area for the U.S.. Comparisons to census data is by no means 
the only way, or even the best way, to evaluate the DMSP 
data. It is, however, a useful comparison to make if the 
desired application is the spatial measurement of urban or 
urbanized area. The problem in making comparisons to 
census data is the radical difference between what the 
DMSP/OLS satellite imagery represents and what the 
census is. The DMSP/OLS sensor only registers emitted 
light. The relationship between the emitted light and any 
land use class or demographic variable will be dependent on 
a complex set of physical and social factors that determine 
the number of watts emitted by lit infrastructure on the 
ground. The placement of lights, their intensity, type, 
density, purpose etc., will all vary depending on 
development guidelines. In developed countries, such as the 
U.S., these development guidelines may be consistent 
enough that a data set showing lit area at night could map 
predictably into some definable land use class that can be 
indexed to a reliable and useful demographic parameter. This 
was our basic hypothesis for our methodology in 
thresholding the gain controlled DMSP data. 
For our experiment, we took the entire gain controlled 
DMSP/OLS image of North America, clipped it in a 
geographic information system to conform to the lower 48 
states of the U.S.. We then generated a cumulative histogram 
of image values where area was represented in the Y axis and 
DN's in the X-axis. We then examined the dynamic range of 
the data and selected an image threshold value that 
represented a total lit area that most closely matched the 
total urban and urbanized area estimate for the U.S. from the 
1990 census. We found that the dynamic range offered by the 
gain controlled DMSP was very good and a single DN 
threshold value (DN >25 which is equivalent to > 1.25x107* 
Watts/cm?/sr/um) could be defined where lit area above the 
threshold summed to 225,596 km”. This was very close to 
the 1990 U.S. Census estimated area of urbanized lands 
which was 222,606.1 km“. The difference between the two is 
only 1.34%. 
This threshold value was then used to regenerate the 
DMSP/OLS image so that only areas with radiance values 
above 1.25x10 Watts/cm^/sr/um would be counted as lit 
area. The hypothesis was that an accurate map of urbanized 
area for the entire coterminous U.S. could be created using a 
single threshold value. 
RESULTS 
Comparing DMSP Data Sets to Census Data 
Since we selected a single threshold value to create the 
urban area estimates from the DMSP data, we needed to test 
for spatial or regional consistency within the continental 
U.S.. To do this we compared the lit area defined by the 
thresholded DMSP image product to the census data on a 
state by state basis. This way we could make a statistical 
comparison to both the census and our previous “stable 
lights” product. 
A statistical comparison of the lit area provided by the 
thresholded gain controlled DMSP/OLS image to the U.S. 
census data showed a very close statistical match (Figure 1). 
There was basically no significant difference between the 
DMSP/OLS map of lit area and the urbanized area 
designated by the 1990 U.S. census. 
436 International Archives of Photogrammetry and Remote Sensing. Vol. 
DMSP/OLS (Gain Controlled) Urban Area 
Estimates vs US Bureau of the Census 
  
   
  
  
   
   
  
20000 ————Ó————— an + 
| [-o- Threshold »25 
~~ O 
"E 
= 415000 } 
o | 
-— } 
© | 
2 i 
2 10000 | 7 
= Yz 381.42 « 0.929*X 
© R=0.94648 
<< 
5000 | 
= No Significant Difference 
Between Methods 
(t-test, t-value=-.579, p-value=.5654) 
  
  
  
0 L ok 1 L Pe 
0 5000 10000 15000 20000 25000 
US Bureau of Census for States 
Total Urban Area (km?) 
Figure 1. Comparison of lit area (above the threshold) for the | 
gain controlled DMSP/OLS data and urban and urbanized 
area designated by the 1990 U.S. census for the lower 48 
states in the United States. There is no significant difference 
between methods. 
The state by state comparison also revealed that the 
thresholded gain controlled DMSP data had less variability 
than the urban map created using the “stable lights” DMSP 
data. The maximum underestimate for any state for the gain 
controlled DMSP product is 46% compared to 78% percent 
for the stable lights product and the maximum over estimate 
(Wyoming) is nearly half (245% vs 417%) of what it was for 
the stable lights product (Imhoff et al. 1997b). 
Table 1: A comparison of lit area from the gain controlled 
thresholded DMSP/OLS data with urban and urbanized area 
from the 1990 U.S. Census. Threshold is >25 DN value. 
  
State Lit Area Census % Difference 
DMSP/OLS Total Urban Census vs 
(>25) (km?) Area (km?) DMSP GC (>25) 
  
Alabama 3741.07 6935.80 -46.06 
Arizona 3486.28 5312.60 -34.38 
Arkansas 2216.53 3167.60 -30.02 
California 17325.92 21174.40 -18.18 
Colorado 3916.30 3411.10 14.81 
Connecticut 3071.20 3246.00 -5.39 
Delaware 926.01 540.50 71.32 
D.C. 225.36 159.10 41.65 
Florida 12248.27  13281.60 -7.78 
Georgia 5810.84 7283.50 -20.22 
Idaho 926.68 834.20 11.09 
Illinois 12631.27 7854.30 60.82 
Indiana 7469.82 4613.60 61.91 
Iowa 3366.39 2850.70 18.09 
Kansas 2050.10 2416.00 -15.15 
Kentucky 3272 44 2733.10 38.2 
Louisiana 4008.77 4137.30 -3.11 
Maine 1236.60 1873.30 -33.99 
Maryland 4538.69 4087.90 11.03 
XXXII, Part 7, Budapest, 1998 
<< DLL DTT ON FF rp) ry) ry ry ry ry rp rp mp pp 
lc