General Disclaimer
One or more of the Following Statements may affect this Document
This document has been reproduced from the best copy furnished by the
organizational source. It is being released in the interest of making available as
much information as possible.
This document may contain data, which exceeds the sheet parameters. It was
furnished in this condition by the organizational source and is the best copy
available.
This document may contain tone-on-tone or color graphs, charts and/or pictures,
which have been reproduced in black and white.
This document is paginated as submitted by the original source.
Portions of this document are not fully legible due to the historical nature of some
of the material. However, it is the best reproduction available from the original
submission.
Produced by the NASA Center for Aerospace Information (CASI)
1
IE
y'
E82."10378
Columbia University
ii), the
City of
New York I N ew
DEPARTMENT OF GEOGRAPHY
York, N. I'. 10027
International Affairs Building
420 west 1 18th street
s imade available !under NASA sponsorship
in the interest of early a nd
urcesdO d'
seminatio n of Earth Reso
*
and without iiagillty,
program information
for any use made thef*"
+
(E82- 10378) APPLICATION OF DIGITAL ANALYSTS
1182-32800
OF MSS DATA TO AGRC-F;NVIPONMFNTAL STUDIES
Semiannual Proq^ess Report, 1 ,an. - 31 Dec.
Unclas
1981 (rolumba.a Univ.) 99 p V A0S/MF A01
CSCL 02C G3/43 00378
APPLICATION OF DIGITAL ANALYSIS OF
MSS DATA TO AGRO-ENVIRONMENTAL STUDIES
Original photography may be purohased
from EROS Data Center
ORIGINAL PAGE 19
OF POOR QUALITY
Siouz Fallq,, SA 5.7198
Semi-Annual Progress Report
For the Period Jan. 1 - Dec. 31, 1981
NASA Cooperative Agreement NCC 5-20
\^
r
`l ^^ 1 ^^32
_^
Principal Investigators
Robert A. Lewis
Samuel N. Goward
'^'
sre^ cv
`^ •,
r
CONTENTS
r
I.
Introduction
II.
Ongoing Research
A.
Simulation of Landsat D Thematic Mapper Observations
B.
Analysis of TM Simulation
C.
Analysis of Snow Cover and Wetlands from Landsat Data
D. Agro-environmental Applications of Satellite Remote Sensor Data
III. Appendix: Papers and rresentations
i
a
I. Introduction
Since 1975 members of the Department of Geography, Columbia University,
have worked with NASA scientists at the Goddard Institute for Space Studies
(GISS) on research concerned with satellite remote sensor observations of
earth resources. This activity is currently supported under NASA Cooperative
Agreement NCC 5-20. During 1981 research activities have concentrated in four
areas;
1) Simulation of Landsat D Thematic Mapper (TM) observations from aircraft
multispectral scanner data and field spectrometer data collected over a
corn-soybeans agricultural region in Webster County, Iowa during the 1979
growing season in support of the NASA/AgRISTARS program;
2) Analysis of the simulations to evaluate the potential utility of the TM
band 5 (1.55 -1.75 um) mid-infrared observations in corn-soybeans
discrimination;
F
3) Analysis of current Landsat data to study snow cover in northern New
England and wetlands in Nebraska and Vermont in cooperation with scientists at
the U.S. Army Corps of Engineers Cold Regions Research Laboratory;
F%
4) Application of satellite remote sensor data in selected additional
agro-environmental research areas pertinent to Columbia staff interests.
Dr. Samuel N. Coward, research associate, and Professor Robert Lewis,
chairman of the Geography Department, are the principal investigators. Dr.
Goward serves as the research director. Professor Leonard Zobler supports the
activity as faculty advisor. Columbia University students work on projects
under investigation as research assistants and aides. During 1981 five
, ygraduate students were funded under the cooperative agreement. Several other
i
7
r
students, funded from other sources, also supported the research activities.
r,
i
a
7
Y^
s.
i
r
f
The research activities are carried out at the
Studies. The Institute provides work space, data,
software to support the research. The Columbia staff interact daily with the
Ni'-$A scientists to carry out the Earth Resources Group research activities.
Principal GISS contacts are Dr. Stephen Ungar and Dr. Richard Kiang. Senior
Columbia staff collaborate with the NASA scientists to plan and direct the
research activities. The research assistants and aides participate in the
research and provide technical support in preparation of presentations and
reports.
II. Research Activities
A. Simulation of Landsat D Thematic Mapper Observations
In the third quarter of 1982, NASA plans to launch Landsat D. A new
observing system, the Thematic Mapper, will be flown on this mission. Thematic
Mapper represents a significant technical advance compared to the current
Landsat observing system (J.L. Engel, 1980). A comparison of relevant TM and
MSS parameters is given in Table 1. Considerable prelaunch research activity
is underway to evaluate the potential information gain TM data will bring to
earth resources observations (ORI, 1981). Columbia/GISS Earth Resources
scientists are contributing to these activities through analysis of field
measurements data collected at AgRISTARS "supersites".
The AgRISTARS field measurements program collects Landsat and ground
observations at 300 selected agricultural sites across the United States
(AgRISTARS, 1981). In addition, at two "supersites" (Webster County, , Iowa for
corn and soybeans and Cass County, North Dakota for small grains) more
intensive observations are collected by the helicopter-based field
spectrometer system (FSS), the NS001 aircraft-based ;multispectral scanner
y
system and by ground observrs. (For a discussion of the analogous LACIE field
t^
k;
2
:.'.
^
s
r.i 1.. u.. s.
Y
u...^
.. ",
r=^,.d...^._.^-_.
.v
LL
rkM
-`"'- Ar¢iswLlflfasfY
y
ORIGINAL PAGE IS
OF POOR OUALITY
TABLE 1
Comparison of Landsat MSS and Thematic Mapper
Sensor Specifications
LANDSAT
Band #
Spectral
Resolution
Spatial
Resolution
Radiometric
Resolution
THEMATIC MAPPER
Spectral
Interval (um)
Band #
Spectral
Interval Gm)
1
0.50-0.60
1
0.45-0.52
2
0.60-0.70
2
0.52-0.62
3
0.70-0.80
4
0.80-1.10
3
4
5
6
7
0.63-0.69
0.76-0.90
1.55-1.75
2.08-2.35
10.40-12.50
80 meters
30 meters
120 meters
26
(64 gray levels)
3
(256
Band
2e
gray levels)
7
`
^
r
measurements program see Bauer, et a'!. 1978.) The supersites consist of 5 x 6
nautical mile regions in which 80 fields are periodically observed by field
enumerators (NASA-USDA/ESS, 1979). These observations are timed to coincide
with Landsat overpasses and are conducted at a 9 to 13 day interval. The
helicopter and aircraft observations are also conducted at the same time
although the aircraft is flown less frequently (once in 1979, 4 times in 1980
and 1981). The aircraft and Landsat observations provide multispectral
scanner data for the entire site. The helicopter FSS observations are
non-imaging high spectral resolution radiance measurements from a flight path
across each of the 80 periodically observed fields.
The Conceptual Framework of the GISS Simulations
During 1981, the Columbia/GISS Earth Resources Group, as past of their
l
AgRISTARS research effort, have developed techniques to simulate TM
observations from the field measurements data. The satellite, aircraft, and
helicopter observations serve coimplimentary roles in the TM simulation. As
shown in Figure 1, the aircraft data serve as a means +,o simulate both TM
and Landsat observations. The Landsat simulation may be compared to the real
Landsat data to confirm the simulation technique. The FSS data serve as a
calibration standard to verify radiometric adjustment procedures, and as the
basis for extending the aircraft observations to satellite observing
conditions through simulation of atmospheric effects and consideration of the
satellite TM engineering parameters. In addition, the FSS observations permit
mporal evaluation of crop radiance behavior in TM bands at those times when
e aircraft is not flown.
The GISS-developed techniques include radiometric, geometric, and
solution processing of the aircraft scanner data and radiometric and
atistical processing of the FSS data. These techniques were tested using
4
n
r
ORIGINAL PAGE IS
OF POOR QUALITY
I
f55
RADIOMETRIC
CALIBRATION
u
NS001
SIM TM
DATA
1
I
SPECTRAL
ADJUSTMENT
"LEGITIMATE"
COMPARISON
i
INFERRED
COMPARISON
r
NSO01
i
SIM
MSS
u
TECHNIQUE
VALIDATION
LANDSAT
MSS
i
i
FIGURE
1
TM
SIMULATION
APPROACH
From Presentation material used by S. Ungar; Renewable Resources Thematic
Mapper Simulator Workshop (ORI, 1981).
5`
f
k
t
F,
ORIGINAL PAGE IS
I
OF POOR QUALITY
1979 field measurements data from the Webster County, Iowa supersite (Figure
2). One NS001 flight, collected on 8/30/79 and four FSS observations
collected on 6/11, 6/29, 7/16, and 8/30/79, were processed to simulate TM
observations.
TM Simulations From NSOO1 Data
The NS001 (NASA scanner 1) instrument was constructed by NASA engineers at
the Johnson Space Center to emulate, as much as possible, the satellite TM
instrument (Richard, Merkel and Meeks, 1978). The NS001 instrument parameters
are given in Table 2. The NS001 is typically flown on the NASA C-130 aircraft
at altitudes between 5 and 8 kilometers. With a 2.5 milliradian detector
IFOV, the scanner observes, at nadir, a 19 meter square ground resolution
element from 7,620 meters (25,000 ^eet) altitude. At this altitude, the
scanner normally operates at 15 revolutions per second which, for the nomimal
aircraft ground speed of 135 meters/sec, produces a 50% overlap between
adjacent scan lines. As the instrument was constructed prior to final
selection of TM bands, eight rather than seven bands are included in the
detector package (NSOO1 band 5, 1.00 - 1.30 um is the non-TM band).
Although the NS001 scanner was constructed to emulate the satellite TM
scanner, no aircraft scanner system is fully capable of simulating satellite
observing conditions. Not only does the aircraft fly at much lower altitudes,
thus not observing the full effects of atmospheric attenuation, but also,
because of the altitude difference, the scanner must observe the surface at
more extreme scan angles to view a ground area comparable to a small subarea
K
in one x,%tellite scene (ORI, 1981).
The NSOO1 maximum scan angle from nadir is 50 degrees. This permits the
P
instrument to view an across track ground swath of 18 kilometers (± 9
kilometers from nadir) at an altitude of 7,620 meters. This variation in look
6
r
I
r
ORIGINAL, PA r,, E IS
OF POO R QUALITY
o
N
•' 07
V)
Q)
00
c
N
k
(1)
cr
U) o
CT
Q
03
c
0
V
^
o
^ CO.
r .
Z
0
N
V,
"l
ORIGINA L PAGE
OF POOR QUALITY
TABLE 2
NS001 TECHNICAL SPECIFICATIONS
sr c7 RAL BANDS
Gan'!
Detector
P,
NEI
Spectral bandwidth, :m
1
S1
0.15 - 0.52
0,5
2
S.
0,52 - 0 1 60
15
3
Si
0.63 - 0.69
.5
ti
ai
0176 - 0.90
5
^e
1.00
!
ie
1.55 • 1.75
1.0
- 2.3t
2.0
6
100
InAs
2.08
HgCdTe
10.1 - 12, G'
.5
1.0
NE:,T
G,25°Y
D3 X14 !. DATA
Instantaneous field of view
2.5 ' 2,5 milliradians
(IFoV)
Across- tract; ficld of vies
,,Tina! aperturc diameter
rffecr;ve aperture area
10,16 cm
72,1 rm2
f n,m%ver
1.85
100°
16.8 cm
focal length
nfliart calibraticn
rrir.ary
Integrating sphere and two
controllable ' blackbodiea
Short waveler.ct), array temperature
1980Y
V.'H rance
Variable 0,025 to 0.25
Scar, rate
Variable 10 to 100 scans/sec
S.a , . speed star:lity
one --third of the IFOV, scan line to
scan line
8 bits (156 discrete levels)
Cata guar,tiration
!. -ber cf video sa-^,ples:scan line
R::1 compensation
700
Data format
:15°
Compatible with N2s
Scar. r• irr
156
Tape packing density
rotating mirror
10 000 bits-per inch (BPI) constant
Recording code
Bi-C-L
r
8
F
angle introduces significant across track variations in ground resolution and
the radiometric conditions observed by the sensor that will not be present in
the TM observations.
An additional factor, which should not be present under satellite
conditions, is complex distortion of image geometry due to short term changes
in aircraft motion. At 8,000 meters altitude, the C-130 aircraft is subject to
uncontrolled motion due to buffeting in the turbulent; lower atmosphere. Short
term changes in the velocity, altitude, pitch, yaw and roll of the aircraft
can significantly alter the geometry of the image data in comparison to
satellite observations (and other information available such as photography
and maps). The NS001 is roll-compensated (Richard, Merkel, and Meeks, 1978)
However, other mo"i,^.ns of the aircraft during data acquisition significantly
affect the image g eometry . Figure 3 presents a grayscale image of raw NS001
Band 6 (TM5) data from 8/30/79 which displays the resolution, radiometric and
geometric prperties of a typical aircraft scanner image.
Aircraft Scanner Data Processing Techniques Developed
Although processing techniques developed by the Columbia/GISS staff were
specifically directed to processing the NS001 data, these sane techniques are
applicable to aircraft scanner data from other systems such as the NASA/ERL
TM simulator and the NASA/Ames Daedalas system. The approach is to develop
techniques, where possible, that systematically account data variations
introduced by the scanner system. For example, ground resolution variations
are, in general, an explicit function of scan angle. However, certain
characteristics of the data must be treated empirically. Variations in
atmospheric optical depth, scattering, and surface bidirectional reflectance,
although they appear to be related to scan look angle, are not sufficiently
known to be subject to systematic correction. In this case, empirical
9
ORIGINAL rAGE i
s
OF POOR QUALITY
•T
r« rr ^v.^l
^^ f
L
Mo ; wk^
s
A♦
/f
.
^
#
^• ^^I^^ ^
If
fir,`
- ,
#^, , tog
!
r
4i
WEBSTER Co.
IIS001
8/30/19
1.55 - 1.15
Nr
FIGUPE 3
a
10
i
.
e
A
techniques were developed to compensate or adjust the data in a reasonable
fashion. Correction of the effects of platform velocity, altitude and attitude
variations on image geometry presents a particularly difficult problem.
Although systematic techniques are possible, practical implementation requires
a high quality source of aircraft navigational data during ,scanner
observations. Under relatively stable flight conditions, empirical
"rubbersheeting" techniques app^ar to be adequate. However, under unstable
flight conditions a more systematic approach is needed. Expermental activities
in both areas were explored by the Columbia/GISS group in 1981.
Ground Resolution
Whereas the NS001 ground resolution is 19 meters at nadir from a 7,620
meter altitude, at the 50 degree maximum look angle, the ground resolution
element is approximately 30 meters in the along track direction by 46 meters
in the across track direction. Since the NS001 instrument is roll-compensated
and pixels are contiguous in the scan direction, calculation of the scan angle
position of each pixel in a scan line is straight forward. The central pixel
in each scan line represents the nadir pixel, each adjacent pixel is 2.5
milliradians of scan angle away from nadir. The around resolution of any
pixel in the scan may be calculated by (Baker and Mikhail, 1975):
R along track - (h/cos A)a
R across track
(^i/cos 2 A)a
(1.1)
(1.2)
where R = linear resolution
altitude
h
A
look angle
a = angular resolution of the detector.
To aid in later geometric processing, the data are resampled to nominal 10
meter pixels. In the scan direction resampling is conducted using equation
1.2. Since adjacent scans are 50% overlapped each scan is assumed to
f
.
0
represent a 10 meter data sample. After geometric correction the 10 meter
pixels are aggregated to 30 meter pixels to simulate the nominal satellite TM
ground resolution, Figure 4 is a grayscale of the 8/30/79 NS001 band 6 (TM
band 5) data which has been resampled to 10 meters and radiometrically
adjusted.
Radiometric Adjustment
Because the NS001 makes observations to 50 degrees either side of nadir
the radiance measured varies not only as a function of the ground cover
present but also due to the bidirectional radiance properties of the ground
cover, variations in atmospheric path length and anisotrophic scattering of
the radiant flux by atmospheric aerosols. The relation between incident solar
radiation and the NS001 maximum look angles for the 8/30/79 observation is
plotted in Figure 5. Note that the solar flux is nearly perpendicular to the
eastern look angle and nearly parallel to the western look angle. Figure 6 is
a plot of the mean and standard deviations for each column of the western
portion of the 8/30/79 NS001 band 1 (0.45 - 0.52 um) data. The general
pattern displayed is increasing radiance with increased westerly look angle;
less significant variations occur in the easterly direction. The reduced canopy
shadowing and increased canopy cover along with increased observed aerosol
scattering due to both increased path length and increased back scatter
parallel to incident solar illumination, cause the image to brighten in the
westerly direction.
The Column-Averaging Approach
The landscape and atmospheric conditions that contribute to this effect
are not sufficiently known to explicitly compensate for their variation as
function of look-angle. A simple empirical approach to radiometric
12
ORIGINAL PAGE IS
OF POOR QUALITY
'IV,
w
^^
`
Q I've
1W4
^ Imo .
^ .+.^t^:^ i 9f
40
s
x,.
I
led
A Lws
Ills.
4
i
rA
L
q3h
INIM
Jul
iv AM
_ ^,
_
_All
SCAN-ANSLE ADJUSTED
FIGURE 4
13
.4*w
L
li
ORIGINAL P,AGS IS
O POOR QUALITY
F
W
0
0
Jam, fh
0
cZ
J LLJ
H
o
•
C/)
^z^
J
>
Z
2a
CD
W
Z
=)
Z
—
= Y
0 0
li
I
p
U)
Cf)
Z
^A
0
cn
H m
W
cn
5E
3
CP
Q
14
•
f
t,••s
^.t
t•
f•
^
f
ORIGINAL 'AGE IS
()F POOR QUAI rrf
i
. tt
t
t•tt
ss
s
i
s
i,
x
w
+I
=
rI
Coll —_-^
^ W
J
Z
J c.^
O
U
U—
O
O
J
C1.
L- -
w
i
s
s
LL-
adjustment, which
will
insure a uniform tonal distribution across the image,
is to derive a least-squares fit to the column-average distribution. As shown
in Figure 6, this least-squares line can be
used to
calculate a
column-by-column adjustment of the data that produces the uniform tonal
appearance of the image data displayed in Figure 4.
The assumptions that underlie this column-average radiometric adjustment
approach are that landscape materials of differing reflectance properties are
randomly distributed
in
the scan direction and that the bidirectional
reflectance of the landscape is independent of landscape type (e.g., corn and
soybean canopies). From visual inspection of the 8/30/79 NSO01 data
former assumption is reasonably satisfied for this data
assumption was statistically examined for the 8/30/779
set.
the
The latter
data.
Following the classification procedures discussed later in this report,
the radiometrically adjusted data were classified, on 0% commission error
basis, to identify corn and soybeans pixels. The classification map was used
as a mask on the raw data to select corn and soybeans pixels. Data
column-averages and least-squares fits were then computed independently for
corn and soybeans. Any difference in the corn and soybeans column average
patterns would suggest differential corn/soybeans bidirectional reflectance
as
a function of look angle. Visual inspection of the comparable corn/soybeans
plots revealed little difference in
the
two curves. Statistical evaluation of
the least-squares fits indicated that there is a slight but statistically
significant difference. Thus, although differences
in
corn and soybean
canopies during this 8/30/79 observation appear to differentially effect the
column-average }pattern of the data, the differen
this
data set, ignored.
16
y
Comparison of Raliometrically Adjusted NSO01
Observations
The FSS observations may be used
as
Data
and
FSS
an independent check of the
radiometric adjustment procedure. The FSS instrument acquires nadir
observations of the same ground area the NS001 scans. FSS data are processed
to NS001 sprectral band intervals (see "TM Simulations from FSS Data"). By
comparing the two data sets before and after NS001 data radiometric adjustment
the effectiveness of the adjustment may be evaluated.
The locations of the helicopter flightlines were derived by analysis of the
FSS ooresight color photography and CIR photography acquired during the NS001
overflight. The FSS instrument, when flown at 60 meters, observes
a
nominal
24 meter spot size (Bauer et al., 1978). However, one rotation cf the filter
wheel takes about one second, during which time the helicopter moves
approximately 27 meters with respect to the ground. To
select
equivalent
NSOO1 observations every other NSO01 pixel along the helicopter flightline was
extracted from the NS001 data. The mean value per field of the FSS and NS001
data were correlated
to study
the impact of the radiometric adjustment
procedure.
Table 3 presents the results from the correlation analysis. Correlations
between the 'two data sets prior to adjustment are consistently lower than
following adjustments. This indicates that radiometric adjustment makes the
NS001 data appear more like nadir observations. The relation, after adjustment
for all crops shows a strong relation (R 2 > 0.9) between the two data sets.
Correlation between corn and soybean means taken separately shows somewhat
lower final R2 values. This is most likely the result of low within-crop
variability in the reflectance of corn and soybean canopies, close to the
sensitivity limits of the two instruments. However, the differential
bidirectional reflectance of the corn and soybeans canopies may also affect
this relation. Scatterplots for NSOOI bands 1 and 4 of the before and after
17
,.
i
ORIGINAL i
OF POOR
4
W O
t
r
k
f
E
E
M
O
m
O
1ul
w
0
co
O
ONO
O
Ln
rNl
C
O
U)
W
mot'
O
O
O
1^+
O
C
WO
J V
Z
O
O
O
^
O
CT
O
00
O
0)
O
cn
O
O
FH
3
O
Q
z
N
Z
W
N
^ J
N LiJ
Z
O
O L'•
N Q
LL
M
LU
J
0
C3
O
O
O
N
N
uJ Z
I-
Z
CIS
U- C]
O
•-•
N
Z
O
CL
N
^ uj
0
O ^
O
F-
J
O
Vim'
LD
L
Ln
N
^M-1
O
O
O
O
O
O
r-i
O
O
N
O
O
O
"-4
O
lD
M
Ca
r-1
O
O
O
O
O
Q
F--
Q Z
G1 CL'
O
3 U
9
J
c.'[
r-i
d
Ln
00
P^
co7
.
O O O O O O
LU
w
ce-
O
C>
J
O Z
CD
Lo =
V
rl
N
M
ct°
LD
1^
a
;
adjustment relations
between FSS and NS001 data are given in Figure 7. These
plots confirm the tabular results from the correlation analysis.
Geometric Correction
Aircraft altitude, velocity, and attitude variations during scanner
observations introduce complex geometric distortions in the scanner imagery.
These distortions will generally not be present in the satillite data and make
comparison of the aircraft
data with maps and photos of the area, as well as
construction of registered temporal
data sets, difficult• Two approaches,
one empirical and the other systematic, to geometric correction of the
ai-rcraft scanner data have been investigated by the Columbia/GIBS staff during
1981.
The empirical geometric correction approach uses map and image control
points to statistically derive a "rubbersheet" transform of the image data
that matches, as much as possible, the Sieometry of the map control point
source (Raker and Mikhail, 1975). The systematic approach use aircraft
altitude, velocity, and attitude information, recorded during the observation,
to compensate for the effects of these variations on the image geometry
(Spencer, Wol f , and Schall, 1974). Although
the
latter approach is more
desirable, since it represents a "true" correction of these distortions rather
than a statistical estimate, it is highly sensitive to the precision and
accuracy of the recorded aircraft altitude, velocity, and attitude
information. The experiments completed at GISS during 1981 suggest that for
the current NS001 observations the empirical approach is more sucessful than
the systematic approach.
19
ORIGINAL PAGE I
FIGURE 7
COMPARISON OF FSS AND NS001 MEANS BEFORE AND AFTER OF POOR QUALITY
NS001 RWOMETRIC ADJUSTMENT
NS001 VERSUS INTEGRATED FSS FIELD AVERAGES
TM BAND I (0,45.0.52µm)
ADJUSTED NS001 VERSUS INTEGRATED
FSS FIELD AVERAGES
"
TM BAND I (00.012µm)
F
E
4
€
G
y
C
C
-
^E
t
E
V^
Vs
^ Ly»
y
C
^
^
^
f
f
``s: .- c'r "" r t
r r
r .c c
FSS EFFECTIVE REFLECTANCE
^F55 EFFECTIVE REFLECTANCE ` f'
ADJUSTED NS001 VERSUS INTEGRATED
NS001 VERSUS INTEGRATED FSS FIELD AVERAGES
FSS FIELD AVERAGES
TM BAND 4 (0.76-0,90µm)
TM BAND 4 (0,76-0,90µm)
I'
G
G
C
1r.
4
G
Ic a ty
Y
GArrF
1.
-"
0=
U
N.
V5 z
2 ^
.^
C
u* k
_
•
^i^* e
c^
€w!
c
1.1
t.
FSS EFFECTIVE REFLECTANCE
FSS EFFECTIVE REFLECTANCE
20
O
t ^'
Empirical Geometric Correlation
The empirical "rubbersheet" approach to geometric correction requires a
dense network of control points that can be easily located in the image data
and a ground reference source. For the United States, USGS 7.5 minute
quadrangles provide an accurate source of ground information. In the Webster
County, Iowa, region one feature, which is common to the maps and and the image
data, is the road network. The road system follows the Township and Range
survey system at one mile intervals. The intersections between the N-S/E-W roads
were selected as ground control points. The road intersections are uniformly
distributed in the data and easily located in the maps and in the image data.
To encompass Webster County supersite 59 road intersections were selected as
ground control points (figure 8).
The map coordinate locations of the road 'intersections are derived from
the t)SGS maps by locating the center of each intersection oil
map and
digitizing the locations via the flatbed digitizer at the Columbia University
La ► nont-Oorherty Laboratories. The digitizer table precision is 1/1000 of an
inch. For the USGS 1/24,000 scale map this results in a ground coordinate
location precision of better than one meter.
The image coordinates are located on line printer grayscales of the ten
meter data. Preliminary experiments were conducted using grayscales of NS001
bands 3,4, and 6 to determine which band or sands were best suited for
identification of the road intersections. Band 3 was found to be most
effective, followed by band 6 and band 4. Two analysts, each working with a
different bands, independently identified the line and column locations of the
59 road intersections in 10 meter data. The road intersection is defined as
the point at which two lines, one located at the center of the N-S road and
one located at the center of the E-W road, intersect in the grayscale (Figure
9). For the 8/30/79 data the two analysts generally selected the same ten
21
ORIGINAL PAGE 19
OF POOR QUALITY
woova
O
00
UJ
O
O
*iA
Vl-^%-
22
O
ORIGINAL PAGE IS
OF POUR QUALITY
•• ► ..••• ► ••1. •1•..••.•.•....•• ► .0 a\\t • • •••Y Na.W11rYU \q Va1t^1.rJJ M11
.••1..r1 N •.• ► ••.•1. •.......r.. n .. \\\J,^
..•I.JNW 11yy V\a. e lYtl.1' • N11
./•. ► . ► .. • ••.••.•.........•......•u\Jt1.1
--•IeWwlrLw\ \e.11/ -. b11
1.•.r•....ra • e•m.••1.
.................
aUre1-w11a1 lIJJJaM/IhI YJ ••e•e W./
m.1r.rrr..•..
•..m.uJJ' - .•JJV.1...WYJ..•.• aJ11
. ► . ► ..aaV 4 11e•••••rM...
..r...•. r•.I a1. W / •1•W..m1.mJJJ...•.•• ^^1J
1r ► r111C4.44JOJ ► • ►
a ► rrrw ..••u.••. ...•.•m•.. a&
11.:1\tO ki -a.JM /11.\Jdr uwm.• JI11
u 11 a a1\ aa.ou . n ► ...•ra..r.••....a•• u.. a\\d••. . N ..m..J JJ ......• •IIf
.I.a1 ► Cta\uaa..r. r r..urr...rru nn .. uJJr.... • •f ... rAIYJ.••r•.• •hJ
n ••• q 1J w11m•1.e •/.•
• lI 111.1•\JJW \11r. .r•..••••.r..... r•• nn .\\JJ r•
1 1.1\\ \00\r r.. .r.•.r.... ..r •r.. N \ \ WV . .. • .••e1\W IIeease t /1•
'11111
U1a1/ • n .aJJJJ. n ••ru.ur.0 r... . .14. .. a. Y • ' ..rr.•••...•......-. ^t1•
.. u.•.. .•-......
• 111\11r • •a JaW ra ► .•.•r rrrrll•uuu u..•a ua/'
•Y\NIr••m\\\\1 n .. n . n .1. ► 11.H a... u1•u.. J.1• .w.. • •.. J...1..11e.. N1•
•11 ►\\• ► .ra\\\\aJa nn ar ...•..r.•11.••u. w nn JJJ• Wr..,,w.•-••»•• X111
arrau JJ•1•1...... •Ilr
.1r11Ur..m\\VU U .•• n ........ n •.. ..•rr. rr.J_
.111111 ► ...\\JO\ \U rr. r•.•..r^.. u.las. r.rr a..a►..^I YUJ•11-»•.1. ^/II
•. n 1\\•1./•\\\\\ n eam n .r M..••.....•r... r .. rI 1. • rJ Jr J•1 ► •--•-•• IN//
•..w•1e/«...... N ► I
'. ► aa n . ► r 1 n \ \OC7 \\aa.a.rr n u•. ....•....1.....a.
. .•••.. Ie.•..... r/ ► •
..a.. ► n r...\100\a a n u. u.•r..r....• Nrr... •..
rr...r.••a.\rJOia• n marr•....••..u•.e ► . ^.. ..
lb
•.111-..-...... ^I/1
ru ► .a. I• n a1\\ n 1\a..J.U..err.l.r..a1e•.•.•u\,
.........«.. ^I11
• 1 ► r n 1.ra!!r\\ \u n urti a•ru ..1•...1.. .•.. r• \r• . ^ .........
.. ».. .... .... ^1I•
•1/1.r\11.:Noesa\a//\.....r1•.•rr.•a.0 •.r. Yr..
raC are\ra.1.l1111\1.1rrr... 4.0 ..... .......•Ir ..-- •11.....11...1• ^t1•
•..........1111111 N .J
.a.lala ► 11• .. . r. .al. r a .Iw ..1•....•...••• ••.r.
^ ../11111••e1 y1.1
.aa„ l ^r. •.... .rlr. n . u•.... r... r••. ......
.r•r••. •• - •• 111111.• .. ads.
ra.. ►C. ••ur.. ..r.. urr. ... n •
1a. ► .a.1•.•.rr...e•r. n.. r.r1...••.••••4r.. r 1. 1 .•11..1• N ••1.•.• at 1J
..1 .. n . J•.• Y111•.. 1. .1.••1e a/11
I.. r. nn ► \. n • n rr n ...... r.. •..
.. 1. Y 1..1..1 .r/o.
1.1 ► a.\ ► aa..rr.l.aa..r rY.rrr..1.... .r.. 1000..\.•/•
.11.111•..•1• :/a1
I. n am1\ ► 1.••.•\\\a•aaaa.•m.....e.•e11.. r.0 1•\re•
. .. ► a.J la.r .r ra\\ n a n u.. glu ..r....••'u.e. .. n .\• - • .11-.• w el r..e fit./
.11....11••.• Jt1e
/al a ► .... n .. n . \1.• • aa.r•.•rr... .y 1r.. .....IJJ.•
...11....1•.. /f.•
• ► ... r ••.JJ
.
..•meta:.. .. a \1\.r n .! N...11..
••111.••• ti .. of •V
... • C Jla . a r Y .r..1.......•.•.erV •.•.•
.10...01
1leM •..
be Us 0.
• -••/arll•11.•• •1.4
•••••
.1.--0000../ ate.
. ..111111/11 •i iu r
0000...
..
•.m•1 n •1..1rrr11.•.•^.r+•
N •1
• 11••.••. •.•++•r•.1.. 11.•• •. ..Jr rJr
^fvl
_
.r. ..
^
^.1.r.J01••
-
. • 1 .. • • • • • ... • • • r 0000. ► ► 0 1 11 0 • e• .. e 1 \ 1 / \
`- ^
Ot v . ^ ' IV
11 r 110 n W•
••.1e..•..•...rr n area.•.. •r•r.•••..•• r •• .rJ
aJJJ
....e1..... .. n ......r...r.•.....•1..••a..
u u ..........1r. M,
0000. rr•a fir. • M :0•.0.00 0.0•.01 JJJ1
. ► 010.••: 0 0... • . • 0 0 0 r..•.•
.,1111... r....rr...
r1.....r...••.• JJ r/
.•..
00:00.:-.
.
. ...r...r .r.r1161
r r.e• rrr. • n ..
•t1...e e n .rr. ......• JJJe
.1.0....00• r .r..
•.rrr.•
. ..,
.•.•
.•• V a.\
Jlrr
.
... •0001. •• .Jr
"a
00.00. /rrr.lr . •r. 0000..
. .0
1. 0000
0.00••0.•
0000
.••
...ea.• JJr.
.••. a•••a11• •100.0••
.
00••00
•... JJIJ
...
-•r0eelrr••
..
.•.•r..
.. ••.. r.
r . ► •••...1•.... •.......r.. ...•
w r. ► n .....•....• reel
► 1 ►► 11•. ► m...r.1... w 11.• a r.... n r. •rr. .. ..r1.J1-. r..ler.r ..a•..• Jllf
0.000 . ► 0••000• rra\ • • •J1 ur..r. .1•.• JJIJ
.^•e.rrr..
►► 100.••0•.
-J. a r••.. 0000 JJI•
r. ► m.r11..r.rre...m• rr....•r•..0. 0.r. rrr ^. ..
•..:.1•^•e•mr...1.1•rrr.r..•.. ► 11e..•••1•.•1r.. • Ok.N 1.err.e.1.•.• JJI!
........•a.........r..r...•e •....re..•... JI.
.a•.• r ...
J
.r .••.•e.rrl /..r•.r. .•..•u.....
.. a...ur.
u. r . • , 1 . n 00.00 J ll/
•rr0•r•e.e•0r..e •.. .•nr ...•r..0.r•. arr.
............ J1
• n •r....•..0•r••.•r.v.• .. n •.r... n r.. .e... 1•JU....•.....• aJl.
ru.••... ..a r0 . ► ••0000 r. u.•....r ... .. r.. •. ..
dy .. r.• n .. .. ..••. JJtJ
dalWr.r. 1•. .•.•• JJIa
r 0 ► •r.•ra Nr•Or Nr. w...... n. .. y r.r ...•1 1•a,
r• ► r•..rr.al...tre.•rr.•. n r ...r ..0 .rr.r.0 .•.. Clu .lr.r..r..ea•.• JJf.
1.••.01001.rr. n a. n •rr•. ........ . n .rr• ..r n . •e.J.01•rr. n r•e.•• JJda
.r.r....•r. YJJ.••ru.•••.r• J1f1
. ► ....•.mr..00 ► .•.1w•...••...
... ..•. 0 0 00.. Wa...r..r ..a. • .• JJf1
••
.r• rr.. n Nr•rr•.
a ► .......•.101. n JJ1r
.•r.... N •... nn a..rrr.r..r......
. w .r0.
r ... n ► . ► 1 N . ...eaaa a..a.r.e••.r......•.r.•. I.r. YJ....... .•..1•. JJ. I
► r.0. ► • ► 11 nn u e t a \O{. _)11. ..rte.•.•..•.... r. n. n e/• .N..J• rr.....e... 1 Jll0
.
. 11 ..
40 ..m1 n .. n ..a a\ \u .. ..a..•.11u. ur. .. w..r.. 0.•ar ►r..........•.••
....•.•• JJ1J
•I...•em•.•rr n r n .a\000a. Mrer.a• a.a ► urw.•.e .. JJ. ►u
JJII
•Ia1 n ► .••••rrrre.aa•N1 r .. r.uru a ► n ...•.1...,,.. -•all....•..•......•
.r01...•.•I.rr..I.e1rY.r.•.... ..\• ► ..a••\. 1e.,.• -.!la.. rr .0.1..0eJJ JJI/
. r n ......1.r.,.e01rlr...•r....... n rra.aa u•r..- •11..•m.r••.•..•1.a 11JJJ
.....r0.. •••r1 ... 0000
.0000 .. r..l--••-\. nn •..•••1•.•r.• aJl1
.r u•..
• ► ..r.••••a•rr• ► •r••r.00••00u.• n r....•ru.r............•.urr•mrr•.• JJJa
alit
•.e. n a•.•.a.r.r••0•.r..rw•••.•. 0000. .a•r n .. .. .,. n ....r ur1.a. n
aJJ/
.0 • ► ..m.a•...r.1..rrr.r...•.u.•..•u••r.a rJU. ..1lrr rr n .ru ... n
*lie
•..•.••m1` n 1..r..•.. r1..•.1...• n rr • n r..•F.... .111..••. r. ...N
a..••......... r.• all.
r•.a••r.1u.0 n a1 ► 1.•.. a.•...r.. u..0rr1 .a./ 1/..
I • • .•Ja ►r r.. n ..•ar•• .1 leJ
r.0•.0.1 ► ....•re.. •..r .......•.•1e.JJJ. J.
"J oe
FIGIIRE 9
LOCATION OF A ROAD INTERSECTION IN
NS001 RAND 3 GRAYSCALE
23
R
9
meter pixel. In those cases where they disagreed the difference was one
coordinate location in one direction only.
The rubbersheet transform is derived by calculating a least-squares
polynomial relation between the image and map coordinates. The general form of
the relation is:
n
X
1
n ao +
a i x i(2
1)
=l
Y'
x
where x',y' represent the
bo + n b i xi
ixi
(z•?)
map coordinates and x,y represent the image
coordinates. This relation is used to
reassign data from the image coordinate
system to the map coordinate system. The nearest neighbor criterion is used
to specify which image data are assigned to each map coordinate locationt.
Transformation experiments conducted with the 8/30/79 NS001 data led to
the conclusion that a third order polynomial fit produces visually the most
acceptable solution. Although higher order polynomials result in lower root
mean square (RMS) error relations between the image and map control points
than the third order, they also introduce geometric distortions in the
resultant image that are not present in the raw data. The RMS error to a
third
order fit on the 59 control points in the 8/30/79 data is approximately
60 meters (two TM pixels).
Local Rubbersheeting
Analysis of the third-order transformed 8/30/79 data indicated that
the greatest residual error was a high frequency variation of the distance
between control points in the flightline direction (Figure 10); possibly due
to aircraft pitch motion. Nigher order polynomials had been judged
ineffective. An alternate approach in this case is to rubbersheet local
24
ORIGINAL PAoE IS
OF POOR QUALITY
FIGURE 10
RATIO OF VERSION 1 IMAGE TO MAP DISTANCES
FOR ADJACENT GCP's (IN Y DIRECTION)
I
Wl
•,
—•
00
a"
ss
4
• \1
-OWNS
r
•
6.55s
0
IL
a p1^ti
e
!.'I
ey^^
25
ds, ^
e^,1
G! Vito
a
strips of the data in which this residual effect could be treated by lower
order polynomials (Figure 11). To insure continuity
of the
image strips
follow N g transformation, the transformation of each strip #s constrained to
fit
selected control
points in adjacent strips. This insures that the
transformed image strips
transformation
is
will
match at these points and, if the local
reasonable, that the strips will properly
rubbersheeting of the 8/30/79
data produced
mosaic.
Local
a visually acceptable image
an average RMS error for the 59 control points of less
than 10
with
meters (Figure
12).
Systematic Gcometric Correction
Experiments with the 8/30/79 NS001 data demonstrated that the empirical
rubbersheet approach to geometric correction is not well suited to aircraft
scanner images where short term variations in platform motion have introduced
local geometric distortions. Preliminary screening of the 1980 AgRISTARS NS001
observations indicated that this high frequency local geometric distortion is
typically present in the NS001 data. The 9/10/80 flight over Webster County,
Iowa presents an extreme case of this ; p roblem (Figure 13). There is a great
need to establish a more systematic approach to geometric correction
of
aircraft scanner data which draws upon recorded aircraft motion information to
correct for motion-introduced distortions.
NERDAS
The navigational parameters of the NASA C-130 aircraft are recorded
during observation missions by the NERDAS (NASA Earth Resources Data
Annotation System). Signals from the Litton 51 navigational computer
(LTN-51), radar altimeter, and attitude sensors (pitch and roll) as well as
clock time are recorded by the NERDAS (NASA, 1980). The relevant NERDAS
recorded parameters and their precision is given in Table 4.
26
►—
pRIGINAL PA GE
p IV
OF POO R Q
n a ap i
•
^^^
e
-~
Im
,^
•
'- - ";W:
iii/
^k
IL
cow
SJW
^
^
•
, Na^ti r
AO
^r
ow
AZ
t
A
_• ^,
FIGURE 11
STRIPS USEL FOR LorALLY
RUBBER-SHE T ]Nr-
27
2
^;
ORIGINAL
I' atir "
OF POOR QUALITY
ik
^
Rz
r
r`
wit
PC
r
^r
M7
z ..A(
WAS
a
Ilk
s
^1
i
^y^N•
FIGURE 1^
PIECEWISE RUBBER-SHEETED
28
% ^►.^^
1
[
'\mv Y, Y^IM
t
ORIGIN AL F:
OF pOOR Q -;'-
r
r
wir `^
01.
r
'
^'•
t
1
qa
^'
+
a
A
!VULVpp,
y
t
_
a
>r-
AL
r
AGRISTARS ITS
FIG;iRE 13
,)g
NS001
SEPT. K, 19^:
TABLE 4
NERDAS-RECORDED NAVIGATIONAL PARAMETERS NEEDED
FOR SYSTEMATIC GEOMETRIC CORRECTION
NERDAS PRECIS ION
PARAMETER
Ti me
0.1 secon d
A 1 6'i Lucie
10 Feet
Headi ng
0.10
Drift
0.1'^
Roll
Oat°
Pi tch
0. 10
Ground Speed
1 knot
30
9
In the
ideal case the NERDAS data should be updated for
(pixel) of each scan. At
each
observation
a scan rate of 15 revolutions per second (rps) an
update rate of 0.2 milliseconds would be
the aircraft is stable during one
required. Under the assumption that
scan, an update rate of 67 milliseconds is
needed at 15 rps to account for motions
between scans. The NERDAS records are
updated every 40 milliseconds, however individual parameters
have differing
update rates (Spencar, Wolfe, and Schall, 1974). In particular, information
from the LTN-51,
1.8 seconds.
including
ground speed, heading, and drift, are updated every
This is convoluted with the 1 second LTN buffer update rate
which creates at least a 0.5 second uncertainty in what time the measurements
were taken. At the 15 rps scan rate 27 + 7.5 scans will occur between updates
of these critical motion parameters.
The equations employed to systematically correct the geometry of the
NS001 data are given in Table 5. Note that since the NS001 instrument is
roll-compensated this parameter is not included in the equations. Clock time
recorded in the NERDAS and NS001 data are used to register the data sets. The
LTN-51 derived parameters are cubicly interpolated and
linearly
smoothed to
reduce the discontinuous nature of the observations. g ad NERDAS records, as
noted in the J SC NERDAS data quality report, are replaced with interpolated
values.
Preliminary experiments employing the NERDAS data to systematically
correct the NS001 image geometry have been dissappointing. The resultant
image for the 8/30/79 data is shown in Figure la. Analysis f this image
comp ared to ground conditions indicates that the technique and/or NERDAS data
do not properly account for aircraft variations in drift and/or ground speed.
Although the LTN-51 data appear suspect, numerous factors, such as the
31
L' -
-
TABLE 5
ORIGINAL PAGE IS
OF POOR QUALITY
SYSTEMATIC GEOMETRIC CORRECTION
USING NERDAS DATA
FOR RO LL COMPENSATED SCAN MIRROR VELOCITY
X = Xo + h tan g sin ¢-h(ion f/cos8)cos
Y =Yo-h tan 8 cos ¢-h (ton ^ / cos O) sin
Xo(1 +0t ?=Xo(t)+f v sin(0+8)dt
Yo(t+At)=Yo (t) j v cos (O+S)dt
where; X 0 , Yo =nadir coordinates
*--scan angle
At --scan period
h = altitude
8 pitch
v --ground speed
$ heading
f = frequency of
observation
S = drift
UPDATING Xo,Yp WITH NAVIGATION DATA ( NERDAS)
(1) Find nearest observation times from NERDAS
where, i = INT [(t-to)/f+0
t,= t=ti.,
(2) Obtain values for each scan by cubic Interpolation
Z(t) = o o " t"
where: Z represents [041841111 or v
using NERDAS values for Z(t,) i-I s k :-i+2
solve for a.
nz 0,...4
i
(3) Calculate nadir displacement between scans assuming
4
j di E a [(t+i/f )" -t")
#
1
4
1
where: Z represents v sin (^+S) or -v cos (0+8)
and a " is determined by technique used in step (2)
i
32
i
O
RIGINAL PAGE IS
OF POOR QUALITY
•;
Z 11m J
61
:;
I
oil "`
^.
m
a r0if r
`
-,
-.^.,.
dw
0z
IL
I
W.m.
Figure 14
8/30/79 NS001 Data Systematically
Corrected Using NERDAS Data
33
^
•
•
a—
`
^.
alignment between the NS001 and the various motion detectors or innaccurate
treatment of velocity effects are other possible sources of error.
Further analysis of these problems is required. the systematic approach
to geometric correction is the best general solution to rectification of
aircraft scanner imagery. The Columbia/GISS staff will explore a number of
alternatives during 1982. One approach to resolving the current discrepancies
in the systematic technique will be to examine how well that data can be
brought into accord with the control points by varying the magnitude of
individual aircraft motion parameters.
TM Simulation
Once the ground resolution, radiometry, and geometry of the NS001 data
are processed a simulation of TM observations may be constructed. NS001 band
5 is deleted from the data and the remaining bands, with the exception of
NSOOI band 8, are aggregated to the TM 30 meter ground resolution. NS001 band
8 (10.4 - 12.5 },m) is aggregated to 120 meters to simulate the TM thermal
infrared band ground resolution. On the basis of the TM orbital parameters the
data array is rotated and an array of 366 by 318 pixels , representing the 5
by 6 nautical mile area of the AgRISTARS supersite, is extracted from the data
(Figure 15). The data are then convoluted with appropriate header information,
in JSC Universal format, and forwarded to AgRISTARS scientists in Houston as
well as subjected to analysis at GISS.
Further steps in processing are possible but have not been carried out in
1981. Examples include calibration of the NSO01 data to bidirectional
reflectance factor by regression to the F9S data; conversion to nominal
radiance and inclusion of expected atmospheric attenuation at satillite
.
alti^ude; and inclusion of the TM engineering instrument parameters. These
additional data processing steps will increase the realism of the simulation
34
i
It
ORIGINAL PAGE 19
OF POOR QUALITY
1p
-
I
is
"Co
W-
\^
I
VIGURE 15
ROTATED, TM,
RESOLUTION
35
kt
T
of the expected satellite TM observations. Further progress In this area
is
expected in 1982.
Simulation of TM Observations from FSS Data
The Field Spectrometer System (FSS),
a high spectral resolution (0.02 um,
0.4-1.1 um; 0.05 um for 1.1-2.4 um; 0.5 gym, 8.0 -14.0 um) nonimagi ng
instrument, periodically observes the AgRISTARS supersites from a helicopter
platform at a nominal altitude of 60 meters. Missions
are typically flown on
a 9 to 18 day cycl%i during the growing season, in phase, as much as possible,
with Landsat observations of these locations. A set of 10 flightlines are
flown during each observation period. The flightlines are planned to pass
over 80 fields that are also periodically observed by USDA field enumerators.
Instrument calibration is achieved by observing a panel of known reflectance
before and after each pair of adjacent flightlines are completed. A
boresighted 70 mmii camera is operated in conjunction with the FSS instrument to
provide a visual assessment of the ground conditions being observed by the
spectrometer.
The FSS data are preprocessed at the Johnson Space Center to selectively
extract only observations of the 80 periodically observed fields and the
calibration panel. These data are forwarded to LARS-Purdue University, where
the data are converted to bidirectional reflectance factor by
calibration
to the panel observations. In addition, the field enumerator observations are
appended to the appropriate FSS observations. The FSS instrument generally
acquires multiple observations of each field (the number depends on the field
dimension in the flight direction). Both single-scan and field
average data
are produced by the LARS staff. LARS processing of the field average data
includes a statistical method to delete "'unlike" observations prior to
computation of the field means (Kiehl, 1979).
36
k
.
The Columbia/GISS staff acquire the processed FSS data from LARS and carry
out further
data processing to simulate TM, Landsat and other sensor Systems
spectral configuration. The nominal spectral configuration of the FSS
instrument is 0.02 um between 0.4 -1.1 jim and 0.05 um between 1.1 -2.4 jam.
Studies by GI,SS and LARS scientists suggested the actual configuration
deviates from these specifications. A bench test of the FSS filter wheel
revealed that the spectral transmission of the filter wheel is a nonlinear
function of the filter wheel position (5arrnett, 1950). Based on JSC
recommendations, the recalibrated FSS spectral intervals are used at GISS for
data processing. TM, Landsat MSS and other spectral reflectances are computed
from the FSS data by integration of the appropriate FSS observations weighted
by the proportion of solar illumination in each FSS spectral interval. This
computation produced the "effective" spectral bidirectional reflectance factor
that the TM, Landsat MSS or other instrument would observe. Figure 16
provides an example plot of FSS data and integrated TM
band reflectance from
Webster County, Iowa AgRISTARS supersite. Additional simulation steps that
may be included to produce more realistic results include: conversion to
nominal observed radiance, simulation of atmospheric effects observed from
satellite altitude, inclusion of appropriate sensor filter functions, and
conversion of radiance measurements to count data based on sensor engineering
data.
B. Analysis of TM Simulations
One of the major reasons to simulate TM satellite observations is to
provide a data base with which to evaluate the potential information content
of TM observations for earth resources research. One pressing issue is whether
the TM band 5 (1.55 -1.75 um), mid-infrared (or shortwave infrared) radiance
data will provide information about earth resources that current Landsat MSS
37
F
FIGURE 16
ORIGINAL PA.QV IS
'OF POOR QUALITY
SOYBEANS
F I E L D 40
PLANTED: 5/14/79
WEBSTER CO,. IOWA
.0
so
135/1Q79
INKY 151
143/1979
%D
%a
20
20
0
0
6"
i0
162/1974
r
180/1979
tJ^hE* 111
4:^
1 m A y 25:
(JUNE 29)
4a
M
t
}
I
t
V s-.a wt : ws -n.r+e 1*^ire.e*n^+-- --.
.... ^.. ,sn..n^ ^cr^w ,
C
E.^^^...-
6C
t
357.';; ?G
196/1979
(JL..Y IE:
4C^^
4D
2i
2C
i0
!0
2G
U
L,4
L
L
0
0.4
0
D.8
1.2
1,6
2.D
2.4
38
IJL_y Ili
0.°
1.?
1.6
2.?
2.4
f
i
bands do not. This band was selected for observation of vegetated landscapes
on the basis that leaf reflectance
variations in these wavelengths are
conte:it, versus pigments in the visible
principally a function of leaf water
and leaf mesophyll structure in the
near-infrared. However, only limited
research has been conducted to access
the comparative utility of the '1.55 -
vegetation analysis.
1.75 um band for
In 1981 Columbia/GISS staff conducted a
preliminary study
of the mid-lR
(1.55 -1.75 Nm) reflectance behavior of corn and soybeans from the
collected
in
data
Webster County,Iowa• The investigation included histogram
analysis and parallelapiped classification of the processed NS001 data; a
comparative analysis of the NS001 and FSS data for 8/30/79; and temporal
analysis using the FSS data. The principal conclusion of this work is that
the 1.55 -1.75 um TM band should improve corn-soybean discrimination in
comparison to observations restricted to visible and near-infrared portion of
the spectrum. In particular, for these observations of Webster County, Iowa,
the 1.55 -1.75 l,m measurements
provide corn-soybean
separability earlier in
the season than visible and near-infrared observations.
Experiment Design
NS001 data, FSS data and ground periodic
1979 for the Webster
observations collected during
County, Iowa AgRISTARS supersite
were used
for analysis.
Eighty fields within the site, selected by the field enumerator prior to the
growing season,
conjunction with
are observed every 9 to 18 days; as
other observations
To produce ccinparative
results,
much as possible, in
by the FSS, NS001, and Landsat sensors.
the analysis was constrained to those fields
which were observed by both the NS001 and FSS instruments and which could be
unambigiously located in
the NS001 data. Four fields (29, 30 0
37 and 79)
were
not observed by the FSS (apparently due to chickens that were disturbed by the
39
l
helicopter) and three, fields (8,17, and 22) could not be accurately located in
the NS001 data due to ambiguities in the available ancillary information. Of
the 73 remaining fields, 36 were corn, 33 were soybeans and 4 were oats.
Analysis of the 8/30/79 NS001 Data
The NS001 data were radiometrically adjusted and resampled to 30 meter
ground resolution, as previously described. The 73 fields were located in
printer grayscale maps of the data by consultation with AgRISTARS field
boundary maps and aerial photography taken during the NSOOI overflight and
earlier in the 1979 growing season. The analysis was constrained to "fieldcenter" pixels to study the typical reflectance behavior of corn versus
soybeans canopies; indepedent of mixed pixels at field borders. Field-center
pixels were defined by the largest single rectangle which could be located
within the field without including farmsteads, drainage ditches or other
non-crop features.
Classification Approach
Visual inspection of the NS001 data on the GISS image display system
suggested that corn and soybeans could be more easily discriminated in TM band
5 than in TM bands 2, 3 or 4. To verify this observation, the data were
subjected to numerical classification using the GISS-ISURSL parallelapiped
classifier (Hyde, Goward and Mausel, 1977) . 10 corn fields, 10 soybeans
fields and 1 oats field were randomly selected, on the basis of field number,
to serve as test fields. The remaining fields served as training fields.
Several band combinations, specifically bands 2 and 3; bands 2, 3, and 4;
bands 2, 3 and 6; and bands 2,3,4, and 6, were used to classify the data.
Preliminary class signatures were extracted from the test field data by
computing band means and standard deviations for each crop. The parallelapiped
class signature was then defined, for each band, as the range of data within
two standard deviations from the mean. Class order is a significant component
40
of the GISS-ISURSL parallelapiped classifier because where class signatures
overlap a pixel will be assigned to the first class in which it is a potential
member. The strategy employed is to order class signatures within the
classifier with respect to signature range magnitude. The class with the
smallest signature range is entered first. In this case, the corn signature
was the least variable and the oats signature the most variable. The results
of training field classification for the preliminary signatures are given in
Tabl a 6.
To evaluate the performance of the paral'ielapiped classifier a proportion
estimate is calculated. This proportion compares the actual number of pixels
in the class training fields with the numbers of pixels classified as being
members of that class. Where the proportion is 100, the classified and actual
are the same and the classification is considered accurate. Proportions
larger than 100 indicate an overclassification of the category and smaller
indicate underclassification. Because the parallelapiped classifier is order
dependent the overclassification of the corn for the bands 2 and 3 combination
indiacted that the corn signature was too broad and could be impoved by
reducing the signature range. This signature refinement is conducted
interactively by reducing the upper and lower bounds of each band until the
best proportion estimate is achived. Table 7 presents the results of this
analysis. Note that a significant improvement is achieved in corn/soybeans
classification for the bands 2 1 3 combination. A minor improvement is achieved
for the 2,3,4 combination and no change was necessary in the remaining band
combinations.
l
Classification Results
The test field classification results for the 2,3,5 and 2,3,4,5 TM band
combinations are of particular interest. In those cases where the TM 1.55 -
41
ORIGINAL PAGE 1S
OF POOR OUALITY
00 °
00
° 00
Z
a
..J
W
Li
C
o0
0
M
0
M
0s
00
O 0
4^
dump
cc
a
V co O O 0
— 4- — —
z
N
^
Z
z
C=)
M
00
LL.
O
Z
LLJ
^
F-•
4
U
J
U- ^
F-
aN
c~
i^
z
^+
" w
LU
a0.
^,
J
w
J
UW
U
z
Q
2
0
f.i^
_J
J
c
cr
>W
w M M
0
CL
d.
W
Z
c0
Z M
=
M
N N
M
..
q
M
N N
V
3
a
4
42
r
ORIGINAL. PAGE I
OF POOR QUALITY
000 i
0000
2
O
H
O
cv
J
LJ
0
CD
Ln
ch
o"
^=
Q
W^
d^
00 O O
00 0 0
M
0
lx
1
CL
V
0 M — 0
0 0 0 O
a
wZ
t^- —
^ Z
o
d
z
0
Ln
LLJ
°w
Z
CL U
w
cQ
J
Q-
C4
J
Q
0
01
07
Q,
W
a)
M
01
M
w
0
0
u-
iii
W
a.
N
-J
Z
Z
49
=
V
43
tt c0ob
M M M
Ww
tt
M
N N N N
1
1.75 pin band is included classification accuracy is improved by approximately
2%. This suggests that the mid-infrared band will improve our ability to
discriminate corn and soybeans during
this period
of the growing
season.
Evaluation of the training fie l d performance statistics confirms this
conclusion (Table 3). Classification accuracy is approximately 6% higher with
TM band 5 included.
Discussion
Laboratory investigations of corn and soybeans leaf spectra (LARS, 1968)
in the 1.55 -1.75 j,m spectral range have observed the same differential
reflectance behavior that is observed in this classification study of NS001
data.
In essence, the reflectance behavior of corn and soybeans leaves
differs most sharply in the 1.55 -1.75 um portion of the electromagnetic
spectrum. Little is known
about how this leaf reflectance behavior translates
to canopy reflectance behavior. Examination of histograms of the corn and
soybeans
data extracted
from the NS001 data (Figure 17) suggests that although
on average the reflectance difference between corn and soybeans about the same
in the near-infrared and mid-infrared that the variability of the reflectance
in the mid-infrared is less than in the near-infrared.
Comparison of 8/30/79 FSS and NS001 Data
As previously
discussed,
the radiometrically adjsuted NS001 data and the
FSS data, integrate to TM spectral bands, are highly correlated when field
means are compared. Further evidence of their equivalence can be noted by
comparing
9 histograms
P
of individual observations from each data set. The
regression relation between the NS001 and FSS field means was used to convert
k
the FSS bidirectional reflectance factor data to 8-bit count NS001-like
measurements. Using the FSS single-scan data, histograms of
corn and soybeans
44
i
r
E
r
ORIGINAL PACV.* E
OF POOR QUALITY
O
W
CD
tD
co
in
U
Z
Q
N M
4
p
_
Q- O
W
^
W
Cl)
a
C)
CD
(n
L
W
M
1
U
OD M M
LLC)
Z
W
O
W
F--
co
u
H
Li.
^
O
u-
in
w
Cl.
U
Q
Q
Q
O
C?
--
CD
c
W
CD
CD
0)
0^
N
N
Q
J
U
^^ss
LL
W
Z
LU
J
J
9a
CL.
0
0
cr-
W
0.
J
2
Z
^
ip
ff) M M M
N N N N
U
ORIGINAL
OF POOR
s0
u
o
op
N
_
N
4
" ^o
^M
wm
rV^
Vf
^
D
0^
0
O
t
m
^
n
p
H
k
0
gp u c
t
i
o a
Q
r
u
u
i
Cn
O
FC
H
Z
N
Z
^
f
f
r
ti
r-1
LLJ
LU
w
C7
>O
y
CL's
LL
CS:
V
O
O
Z
N
0)
0
C
O
M
00
i
O
iq
ti E
O
N
\ O
N
P1
^
gN
a
o
s
v
O
0.-
G
u
Co. =
N
fin
p
c
N O T
u
0
p ^
^
N_
a
r
S
s
D
0
a
0r
u
u
46
3
zat
0u
Ln
^,
i
o
r=
Y
0
to
.
t
the
observations were compiled. Because
produces
highly
serrated
histograms presented
in
8-bit conversion
histograms which
of the FSS
data
disrupts visual inspection, the
Figure 18 have been smoothed using a 3 count window
moving average.
The FSS histograms
in Figure
18 and the NS001 histograms in Figure 17 are
quite similar in appearance. The mean and variance behavior of the corn and
soybeans canopies display the same patterns in
each data
set.
Note in
particular, that the mid-infrared band distribution shows the same
comparatively low within-crop variance when compared with the near-infrared
band in the FSS data as
w?as noted
in the NS001 data.
Temporal Analysis of FSS Data
The comparison of the FSS and NS001 8/30/79 data indicates that they
provide common measures of the reflectance behavior of corn and soybeans
canopies. The 8/30/79 data suggests that TM mid-IR observations may improve
corn/soybeans discrimination on this data. However, analysis of current
Landsat data shows that at this time of the growing season, after corn has
tasselled,
corn and soybeans are not difficult
to
separate using the visible
and near-IR bands.
FSS single scan observations for 6/11, 6/29, 7/16/79 were also _processed
to construct histograms of the corn and soybeans reflective patterns. Figures
19-22 present the histo g rams of corn and soybeans FSS observations on these
dates and the 8/30/79 observations. Table 9 provides the percent ground cover
observed by the field enumerators on these dates.
The histograms reveal a particularly interesting pattern of mid-infrared
relectance for corn and soybeans when
compared to
near-infrared reflectance:
On 6/11/79 the FSS observes primarily bare soil and there is little difference
in the comparative reflectance of corn and soybean fields
4 7'
in
any of the bands.
ORIG ►AL PAGIZ 13
OF
POOR QUALITY
N
N
r
N
A
^
r
CE,
^
M
►
e
N
t/)
I ""
Y
a
r
(D
lla
U..
•
A
O
a
1
O
^
•
^
N
A_
i
Z
10
S
W
r,
O
iA
r
O
s
•
N
N
s•
N
r'
o
N
M
W
ca
r
p
N
!
Q
t
t
s
u
f-I
^
P
^
W
N
r
N
:
M
r
^
C)
V)
=
O
u
to
V)
LL.
tT
i
C:)
M
co
N
r
^
t
r
p
J
^
N
N
N
a
r ^
ti
O
^
o
s
E^
^
O
Li
^
O
W
Ctm
h
O
N
r
S
o
^
^
u
I-
N
s
^c
a
H
cc
N
z
p O
i I1
y ^
•
D
^
W ^ ^
J
^
OI
^
O
^O
N u e
N ^
r W a
^ N p
^ = a
M •
Z „
u a.
41
u
48
7
€ r POOR QUALITY
TABLE 9
PERCENT 6ROUHD COVER FOR CORN
AND SOYBEAN
FIELDS IN 1979
a
CORM
PERCEPT
COVER
0 — 10
11 - 20
21-30
31 — 40
41 - 50
51 — 60
61-70
71 — 00 0
91 - 90
91 — 100
6/11
17
21
1
6/29
SOY
7/16
8/30
6/29
7/16
8/31
30
6
2
6
13
9
5
2
36
38
n
49
i
6/11
7
10
3
1
1
1
1
3
6
12
7
3
33
r,.
CROP HISTOGRAMS FOR TM SIMULATED FROM FSS
,AgRISTARS ITS WEBSTER CO. IOWA 1979 CROPPING SEASON
TM BAND 2 (0.52-0.644)
COIN
S0Y
ORIGINAL PAG*» 1
OF POOR QUALITY
JUNE 29
JUNE 11
AUGUST 30
JULY 16
FIGURE 19
50
J
CROP HISTOGRAMS FOR TM SIMULATED FROM FSS
ApRISTARS ITS WEBSTER CO. IOWA 1979 CROPPING SEASON
TM BAND 3 (0.63-0.69pm)
CORN
ORIGINAL PA01, I
soy
OF POOR QUALITY
JUNE it
JUNE 29
JULY 16
AUGUST 30
a
FIGURE 20
51
CROP HISTOGRAMS FOR TM SIMULATED FROM FSS
AiRISTARS ITS WE13STER CO, IOWA 1979 CROPPING SEASON
TM BAND 4 (0.76 - 0.90sm)
CORN
soy
JUNE 11
JUNE 29
JULY 16
AUGUST 30
FIGURE 21
52
ORIGINAL PAGF,4' IS
OF POORUAL
Qf Ty
rCROP HISTOGRAMS FOR TM SIMULATED FROM FSS
'AgRISTARS ITS WEBSTER CO. IOWA 1979 CROPPING SEASON
TM BAND S (1.55-135pm)
CORN--SOY---.
JUNE 11
JUNE 29
ORIGINAL PAGE IS
OF POOR QUALITY
AUGUST 30
JULY 16
I
FIGURE 22
53
I
/
Conclusions
The GISS/Columbia Earth Resources group has been intensively involved,
during 1981,with activities directed at simulating TM observations and
evaluating the potential information content of these observations. The
NASA/AgRISTARS field measurements program continues to provide extensive data
to conduct this research. The GISS/Columbia staff have developed processing
and analysis techniques to utilize this data base. The 1981 research program
has successfully addressed several outstanding NASA research objectives
concerned with preparation for the analysis of satellite acquired TM
observations. The more extensive field observations collected at the
AgRISTARS supersites during the 1980 and 1981 growing season present a major
analysis challenge. The experience gained by the GISS/ Columbia staff in
analysis if the 1979 AgRISTARS data during this year will be invaluable to
exploitation of the more recent field measurements and in analysis of the
soon-to-be acquired satellite TM observations.
C. Analysis of Snow Cover and Wetlands from Landsat Data
The GISS/Columbia group,, in cooperative research program with U.S. Army
Corps of Engineers applications scientists at the Cold Regions Research
Laboratory, Hanover, New Hampshire, have, during 1981, carried out analysis of
selected La.ndsat scenes. This research objective is to determine the utility
of Landsat numerical observations for mapping and analysis of snow cover and
wetlands. Michael hiller, a graduate student of the Department of Geography,
has served as senior analyst for these activities. Drs. Goward and Ungar of
the GISS/Columbia group provide research direction in consultation with CRREL
p
^a
scientists, Carolyn Merry, Dr. Jerry Brown and Dr. Ike McKim. The following
reports describe ` , he research activities.
54
NIL—
j
q^
k
i
fi
By 6/29/79, corn fields average 50% ground cover and soybeans 35% ground
cover.
In the mid-infrared corn fields are less reflective than the soybeans
fields whereas in the near-IR the reverse is true.
On 7/16/79 corn canopies
are reported to have 100% ground cover whereas the soybeans fields average 65%
but are highly variable.
The mid-IR data show that corn is still
relective than soybeans and better separated than on 6/29/79.
j
less
However, in the
near-IR the soybeans reflectance is highly variable and mixed with the corn
reflectance.
By 8/30/79 both crops are at 100% ground cover and soybeans are
more reflective than corn in both TM bands. These histograms suggest that the
i
TM mid-IR observations will significantly improve early season corn/soybeans
discrimination when compared to currant Landsat observations.
Di scussi on
This investigation of the mid-IR reflectance behavior of corn and soybeans
k
canopies suggests that TM observation in this wavelength interval will improve
both early season and through-the-season discrimination of corn and soybeans.
k
These conclusions are drawn, however, from one growing season in one location.
a'
u
a
Further analysis of observational data will
be required to establish the
general applicability of these findings.
In depth evaluation of these results will
physical processes that lead to differential
require examination of the
reflectance behavior of
c
^
corn/soybeans canopies in the mid-IR.
Interactions between leaf reflectance
a
properties, canopy architecture and background soil
z.
need to be examined.
3
Although further empirical studies may be able to confirm the behavior
•
j
observed in this study, prediction of where and under what conditions this
t
z
behavior will be observed will require physical
understanding.
E
a
55
f
ORIGINAL PAGE H
OF POOR QUALIT
Progress Report: Remote Sensing of Snow
Michael S. Miller
July - December, 1980
,S^t,wtS q^Z
Introduction: Landsat is being investigated in conjunction with
A
ground snow course measurements to determine the utility of the
remotely sensed data for snow cover evaluation. Although dptogd
qV'^.
satellite data
iu nR proven useful for estimating areal extent
of snow, applicability in forested hydrologic basins, and for other
descriptors of snow cover (e.g. depth, water equivalent) are less
obvious.
Study Site: Allagash, St. John Rivers confluence area, Maine. Land
Cover is a mix of deciduous, coniferous, and mixed forests, clearings,
agriculture, and water surfaces.
Achievements (July - Dec., 1980)
i
1.
Using a May 31, 19078 Landsat image, the 300 x 300 pixel study
area
(
153 sq. miles) was classified using the CISS - MAP1 classification
routine. Results, in percent cover of the area,ar e:
Mixed Forest
Clearings
Water
2.'L%
Deciduous Forest 16.3%
28.2%
Coniferous
2.
Agriculture
52.%
.s%o
.5%
Using these percents as weighting factors for ground snow depth
f
measurements, average snow depths for the study region were determined
for four days for which good quality
Landsat
i
`
February 11, 1978
29.01"
January 11, 1979
December 24, 1978
23.63"
20.94"
January
17.06"
6, 1978
56
coverage is available:
t
i
ORIGINAL PAI5 I'S
OF pOOR QUALITY
3. Landsat MSS digital data have been investigated to determine whether
there is a crrelation between snow depth and the measured intensities
in the four spectral bands. All sites were thus evaluated. Band
ratios and summations were considered, as was the inclusion of a sun
elevation factor to compensate for differences in incoming radiation
on different days. In most cases (see Figure one a-d) little
correlation between snow aepth and Landsat counts/energies was noted.
4.
Although radiances measured in reflected visible and near-infrared
wavelengths within an individual pixel do not increase (or decrease)
with increased snow depths beyond several inches, it is believed that
the regional landscape does change. In the northeast forests, the
hav u%, 0ujr%
vegetation canopy is
Mixtures of clearings and
coniferous and deciduous forests at various stages of growth allow
upper canopies and understories of differing heights and densities.
As snow accumulates, different levels of vegetation would be expected
to be covered with snow. Figure two a-d show band 7 grayscales for the
300X300 pixel Allagash area for four days. As is seen, increasing
snow depth results in increasing 'whiteness' of the area.
5. Histograms of the energies have been investigated to quantify this
change (figure 3 ^. A no snow scene in autumn shows a band 7 histogram
that is unimodal and of low variance (3-a). Ms snow depth increases
(3-b to 3-e), there is a decreasing in peakedness and increasing of
skewness toward the bright end of the distribution. This indicates
57
-^r—
r
that as snow depth in a diverse land ccver increases, increasingly
more pixels will become 'snow pixels', migrating in the regional
histogram from the apparantly snowless vegetation peak to the
brighter (snow) intensities.
Projected work (4an. - June 1981)
Examination of gray scales and histograms for additional snow
scenes of the Allagash/St. g'ohn area will continue. This is to
dtermine the validity of the preliminary observations.
r
In addition, gray scales will be examined in conjunction with
the derived classification map to determine which vegetation types
contribute to increased brightness as snow depth increases.
Similar examination will be initiated in the Danville, Vermont
area in the Sleepers River. Basin. Records of snow depth for a
variety of land cover types are available for this intensively monitored
basin in northeastern Vermont.
Problems encountered to date: The major obstacle to this research into
snow cover is the low number of available Landsat scenes for this region
during snow seasons. This is due primarily to the satellite's cycle
and to the frequency of cloud cover, particularly in the important
snow melt period.
l
!Y
:Y
58
Y
OF r,)t)ri QUALITY
Figure one. Snow Course and Landsat MSS data for four sites.
A.
3.0
N
V
AA
E I.S
'EW
1.4
E
914
W
2
W
MSS BAND
B.
SITE II HARDWOODS
NORTH 130011.3-4% SLOPE
FOUR SAND/ flllwEO
I
3.0
11.6
N
2.2
RATIO: SAMO 1
"NO 5
I
1.0
'EW
E 1.4
W
2
W
1.0
It/N
n n • MT[II
j
.-
4
ORIGINAL FA. i;3
OF rOOR QUALITY
Figure one, continued.
c.
SITE 12 SOFTwOOD
LEVEL 1140 ft. SHELTERED
FOUR 861110• eum" 0
10
c
M
RATIO SA W ►
./u
14
we
E
U
E
I
M
O•NOw ^/TMIrGIt•1
IA
r
• MT[^
W .^
or 1.0
[OU1V^L[rT
W
2
1 ••CNl [ 1
^
W .e
4
d
6
7
Y• LA N
ri••
MSS BAND
D.
SITE 13 CLEARED
LEVEL 880 ft STEEP SLOPE RISES
DIRECTLY EAST
F" "BIOS VJMW D
m
•
5.2
r
.n
44
'M
UN,
36
'E
u
31r !.8
an
r
c^
N
0C 24
W
2
W
r /e•
1.2
• allow arT" MCM 111 .
•rT[ n
awwAL[NT
Iwc»tn
•
n•
4
6
6
MSS BAND
60
T
l^ ^^ v^ rn•
ORIGINAL PAGE IS
OF POOR QUALITY
Figure two, continued.
C. January 11, 1979
Average snow depth: 23.63"
IT
A
i
^.. Fry
D. February 11, 1978
Average snow depth: 29.01"
61
ORIGINAL FACE IS
OF POOR QUALITY
Figure two. *,andsat MSS
Band
7, gray scales.
A. January 6, 1976
Average snow depth: 17.06"
B. December 24
Average sna0
1
1
11
1
1
i
ORIGINAL FATE IS
o
OF POOR QUALITY
1
t7
0°
OD
go
gN
o
—
v
N
0
o
o
0
C:)
V
0
_•
_
00
Oa
o
ti
O[^ o
^
G
r-,
-----
z
m
0
o
o
-----^
m
Q
O
W
m
O
r-- -- p_
1O
(10
W
m
Q
L)
'
z
0o No
Q
Cn
Q
CD
0
0
o'
0
z
°2
c1
Q
O
FV)
2
o
c^
0
ltJ
i o
1
o
--'
z
aN
c°
u
Mo
CD
i
I.O)
i
tN
oo
rn
—
c----
G^
D
o
^— -
C=)
4)
p
C
^'
0
^
Io
O
^.
^•
O
^o
O
O
O
O
O
o
N
O
O
O
O
O
An
o
W)
—
o
ADN3n0363
O
O
63
zD
U
^______^
m
D
z
O
N
O
O
0
—
O
O
O
O
O
O
O
U
V
USE OF LANDSAT DIGITAL DATA IN SNOW COVER MAPPING
FOR THE UPPER SAINT JOHN RIVER BASIN, MAINE
Carolyn J. Merry
U.S. Amy Cold Regions Research and Engineering laboratory
72 Lyme Road
Hanover, New Hampshire 03755
ORIGINAL PAGE IS
OF POOR QUALIT Y
Michael Miller
NASA Goddard Institute for Space Studies
2880 Broadway
New York, New York 10025
ABSTRACT
Each spring the SSARR (Streamflow Synthesis and Reservoir Regulation)
model is used ivt the Upper Saint John River Basin, Maine, to forecast runoff
due to snowmelt and precipitation. One of the main inputs to the model is
the extent and distribution of the snowpack. Using point snow course measurements to determine the extent and distribution of the snowpack is a commonly
recognized problem. A technique for using Landsat digital data to map the
distribution of snow cover/land use categories has been evaluated for use in
the SSARR model to improve the accuracy of spring runoff predictions.
Landsat MSS digital data has been investigated to determine whether there
is a correlation between snow depth and the measured intensities in the four
spectral bands. Band ratios and summations of the four bands were considered,
as was the inclusion of a uun elevation factor to account for differences in
incoming radiation on different days. A direct correlation between snow depth
and Landsat counts/energies does not occur.
Although the Landsat radiances measured in the reflected visible and near
infrared wavelengths fora single pixel do not increase (or decrease) uniformly
with increased snow depth beyond several inches, it is believed that Landsat
radiance values on a regional scale do change. Histograms of the Landsat radiance
values for a 300 x 300 pixel (153 sq. mi.) study area have been investigated to
quantify the change. A histogram for a no snow scene in the late autumn (27 October
1978) exhibits a E,nimodal distribution with low variance. As the snow depth
increases over the total landscape, there is a decrease in peakedness and a
corresponding increase of skewness toward the bright end of the Landsat distribution of radiance. This fact indicates that as snow depth in a landscape
increases, additional pixels become 'snow pixels', migrating in the regional
histogram from the apparently snowless vegetation peak to the brighter 'snow'
intensities.
Various indices of skewness and peakedness (kurtosis) have been examined to
quantify the above observation. An index of potential use is defined as:
x .6
a 3 * pop(total) *COS( c
pop(mode)
64
F
:
.
ORIGINAL PAGE IS
Up pOoR QUALITY
In which I s is the index of anowness, • 3 is the moment coeffient of skewness,
pop (mode) is the pixel population of the histogram mode, pop (total) is the
total pixel population and COS(** is the cosine of the solar seo*" angle.
Through linear regression techniques, it was determined for t4e ` 360 x 300 pixel
Landsat area that:
Snow Depth - 2.079*(Xa) -1.111
in which snow depth is measured in inches. The index of snowness, Is, was
found to predict mean snow depth with a standard error of estimate of ± 4.2
inches at a 95% confidence level. For different mixtures of canopy cover,
the slope and y-intercept Would vary.
As a preliminary test of this index of snowness,two
winter Landsat passes from previous years were examined for
the Allagash - St. Oohn region. Using the index and equation
of regressionl mean snow depth for February 11, 1973 and
April 19, 1974 were eatimated at, respectively, 35.49" and
15.89", ±4. =44 ". During those years, ground snow course measurements
were not available for the Allagash stations. Closest measurements
were taken at XX# Connors, New Brunswick, 12 miles east on the
St. John River. Snow depths at Connors were:
February 1, 1973
February 19, 1973
April 16, 1974
May 7, 1974
33.3"
37.9"
23.7"
11.31'
II. Bibliographic search for literature concerning the
remote sensing of snow pack characteristics has continued
during the report period.
111. Visit was made on March 25 to the Sleeper's River
Research Watershed. Snow courses, meteorological stations and
streamflow measurement sites were examined in anticipation of
studying the land cover snow pack relationship for this basin
in Vermont; Search of Lander: scenes is underway.
65
_.
l
a
r
PROGRES §„ REPORT
Pro e t: Simeon, Nebraska - Wetlands Classification
lr or kgriod: September/October, 1980
Qbiectivet To classify Landsat MSS digital data for June 12, 1978
(scene-i.d. 3099-16494) for Simeon (Nebr.) SW and SE Quadrangles.
This area lies within the Valentine National Wildlife Refuge.
Three cover types are classified:
subirrigated meadow. Unclassified
open water,
areas are
marsh,
'classed' as
uplands. This classification is performed for the
Experiment Station, Army Corps of
and
Waterways
Engineers,
Procedure: 1. The MAP1 classification algorithm developed by
Dr. S. Ungar of the NASA Goddard Institute for Space Studies
is used in this ,project. A description of this algorithm is
provided in the appendix to this report.
In MAP1, the digitized Landsat counts are converted to
energy values (mw cm-2 sr
a signature, or group
of
-1,
).
For each classified cover type
signatures is required. The area
within the Landsat multispectral space surrounding each signature
and classified to its cover type is determined by user specified
parameters:,
M AX
i
and WHRT (maximum angular dolts, and brightness
weighting) .
2.
Signatures were
selected by the evaluation of test sites
a
g
66.
.
.
within the larger study area. Six to eight test sites
selected from
the quadrangles for each of
were
the three cover types.
This was accomplished by using the wetlands map overlay (scale
,1:24,600) provided by the Omaha District, Corps of Engineers.
The overlay was superimposed onto a MSS band 7 grayscale printout.
Test sites were delineated for
photoquad for reference. The
each
cover type using the ortho-
test sites ranged from
5 hectares
(12 acres) to 40 hectares (100 acres). To maximize 'purity'
of cover type, the
larger
sites were selected
from the interior of
homogeneous areas.
3. Unsupervised classifications were run
to distinguish distinct
o::nrgies
classes within each test site,. Printouts of the Landsat
were used
to
derive signatures for each
classifications it
necessary
was determined
to most accurately
class. From the
that two signatures
were
describe each of the cover types.
signature( in energy:
Cover Type
unsupervised
MW cm
-2 sr-1u-
Band_4_ _ Band_5" _`sInd_6_ _ Band
.08
.20
.09
.21
.26
.37
1.17
.47
.32
.50
1.45
.47
.46
.32
.58
.61
1.73.
1.er
1. Water
(a)
(b)
.29
.42
.15
.25
2. Marsh
(a)
(b)
.41
(a)
(b)
3. Subirrigated
Meadow
.31
67
•
.
.
4.
The signatures derived in the unsupervised runs were
used in the supervised procedure for a ;Larger area approximating
the Simeon SW Quadrangle. The supervised classifications were
run with a variety of A MAX / WBRl
the set
of pixele
combinations, thus
altering
classed within each cover type. Considerable
visual agreement between the classification and wetland map was
achieved after fifteen runs. Refinements in agreement resulted
after ten additional runs in which signatures were adjusted and
rearrangements made in the order in which the signatures were
assessed for each pixel (:Tote: order is of importance in MAP1,
as the algorithm is deterministic rather than-probablistic).
It was found possible to eliminate the need for one of the water
signatures without a reduction in classification accuracy.
5. In conversation (mid-September) with Horton Struve of the
Waterways Experiment Station, it was decided to extend the
classification to include the Simeon SE Quadrangle. Land cover
within the two Quadrangles is very similar. Nevertheless,
several additional, runs were required to take into account some
spectral variations across the region.
6.
Final signatures, in order of their evaluation within
MAP1 are:
r
t
4
Signature_ _
Cover 'Pype
4
1.
2.
3.
4.
5
6
WBRT
7
47
Marsh
Sub. Meadow.47
Water
.42
Sub. Meadow.40
.32
.32
.25
.30
.50 1.38 .041
.58 1.71 .08
.20 .21 .52
.58 1.80 .12
.40
.32
.30
5. Marsh
.1
.1
.4
.1
.88 .12
.1
7. Percent cover was evaluated for each of the following
classes: open~ water, wetlands (marsh + subirrigated meadow),
and ups-nds for the entire Simeon SE, SW area. In the table
below, these are compared, to the corresponding figures for the
Simeon SW area use.5 at the Waterways Experiment Station for a
ground truth standard. Comparable figures for the entire a^rea
were not available.
Cover Type
GroMnd-Truth
Water
Wetland
Marsh
8.0•%
8.8%
35.7%
32.6%
4.6%
55,5%
59.4%
Difference
.8
3.1
28.0°jo
Subirrigated Meadow
Upland
MSS
3.9
B. A tape of the digitized final classification is being
prepared for Horton Struve, Waterways Experiment .Station.
69
r
E
The following description of the GISS MAP1 classification
algorithm is taken from.
Merry, C . J. et al., Preliminary analysis of water
equivalent/snow characteristics using Landsat digital
processing techniques, 1977, Cold Regions Research and
Engineering Laboratory, Hanover, NH.
The LATDSAT MSS observation may be thought c,f as a point in a fourdimensional "color" space, where the values along each axis represent
the radiant energy received by the satellite in one of the four ba.ds
(this is illustrated for three bands in Fig. 2a). Observations lying in
a similar direction from the origin in this four-dimens ixonal color space
4W
s ^
'r
^v
are said,to be similar in color regardless of thoir total radiant
energy. The distance of an observation from the origin is a measure of
the total radiance associated with that data point. The algorithm is
p rimarily designed to combine observations similar in color into the
same classification category. There is provision for evaluating brightness
differences and for weighting these differences in with the color discrimination when constructing the classification categories.
Discrimination based solely on color is obtained when one examines
the difference in direction between the color vectors (observations).
If the angle between 'the observations is smaller than some user-defined
criterion the vectors are considered to be lying in the same direction
and, therefore, the observations are placed in the same category.
There are two modes in which this classification scheme may be
employed, supervised and unsupervised. In the supervised mode the user
s p ecifies a signature (the energy, distribution in the four LANDgAT bands).
I° an observation lies within an angle smaller than the user-defined
criterion, dm a x, it is said to belong to the category represented by the
signature (Fig. 2b). Therefore, all vectors lying within a cone of
angle, dmax, about the signature representing category X belong to
category X.
In the unsupervised mode the color vector corresponding to the first
observation is compared with all subsequent observations. If color vector
1 is similex in direction to color vector 2 (i.e. de < 6 X ), observation 2
is placed rn the same category as the first observation TMig. 2c). In a
similar .fashion observations subsequent to observation 2 are compared to
the second observation and so on rigs,t up to the last observation. If in
the process of constructing categories, a member is found which belongs
to a previous category, the new category is chained to the original
classification category forming one ,joint category. In effect the unsupervised classification will farm several categories based on a criterion
specifying msaimum colter .difference permissible between members of the
sate category.
In addition tc discrimination based solely on color, the GI22 algo-"ithm provides the capability of weighting total radiance (aZbedc) differences into the discriminant. The percent albedo difference between
two observations is computed. This normalized albedo difference is then
combined with the color difference angle (expressed in radians) by per-
f'or,^. ng .a weighted average in the RMS (root mean square) sense. This
abedowweir.ted quantity is ' n ow compared with the user-defined criterion.
A relative v small albedo weighting allows very la.ge albedo differences
to disqual-ify observations that are similar in color from membership in
the s -n-me category, thereby adding a second level of discrimination.
a
ORIGINAL PAGE IS
OF POOR QUALITY
71
Mi
ORIGINAL PAGE 1
Bond 3
OF POOR QUALITY
e
j
I
/
f
^
1
1
1
I
^
fond I
I
1
^
f ^'
1
I
I
'
I
^
62- - - - - - -
1
Bond 2
B, . Emirp In Band
I
4
B, "' Total Radiance
8 To1^I
a. A
co lor v ect or
which is i 1 us-, 'ate`1 --%r
three bands.
Bond 3
blegory X
Sand 1
C-4ner,risea mcde.
defines category X ab-- ,--, the
Any color vector that lies within this cone beor.zs to
The user-defined criterion., 6.,
specifiel siF;ature,
category X.
Sand ,
V
jory 2
Oand t
land 2
W
%
:unervised mode. B j is similar in direction to SI (bA t d X ) and placed in category
1. 34 is similar in airection to S2 and placed in category 2. However, 14 is also
similar in direction to 7 3 (category 1). Therefore, category 1 is merged with category 2.
Fig,are 2. Concept of the four-dimensional "color" space, represented in three bands.
4
72
r
f^
t
i
a
Project: Burlington, Vermont Quai
Objective: To map potential and definite wetlands; evaluate
seasonal images.
Procedure: Two Landsat images of the Burlington area were
classified using the GISS MAP1 classification package
The dates of the two passes were 13 April 1977 (SCNID
20812-14372) and 26 June 1975 (SCNID 5068-14433).
The first stage in the classification of the two dates was
the development of a standard routine for mapping potential
wetlands. Potential wetlands are here defined as that set of
pimein in the scene which include all wetlands. That is, errors
of omission are minimized. However, there may be large errors
of comission. The second stage was to perform a second classifi-
cation on these potential wetlands in which only defitr,ite areas
of wetlands are identified; errors of comission are minimized.
In this second stage, areas of open water, emergent vegetation,
and forest/shrub-scrub wetlands were segregated. The two classifications used in conjunction may give a good indication on the
locations of the wetlands.
13 April 1977
1) After examining numerous classification possibilities for
potential wetlands, the approach finally selected was a one
signature, two-band classification using MSS-4 and -5. In the
unsupervi.^ed mode of MAP1, pixels corresponding to open water
were found to cluster around a signature of 0.44, 0.25, 0.00,
0.00 when weighting of bands equalled 1.0, 1.0, 0.0, 0.0. This
signature was then applied to the supervised mode with a range of
increasing delmax values. The final namelist parameters selected
73
poi s
s
}R
for this classification were SIG 0.44, 0.225, 0.00, 0.001
nn
KG
1.0, 1.0, 0.0, 0.0.
DELMAX - 0.215; WBRT - 0.Sr WBND
2) To classify definite wetlands three classes were usero,,, open
water, emergent vegetation and forest/shrub-scrub wetlands. All
areas identified as potential wetlands in step 1 were classified
using a one-band (MSS-7) classification. The namelist parameters
used were: a) water: SIG
0.00, 0.00, 0.00, 0.17; DELMAX - 0.7;'
4r'
WBRT - 1.0; WEND - 0.0, 0 . 0, 0.0, 1.0; b) emergents: SIG 0.00 1 0 . 00, 0.00, 0 . 53; DELMAX - 0.25 WBRT 6 1.0; WEND - 0.0, 0.0,
0.0, 1.0; c) forest - shrub-scrub: SIG = 0.00, 0.00, 0.00, 0.55;
DELMAX
=
0.444.
26 June 1975
3) Using the same procedure as in (1), MSS-4, -5 signature of
water (0.40, 0 . 15, 0.00, 0.00) was applied in the supervised mode
with DELMAX = 0.3, WBRT = 0.5, and WBND = 1.0, 1.0, 0.0, 0.0.
4) The nature of the landscape in June made it possible to
successfully classify only open water of the three wetland
classes. In these cases a four-band classification most accurately
identified water: SIG
WBRT = 0.1 WBND
0.40, 0.151 0.10, 0.11; DELHAX
0.4;
1.0, 1.0 1 1.0, 1.0.
ti
Comments
1) Seasonal variations are highlighted here. Spring images are
best for identifying wetlands
i
n the Northeast - water table
heights are highest in this season. Additionally, deciduous trees
have not leafed - out, allowing surface conditions to be viewed.
j^
I
74
L_
ORIGINAL PAGE- )
OF POOR QUALIT Y
;.
During the summer, canopy closure is more complete and the soil
is mostly dry.
'`,
3) The class of forest/shrub-scrub wetlands in the classification
includes deciduous species. Wet evergreen woodlands cannot be
separated from dry evergreen woodlands in either season.
3) The category of intermittent exposed streams as included in the
`
^ss f
ground truth map of wetlands cannot be identified with Landsat
because of the features' small size.
The Tape
The tape is 9-track, 800 B.P.I., and contains 5 files
separated by End-of-File marks. All records are 240 bytes in
length. The first file contains a single " header" record in
EBCDIC format which describes the subsequent files' characteristics.
Files 2, 3, 4, and 5 contain the classification results stored as
EBCDIC symbols. Each file contains 224 records. File 2 contains
potential wetlands classification for April 13, 1977. File 3
contains definite wetlands classification for April 13, 1977.
File 4 contains potential wetlands classification for June 26,
1975. File 5 contains definite wetlands classification on June 26,
1975.
For the potential wetlands classifications ( files 2 and 4)
the symbol ' 1' represents potential wetlands; '-' represents nonpotential wetlands.
For the classification of definite wetlands on April 13 (file
3) 0 '-' represents non-wetland, '1' represents open water, 12'
represents emergents, and 0 3' represents forest/shrub-scrub wetlands.
For definite wetlands on June ;26 (file 5)• only open water
is identified as 1 1 1 .
Non-wetlands are represented by
75
r
Conclusions
These investigations have contributed to the development of U.S.
Army
Corps of Engineers applications of current Landsat data. Highlights of the studies
include the GISS algorithm performing the best
is a comparative study of
classification algorithms on the Nebraska Wetlands study and the innovation of
a new analysis concept for
snow cover applications. This cooperative effort
has proven highly productive and will continue
during 1982 with emphasis being
placed on refinement of the snow cover analysis techngiues.
D. Agro- Environmental Application of Satellite Remote Sensor Data
During 1981 GISS/Columbia staff
members continued pursuit of additional
applications of satellite remote sensor data to selected projects of personal
interest and that support their educational and professional objectives. In
general, these studies address specific questions of interest to geographers
which will also contribute to increased knowledge of numerical remote sensor
data. Presentation of this research has occurred at regional and national
meetings of the Association of American Geographers as well as at the
NASA/NOAA cosponsored Remote Sensing Educator's Conference (GORSE-81) held at
Purdue University (Appendix A).
Landscape "Greenness"
Helene Wilson has proposed that Landsat observations provide a unique and
universal measure of landscape vegetated condition, called landscape
"greenness". Considerable research has been conducted in agricultural
applications of Landsat data that shows a strong relation between various
attributes of crop canopies (e.g., percent ground cover, leaf area index,
biomass) and various transforms of the Landsat visible and near-infrared
observations. Although it has not been proven, this approach to Landsat data
a
would appear to be generally applicable to all landscapes. That is, the
degree to which a landscape is vegetated should be related to its "greenness"
76
t
.
.
as observed by one or more vegetation indices derived from Landsat
observations.
Helene has selected a study site in the Hartford, Conn. region, where a
complex variety of land cover conditions including forest, agriculture, and
urban are present. Her hypothesis is that regardless of the cover type, one
or more of the Landsat-derived vegetation indices will indicate the degree to
which the landscape is occupied by photosynthesizing vegetation. If this is
true then Landsat data can be used as a universal measure of landscape
conditions where the vegetation "greenness" may serve as an indicator of
current environmental conditions.
During 1981 Helene acquired low altitude color infrared stereo photography
and conducted a field measurements program to evaluate the relation between
ground conditions and the air photo observations. The aerial photography will
provide Landsat subpixel measures of percent area covered with living
vegetation and vegetation height. Through intensive analysis of the
photography she will establish the vegetated conditions within Landsat
observations of various land cover conditions. Through comparison of the
photo-derived observations and the Landsat Data, she will test the research
hypothesis.
Albedo Studies
Chris Brest, during 1979 and 1980, collected field
measurements of the
reflectance of building roof tops and parking lots in the Hartford, Conn.
region. These observations are being used to calibrate a set of 28 Landsat
observations of the Hartford region acquired between 1972 and 1978 and
covering eleven months of the year. Calibration of these
data to
percent
reflectance and the combination of Landsat visible and near-infrared
measurements provides a good estimate of surface albedo. This further permits
77
.
.
analysis of urban-rural
albedo differences as a function of season and cover
type. Albedo has been suggested as a
significant factor in causing
urban/rural climate differences. Chris's analysis will evaluate this
hypothesis.
A preliminary analysis of the data (Appendix A) has confirmed that the
calibration procedure is effective (error rate of 2-1%) and that
urban-rural
albedo differences may be smaller (approximately 5%) than is frequently
reported. This appears to be related to the inverse variation of
visible/near-infrared
reflectance through the
season; a phenomenon generally
not investigated in albedo studies. During 1981 Chris completed analysis of
the entire 23 Landsat observations and is currently drafting his research
results for submission to the Department of Geography faculty as his doctoral
dissertation.
Michael Miller has approached derivation of surface
albedos from Landsat
data through numerical modelling of atmospheric radiative transfer. By
calculating the effects of scattering, as described in the U.S. Air Force
Cambridge Laboratories research, and absorption, as described in the
Smithsonian Tables, he evaluates the relation between ground reflectance and
observed Landsat radiance. Seasonal variations in atmospheric conditions are
considered but not short term meteorological conditions.
Michael has applied the model to Landsat observations of eastern Long
Island, N.Y. where diverse land cover conditions present a variety of surface
albedos. The results, although promising, suggested that his method for
handling atmospheric absorption of radiation is less realistic than desirable.
The albedos calculated are the right order of magnitude but appear to be
overestimated. Using only the scattering model he derives more reasonable
78
k
F
4
figures.
He is
currently drafting his research results for submission, as his
master's thesis, to the Department of Geography.
Kenyan Agricultural
Systems
Tina Cary is conducting a study of agricultural systems in western Kenya
(east Africa).
variations in
Her hypothesis is the Landsat observations record regional
the agricultural systems
present.
that, because of striking regional variations
development and regional
variations in
a
in the intensity of agricultural
and interaction between cultural and
agricultural practices, Landsat
physical factors that effect local
observations
in particular, she proposes
may be used as an indicator
of the agricultural
system present in
location.
To identify the relation between regional variations of agricultural
systems and Landsat observations she will examine the numerical properties of
spatially aggregated Landsat data.
Through analysis of the data at various
levels of aggregation she will seek to isolate those scales (or resolutions)
at which various cultural
landscape appearance.
and/or
physical
factors are most
highly related
S
to
Tina will use this information to develop a systematic
approach to landscape regionalization which will
regional patterns of agricultural
permit analysis of the
systems.
1
a
During 1982 Tina will travel to Kenya, under a Fullbright scholarship, to
conduct field observations to support her analysis.
She will
be
in
the field
for one year traversing the 10,000 square mile area that represents her study
site.
The study site consists of the nominal
ground coverage of one Landsat
scene and thus represents an ambitious field project.
field experience in Kenya
should be
with Dr.
highly successful
in
Because of her previous
Frank Conant of Hunter College, Tina
accomplishing
her
objectives.`
79
i
4
.
2i
e
Conclusions
The diverse research projects
agro-environmental research provide major contributions to the GISS/Columbia
research program with a minimum cost to NASA. The activities in general, do
not draw heavily of the group resources and often, as in Tina's case, are
significantly supported by other funding sources. However, the outcome of the
research significantly contributes to advancing understanding of satellite
earth observations. The success of these projects are dependent on the
working relation between CISS and Columbia scientists and demonstrate the
productive nature of the cooperative agreement.
80
References Cited
AgRISTARS, 1981. AgRISTARS Annual Report - Fiscal Year 1980, NASA, Johnson
Space Center, Hous o , Tf'exas.
Baker, J.R. and E.M. Mikhail, 1971,, Geometric Analysis and Restitution of
Digital Multis pectral Scanner Data` Arra s, LARS Inf6Ra-TioFW6`t
e05_Z875,
ur de UniversRY, West afaay ette, ndiana.
Barrnett, T,L „ 1980, "Recalibration of FSS Data", NASA-JSC Memorandum to
D. Pitts. Johnson Space Center, Houston, Texas,
"Design,
Implementation, and Results of LACIE Field Research", Proceedin g s of
Bauer, M.E,, M.C. McEwen, W,A. Malila, and J.C. Harlan, 1978.
the LACIE Symposium Vol, II. Johnson Space Center, Houston, - Teiex
-Oct, MR, 197 ; pp. 1037-1066,
Bieh1, L.L „ 1979. "Algorithm for Detectin Scans for Field Average Version
of NASA/JSC Field Spectrometer System ?FSS) Data, Memorandum dated 8/22/79
from LARS-Purdue University, West Lafayette, Indiana.
