ept for a 
ed areas as 
ATATO 
figure out 
| data by 
|. Defense 
OLS hasa 
le to near 
at sensing 
magnitude 
VHRR or 
ne scenes. 
uipposed to 
. Elvidge 
correlation 
b), though 
estimating 
ido 
A fairly cloud-free night-time image of Japan, obtained by 
DMSP/OLS on 11 November 1994 (Figure 1), was 
provided by the National Geophysical Data Center of 
NOAA (in Boulder, Colorado, U.S.A.) to the author for this 
analysis. The Hokkaidoisland, in the northern part of Japan, 
was selected as the case study area, for (a) this region has 
relatively lower population density than other parts of Japan, 
and (b) most of the pixels covering the case study area are 
supposed to be free from saturation caused by extremely 
bright city lights, which is often the case with pixels 
covering major Japanese cities (e.g. Tokyo or Osaka). 
DEVELOPMENT OF REGRESSION MODEL 
The satellite imagery in digital format was re-sampled to the 
thirdorder standard mesh system of the Japanese 
govemment (approximately 1 km by 1 km in size.) The 
population data of the Hokkaido island in the same mesh 
system (Figure 2), after the National Census conducted in 
1992, were obtained from the Japanese Government. 
The coefficient of correlation between the digital number of 
the DMSP/OLS imagery and the logarithm of the 
population (POP-L) was 0.531 for non-saturated 20974 
pixels. Many "bright" pixels were found, within the 
DMSP/OLS imagery, in areas where few residences 
actually live in. 
Such bright pixels were mostl y observed in the coastal zone, 
both on land and sea. It turned out, as Croft suggestedin his 
initial paper on the potential use of nighttime DMSP/OLS 
data for observation of city lights (Croft, 1978), that these 
pixels were due to presence of fishing boats, mostly 
squid-fishing vessels. Such boats, weighting between 60 
and 100 tons, are especially equipped with as many as 50 
incandescent lamps with an average power of 3,500 watts 
per lamp (Croft, 1978), in order to attract squids in the dark. 
Half of the bulbs have no shades, while the other half have 
only small shades. Itis the way of fishing squids commonly 
practiced in the Japan Sea as well as the coastal zone of the 
Hokkaido island. Very bright lights are in fact observed in 
the midde of the Japan Sea, where no island exists. On 
coastal zones, some of the light emissions from these 
squid-fishing vessels could be mistakenly recognized as city 
lights. 
On the other hand, some of the populated areasin the inland 
Zones failedto be seen as bright pixels. It was assumed that 
the emission of city lights, as seen from the space, may 
depend on the way of life in a given area. For example, 
urban areas may have more light emissions than in rural 
zones due to such sources as streetlights andilluminations 
for advertisement. 
APPLICATION OF VEGETATION INDEX AND 
TOPOGRAPHY 
In the subsequent stage of the analysis, a couple of 
additional data sets, namely normalized vegetation index 
(NV) and topography (DTM), were obtained to find out if 
these data sets could improve the accuracy in estimating 
population. The NVI data for the Hokkaido area in mid 
1992 were obtained out of the DAAC's 1 km NVI project 
ftp site, and the DTM data in the third-order standard mesh 
system were obtained from the Japanese government. The 
former was resampled to the latter. 
The correlation between POP-L and NVI was -0.272, while 
the same between POP-L and DTM was -0.229. These 
numbers suggest that the digital number of the DMSP/OLS 
imagery much significantly represents the amount of 
population in a given area than NVI or DTM. 
A multiple regression model for POP-L with the digital 
number of the DMSP/OLS imagery, NVI and DTM has the 
R value of 0.577, which is only slightly better than the 
same in the simple regression model with the digital 
number of the DMSP/OLS imagery alone (R =0.531). The 
results suggest that neither NVI nor DTM does not 
significantly contribute to improve the accuracy in 
estimating the population with the DMSP/OLS imagery. 
  
Figure 2: Population Distribution in 1 km by 1 
km Mesh (Grid) Database 
ESTIMATING POPULATION IN GRID 
SQUARE 
Analysis was also made for population in square grids of 10 
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 457