Image Processing Reference
In-Depth Information
Figure
17.3
below demonstrates how the varying levels of light intensity within
an urban area (Los Angeles) can be used to approximate intra-urban 'ambient'
population density. Census data typically records residential population density
which is ironically a measure primarily of where people are at night when they are
sleeping. To truly capture 'ambient' population density one would need spatially
explicit human behavior data. The ground truth in Fig.
17.3
tries to capture this by averaging residential and employ-
ment population density data for Los Angeles. The corre-
lation between the nighttime imagery is statistically
significant with both employment and residential mea-
sures of population density. However the strongest corre-
lation is with the average of the two which suggests that
the nighttime imagery is probably a better measure of
'ambient' population density then either residence or
employment based measures. To appreciate this, imagine
an airport. The census will usually describe the area in and
around an airport as having a very low population density
because very few people live there. However, the nighttime imagery usually picks
up airports as having a significant nighttime light signal. Airports have a high
'ambient' population density because of the many travelers that travel through them
on a daily basis and the many people who work at airports.
the concept of
'ambient'
population
density is a
representation of
population
density based on
the mobility of
people through
time
Application of Nighttime Data for Urban Change Analysis
Another powerful aspect of the DMSP OLS image data products is their avail-
ability in time series. This allows for change detection. An obvious example
would be attempts to measure increase in urban extent. The India-Pakistan
border presents an interesting twist on this concept. Figure
17.4
shows changes
in light emissions for the border between India and Pakistan. To create the
image, a composite image derived from orbits in 1992-1993 was compared to
a composite image derived from orbits in 2000. The interpretation of colors in
the change-detection images is as follows: black represents bright lights
(saturated) in both time periods, red represents lights much brighter in 2000,
yellow indicates new lights in 2000, blue indicates lights missing or substan-
tially dimmer in 2000, and grays indicate that no light was detected in both
time periods. The bright linear feature on the left side of the image about half-
way up isn't a river; it's the border. The blue area in the middle of the image
seems to be confined to the state of Uttar Pradesh, raising interesting ques-
tions about how the demographic and economic changes that took place in
Uttar Pradesh may have been different than those of other Indian states. The
states of Haryana and the Punjab to the north of Delhi seemed to have dramatic
increases in the spatial extent of lit areas (more red and yellow), whereas
