3.3
5. Student’s t-values are usually referred to
by t. In this study t stands for the number
of trees; in order to avoid confusion we
refer to Student’s t-values by St.
STATISTICAL METHOD
For all statistical analyses a time interval of 150 years was used,
running from AD 1316 to 1465. In order to obtain a clearer understanding of
the geographical domain of each regional chronology, the correlation between
the regional chronologies was calculated. Then, standard statistical dating
techniques were applied to determine which available regional chronologies
crossdated best with the mean chronology from ’s-Hertogenbosch. The
parameters describing the goodness-of-fit are PV (the coefficient of parallel
variation); r (the correlation coefficient between two series), and St (the value
resulting from a Student’s t-test on the highest value of r found when comparing
the two series5; Hollstein 1980; Baillie 1982; Munro 1984; Chapter 5, section
5.4.1). The regional chronology showing the best fit was renamed Creg.
For Creg, the similarity between the individual tree-ring series in this time
interval was assessed using COFECHA, a computer program designed for
dendrochronological quality control at the Laboratory of Tree-Ring Research,
Univ. of Arizona, USA (Holmes 1983). Using this program and a collection of
individual tree-ring series, each series was broken up into segments of 50 years
and then compared to the mean chronology of all other series. Individual treering series that did not agree with the overall signal, and series that appeared to
contain measurement errors, were removed from Creg.
The signals in the corrected regional chronology Creg*, the ’s-Hertogenbosch
chronology CsH and their combined chronology Call were analysed by means of
dendroclimatological procedures. The signal of a chronology is defined as a
statistical quantity that represents the common variability present in all of the
tree-ring series at a particular site (Briffa and Jones 1990). The climatological
research of tree-ring patterns, of which this and similar statistics are an inherent
part, usually involves living trees that grow at sites whose environmental
conditions are well known. Because working with living trees implies the
possibility of site selection (i.e., the selection of trees that optimally respond to
the environmental factor(s) one seeks to investigate or reconstruct), the
statistical criteria that climatological tree-ring data have to meet are quite strict.
One of two methods is in general used to assess in detail the strength of the
signal that is contained in a tree-ring chronology. Both methods are based on
the assumption that a group of tree-ring series from a specific site makes up one
sample from a hypothetical population that represents the perfect chronology
(Briffa and Jones 1990). The first method is Analysis of Variance (ANOVA;
Fritts 1976). This method of measuring the common variability within and
between trees can only be applied to sections where all tree-ring series overlap
(the common interval). Since some of the series in both Creg* and CsH are short
and the series in general do not overlap very well, the common interval in this
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
case covers only a few decades. We did not, therefore, apply this method. The
second method is the Mean Correlation Technique. This method is discussed
in detail in Briffa and Jones (1990). It consists of calculating the correlation
between all pairs of series, using the maximum overlap between them. We
used a minimum overlap of thirty years between each pair of series, hence
correlations between series with a shorter overlap were omitted from the
calculations. Second, it requires that the means of the individual series are the
same. This demand was met by detrending all series in CsH and Creg* before the
analysis took place.6
The mean correlation between the series that represent different trees ( r̄bt )
was calculated for Creg*, CsH and Call. For CsH the mean correlation was
calculated between series that represent the same trees ( r̄wt ). It was impossible
to calculate r̄wt for Creg because here all trees were represented by a single series
of ring widths. Consequently, the value for r̄wt was the same for CsH and Call.
The effective mean correlation r̄eff , which is an estimate of the chronology
signal that includes both the signals within trees ( r̄wt ) and between trees ( r̄bt ),
was calculated using the method described by Briffa and Jones (1990). First,
one computes ceff, the effective number of measurement series per tree:
(1)
where t is the number of trees, ci the number of samples taken from tree i, and
ceff the effective number of samples in the chronology. Briffa and Jones (1990)
define r̄eff by
(2)
Note that when each tree is represented by one single sample, ceff equals 1 and
r̄eff equals r̄bt .
For Creg*, CsH and Call the degree was quantified to which the chronology
signal is expressed when the individual series in the respective data sets are
averaged. When expressed as a fraction of the total chronology variance, the
chronology signal quantifies the degree to which this particular sample
chronology reflects the hypothetical perfect chronology. This Expressed
Population Signal (EPS; Briffa and Jones 1990) is calculated as:
(3)
6. The ’s-Hertogenbosch chronology
consisted of detrended and pre-whitened
growth indices of which the variance was
stabilized. The regional chronologies used in
the analysis were standardized in exactly the
same way. The detrending of tree-ring
patterns, in this case the removal of low
frequency variations in order to enhance the
high frequency climate signal, is discussed in
Fritts (1976) and Cook et al. (1990).
Pre-whitening, i.e. the removal of the autocorrelation in tree-ring series, is discussed in
Box and Jenkins (1970) and Cook et al.
(1990). In tree-ring patterns, autocorrelation is a function of the growth
conditions in previous years. Its removal
enhances the climate signal.
where t is the number of separate trees included in the chronology. Wigley et
al. (1984) suggest that values for EPS of 0.85 and higher are of acceptable
statistical quality for dendroclimatological studies.
The standard error (SE) of the chronologies is calculated as
(4)
with t the number of separate trees.
chapter 3 | The origin of oak timber (’s-Hertogenbosch)
From (3) it can be seen that the values of EPS and SE are dependent on both
the value of r̄eff and on the number of trees (t) included in the chronology. As t
becomes smaller, the chronology error increases. How well a chronology based
on a subset of t' samples (here Creg* and CsH) estimates a chronology based on t
samples (here Call), is expressed by the sub-sample signal (SSS):
(5)
Briffa and Jones (1990) suggest that SSS values of 0.85 or higher are of
acceptable statistical quality when reconstructing climate from tree rings. They
are less explicit about advisable levels of SE (standard chronology error).
However, values of SE equal to or lower than 0.15 should be adequate in most
cases.
3.4
TABLE 3.3 - Correlations between the
regional chronologies; n = 150 except
in comparisons of DLWS (n = 97) and
DLLR (n = 139); ◆ = correlation not
significant at level of significance ␣ =
0.01; abbreviations: see Table 3.1
Abbrev. DLAE
NLCE
DLEW
RESULTS
Of the thirteen regional chronologies included in the analysis (Table 3.1,
Fig. 3.2; AD 1316 to 1465), in particular the Lower Rhine and Ems-Weser
chronologies crossdate poorly with the chronologies from the adjacent areas
(Table 3.3). The Lower Rhine chronology (first year: AD 1327) correlates
negatively with five chronologies, the lowest correlation (r = -0.60) occurring with
the Ardennes-Eiffel chronology. The Ems-Weser chronology correlates negatively
with three chronologies, the lowest correlation (r = -0.52) occurring with the
Rhine-Main chronology. Both chronologies crossdate well with the WesterwaldSauerland chronology (first year: AD 1369), the Lower Rhine chronology giving
a correlation of 0.72 and the Ems-Weser one of 0.68 (Table 3.3).
DLLR
DLNS
BMBA
DLOF
DLRM
DLSH
DLSM
DLWB
NLCE
0.48
DLEW
-0.50
◆
DLLR
-0.60
-0.44
0.42
DLNS
◆
◆
◆
◆
BMBA
0.25
◆
◆
◆
◆
DLOF
◆
◆
◆
-0.29
0.89
◆
DLRM
0.32
◆
-0.52
◆
0.27
0.30
◆
DLSH
0.31
0.52
◆
-0.50
0.48
◆
0.66
◆
DLSM
0.45
◆
◆
◆
◆
0.54
◆
0.53
◆
DLWB
◆
-0.27
◆
◆
0.76
0.23
0.55
0.40
◆
◆
DLWF
0.45
0.52
-0.30
-0.40
◆
0.29
0.34
◆
0.53
0.23
◆
DLWS
◆
◆
0.68
0.72
-0.37
0.31
-0.42
◆
-0.36
0.48
-0.40
DLWF
◆
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Oak timber from Dutch fifteenth century contexts usually crossdates well with
the Westphalia chronology. The Westphalia chronology in fact constitutes one
of the most useful chronologies for obtaining absolute dendrochronological
dates for fifteenth century timber from Dutch contexts (Jansma, unpublished
data; De Vries, unpublished data). However, the material from ’s-Hertogenbosch does not crossdate well with this chronology (r = 0.22; Table 3.4). The
’s-Hertogenbosch chronology crossdates best with the Saar-Mosel and the
Meuse Basin chronologies (r = 0.50 and 0.51 respectively). Of these two
chronologies, the Meuse Basin chronology yields the highest values of PV and
St (Table 3.4) and is included in the subsequent analyses as Creg.
TABLE 3.4 - CsH compared to regional
chronologies; n = 150 except in
comparisons made with DLWS
(n = 97) and DLLR (n = 139);
PV = coefficient of parallel variation;
r = correlation coefficient;
St = Student’s t-value
Abbreviation
PV
r
St
C reg *
79%
0.67
11.7
BMBA
74%
0.51
10.4
DLSM
71%
0.50
9.4
DLRM
64%
0.26
7.0
DLWB
66%
0.25
6.6
DLAE
66%
0.20
5.8
DLWF
62%
0.22
5.0
DLWS
⬇50%
-
-
DLNS
⬇50%
-
-
DLSH
⬇50%
-
-
DLEW
⬇50%
-
-
NLCE
⬇50%
-
-
DLOF
⬇50%
-
-
DLLR
<50%
-
-
It proved possible to correct Creg by the removal of individual tree-ring series
that showed no agreement with the overall signal. In the corrected chronology
Creg* the average correlation between each separate series and the mean
chronology of the other series has improved from 0.38 to 0.55 (program
COFECHA; Holmes 1983). The correction of Creg caused the correlation with
CsH to increase from 0.51 to 0.67, and the coefficient of parallel variation (PV)
from 74% to 79% (Table 3.4). The corrected chronology Creg* is shown
together with CsH in Fig. 3.4.
The chronology statistics are listed in Table 3.5. All three chronologies CsH,
Creg* and Call have a value for EPS (Expressed Population Signal) that is higher
than 0.85; all three therefore adequately represent the hypothetical perfect
chronology. Both CsH and Creg* have sufficiently high values for SSS (Sub
Sample Strength). However, their values for SE (Chronology Standard Error)
exceed 0.15. The highest value for EPS and the lowest value for SE belong to
the combination of the two chronologies, Call.
chapter 3 | The origin of oak timber (’s-Hertogenbosch)
TABLE 3.5 - Chronology statistics
Number of trees (t)
Number of cores
ceff
r̄bt
r̄wt
r̄eff
EPS
SE
SSS
3.5
CsH
Creg*
Call
21
46
2.10
0.37
0.81
0.43
0.94
0.20
0.98
17
17
1
0.32
0.32
0.89
0.23
0.93
38
63
1.78
0.35
0.81
0.39
0.96
0.15
-
DISCUSSION
Strong negative correlations like those that occur between the Lower
Rhine and Ems-Weser chronologies and some of the other regional chronologies
used in this study (Table 3.3), cannot be completely explained in terms of the
climatological and ecological differences between the regions. The samples used
to construct the Ems-Weser chronology, for instance, were in part derived from
sites in the province of Westphalia (Hollstein 1980, 4), whereas this chronology
does not crossdate at all with the independently constructed Westphalia
chronology. A similar problem exists for the chronology of the Lower Rhine.
Although this chronology is considered to include Dutch tree-ring patterns
(and therefore to represent Dutch growth conditions; Hollstein 1980), it is in
our experience not useful for dating wood from Dutch archaeological and
historical contexts. It is possible that the Lower Rhine chronology was compiled
from wood from multiple climatological regions and types of site, and that the
resulting signal is amorphous. However, between AD 1369 and 1465 the Lower
Rhine chronology is almost identical to the Westerwald-Sauerland chronology
(r = 0.72), which implies that the Lower Rhine chronology in this time interval,
when oak was becoming scarce, most likely represents trees from the forested
Westerwald-Sauerland area. In other words, between AD 1369 and 1465 the
origin of the tree-ring series that were used in the Lower Rhine chronology is
different from that previously assumed. In addition, this chronology is clearly
not suitable for determining the origin of oak from this period that did in fact
grow in the region of the Lower Rhine.
FIGURE 3.4 - The detrended
chronologies from the Meuse Basin
and ’s Hertogenbosch between AD
1316 and 1465; upper: ’s-Hertogenbosch chronology; lower: corrected
Meuse Basin chronology
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
From the documentary evidence it is clear that during the fifteenth century
the demand for construction wood in ’s-Hertogenbosch could in part be met
by using trees from the properties of local land-owning institutions (Leenders
1991; Vink 1993). Since no fourteenth and fifteenth century absolutely dated
oak chronologies exist for the vicinity of ’s-Hertogenbosch, it is impossible to
assess whether CsH (the ’s-Hertogenbosch chronology) represents local growthconditions. However, many of the fifteenth and sixteenth century oak samples
from ’s-Hertogenbosch cannot be dated by means of either CsH or any of the
available regional chronologies (section 3.1). Given the fact that locally grown
oak provided the fifteenth century institutions in ’s- Hertogenbosch with part
of the timber that was needed, this undatable timber may well have come from
the vicinity of ’s-Hertogenbosch. If this is true, the lack of agreement between
this material and CsH would indicate that the trees included in CsH did not
come from the vicinity of ’s- Hertogenbosch.
The values of r̄bt , r̄eff , EPS, SE and SSS are higher for CsH (’s- Hertogenbosch chronology) than for Creg* (Meuse Basin chronology), which means that
CsH contains the stronger signal. This might in part be caused by the larger
number of samples included in CsH. In addition, more series of measurements
per tree are available for CsH. This allowed an estimate of an improved r̄eff for
CsH, which resulted in more accurate and higher values for EPS, SE and SSS.
The highest value for EPS and the lowest value for SE belong to the combination of the two chronologies, Call. This means that no statistical objection
exists against averaging the series constituting CsH and Creg* into a new
chronology, Call. If the signal in Creg* is indeed a climatological one and reflects
the growth responses of oak that grew in the Meuse Basin between AD 1316
and 1465, then the extended chronology Call reflects these responses more
accurately and is a better Meuse Basin chronology for oak than Creg* .
The higher values for EPS and SE in the combined chronology Call mean
that in this chronology the ring-width variations that were caused by
exogenous and endogenous factors are less marked than in the original CsH and
Creg* chronologies. The removal of such non-climatic variations improves the
usefulness of a chronology for the purposes of dating. Even if both CsH and
Creg* would reflect a wide range of climatological conditions (which is unlikely
because of their high values for EPS), their combined chronology Call should,
on the grounds of its stronger signal, be better suited for dating purposes.
The documentary evidence of the use of timber from eastern Belgium is
indirect: occasionally the sources mention the Meuse as a route of
transportation (Vink 1993), but the exact geographical origin of the timber
that was transported along this river remains implicit. The high degree of
crossdating between CsH and the Meuse Basin chronology, and the improved
quality of their combined chronology Call , indicate that the timber used in
’s-Hertogenbosch after the fire of AD 1463 came from eastern Belgium. The
application of the Mean Correlation Technique, in other words, results in
independent data that complement the information contained in documentary
sources on the trade in wood in ’s-Hertogenbosch during the fifteenth century.
3.6
CONCLUSION
The dendrochronological quality of the Meuse Basin chronology
improves when tree-ring series of fifteenth century timber from ’s-Hertogenbosch are included. If, for the period between AD 1316 and 1465, the Meuse
Basin chronology is indeed compiled from oaks that grew in the Meuse Basin,
and the exact soil and weather conditions that influenced the growth of oak in
the Meuse Basin did not occur elsewhere, then the improvement of the Meuse
Basin chronology points to an eastern Belgian origin for the timber. This
means that during periods of increased building activities (e.g. after the fire of
AD 1463), locally grown oaks could not fulfil the demand for timber in
’s-Hertogenbosch.
Current dendrochronological research in the Netherlands aims to construct
average chronologies of oak for dating purposes. The stronger the signal in a
chapter 3 | The origin of oak timber (’s-Hertogenbosch)
chronology, the better suited it is for dating tree-ring patterns from the same
climatological region and/or type(s) of site. The dendroclimatological
techniques used in this study prove useful for grouping tree-ring series of
unknown origin into climatologically homogeneous chronologies with a strong
signal, and in this respect contribute to current tree-ring research. Moreover,
these techniques constitute a quantitative and verifiable approach to questions
that concern the origin of timber.
4
AN 1100-YEAR TREE-RING CHRONOLOGY OF OAK FOR THE
DUTCH COASTAL REGION (2258 - 1141 BC)1
ABSTRACT - The Dutch Sub-Fossil Forests (SFF) Project was initiated in 1992. The
aim of the project was to extend oak tree-ring chronologies in the Netherlands back
in time, using tree-ring data of known origin. One of the results of the project is an
1100-year bog oak chronology, that runs from the Late Neolithic Period to the
Middle Bronze Age (2258 to 1141 BC). The overall value for its Expressed
Population Signal (EPS) is high. The values for EPS and sample depth at different
times, however, indicate that more samples should be included in order for EPS to
reach acceptable levels at some intervals of the chronology. The results of the
statistical analysis indicate that for a long chronology, which in part consists of
series that do not overlap, an overall estimate of the signal may result in values that
overestimate the actual chronology signal.
4.1
1. Jansma, E., 1995. In: J. S. Dean, D. M.
Meko and T. W. Swetnam (eds.), Tree-Rings,
Environment and Humanity - Proceedings of the
International Tree-Ring Conference 1995, Univ.
of Arizona (provisional title). Radiocarbon,
Tucson (in print).
INTRODUCTION
During the last decade, dendrochronology in the Netherlands has
become an accepted tool for archaeological, historical and environmental
studies. Currently, in this country, two organizations are engaged in treering analysis: Holtland Dendroconsult (Veenendaal), which deals with
living trees (e.g. forestry, climate), and the Centre for Dendrochronology
RING (Amersfoort), which researches oak and other tree species from
archaeological/historical and natural contexts from the past.
RING was founded in 1992 to replace the dendrochronological
laboratories of the University of Amsterdam (‘Albert Egges van Giffen’)
Institute for Pre- and Protohistorical Archaeology (IPP, Univ. of
Amsterdam) and the State Service for Archaeological Research (ROB). Its
activities include dating and chronology development (Fig. 4.1), the
reconstruction of forest management in the past (Jansma and Casparie
1993) and, occasionally, the study of living trees (e.g. the reconstruction of
former layouts of parks and gardens on the basis of the planting dates of
living trees (Jansma 1993).
Until 1992, the efforts to generate Dutch reference chronologies, i.e.,
chronologies that are useful for crossdating undated tree-ring patterns from
contexts in the Netherlands, mainly involved oak timber from
archaeological and historical sites. This approach is subject to several
restrictions. First, the provenance of the majority of timbers from Dutch
cultural contexts is unknown. In cases where the region of origin could be
estimated with statistical methods, the studied material (fifteenth century
timber) has proven to be unsuitable for the construction of Dutch reference
chronologies, because the statistical evidence implied that it was brought in
from neighbouring countries (Chapter 3). Second, the selection of
dendrochronological samples from these sites, which in general is made by
archaeologists and historians, often does not correspond to the dendrochronological aim of building well-replicated chronologies. Due to lack of
funds, for example, there is a tendency to keep the number of dendrochronological wood samples from archaeological and historical structures at
chapter 4 | An 1100-year tree-ring chronology of oak (2258 - 1141 BC)
a minimum. As a result, many data sets of absolutely dated tree-ring series
from Dutch sites are at this moment too small to be used independently to
generate average chronologies. Last, the supply of oak samples from
archaeological sites depends on the extent to which the material has been
preserved. In the Netherlands, prehistoric sites are mainly located in soils
that have been well-drained since the fifties and sixties (sand and boulder
clay; Fig. 4.2), and the organic component of these sites has been
deteriorating quickly. As a consequence, in the Netherlands there is only a
limited supply of prehistoric wood samples from cultural contexts. This
means that other sources of samples are needed.
FIGURE 4.1 - Oak tree-ring chronologies with 2 ≥ t ≥ 40 ( t = number
of series; data set of RING (1994))
German and Irish laboratories have in the past overcome these and related
restrictions of their data sets by collecting samples from bog and river oaks
that belonged to natural forests and were deposited in the soil without
human interference (Pilcher 1973; Baillie 1982; Baillie et al. 1983;
Leuschner et al. 1987; Becker 1993). Bog oaks are particularly well-suited
for chronology development. First, their provenance is known; it coincides
with, or was situated near, the location where their remains are found today.
Provided that they can be absolutely dated, they can therefore be used to
generate chronologies whose environmental context is known. Second, bog
oaks are often found together in large numbers; they appear in former bog
peat, where they have been preserved in large numbers below the water
table. In terms of sample size, therefore, tree-ring data that are derived from
bog oaks are well suited for the generation of average chronologies.
In addition, bog oaks that are found together reacted to the same general
environmental conditions. This means that their patterns as a rule have a
strong common signal. They are therefore easy to crossdate with each other
and with absolutely dated chronologies that represent similar environmental
conditions (in this case existing bog oak chronologies from North Germany),
and they are well suited for the generation of average chronologies with an
unambiguous environmental signal. Last, the number of growth rings in bog
oaks is relatively high, compared to the number of rings in oak from
archaeological and historical sites. This means that relatively few samples
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
FIGURE 4.2 - Soil types in the
Netherlands between 1800 and 1200
BC ( after van Es et al. 1988;
schematic representation):
1 = bog and fen peat;
2 = sand/dunes;
3 = fluvial deposits;
4 = marine deposits;
5 = boulder clay and sand;
6 = löss
Bog oak sites:
A = Alphen aan de Rijn;
H = Hazerswoude;
Z = Zoeterwoude
are needed in order to generate average chronologies that span centuries.
Based on the Dutch situation and on the experiences of foreign
laboratories, in 1992 I decided to broaden the strategy of sample collection
and drew up plans for the Sub-Fossil Forests (SFF) Project. In its first year,
the SSF project included a survey of locations in the Netherlands where
tree trunks had been reported, sample collection and measurement, and the
generation of average chronologies from those samples that could be
crossdated. This was followed by efforts to absolutely date the
chronologies.
4.2
MATERIAL
In the province of Zuid-Holland (western coastal region) samples
were taken at three sites situated in former bog peat (Fig. 4.2): Alphen aan
de Rijn (18 trees), Hazerswoude (24 trees) and Zoeterwoude (13 trees).
Sapwood is absent on most of the trunks.
chapter 4 | An 1100-year tree-ring chronology of oak (2258 - 1141 BC)
4.3
METHODS
We measured the ring widths in hundredths of millimeters. In most
cases, two radii were measured per cross-section. Whenever the tree-ring
patterns were difficult to read or measurement errors were suspected, more
radii were measured. Occasionally, we used microscopic slides. The quality
of the measurement series was verified visually and with COFECHA
(Holmes 1983).2
The samples were relatively dated by visual comparison of their plotted
curves and with computerized dating methods. Computerized crossdating
involves sliding two tree-ring series past each other in increments of one
year; at each position of overlap, three values are calculated (see Chapter 5,
section 5.4.1). The first value is PV: the coefficient of parallel variation
between two tree-ring curves (Baillie 1982). Whether a value for PV is
meaningful depends mainly on the length of the overlap between the series.
The second value is the correlation coefficient (r) between the two series. To
assess whether a value for r deviates from zero, the Student’s t-test is
invoked. The Student’s t-value (St) indicates the probability that the
observed and more extreme values for r deviate from r = 0 for a sample size
of n (n being the overlap between the series). This test requires that the
series are pre-whitened, i.e., that the auto-correlation has been filtered out.
In practice, this requirement is often ignored.
The ring-width series were detrended and standardized into indices with
ARSTAN (Cook 1985). I used a flexible spline in order to suppress the low
frequency variance and to enhance the dating potential of the series.3 In
order to arrive at absolute dates, the resulting average chronologies were
compared to bog oak chronologies from North Germany (Leuschner et al.
