BACKGROUND 
Work to date has shown that wavelength (Bryan, 1975; Haack, 
1984; Hender$on and Anuta, 1980), incident angle (Brisco 
ct.al, 1983; Henderson, 1995; Hussin, 1995; Kessler, 1986), 
and polarization (Haack, 1984; Henderson and Mogilski, 1987; 
Lewis, 1968) all influence detection of human settlement. Thc 
role of environment has also been explored (Forster, 1985; 
Henderson and Anuta, 1980; Lo, 1986; Trevett, 1986). A 
review of these system and environmental variables and a 
review of radar research in these areas can be found in 
Henderson and Xia (in press); and Xia and Henderson (1995). 
One of the problems to date has been the unavailability of SAR 
data sets that permit an analysis of more than a single variable. 
Most SAR systems acquire data at a single wavelength, 
polarization, and incident angle. This study provides the 
opportunity to study these variables simultancously in a single 
study area. 
STUDY AREA 
The study area, approximately 20 by 25 km, is located 
southwest of Munich, Germany. The glacial and fluvial-glacial 
landscape is part of the Ammer Loissach Hucgelland. The 
undulating hills reach heights of 750m with a main gradient 
from south to north. Land cover consists primarily of mixed 
hardwood and conifer forests, wetlands, and meadows and 
pastures. "The arca is popular with tourists and as a commuting 
suburb to Munich. As a result the arca population has increased 
dramatically over the last thirty years. "Village populations 
range from a high of about 10,000 to small groups of only two 
or three buildings. 
DATA 
The study area was imaged by SIR-C SAR on ascending passes 
on April 11, 1994 and April 18, 1994. The former pass was 
acquired with an incident angle at image center of 28 degrees; 
the latter with an incident angle of 57.7 degrees. The data 
were resampled to produce a 12.5m resolution ground range 
projection and hard copy images at a scale of 1:110,000 
generated for interpretation. The entire data set consisted of 17 
scenes. The April 11 data set (small incident angle) consisted 
of: (1) X-VV; (2) C-VV; (3) C-HH; (4) C-HV; (5) L-VV; (6) 
L-HH; and (7) L-HV. The April 18 data set (large incident 
angle) consisted of: (1) X-VV; (2) C-VV; (3) C-HV; (4) L-VV; 
(5) L-HV; (6) C-total power; and (7) L-total power. In 
addition, three multispectral color composite SAR scenes were 
produced: (1) X-VV/C-HV/L-VV; and (2) X-VV/L-HV/C-tp 
from the large incident angle data and (3) X-VV/C-HH/L-HV 
from the small incident angle data. 
METHODOLOGY 
Each scene was visually interpreted and all sites thought to be 
settlements were delimited. SAR scenes were selected for 
interpretation in a random order. The images were interpreted 
over a three week period with the color composites analyz 
last to minimize any learning curve bias. The results were 
compared with 1:50,000 topographic maps of the study arca t 
determine accuracy and errors. Fifty-three settlements in the 
study arca comprised the data base. Settlements listing no 
population in the census reports were excluded from the study 
as being too small. 
RESULTS 
Wavelength: The best results from the single wavelength 
scenes were obtained with the L-band imagery. The poorest 
results were from the X-band imagery, with the C-band 
accuracies falling in between. This was true for both incident 
angles. 
Polarization: For the large incident angle data the C-tp image 
was the most accurate of the C-band data, followed by the HV 
and the VV images. A slight reverse in this ranking was 
cvident for the L-band data. Here, the HV image was the best, 
followed by the L-tp and the VV images. At the small incident 
angle the best C-band image was the HV, followed by the HH 
and VV; the rankings werc identical for the L-band data. For 
all single image data sets the best polarization was the L-HV 
imagc. 
Incident Angle: — The poorest accuracy was produced 
consistently by the small incident angle images. Steep incident 
angle imagery was not conduciveto settlement detection. When 
imagery with the same wavelength and polarization but 
different incident angles was compared (5 cases), the large 
incident angle was significantly superior in every case. 
Color Composite: There were three multispectral SAR data 
sets generated as color composite images. The small incident 
angle image was comprised of X-VV/C-HV/L-VV data, and 
the two large incident angle images were combinations of X- 
VV/L-HV/Ctp and X-VV/L-HV/C-HH. Each of the color 
composite images evidenced a synergetic effect, compared to 
the individual images. The accuracy of the multispectral SAR 
image ranged from 11 per cent to 64 per cent better than the 
individual bands that comprised the composite. The two large 
incident angle color composites were the most accurate images 
of all 17 SAR image data sets. 
Table 1 lists the respective images, their accuracy, and number 
of commission errors. As can be seen, the detection ranged 
from a high of 96.2 per cent for the X-VV/C-HV/L-VV LIA 
(Large Incident Angle) composite to a low of 17.0 per cent for 
the X-VV SIA (Small Incident Angle) image. The L-HV LIA 
image produced the third highest accuracy (83.0%), and the L- 
tp image was tied with the SIA color composite (81.1%) for 
fourth best. 
The worst image was the X-band, VV polarized imag 
. acquired at the small incident angle. The greatest number of 
288 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996 
eeu a pte) "i deu M aar: