Purdue University

Purdue e-Pubs
LARS Symposia Laboratory for Applications of Remote Sensing

1-1-1976

The Tasselled Cap -- A Graphic Description of the

Spectral-Temporal Development of Agricultural
Crops as Seen by LANDSAT
R. J. Kauth

G. S. Thomas

Follow this and additional works at: http://docs.lib.purdue.edu/lars_symp

Part of the Agriculture Commons

Kauth, R. J. and Thomas, G. S., "The Tasselled Cap -- A Graphic Description of the Spectral-Temporal Development of Agricultural
Crops as Seen by LANDSAT" (1976). LARS Symposia. Paper 159.
http://docs.lib.purdue.edu/lars_symp/159

This document has been made available through Purdue e-Pubs, a service of the Purdue University Libraries. Please contact epubs@purdue.edu for
additional information.
 Reprinted from

Symposium on
Machine Processing of
Remotely Sensed Data

June 29 - July 1, 1976

The Laboratory for Applications of

Remote Sensing

Purdue University
West Lafayette
Indiana

IEEE Catalog No.

76CH1103-1 MPRSD

Copyright 1976 IEEE

The Institute of Electrical and Electronics Engineers, Inc.

Copyright 2004 IEEE. This material is provided with permission of the IEEE. Such
permission of the IEEE does not in any way imply IEEE endorsement of any of the
products or services of the Purdue Research Foundation/University. Internal or personal
use of this material is permitted. However, permission to reprint/republish this material
for advertising or promotional purposes or for creating new collective works for resale or
redistribution must be obtained from the IEEE by writing to pubs-permissions@ieee.org.

By choosing to view this document, you agree to all provisions of the copyright laws
protecting it.
 THE TASSELLED CAP -- A GRAPHIC DESCRIPTION OF THE SPECTRAL-TEMPORAL
DEVELOPMENT OF AGRICULTURAL CROPS AS SEEN BY LANDSAT*

R. J. Kauth and G. S. Thomas

Environmental Research Institute of Michigan

Ann Arbor, Michigan

I. ABSTRACT farious external effects influence the pattern of

the crop in question as well as the pattern of all
The time trajectories of agricultural data other crops. The number of variables is very large.
points as seen in LANDSAT signal space form a pattern In this environment how do we proceed in the design
suggestive of a tasselled woolly cap. Using this of a processor?
easily visualized three dimensional construct most
of the important phenomena of crop development and The first step is to find a variety of ways to
observation variables are pointed out, named, dis- display the data. Given the displays, one will
cussed and measured. The important crop phenomena notice the structure of the data, will name the data
described are the distribution of signals from bare structures, will create some visual model of the
soil, the processes of green development, of yellow data structure. Such a model may suggest a physical
development, of shadowing and of harvesting. The interpretation or a quantitative test. One then sub-
important external variables include view angle, sun jects the description to quantitative tests, result-
angle, atmospheric haze, and atmospheric water vapor. ing in gradual elaboration and refinement of the
description, or, in some cases, res~lting in its
The development of a quantitative picture of destruction.
the tasselled cap depends upon crop and atmosphere
effects modeling and upon empirical observation of Let us proceed then to look at the structure of
data signals. Quantitative data from a composite LANDSAT data of agricultural crops, leading to the
of these sources is given. The tasselled cap struc- visual model of the tasselled cap.
ture is illustrated with cluster plots and diagrams.
III. THE TASSELLED CAP
The tasselled cap is a fertile source of ideas
for processing techniques. Examples discussed Figure l(a) shows a two channel scatter diagram
include of LANDSAT data in an agricultural scene in Fayette
Co., Illinois. The data has been compressed by
a) A linear preprocessing transformation which
unsupervised spectral clustering of the data pOints
isolates green development, yellow develop-
in all 4 LANDSAT MSS channels. The ellipses shown
ment and soil brightness and allows the
are the unit contour ellipses of the normal density
reduction of the dimension of the feature function describing each cluster. The channels shown
space. are CH 2 and CH 3.
b) The use of specific measurable pattern ele-
Notice in Figure l(a) the definite boundary
ments of the tasselled cap structure to region near the diagonal of the two channel presen-
estimate and correct atmospheric haze and tation. All of the agricultural data lies to the
moisture effects. left of this boundary. To the right of the boundary
there is no data. The region to the left shows a
II. INTRODUCTION definite triangle like shape, with two vertices on
the diagonal and one near the CH 3 axis.
The problem is to be able to identify some
agricultural crop systematically on a large scale, Figure l(b) shows a similar cluster plot of
based on remotely sensed data acquired at a number CH 1 vs. CH 2. Here, all of the data lies near a
of times: The pattern of the particular crop must diagonal of the space again. Thus we can infer that
be distinguished from the patterns of multiple other the triangle shaped region of Figure l(a) is shown
@rops which mayor may not always be present. Multi- edge on in Figure l(b). The three-dimensional shape