1987; Leuschner and Delorme 1988).
Two methods are available to assess in detail the strength of the signal that
is contained in an index chronology: analysis of variance (ANOVA) and the
Mean Correlation Technique (Briffa and Jones 1990). ANOVA requires that
the series have a large interval in common. The Mean Correlation
Technique, which already proved useful for the analysis of the provenance of
oak timber (Chapter 3), only requires a minimum of 30 years of overlap
between the series. Sub-fossil trunks are the residual of natural forests that
existed for centuries or even millennia, and their index series do not meet the
common interval criterion of ANOVA. I therefore used the Mean
Correlation Technique to estimate the Expressed Population Signal (EPS) of
the chronologies. This method requires that the means of the individual
series are the same. This demand was met by standardizing all series before
the analysis was carried out.
The Mean Correlation Technique uses an improved estimate of the
average cross-correlations among all series. This effective mean correlation,
r̄eff , is an estimate of the chronology signal that includes both the signals
within trees (r̄bt ) and between trees (r̄wt ). First one computes ceff , the effective
number of measurement series per tree,
(1)
2. The computer programs COFECHA
(quality control and crossdating), ARSTAN
(chronology development) and CHRONOL
(chronology development) were provided
by the International Tree-Ring Data Bank
(ITRDB); we also used CATRAS
(measurement and crossdating; Aniol 1983)
and programs developed by the author (for
the calculation of correlations and values
for EPS).
where t is the number of trees and ci the number of samples taken from tree
i. Briffa and Jones (1990) define r̄eff as
(2)
3. The concept is described in detail by
Fritts (1976), Cook (1985) and Cook et al.
1990.
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Note that when each tree is represented by one sample, ceff equals 1 and r̄eff
equals r̄bt .
When expressed as a fraction of the total chronology variance, the
Expressed Population Signal (EPS) quantifies the degree to which this
particular sample chronology reflects the hypothetically perfect chronology
(Briffa and Jones 1990). EPS is calculated as
(3)
where t is the number of separate trees included in the chronology. Wigley
et al. (1984) suggest that values for EPS of 0.85 and higher indicate that a
chronology is of acceptable statistical quality for dendroclimatological
studies.
The standard error (SE) of the chronologies is calculated as
(4)
with t the number of separate trees.
From (3) it can be seen that the values for EPS and SE are dependent on
both the value for r̄eff and the number of trees (t). As t becomes smaller, the
chronology error increases.
The Mean Correlation Technique was then used to estimate the changes
within the signal. The values for EPS of the standard and residual
chronologies were calculated in 100 year segments (using lags of 50 years),
and compared to (a) changes within the sample depth (the number of treering series included in the chronology), and (b) changes within the
correlation with the reference chronology used to absolutely date the
material.
4.4
RESULTS
Of the 55 trunks that were sampled in the province of ZuidHolland (Fig. 4.2), at first only 26 could be relatively matched. From this
material three separate index chronologies were generated. Although I
noted that these chronologies probably overlapped, the length of the
overlap (and therefore the strength of the match) was minimal and I
refrained from averaging the material into a single chronology.
The three chronologies were recently matched against the North German
bog oak chronology that was developed by the tree-ring laboratory in
Göttingen (Germany; Leuschner and Delorme 1988). This chronology,
which runs from 6069 BC to AD 928, represents about 1600 bog oaks from
100 sites in Lower Saxony and the Emsland region, including sites at the
eastern border of the Netherlands (Leuschner, personal communication).
The statistics that accompany the match are given in Table 4.1. The match
showed that the expected overlap between the ZH chronologies was
correct. At this stage, the measurement series of ten as yet undated bog
oaks from the same sites could be dated and included in the data set. This
extended the chronology from 1163 to 1141 BC.
chapter 4 | An 1100-year tree-ring chronology of oak (2258 - 1141 BC)
TABLE 4.1 - The match with the
North German bog oak chronology;
n = length of compared interval;
St = Student’s t-value;
PV = coefficient of parallel variation
First Year
Last Year
n
St
PV
ZH Interval 1
2258 BC
1690 BC
569
7.2
61%
ZH Interval 2
1727 BC
1358 BC
370
6.7
62%
ZH Interval 3
1393 BC
1163 BC
231
6.2
61%
The dated series, 70 in total, were converted into indices and compiled
into the 1118 year NLPre_ZH chronology (Appendix C).4 Its match with the
North German bog oak chronology is given in Table 4.2.
TABLE 4.2 - The match between
NLPre_ZH and the North German
bog oak chronology;
n = length of compared interval;
St = Student’s t-value;
PV = coefficient of parallel variation
4. Because of the large number of series,
the program CHRONOL was used at this
stage and no ‘arstan’ chronology could be
produced. Program ARSTAN can now only
be applied to data sets that consist of less
than 60 series, whereas CHRONOL
accepts larger data sets. CHRONOL
produces two types of chronologies:
(a) a ‘standard’ chronology, compiled of
detrended tree-ring series; (b) a ‘residual’
chronology, consisting of series that also
have been pre-whitened, i.e. from which
the autocorrelation has been removed.
ARSTAN also produces an ‘arstan’
chronology: the residual chronology in
which the estimated autocorrelation has
been re-introduced.
Chronology
n
St
PV
Standard NLPre_ZH
1118
11.7
61%
Residual NLPre_ZH
1116
10.2
60%
Figure 4.3 shows the sample depth of NLPre_ZH, its running correlations
with the North German bog oak chronology, and its values for EPS versus
time. Here, the sample depth reflects the actual number of samples at each
point in time shown on the X-axis. It varies from 1 to 21 series, with minima
at the beginning and end of the chronology and at 1750, 1700, 1550 and
1400 BC (3, 5, 5 and 4 samples respectively).
The running correlation with the North German bog oak chronology
varies between 0.48 and 0.02, and is somewhat higher for the standard than
for the residual chronology (Fig. 4.3). Minimal correlations occur at the end
of NLPre_ZH and around 2050 - 1950, 1750 - 1700, 1300 and 1200 BC.
Maxima occur at 2200 - 2150 BC, 1800 BC, 1650 BC and 1400 - 1350 BC.
The running values for EPS are somewhat higher for the standard than for
the residual chronology; they vary around 0.82 and 0.80 respectively (Fig.
4.3). Minima occur at the end of NLPre_ZH and at 2100 - 2050 BC, 1750 1650 BC and 1400 BC (residual chronology only).
Figure 4.4 shows the relationship between sample depth and EPS, again
calculated for intervals of 100 years (lag = 50 years). Here, the sample depth
reflects the number of series that have contributed to each value for EPS,
i.e., the total number of series that during each 100 year interval overlap
more than 30 years with any of the other series. It is clear that the value for
EPS depends on the sample size: for intervals where the chronologies consist
of 5 to 10 series, most values for EPS are lower than 0.85, whereas intervals
that contain 10 or more series are mostly, and intervals containing 15 and
more are always characterized by values for EPS higher than 0.85.
The chronology statistics are listed in Table 4.3. The overall value for
EPS is 0.96 (SE = 0.15) for both the standard and residual chronology. This
is markedly higher than the average values for EPS in intervals of 100 years.
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
FIGURE 4.3 - The standard and residual
ZH chronology versus time; sample size
(t), correlation with the North German
bog oak chronology (r) and values for
EPS (100 - yr. intervals,
lag = 50)
FIGURE 4.4 - The relationship between
sample depth and values for EPS in the
ZH standard chronology
chapter 4 | An 1100-year tree-ring chronology of oak (2258 - 1141 BC)
TABLE 4.3 - Chronology Statistics
NLPre_ZH (standard)5
Length in years
1118
Number of trees (t)
36
Number of cores
70
Effective number of cores/tree (ceff )
1.71
Between-tree signal (r̄bt )
0.38
Within-tree signal (r̄wt )
0.74
Effective chronology signal (r̄eff )
0.42
Expressed Population Signal (EPS)
0.96
Standard Error (SE)
0.15
4.5
5. The residual chronology shows slightly
higher values for r̄wt and r̄bt , and a lower
value for r̄eff .
DISCUSSION
The values for EPS indicate that the NLPre_ZH chronology contains
a strong environmental signal. The investigated sites in Zuid-Holland are
situated below sea level in an area of former bog peat (Fig. 4.2), where the
growth limiting environmental conditions must have included periodic high
ground water levels. Provided that ground water levels in the Dutch coastal
region were in the past related to large scale phenomena such as oceanic
transgressions/regressions and changing precipitation levels and/or
evaporation rates, the NLPre_ZH chronology probably contains information
about these phenomena.
Low correlations between NLPre_ZH and the North German bog oak
chronology occur simultaneously with small sample sizes and low values for
EPS (for NLPre_ZH) during the intervals 2100 - 2000 BC, 1800 - 1650 BC
and 1250 - 1150 BC (Fig. 4.3). For these intervals the low correlation
between the chronologies is most likely the result of small sample sizes in
NLPre_ZH. The good match between NLPre_ZH and the German
chronology during the interval 1850 - 1750 BC is probably also related to
the number of series in NLPre_ZH (> 10). However, the relatively high
correlation between the chronologies that occurs between 1450 and 1350 BC
does not coincide with maxima of sample depth and EPS for NLPre_ZH.
During this interval the low replication of NLPre_ZH (4 ≤ t ≤ 8) does not
impair its match with the North German bog oak chronology. This means
that between 1450 and 1350 BC similar growth limiting conditions may have
influenced the growth of the oaks in both regions. Given the locations of the
sites this suggests raised ground water levels and an increased development
of bogs during this period in both regions.
For the NLPre_ZH chronology a minimum sample depth of 10 to 15 is
required in order to obtain a chronology signal of 0.85 or higher (Figs. 4.3
and 4.4). In about one third of the chronology, between 2020 - 1790 BC,
1680 - 1580 BC and 1320 - 1250 BC (400 years in total), this requirement is
met. In the other two thirds of the chronology (700 years) the number of
samples is too small. Provided that the Expressed Population Signal
adequately measures the quality of a chronology, this means that more
samples from the investigated sites, or from similar ones, should be included.
This may be difficult to attain for periods during which oaks could not
germinate in this region because of an increased development of bogs.
However, due to stress the few oaks that remain from such periods should
have an environmental signal that is strong enough to be expressed by the
variability of their annual ring widths and therefore in their average
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
chronology, even if the small number of series in such a chronology
prevented this quality from being expressed by values for EPS.
Although NLPre_ZH has an overall estimated value for EPS of 0.96
(Table 4.3), its values for EPS through time vary around 0.81 (Fig. 4.3).
It has already been noted that EPS is in part a function of sample size: the
overall value for EPS is related to the total number of series that are
included in a chronology and does not reflect the strength (and periodic
weakness) of the signal in different intervals of the chronology. When the
Mean Correlation Technique is applied to long chronologies that largely
consist of series that do not overlap at all, it is therefore advisable to treat
the results with caution.
4.6
CONCLUSION
The SFF project, which the Dutch Centre for Dendrochronology
(RING) initiated in 1992, has already contributed to the data set of
indigenous absolutely dated oak tree ring series. For the Late Neolithic
period to the Middle Bronze Age, the project has resulted in an 1100 year
chronology for the Dutch coastal region. This chronology, which runs from
2258 to 1141 BC, may contain information on changes in the water table
during this period. However, its Expressed Population Signal is insufficient
during at least half of this time interval. Given the relationship between
values for EPS and sample size, and provided that the signal of a
chronology is related to its quality as a reference chronology, i.e., as a tool
for dating as yet undated tree-ring series, this means that we have to sample
more bog oaks in the Dutch coastal region and include them in the
chronology.
The value for EPS overestimates the actual signal in long chronologies;
with such chronologies it is better to analyse the signal in shorter intervals.
The advantage of such interval-analysis is that the results show clearly
which parts of the chronology have a sufficient sample size, and which parts
require additional series.
5
OAK TREE-RING CHRONOLOGIES FOR THE NETHERLANDS
BETWEEN 325 BC AND AD 563 1
ABSTRACT - An 888-year oak chronology is presented, developed from
archaeological material as well as from bog oaks collected through the Dutch SubFossil Forests Project (RING/ROB). This chronology, NLRom_R (46 trees), runs
from 325 BC to AD 563 and represents the growth of oak on low, wet sites in the
central Netherlands. In addition, three shorter chronologies are presented from oak
timber from Iron Age/Roman sites in the East and the coastal region in the West of
the Netherlands: NLRom_E (92 trees; AD 190 - 395 (Cuyk, Gennep, Heeten));
NLRom_W1 (14 trees; 84 BC - AD 50 (Leidschendam, Velsen)); and NLRom_W2
(42 trees; 140 BC - AD 87 (Nieuwenhoorn, Velsen)). These four chronologies
contain a strong signal and are used as a reference for dating oak timber from this
period in the Netherlands.
5.1
1. Jansma, E., 1994. Helinium 34 (in print).
INTRODUCTION
The growth of oak trees is affected by yearly fluctuations in weather
conditions. In large numbers of trees from the same area, favourable and
unfavourable conditions are recorded by the trees in the form of sequences, or
patterns, of wide and narrow growth rings. Dendrochronology is a method of
dating these patterns absolutely. To that end patterns in undated wood are
matched with absolutely dated chronologies that represent the average growth
of the same tree species.
In the Netherlands the dating of Roman structures with dendrochronology
goes back about ten years (Jansma 1985). In order to date tree-ring patterns,
suitable chronologies must be available as a reference. Best suited are
chronologies that represent the same growth conditions as samples bearing the
undated patterns. Dutch local and regional oak chronologies are best suited
for dating oaks that once grew in the Netherlands. This is why
dendrochronology in the Netherlands focuses not only on the dating of oak
from Dutch excavations, but also on the development of average chronologies
from this material.
Until 1992, efforts towards generating Dutch chronologies for the Iron Age
and Roman period mainly involved dated patterns from oak derived from
archaeological sites. Various short archaeological chronologies were constructed
(Jansma 1985; Van Rijn 1987; Van der Sanden 1987; Bult et al. 1989;
RING, unpublished data). These, however, do not overlap sufficiently in time
to allow the construction of a long, well-replicated, chronology. To improve
the data set the dendrochronological laboratory of the State Service for
Archaeological Research (RING/ROB, Amersfoort) broadened its strategy of
sample collection in 1992 through the Sub-Fossil Forests (SFF) Project, which
involved the collection and research of ‘bog oaks’, i.e., oak trunks preserved
in bogs.
Bog oaks are well suited for chronology development. First, bog oaks that
are found together grew on the same site and often at the same time.
This means that their patterns as a rule crossdate well, and can be used to
construct average chronologies with a strong signal. We found that
chapter 5 | Chronologies between 325 BC and AD 563
chronologies from Dutch bog oaks crossdate well with absolutely dated
chronologies from different locations where the environmental conditions were
similar (Chapter 4).2 Second, bog oaks contain more rings than oak timber (the
trees grew older), so fewer crossdated bog oak series are needed in order to
generate long chronologies. Last, bog oaks grew near the bogs (or former bogs)
in which their remains are found today, i.e., they can be used to generate
chronologies from trees whose provenance is known.
This paper presents the average chronologies that are now used to date oak
from Iron Age/Roman and Early Medieval excavations in the Netherlands.
The longest chronology was developed through the SFF project and represents
bog oaks and archaeological timber. Three shorter chronologies represent
archaeological timber derived from Dutch Iron Age/Roman sites.
5.2
2. The water table is the main condition
assumed to have governed the growth of
‘bog oaks’ in North Germany (Leuschner et
al. 1985; Leuschner 1990). Research into
the environmental growth responses of oaks
that have been deposited in bogs is only
just beginning (EC Research Project EV5V
CT94 0500; ‘Temperature Change over
Northern Eurasia during the last 2500
years’). The growth responses of trees are
usually assessed by the correlation of their
ring widths to monthly temperature,
precipitation, ground water levels, etc.
Given the time interval spanned by the
meteorological record (a few centuries at
most), this is only possible with living trees.
Research on the growth responses of bog
oaks is complicated by the fact that no
living stands of oaks that border on bogs
are known today throughout Northwest
Europe. Because the environmental causes
of the ring-width variability in bog oaks
cannot be assessed through the study of
living trees, these causes will to a degree
remain a matter of conjecture.
TERMINOLOGY
A tree ring is the layer of wood that most trees in the temperate climatic
zone produce annually. The distance from one ring boundary to the next,
nearly always expressed in hundredths of millimetres, is referred to as a ring
width. A tree-ring pattern is the sequence of growth rings in wood. When the
successive ring widths are measured, a tree-ring measurement series results which
can be plotted as a tree-ring curve.
A tree-ring pattern reflects the growth response of a tree to various types of
environmental influences. In a tree-ring curve this response shows up as abrupt
and gradual changes in the ring widths. A tree-ring curve is regarded as the sum
of various growth responses, or signals, with different periods (Cook 1990).
Growth signals comprise ring-width variability related to (1) the age of the tree
(the ‘age trend’), (b) local ‘endogenous’ factors (e.g. competition of a tree with
its neighbours for light/food), (c) stand-related ‘exogenous’ factors (e.g., insect
attacks, forest fires), and (d) climate (fluctuations of temperature and
precipitation).
The subject of a dendrochronological study determines which signal is
studied. For dating purposes the most important one is the climate signal, which
in a tree-ring curve is expressed by the variation of the ring widths from year to
year.
Detrending entails the removal of growth signals that obscure the studied
signal from a tree-ring measurement series. A detrended measurement series is
called a growth-index series. The common signal of crossdated growth-index series
is the fraction of the growth signal(s) the series have in common. It is estimated
from the correlation coefficients between the series.
TABLE 5.1 - Classification of tree-ring
chronologies
Chronology type
Living trees
Dead wood
Observation
valid for
Geographical
Scale
Tree curve
1 tree
1 tree
1 tree
point
Site chronology
1 forest stand
1 bog oak site
1 forest stand
micro scale
Object chronology
-
1 arch./hist. object
1 forest stand?
Local Chronology
2 to 5 forest stands
2 to 5 arch./hist. objects
≥ 1 forest stand
in ≥ 1 forest?
local scale
Regional chronology
≥ 6 forest stands in
well-defined
geographical region
≥ 6 arch./hist. objects and bog
oak sites in broadly defined
geographical region
> 1 forest stand
in > 1 forest
regional scale
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
FIGURE 5.1 - Soil types in the
Netherlands during the Iron Age and
Roman period (after van Es et al. 1988;
schematic representation):
A =
bog and fen peat;
B =
marine deposits;
C =
fluvial deposits;
D =
sand;
E
=
boulder clay and sand;
F
=
dunes;
▲ =
archaeological site;
● =
bog oak site;
1
=
Abcoude (ABC);
2
=
Colmsgate (COW);
3
=
Cuyk (CUY);
4
=
Emmeloord (EOF);
5
=
Empel (EMP);
6
=
Flevopolder (FLE);
7
=
Gennep (GENa; GENb);
8
=
Heeten (HRWa; HRWb);
9
=
Leidschendam (LRC);
10 =
Mariënberg (FMB);
11 =
Nieuwenhoorn (NWH);
12 =
Olst (OLF);
13 =
Ouderkerk a/d IJssel (OKF);
14 =
Velsen (VELa; VELb; VELc)
Dendrochronological dating is referred to as crossdating. It involves finding
the correct match of two series. If one of the series is dated and the other one
is not, this process results in a dendrochronological date for the latter. It is
easier to date an average chronology than single tree-ring series, since
averaging reduces part of the non-climatic variability that is present in single
series.
An average chronology consists of the yearly averages of crossdated growthindex series. In this chapter several types of tree-ring chronologies are
distinguished (Table 5.1). A site chronology represents series from a single,
known location. This term is used for chronologies of living trees and the in
situ remains of past forests. Archaeological and historical object chronologies
represent series from one single archaeological or historical object (e.g., a
water well, bridge, ship or (phase of) a building). Archaeological and historical
local chronologies represent series from two to five archaeological or historical
chapter 5 | Chronologies between 325 BC and AD 563
objects. A regional chronology represents series from six or more locations in a
broadly defined geographical region, such as the province of Ostfriesland in
Germany (Leuschner, unpublished data) or the Meuse Basin in eastern
Belgium (Hoffsummer 1989). This concept is used for chronologies of bog
oaks, archaeological/historical timber, living trees, and combinations thereof.
5.3
MATERIAL
5.4
METHODOLOGY
The aim of this study was to produce chronologies with a strong
common signal, so tree-ring series from the same archaeological object that did
not crossdate with each other were excluded from the analysis, as were
deviating tree-ring series from imported barrels and ships’ timbers. Eight
archaeological and six bog oak sites contributed to the current study (Fig. 5.1).
The archaeological sites are: Colmsgate (a water well), Cuyk (a bridge;
Goudswaard 1995), Empel (a water well; Hiddink 1994), Gennep (two water
wells; Heidinga and Offenberg 1992), Heeten (two water wells; Erdrich and
Verlinde 1995), Leidschendam (the Canal of Corbulo), Nieuwenhoorn (a
farmhouse; van Trierum 1992) and Velsen (the Roman fortress Velsen 1: the
western jetty (Morel 1988), ships’ timbers and the foundation of a watch tower
(Bosman, unpublished data)).3 The bog oaks were derived from Abcoude
(Jansma 1987), Emmeloord, the Flevopolder, Mariënberg, Olst and Ouderkerk
a/d IJssel (RING, unpublished data).
5.4.1 Methods of dating
. The reference chronologies used for dating
Only a few oak chronologies available to RING extend back as far as the Iron
Age/Roman period. These are the North German bog oak chronology (6069 BC AD 928; Leuschner and Delorme 1988; Leuschner, unpublished data) and the
Central German oak chronology (690 BC - AD 1975; Hollstein 1980). The
former reflects the average growth of oak in oceanic regions east of the
Netherlands (Ostfriesland and Lower Saxony), the latter the growth in regions
southeast of the Netherlands, where a continental climate prevails. Both are
well suited as a reference to date oak from Dutch archaeological and natural
contexts.
. Visual comparison
The most important crossdating technique is visual comparison of plotted
curves (tree rings as well as indices), which is done by sliding curves alongside
each other on a light table. An experienced dendrochronologist can at a glance
shift the two curves for a proper match, at least with series up to a few hundred
years in length. When longer series are compared, a computer is used to give a
first indication of possible matches, each of which is checked on the light table.
. The coefficient of parallel variation
3. The ships’ timbers from Velsen were
included because their measurement series
crossdate well with the series from
Nieuwenhoorn.
A non-parametric test for dating involves sliding two series alongside each
other and calculating the coefficient of parallel variation (PV) for each position of
overlap. This test can be used to compare undetrended series, because it only
takes into account the differences between directly adjacent ring widths. PV
expresses the fraction of ring widths that at a given position simultaneously
show an increase or decrease relative to the preceding width. It is usually
expressed as a percentage. If there is no match, the expected value for PV is 0.5,
with a standard deviation (S) of 1/(2公n), n being the overlapping years
between the curves. The significance of an observed PV is calculated by
transforming it into a z-score:
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
(1)
The standard normal curve is used to determine the probability (P) that the
observed or an extremer value of z occurs when in reality no match exists
between the series (the probability of exceedence).4
The dating of a tree-ring curve, by sliding it alongside dated chronologies,
involves the examination of hundreds of possible matches. For example, 900
possible matches exist between a curve of 100 rings and a 1000 year master
chronology, and non-matches with a probability of exceedence (P) of 0.01 will
crop up about 9 times when these series are compared. Therefore, in
dendrochronologocal dating only those matches that have a value for P smaller
than 0.001 are considered, and also other dating techniques are used.
. Correlation coefficients and Student’s t-values
The second statistical procedure involves calculating the coefficient of
correlation, r, between the two series at each position of overlap. This
procedure is unsuitable for undetrended ring-width series, because non-zero
correlations may occur between their non-climatic low frequency components.