Engel, J,L,, 1980. "Thematic Mapper ¢ An Interim Report of Anticipated
Performance" Enclosure 5 in NASA Applications Notice "Landsat-D Image
Datr Quality Analysis Program 10/23/81, Goddard Space Flight Center,
Greenbelt, Maryland,
Hyde, R.F,, S.N. Goward and P.W. Mausel, 1977, "ISURSL Levels °Classification:
A Low Cost Approach to M:oltispectral Data Analysis", Proceedings of the
1977 Machine Processing of Remotely Sensed Data Sympos um,—Purdue University,
LARS (Laboratory for Agricultural Remote Sensing), 1968. "Remote Multispectral
Sensing in Agriculture," Vol. 3y Research Bulletin No. 844, Agricultural
Experiment Station, Purdue University, West Lafayette, Indiana, Cited
in P.W, Swain and S.M. Davis, editors, 1978, Remote Sensing the quantitative
Approach, McGraw-Hill, New York, NY, p. 14.
NASA, Johnson Space Center, 1978, Data Format Control Book, Airborne Instrumentation
Research Program, Experiments Systems Division, Oct. 1978. NASA/JSC,
Houston, Texas.
.
NASA/JSC, USDA/ESS, 1979.
Enumerator's Manual, 1979 AgRISTARS Ground Data
Survey, Johnson Space Center, Houston, Texas,
ORI, 3:981, Renewable Resources Thematic Mapper Simulator Workshop, 2/23/81,
prepared for NASA /GSFC, ORL,Silver Spring; Maryland.
Richard, R.R., R.F. Merkel, and G.R. Meeks, 1978. "NS001MS - Landsat D-- Thematic
Mapper Band Aircraft Scanner", Proceedings of the 12th International
Symposium on Remote Sensing of Environment, Manila, Philippines, April 1978,
pp 719-728.
Spencer, M.M., J.M. Wolf, and M.A. Scholl, 1974. "A System to Geometrically
Rectify and Map Airborne Scanner Imagery and to Estimate Ground Area".
NASA Contractor Report, pprepared by Environmental Research Institute of
Michigan, Ann Arbor, Michigan.
81
9
j
ABSTRACTS OF PAPERS PRESENTED BY COI
at the
ASSOCIATION OF AMERICAN GEOGRAPHERS i
April 19-22, 1981
Los Angeles, California
er L. Brest
DYNAMICS OF SURFACE ALBEDO by Christopher
'his study assesses the role of albedo as a co^ributing factor
o the formation of urban heat islands. Field spectroradiometric
measurements are used is calibrate twenty-seven Landsat observations of the Hartford, Connecticut region, Calibration of these
remotely sensed data permits analysis of the ;spatial and temporal
variations of surface albedo for the entire region. Thestudy investigate5 albedo differences between man-made urban surfaces and
natural vegetative surfaces. Additionally, spectral reflectance
in the visible and near-infrared are being studied to determine
their contribution to net shortwave albedo_.
SEASONAL TRENDS IN LAND COVER ALBEDO ON EASTERN LONG ISLAND by
Michael S. Miller
Land cover albedo is evaluated using mu'ltitemporal Landsat MSS
data. This is accomplished by applying the Elterman atmospheric
attenuation model, Temporal variations in incoming radiation at
the surface are first determined with the model and compared with
ground pyrenometric measurements. Assuming 100 percent reflection, the model is used to estimate radiation reaching the satellite sensor. The ratios of Landsat energy measurements to these
estimates are used to derive albedo values. 'The Riverhead,
Long Island study site includes agriculture, soak forest, pine
barrens, water, and urbanized lands. These land cover types are
examined for spatial and seasonal variations in surface albedo.
PATTERNS OF AGRICULTURE: A FUNCTION OF SCALE by Tina K.
Studies in agricultural geography have generally occupied two extremes on a scale continuum: detailed field mapping, and large-area
regionalization.
The opportunity now exists to investigate spatial
patterns of agriculture at meso-scales as well, using the Landsat
data base.
'n this study, patterns of agriculture in the tropical
highlands of western Kenya are being investigated for a range of
scales.
Landsat digital data are spatially aggregated to generate
the range of scales, Using this approach, the gap in the scale continuum can be bridged, permitting explicit consideration of how patterns of agricultural land use appear to vary with scale,
LANDSAT MEASUREMENT OF LANDSCAPE GREENNESS by Helene Wilson and
Samuel N. Goward
Th e principal Information contained in Landsat multispectral data
is the degree to which any landscape is occupiad by photosynthesizing surfaces and is expressed in the relative differential between visible and near-infrared radiances.
Numerous studies have
demonstrated the utility of vegetation indices based on this differential for asse W ng a variety of land cover conditions. This
investigation extends previous work by explicitly relating the
amount of actively producing vegetation rover within a pixel to ee—
coroed radiances for a heterogeneous landslape. The "greenness"
approach offers potential as a nomothetic solution to the problei4
of exploiting the uniquely global features of Landsat rbservations.
ORIGINAL F
OF POOR C
Pape° presented at 1980 Annual Meeting#
Association of American Geographers Middle States Division,
University of Delaware, Newark # Aelow4ryt, October # 19801
OBSERVATION OF URBAN/RURAL AbBEDO CONTRASTS
Christopher L. Brest and Samuel N. Goward"
Columbia University
*Columbian University and NASA Goddard
Institute
for Space Studies
INTRODUCTION
Since albedo is a significLnt determinant of net radiation it
been suggested that
the
has
modified urban climate maybe due to the differences
in albedo of urban and rural surfaces. This research is an investigation
of variations in surface albedo in the Hartford, Connecticut region. The
intent of
the study
is to assess urban/rural surface albedo differences
and to ascertain the seasonal variability of these differences.
It is hypothesized that the albedo of a mid-latitude city in a humid
climate is significantly lower than that of its surrounding rural environs$.
The higher albedo of the rural area is attributed to vegetation in the
summer and snow cover in the winter. It is further hypothesized that the
urban/rural differences vary seasonally due to changes in rural albedo.
The methodology employed in this study allows measurement of the
albedo of the entire .study site, not just 'representative;' samples. Urban
areas are mosaics of various man-made and natural surfaces and only by a
study of the entire region can an understanding of the spatial and temporal
characteristics of albedo be acquired. Until recently equipment and techniques have not
been available to undertake the type of study proposed here.
Use of remote sensing techniques avoids the problem
oZ spot measurements
ii
w
,
or representative samples. The utility of remotely sensed, image formatter'.
data to measure solar and terrestrial radiation values has been demon-
ac
-'rated
Y
Y
1
i
k
AVAILABLE DATA
The primary data for this study consists of 25 observations of the
Hartford, Conn. area acquired
between 1972
and 1978 by the Landsat multi-
spectral scanner. The data are available in digital form and are being
for Space Studies. The 25 observations
analyzed at NASA's Goddard Institute
give excellent seasonal coverage. The number of observations per
month
are shown below:
JAN. FEB. MAR. APR. MAX JUN. aUL. AUG. SEP. OCT. NOV. DEC.
2
3
2
1
1
3
2
4
2
4
1
-
Additional data includes the radiometric measurements of selected
urban targets collected by the Authors which are to be used i+ the calibration procedure.
RESEARCH DESIGN
To successfully determine surface albedo from spectrally selective,
remotely sensed satellite observations three questions must be addressed:
1)
How do you derive al.bedo from spectral measurements?
2)
How do you convert radiance to percent reflectanc.e?
3)
How do you account for atmospheric effects?
Albedo is defined as the ratio of reflected to total radiation in the
short wave portion of the spectrum (<4.0 pm) 5 . It is therefore necessary
to consider the reflectance of the surface
in this
entire range. Spectral
intervals must be chosen which are representative of reflectance in this
range. This necessitates a knowledge of the reflectance characteristics
of the surfaces involved. Man-made surfaces such as concrete and asphalt
display a fairly constant reflectance with changing wavelength. Vegetation
has a very distinct reflectance curve, with a significant rise centered
3
c
F,
r
at about 0 . 725 µm separating the .response of vegetation into two distinct
segments: low reflectance in the visible and high reflectance in the nearinfrared. This 'variation in reflectance of a vegetative sur f ace i s account
for in this study by constructing a weighted average of the reflectance in
the visible and near-infrared. The weighting factors, 60% for the visible
and 40$ for the near-infrared, are derived from the proportion of incident
i
solar
radiation which lies below and above the vegetation rise at 0.725 pm6.
The Landsat multispectal scanner operates in four spectral intervals:
0.5-0.6, 0.6-0.7, 0.7-0.8, 0.8-1.1 4m.
The use of two Landsat bands which
straddle the rise in vegetation reflectance can be employed to construct
an accurate albedo measurement from Landsat spectral observations.
The issues raised by the .last two questions stated above are accounted
for by the calibration procedure used t;i this study.
The methodology employed
i
utilizes a radiometer which provides multispectral reflectance readings
of selected sites in the form of percent reflectance.
The radiometer was
Ii
designed specifically to provide these readings in support of remote sensing
miGsions and is described elsewhere .
Once computed the percent reflectance
of the targets is independent of atmospheric conditions.
I^
This is an impor-
s
tant feature because once a target has been measured, it can be utilized
in the calibration of satellite observations fron different passes.
It
eliminates the need to have ground based measurements taken coincidently
with each satellite pass.
A necessary condition for the success of this
procedure is that the targets selected must be temporally stable. The
targets chosen are large man-made surfaces: builaing roofs and parking lots.
This will test a new type of target not previously used in calibrating
satellite data.
Lj
.
For this preliminary paper 14 targets whose reflectances were mea"
cured in the summer and fall of 1979 are being used. Field work is
continuing in the summer of 1980 to increase the number of calibration
targets. Tha 14 targets range from dark to brightc percent reflectances
from approximately 4 to 50%.
The radiometrically observed spectral reflectance measurements are
used to calibrate the satellite data. Regression analysis is used to
determine the relationshi p between ground and satellite observations of
the targets sites (Fig. 1). Each band is calibrated separately to transform
the satellite observed radiances to reflectance and to remove atmospheric
effects. The two bands are combined according to the weighted average
previously discussed. Calibrated albedo maps can now be produced either
in the ,form of computer, printouts or video display.
PRELIMINARY RESULTS
Using the data from the 14 targets ,h began to evaluate our research
design using 4 of the 25 satellite passes as representative of the seasons.
The statistical relationships are shown below;
TARGET LANDSAT STATISTICAL RELATIONSHIPS
(Measured percent reflectance vs. Landsat radiance)
EXPLAINED
VARIANCE
(R2 )
MAY
AUG
NOV
FEB(No snoo cover)
BAND 4
(Visible)
0.90
0.89
0.90
0.90
BAND 6
(Near IR)
0.83
0.86
0.88
0.86
1
BAND 4
STANDARD
ERROR OF
BAND 6
ESTIMATE
(% REFLECTANCE)
2.6
2.8
2.6
2.6
3.9
3.3
3.2
3.5
ORIGINAL PAGE 15
OF POOR QUALITY
j
'ti
a
5
L.
R
#
The R2 values (0.83-0.9) are highly encouraging and indicate the
potential of our procedure for calibrating satellite data. The standard
error of estimate is higher than we had hoped but we are confident that
this can be lowered to approximately It through the addition of more
calibration targets.
Determination of the albedo of natural versus man-made surfaces was
conducted using the USGS landuse-landcover classification 1:250,000 scale
map of Hartford. For purposes of this paper we have selected six landcover types to demonstrate our preliminary conclusions. They are:
DR = Dense Residential
A - Agriculture
LR = Low Density Residential
F - Forest
CD = Commercial,Industrial,Transportation WN - Wetland,Nonforest
(Downtown)
In general, albedos of man-made surfaces are lower than those of
vegetative surfaces. The seasonal variation of albedo for the six landcover types is shown in Fig. 2. The variation is the greatest for the
landcover types which have a high proportion of vegetation, i.e., A,F,WN,
LR. The landcover types which are comprised mainly of man-made surfaces
are lower and have much less seasonal variation. The driving force behind
these albedo differences is the high infrared reflectance from vegetation
foliage. Reflectance in the near-infrared also shows seasonal variation.
It is this high reflectance in the infrared which produces the summertime
peak in the seasonal albedo shown in Fig. 2. This point is further reinforced by looking at Fig. 3 which shows the ratio of BAND 6/BAND 4. There
exists a high ratio in the spring and. summer. The LR landcover type falls
into this grouping because of the large proportion of vegetation present
it
i
I
e
between the man-made structures.
The three landcover types which might be considered 'urban' are
shown
in Fig. 4. This figure shows the individual band reflectances for
the three types. Again the type with the highest
proportion of vegetation,
LR, has a much higher reflectance in the infrared during spring
and summer
and is more variable. Notice the relative uniformity of the two landcover
types comprised
mainly of man-made surfaces, CD and DR, during the course
of the year.
color video displays of the computer generated albedo maps clearly
demonstrate the considerable seasonal dynamics of landcover albedos.
Rural, vegetative, surfaces have a high albedo in spring and summer.
Urban areas, or those comprised mainly of man-made surfaces, have a lower
albedo and remain more consistent throughout the four passes. In the fall
and winter rural surfaces return to a lower albedo, similar in magnitude
to that of urban areas.
At the present time the albedo values derived are low
in comparison
with those stated in the literature. The reason for this may be the
'i^
r
calibration panel which is used to derive the percent reflectance values
for the targets. We believe the furnished percent reflectance value for
the panel is too
low. At the conclusion of the summer 1980 field measure-
ments program the panel will be sent out for laboratory testing to confirm
this suspici.an.
y
CONCLUSIONS
Results generated demonstrate that seasonal variation in rural albedo
is significant. To understand urban/rural albedo contrasts seasonal
t
variations must be studied# Certain technical limitations in the calibration procedure need to be
resolved before the
absolute magnitude of
urban/rural albedo contrasts can be assessed. The
research show grest promise
of
methods applied
in this
in providing appropriate data for analysis
both temporal and spatial variation of surface albedo.
Final results
of this research should provide a valuable contribution to studies of
variations in interface energy budgets that result from variable
land-
cover conditions.
ACKNOWLEDGMENTS
The authors are grateful to Prof. Francis Conant of Hunter College,
C.U.N.Y., for the generous loan of
the field
radiometer. This research
was Supported in part by NASA grant NSG-5080.
REFERENCES CITED
Robert W. Pease and David A. Nichols, "Energy Balance Maps from Remotely
Sensed imagery," Photogrammetric Engineering and Remote Sensing 42
(November 1976); 1367-73.
2 Robert W. Pease et al.,"Urban Terrain Climatology and Remote Sensing,"
Annals AAG 66 (December 1976): 557-69.
3 William A. Malila, "Radiation Balance Mapping With Multispectral Scanner
Data," Remote Sensing of Earth's Resources Vol. 1 (1972): 769-83.
4 Robert H. Alexander et al., ,Applications of Skylab Data to Taiid Use and
Climatological Analysis, Final Report Skylab/EREP Investigation no. 469
Reston, Virginia: U.S, Geological Survey, 1976.
5
William D. Sellers, Physical. Climatology (Chicago: University of Chicago
Press, 1965), pp. 19-20.
6 Robert W. Pease and Stephen R. Pease, Photographic Films as Remote
Sensors for Measuring Albedos of Terrestrial Surfaces, Technical
Report V, U.S.G.S. Contract 14-06-0001-11914, University of California,
Riverside, 1972.
7
Dwight D. Egbert, Spectral Reflectivity Dataa: A Practicz,l, Acquisition
University of Kansas, Center for Research, Inc. 1970.
Procedure,
WMA
. ..
N
N
ABSTRACT
In this study an examination is made of spatial and temporal
variations of urban/rural albedo in the Hartford, Conn. region. Such
contrasts have been proposed as one cause of observed differences in
climatic phenomena. Due to the complexity of ground cover, the determination of spatial variability of surface albedo is difficult with
with traditional ground based observations. In this study calibrated
Landsat acquired remote sensor measurements are used for albedo assessment. Results indicate that urban/rural albedo differences vary
temporally. This results primarily from seasonal changes in the
reflectance properties of vegetation canopies. Greatest albedo
differences between urban and rural areas occur during the summer
due to the high near-infrared reflectance by vegetation. The smallest
differences occur during the winter, with the absence of snow.
:
•
ORIGINAL PAGE is
OF POOR QUALITY
I
FIG.
SCATTER PLOT OF LANDSAT BAND COMPARED
TO TARGET REFLECTANCE VALUES
(BAND 4 MAY)
30
25
h
1201s
s .