* The effort described herein was supported by the Earth Observations Division of the NASA/Johnson Space
Center under Contract NAS9-l4l23.

4B-41
 of the data structure is that of a flattened tri- The largest principal component of soil reflectance
angular shape having little thickness. Figure l(c) is nearly parallel to the diagonal. The square
.shows a cluster plot of CH 3 vs. CH 4. Again the root of the eigenvalue associated with the first
data lies closely along a diagonal. Viewing only component is about 35 units, (i.e., one standard
Figures l(a) and l(c), one would conclude that, seen deviation of the data projected onto the first prin-
in the 3 space of Channels 2, 3 and 4 the three- cipal component is about 35). The second principal
dimensional shape of the data structure is a flat- component, normal to the first, has a standard devi-
tened triangular shape. One then can conclude that ation of about 5 units the third of about 3 units
the data structure forms a flattened triangular and the fourth of about 1 1/2 units. The unit con-
shape in 4 dimensions, and that is correct. tour ellipsoid describing the distribution of sotls
forms a four dimensional flattened cigar shape,
Figures l(d), l(e) and l(f) show the same data about seven times as long as it is wide, about twice
in channel pairs 1 vs. 3, 2 vs. 4 and 1 vs. 4. If as wide as it is thick, and twice as thick as it is
one assumes that CH 1 is highly correlated to CH 2 thin (what is the name for distance in the 4th direc-
(as it seems to be, based on Figure l(b and that tion?). Hence, for some applications we would be
CH 4 is highly correlated to CH 3 (as it seems to justified in describing the data from soil points
be, based on Figure l(c, then these last 3 figures as the "line of soils", ignoring all but the major
offer no particular surprises; they are in a manner component. In other cases we might speak of the
of speaking first and second cousins of Figure l(a). "plane of soils", referring to the first and second
(The fact of the high correlation of CH 1 with CH 2 component.
and of CH 4 with CH 3 has sometimes stimulated the
comment that LANDSAT MSS is essentially a two channel Returning now to Figure 2, we notice again that
system; that no information would be lost by throw- after the initial development stages the two crop
ing away CH 1 and CH 4. On the contrary, there is canopy trajectories join and fall back towards the
significant information of several types contained soil line. What cannot be seen in this figure is that
in the 4 channels, as we shall see as this dis- the line of falling back is not in the same plane
cussion develops.) (in the 4 space of LANDSAT data) as the two develop-
ment lines up to the point of joining. The crop is
What is the physical reason for the data to yellowing, and yellow colored things lie in adif-
lie in this flattened triangular structure? Figure ferent direction away from the soil line than do
2 shows a model calculation of the reflectance of a green colored things.
crop canopy at two wavelengths, .65 nm and .75 nm,
corresponding to the centers of CH 2 and CH 3. The We now have sufficient information to create
calculations were made for two soil samples, one the basic image of the tasselled cap, shown in
dark, the other light, through the life of the crop. Figure 4.
Notably, the triangular shape is outlined by the two
crop life development lines. After the crop canopy The basic tasselled cap shown in Figure 4 is
cover the soil completely the two canopies look created by combining soil reflectance and green
identical. Figure 2 is extracted from Reference 1. stuff and then adding yellow stuff. We say that
The canopy model u~ed was developed by G. Suits 2 . the crop starts growing on the plane of soils. As
Roughly what seems to be occurring is that the crop it grows it progresses outward, roughly normal to
starts its growth on the line of soils. As it grows the plane of soils, on a curving trajectory towards
the composite reflectance of soil and crop increases the region of green stuff. Next the trajectories
in CH 3 because of the presence of cellulose in the fold over and converge on the region of yellow stuff.
plant. The composite reflectance of CH 2 decreases Finally the crop progresses back to the soil from
because of the chlorophyll in the plants is highly whence it came (dust from dust?) by any of several
absorbing. Hence the radiance typical of green routes, depending on the crop and the harvesting
plants is located to the left, at the tip of the practices.
triangle.
Initially we spoke of a flattened triangle,
Figure 2 attempts to span the. range of soil now we are likening the data structure to a tasselled
conditions by the terms "light" and "dark". Is this cap. To fit both of these images the yellow point
all there is to soils as seen in LANDSAT data? must be quite close to the side of the cap, and in-
Condit 3 ,4 has measured the spectral reflectance of deed that is true. For wheat the yellow is also
soil samples from allover the United States, and accompanied by shadowing so that the yellow point is
analyzed them in terms of their principal spectral found near the dark end of the plane of soils.
components. We have used Condit's data to calculate
the soil distribution that would be seen by the The "front" of the cap looks down toward the
LANDSAT MSS spectral filters. Table 1 shows the origin of all data otherwise called THE ORIGIN. On
soil reflectance mean vector and principal compon- the front of the cap is the badge of trees. Why the
ents in LANDSAT data. We will summarize those reflectance of trees is located just here will be
results in the following discussion. explained a little further on.