The correlation between the detrended series x and y is estimated by:
(2)
with n the number of values in each series; xi and yi the observed values in year
i; x̄ and ȳ the mean of the series; and Sx and Sy their standard deviation. To
assess whether rxy deviates from zero one uses the fact that the stochast
4. The foregoing holds only if approximately
0.2 < PV < 0.8, since PV is not normally
distributed. For PV < 0.2 and PV > 0.8,
which seldom occurs in dendrochronological
dating, more precise estimators should be
used, e.g. arcsine transformation of PV,
which is a variance stabilization transformation (Strackee, personal communication).
5. This applies to series that have not been
pre-whitened, i.e. auto-correlated series. Prewhitened series require less high St-values in
crossdating.
5. COFECHA computer program (Holmes
1983).
has a Student distribution (St distribution; see Chapter 3, footnote 5) with
(n-2) degrees of freedom. The distribution of this statistic is well tabulated
(e.g. Thomas 1976). In dendrochronological dating an St-value of less than 3
is meaningless and a value over 6 almost always indicates that two patterns can
be matched visually.5
5.4.2 Chronology development
For all object/site chronologies we estimated the average correlation
(r̄) between each individual growth-index series (gi) and the average
chronology of the remaining series (all series in the data set except gi; with
i = 1 to t and t the number of trees in the chronology).6 Next, we looked at the
degree of crossdating among the chronologies. The index series from
chronologies that crossdate were combined, after which the signal was again
assessed. If all series fitted well, i.e. r̄ did not decrease, the combination was
considered a success. Then, the same procedure was repeated with the
extended chronology and the remaining (unclustered) site/object chronologies.
This procedure was repeated until the cluster chronologies could not be
improved further with the available material.
chapter 5 | Chronologies between 325 BC and AD 563
5.4.3 Estimating the chronology signal
The common signal of the average chronologies was estimated
according to the Mean Correlation Technique. This technique is well-suited for
analysing tree-ring series from the past, because it does not require that the
series completely overlap. The Mean Correlation Technique was first presented
by Wigley et al. (1984). They defined the Expressed Population Signal (EPS) of
a chronology as
(3)
with t the number of trees and r̄bt the average correlation between the series.
This definition only considers data sets that consist of one measurement series
per tree. If a tree is represented by more series, these are averaged before the
analysis takes place.7 Wigley et al. (1984) suggest that values for EPS of 0.85
and higher indicate that a chronology is of acceptable quality for
dendroclimatological analyses.
5.5
RESULTS
5.5.1 General
FIGURE 5.2 - The position of the
object and site chronologies in time
(gray bar = bog oaks)
A total of 194 trees were used in the analyses. The series represent 12
archaeological objects and 6 bog oak sites. One archaeological site (Colmsgate)
and one bog oak site (Mariënberg) are represented by a single tree curve only.
The two tree curves and 16 object/site chronologies are listed in Table 5.2,
together with the number of trees, the first and last year, details of their signal
(r̄ and EPS), and their match with the Central German oak chronology
(Hollstein 1980) and the North German bog oak chronology (Leuschner and
Delorme 1988, Leuschner unpublished data). Figure 5.2 shows the position of
the chronologies in time.
Three local chronologies and one regional chronology were constructed
based on (a) the degree of crossdating among the Dutch object/site
chronologies (Table 5.3) and (b) the correlation (r̄) among the index series in
combined data sets.
7. An alternative that considers more than
one measurement series per tree was
developed by Briffa and Jones (1990).
It involves estimates of the average
between-tree signal (r̄bt ; the average
correlation between cores from different
trees), and the within-tree signal (r̄wt ; the
average correlation between cores from the
same tree over all trees). From these two
estimates, the effective chronology signal (r̄eff )
is estimated, from which EPS is derived.
In Chapter 7 it is shown that the
mathematical relationship between the
within-tree signal and the effective
chronology signal is statistically
problematic. This approach was not,
therefore, used.
Location
Context
t
n
Number
Length
of Trees
in Years
First Yr.
Last Yr.
Chronology Statistics
The match with the Central
The match with the
German chronology
North German chronology
r̄
EPS
St
PV
P < than
St
PV
P < than
ABC*
Abcoude
bog oaks
10
355
325 BC
AD 30
0.53
0.92
5.67
62%
0.0001
6.16
60%
0.0002
COW
Colmsgate
water well
1
217
AD 35
AD 251
-
-
2.09
54%
◆
5.35
62%
0.0005
CUY
Cuyk
bridge
74
143
AD 245
AD 387
0.34
0.97
5.96
61%
0.01
5.62
65%
0.0005
EOF*
Emmeloord
bog oaks
2
144
AD 305
AD 408
0.74
0.85
4.07
66%
0.0002
4.50
63%
0.005
EMP
Empel
water well
9
173
67 BC
AD 106
0.36
0.84
5.70
67%
0.0001
6.23
62%
0.005
FLE*
Flevopolder
bog oaks
8
410
AD 68
AD 477
0.39
0.84
6.60
60%
0.0001
10.22
68%
0.0001
GENa
Gennep A
water well
7
111
AD 248
AD 373
0.50
0.87
5.51
65%
0.005
9.30
79%
0.0001
GENb
Gennep B
water well
6
112
AD 284
AD 395
0.65
0.92
3.55
59%
◆
3.26
51%
◆
HRWa
Heeten A
water well
5
126
AD 190
AD 315
0.40
0.77
4.41
70%
0.0001
6.74
76%
0.0001
HRWb
Heeten B
water well
4
200
AD 136
AD 335
0.45
0.77
3.75
58%
0.025
6.25
65%
0.0001
LRC
Leidschendam
canal
8
134
84 BC
AD 50
0.44
0.86
4.55
65%
0.001
2.39
57%
◆
FMB*
Marienberg
bog oak
1
149
AD 415
AD 563
-
-
2.35
59%
0.03
5.99
64%
0.001
NWH
Nieuwenhoorn
farmhouse
39
227
140 BC
AD 87
0.51
0.98
8.26
65%
0.0001
7.29
66%
0.0001
OLF*
Olst
bog oaks
2
117
AD 424
AD 540
0.78
0.88
1.87
61%
0.02
2.96
61%
0.02
OKF*
Ouderk. a/d IJssel
bog oaks
2
244
AD 81
AD 324
0.63
0.77
4.33
62%
0.0005
5.91
65%
0.0001
VELa
Velsen
Velsen 1 Western jetty
7
109
88 BC
AD 21
0.53
0.89
3.56
62%
0.02
3.95
60%
0.04
VELb
Velsen
Velsen 1 ships timbers
3
121
91 BC
AD 28
0.57
0.80
5.88
60%
0.0001
4.85
64%
0.005
VELc
Velsen
Velsen 1 watch tower
6
65
28 BC
AD 37
0.57
0.89
3.68
65%
0.02
5.89
68%
0.005
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Code
COW
-
ABC*
CUY
EOF*
EMP
FLE*
GENa
GENb
HRWa
HRWb
LRC
FMB*
NWH
OLF*
OKF*
VELa
VELb
VELc
-
-
63%
-
-
-
-
-
65%
-
60%
-
-
66%
63%
55%
-
-
53%
61%
-
-
63%
67%
-
-
51%
-
58%
-
-
-
54%
-
54%
72%
70%
71%
51%
-
-
-
-
54%
-
-
-
-
60%
66%
58%
-
52%
-
53%
-
-
-
-
-
-
COW
-
CUY
-
-
EOF*
-
-
3.79
EMP
3.13
1.99
-
-
FLE*
-
6.70
2.82
4.71
3.27
GENa
-
-
7.97
3.96
-
4.92
GENb
-
-
7.81
4.49
-
3.28
4.23
HRWa
-
1.97
4.29
-
-
4.53
4.77
3.21
HRWb
-
5.08
1.27
0.41
-
4.14
0.86
0.80
6.63
LRC
2.29
-
-
3.38
-
-
-
-
-
FMB*
-
-
-
0.81
-
1.94
-
-
-
-
-
NWH
4.28
0.80
-
-
4.24
-
-
-
-
-
3.15
-
OLF*
-
-
-
-
-
4.73
-
-
-
-
-
4.25
-
OKF*
-
2.74
1.34
-
-
4.77
1.83
0.71
2.74
6.39
-
-
-
-
VELa
5.16
-
-
-
4.10
-
-
-
-
-
2.50
-
4.78
-
-
VELb
2.88
-
-
-
4.39
-
-
-
-
-
3.43
-
10.75
-
-
VELc
1.24
-
-
-
3.00
-
-
-
-
-
4.21
-
1.86
-
74%
-
-
-
-
60%
-
67%
-
-
66%
69%
59%
61%
60%
64%
62%
-
58%
-
79%
63%
-
-
-
67%
74%
54%
-
-
-
-
61%
-
-
-
68%
62%
-
-
-
-
57%
-
-
-
75%
-
-
-
-
57%
-
-
-
-
-
-
-
58%
-
-
-
-
65%
-
-
61%
60%
73%
-
60%
-
-
-
-
-
-
65%
76%
57%
-
-
-
-
-
-
-
65%
48%
4.00
1.09
53%
1.56
chapter 5 | Chronologies between 325 BC and AD 563
TABLE 5.2 - Oak object and site
chronologies from Dutch locations
between 325 BC and AD 563; * = bog
oaks; r̄ = average correlation between
each individual index series and the
average chronology of all other series
in the data set; EPS = Expressed
Population Signal; St = Student’s
t-value; PV = coefficient of parallel
variation; P = probability of
exceedence; ◆ p > 0.05; - = single
curve
TABLE 5.3 - The degree of crossdating
between the Dutch object and site
chronologies; upper right: PV =
coefficient of parallel variation; lower
left: St = Student’s t-value; 30 ≤ n ≤ 244
(n = length of overlap between the
series); * = bog oaks; - = n < 30
ABC*
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
5.5.2 The local eastern chronology NLRom_E (AD 190 - 395)
Chronology NLRom_E (Fig. 5.3; Appendix C) consists of the series
from Cuyk (Roman bridge; 74 trees), Gennep (two water wells; 13 trees) and
Heeten (HRWa; a water well; 5 trees). Their object chronologies crossdate
with values for PV that lie between 67% and 74%; the values for St vary
between 3.21 and 7.97 (Table 5.3). NLRom_E consists of 92 trees and has a
value for EPS of 0.98 (r̄ = 0.57; Table 5.4).
5.5.3 Local western chronology NLRom_W1 (84 BC - AD 50)
FIGURE 5.3 - Oak tree-ring
chronologies for the Netherlands
between 325 BC and AD 563
Chronology NLRom_W1 (Fig. 5.3; Appendix C) consists of the series
from Leidschendam (the Canal of Corbulo; 8 trees) and Velsen (Roman fortress
Velsen 1 (watch tower foundation); 6 trees). The two object chronologies
crossdate with PV = 73% and St = 4.21 (Table 5.3). They do not crossdate
well with any other chronology from the same period. NLRom_W1 consists of
14 trees and has a value for EPS of 0.92 (r̄ = 0.59; Table 5.4).
5.5.4 Local western chronology NLRom_W2 (140 BC - AD 87)
Chronology NLRom_W2 (Fig. 5.3; Appendix C) consists of the series
from Nieuwenhoorn (farmhouse; 39 trees) and Velsen (Roman fortress
Velsen 1 (ships’ timbers); 3 trees). The two object chronologies crossdate well:
PV = 76% and St = 10.75 (Table 5.3). Their plotted curves are very similar; it
is likely that the trees were collected in the same forest. NLRom_W2 consists
of 42 trees and has a value for EPS of 0.98 (r̄ = 0.66; Table 5.4).
5.5.5 The regional low altitude chronology NLRom_R
(325 BC - AD 563)
Chronology NLRom_R (Fig. 5.3; Appendix C) consists of the archaeological
series from Colmsgate (a water well; 1 tree), Empel (a water well; 9 trees),
Heeten (a water well; 4 trees) and Velsen (Roman fortress Velsen 1 (western
jetty); 7 trees), and all bog oak series: Abcoude (10 trees), Emmeloord (2
trees), Flevopolder (8 trees), Mariënberg (1 tree), Olst (2 trees) and
Ouderkerk a/d IJssel (2 trees). The crossdating between their chronologies in
terms of PV and St is shown in Table 5.3. NLRom_R consists of 46 trees and
has a value for EPS of 0.96 (r̄ = 0.53; Table 5.4).
chapter 5 | Chronologies between 325 BC and AD 563
Name
Sites/Objects
First Yr.
Last Yr.
Number
Avgerage
of Trees
Tree Age
r̄
EPS
NLRom_E
CUY, GENa, GENb, HRWa
AD 190
AD 395
92
69
0.57 0.98
NLRom_W1
LRC, VELc
84 BC
AD 50
14
81
0.59 0.92
NLRom_W2
NWH, VELb
140 BC
AD 87
42
102
0.66 0.98
NLRom_R
ABC*, COW, EMP, EOF*, FLE*,
HRWb, FMB*, OLF*, OKF*, VELa
325 BC
AD 563
46
131
0.53 0.96
TABLE 5.4 - Local and regional
chronologies for the Netherlands
(325 BC - AD 563); * = bog oaks;
r̄= average correlation between each
individual index series and the
average chronology of all other series
in the data set; EPS = Expressed
Population Signal
5.6
INTERPRETATION AND DISCUSSION
The Expressed Population Signal (EPS) of the 16 separate object/site
chronologies varies between 0.77 and 0.98 (Table 5.2). If the 0.85 criterion of
Wigley et al. (1984) is applied, six out of these 16 chronologies are of
insufficient statistical quality. These are the chronologies from Empel (EMP),
the Flevopolder (FLE*), Heeten (HRWa, HRWb), Ouderkerk a/d IJssel
(OKF*) and Velsen (VELb), which all consist of less then 10 trees. These
chronologies do, however, crossdate with at least one of the two German master
chronologies with a probability of exceedence (P) < 0.0001 (Table 5.2). This
indicates that they are very similar to the master chronologies and that their low
values for EPS reflect small sample sizes only (equation (3)). The clustered
chronologies NLRom_E, NLRom_W1, NLRom_W2 and NLRom_R, which
consist of between 14 and 92 trees, all have an Expressed Population Signal
over 0.90.
The provenance of the trees used for archaeological chronologies cannot be
established in detail by dendrochronological methods. The general region from
which the timbers were derived can be determined by combining various types
of information: (a) the degree of crossdating with available regional and/or local
chronologies; (b) characteristics of the timbers themselves (e.g. the age of the
trees); and (c) the wood spectra of the archaeological sites where the timbers
have been found.
NLRom_W1 represents timber from Leidschendam (the Canal of Corbulo)
and Velsen (Roman fortress Velsen 1, watch tower foundation); NLRom_W2
timber from Nieuwenhoorn (farmhouse) and, again, Velsen (Velsen 1, ships’
timbers). These chronologies, both from archaeological sites situated directly
behind the western coastal dunes, overlap between 84 BC and AD 50 (Fig.
5.3), but do not crossdate well (St = 3.15; PV = 63%; P < 0.005; Table 5.5).
This indicates that NLRom_W1 and NLRom_W2 represent different
environments.
Several arguments exist for a more eastern provenance of NLRom_W2. First,
NLRom_W2 crossdates better with the German chronologies than NLRom_W1
(Table 5.5). Second, NLRom_W1 in part represents young undressed timber
(Velsen 1, watch tower foundation), whereas NLRom_W2 represents older and
more slowly grown trees, i.e., timber that is rarer and more expensive, and
therefore more likely to have been transported over a distance. Last, tree species
that are less suitable for construction than oak, such as ash (Fraxinus excelsior
L.), were utilized in the harbour works of Velsen 1 (Morel 1988). This means
that oaks in the vicinity of Velsen must have been scarce in the first century.
Given the many building activities of the Romans along and below the Central
Dutch rivers, it is also unlikely that in the vicinity of Nieuwenhoorn old-growth
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
TABLE 5.5 - The degree of crossdating
between the local and regional
chronologies; upper right: PV =
coefficient of parallel variation;
lower left: St = Student’s t-values;
P = probability of exceedence;
X = n < 30; n = length of the overlap
between the chronologies
Chronology
forests still existed in the last quarter of the first century AD.
One object chronology from Heeten is incorporated in NLRom_R (HRWb),
whereas the other belongs in the eastern chronology NLRom_E (HRWa).
Heeten is situated in a transitional area between the IJssel and higher sandy
soils (Fig. 5.1). In the fourth century the inhabitants of Heeten must have
derived their timber from different forests situated to the west and east of their
village.
NLRom_E crossdates well with NLRom_R (P < 0.001), but better with the
German chronologies (P < 0.0001; Table 5.5). This chronology contains the
ring patterns of both young, undressed trees (the bridge in Cuyk) and worked
timber (two water wells from Gennep and one from Heeten). Given the
difference between its signal and the signal in NLRom_R, the trees did not
grow on low, wet sites. The young age of the trees from Cuyk (Table 5.4;
Appendix A) indicates that they were collected close to the building site,
which means that the trees grew on well-drained sandy soils near Cuyk in the
southeastern Netherlands. It is possible that NLRom_E represents trees from
the same area as NLRom_W2, but because the chronologies do not (yet)
overlap in time, this cannot be confirmed.
NLRom_E
NLRom_E
NLRom_W1
NLRom_W2
NLRom_R
Central
Germany
North
Germany
X
X
62%
P < 0.001
67%
P < 0.0001
66%
P < 0.0001
63%
P < 0.005
66%
P < 0.0002
64%
P < 0.005
59%
P < 0.04
63%
P < 0.0002
64%
P < 0.0001
67%
P < 0.0001
63%
P < 0.0001
64%
P < 0.0001
NLRom_W1
X
NLRom_W2
X
3.15
NLRom_R
6.97
4.65
5.78
Central Germany
7.89
4.61
8.37
11.24
North Germany
8.03
3.55
7.35
15.22
62%
P < 0.0001
17.13
The chronologies from bog oaks, which all fit into NLRom_R, represent sites
that were situated near bogs and lakes in the central Netherlands, where high
ground water levels are assumed to have been the main growth limiting factor
(Fig. 5.1). The archaeological material included in NLRom_R must represent
similar conditions, i.e., the trees used in Colmsgate (a water well), Empel (a
water well), Heeten (a water well; HRWb) and Velsen (the Roman fortress
Velsen 1; western jetty) grew in the vicinity of lakes and bogs in the central
Netherlands.
5.7
CONCLUSION
After ten years of dendrochronology in the Netherlands, the data set
of absolutely dated tree-ring series from archaeological Iron Age/Roman
contexts has become large enough to construct meaningful average chronologies. Three archaeological chronologies have been produced (Fig. 5.3).
chapter 5 | Chronologies between 325 BC and AD 563
NLRom_E, which runs from AD 190 to 395, represents timber derived from
forests in the southeastern Netherlands. NLRom_W1, which runs from 84 BC
to AD 50, represents timber from forests in the western Netherlands.
NLRom_W2, which runs from 140 BC to AD 87, most likely represents timber
derived from forests in the eastern Netherlands and/or adjacent areas of
Germany. A fourth chronology, NLRom_R, was developed through the SubFossil Forests (SFF) Project (RING/ROB) and represents bog oaks as well as
archaeological timbers (Fig. 5.3). It runs from the Middle Iron Age to the Early
Middle Ages (325 BC to AD 563) and reflects the growth of oaks on low, wet
sites in the central Netherlands. The chronologies have a strong signal and
crossdate well with existing regional chronologies from Germany.
6
MEDIEVAL TREE-RING CHRONOLOGIES OF OAK
FROM DUTCH ARCHAEOLOGICAL AND HISTORICAL SITES
(AD 427 - 1752; AD 1023 - 1666; AD 1041 - 1346)
ABSTRACT - Absolutely dated tree-ring series from the Early Middle Ages and later,
derived from Dutch archaeological and historical contexts, are clustered into three
master chronologies. In geographical terms, the distribution of the sites represented
by the chronologies differs markedly. Chronology NLHist_1 (AD 427 - 1752; 259
series) mainly represents sites in the south of the Netherlands; NLHist_2
(AD 1023 - 1666; 195 series) consists of timbers applied in the central and
northern parts, and NLHist_3 (AD 1041 - 1346; 30 series) is composed of timbers
from sites along the coast and in the IJssel and Vecht Valley. The new chronologies
are well suited as a reference to date oak from Medieval archaeological sites and
historical buildings in the Netherlands.
6.1
INTRODUCTION
6.2
MATERIAL
6.3
METHODOLOGY
The aim of the research presented here was to produce average treering chronologies of oak for the last 1500 years that are suited for dating oak
from Dutch contexts. Such chronologies should meet the following criteria:
(1) they should represent absolutely dated tree-ring series of oak from Dutch
archaeological and historical contexts; (2) they should have a strong signal.
Because the signal in a chronology depends both on the between-tree
correlation as well as on the number of trees included in the chronology
(Wigley et al. 1984), criterion (2) implies that the chronologies should consist
of tree-ring series that crossdate well in terms of their correlation coefficients,
and should include as many tree-ring series as possible.
At the end of 1994, the dendrochronological data set at the Dutch
Centre for Dendrochronology (RING/ROB, Amersfoort) consisted of 611
tree-ring series with end-dates in the seventh century or later (Appendix A).
The series were dated in previous years using chronologies from abroad
(Appendix B) and our own chronologies.1
6.3.1 Quality control
1. Jansma, unpublished data, in part
available through the International Tree-Ring
Data Bank (ITRDB, NOAA/NGDC;
Boulder, Colorado).
Usually, several wood samples are available from the same cultural
object, and the ring widths in each sample are measured at least twice along a
different radius of the wood. The ring widths are plotted logarithmically along
the y-axis and the (as yet undated) years along the x-axis. The graphs are
compared visually, using a light table. By the comparison of independent
measurement series from the same sample, errors can easily be detected and
corrected. The similarity between graphs of different samples from the same
chapter 6 | Medieval tree-ring chronologies of oak
object is a first indication of the degree of crossdating between the samples and
also of the likelihood that the samples can be absolutely dated. After a wood
sample has been dated with statistical methods, the date is checked visually by
comparing its tree-ring curve with the curve of the chronology that produced
the date.
6.3.2 The detrending of tree-ring patterns
The dendrochronological dating of ring-width patterns makes use of
the variations of the ring widths that have been caused by climate, i.e. changes
in temperature and precipitation. A series of values that represents the exact
width of the growth rings in a tree often reflects more than the yearly
fluctuations of the weather conditions: e.g., the increasing age of the tree, which
is expressed by a gradual decrease in the ring widths towards the outside of the
stem, the effects of competition with neighbouring trees for food and light, or
the long-term influences of a sudden disaster like fire or an insect outbreak
(Cook 1990). If the non-climatic component in a tree-ring pattern masks the
ring-width variability caused by climate, the pattern does not resemble that of
other trees from the same region and period and cannot be dated.
Detrending is the removal of unwanted information from tree-ring series. The
assumption is that a ring-width pattern is the sum of various signals with
different periods (Cook 1990). Detrending involves fitting a more or less
flexible estimated growth curve (a model fitted to the data), to the ring-width
series. A detrended series of growth indices is produced either by division of the
original values by the estimated ones or by subtracting the estimated values
from the original ones.
The focus of the dendrochronological study (e.g. climate, forest dynamics,
dating) determines how the growth curve is estimated and which ring-width
variability it is designed to remove. The influence of climate is mainly expressed
in the difference between ring widths from year to year. When the objective is to
date a pattern, detrending methods are therefore used which remove some of
the gradual increases and decreases in the ring widths.
6.3.3 Methods of dating
Tree-ring patterns are dated by matching their growth-index series with
dated master chronologies. The master chronologies used by RING are listed in
Appendix B. The most useful Medieval chronologies are those of Lower Saxony
and Ostfriesland (Leuschner, unpublished data), Weserbergland (Delorme
1972), Twente (NL) and Westphalia (Tisje, unpublished data), the Meuse
Basin (Hoffsummer 1989), and the Central German master chronology
(Hollstein 1980).