10-7
•
1-
•
W 5
o
^
1
1
^
1
1
i
1
25
30
35
40
45
50
58
BAND 4
FIG, 2 SEASONAL VARIATION OF ALBEDO
A
15
14
A
13
s
12
—A
LR
R-
F
LA
Wk
=11
s 10
CD
cp
DR
°R
WN
C
•'
WN
""•LR
w
r;
•
May
Aw
Nov
Feb
x
OR
ORIGINAL PAGE IS
OF POOR QUALITY
7
FI6.3 RAT10 OF SAND 4
s
LR
a
'LR//'^A
N
o
;4
s
s
a
2N^\IF
c0 _ --._ ^
1
May
M.
25
Fed
Nov
Avg
w
s-
FIG-4 LANDSAT BAND
REFLECTANCE
LR
W
BAND 6
BAND 4
%%
1920
W
J
W
W
X15
sW
va
W
C D^ .
p
LR ^.
DO-
a4
OR
Dot "'---.^ OR
SLR
d
10
d
LR
LR
Mop
Avg
Mm
e
Fr-'
h^
/%c.
A
N
or
ORIGINAL
PAGE
%7
1
I3
1a"
OF POOR QUALI TY
^
0
The 1981 Conference On Remote Sensing Education
May 18-22, 1981
Session No.
REMOTE SENSING RESEARCH IN GEOGRAPHIC EDUCATION: AN ALTERNATIVE ViEW
Helene Wilson, Tin,) K. Cary, and Samuel N. Coward
Department of Geography, Columbia University
and
NASA/Goddard institute for Space Studies
THE PROBLEM
Geography currently plays a major role in remote sensing education
st ol.,
1977),
this country (Estes,
in
with the emphasis today on
troining studonto in the application of remotely sensed observations
to
problems.
Th;o
emphasis
view of
reflects
a
prevolling
geogra p hic
data
remotely sensed
interpretation as a well-established body of
be applied
in
solving
techniques
to
o broad spectrum of geographic
This view is also apparent in what we
problems.
see as
the current
emphasis
in geographic remote sensing research on applications studies.
The underlying assumption of this applications orientation is
the
that
information
content rf remotely s.e.nsed data is known and coincides with
the types of 'information raquired for porti`culor purposes.
to which analysis of
for applications purposes
is
There is some question as to the degree
contemporary
remotely
sensed
data
mot'iefoctory.
The categories of information typically
of
interest
in
applications work
(e.g., Anderson., et al., 1976) have not proved to be
interpretable from current remotely sensed data with consistency and
confidence. This situation raises doubts as to the degree to which the
information contained in the current generation of sensor observations
of the earth's surface is understood.
SOURCE
OF THE
PROBLEM
Modern remote sensing systems have been criticized Uy
some
as not being
investigators
effective
in providing
of
the
types
information for which the use of traditional observation systems
(e.g.,
aerial
photography)
has
been proven successful.
While this criticism
may be viewed as merely a conservative reaction to the appearance of new
technology, it is a valid criticism.
The advent of remote sensing
from
space
occurred
during
a period
in which American geographical
photointerpretation work emphasized applicat'ions (Stone, 1974).
investlgotore appear to have turned to the new data as another means of
solving the some types of geographical problems to which they hod been
^1
ORIGINAL PAGE 19
OF POOR QUALITY
applying
aerial
photogr:oriol
phy.
There was n
phase of thei
investigation of these newdata comparable to that which hod I
the applications phase in
photographic interpretation.
Not surprfsfnglyr Tnvastigdtors hove oneountored d^ffiaulviva
,n
addressing o not of problems that hove remained the some with ge to that
have changed markedly.
The spatial, temporal, and spectral/rdd`iom*trio
po,rometers of new *beer y otian systems ore significantly different from
those of what may now be regarded on "trodItionol" systems,
The major
innovations associated with remote sensing from space are as follow ;
1) spatial.,uniform observations are acquired on a g1obol scale,
and are more generalized;
2) temporal;
coverage is of significantly higher frequency,
3 spectral/rodfometric;
observations are numerical measurements,
In many wavelength bones.
These features have o number of implications for the extraction of
geographic
information
from these new data. For example, the spatial
resolutions typically encountered in these new dota are coarser
than
those characteristic of most aerial
photography,
The relationship
between images in photography and features on the ground is fairly well
un4er3tood as
a result of accumulated experience in extending visual
perception of the surface to aerial perspectives.
However, there is no
reason
to assume that the associations which c yan be made at one scale
are valid at another, Furthermore, sensor systems do not necessarily
generalize
landscapes
in ways consistent with our conceptual
or
cartographic
gonerorizations.
Similarly,
increased
the
temporal
rejeoiution possible with 94teilite observing systems is quite different
from that of "traditional" systems.
Aerial
photographic missions are
usually planned for a time of year considered t be optimal for a
particular application (e.q., tres identification), and repeat coverage
To rarely obtained more than once every five years, As a result, the
selection of on appropriate observation date from among the many
available
for
a given
geographic location becomes problematical with
these new data.
Analysis of remotely sensed data in numerical format has been put
forward
as o poaalble solution to some of the difficulties presented by
the cha r acteristics of these new observations.
In particular,
analysis
the numerical
of
data
at the resolution
limit of
the sensor and
utilization of all wavelength bands simultaneously has been expected
to
provide information about the surface in greater detail than that which
human beings ore able to extract from visual
assessment of imagery.
Automated interpretation hoe also been seen as a means of affic lent Iy
utili «ing a data base that has been accumulating with
unprecedented
rapidity.
The spectral signature paradigm, widely employed in computer•bosed
Into rpre tot ion
of
the
multivoriote,
numerical data, has not met with
resounding success.
For example,
a
survey
(Joyroe,
1978)
of
investigations
using
digital
Landsot
data
to
classify
land
cover
indicates that classification accuracy
figures are on
the average
significontly lower then the minimum criterion of e5 percent often used
for
visual image interpretation (Anderson, et al„ 1976).
Furthermore,
spoctrol signature methods give acceptable
results only over very
Limited
areas
and
only
at certain times.
In addition, these methods
nearly
always
require- significant
amounts
of
ancillary
ground
information.
As a result, automated interpretotion has yet to be proven
es a means of exploiting the global and temporal coverage afforded by
satellite do to.
Apparently, the numerical analysis techniques
wide
fn
use are unable
to address effectively
the problems raised by the
features of contemporary remotely sensed earth observations.
We believe that others, like ourselves, have found the
of analysis of these new data have not
conformed
the results
initial
to
ORIGINAL PAGE IS
OF POOR QUALITY
3
expectations:, However, we
do not believe that
systems should be rejected+
We suggest that' the
incited understanding of t he nature of the data
somprehensive understond.ing
of the spectral,
as
landscapes,
observed
properties of
with
observation systems, has not yet been developed,
these now obsorvotion
problem lies to our
new being acquired. A
spatial, or temporal
enrth
contemporary
WHAT NEEDS TO BE DONE?
We see a need for basic research to determino the inherent
Information content of the data. An approach that we propose is to
consider remotely sensed data as a Measure of one or more unknown or
poorly-understood landscape attributes. The hypotheses for this
research should originate in exominotion of the remotely sensed do to
rather than from external objectives. The pool of this resoorch to to
establish models of landscapes that may be used to explain the
information content of remotely sensed data
One hypothesis that should be investigated to whether remotely
sensed observation n provide measures of continuously distributed landfor example, moisture content,
scope physical attributes thermal
end chloraphyll
content -- rather than simply indicating the
inortio,
presence and extent of discrete categories or objects,
To test this
hypothesis requires a thorough knowledge of the interactions between
energy
and
of
the
electromagnetic
landscapes,
in
expressed
spectral/rodiometric component of remotely sensed data. Study of these
is prevalent in remote sensing research conducted in the
T
men-photographic regions of the electromagnetic spectrum, For ekomple,
microwave remote sensing research is concerned with the influence of
moisture on observations; thermal properties ore considered to be key
landscape attributes in explaining thermal infrared observations. This
oppr000h is not often taken in the peak energy range of the solar
spectrum (visible and near-i'nfrored), because data such as those
acquired by the Landsot system tend to be viewed an merely on extension
of *hot is already known from experience with photographic observations.
New remotely
However, we believe that this presumption is not correct.
data cannot be viewed simply as a direct s. g tenslon of our visual
sensed
capab i 1 i ti ell.
Additional study is needed to investigate the relotionships between
and
rod'ometrc measurement of
landscape
and
spatial
*<
attributes
For example, one hypothesis might be that the ability
t gkmporol factors.
ta, measure a particular landscape attribute in invariant as a function
Although one might
of the spatial resolution of the observing system.
different attributes
to be measurable at different scales of
expect
ohaervation, relatively little is known about the ways in which sensors
If the
the
surface as a function of observation cell size.
generalize
scale,
obility to measure a ,particular attribute is found to vary with
focus for
investigation becomes the determination of the
a new
then
sco p e range over which the attribute is measurable.
With respect to temporal factors, an initial hypothesis that should
at all
is measurable
be examined is that a given landscape attribute
The consistency and precision of that measure may in fact vary
times.
with time -- for example, because of variutions in factors external
to
intensity of solor
as
such
differences
seoiAonal
landscape,
in
the
radiation,
WHO SHOULD 00 THIS RESEARCH?
in
advances
Specialists in many disciplines have
to
contributed
of
including engineers,
sensing
the
earth's
surface,
remote
others.
physiciots,
and
geologists.
agronomists,
mathematicions,
investigators
in tech of V7.2se disciplines understand and are
However,
concerned with only o port of the abaerved landscape and/or of the data.
integrated
represent
The dato, in contrast, are holistic, in that they
observations of landscapes.
Geography is the discipline that claims an
Therefore,
integrated approach to
landscape:
(Fennemon,
1919).
e
LAi
Fl-
4
ORIGINAL PAGE IS
OF POOR QUALITY
geog r aphers bear a major responsibility' for basic research on the
remotely sensed observation of landscapep.
Wh'lo , .-.r contention about the unique approach of geographers would
, ttle disagreement, the conclusion we draw with
meet v,1i l,
to
respect
r-..
the
geographers should be taking in basic remote sensing research
To parhra,-w lose widoiy accepted.
Geography as a discipline
to
oppeora
largely
os
o
ttaulf
passive
user
view
of
remotely
data,
sensed
principally
interested
in using
data as on adjunct to current
the
research objectives.
The development of analysis techniques,
theories,
and
,opeelficatione
for
future
system
designs
to
other
left to
Many geographers view bosh
disciplines.
remote sensing
research as
"non-geogrophie,"
since
it is seen as being concerned with "technique"
rather than with what is regarded as substantive geographic inquiry,
This attitude has hindered geographers
to making
remote sensing
research contributions commensurate with
the breadth of geographers'
perspectives on landscapes.
to develop
We cannot depend on others
theories and paradigms
for us.
The concerns of the systematic disciplinea are not necessarily coincident with our own.
the
Geography
is
only discipline which takes on integrated approach to the explanation of
areal
differentiation on the earth's surface.
We routinely apply this
approach
in
to explain
attempting
field
observations
ground
and
measurements.
Remote sensing systems provide new observations and
meosurements
of
aren'
differentiation.
must
assume
our
We
responsibility to seek view theories to explain the data and
provide
to
others with the benefits of our insights.
SIGNIFICANCE OF THIS APPROACH
Remotely aenaod observations of the earth 'a surface raise questions
that are particularly geographic to their form and
Geographers
scope.
to
have
contributed
significantly
aerial
development
of
photointerpretotion
techniques
by
detailed
the
examination
of
information
contained
the
generation
in
photographs.
new
of
The
observations has yet to be examined with such rigor.
remotely sensed
This failure to conduct the needed basic reseor ,sh
our
constrained
has
and
to
extract
apply the geographic information rontained in
ability
proposing
these new data.
We contend that the research Approach we ore
will
not
only
improve
applications
observotions
to
of
these
new
problems
of current interest but will also provide new ways
geographic
of
of examining landscapes that may directly contribute to
advancement
geographic methodology and theory.
The demonatrotion that remotely sensed observations of the earth's
consistently measure
surface
selected attributes of landscapes should
construction
of models
describe
relationships
that
the
enable the
between landscape factors (e.g., spattol arrangement, vertical extent of
within
the observation cell) and the attributes as
elements
landscope
the
to enhance
measured in the data.
These models should then
serve
utility of remotely sensed observations for collecting information about
For example, this approach should lead to on improvement
surfaca,
the
by
in our ability to use the data to identify and map
nominal
classes
providing
a
physical basis for explaining and predic^sing the degree to
the
which any ground category is distinguishable from others present in
This knowledge should also serve as a guide in the selection
londscope.
sensed
data
appropriate
problem.
In
remotely
a
particular
to
of
consistent measurements
landscape
addition, if the
data
provide
of
of the dato
attributes,
nomothetic
solutions
int:orpretotion
to
the
zany
allow
c>f
the
from
should be possible which will
analysis
data
geographic location, with a minimum of ancillary information.
}
j4
a new source of geographic
The potentiol of remote sensing as
about
the
environment has not been fully exploited (Bowden.
knowledge
bensed
We expect that the attributes measurable
with
1977).
remotely
For example,
data will serve as new indicators of landscape conditions.
provide
a
measure
of
degree to which a
observations
the
these
if
I^
ORIGINAL PAGE iS
OF POOR QUALITY
landscape iir photosynthetically active, we have a now moons to study the
functioning
biophysical
of
to
natural
landscapes
rolotion
and
in
cultural factors.
This
integrated expression of
londsco,e dynamics
should
highly
be
amenable to modeling, and model refinement should be
facilitated uy the availability of a large data bass
against which
to
test model
predictions.
The
results
of this research could lead to
significant now geographic :oneepts thut would assist
in environmental
and resources assessment as well as contribute to ndvonced
analysis
understanding of landscopec on a global basis,
IMPLICATIONS FOR REMOTE SENSING EDUCATION
Within many geography departments remote sensing
is viewed as a
more "technique"
a student should learn in order to carry out "true"
geographic research.
This view inhibits both students and faculty
from
Investigation of
remotely sensed data
as a now source of geographic
knowledge that may alter our understanding of the earth.
tendency
The
has been for geographers to accept these new data and analysis
techniques
from engineers and mathematicians without questioning the
accompanying premises,
"block-box"
approach
hos
hindered
This
geographic applications of the new remotely sensed data and has limited
the geographer's contribution to further development rf remote sensing
observation systems.
We suggest that geographers accept their inherent responsibility to
contribute to the development of remote sensing through pursuit of basic
research along the
lines
we
hove
proposed.
research
can
This
be
encouraged,
particularly
among
students,
demonstrating the links
by
between geographic theory and remotely sensed observations,
encouraging
a healthy skepticism concerning our current understanding of these data,
and
suggesting
possible
avenues
which
no), improve our
of
research
understanding.
The incorporation of the framework of
inquiry
proposed
into current geographic
here
remote
research and education
sensing
presents a challenge.
by
Rising to this challenge will,
bringing
the
weight
the
geographic
bear
on
of
pArspective
to
these
full
new
observations,
contribute
the
realization of the inherent value of
to
some
contemporary and future remote
At the
sensing
systems.
time,
of answers
questions
pursuit
such
as those we have posed should
to
enhance our understanding of landscapes.
REFERENCES
Anderson, J.
Use and
Data."
D. C.:
Bowden, L.
"A Land
R., E. E. Hardy, J. T. Roach, and R. E. Witmer.
Land Cover Classification System for Use with Remote Sensor
904.
Geological
Survey
Paper
Washington,
Professional
U. S, Government Printing Office, 1976,
"Remote Sensing and Geography."
RSEMS 4(1977):17-21.
"The Impact of Remote
Estes, J. E., J. R, Jensen, and D. S. Simonett,
The Post in Peespective,
Sensing on United States' Geography:
Proceedings of the 11th
Present Roof 1ties, Future Potentials."
Ann Arbor,
Int'I. Symp, on R. S. of Env't., 25-29 April 1977,
Michigan:
Environmental Research Institute of Michigan, 1977.
-
Fennemon, N, "The Circumference of Geography." Annals AAG 9(1919):3-11.
Joyroe, R. R., Jr.
Some observations about Landsot digital analysis,
George C.
NASA Technical Memorandum TM 78184.
MSFC, Alabomc:
Marshall Space Flight Center, August 1978,
In Remote Sensing:
Stone, K.
"Developing Geographical Remote Sensing."
Edited by J. E.
Techniques for Environmental Analysis, pp. 1-13.
Hamilton, 1974.
Estes and L. W. Senger.
Santo Barbara, Cal..
V