Figure 3 gives an idea of the distribution of Effects of Shadow

soil reflectances projected into the 4'dimensions
spanned by the 4 LANDSAT MSS channels. That space As the crop canopy develops away from the soil
has a "diagonal", i.e., a line along wh.ich the nor- the average reflectance becomes more green, but at
malized reflectance of all channels is equal. The the same time shadows develop. Initially, much of
mean reflectance of soils lies near that diagonal. this shadow will appear on the soil portions of the

4I3-42
..

composite canopy. Thus the reflectance of a crop R is a unitary matrix, i.e., the columns
planted on bright soil will initially migrate mainly of R are unit vectors Rl , R , R3 and R ,
in the direction of the origin. 2 4
which are all orthogonal to eacn other.
SuperscriptT indicates the transpose.
A crop which is planted on dark soil will not Thus the application of the transformation
show this beha,~or significantly. After all there to the data x results in a pure rotation
is little difference between the radiance of dark plus a pure translation.
soil and the radiance of shadowed dark soil.
The components of R are chosen in the following
Once maximum shadowing on the soil has been way:
reached the reflectance is more strongly influenced
by the addition of green elements to the canopy. is cho~en to point along the major axis of
Thus the trajectory of reflectance values sweeps soils, in the LANDSAT data. A particular
away from the plane of soils. Initially many of the sample of LANDSAT data was chosen to derive
green elements that are added are shadowed green Rl , namely Fayette County, Illinois, June
elements. Hence the total reflectance remains low 1973. Visual inspection of Figure lea) was
until most of the ground is covered. used to pick out 12 soil line clusters.
The best fit line to those 12 clusters was
In the next stage the canopy loses most of its chosen as the direction of Rl Rl is
shadows, reaching a state of full green development.
called the soil brightness unit vector.
Whether a crop actually reaches this stage depends
The projection of a data point onto Rl is
upon the planting density and upon the way its
a feature called brightness.
leaves form together to make a canopy.
is chosen to point orthogonal to Rl and
This curving trajectory has been documented
toward a green cluster in the same data
by F. Johnson5 in Fayette County corn field data,
set. Visual inspection of Figure 1 was
and also has been shown in the results of a detailed
used to identify the cluster. RZ was
modelling exercise being presented at this sym-
generated using the Gram-Schmidt ortho-
posium6 Interestingly, Johnson has found that
corn planted in East-West rows does not show this gonalization procedure. R2 is the green
stuff unit vector. The projection of a
behavior significantly, whilst corn planted in
data point onto R2 is a feature called
North-South rows does show a very strong shadow
"green stuff".
effect'. The reason is clear. At the time of the
LANDSAT overpass the Sun's rays are coming mainly
from the east. Sunlight falls down the East-West is chosen orthogonal to both R1 and R2 and
rows and shadows fallon the sides of other corn points toward a yellow stuff point. There
plants rather than in the open rows. was no yellow stuff in the Fayette segment,
hence an approximate spectrum of yellow
corn was used to simulate or predict the
Now we can see why trees occupy the place they
yellow point in the Fayette data. Again
do in reflectance space. Trees are green canopies the Gram-Schmidt procedure was used to
structured so as to create a good deal of shadow.
derive the yellow stuff unit vector.