The most important dating technique is visual comparison of plotted curves.
When long series are compared, the computer is used to give an indication of
possible matches through (a) the coefficient of parallel variation (PV), and (b)
the Student’s t-value (St) derived from the correlation coefficient between the
series. All possible matches are inspected visually. Further details are given in
Chapter 5 (section 5.4.1).
6.3.4 The clustering of tree-ring index series
In view of the aim of the study, which was to produce average oak
chronologies for the Netherlands that can be used to date as many
dendrochronological samples as possible, the series were not clustered prior to
the analysis according to their date or other a priori defined criteria. Instead, all
611 tree-ring series were combined into one large group, and those series that
show least agreement with the behaviour of the majority of the series were
identified. These series were removed from the data set. The assumption
underlying this procedure is that no matter how heterogeneous the Dutch
collection of absolutely dated tree-ring series is, the average variation of the
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
annual values is most likely determined by some dominant environmental
signal(s); these should become stronger as more and more tree-ring series with
deviating characteristics are removed. Furthermore, any climatic signal that is
not dominant in the complete data set, but is nonetheless shared by a
substantial number of trees, should become recognizable by the clusters that
occur among the deviating series.
The estimates of the degree of similarity between the tree-ring series are
based on the correlation coefficient (ri) of each single growth-index series gi
with the average chronology of all series except gi (1≤ i ≤ t, with t the total
number of trees in the data set).2 Data sets of index series from the
Netherlands that, when averaged, result in a chronology that is a suitable
reference for dating, are characterized by a mean correlation (r̄ , the average of
all ri’s) of 0.50 to 0.65 (Chapter 5, Table 5.4). Therefore, during the analysis a
minimum criterion of 0.50 is used, and clusters of tree-ring series that show a
lower value for r̄ are rejected. To determine which series should be removed
from a cluster to increase r̄, the correlation between gi and the average
chronology excluding gi is calculated in 50-year intervals (rik), using an overlap
of 25 years. Here, k is 1 to q and q is the number of 50 year intervals in gi.
Values for rik equal to or lower than 0.32 are not significant at ␣ = 0.05. Series
with more than two intervals showing values for rik ≤ 0.32 are rejected.
6.3.5 Assessment of the reliability of tree-ring chronologies
The Mean Correlation Technique is used to assess the strength of the
chronology signal (Wigley et al. 1984; Chapter 7). With this technique the
Expressed Population Signal (EPS) is calculated. EPS is a function of both the
number of series in a chronology and the average correlation between these
series. A value for EPS of 0.85 or higher is taken to indicate that a chronology
is of sufficient quality for dendroclimatological analyses (Wigley et al. 1984).
The value for EPS overestimates the signal in long chronologies, i.e.,
chronologies in which not all tree-ring series overlap (Chapter 4). EPS is
therefore calculated in 100-year intervals (using an overlap of 50 years).
Intervals that have a value for EPS lower than 0.85 require more samples and
better internal crossdating.
To determine which intervals of the new chronologies are suited as a
reference for dating regardless of EPS, the correlation with the available
master chronologies from adjacent countries is calculated in hundred-year
intervals. Chronology intervals that have a low value for EPS and a high
correlation with the master chronologies probably reflect climatic conditions
well and do not require additional sampling, in contrast to intervals that have a
low value for EPS and show low correlations with the existing chronologies.
6.3.6 The environmental signal of the Dutch cluster chronologies
Two types of information are used to determine the geographical
domain represented by the new chronologies. The first is their degree of
crossdating with available chronologies from adjacent countries. The second is
the distribution of the locations where the timbers represented by each
chronology were put to use. The locations are classified according to the
criteria defined by Wolff, who designated the major ecological systems in the
Netherlands according to the so-called information-carrying functions of
nature, as expressed by a large range of geological and ecological criteria, and
on these grounds divided the Netherlands into 26 geogenetical growth regions
(Figure 6.1; Wolff 1989).3
2. COFECHA computer program (Holmes
1983).
3. This classification was chosen in
consultation with Dutch foresters (Maessen,
personal communication).
chapter 6 | Medieval tree-ring chronologies of oak
FIGURE 6.1 - Growth regions in the
Netherlands (after Wolff 1989):
(I) Pre-Pleistocene:
a = cretaceous deposits;
b = löss;
c = Winterswijk sediments;
(II) Pleistocene:
d = Drenthe Plateau (including
northern lateral morenes);
e = central Netherlands lateral
moraines;
f = Twente Formation;
g = Hunze Depression;
h = IJssel and Vecht Valley;
i = ‘Gelderse Vallei’;
j = western Brabant;
k = central Brabant;
l = eastern Brabant;
m = Zeeuws Vlaanderen Pleistocene
sand area;
n = Meuse terraces;
o = Oude IJssel valley;
p = Vecht valley;
(III) Holocene:
q
r
s
t
u
v
w
x
y
z
= central Netherlands river area;
= (former) perimarine peat area;
= Dutch Wadden;
= (closed) estuaries;
= former sea;
= northern marine sediments;
= southwestern marine sediments;
= IJsselmeer polders;
= IJsselmeer and peripheral lakes;
= coastal dunes
TABLE 6.1 - Chronology statistics
before and after clustering;
r̄ i = average correlation between
single series and mean chronology of
all other series; rik = correlation
between single series and mean
chronology of all other series in
50-year intervals (overlap = 25).
No.
Trees
First
year
Last
year
Length in
years
Total No.
of Rings
r̄ i
rik ≤ 0.32
All series
611
AD 427
AD 1752
1326
63 157
0.44
24%
NLHist_1
259
AD 427
AD 1752
1326
27 100
0.53
4%
NLHist_2
195
AD 1023
AD 1666
664
17 326
0.53
4%
NLHist_3
30
AD 1041
AD 1346
306
3359
0.55
6%
Unclustered
127
AD 906
AD 1749
844
15 373
0.30
59%
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
6.4
RESULTS
6.4.1 General
The averaging of all 611 series that cover the period between AD 427
and 1752 results in a chronology of which the quality is doubtful. The
correlation between each individual series and the average chronology of all
other series in 50-year intervals (rik) is equal to or lower than 0.32 for 24% of
the intervals and r̄ is 0.44, i.e., lower than the 0.50 criterion employed
(Table 6.1).
After the series that show low correlation coefficients are removed, the data
set is reduced to 259 tree-ring series (Table 6.1; NLHist_1). Like the complete
data set, this cluster covers the interval between AD 427 and 1752, but now
only 4% of the 50-year intervals of the tree-ring series have a value for rik ≤
0.32, and the value for r̄ has increased to 0.53.
Within the remaining data set of ‘dissimilar’ series (469 trees), two clusters
can be distinguished (Table 6.1). NLHist_2, representing 195 trees, has a
value for r̄ of 0.53, and only 4% of the 50-year intervals of the individual treering series show values for rik ≤ 0.32. The respective values for NLHist_3,
representing 30 trees, are 0.55 and 6%.
The three clusters contain a total of 484 series; 127 series remain
unclustered. Figure 6.2 shows the shape of the average chronologies; the
number of trees in the chronologies is shown in Figure 6.3.
FIGURE 6.2 - Historical oak tree-ring
chronologies for the Netherlands:
NLHist_1 (AD 427 - 1752);
NLHIst_2 (AD 1023 - 1666);
NLHist_3 (AD 1041 - 1346)
6.4.2 The signal within the chronologies
NLHist_1 (AD 427 - 1752) has a value for EPS of 0.85 or higher
between AD 1150 and 1650 (Fig. 6.4; calculation interval AD 1100 - 1700).
Before AD 1150 the signal is generally lower. However, the correlation
between the chronology of NLHist_1 and the available chronologies from
abroad is not much lower before AD 1150 than afterwards (Fig. 6.5). The
lowest correlation between NLHist_1 and the master chronologies occurs at at
the end of NLHist_1 (AD 1650 - 1750). The correlation between NLHist_1
and the master chronologies before AD 900 is higher than the correlation
among the master chronologies themselves.
NLHist_2 (AD 1023 - 1666) has a value for EPS of 0.85 or higher from
1300 onwards (Fig. 6.4; calculation interval AD 1250 - 1650). The correlation
with the chronologies from abroad before AD 1300 is low.
NLHist_3 (AD 1041 - 1346) shows values for EPS of 0.85 or higher for all
intervals (Fig. 6.4; calculation interval AD 1050 - 1346). The correlation with
the master chronologies is low (Fig. 6.5).
chapter 6 | Medieval tree-ring chronologies of oak
FIGURE 6.3 - The number of samples
in the cluster chronologies
FIGURE 6.4 - The Expressed
Population Signal (EPS) in hundred
year intervals (lag = 50 years)
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
FIGURE 6.5 - The running correlation
between the reference chronologies,
and between the cluster and reference
chronologies. Calculation interval: 100
years (lag = 50). Table 6.2 shows the
reference chronologies used in the
analysis
6.4.3 Growth regions
Figure 6.6 shows the number of samples per cluster and per growth
region. Geographically, NLHist_1 is dominant in regions a, b and n (Fig. 6.1;
province of Limburg); k and j (province of Noord-Brabant); z (the coastal
dunes, in this case in the province of Zuid-Holland); and w (province of
Zeeland). In other words, NLHist_1 is dominated by oak from sites that are
located south of the central Dutch rivers. The trees in this cluster, which
represent timber from ’s-Hertogenbosch (region k) and have felling dates
between AD 1463 and 1465, belonged to forests in the Belgium Meuse Basin
(Chapter 3). NLHist_2 dominates in regions c, d, e, f, i, q, r, s, v and x: all of
the Netherlands north of the provinces of Limburg, Noord-Brabant and
Zeeland, except region h. NLHist_3 mainly represents timber from regions h
(IJssel and Vecht Valley), v (marine sediments in the northern Netherlands
and x (IJsselmeer polders), and only dominates in the first region. In Figure
6.2 the growth regions that are represented by 10 tree-ring series or more are
shaded according to the dominant chronology.
chapter 6 | Medieval tree-ring chronologies of oak
FIGURE 6.6 - The number of samples
per clusters and per growth region
6.4.4 Crossdating with existing master chronologies
The correlation between the cluster and master chronologies, which is
a measure of their degree of similarity, varies considerably (Table 6.2; see
Appendix B for the abbreviations of the master chronologies). NLHist_1
crossdates best with the Central German oak chronology (DLCE; r = 0.62;
overlap = 1326 years). The next best match is with growth region 9 in Lower
Saxony (DLS9; r = 0.50; overlap = 877 years), which represents the area of
Lower Saxony adjacent to the Dutch province of Overijssel. NLHist_2
crossdates best with chronologies of more northern and easterly regions in
Lower Saxony (DLS5, 6 and 7), and with the Ostfriesland chronology (DLOF;
Leuschner, unpublished data; r ⬇ 0.60). NLHist_3 does not crossdate well with
any master chronology: not a single correlation over 0.40 was found.
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
NLHist_1
NLHist_2
NLHist_3
St
PV
r
n
St
PV
r
n
St
PV
r
n
DLS1
21.71
71%
0.49
838
16.59
73%
0.51
644
6.08
62%
0.21
306
DLS3
21.60
70%
0.49
838
17.32
74%
0.54
644
6.83
62%
0.22
306
DLS4
16.83
67%
0.41
878
16.64
73%
0.56
644
6.77
60%
0.23
306
DLS5
17.45
69%
0.43
873
18.34
74%
0.61
644
7.28
64%
0.27
306
DLS6
13.70
65%
0.31
872
16.19
71%
0.60
644
7.26
62%
0.24
306
DLS7
15.91
65%
0.41
888
17.44
72%
0.59
644
7.74
61%
0.27
306
DLS9
22.57
71%
0.50
877
17.79
74%
0.55
644
7.11
62%
0.23
306
DLNS
16.63
66%
0.26
838
18.42
74%
0.45
644
8.58
64%
0.22
306
DLOF
17.18
64%
0.36
1326
17.97
72%
0.59
644
7.74
62%
0.25
306
DLWB
19.11
72%
0.24
749
14.62
71%
0.24
644
4.04
56%
0.15
306
DLWF
12.44
69%
0.30
410
19.36
76%
0.47
407
1.76
56%
-0.04
87
NLTF
12.76
70%
0.33
713
11.11
74%
0.46
627
7.24
66%
0.30
306
NLTW
9.49
68%
0.23
368
18.17
76%
0.54
310
-
-
-
0
BMBA
23.85
73%
0.28
1081
10.32
64%
0.18
644
6.13
61%
0.24
306
DLCE
35.81
76%
0.62
1326
12.59
67%
0.42
644
6.46
62%
0.36
306
NLHist_1
-
-
-
-
12.87
71%
0.42
644
6.70
62%
0.34
306
NLHist_2
12.87
71%
0.42
644
-
-
-
-
4.50
57%
0.26
306
NLHist_3
6.70
62%
0.34
306
4.50
57%
0.26
306
-
-
-
-
TABLE 6.2 - The crossdating between
the Dutch cluster chronologies and
available reference chronologies;
St = Student’s t-value;
PV= coefficient of parallel variation;
r = correlation coefficient;
n = length of compared interval (years)
6.5
INTERPRETATION AND DISCUSSION
6.5.1 Chronology NLHist_1 (AD 427 - 1752)
Geographically, NLHist_1 represents oak timber that was used in the
provinces of Limburg, Noord-Brabant and Zeeland (Figs. 6.6 and 6.7). The
interval AD 1316 - 1465 mainly consists of trees derived from eastern Belgian
forests (Chapter 3). However, the strong agreement with the Central German
chronology (r = 0.62) indicates that outside this interval the timbers were
derived from a wider area, viz., eastern Belgium and adjacent Germany.
Earlier, we defined the chronology intervals for which more samples are
needed as those intervals where minimum values occur for both the
chapter 6 | Medieval tree-ring chronologies of oak
FIGURE 6.7 - The dominance of the
Dutch cluster chronologies in growth
regions represented by ≥ 10 samples
chronology signal (EPS) and the correlation with other chronologies.
Problematic in terms of EPS are the intervals AD 450 - 850/900, AD 900/950 1100/1150 and AD 1650 - 1750 (Fig. 6.4). Less than average correlation with
existing chronologies occurs during the interval AD 1650 - 1750 (Fig. 6.5).
This interval is covered by two series only (Fig. 6.3), and the conclusion that
more samples are needed for this interval is justified.
It is interesting to note that some intervals with a low signal synchronize with
known atmospheric 14C anomalies. These anomalies, above average 14C
production related to sunspot minima, are termed 14C minima. For some
periods, an association exists between 14C minima and global cooling events
(Davis et al. 1992). 14C minima in the first millennium BC coincide with low
similarity between existing chronologies of oak, which is interpreted in terms of
a climatic change towards cooler and wetter conditions (Schmidt and Gruhle
1988). The 14C minima that occurred during the interval covered by NLHist_1
are: the Roman Minimum (AD 660 - 770); the Medieval Minimum (AD 940 1140); the Wolff Minimum (AD 1290 - 1350); the Sporer Minimum (AD 1400
- 1510); and the Maunder Minimum (AD 1645 - 1715; terminology by Davis et
al. 1992). The chronology intervals that simultaneously show a weak value for
EPS are: (a) AD 660 - 770; (b) AD 950 - 1150; and (c) AD 1650 - 1715. EPS
is, however, in part a function of sample size, and each year of the chronology
should represent the same number of tree-ring series (preferably ten or more)
for a valid attempt to explain the values for EPS in terms other than sample
size. Given the low replication of the first and last part of NLHist_1, it is too
early for such an attempt.
6.5.2 Chronology NLHist_2 (AD 1023 - 1666)
The timbers in NLHist_2 mainly were used in the central and northern
Netherlands, and its match with NLHist_1 is not optimal (r = 0.42; Table 6.2).
It is therefore unlikely that the timbers in NLHist_2 have a southern
provenance. NLHist_2 crossdates best with several Lower Saxony chronologies
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
(r ⬇0.60) and the Ostfriesland chronology (r = 0.59), closely followed by the
Twente/Westphalia chronology (r = 0.54). This indicates a more northern
provenance than NLHist_1, most likely Twente/Westphalia and adjacent
North Germany.
The chronology signal is mainly determined by the sample size: the earlier
part of NLHist_2 consists of a few trees only (Fig. 6.3), and the low correlation
with the master chronologies during this earlier interval, which coincides with
low values for EPS, must also be related to sample size (Figs. 6.4 and 6.5).
The first 200 years of this chronology would therefore improve if tree-ring
series were added.
6.5.3 Chronology NLHist_3 (AD 1041 - 1346)
The signal in NLHist_3 is high and cannot be improved further by
adding samples. Its correlation with the existing master chronologies is low
(Table 6.2; maximum r = 0.36 (Central German chronology)). NLHist_3
includes ship timbers from Zeeland (Fig. 6.1; growth region w), a tree that
died naturally near Baarn and was found in situ (region r), thirteenth century
posts from Hoorn (region v), staves from a barrel found in Alkmaar (region v)
and foundation timber from Apeldoorn (region h). In terms of provenance,
the fact that this chronology includes a tree found in situ might indicate that it
represents indigenous oak. On the other hand, the chronology also includes
ship timbers and a barrel, i.e. wood from relocatable objects, which means that
the provenance of the wood could be further away. At this point of our
research, the provenance of NLHist_3 cannot be established with certainty.
6.5.4 The use of the cluster chronologies for dating Dutch sites
and objects
Oak timbers that were used in the Netherlands during the Middle Ages
came from different regions (e.g., De Vries 1994; Jansma 1995; Chapter 3), and
their tree-ring patterns reflect a variety of different growth conditions (e.g. soil
and forest types and meteorological conditions). By clustering these patterns
into separate groups, chronologies result that also reflect different conditions or
combinations of conditions. The precise location of the forests where these
conditions occurred cannot be ascertained. Fortunately, the location of the
forests does not need to be established in order for the chronologies to be useful
for dating. Together, NLHist_1, NLHist_2 and NLHist_3 represent 80% of the
tree-ring series from Medieval contexts that were available at the time of the
analysis. Separately, they summarize the ring-width characteristics of three
different groups of dated timbers used in the Netherlands.
Our experience with the new chronologies is positive: at present more than
half of the datings by RING are achieved using these chronologies as a
reference.4 Noteworthy is the fact that it proved possible during the clustering
to date some timbers that were previously undatable, and include them in the
new chronologies.5
6.6
4. The next most successful chronology is
the Ostfriesland chronology (Leuschner,
unpublished data), followed by the
chronologies of Twente (NL) and Westphalia
(Tisje, unpublished data).
5. Dendrochronological codes
(RING laboratory): NLE01005; NLE08001;
NLE08002; NLR33001; NLR34001;
NLV17007 (Appendix A).
CONCLUSIONS
Three historical chronologies of oak have been produced.6 Together,
they represent 80% of the 611 medieval tree-ring patterns from Dutch
locations available at the time of the analysis. NLHist_1, which runs from AD
427 to 1752, represents oak that grew in the southern part and
south/southeast of the Netherlands. NLHist_2, which covers the interval
between AD 1023 and 1666, represents oak from the eastern part of the
central and northern Netherlands and areas to the east of the country. The
geographical domain of NLHist_3, which runs from AD 1041 to 1346, is
unknown. These chronologies are now used as a reference for dating oak from
Dutch archaeological and historical contexts.
6. The unpublished chronologies are
property of the RING tree-ring laboratory
(ROB; Amersfoort (NL)).
7
THE MEAN CORRELATION TECHNIQUE: THE ‘EFFECTIVE
CHRONOLOGY SIGNAL’ AS AN ESTIMATOR OF THE SIGNAL IN
TREE-RING CHRONOLOGIES
ABSTRACT - The Mean Correlation Technique estimates the signal that a tree-ring
chronology contains. Unlike ANOVA, it allows for different numbers of cores per
tree. In this context it involves the definition of an ‘effective number of cores’ (ceff).
Using (1) ceff, (2) the average correlation between cores from the same tree over all
trees (r̄wt) , and (3) the average correlation between cores from different trees over
all trees (r̄bt) , an ‘effective chronology signal’ (r̄eff) is estimated. From r̄eff and the
number of trees (t), the chronology signal (EPS) is derived. The relationship
between and domain of r̄bt, r̄wt and r̄eff is analysed. It is shown that the effective
chronology signal (r̄eff) is inversely related to the within-tree signal (r̄wt), and
therefore cannot be an adequate measure of the chronology signal. The fact that r̄eff
is not restricted to the domain [-1,1] is problematic. Although anomalous values
occur only when r̄wt< r̄bt , i.e., outside the domain that is relevant for
dendrochronological analyses, it disqualifies r̄eff as a ‘correlation coefficient’.
7.1
INTRODUCTION
All forms of tree-ring research include (1) data collection and
measurement, (2) assessment of signal and noise in the tree-ring series
according to the aims of the study, (3) detrending and standardization of the
series into indices, and (4) the development of average chronologies (Cook
and Kairiukstis 1990). Before a chronology is used for the purpose for which it
was developed (i.e. environmental studies or crossdating), its quality, which
depends on the strength of its signal, is estimated. The chronology signal is a
stochastic quantity which expresses the fraction of ring-width variability that
the set of tree-ring series has in common (Briffa and Jones 1990).
One may choose from two methods to calculate the chronology signal. Both
measure the amount of variability that is shared by a set of indices from
different trees (the between-tree signal) and the amount of variability that is
shared by the set representing the same trees (the within-tree signal). The first
method, analysis of variance (ANOVA; Fritts 1976), must have fixed time
intervals for which all individual series overlap completely. This means that
series that are shorter than the selected interval have to be omitted from the
analysis. ANOVA furthermore requires that all trees in the data set are
represented by the same number of measurement series. The second method,
the Mean Correlation Technique (MCT; Wigley et al. 1984; Briffa and Jones
1990), is less demanding in terms of replication; it does not require a common
interval and can be applied to any interval where two or more series overlap
for a number of years. In addition, it allows for a different number of
measurement series per tree.
Tree-ring data from living trees often have a common interval that is larger
than a hundred years, running from the year in which the trees reached ‘breast
height’ (cores as a rule are taken at a stem height of 1.30 m.) to the year in
which they were sampled. In addition, for living trees the sampling strategy is
generally determined by dendrochronologists, which means that an equal
chapter 7 | The ‘effective chronology signal’
number of cores is taken from each tree. Therefore ANOVA is well suited to
estimate the signal in chronologies from living trees.
Many tree-ring series from archaeological, historical and bog oak sites, on
the other hand, do not overlap for long intervals. In addition, the dendrochronologist often cannot control the number of samples that is taken from this
material. When analysing the signal in chronologies that represent bog oaks or
archaeological and historical timber, MCT, with its less strict demands, is often
preferred to ANOVA (Chapter 3).
When estimating a chronology signal from correlation coefficients, one would
expect the chronology signal to be positively related to both the between-tree
signal (the average correlation between cores that represent different trees) and
the within-tree signal (the average correlation between cores that represent the
same trees). However, during a recent application of MCT I noted that while
the values for the between-tree signal and the overall chronology signal showed
a positive relationship, the values for the within-tree signal and the chronology
signal were negatively related. In order to understand this phenomenon more
fully, I analysed the domain of and the relationship between the variables that
are used to calculate the chronology signal according to the Mean Correlation
Technique.
7.2
THE MEAN CORRELATION TECHNIQUE
The Mean Correlation Technique was first presented by Wigley et al.
(1984). They define the ‘Expressed Population Signal’ (EPS) of a chronology as
(1)
with N the number of trees and r̄ the average correlation between the series.
This definition only considers data sets that consist of one measurement series
per tree. However, Wigley et al. (1984, 211) suggest that this method can also
be applied when several cores per tree are available. In such a case, the series of
each tree should be averaged into one single time series before the analysis takes
place.