IV. A FIXED LINEAR TRANSFORMATION is chosen orthogonal to RI , R2 and R3 The

projection of a data point onto R4 is a .
It is difficult to look at LANDSAT data and feature called "non-such".
see all of the features so far described. After
all, this is a 4-dimensional space we are looking The values for Rl , R2 , R3 and R4 are, to the
at, and it is hard to be sure we are seeing every- third decimal place,
thing. Therefore we have developed some transfor-
mations of the data which assist us to see it
better.* The only one of these we will discuss at -.290 )
.433 )
this point is a fixed affine transformation, R .. 632 R" -.562 .
1 ( .586 2 ( .600
u"RTx+r (1) .264 .491

where
x is the LANDSAT MSS signal vector -.829 ) .223 )
expressed in counts R. .522 .012
3 R4 -
( -.039 ( -.543
u is the transformed vector, also .194 .810
expressed in counts
r is an offset vector, introduced to avoid The offset vector is r-bitrary. All compon-
negative values in the transformed data ents equal to 32 seema to work well.

* The transformations we have developed have depended in part on the work of F. Johnson. 5

48-43
 The fixed linear transformation has several numerous winter wheat fields. The region at the
potential uses. center is rangeland. Figure 7 is the soil bright-
ness image of the same data. Figure 8(b) shows non-
a) Simply by projecting the clustered data in such in the 4th biophase and is reasonably typical
terms of the features of Equation 1 we can see the of non-such and yellow stuff in all of the biophases,
data structure easily. We can also examine it to i.e., mainly noise with almost no discernible struc-
determine to what extent it actually behaves accord- ture. Figure 8(a) is yellow stuff in the 4th bio-
ing to our imaginary picture. phase. Although the dynamic range of the data is I
only about 10 counts, which is comparable to Figure
b) Potentially there is significantly less 8(b), the strong spatial structure is evident.
information in some of the transformed channels
than in others, whereas each of the original chan- Returning to Figure 6(a), we note the range-
nels is about equally information carrying. Thus land is somewhat green, but the fields are not green
one might be able to ignore certain of the trans- at all. The roads show up, if at all, as sljghtly
formed channels and this could lead to cost reduc- green, due to the grass on the roadside. In Figure
tion in processing. 6(b), the second biophase, the fields show up
strongly green, while the rangeland is still only
c) The transformation of the data allows cer- somewhat green. In Figure 6(c), the third biophase,
tain diagnostic features to be extracted which are both rangeland and winter wheat are green; one can
symptomatic of external effects, such as haze, H20 imagine that the rangeland has caught up with the
vapor, illumination angle and viewing angle. wheat. Finally in Figure 6(d), the 4th biophase,
the wheat is again not green.
In order to picture the data resulting from
the fixed linear transformation we show Figures 5(a) Returning to Figure 7(a), we see the soil
through 5(f), which are cluster plots of the data brightness during the first biophase. One striking
presented in the pairs of transformed channels. affect is the way the roads stand out in this image.
The data shown is from the Ellis Co. ITS, dated Notice that the wheat fields are generally, but not
June 13, 1973. (Recall that the transformation was entirely, dark. Basically these are bare soil
developed on Illinois data.) Notice that trans- fields and we could expect some to be light and
formed channell, (TCH 1), which is soil brightness, some dark.
and TCH 2, green stuff, contain almost all of the
variation within the sample segment. In Figure 7(b) the wheat fields are dark; we
interpret this to mean that the fields have developed
Figure 5(a) shows these two channels plotted shadow in them in the process of growing. The range-
against each other. The basic triangle shape is land is substantially unchanged between biophases 1
easily noted, now rotated to the right so that the and 2, in the soil brightness feature. The roads
soil line is parallel with the soil brightness axis. are still bright.
One noticeable effect of the transformation is to
increase the apparent size of the tasselled cap, In Figure 7(c), the 3rd biophase, the wheat
even though there was not any scale factor built in fields have brightened up, as has the rangeland.
to the transformation. The reason is that in the There is little contrast between the two.
transformed data we are seeing the tasselled cap
directly from the side. In Figure 7(d), the 4th biophase, some of the
wheat fields are bright, others are not. We inter-
Figure 5(b) shows the yellow feature plotted pret this to mean that some are harvested (no shad-
versus the green feature. Notice that the data is ows) and others are not harvested. Notice that all
greatly compressed in the yellow direction. There of the areas that appear to be wheat are yellow in
are a few small clusters to the side of the main the 4th biophase, but only some are bright (see
group of clusters, which are, perhaps, examples of Figure 8(a. In the 4th biophase the rangeland is
yellow materials . again moderately bright. The roads stand out by
bright contrast.
Figure 5(c) shows non-such versus brightness.
There is evidently no structure at all in the non- Earlier we commented that there was more than
such direction. Figure 5(f) shows non-such versus two channels worth of information con~ained in
yellow stuff. One could easily believe that these LANDSAT data. Here we have shown an example. The
channels together carry only a tiny fraction of the green feature, the yellow feature, and the bright-
information available in LANDSAT data. However, ness feature are three independent measurements.
yellow stuff does show definite spatial structure They could not all be measured and represented by a
at some times, as we will see later. two channel LANDSAT.
A second method of presentation of transformed A third method of viewing transformed data is
data is by viewing transformed imagery. Figure 6 by looking at tables of cluster statistics or train-
shows green stuff images of a LACIE sample segment ing statistics. This approach is utilized in the
in Kansas, during 4 successive biophases.* The discussion in the following section.
region at the top and bottom of the segment contains

* Q. Holmes, NASA/JSC, transformed the data used in this and the next example, and created the imagery
we have used in Figures 6 through 8.