Briffa and Jones (1990) have adapted the definition of EPS to data sets that
include more than one series per tree. They first calculate the correlation
coefficients between all series of indices in the data set. Then, a total correlation
mean is computed. The total number of coefficients involved is:
(2)
with i = 1 to t trees, each tree is represented by j = 1 to ci #cores/tree, and Ntot is
the number of correlations that is computed.
A within-tree signal is then estimated by averaging the correlation coefficients
between series from the same tree over all trees. It is expressed as
(3.1)
where
(3.2)
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
A between-tree signal is estimated as the average correlation between all
possible pairs of series that represent different trees:
(4.1)
where
Nbt = Ntot -Nwt
(4.2)
If the number of cores per tree is unequal, ci is replaced by ceff , the effective
number of trees:
(5)
where i = 1 to t trees and ci is the number of series from tree i.
Using ceff, an effective chronology signal is calculated that includes both the
between-tree and the within-tree signal:
(6)
If only one core is available per tree, ci equals 1 and r̄eff equals r̄bt.
Briffa and Jones (1990, 143) state that since by definition r̄wt has a lower
limit equal to r̄bt it can be shown that equation (6) gives an estimated value of
the chronology signal that is almost invariably higher and more accurate than
measures derived solely from r̄bt.
The Expressed Population Signal (EPS) quantifies the degree to which the
chronology signal is expressed when series are averaged:
(7)
with t the number of trees.
chapter 7 | The ‘effective chronology signal’
7.3
THE BEHAVIOUR OF
r̄eff
In equation (6) r̄eff depends on r̄bt, r̄wt and ceff. Because r̄bt is in the
nominator, its relationship to r̄eff is proportional, i.e., y times larger values for r̄bt
result in y times larger values for r̄eff. Because r̄wt is in the denominator, this
variable is hyperbolically related to the value for r̄eff.
FIGURE 7.1 - The domain of r̄eff (ceff=
2 and 3)
Figures 7.1, 7.2 and 7.3 show the relationship between r̄eff, r̄bt and r̄wt for
several values of ceff. This relationship has the following characteristics:
i
r̄eff is discontinuous for r̄wt = 1 / (1 - ceff)
(Fig. 7.1)
ii
if 0 ≤ r̄wt ≤ 1 then ceffr̄bt ≤ r̄eff ≤ r̄bt
(Fig. 7.2)
iii
if r̄wt = 1 then r̄eff = r̄bt
(Fig. 7.2)
iv
if r̄wt = 0 then r̄eff = ceffr̄bt
(Figs. 7.2 and 7.3)
FIGURE 7.2 - The domain of r̄eff
(r̄wt ≥ 0; ceff = 1, 2 and 3)
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
FIGURE 7.3 - The domain of r̄eff (r̄wt ≥ 0,
r̄bt ≥ 0, r̄wt ≥ r̄bt ; ceff = 1.78 (arbitrary
value)); the right Y-axis shows the values
for r̄bt that characterize the adjoining
curves
In other words, (i) in the negative domain of r̄wt the value for r̄eff can become
indefinite (Fig. 7.1). This occurs when the denominator of (6) equals 0; (ii) in
the positive domain of r̄wt the value for r̄eff becomes no larger than ceffr̄bt and no
smaller than r̄bt (Fig. 7.2); (iii) in the positive domain of r̄wt the value for r̄eff
reaches its minimum value of r̄bt when r̄wt = 1 (Fig. 7.2); (iv) in the positive
domain of r̄wt the value for r̄eff approaches its maximum value of ceffr̄bt if the
value for r̄wtconverges to 0 (Figs 7.2. and 7.3). Figures 7.1 and 7.2 show
furthermore that if r̄bt < 0, the values for r̄eff are positively related to r̄wt and
always lower than the corresponding values for r̄bt .
Figure 7.3 shows r̄eff for r̄wt > 0, r̄bt > 0 and r̄wt ≥ r̄bt, which is the domain of these
variables when the chronology signal is calculated. In this case the values for r̄reff
are always higher than the values for r̄bt , and do not become larger than 1.
7.4
DISCUSSION
7.4.1 The domain of r̄eff and r̄bt
In practice, only tree-ring patterns that match well are used for
analyses of the chronology signal. This means that the correlation between the
series is verified before the analysis takes place. Although the domain of r̄eff
that involves negative values for both r̄bt and r̄wt (Fig. 7.1) is not relevant in a
dendrochronological context, its definition, as given by equation (6),
disqualifies the variable as a correlation coefficient: it is not restricted to the
domain [-1,1]. The problem does not occur if both r̄wt and r̄bt are positive and r̄wt
is larger than or equal to r̄bt (Fig. 7.3).
7.4.2 Estimating the mean of correlation coefficients
1. The transformation of r is given by zr =
0.5 loge((1+r)/(1-r)) (Thomas 1976, 393).
The population correlation coefficient () applies only to correlation
in a bivariate normal distribution. The Pearson correlation coefficient (r)
estimates from a sample of variates. Finding the confidence limits about r is
complicated by the fact that must be known prior to finding the standard
error of r. This led Fisher (1935) to derive a second index, known as z.1
To average correlations without previously transforming them to Fisher’s zscores is problematic, especially when the correlations approach the limits of
the domain of r [-1,1]. Averaging always results in an underestimation of the
true correlation: if r1 is 0.7 and r2 is 0.9, then r equals 0.82 (not 0.80); if r1 is
0.4 and r2 is 0.6, then r equals 0.507 (not 0.50). It will be clear that the
calculation of a chronology signal from r̄wt and/or r̄bt as defined in equations
(3.1) and (4.1), whether by the Mean Correlation Technique or ANOVA,
leads to an underestimation.
chapter 7 | The ‘effective chronology signal’
r̄wt and r̄eff
Briffa and Jones (1990, 143) state that since r̄wt has a lower limit equal
to r̄bt , the value for r̄eff gives a measure of the chronology signal that is almost
always larger than measures derived solely from r̄bt. Within the domain given by
r̄wt > 0, r̄bt > 0 and r̄wt ≥ r̄bt, the one exception occurs when r̄wt equals 1: then, r̄eff
reaches its lower limit of r̄bt (Fig. 2). This is because the values for r̄eff and r̄wt
are hyperbolically related; a stronger within-tree signal results in a weaker
chronology signal as calculated from equation (6).
The implications of this relation can be illustrated by an example from our
own research. When estimating the signal in a bog oak chronology that runs
from 2258 to 1141 BC (Chapter 4), the following values were found: detrended
(‘standard’) chronology:
7.4.3 The hyperbolical relationship between
ceff
= 1.708;
r̄bt
= 0.377;
r̄wt
= 0.740;
detrended and pre-whitened (‘residual’) chronology:
ceff
= 1.708;
r̄bt
= 0.379;
r̄wt
= 0.764.
These values result in an effective chronology signal ( r̄eff ) of 0.423 (standard
chronology) and 0.419 (residual chronology). In other words, although the
residual chronology has both a higher within-tree signal and between-tree signal due to
the removal of auto-correlation, its effective chronology signal as estimated by the
Mean Correlation Technique is lower.
According to the linear aggregate model of tree growth defined by Cook
(1990), on which many forms of dendrochronological signal improvement (e.g.
detrending, pre-whitening) are based, any ring-width series (t) is the sum of
various sub-signals. These are: (1) the age-size-related trend in ring width (At);
(2) the climatically related environmental signal (Ct); (3) the disturbance
impulse caused by a local ‘endogenous’ disturbance (D1t); (4) the disturbance
impulse caused by a standwide ‘exogenous’ disturbance (D2t); and (5) the
mainly unexplained variability not related to the other signals (Et). Because
truly endogenous disturbances will be random events in space and time within a
forest stand of sufficient size, the endogenous disturbance impulse in the ring
widths of a given tree will be largely uncorrelated with endogenous disturbance
impulses in other trees from the same stand (Cook 1990, 100). The detrending
of single measurement series before they are averaged into tree curves is in part
based on the fact that these disturbance impulses can even be limited to a
restricted area in the stem of a tree, in which case they are uncorrelated with the
disturbance impulse in other stem areas of the same tree.
From the linear aggregate model it can be seen that in a ring-width series the
proportional contribution of the exogenous and climatic components (D2t and
Ct) should increase as the endogenous signal becomes weaker. The removal
from ring-width series of fluctuations that reflect growth responses limited to
restricted areas in the trees’ stems, therefore not only improves the within-tree
signals (r̄wt), but also the signal shared by different trees from the same forest
stand (r̄bt and r̄eff). The hyperbolical relationship between the within-tree and
effective chronology signal defined by the Mean Correlation Technique
contradicts the fact that the ring-width variability caused by tree-specific
‘endogenous’ influences reduces or conceals the variability caused by more
general environmental (exogenous and climatological) influences. If the
chronology signal would be in truth hyperbolically related to the within-tree
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
signal, the model by Cook (1990) would be invalid and the application of
noise reducing techniques to tree-ring series from the same tree superfluous or
even counterproductive.
7.5
CONCLUSION
The Mean Correlation Technique (Wigley et al. 1984, Briffa and
Jones 1990) results in ambiguous, not wholly comprehensible estimates of the
chronology signal. The derivation of r̄bt and r̄wt according to equations (3.1)
and (4.1) results in an underestimation. The estimation of the ‘effective’
chronology signal (r̄eff), from which the Expressed Population Signal (EPS) is
derived, contains irregularities that are incompatible with the dendrochronological assumptions and statistical demands. Problems include (1) the
interpretation of the effective chronology signal (r̄eff) as a correlation coefficient,
and (2) the inadequate expression of the within-tree signal (r̄wt) in r̄eff.
On these grounds, we suggest that the within-tree signal be omitted from
the Mean Correlation Technique and that correlation coefficients are averaged
using Fisher’s z-scores. In cases where more than one measurement series is
available per tree, prior to the analysis for each tree the same number of series
should be averaged into a single series. Averaging reduces noisy components,
and the correlation coefficients between averaged (crossdated) series should be
higher than the correlation coefficients between single observations. The use
of averaged series and the transformation of their correlation coefficients into
z-scores prior to estimating their average function should result in more
comprehensive estimates of the chronology signal.
8
SYNTHESIS: DISCUSSION AND CONCLUSIONS
8.1
INTRODUCTION
8.2
METHODOLOGICAL ASPECTS: THE ESTIMATION OF THE SIGNAL IN
This study has looked at different methods and applications of
dendrochronology. The outcomes are discussed in this chapter. Section 8.2
presents a discussion of the methods that were used; section 8.3 discusses the
manner and results of applying the new chronologies to actual data sets. In
this chapter directions for future research are suggested.
TREE-RING CHRONOLOGIES
In the mid-1980’s, not much was known about the sensitivity of oak to climate
in the Netherlands. It had been demonstrated that growth patterns of oak
from living trees, timbers and panel paintings could be matched, which had
resulted in the construction of Chronology 1 and Chronology 2 (Eckstein et al.
1975). No match could be established between these chronologies, indicating
that in the Netherlands distinct groups of indigenous oak existed, which are
characterized by a different climate signal (ring-width variability that trees from
different sites have in common), and that some of these groups were exploited
in the past to such an extent that they became extinct (Eckstein et al. 1975).
Chronology 1 and 2 were not useful as a reference to date oak tree-ring series
from locations in the Netherlands (Chapter 1), which suggested that the
signal in indigenous oak is frequently dominated by local growth signals, viz.,
ring-width variability that only occurs in the trees at one site, or only in one
tree. However, the fact that patterns of bog oaks from the western Netherlands
could be matched with the Central German oak chronology (Jansma 1987),
showed that indigenous oak from oceanic sites could contain a climate signal
similar to that in patterns from more continental regions.
In order to investigate the types of variability present in the ring-width
patterns of Dutch oak, I analysed the descriptive parameter ‘Mean Sensitivity’
() for oak from natural and cultural sites from the Iron Age/Roman period.
was developed by A. E. Douglass (1928) to assess the year-to-year variability
in tree-ring series, which to a degree reflects the sensitivity of tree growth to
climate. High values for were interpreted chiefly in terms of stress caused by
continental climatic factors (Fritts 1976; Fürst 1963; Fürst 1978). Because
trees that respond to climatic stress often have patterns that can be matched,
was furthermore taken as an indication of the value of tree-ring patterns for
dating purposes. The main finding was that series of oak from natural and
cultural sites in the western Netherlands show values for that are as high as
those of series from more continental regions (Jansma, unpublished data), i.e.,
if is related to stress, oceanic sites can be as extreme as continental sites in
terms of tree-growth. Furthermore it was established that cannot reflect
continuous environmental stress, since lasting stress results in sequences of
narrow rings with little variation, hence low values for . This study was not
published because the data set on which these findings were based is small
and, more importantly, proved to be an ambiguous parameter.
chapter 8 | Discussion and Conclusions
We demonstrated that if the underlying time series is lognormally distributed,
which is the case with indexed tree-ring series, is directly related to the
standard deviation and the first-order autocorrelation coefficient of the series
(Chapter 2). A model of this relationship was developed and applied to
experimental tree-ring data. The main finding is that depends solely on
公(1-r1),  being the estimated standard deviation of the series and r1 the
estimated first-order autocorrelation. This means that the value of can be
derived from the standard deviation () and first-order autocorrelation (r1) of
the series; a large value for may result from either a large value of , or a small
value for r1, or a combination of the two. This, by the way, explains why does
not reach a high value for tree-ring series that reflect continued environmental
stress; intervals of narrow rings (small values) are characterized by small values
for and high values for r1, which leads to low values for .
Because had proven ambiguous, I turned to dendroclimatology for
techniques to assess the signal in tree-ring series. ANOVA (Fritts 1976) and the
‘Mean Correlation Technique’ (Wigley et al. 1984) are used in climatological
studies to estimate the signal in series from living trees from the same forest or
forest stand. Unlike ANOVA the Mean Correlation Technique does not require
that all tree-ring series overlap; it is therefore suitable for analysing long
chronologies that consist of short series (e.g. oak). The adaptation published by
Briffa and Jones (1990) distinguishes between the ‘within-tree signal’ (the
average correlation between series that represent the same tree) and the
‘between-tree signal’ (the average correlation between series that represent
different trees). From these the ‘effective chronology signal’ is derived, from which
the ‘Expressed Population Signal’ (EPS) is estimated. Because in climate
studies EPS is required to be 0.85 or higher (Wigley et al. 1984), a threshold
value of 0.85 was used throughout the analyses, i.e., chronology intervals with
EPS < 0.85 were considered unreliable.
The first step was to assess whether or not EPS can be applied to estimate
the strength of the signal in Dutch tree-ring data (Chapter 3). A data set was
used from a limited period and restricted geographical domain, consisting of
timbers used between AD 1463 and 1465 in the town of ’s-Hertogenbosch
(province of Noord-Brabant) to rebuild houses destroyed by fire.
’s-Hertogenbosch is situated near the Meuse. The oak chronology of eastern
Belgium, produced from the patterns in timbers from historical buildings in the
Meuse Basin, had just become available (Hoffsummer 1989) and, upon request,
the individual measurement series were also made available.
The historical information on trade and forest management in the region
shows that fifteenth century oak timber used in ’s-Hertogenbosch was collected
locally as well as being imported along the Rhine and Meuse rivers (Vink 1990;
Vink 1993; Chapter 3). Over a hundred timbers from historical buildings were
analysed. Most of the data set consists of undatable patterns from rapidly
grown, young trees; they most likely represent locally grown oak. Different
groups could be not distinguished in the set of 21 timbers with felling dates
around 1465.
The match of the average ’s-Hertogenbosch chronology with the
chronologies from adjacent countries indicates that the timbers were derived
from eastern Belgium. The average chronology of ’s-Hertogenbosch has a
higher EPS than the eastern Belgium chronology. This might in part be caused
by the larger number of series in the ’s-Hertogenbosch chronology. When the
series from eastern Belgium and ’s-Hertogenbosch are combined, the value for
EPS increases and the match with established German chronologies improves.
This indicates that in the combined chronology the ring-width variations
caused by local factors are less marked than they are in the separate
chronologies, i.e., the combined chronology reflects general climatic conditions
more accurately than the separate chronologies.
Some of the regional chronologies used in the analysis represent different
geographical regions than previously assumed. The most serious problem
concerns the chronology of the Lower Rhine, which in part consists of tree-ring
patterns from Dutch locations (Hollstein 1980). Between AD 1327 and 1465 it
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
shows a correlation of -0.60 with the Ardennes-Eiffel chronology of Hollstein
(1980), and the interval between AD 1369 and 1465 is practically identical to
his more northern chronology of the Westerwald/Sauerland area (r = 0.72).
This indicates that the provenance of the oak series used in the Lower Rhine
chronology is in another region than Hollstein assumed. It also concurs with
my experience that the Medieval part of the Lower Rhine chronology does not
match with patterns of oak from Dutch contexts.
Several conclusions were based on these findings: (1) given restricted,
geographically and chronologically well-defined data sets it is possible to
produce oak chronologies with a homogeneous signal in the Netherlands; (2)
the Mean Correlation Technique can be used to estimate the strength of the
signal in oak chronologies from Dutch contexts; (3) access is required to the
measurement series included in a chronology if the quality of this chronology
is to be assessed; (4) parts of Hollstein’s regional chronologies are not useful
for establishing the provenance of oak timber; (5) many oak timbers used in
the Netherlands were imported, i.e., cannot be used to construct chronologies
of indigenous oak.
The Sub-Fossil Forests project was partly launched in order to collect a data
set of indigenous oak. The methodological purpose was to empirically
establish the strength of the environmental signal, estimated through the
Expressed Population Signal (EPS), in data sets derived from oaks that
simultaneously grew at the same (oceanic) site. The largest data set was
derived from the province of Zuid-Holland, from locations near Alphen aan de
Rijn, Zoeterwoude and Hazerswoude (Chapter 4). Of the 55 trunks that were
sampled, 36 were found to match, resulting in a chronology of 1118 years
(NLPre_ZH, 2258 -1141 BC). NLPre_ZH could be matched with the North
German bog oak chronology (Leuschner and Delorme 1988).
One of the main findings was that EPS should be estimated on short
consecutive intervals of a long chronology, not on the complete interval; in the
latter case, the value for EPS overestimates the signal (Chapter 4). Hundredyear intervals of NLPre_ZH require 10 to 15 series in order to reach values for
EPS higher than 0.85. This contrasts with chronologies of arid-site conifers in
the western United States, which require a minimum of four series, and with
chronologies of deciduous oak in the United Kingdom, which require a
minimum of 25 series (Briffa and Jones 1990). It was noted that although the
residual NLPre_ZH chronology has both a higher within-tree and between-tree
signal than the standard chronology due to the removal of the auto-correlation,
its effective chronology signal as estimated by the Mean Correlation
Technique is lower. This was ignored because the differences between the
values were small.
NLPre_ZH matches well with the North German bog oak chronology
(Chapter 4). Both this match and the values for EPS indicate that the patterns
of bog oaks from Zuid-Holland contain a strong environmental signal. This
means that contrary to former expectations oaks in the western Netherlands
are stressed by non-local environmental factors. Given the fact that bog oaks
are found in former bogs, whereas living oaks in the Netherlands prefer welldrained Pleistocene soils, the bog oaks started growing in relatively dry
conditions and became stressed as a result of raised water tables (which also
increased the development of local bogs, in which the remains of the trees
were preserved). The water table is related to precipitation and temperature,
i.e. climate. This means that the variability in bog oak chronologies reflects
climate as well. The main conclusion is that, contrary to the dendrochronological findings of the 1960’s and 1970’s, oaks from Dutch sites are
well-suited for developing climate-sensitive, indigenous chronologies.
This finding was tested by the analysis of a heterogeneous data set from the
Iron Age/Roman period, consisting of 195 series of dated bog oaks and
archaeological timbers from a wide variety of sites (Chapter 5). Using
correlation techniques and dendrochronological matching, four groups could
be distinguished. NLRom_R, from 46 trees derived from six natural and four
archaeological sites throughout the Netherlands, runs from the end of the
chapter 8 | Discussion and Conclusions
Middle Iron Age to Early Merovingian times (325 BC - AD 563; EPS = 0.96).
Given the fraction of bog oaks in this chronology, NLRom_R represents
indigenous oak. The three other chronologies exclusively represent timbers
from archaeological sites. The first, NLRom_W1 (84 BC - AD 50), consists of
14 series from Leidschendam and Velsen and has an Expressed Population
Signal (EPS) of 0.92. This chronology most likely represents indigenous oak
from the western Netherlands, since it contains young, undressed timber and
matches better with NLRom_R than with established chronologies from
neighbouring countries. The second chronology, NLRom_W2 (140 BC AD 87; EPS = 0.98), represents 42 timbers from Nieuwenhoorn and Velsen.
In Chapter 5 it was argued that this chronology represents trees from the
eastern Netherlands and adjacent areas of Germany, but this is contradicted by
new evidence (see below). The third chronology, NLRom_E (AD 190 - 395;
EPS = 0.98), represents the series of 92 trees from archaeological sites at Cuyk,
Gennep and Heeten. Given the distribution of the sites and the number of
young, undressed trees (Cuyk), this chronology most likely represents forests in
the southeastern Netherlands that were situated close to the archaeological
sites. In other words: (1) in Dutch tree-ring data from cultural contexts (about
which no a priori assumptions can be made regarding the provenance) subgroups are characterized by different signal contents; (2) the most likely
provenance can be deduced from the geographical distribution of the sites
represented by these groups and from additional evidence such as the age of the
trees and whether oak was commonly used in this period and district.
Over 80% of the dated series of timber from Dutch Medieval contexts were
clustered according to their match as expressed by correlation coefficients
(Chapter 6). Three historical chronologies resulted: NLHist_1, running from
AD 427 to 1752 (259 trees), NLHist_2, running from AD 1023 to 1666 (195
trees) and NLHist_3, running from AD 1041 to 1346 (30 trees). The analysis of
EPS showed that NLHist_1 and NLHist_2 contain too few samples before AD
1100 and 1300, respectively, but that NLHist_3 contains sufficient samples.
It was noted that intervals in NLHist_1 with low values for EPS synchronize
with intervals in the radiocarbon calibration curve that are characterized by
anomalous values for 14C. Based on the match with chronologies from
neighbouring countries it was established that (1) the variability in NLHist_1
closely matches that in the established chronologies, with the exception of the
first and last century. Excluding the first and last part NLHist_1 should be
useful as a reference for dating, even if no series are added; (2) NLHist_2 does
not match well with established chronologies before AD 1300. It requires more
series between AD 1023 (its first year) and AD 1300; (3) NLHist_3 does not
match with any established chronology and this situation is not likely to
improve if series are added.
In the Netherlands the provenance of Medieval timbers is heterogeneous,
which is illustrated by the problems of interpretation regarding the provenance
of Chronologies 1 and 2 (Eckstein et al. 1975). In order to investigate the
provenance of Dutch timbers, and to reduce the chance that chronologies of
locally grown oak would not be recognized, a regional analysis of the new
timber chronologies was attempted. To this end, the classification by Wolff
(1989) was used; he divides the Netherlands into 26 geogenetical growth
regions. This classification does not result in detailed information, but
nonetheless shows that NLHist_1 dominates in the southern Netherlands,
NLHist_2 in the central and northern Netherlands, and NLHist_3 in coastal
regions and an occasional river valley. From this it can be concluded that the
signal in the tree-ring patterns of Medieval timbers used in the central and
northern Netherlands differs markedly from the signal in the patterns from
timbers used in the southern Netherlands. This difference might be caused by
(1) different trade routes for importing oak in the southern and central/
northern Netherlands, and/or (2) differences in the environmental conditions
that govern the growth of indigenous oak in both regions.
Given the early date of NLHist_3, and given the fact that it does neither
match with established chronologies nor with NLHist_1 and NLHist_2, the
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
timbers in this chronology may be indigenous. If this is the case,
dendrochronology in the Netherlands should accept the explanation for
deviating characteristics in Dutch chronologies offered by Eckstein et al.