4B-44
i
V. THE PROBLEM OF CORRECTION FOR are experimenting with a cloud detector based On
EXTERNAL EFFECTS this idea, i.e., if the quantity u - u3 passes a
certain threshold a pixel is label~d cloud. (ul
We have discussed the Tasselled Cap as a way and u3 are the brightness and yellow stuff com-
of integrating the spectral reflectance structure ponents of the vector u shown in Equation 1.)
of a LANDSAT MSS agricultural scene. The reflec-
tance has, for some specified conditions of obser-
vation, a corresponding radiance and a corresponding VI. POINT OF ALL SHADOW
representation in LANDSAT counts. As the conditions
of observation change, however, the relationship Figure 4 shows a feature, a point located
between reflectance and LANDSAT counts changes. By somewhere back of the origin, called the point of
observation conditions we mean such items as the all shadow (see also Appendix B). As we change
viewing and illumination geometry, the amount of viewing angle and illumination angle the amount of
haze in the atmosphere, the amount of H20 vapor, shadow which can be seen in the canopy varies. The
the amount of cirrus cloud and the height distri- reflectance of the canopy therefore changes,
bution of these in the atmosphere; also the average becoming lighter or darker. However, the changes
ground albedo in the neighborhood of the particular are not merely towards or away from the origin, as
observed points. they would be if only a change in illumination
level were involved; there is also a color shift,
Some combination of these effects is without since the radiation reflected from within the shadow
doubt extremely significant to the problem of region is more strongly colored than that reflected
identifying field types in LANDSAT data. The very from the unshadowed region.
fact that the data within a local area is confined
to an extremely flattened structure within the By making a shift of coordinates to the point
LANDSAT signal space makes it easier in a certain of all shadow the data can be treated as though the
sense to make errors in classification. Figure 9 changes in illumination and viewing .angles did not
shows a hypothetical two-channel example in which induce any color shift, but only a brightness change.
external effects (haze perhaps) can shift the
entire data collection sideways. Two crops, Wand The key idea about the point of all shadow is
V, occupy a narrow region of the space, and are that all points lying on any radius from this point
easily separable in that region. Assume that we are at the same stage of crop development. This is
train a classifier on the data from one sample seg- no doubt not perfectly true -- it is in fact only
ment, obtaining the signatures Wand V. Then assume an idea. But a slightly simpler version of this
that the conditions change and the entire region same idea forms the basis for the red to infrared
shifts to the position represented by W' and V'. ratio (i.e., CH 3 divided by CH 2) as a measure of
Classification errors will now occur, but more than green biomass 7
that, there will be a great deal of thresholding.
In fact Figure 9 represents quite well what really To be specific, we propose to use a transfor-
occurs due to the addition of haze to the atmosphere mation of the form
over a scene. The equivalent occurrence in the
4-dimensional case of LANDSAT data would consist v = QT (x - S) + r (2)
of a shift of the entire tasselled cap in the s
yellow stuff or non-such direction. Such shifts,
ranging up to several standard deviations of the where
yellow stuff channel have been observed in randomly
selected LACIE sample segments in Kansas (where x is the LANDSAT signal vector after
standard deviation refers to the thickness of the haze correction
entire tasselled cap in the yellow stuff direction).
S is the point of all shadow

Figure 9 also shows a shift in the brightness s is a brightness measure

direction. In the real case a negative shift in
the yellow stuff direction due to haze is also
accompanied by a positive shift in brightness and
(
e.g., t (x - S.) s)
i=l i ~
=

a negative shift in greenness, as well as a general

contraction in scale (i.e., loss of contrast). The QT is a dimension reducing matrix such that,
interactions are complicated and outside the scope
of this paper. The key point is that the yellow
shift and the non-such shift are diagnostic of a v = (VV2l ) ( green color feature )
= yellow color feature
physical state of the atmosphere. We are attempting
to exploit these diagnostic features for purposes
of correcting the data for the effects of haze and Thus three features would be retained for pro-
viewing angle 7 ,8. cessing -- s and the two components of v.
Imagine now that a cloud represents an extreme
In order to use this idea we have to pick a
case of haze. Then, according to the relationships
point of all shadow to work with. Several comments
suggested in the previous paragraph, a cloud would
are in order:
appear extremely shifted in both the negative yellow
stuff and the positive brightness direction. We

4B-45
 a) The point of all shadow should be chosen
on the extended line of soils, even if that is not
truly on the reflectance diagonal. In this way
natural soil brightness variation will be lumped
together with shadow variation.

b) We can pick a working shadow pOint and use 40

it. If we have some success, then systematic
efforts should be made to establish its position
more accurately. 4A,4B
a
:1
c) The point of all shadow will be modified 0
LI"\
by atmosphere (haze) effects in the same way that r-
any other pOint in the reflectance space will be. ~
U
Any transformed features which utilize the point
of all shadow as an origin will be dependent upon ~
Eo< lA DARK BARE SOIL
the haze level. Therefore it will be necessary to u lB LIGHT BARE SOIL
~
carry out a correction for haze in order to properly ....:l
~ 2( ) ~mEAT AT EMERGENCE
exploit the color feature representation. ~ 3 ( ) WHEAT AT INTERMEDIATE STAGE
N 4( ) HHEAT AT MATURE STAGE
5 ( ) WHEAT AT CHLOROTIC STAGE
VIr. SUMMARy
6( ) WHEAT AT SENESCENT STAGE
The purpose of this paper is to illustrate the 0
usefulness of obtaining insight and understanding 0 10 20 30
of the overall data structure in a pictorial/geo- % REFLECTANCE 650 ~m
metric sense. Introducing precise but colorful
language is useful because it allows groups of
workers to communicate their developing ideas in a Figure 2. Phenology for Wheat (Ionia Variety)
Based on Canopy Model
graphic shorthand manner.