(1975), namely that some areas in the Netherlands were exploited to such an
extent that in later times no wood from these areas was available, and hence
chronologies that represent wood from these areas cannot be matched with
established chronologies. Before doing so, however, one should investigate the
possibility that NLHist_3 represents trees from one of the neighbouring
countries.
I then turned to the question of why, compared to the standard chronology,
the residual NLPre_ZH chronology is characterized by a higher within-tree and
between-tree signal, but by a lower effective chronology signal. The introduced
hyperbolical relationship between these variables (Briffa and Jones 1990)
contradicts the fact that ring-width variability caused by tree-specific
endogenous influences reduces and/or conceals the variability caused by more
general environmental (exogenous and climatological) influences. If the
chronology signal is indeed hyperbolically related to the within-tree signal, the
application of noise reducing techniques to series that represent the same tree
would be superfluous or even counterproductive. In order to investigate the
relationship between the within-tree and effective chronology signal the domain of
these variables was analysed (Chapter 7). It was found that: (1) the derivation
of the signals according to the definition by Wigley et al. (1984) and Briffa and
Jones (1990) results in an underestimation of the actual signal, especially
when the correlations approach the limits of the domain of r [-1,1].
The correlation coefficients from which the signal is estimated should be
transformed to Fisher’s z-scores before they are averaged; (2) the effective
chronology signal, from which the Expressed Population Signal (EPS) is derived
(Briffa and Jones 1990), is treated as, but in fact is not, a correlation
coefficient; (3) the within-tree signal is inadequately expressed in the effective
chronology signal.
In view of conclusion (3), the values for EPS in Chapters 5 and 6 are
derived from the between-tree signal only. Conclusion (1) was established too
late during the analysis to be put into effect. The implication of conclusion (1)
is that the values for EPS established throughout this study underestimate the
actual strength of the signal in the chronologies (the resulting error is largest
for the highest values for EPS). Conclusion (2) was also established too late.
Once more tree-ring series have been added to the chronologies, i.e. the
overlap between the series has improved, the chronologies should be analysed
using other techniques for time-series analysis.
8.3
THE APPLICATION OF CHRONOLOGIES FROM DUTCH CONTEXTS
8.3.1 Prehistory
The prehistoric chronology NLPre_ZH, developed from bog oaks
found in the province of Zuid-Holland, contains series of 36 trees and runs
from the Late Neolithic/Early Bronze Age to the Early Iron Age (Chapter 4).
The chronological position of this and all other chronologies is shown in
Figure 8.1; further details are given in Table 8.1. We refer to these summaries
in the sequel.
The signal contained in NLPre_ZH, as estimated by EPS, is insufficient over
two-thirds of the interval. This indicates that more series should be added.
Recently, the series from three bog oaks from Papendrecht (province of ZuidHolland) and one from Wageningen (province of Gelderland) were dated
against this chronology. The Papendrecht series cover the period between
1406 and 1035 BC, i.e., since its construction NLPre_ZH has been extended
from 1141 to 1035 BC. Interestingly, a fourth bog oak series from Papendrecht
matches well with archaeological series from Ede, Oss-Mettegeupel and
chapter 8 | Discussion and Conclusions
Spijkenisse (provinces of Gelderland, Noord-Brabant and Zuid-Holland) and
with a single bog-oak series from Weesp (province of Noord-Holland; RING,
unpublished data). The combined chronology of the five locations, which was
matched against the North German bog oak chronology, runs from the eleventh
to the seventh century BC; a gap of seven years (1034 - 1028 BC) separates it
from the extended, as yet unpublished, NLPre_ZH chronology. These new
results show that (1) NLPre_ZH is suitable as a reference for dating bog oak
series and (2) the tree-ring patterns in bog oaks from the first 350 years of the
first millennium BC do match with the patterns in timber from cultural
contexts. This is especially important in view of the chronological gap in the
Dutch chronologies between 1141 (currently 1035) and 325 BC.
FIGURE 8.1 - The chronological
position of oak tree-ring chronologies
for the Netherlands. Chronologies 1
and 2 are described by Eckstein et al.
1975
TABLE 8.1 - Overview of new treering chronologies for the Netherlands
Chronology
First Year
Last Year
No. of trees
Context
natural
archaeol.
Reference
historical
NLPre _ZH
2258 BC
1141 BC
36
+
-
-
Chapter 4
NLRom_R
325 BC
AD 563
46
+
+
-
Chapter 5
NLRom_W2
140 BC
AD 87
42
-
+
-
Chapter 5
NLRom_W1
84 BC
AD 50
14
-
+
-
Chapter 5
NLRom_E
AD 190
AD 395
92
-
+
-
Chapter 5
NLHist _1
AD 427
AD 1752
259
-
+
+
Chapter 6
NLHist _2
AD 1023
AD 1666
195
-
+
+
Chapter 6
NLHist _3
AD 1041
AD 1346
30
+
+
-
Chapter 6
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
The first millennium BC is somewhat of an enigma in palaeodendrochronology. First, established oak chronologies that cover this period show
minimal replication at various intervals between 800/700 and 400 BC
(Schmidt and Gruhle 1988; Baillie 1993; Becker 1993). Second, during this
period the similarity between the established chronologies is low (Schmidt and
Gruhle 1988). Of course, these phenomena could be related, since the
reliability of the signal in a chronology (in part expressed by the strength of the
match with other chronologies) is, among other things, related to the number
of samples. On the other hand, they could also be the expression of climatic
factors, such as a sudden environmental change that caused (1) the death of
oaks on low sites (fewer bog oaks) and (2) an increased local signal in
surviving oaks (decreased correlations among established chronologies).
In view of the anomalous 14C production during this interval (the ‘Hallstatt
Plateau’), Schmidt and Gruhle (1988) opt for the latter explanation and state
that the climate in general became cooler and wetter. They argue that
established chronologies do not match well during this period as a result of the
increased locality of the signal in patterns of oak. The recently found match
between series from Papendrecht, Weesp, Ede, Oss-Mettegeupel and
Spijkenisse (and between their average chronology and the North German bog
oak chronology) contradicts this line of reasoning, at least for the period
between 1027 and 658 BC; it proves that oak series from this period, derived
from different natural and archaeological sites, can be matched. In other
words, oak series from the first 350 years of the first millennium BC contain a
strong, and similar, environmental signal. This indicates that it is only a matter
of time before NLPre_ZH can be linked to the newly dated series and
extended into the first millennium BC. In the Netherlands no bog oaks have
been found that lived from the mid-eighth century to the fourth century BC.
This may indicate that oaks growing on sites at low elevations did not survive
past the eighth century BC. Possible explanations are a raised water table,
increased marine/fluvial activity at low sites, and increased development of
bogs. However, the lack of bog oak samples in the Netherlands for this period
may also, at least in part, be the result of sample bias.
Linking NLPre_ZH to the more recent chronology NLRom_R (325 BC AD 563) is one of the current priorities in Dutch dendrochronology, because
it will allow not only a closer look at the impact of the assumed environmental
change on the distribution and growth characteristics of indigenous oak
during this period, but also enhance the possibility of dating archaeological
contexts from the first millennium BC. To this end two floating chronologies
have already been established that contain oak tree-ring series from
archaeological sites dated to the first millennium BC. The first chronology
represents archaeological series from the Velserbroekpolder (province of
Noord-Holland); the second chronology represents one archaeological series
from Wommels (province of Friesland) and bog oak series from Papendrecht
(province of Zuid-Holland) and Vriezenveen (province of Overijssel).
8.3.2 The Iron Age/Roman period
NLRom_R represents bog oaks and archaeological material from a
variety of locations. Its signal was not analysed for different intervals. Initially,
its use as a reference chronology for dating appeared to be limited; the first
and last parts of NLRom_R (the segments before 100 BC and after AD 400)
do not match well with established chronologies, and its use as a reference
chronology did not result in any new dates for undated oak series for this
period. Given that the first and last interval of NLRom_R consist of only a few
series, more samples should be included. Nonetheless, the dendrochronological dates of about ten archaeological timbers from sites in Oss-Ussen and
Valkenburg (provinces of Noord-Brabant and Zuid-Holland), which I
established in the earliest phase of my research, were confirmed by matching
with NLRom_R (Appendix A). And, more recently, two archaeological
samples from Eme (Zutphen, province of Gelderland), which were undatable
chapter 8 | Discussion and Conclusions
with any other chronology, could be dated to the sixth century by matching
with NLRom_R (RING, unpublished data). This indicates that also in its
current form NLRom_R is suitable as a reference chronology for dating.
No problem exists regarding the sample size of the three archaeological
chronologies from this period. Only one timber, derived from Valkenburg
(province of Zuid-Holland), has been dated so far by a match with NLRom_W1.
NLRom_W2 was recently used as a reference to date timbers from
archaeological sites at Alphen aan de Rijn and Spijkenisse (province of ZuidHolland; RING, unpublished data). In Chapter 5 it was argued that, in view of
the match with established German chronologies, NLRom_W2 represents trees
that grew in the eastern Netherlands. The new dates established through this
chronology, however, exclusively apply to archaeological sites in the province of
Zuid-Holland. This indicates that the interpretation presented in Chapter 5
may be wrong and that NLRom_W2 instead consists of series from ZuidHolland, i.e. the western Netherlands. NLRom_E has been used as a reference
to date timbers from archaeological sites at Bergeijk (province of NoordBrabant) and, in an earlier version, Wehl (province of Gelderland). Given the
distribution of the sites represented by NLRom_E (situated in the provinces of
Limburg and Noord-Brabant) as well as that of the sites that can be dated by it,
this chronology represents forests located in the southeastern Netherlands.
Linking the Iron Age/Roman period chronologies to the historical timber
chronologies NLHist_1, NLHist_2 and NLHist_3 is a second priority in current
Dutch dendrochronological research. Although NLRom_R overlaps with
NLHist_1 for more than a hundred years, the chronologies do not crossdate and
cannot be combined. This could be related to differences in their geographical
distribution; NLRom_R represents oak from low, wet locations in the central
Netherlands, whereas NLHist_1 represents oak from a broad geographical
region that ranges from the Pleistocene soils in the southern Netherlands and
Belgium to Central Germany. If NLHist_1 can be linked to an earlier
chronology, the most likely candidate is NLRom_E, which represents oak that
grew in the southeastern Netherlands. NLRom_E runs no further than AD 395
(Table 8.1). Given that dendrochronological matching requires an overlap of 80
years or more, this chronology needs to be extended at least a hundred years
before the link with NLHist_1 can be established. It is, however, uncertain
whether an extended version of NLRom_E will match with NLHist_1, because
the size of the regions represented by both chronologies differs considerably.
8.3.3 The Merovingian and Carolingian periods
Given the geographical differences between the newly established
historical chronologies and the fact that the more recent chronologies among
them are either of insufficient quality before AD 1000 (NLHist_1) or do not
even extend back this far (NLHist_2 and 3), more dendrochronological data are
needed for the period between AD 400 and 1000. It has been shown that
archaeology will profit greatly from precise dating for this particular interval
(Chapter 1). Dendroclimatology provides additional reasons to focus on this
period. In AD 536, either a volcano erupted or meteors struck. A cooling of the
climate resulted that lasted at least a decade, and is recorded in tree-rings from
Ireland and Germany (Baillie 1994). We cannot look at the environmental
impact in the Netherlands, because this particular year occurs only in two bog
oak series (NLRom_R) and one archaeological series (NLHist_1). The tenth
century, also characterized by climatic anomalies (a cooling of the climate
caused by an eruption of the Icelandic vulcano Eldgjá in about AD 940
(Zielinski et al. 1995); extreme drought in the Netherlands (Heidinga 1987)), is
represented by about ten series (NLHist_1). Given their weak match
(represented by a slight dip in the value for EPS; Chapter 6), more series should
be added to this interval, both to strengthen the chronology signal and its
quality for dating and to improve the quality of NLHist_1 for climate studies.
A recent success in this respect was the dating of four oak posts from Tiel
(province of Gelderland) to the tenth century by matching their series with
NLHist_1 (RING, unpublished data).
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
8.3.4 The Late Middle Ages
NLHist_1 is the longest and best-replicated of the newly established
Medieval chronologies. It has, however, resulted in only a few new
dendrochronological dates after AD 1100. This is related to a tautology that
was implicit in the research from the very beginning. The undertaking was to
match samples from Dutch locations with established chronologies from
neighbouring countries, and then to compile the dated series into new Dutch
chronologies. In this way, a ‘Dutch’ chronology resulted that mirrors an
established chronology abroad: NLHist_1 strongly resembles, and possibly
repeats, Hollstein’s Central German oak chronology (r = 0.62; Chapter 6), and
to a lesser extent resembles the chronology of eastern Belgium, which was
produced from patterns matched with the Central German chronology
(Hoffsummer 1989). In other words, the reason that the interval of NLHist_1
after AD 1100 is not important in dendrochronological dating is because wellreplicated chronologies for eastern Belgium and adjacent areas of Germany
already exist for this period. Tree-ring series that can be matched with the
established chronologies of eastern Belgium and Central Germany can also be
matched with NLHist_1, and vice versa; hence the lack of unique dating
results through the application of NLHist_1.
The historical chronology NLHist_2 has proven to be the most useful of the
new Medieval chronologies in recent dating research; quite a few tree-ring
patterns that were undatable with established chronologies could be matched
with NLHist_2. Although its Expressed Population Signal (EPS) between AD
1023 and 1300 is weak, this early interval is nevertheless suitable for dating
oak from cultural contexts. This is illustrated by recent datings for the N.H.
Kerk in Oudewater (province of Utrecht), a church with building phases in the
fourteenth to sixteenth century. The oldest timbers in this church are in the
roof of the tower and the nave. Their tree-ring series match well with
NLHist_2 and run back to AD 1182, i.e., well into the interval of NLHist_2
that is characterized by low values for EPS.
In summary, it can be stated that the objectives as outlined in the
introduction were well achieved. New oak chronologies have been produced
from series derived from natural sites and archaeological and historical
structures in the Netherlands. These chronologies allow excellent dating of
archaeological and historical material for the periods 2258 - 1141 BC and 325
BC - AD 1752.
SAMENVATTING
In dit proefschrift komen diverse methoden en toepassingen van
dendrochronologie aan de orde. Hieronder zijn de resultaten samengevat.
Paragraaf A bevat een discussie over de gebruikte methoden, paragraaf B een
discussie over de wijze waarop de nieuwe jaarringkalenders sinds hun
vervaardiging worden toegepast.
A
METHODOLOGISCHE ASPECTEN: HET SCHATTEN VAN HET SIGNAAL IN
JAARRINGKALENDERS
Hoofdstuk 1
In het midden van de 80er jaren was weinig bekend over de reactie van
Nederlandse eiken op het klimaat. Gebleken was dat de groeipatronen van
levende bomen, bouwhout en schilderijpanelen (alle eik) dateerbaar waren met
Duitse kalenders; dit had geleid tot de samenstelling van Chronologie 1 en
Chronologie 2 (Eckstein et al. 1975). Chronologie 1 en 2 vertoonden echter geen
onderlinge overeenkomst, wat erop leek te wijzen dat in Nederland in het
verleden duidelijk onderscheiden bosbestanden bestonden, elk gekenmerkt
door een ander klimaatsignaal (de variatie van ringbreedten die bomen op
verschillende standplaatsen gemeen hebben), en dat deze bestanden ten dele
zodanig werden geëxploiteerd dat ze later voorgoed verdwenen waren. Voorts
was gebleken dat Chronologie 1 en 2 niet goed bruikbaar waren als standaard
ter datering van ongedateerd eikehout uit Nederlandse vindplaatsen en
gebouwen. Dit leek erop te wijzen dat het signaal in Nederlandse eiken in het
verleden gedomineerd werd door lokale groeisignalen (ringbreedtevariaties die
alleen voorkomen in bomen op een enkele standplaats, of slechts in een enkele
boom). Aan de andere kant was het mogelijk gebleken om de jaarringpatronen
van veeneiken uit het West-Nederlandse kustgebied te correleren met de
Centraal-Duitse eikkalender van Hollstein (Hollstein 1980; Jansma 1987); dit
wees er juist op dat de groeipatronen van in Nederland gegroeide eiken wèl
een algemeen klimaatsignaal kunnen bevatten, en zelfs een signaal dat
overeenkomsten vertoont met het signaal in patronen van eiken uit regio’s met
een landklimaat.
Hoofdstuk 2
Om meer inzicht te verkrijgen in de soorten variabiliteit die aanwezig zijn in
de jaarringpatronen van Nederlands eiken, analyseerde ik de ‘Mean
Sensitivity’ () van eiken uit natuurlijke en antropogene contexten uit de
IJzertijd/Romeinse Tijd. De beschrijvende parameter werd door
A.E. Douglass (1928) ontwikkeld als schatter van de variatie van ringbreedten
van jaar tot jaar; deze variatie reflecteert tot op zekere hoogte de mate waarin
boomgroei afhankelijk is van, en onderdrukt wordt door het klimaat (de
sterkte van het klimaatsignaal). Hoge waarden voor werden vooral verklaard
Samenvatting
in termen van stress die samenhangt met continentale klimatologische
omstandigheden (Fritts 1976; Fürst 1963; Fürst 1978). Omdat de patronen van
bomen die reageren op klimatologische stress vaak dateerbaar zijn, werd de
waarde voor daarnaast opgevat als een indicatie van het daterend potentieel
van jaarringpatronen. Mijn belangrijkste bevinding was dat jaarringpatronen
van eiken uit natuurlijke en antropogene vindplaatsen in West-Nederland
gekenmerkt worden door even hoge waarden voor als patronen uit gebieden
met een landklimaat, zoals Centraal-Duitsland (Jansma, ongepubliceerde
gegevens). De conclusies waren: (1) de waarde voor weerspiegelt niet alleen
continentale, maar ook oceanische omgevingsinvloeden; (2) indien de waarde
voor gerelateerd is aan stress kunnen de groeiomstandigheden voor eiken in
regio’s met een zeeklimaat even extreem zijn als in regio’s met een landklimaat.
Verder werd vastgesteld dat langdurige stress niet kan worden uitgedrukt in ,
omdat langdurig onderdrukte groei resulteert in reeksen jaarringen die zeer
smal zijn en daardoor weinig variatie vertonen van jaar tot jaar, dus in lage
waarden voor .
Dit onderzoek werd niet gepubliceerd; het gegevensbestand waarop de
analyse was gebaseerd, bleek te klein te zijn en bleek als parameter
problematisch. In samenwerking met Prof. Dr. J. Strackee (Afd. Medische
Fysica en Informatica, Universiteit van Amsterdam) werd vastgesteld dat een
tijdreeks die een lognormale verdeling heeft, zoals een reeks geïndexeerde
jaarringbreedten, een waarde voor heeft die afhankelijk is van de standaarddeviatie en de eerste-orde autocorrelatie van de reeks. Er werd een model van
deze relatie ontwikkeld en toegepast op bestaande jaarringseries. De belangrijkste conclusie was dat volledig afhankelijk is van 公(1-r1), waarbij  de
geschatte standaarddeviatie van de reeks is en r1 de geschatte autocorrelatie. Dit
houdt in dat de waarde voor afgeleid kan worden uit de standaarddeviatie ()
en de eerste-orde autocorrelatie (r1) van een reeks (een hoge waarde voor kan
het resultaat zijn van een hoge waarde voor , een lage waarde voor r1, of een
combinatie van beide). Dit verklaart overigens waarom slechts lage waarden
bereikt voor jaarringpatronen die langdurige onderdrukte groei vertonen;
reeksen smalle ringen (lage meetwaarden) worden gekenmerkt door lage
waarden voor en hoge waarden voor r1, hetgeen tot lage waarden voor leidt.
Hoofdstuk 3
1. Veldatum: het jaar waarin een boom is
geveld.
Omdat problematisch was gebleken, richtte ik mij op dendroklimatologische methoden om het signaal in jaarringreeksen te schatten. ANOVA
(variantieanalyse; Fritts 1976) en de ‘Gemiddelde Correlatietechniek’ (Wigley
et al. 1984) worden in de dendroklimatologie gebruikt om het signaal te
schatten in jaarringreeksen van levende bomen die tot hetzelfde bosbestand
behoren. In tegenstelling tot ANOVA vereist de Gemiddelde Correlatietechniek
niet dat alle jaarringreeksen volledig in de tijd overlappen; daarom is deze
methode goed toepasbaar bij de analyse van lange kalenders die uit korte
jaarringreeksen zijn opgebouwd, zoals kalenders van eik. Aanvankelijk werd de
aangepaste correlatietechniek van Briffa en Jones (1990) gebruikt, waarbij het
in-de-boom signaal (de gemiddelde correlatie tussen jaarringreeksen die dezelfde
boom vertegenwoordigen) wordt onderscheiden van het tussen-de-bomen signaal
(de gemiddelde correlatie tussen reeksen die verschillende bomen
vertegenwoordigen). Uit deze beide wordt het effectief chronologiesignaal afgeleid,
waaruit het ‘Populatie Signaal’ (EPS) wordt geschat. Omdat EPS in
klimatologische studies 0.85 of hoger moet zijn (Wigley et al. 1984), werd
tijdens alle analyses een drempelwaarde van 0.85 aangehouden, wat inhoudt dat
(intervallen van) kalenders met lagere waarden voor EPS als onbetrouwbaar
werden beschouwd.
Als eerste werd de bruikbaarheid getest van EPS als schatter van het signaal
in Nederlandse jaarringreeksen. Hierbij werd een gegevensbestand gebruikt uit
een beperkt geografisch en temporeel domein: bouwhout met veldata1 tussen
1463 en 1465 n. Chr., dat in ’s-Hertogenbosch (Noord-Brabant) was gebruikt
om huizen te renoveren die in 1463 door brand waren verwoest. ’s-Hertogenbosch ligt vlakbij de Maas. De Oostbelgische eikkalender, vervaardigd uit de
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
patronen in bouwhout uit de Maasvlakte, was zojuist beschikbaar gekomen
(Hoffsummer 1989), en op verzoek werden ook de individuele metingen
beschikbaar gesteld.
De historische informatie over de houthandel en het bosbeheer in de regio
laat zien dat in ’s-Hertogenbosch in de vijftiende eeuw zowel lokaal gegroeide
eiken werden gebruikt, als eikehout dat van elders werd aangevoerd via de Rijn
en Maas (Vink 1990; Vink 1993). Meer dan honderd stukken bouwhout
werden geanalyseerd; de meeste hiervan vertegenwoordigden ondateerbare,
jonge en snelgegroeide bomen uit de onmiddellijke omgeving van ’s-Hertogenbosch. Binnen het bestand van 21 houtmonsters met veldata tussen 1463
en 1465 konden geen verschillende groepen worden onderscheiden; uit deze
jaarringseries werd een ’s-Hertogenbosch kalender vervaardigd.
De overeenkomst van de ’s-Hertogenbosch kalender met bestaande
kalenders uit de omringende landen geeft aan dat het bouwhout met veldata
tussen 1463 en 1465 afkomstig was uit Oost-België en daarom waarschijnlijk
via de Maas moet zijn aangevoerd. De ’s-Hertogenbosch kalender heeft een
hogere waarde voor EPS dan de Oost-België kalender. Dit zou ten dele
veroorzaakt kunnen zijn door het grotere aantal jaarringpatronen in de
’s-Hertogenbosch kalender. Wanneer de reeksen in beide kalenders worden
gecombineerd, neemt de waarde voor EPS toe en wordt de overeenkomst met
de overige kalenders sterker. Met andere woorden, in de gecombineerde
’s-Hertogenbosch/België kalender zijn de door lokale invloeden veroorzaakte
ringbreedtevariaties zwakker, en de door algemeen klimatologische invloeden
veroorzaakte variaties sterker, dan in de afzonderlijke kalenders.