The following summary statements can be made:

a) The data from an agricultural scene lies

substantially on a plane in LANDSAT signal space, Diagonal of
spanned by brightness and green development, except 4 Space
that some protrusions out of the plane in a yellow
direction are observable. /
b) Within the plane the data are bounded in /
a triangular region.
1
c) The position of the entire plane and the
boundaries within it are indicative of the external
conditions of observation and can be used to esti-
mate those conditions.

d) A fixed linear transform can be used to

aid in viewing the data and to create diagnostic
features from the data.

e) The primary description of agricultural /

crops is in their color and brightness -- non-linear
functions of the LANDSAT channels. A transformation
to a color-brightness feature space is suggested.
Haze correction should be accomplished prior to
application of the non-linear transformation.

Figure 3. Concept of Soil Distribution

in 4-Dimensions

4B-4(i
 FAYETTE 11 JUN
g
FAYETTE 11 JUN
CUJ",ER NO, AND PERCf\lTRGE (a) ii CLUS TER NO, AND PERCENTAGE
(d)
~
Ii
C)'D'" ~oo
0-00 (3'/~.
0 n -:;,y '. G'oo ~
~',,' ."
()"~0~oo G'oG~.~~ ';'~. .

~~~
,

~P030'" 0r..
0:;!)0s0'03 0"0'''? ~. '."
s 0- 064 fo 04 g ~W"
(J",
G'''G''0'0~
I I
""s! ~~

o~"
Iu' GOl(!r.(2Yo'
~-O2(!}-OO
I~ C3-&@'03
0' &:0'00
~oo . (!j'OO

g
'.00 ".00 ".00
C_L
so."
~ "."
8
'00 40.00 :,0.00 70.00
CJiRNNfL I

FAYETTE 11 JUN
FAYETTE 11 JUN
"
~
CLUSTER NO, RND ~ERCENTRGE
(b)
CLUSTER NO. RND PERCENTRGE
(e)
8
Ii

8
(j"
~ 8
~ ~
~.oo
0' .",

8
"'51
# 8

,.'
~ ~" .5i

~8
u' ~ !8
u' ~oo02
. .f2}'00
~eg'G'00'''0Q'~P'00
~". , ':i'CY<~
I-OJ

0,{!J
'~~
.,
JOB ()4 -0; ,

g
&00 00 Goo
2

8
'00 10.00 ".00 ".00
CHANNEL
".00
t
".00 .,,' !Kt.VO
8
'.00 40.00 SO.OO
CHANNEL 2

fAYETTE 11 JUN
CLUSTER NO, RND PERCENTRGE g FAYETTE 11 JUN
(c) CLUSTER NO. AND PERCENT AGE
Ii
(f)

8
Ii

8
.5i

'L C6'OO
8
,; 8
Ii
Q!~"~'~
'
(}'. ..
"

, ,
\- - .-
8
& J;D'
, ~j,?J'OO
8

ItI.OO ~.OO
CHANIlL
!oIl.DD
3 '.00
. " 40.00
CHANNEL
SO.OO
1

Figure 1. Unsupervised Cluster Plots, Fayette Co., Illinois

4B-47
 ELLIS 13J T..2C ELLIS 13J T.::2C
8
i LlNEFfI LINEAR
(a) (d)

8
~

..
8
~

Goo 8
Ii

~
.-., 'GO
~~
s
''''{p~' ~ 2

~~

. oo~~~~~~.~~
8~~--~~~--~~,,~~~~~.~~.~ 8
ClIIIIMl , '.~ 10.00 ".~ 01
-.,
40.00 SO.OO

ELLIS 13J T=2C' ELLIS 13J T=2C

8
LlNEA/\
i
(b) I LINEAR
(e)
g
i 1
1
8
S I
..:
11
8
II
Ii

2
~

ELL! S 13J T=2C

-:00
-- ...............
"
LINEAR ELLiS 13J T=2C
LLNEAR
(c)
(f)
8
Ii

Ii 8
~

8
:j

...,.8 '"';
8

8

8

8
"
,;

8~~--------~N~.,,~-~~.~~~W.~~--~~~~~~
C_EL ,

Figure 5. Cluster Plots of Linearly Transformed Data, Ellis Co., Illinois

4B-48
 YELLOW SnrFF
FOLD F GREEN STUFF)A~~~
Fe.tllre 2, Blopl\ue 1, Green,tull, 55 No. 1172

\
LIGHTER