Enkele van de regionale kalenders die tijdens de analyse werden gebruikt,
bleken een andere regio te vertegenwoordigen dan algemeen werd
aangenomen. Het grootste probleem betrof de Neder-Rijn kalender, die ten
dele bestaat uit patronen van hout uit Nederlandse vindplaatsen (Hollstein
1980). Tussen 1327 en 1465 n. Chr. heeft deze kalender een correlatie (r) van
-0.60 met Hollstein’s Ardennen-Eiffel kalender, en het interval tussen 1369 en
1465 n. Chr. is praktisch identiek aan Hollstein’s meer noordelijke
Westerwald-Sauerland kalender (r = 0.72). Dit houdt in dat de herkomst van
het hout waarop de Neder-Rijn kalender is gebaseerd, in een ander gebied ligt
dan Hollstein aannam. Ook komt het overeen met mijn ervaring dat deze
kalender onbruikbaar is ter datering van eik uit Nederlandse vindplaatsen.
Deze bevindingen leidden tot de volgende conclusies: (1) gegeven beperkte
en in geografisch en chronologisch opzicht goed gedefinieerde
jaarringbestanden is het in Nederland mogelijk om jaarringkalenders te
vervaardigen met een homogeen signaal; (2) de Gemiddelde
Correlatietechniek kan inderdaad gebruikt worden om de kracht van het
signaal te schatten in kalenders van hout uit Nederlandse vindplaatsen; (3) om
de kwaliteit van een kalender te kunnen vaststellen, is toegang tot de
individuele jaarringreeksen waaruit de kalender is opgebouwd onontbeerlijk;
(4) Hollstein’s regionale kalenders zijn ten dele onbruikbaar als standaard ter
bepaling van de herkomst van eiken bouwhout; (5) veel van het in Nederland
toegepaste bouwhout is van elders afkomstig en is daarom onbruikbaar als
uitgangspunt voor Nederlandse jaarringkalenders.
Hoofdstuk 4
Het Subfossiele Bossenproject (RING/ROB) werd onder andere gestart om
een gegevensbestand van inheems eiken te verzamelen. In methodologisch
opzicht was het doel empirisch vast te stellen hoe sterk het omgevingssignaal,
zoals geschat door EPS, is in patronen van inheemse eiken die tegelijk
groeiden op dezelfde oceanische standplaats. Het grootste bestand werd
verzameld in de provincie Zuid-Holland, op lokaties bij Alphen a/d Rijn,
Zoeterwoude en Hazerswoude. Van de 55 bemonsterde stammen konden 36
onderling gedateerd worden, hetgeen resulteerde in een 1118 jaar lange
kalender, NLPre_ZH, die loopt van 2258 tot 1141 v. Chr. NLPre_ZH kon
absoluut gedateerd worden met behulp van de Noordduitse veeneikkalender
(Leuschner and Delorme 1988).
Samenvatting
Een van de belangrijkste bevindingen was dat EPS geschat dient te worden
voor korte opeenvolgende intervallen van lange kalenders, niet voor het hele
interval; in dat laatste geval overschat de waarde voor EPS het feitelijke signaal.
In NLPre_ZH moeten intervallen van 100 jaar gerepresenteerd worden door
minstens 10 à 15 jaarringpatronen, om waarden voor EPS te bereiken die hoger
zijn dan 0.85. Dit in tegenstelling tot kalenders van naaldbomen op droge
lokaties in het westen van de Verenigde Staten, die minimaal 4 patronen nodig
hebben, en kalenders van eik in Engeland, die gerepresenteerd moeten worden
door minstens 25 patronen (Briffa en Jones 1990). Voorts werd opgemerkt dat
het effectief chronologiesignaal in de ‘residu’-kalender van NLPre_ZH (de versie
waaruit de autocorrelatie is gefilterd) lager is dan in de ‘standaard’-kalender
(waarin de autocorrelatie aanwezig is), terwijl zowel het in-de-boom signaal als
het tussen-de-bomen signaal in de eerste sterker is dan in de tweede. Dit werd
genegeerd, omdat het verschil tussen de waarden klein was.
NLPre_ZH komt goed overeen met de Noordduitse veeneikkalender. Zowel
deze overeenkomst als de waarden voor EPS geven aan dat de groeipatronen
van de Zuidhollandse veeneiken in sterke mate een door maritieme omgevingsinvloeden gestuurd signaal bevatten. Met andere woorden, in tegenstelling tot
wat men eerder aannam werd de groei van eiken in West-Nederland niet
gedomineerd door lokale, maar over grotere gebieden werkzame omgevingsinvloeden. Gegeven het feit dat deze eiken in (voormalige) venen worden
aangetroffen, terwijl eik nu vooral groeit op de Pleistocene zandgronden,
moeten de in venen gevonden eiken in relatief droge omstandigheden zijn
ontkiemd, waarna ze in problemen kwamen door een verhoging van het
grondwaterniveau (die eveneens resulteerde in een uitbreiding van de bestaande
venen). Het grondwaterniveau is gerelateerd aan het klimaat, wat inhoudt dat
de variatie van de ringbreedten in veeneiken een klimatologische component
heeft. In tegenstelling tot de dendrochronologische bevindingen van de 60er en
70er jaren is de belangrijkste conclusie hieruit dat eiken uit Nederlandse
vindplaatsen goed bruikbaar zijn voor de ontwikkeling van aan het klimaat
gerelateerde kalenders.
Hoofdstuk 5
Bovenstaande conclusie werd geverifieerd met behulp van een heterogeen
bestand van jaarringpatronen uit de IJzertijd/Romeinse Tijd, bestaande uit 195
patronen van zowel veeneiken als archeologisch materiaal uit zeer diverse
vindplaatsen. Met behulp van correlatietechnieken en dendrochronologische
dateringsmethoden konden in dit bestand vier groepen worden onderscheiden,
die ieder tot een afzonderlijke kalender leidden. De kalender NLRom_R
vertegenwoordigt 46 bomen uit zes natuurlijke en vier archeologische
vindplaatsen en loopt van de Midden IJzertijd tot de vroeg-Merovingische
periode (325 v. Chr. - 563 n. Chr.; EPS = 0.96). Gegeven het feit dat hier
veeneiken in opgenomen zijn moet deze kalender inheemse eiken vertegenwoordigen. De overige drie kalenders vertegenwoordigen uitsluitend bouwhout
dat in archeologische vindplaatsen is aangetroffen. De eerste archeologische
kalender, NLRom_W1 (84 v. Chr. - 50 n. Chr.), bestaat uit series uit
Leidschendam en Velsen (14 totaal) en heeft een waarde voor EPS van 0.92.
Gegeven de sterke overeenkomst met NLRom_R en het aandeel van jonge
onbewerkte bomen in deze kalender vertegenwoordigt NLRom_W1
waarschijnlijk eiken die in West-Nederland zijn gegroeid. De tweede
archeologische kalender, NLRom_W2 (140 v. Chr. - 87 n. Chr.), bestaat uit
series uit Nieuwenhoorn en Velsen (42 totaal) en heeft een waarde voor EPS
van 0.98. In Hoofdstuk 5 wordt beargumenteerd dat deze kalender eiken uit
Oost-Nederland en aangrenzend Duitsland vertegenwoordigt, maar dit wordt
tegengesproken door nieuwe gegevens (zie onder). De derde archeologische
kalender, NLRom_E (190 - 395 n. Chr.; EPS = 0.98), bestaat uit series uit
Cuyk, Gennep en Heeten (92 totaal). Gegeven de lokatie van de vindplaatsen
en het aandeel van jonge onbewerkte bomen vertegenwoordigt deze kalender
waarschijnlijk eiken die in Zuidoost Nederland groeiden, niet ver van hun
archeologische vindplaatsen. De conclusie was dat in bestanden van jaarring-
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
series uit Nederlandse antropogene vindplaatsen subgroepen kunnen worden
onderscheiden die gekenmerkt worden door verschillende signalen. De regio
waar de bomen in deze subgroepen groeiden kan worden afgeleid uit de
geografische ligging van de vindplaatsen en aanvullende gegevens zoals de
leeftijd van de bomen en de mate waarin eik in het betreffende gebied en
dezelfde periode als bouwhout werd gebruikt.
Hoofdstuk 6
Vervolgens werd ruim 80% van de gedateerde jaarringreeksen uit
middeleeuwse vindplaatsen in Nederland geclusterd naar aanleiding van de
overeenkomst tussen de reeksen zoals deze tot uitdrukking komt in correlatiecoëfficiënten. Hieruit resulteerden drie historische kalenders: NLHist_1 loopt
van 427 tot 1752 n. Chr. (259 bomen), NLHist_2 van 1023 tot 1666 n. Chr.
(195 bomen) en NLHist_3 van 1041 tot 1346 n. Chr. (30 bomen). Uit de
waarden voor EPS blijkt dat NLHist_1 en NLHist_2 te weinig jaarringreeksen
bevatten voor respectievelijk 1100 en 1300 n. Chr., en dat NLHist_3 over zijn
gehele lengte uit voldoende reeksen bestaat. Interessant is dat intervallen van
NLHist_1 die worden gekenmerkt door lage waarden voor EPS,
synchroniseren met uit de calibratiecurve bekende perioden van afwijkende
14
C-produktie. Op basis van de overeenkomst met bestaande kalenders uit de
omringende landen werden de volgende conclusies getrokken: (1) met
uitzondering van de eerste en laatste honderd jaar komt de variatie van de
jaarlijkse groeiwaarden in NLHist_1 overeen met die in de bestaande
kalenders; dit betekent dat NLHist_1, met uitzondering van het eerste en
laatste deel, goed bruikbaar is als standaard voor dateren, zelfs indien geen
nieuwe series worden toegevoegd; (2) het interval voor 1300 n. Chr. van
NLHist_2 correspondeert niet goed met de bestaande kalenders; met andere
woorden, aan de eerste tweehonderd jaar van deze kalender dienen series te
worden toegevoegd; (3) NLHist_3 correspondeert slechts zwak met de
bestaande kalenders en deze situatie zal niet verbeteren wanneer reeksen
worden toegevoegd.
De herkomst van eikehout dat in de Middeleeuwen in Nederland werd
gebruikt is divers, hetgeen geïllustreerd wordt door de interpretatieproblemen
rond de herkomst van Chronologie 1 en 2 (Eckstein et al. 1975). Om meer
inzicht te krijgen in de herkomst van Nederlands bouwhout en om de kans te
verkleinen dat kalenders van lokaal gegroeid eiken niet herkend zouden
worden, werd een regionale analyse gemaakt van de nieuwe bouwhoutkalenders. Hierbij werd de classificatie van Wolff (1989) gebruikt, volgens
welke Nederland opgedeeld wordt in 26 geogenetische groeigebieden. Hoewel
deze analyse niet tot gedetailleerde nieuwe informatie leidde, werd toch
duidelijk dat NLHist_1 dominant is in Zuid-Nederland, NLHist_2 in Centraalen Noord-Nederland, en NLHist_3 in de kustgebieden en een incidentele
riviervallei. Hieruit blijkt dat het signaal in de jaarringpatronen van
Middeleeuws eiken dat in Centraal- en Noord-Nederland werd toegepast,
sterk verschilt van het signaal in eikehout dat zuidelijker werd gebruikt. Dit
verschil kan op twee manieren verklaard worden: (1) in Centraal- en NoordNederland werd eikehout aangevoerd vanuit andere herkomstgebieden, en
langs andere routes, dan in Zuid-Nederland; (2) de omgevingsfactoren die de
groei van inheems eiken in Centraal- en Noord-Nederland beïnvloeden, zijn
wezenlijk anders dan de omgevingsfactoren in Zuid-Nederland.
Zowel de vroege datering van NLHist_3 als de zwakke overeenkomst met de
bestaande kalenders (inclusief NLHist_1 en NLHist_2) wijzen uit dat deze
kalender mogelijk inheemse eiken vertegenwoordigt. Als dit zo is moet
dendrochronologie in Nederland de verklaring van Eckstein et al. (1975) weer
aanvaarden, volgens welke bepaalde bosbestanden in Nederland in het
verleden zodanig zijn geëxploiteerd dat ze verdwenen zijn, waardoor kalenders
van patronen uit deze bosbestanden niet overeenkomen met andere kalenders.
Voordat deze verklaring wordt aanvaard, dient echter te worden nagegaan of
NLHist_3 wellicht bomen uit een van de andere Europese landen
vertegenwoordigt.
Samenvatting
Hoofdstuk 7
Tot slot werd de vraag gesteld waarom de ‘residu’-kalender van NLPre_ZH in
vergelijking met de standaardkalender een hoger in-de-boom signaal en tussen-debomen signaal bevat, maar een lager effectief chronologiesignaal (en dus een lagere
EPS; Hoofdstuk 7). Briffa en Jones (1990) introduceerden een hyperbolische
relatie tussen het in-de-boom signaal en het effectief chronologiesignaal (naarmate
het eerste sterker wordt, wordt het tweede zwakker). Dit is in tegenspraak met
het gegeven dat ringbreedtevariaties veroorzaakt door endogene (voor een
enkele boom geldende) omgevingsinvloeden de variaties verhullen die
veroorzaakt zijn door meer algemene (exogene en/of klimatologische) invloeden.
Als het effectief chronologiesignaal inderdaad hyperbolisch gerelateerd zou zijn aan
het in-de-boom signaal, zou de toepassing van ruis-reducerende technieken op
jaarringseries die dezelfde boom vertegenwoordigen, contraproduktief zijn. Uit
de analyse van het domein van, en de relatie tussen, deze variabelen blijkt: (1)
de wijze waarop het signaal wordt geschat door Wigley et al. (1984) en Briffa
and Jones (1990) leidt tot een onderschatting van het feitelijk signaal, met name
wanneer de correlatiecoëfficiënten de grenzen van het domein van r naderen
[-1,1]; de coëfficiënten dienen getransformeerd te worden tot Fisher’s z-scores
voordat hun gemiddelde wordt bereken; (2) het effectief chronologiesignaal (Briffa
en Jones 1990) wordt behandeld als een correlatiecoëfficiënt, maar is dit niet;
(3) het in-de-boom signaal komt onvolledig tot uitdrukking in het effectief
chronologiesignaal.
Gegeven conclusie (3) werden de waarden voor EPS in Hoofdstukken 5 en 6
afgeleid uit het tussen-de-bomen signaal, en bleef in deze hoofdstukken het in-deboom signaal buiten beschouwing. Conclusie (1) werd te laat getrokken om nog
in de berekeningen te worden opgenomen; de hier gepubliceerde waarden voor
EPS onderschatten daarom het feitelijk signaal in de kalenders (de
onderschatting is groter naarmate de waarden voor EPS hoger zijn). Ook
conclusie (2) werd te laat getrokken. Gesteld dat meer jaarringreeksen aan de
kalenders kunnen worden toegevoegd, waardoor de individuele reeksen beter
overlappen, dienen de kalenders geanalyseerd te worden met behulp van andere
technieken van tijdreeksanalyse dan de Gemiddelde Correlatietechniek.
B
DE TOEPASSING VAN KALENDERS UIT NEDERLANDSE VINDPLAATSEN
. Prehistorie
NLPre_ZH is opgebouwd uit 36 veeneiken uit Zuid-Holland en loopt van
2258 tot 1141 v. Chr. Het signaal is onvoldoende in tweederde van de
chronologie, wat betekent dat er meer jaarringpatronen in moeten worden
opgenomen. Drie recent onderzochte veeneiken uit Papendrecht (ZuidHolland) konden met NLPre_ZH gedateerd worden, waardoor de kalender met
ruim honderd jaar kon worden verlengd tot 1035 v. Chr. Ook een veeneik uit
Wageningen kon met NLPre_ZH worden gedateerd. Dit betekent dat
NLPre_ZH goed bruikbaar is als standaard voor het dateren van in Nederland
gegroeide eiken uit de betreffende periode.
Een vierde veeneik uit Papendrecht vertoont een sterke overeenkomst met
archeologische series uit Ede, Oss-Mettegeupel en Spijkenisse (Gelderland,
Noord-Brabant en Zuid-Holland) en een veeneik uit Weesp. Uit dit nieuwe
materiaal is onlangs een kalender vervaardigd die met de Noordduitse
veeneikkalender gedateerd kon worden en loopt van 1027 tot 658 v. Chr., wat
betekent dat NLPre_ZH nog maar met enkele jaren verlengd hoeft te worden
voordat koppeling met de nieuwe, jongere, kalender mogelijk is. Dit is vooral
van belang gezien het feit dat in Noordwest-Europa voor het eerste millennium
v. Chr maar weinig jaarringkalenders bestaan. Een van de prioriteiten van het
huidige dendrochronologische onderzoek in Nederland is het koppelen van
NLPre_ZH, via de nieuwe kalender, aan NLRom_R, waardoor (1) de
mogelijkheid tot het dateren van archeologisch materiaal uit het eerste
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
millennium v. Chr. toe zou nemen, en (2) jaarringonderzoek mogelijk wordt
naar de veronderstelde klimaatverandering in het eerste millennium v. Chr.
(Schmidt en Gruhle 1988).
. De IJzertijd/Romeinse Tijd
NLRom_R vertegenwoordigt veeneiken en archeologisch materiaal uit zeer
diverse vindplaatsen, en loopt van 325 v. Chr. tot 563 n. Chr. De intervallen
voor 100 v. Chr. en na 400 n. Chr. worden door slechts enkele jaarringreeksen
vertegenwoordigd en komen niet optimaal overeen met de bestaande
kalenders; deze delen van NLRom_R dienen daarom verfijnd te worden met
behulp van meer jaarringreeksen.
De dateringen van tien archeologische jaarringreeksen uit Oss-Ussen en
Valkenburg (Noord-Brabant en Zuid-Holland) konden met behulp van deze
kalender bevestigd worden (Appendix A). Onlangs konden twee
archeologische houtmonsters uit Eme (Zutphen, Gelderland) met behulp van
NLRom_R in de zesde eeuw worden gedateerd. De conclusie is dat NLRom_R
bruikbaar is voor dateren, inclusief het laatste segment dat uit slechts enkele
jaarringreeksen bestaat.
De drie archeologische kalenders uit deze periode bevatten voldoende
jaarringseries. NLRom_W1, die uit 14 series bestaat en loopt van 84 v. Chr. tot
50 n. Chr., heeft tot dusver slechts een enkele nieuwe datering opgeleverd.
NLRom_W2 (42 series; 140 v. Chr. - 87 n. Chr.) is zeer bruikbaar gebleken ter
datering van archeologisch materiaal uit Alphen a/d Rijn en Spijkenisse. In
Hoofdstuk 5 werd beargumenteerd dat deze kalender een Oost-Nederlandse of
zelf Duitse herkomst heeft. De nieuwe dateringen echter betreffen alle
materiaal dat, evenals het in NLRom_W2 opgenomen materiaal, afkomstig is
uit vindplaatsen in Zuid-Holland. Een Zuid-Hollandse herkomst van het
materiaal mag daarom niet uitgesloten worden. NLRom_E, die bestaat uit
series uit Cuyk, Gennep en Heeten (92 totaal), loopt van 190 tot 395 n. Chr.
en vertegenwoordigt eiken die in Zuidoost Nederland groeiden. Deze kalender
heeft geleid tot dateringen van archeologisch materiaal uit Bergeijk en Wehl.
Samenvattend zijn ook de archeologische kalenders NLRom_W1, NLRom_W2
en NLRom_E goed bruikbaar als standaard bij het dateren van eikehout uit
Nederlandse vindplaatsen.
. De Merovingische en Karolingische perioden
NLRom_R loopt door tot 563 n. Chr. Slechts een van de drie historische
kalenders, NLHist_1, bestrijkt het hierop volgende interval tussen 564 en 1000
n. Chr. Er zijn met andere woorden meer dendrochronologische gegevens
nodig voor deze periode. Een recent succes in dit opzicht is de datering met
behulp van NLHist_1 van vier eiken palen uit Tiel in de 10e eeuw n. Chr.
. De Late Middeleeuwen
NLHist_1 (427 - 1752; 259 bomen) is de langste van de historische
kalenders en vertegenwoordigt de meeste bomen. Toch heeft deze kalender
voor de periode na 1100 n. Chr. tot weinig nieuwe dateringen geleid. Dit komt
door de herkomst van de jaarringpatronen die in NLHist_1 opgenomen zijn.
De betreffende eiken groeiden in Zuid-Nederland, Oost-België en aangrenzend
Duitsland; globaal hetzelfde gebied als dat waar de kalenders van CentraalDuitsland (Hollstein 1980) en Oost-België (Hoffsummer 1989) betrekking op
hebben. Jaarringpatronen die gedateerd kunnen worden met de Belgische en
Duitse kalenders, kunnen gedateerd worden met NLHist_1, en omgekeerd.
NLHist_2 (1023 - 1666 n. Chr.) is de meest bruikbare van de nieuwe
historische kalenders; met deze kalender, die eiken vertegenwoordigt uit
Centraal- en Noord-Nederland en aangrenzend Duitsland (Westfalen,
Nedersaksen) zijn al vele nieuwe dateringen verricht. Het signaal in de eerste
tweehonderd jaar (1023 - 1300) is zwak. Uit recente dateringen van bouwhout
uit de N.H. Kerk in Oudewater (Utrecht) blijkt echter dat ook dit vroege
interval geschikt is als standaard bij dateringen.
Samenvatting
Samenvattend kan gesteld worden dat de onderzoeksdoeleinden die in de
introductie (Hoofdstuk 1) werden omschreven, zijn bereikt. Nieuwe jaarring kalenders zijn vervaardigd uit eikehout dat afkomstig is uit natuurlijke,
archeologische en historische vindplaatsen in Nederland. Deze kalenders, die
ruim 3000 kalenderjaren bestrijken, zijn goed bruikbaar als standaard voor het
dateren van in Nederland aangetroffen eikehout dat stamt uit de perioden
2258 - 1141 v. Chr. en 325 v. Chr. - 1752 n. Chr.
REFERENCES
Abramowitz, M. and I.A. Stegun, 1965. Handbook of mathematical functions.
National Bureau of Standards, Washington.
Aniol, R.W., 1983. Tree-ring analysis using CATRAS. Dendrochronologia 1,
45-53.
Aniol, R.W., 1987. A new device for computer assisted measurement of tree
ring widths. Dendrochronologia 5, 135-141.
Arts, N., (ed..), 1992. Het Kasteel van Eindhoven. Archaeologie, ecologie en
geschiedenis van een heerlijke woning 1420-1676. Museum Kempenland,
Eindhoven.
Arts, N., (ed..), 1994. Sporen onder de Kempische stad. Archeologie, ecologie en
vroegste geschiedenis van Eindhoven 1225-1500. Museum Kempenland,
Eindhoven.
Baillie, M.G.L., 1982. Tree-Ring Dating and Archaeology. Croom Helm,
London.
Baillie, M.G.L., 1993. Dark Ages and dendrochronology. Emania 11, 5-12.
Baillie, M.G.L., 1994. Dendrochronology raises questions about the nature of
the AD 536 dust-veil event. The Holocene 4(2), 212-217.
Baillie, M.G.L., and J.R. Pilcher, 1973. A simple cross-dating program for
tree-ring research. Tree-Ring Bulletin 33, 7-14.
Baillie, M.G.L., and J.R. Pilcher, 1976. Climate records dug from Irish bogs.
The Geographical Magazine 47(7), 423-426.
Baillie, M.G.L., and J.R. Pilcher, 1982. A British Isles tree-ring chronology.
Tree-Ring Society Newsletter 22.
Baillie, M.G.L., J.R. Pilcher and G.W. Pearson, 1983. Dendrochronology at
Belfast as a Background to High-Precision Calibration. Radiocarbon 25(2),
171-178.
Bauch, J., 1968. Die Problematik der geographischen Reichweite von
Jahrringdatierungen an Beispielen aus der norddeutschen Tiefebene.