TASSELS
YJ)ARKER

>~:;"--''- PLANE OF SOILS

A/
POINT 0
SHADOW

Future 6, Bloph:ose 2, Greenstu!f, 55 No. 1172

Figure 4. The Tasselled Cap

Feature 10, Bi<>ph2,,,,, 3, Greenstuff, 5S No. lt72

Feature 14, Bloph.ase 4, Greenlltulf, 55 No. 1172

Figure 6 . Time Progression of Greenstuff Feature

48- 49
 Future I, Biop .... se 1, Brightness, SS No. 1112
Fea.ture 15, Bloph:lse 4, Yellow Stull, SS No. 1172

Feature 5, Biopha~e 2, Brightnu~. S5 No. 1172

Feature 16, Blopllase 4, Nonsuch, SS No. 1112

Figure 8 . Samples of Yellow- Stuff and

Non-Such Features

Feature 9, Blopl>:lse 3. Brightness, SS No. 1112 /

Future 13, Blophase 4. Br4!;htness, SS No. 1112 Figure 9 . Hypothetical Example of Haze Effect
Figure 7 . Time Progress i on of Brightness Feature

4B- 50
 Table 1. Summary of Various Vectors as Seen
9. Lambeck, P.F. and D.P. Rice, Signature Exten-
by LANDSAT, % Effective Reflectance.
sion Via Transformations of Cluster Statistics,
~s is the mean vector of soils. VI through Environmental Research Institute of Michigan,
V are principal components whose amplitudes Ann Arbor, (to be published).
4
are given as Ii.

CHI CH2 CH3 CH4

~s 15.57 21.83 25.55 31.14 48.389

14.22 17.36 18.981 20.23 35.681

4.23 .018 -2.03 -2.16 5.165

-.57 2.076 -1.54 1.30 2.949

-1.35 .166 .581 -.1408 1.486

REFERENCES

1. Henderson, R.G., G.S. Thomas, and R.F. Nalepka,

Methods of Extending Signatures and Training
Without Ground Information, Appendix 3,
109600-l6-F, Environmental Research Institute
of Michigan, May 1975.

2. Suits, G.H., The Calculation of the Directional

Reflectance of a Vegetative Canopy, Remote
Sensing of Environment, Vol. 2, 1972, pp. 117-
125.

3. Condit, H.R., The Spectral Reflectance of

American Soils, Photogrammetric Engineering,
Vol. 36, p. 955, September 1970.

4. Condit, H.R., Private Communication.

5. Johnson, F., CITARS Vol. 8, Data Processing at

the Earth Observations Division, Lyndon B.
Johnson Space Center, Part 5, Fayette County,
Illinois Supplement: Graphic Study of Corn
and Soybean Data. Report Number JSC 09391,
December 1975.

6. Ma1i1a, W.A. and R.C. Cicone, Empirical and

Simulation Model Analysis of LANDSAT Signature
Variability, Proc. of Third Symposium on
Machine Processing of Remotely Sensed Data,
West Lafayette, Indiana, June 1976.

7. Colwell, J.E. and G.H. Suits, Yield Prediction

by Analysis of Multispectral Scanner Data,
109600-17-F, Environmental Research Institute
of Michigan, Ann Arbor, May 1975.

8. Erickson, Jon D. and Richard F. Na1epka,

PROCAMS: A Second Generating Mu1tispectra1-
Mu1titempora1 Data Processing System for Agri-
cultural Mensuration, Proceedings of the Sym-
posium on Machine Processing of Remotely Sensed
Data, West Lafayette, Indiana, June 1976.

4B-51