Kunstchronik 21, 144-145.
Bauch, J., 1978. Tree-Ring Chronologies for The Netherlands. BAR
International Series 51, 133-137.
References
Bauch, J. and D. Eckstein, 1970. Dendrochronological dating of oak panels of
Dutch 17th century paintings. Studies in Conservation 15, 45-50.
Bauch, J., D. Eckstein and M. Meier-Siem, 1972. Dating the wood of panels by
a dendrochronological analysis of the tree-rings. Nederlands Kunsthistorisch
Jaarboek 23, 485-496.
Becker, B., 1993. An 11,000 Year German Oak and Pine Chronology for
Radiocarbon Calibration. Radiocarbon 35(1), 201-213.
Boekwijt, H.W., 1993. Van woning tot pastorie, ’s-Hertogenbosch.
Box, G.E.P and G.M. Jenkins, 1971. Time Series Analysis: Forecasting and
Control. Holden-Day, San Francisco.
Briffa, K.R., and P.D. Jones, 1990. Basic Chronology Statistics and
Assessment. In: E.R. Cook and L.A. Kairiukstis (eds.), Methods of
Dendrochronology. Kluwer Academic Publishers, Dordrecht, 137-152.
Briffa, K.R., P.D. Jones and F.H. Schweingruber, 1988. Summer temperature
patterns over Europe: a reconstruction from 1750 A.D. based on maximum
latewood density indices of conifers. Quaternary Research 30, 36-52.
Briffa, K.R., P.D. Jones, T.M.L. Wigley, J.R. Pilcher and M.G.L. Baillie, 1986.
Climate reconstruction from tree rings: Part 2, spatial reconstruction of
summer mean sea-level pressure patterns over Great Brittain. Journal of
Climatology 6, 1-15.
Briffa, K.R. and F.H. Schweingruber, 1992. Recent dendroclimatic evidence of
northern and central European summer temperatures. In: R.S. Bradley and
P.D. Jones (eds.), Climate since A.D. 1500. Routledge Press, London, 366-392.
Brongers, J.A., 1973. Dendrochronological investigation of recent oak (Quercus
sp.) in The Netherlands. Berichten ROB 23, 456-459.
Bult, E.J., D.P. Hallewas and P. van Rijn, 1989. Het Praetorium van
Valkenburg. Natuur en Techniek 57, 194-205.
Cook, E.R., 1985. A Time-Series Analysis Approach to Tree-Ring
Standardization. Ph.D. Dissertation Dept. of Geosciences, Univ. of Arizona.
Cook, E.R., 1990, A Conceptual Linear Aggregate Model for Tree Rings. In:
E.R. Cook and L.A. Kairiukstis (eds.), Methods of Dendrochronology. Kluwer
Academic Publishers, Dordrecht, 98-104.
Cook, E.R. and K.R. Briffa, 1990. A Comparison of Some Tree-Ring
Standardization Methods. In: E.R. Cook and L.A. Kairiukstis (eds.), Methods of
Dendrochronology. Kluwer Academic Publisher, Dordrecht, 153-162.
Cook, E.R., K.R. Briffa, S. Shiyatov and V. Mazepa, 1990. Tree-Ring
Standardization and Growth-Trend Estimation. In: E.R. Cook and L.A.
Kairiukstis (eds.), Methods of Dendrochronology. Kluwer Academic Publisher,
Dordrecht, 104-123.
Cook, E.R. and L.A. Kairiukstis, 1990. Methods of Dendrochronology. Kluwer
Academic Publisher, Dordrecht.
Dalrymple, G.B., 1991. The Age of the Earth. Stanfford.
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Davis, O.K., J. Jirikowic and R.M. Kalin, 1992. Radiocarbon Record of Solar
Variability and Holocene Climate Change in Coastal Southern California. In:
K.T. Redmond (ed.), Proceedings of the Eighth Annual Pacific Climate
(PACLIM) Workshop, March 10-13, 1991, California Department of Water
Resources, Interagency Ecological Studies Program Technical Report 31.
Delorme, A., 1972. Dendrochronologische Untersuchungen an Eichen des
Südlichen Weser- und Leineberglandes. Ph.D. Dissertation Univ. of Göttingen.
Delorme, A., 1974. Mooreichen als Forschungsobjecte. Holz-Zentralblatt
100(58), 901-902.
Delorme, A., 1976. Mooreichen - Bausteine für eine absolute Chronologie des
Postglazials. Neue Ausgrabungen und Forschungen in Niedersachsen 10, 225-237.
Delorme, A., H.-H. Leuschner, H.C. Höffle and J. Tüxen, 1981. Über die
Anwendung der Dendrochronologie in der Moorforschung am Beispiel
subfossiler Eichenstämme aus Niedersächsischen Mooren. Eiszeitalter und
Gegenwart 31, 135-158.
Douglass, A.E., 1909. Weather cycles in the growth of big trees. Monthly
Weather Review 37(5), 225-237.
Douglass, A.E., 1914. A method for estimating rainfall by the growth of trees.
In: E. Huntington, The Climatic Factor as Illustrated in Arid America. Carnegie
Institute of Washington, Publication No. 192, 101-121.
Douglass, A.E., 1919. Climatic Cycles and Tree Growth I: A Study of the Annual
Rings of Trees in Relation to Climate and Solar Activity. Carnegie Institute of
Washington, Publication No. 289.
Douglass, A.E., 1928. Climate Cycles and Tree Growth II: A Study of the Annual
Rings of Trees in Relation to Climate and Solar Activity. Carnegie Institute of
Washington, Publication No. 289.
Drunen, A. van, 1993. Bossche gebouwen dendrochronologisch gedateerd.
Bulletin KNOB 92, 57-63.
Drunen, A. van, and R. Glaudemans, 1995. Bouwen en wonen in de schaduw
van de Sint Jan. Bouwhistorische en Archeologische Kroniek van ’s-Hertogenbosch
2 (in print).
Eckstein, D., J. Bauch and W. Liese, 1970. Afbau einer Jahrringchronologie
für Eichenholz für die Datierung historischer Bauten in Norddeutschland.
Holz-Zentralblatt 96, 674-676.
Eckstein, D., J.A. Brongers and J. Bauch, 1975. Tree-ring research in The
Netherlands. Tree-Ring Bulletin 35, 1-13.
Erdrich, M. and A.D. Verlinde, 1995. Eine Germanische Siedlung der
Jüngeren Kaiserzeit mit umwerter Anlage, Eisenindustrie und
Keramikproduktion in Heeten, Prov. Overijssel, Germania (in print).
Es, W.A. van, H. Sarfatij and P.J. Woltering, 1988. Archeologie in Nederland de rijkdom van het bodemarchief. Meulenhof, Amsterdam.
Fisher, R.A., 1935. The Design of Experiments. Annals of Science 1, 115-137.
Fritts, H.C., 1976. Tree Rings and Climate. Academic Press, London.
References
Fürst, O, 1963. Vergleichende Untersuchungen über räumliche und zeitliche
Unterschiede interannueller Jahrringbreitenschwankungen und ihre
klimatologische Auswertung. Flora 153, 469-508.
Fürst, O., 1978. Geographical and temporal differences in the fluctuations of
ring width in Central Europe: climatic implications. In: J.M. Fletcher (ed.),
Dendrochronology in Europe. BAR International Series 51, 65-76.
Goudswaard, B., 1995. A late-Roman bridge in the Meuse at Cuijk, the
Netherlands. Archäologisches Korrespondenzblatt 25(2), 233-241.
Groenendijk, H.A., 1994. Appingedam steeds ouder. Stad en Lande 3, 8-12.
Haslinghuis, E.J., 1986. Bouwkundige termen: verklarend woordenboek der westerse
architectuur-geschiedenis. Bohn Stafleu Van Loghum, Houten/Zaventem.
Heidinga, H.A., 1987. Medieval Settlement and Economy North of the Lower
Rhine. Ph.D. Dissertation Univ. of Amsterdam, Van Gorcum & Comp., Assen.
Heidinga, H.A. and G.A.M. Offenberg, 1992. Op zoek naar de vijfde eeuw - de
Franken tussen Rijn en Maas. De Bataafse Leeuw, Amsterdam.
Hiddink, H.A. 1994. Romeinse waterputten uit de nadagen van het heiligdom.
In: N. Roymans and T. Derks (eds.), De tempel van Empel - een Herculesheiligdom in het woongebied van de Bataven. Stichting Brabantse Regionale
Geschiedsbeoefening, ’s-Hertogenbosch, 58-71.
Hillam, J. and I. Tyers, 1995. Reliability and repeatability in
dendrochronological analysis: tests using the Fletcher archive of panel-painting
data. Archaeometry (in print).
Hoffsummer, P., 1989, L’évolution des toits à deux versants dans le bassin mosan:
l’apport de la dendrochronologie (XIe-XIXe siècle). Ph.D. Dissertation Univ. of
Liège.
Hollstein, E., 1965. Jahrringchronologische Datierung von Eichenhölzern ohne
Waldkante. Bonner Jahrbucher 165, 12-27.
Hollstein, E., 1980. Mitteleuropäische Eichenchronologie. Verlag Phillipp von
Zabern, Mainz am Rhein.
Holmes, R.L. 1983. Computer-Assisted Quality Control in Tree-Ring Dating
and Measurement. Tree-Ring Bulletin 44, 69-75.
Huber, B. 1941. Aufbau einer mitteleuropäischen Jahrring-Chronologie.
Mitteilungen der Hermann-Göring Akademie der Deutsche Forstwissenschaft 3,
137-142.
Huber, B., and W. Holdheide, 1942. Jahrringchronologische Untersuchungen
an Hölzern der Bronzezeitlichen Wasserburg Buchau am Federsee. Berichte der
Deutschen Botanischen Gesellschaft 60, 261-283.
Huber, B., W. von Jazewitsch, A. John and W. Wellenhofer, 1949.
Jahrringchronologie der Spessarteichen. Forstwissenschaftliches Zentralblatt 68,
706-715.
Hughes, M.K., 1987. Dendroclimatology of Pinus sylvestris L. in the Scottish
highlands. In: R.G.W. Ward (ed.), Applications of Tree-Ring Studies: Current
Research in Dendrochronology and Related Subjects, BAR International Series
333, 91-106.
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Hughes, M.K., B. Gray, J.R. Pilcher, M.G.L. Baillie and P. Leggett, 1978.
Climatic signals in British Isles tree-ring chronologies. Nature 272, 605-606.
Hughes, M.K., P.M. Kelly, J.R. Pilcher and V.C. La Marche (eds.), 1982.
Climate from Tree Rings. Cambridge University Press, Cambridge.
Hughes, M.K., F.H. Schweingruber, D. Cartwright and P.M. Kelly, 1984.
July-August temperature at Edinburgh between 1721 and 1975 from tree-ring
density and width data. Nature 308, 341-344.
Jansma, E., 1985. Dendrochronologisch onderzoek aan houtresten uit het
Romeinse fort Velsen 1 (15-30 AD). M.A. thesis Univ. of Amsterdam.
Jansma, E., 1987. Het Wilde Woud zonder Genade. Westerheem 36, 239-240.
Jansma, E., 1993. De drie aanlegfasen van het park rond het Huis van
Warmond (Zuid-Holland). Int. Rep. RING, Amersfoort.
Jansma, E., 1994. Dendrochronologisch onderzoek van middeleeuws
archeologisch eikehout uit Eindhoven. In: N. Arts, (ed..), Het Kasteel van
Eindhoven. Archaeologie, ecologie en geschiedenis van een heerlijke woning 14201676. Museum Kempenland, Eindhoven, 192-199.
Jansma, E., 1995. Groningse chronologieën; datering en herkomstbepaling
van middeleeuws bouwhout door middel van jaarringonderzoek. In:
K. Helfrich, J.F. Benders and W.A. Casparie (eds.), Handzaam hout uit
Groninger grond, Stichting Monument & Materiaal, Groningen, 150-155.
Jansma, E. and W.A. Casparie, 1993. Een Neolithische landnam ontrafeld
door dendrochronologisch onderzoek van de veenweg van Nieuw-Dordrecht
(Dr). Paleo-Aktueel 4, 33-36.
Jansma, E. and H. Kars, 1995. Het dendrochronologisch onderzoek naar het
‘kasteel van Amstel’. Spiegel Historiael 30(1), 9-11.
Johnson, N.L. and S. Kotz, 1969. Distributions in Statistics: Continuous
Multivariate Distributions. John Wiley & Sons, New York.
Kaennel, M., 1992. Reading trees: an approach to dendrochronology and a
provisional English French glossary. M.A. thesis Univ. de Nancy II.
Kapteyn, J.C., 1914. Tree Growth and Meteorological Factors. Recueil des
Travaux Botaniques Neerlandaises 11(1), 71-93.
Kelly, P.M., M.A.R. Munro, M.K. Hughes and C.M. Goodess, 1989.
Climate and Signature Years in West European Oaks. Nature 340, 57-60.
Kortekaas, G.L.G.A., 1989. Opgraving De Hunze. Interne notitie Stichting
Monument & Materiaal, Groningen.
Kortekaas, G.L.G.A., 1991. Opgraving Wolters-Noordhoff-complex te
Groningen (Gr.). Paleo-Aktueel 2, 130-133.
Leenders, K.A.H.W., 1991. Het Schijndelse cultuurlandschap: een detailstudie.
Den Haag.
Leuschner, H.-H., 1990. Subfossil Trees. In: O. Eggertsson (ed.), Tree Rings
and Environment, Proceedings of the International Dendrochronological
Symposium, Ystad, South Sweden, 3-9 September 1990. LUNDQUA Report
34, 193-197.
References
Leuschner, H.-H., and A. Delorme, 1988. Tree-ring work in Göttingen:
absolute oak chronologies back to 6255 BC. In: T. Hackens, A.V. Munaut and
C. Till (eds.), Wood and Archaeology. PACT 22, 123-132.
Leuschner, H.-H., A. Delorme and H.C. Höffle, 1987. Dendrochronological
study of oak trunks found in bogs of Northwest Germany. In: G.C. Jacoby Jr.
and J.W. Hornbeck (eds.), Proceedings of the International Symposium on
Ecological Aspects of Tree-Ring Analysis, US Dept. of Energy, CONF-8608144,
298-318.
Leuschner, H.-H., A. Delorme, J. Tüxen and H.C. Höffle, 1985.
Eichenstammlagen in Mooren und Talauen und die Klima-verschlechterung im
Subboreal. Flora 177, 283-295.
Michels, J.W., 1973. Dating methods in archaeology. Seminar Press, New York.
Morel, J.-M.A.W., 1988. De Vroeg-Romeinse versterking te Velsen 1. Ph.D.
Dissertation Univ. of Amsterdam.
Munaut, A.V., 1966. Réchêrches dendrochronologiques sur Pinus sylvestris 2:
Première application des méthodes dendrochronologiques à l’étude de pins
silvestres sub-fossiles (Terneuzen, Pays-Bas). Agricultura 14, 361-389.
Munaut, A.V., and W.A. Casparie, 1971. Étude dendrochronologique des
Pinus sylvestris L. subfossiles provenant de la tourbière d’Emmen (Drenthe,
Pays-Bas). Review of Palaeobotany and Palynology 11, 201-226
Munro, M.A.R., 1984. An improved alogrithm for cross-dating tree-ring series.
Tree-Ring Bulletin 44, 17-27.
O’Reilly, W., 1984. Rock and Mineral Magnetism. Methuen.
Pilcher, J.R., 1973. Tree-Ring Research in Ireland. Tree-Ring Bulletin 33, 1-5.
Pilcher, J.R., M.G.L. Baillie, B. Schmidt and B. Becker, 1984. A 7,272 year
tree-ring chronology for western Europe. Nature 312, 150-152.
Pratchett, T, 1989. Truckers. Doubleday.
Rijn, P. van, 1987. Dendrochronologisch onderzoek van de tweede Romeinse
weg. In: E.J. Bult and D.P. Hallewas (eds.), Graven in Valkenburg II. Delft,
111-114.
Sanden, W.A.B. van der, 1987. Dendrochronologie en het archeologisch
onderzoek op de zuidelijke zandgronden. In: W.A.B. van der Sanden and P.W.
van den Broeke (eds.), Getekend zand - tien jaar archeologisch onderzoek in OssUssen. Stichting Brabants Heem, Waalre, 45-52.
Sarfatij, H., 1991. Dordrecht - Drievriendenhof. Regionaal-historisch tijdschrift
HOLLAND 23(6), 353.
Sarfatij, H., 1994. Het huis Scharlaken te Dordrecht: de oudste Lakenhal van
de stad, vervolgend woonhuis en WAAG (13e - 16e eeuw). Archeologie en
geschiedenis van een opmerkelijk huis. Bulletin KNOB 93, 41-52.
Schmidt, B, and W. Gruhle, 1988. Klima, Radiokohlenstoffgehalt und
Dendrochronologie, Naturwissenschaftliche Rundschau 41(5), 177-182.
Schweingruber, F.H., O.U. Bräker and E. Schär, 1979. Dendroclimatic studies
on conifers from central Europe and Great Brittain. Boreas 8(4), 427-452.
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Schweingruber, F.H., O.U. Bräker and E. Schär, 1987. Temperature
information from a European dendroclimatological sampling network.
Dendrochronologia 5, 9-33.
Schweingruber, F.H., K.R. Briffa and P.D. Jones, 1991. Yearly maps of
summer temperatures in western Europe from A.D. 1750 to 1975 and
western North America from 1600 to 1982: Results of a radiodensitometric
study on tree-rings. Vegetatio 92, 5-71.
Schweingruber, F.H., H.C. Fritts, O.U. Bräker, L.G. Drew and E. Schär,
1978. The X-ray technique as applied to dendroclimatology. Tree-Ring
Bulletin 38, 61-91.
Thomas, D.H., 1976. Figuring Anthropology: First Principles of Probability and
Statistics. Holt, Rhinehart and Winston, New York.
Trierum, M.C. van, 1992. Nederzettingen uit de IJzertijd en de Romeinse
Tijd op Voorne-Putten, IJsselmonde en in een deel van de Hoekse Waard. In:
A.B. Döbken (ed.), Bijdragen aan de bewoningsgeschiedenis van het
Maasmondgebied, Boor Balans 2, 15-102.
Tulippe, O., 1942. L’Homme et la forêt temperée en Belgique. Collection
Traveaux du Séminaire de Géographie de l’Université de Liège 77.
Verwers, W.J.H., 1994. Archeologische Kroniek van Noord-Brabant. Brabants
Heem 46, 127 - 152.
Vink, E., 1990. Blocken en balcoenen. Bosch’ hout in de vijftiende en
zestiende eeuw. Brabants Heem 42, 103-116.
Vink, E., 1993. Bosch bouwhout onderweg. Bulletin KNOB 92, 129-140.
Vries, D.J. de, 1983. Houten en ijzeren bouwfragmenten van de Voorst. In:
Het Kasteel Voorst, macht en val van een Overijsselse burcht circa 1280 1362 naar aanleiding van een opgraving. Vereeniging tot Beoefening van
Overijsselsch Regt en Geschiedenis, Werken 36.
Vries, D.J. de, 1987. Monumenten dendrochronologisch gedateerd. Bulletin
KNOB 86(2), 85-89.
Vries, D.J. de, 1988. Monumenten dendrochronologisch gedateerd (2).
Bulletin KNOB 87(2), 71-73.
Vries, D.J. de, 1990. Momumenten dendrochronologisch gedateerd (3).
Bulletin KNOB 89(5), 19-16.
Vries, D.J. de, 1992. Monumenten dendrochronologisch gedateerd (4).
Bulletin KNOB 91(1), 27-35.
Vries, D.J. de, 1994. Bouwen in de Late Middeleeuwen; stedelijke architectuur in
het voormalige Over- en Nedersticht. Matrijs, Utrecht.
Wazny, T., 1990. Afbau und Anwendung der Dendrochronologie für Eichenholz in
Polen. Dissertation Univ. of Hamburg.
Wigley, T.M.L., K.R. Briffa and P.D. Jones, 1984. On the Average Value of
Correlated Time Series, with Applications in Dendroclimatology and
Hydrometeorology. Journal of Climate and Applied Meteorology 23, 201-213.
References
Wolff, W.J., 1989, De internationale betekenis van de Nederlandse natuur - een
verkenning. Ministerie van Landbouw, Natuurbeheer en Visserij, Rijksinstituut
voor Natuurbeheer, SDU Uitgeverij, ’s-Gravenhage (NL).
Zielinski, G.A., M.S. Germani, G. Larsen, M.G.L. Baillie, S. Whitlow, M.S.
Twickler and K. Taylor, 1995. Evidence of the Eldgjá (Iceland) eruption in the
GISP2 Greenland ice core: relationship to eruption processes and climatic
conditions in the tenth century. The Holocene 5(2), 129-140.
APPENDIX
A
ABSOLUTELY DATED TREE-RING SERIES
OF OAK FROM ANTHROPOGENIC AND NATURAL SITES
IN THE NETHERLANDS
Appendix A
The dates listed below were established between 1985 and 1995 at the Dutch
Centre for Dendrochronology (RING Foundation/ROB) and the laboratories it
replaced (IPP; ROB). About 50 series that were dated during the earliest stages
of dendrochronological research (1985/1986) could not be retrieved. The
research was performed for many organizations, including the ‘Albert Egges van
Giffen’ Institute for Pre- and Protohistoric Archaeology (IPP/Univ. of
Amsterdam), the State Service for the Preservation of Monuments and Historic
Buildings (RDMZ, Zeist), the State Service for Archaeological Research (ROB,
Amersfoort), the Bouwhistorische Dienst ’s-Hertogenbosch, and universities and
municipal archaeological and historical organizations in the Netherlands.
The statistics that accompany the dates are not always identical to the
information made available to the organization(s) and persons for which the
research was conducted. As part of the EC-programme Temperature Change over
Northern Eurasia during the last 2500 Years (Contract No. CV5V CT94 0500) all
dates were verified in 1994 by the author and her co-workers P. van Rijn and E.
Hanraets (RING). During this process chronologies that have become available
in recent years were used as a reference; this has resulted in new values for St
and PV. When these are listed instead of the original values, it is because they
provide a better argument for a dendrochronological date.
For each archaeological/historical object and natural site the following details
are given:
First line
location; type of object; institute for which the research was
conducted; references;
Object
sample description (e.g., find number, construction element,
assembly-mark); the terms used for construction elements in
buildings were derived from Haslinghuis (1986). Ships’
elements have been named in consultation with Dr. J. H. G.
Gawronski (IPP);
Filename
the name of the computer file at RING;
Dendro-code the dendrochronological code of the sample; this code is based
on the country (first two digits), ecological growth-region
(third digit, after Wolff et al. 1989), site number (fourth and
fifth digits) and sample number (last three digits);
No.
the number of rings that were measured on the sample;
Date
the calender date of the last, outer, ring that was measured; in
most cases this date is earlier than the actual felling date of the
tree, as a result of the fact that rings are missing on the
outside of the sample.
Reference
the code of the chronology used to date the sample (see
Appendix B for a description of the chronologies);
St
the Student’s t-value that accompanies the match with the
reference chronology;
PV
the Coefficient of Parallel Variation that accompanies the match
with the reference chronology;
NL
the name of the Dutch chronology into which the series has
been incorporated;
Remarks
further details.
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix A
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix A
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix A
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix A
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix A
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix A
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix A
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix A
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix A
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix A
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix A
APPENDIX
B
TREE-RING CHRONOLOGIES OF OAK USED FOR DATING
Appendix B
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix B
APPENDIX
C
AVERAGE GROWTH-INDEX VALUES OF SOME DUTCH
CHRONOLOGIES (2258 - 1141 BC AND 325 BC - AD 563)
Appendix C
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix C
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19
Appendix C
Esther Jansma RemembeRINGs | Nederlandse Archeologische Rapporten 19