Image Processing Reference
In-Depth Information
nighttime imagery are shown on the LandScan population density dataset. The regres-
sion on the left shows that there is a moderate log-log relationship between the
population and area of the urban clusters (Stewart and Warntz
1958
; Tobler
1969
).
The regression on the right contrasts with similar analyses shown in Fig.
17.3
in that
there is only a very weak relationship between light intensity and population
density on a pixel by pixel basis.
The clusters of light shown in Fig.
17.7
clearly illustrates; however, areas of high
population density. The DMSP performs poorly as tool to measure rural population
in Guatemala. The authors know of many rural areas (in a sense the Guatemalan
equivalent of the North American exurbia) that do not appear using DMSP simply
because rural folk in Guatemala use their light conservatively. Future research
might combine night-time light data with rural fire data to get an idea of rural popu-
lation. That is, use fire locations and areal extent detected by the satellite as an
indicator of rural population density.
Figure
17.8
illustrates how different urban settings in Guatemala emit different
light levels and how these light intensities are captured by the sensor. For example,
a clear difference is noted in both the light levels and the ground shots between
central Guatemala City and the northern reaches of the same city where recent popu-
lation growth has taken new residents into unsuitable building areas that cannot
support the same density of people and buildings as the traditional downtown area.
Unlike many North American cities, Guatemala's downtown is a high-density resi-
dential and commercial zone. Overall, the imagery performs well in depicting popu-
lation density in urban settings in Guatemala, regardless of the size of the community
(see Fig.
17.8
). Also, because Guatemala's population is still largely rural (60%
rural, which is high for Latin America), the imagery also performs well in illustrat-
ing the distribution of Guatemala's rural population, who live in dispersed house-
holds or hamlets.
If we know where people live and we also have information about country-
level GDP, we can begin to create new maps using DMSP to map GDP per capita
at finer spatial resolutions. Basically, we can map areas of poverty and wealth in
developing nations. Figure
17.9
is the result of an exploratory exercise combin-
ing nighttime satellite imagery and LandScan data in a new way. Basically the
nighttime imagery is used to allocate the GDP of Guatemala to 1 km
2
spatial
resolution. This is accomplished by spreading the roughly 50 billion of GDP
around the country based on the lights. Agriculture is about 25% of Guatemla's
GDP and this is uniformly allocated to all the 'dark' areas in the DMSP OLS
image. The other 75% of GDP is linearly allocated to the lit areas of the DMSP
OLS image based on the light intensity or DN value. This produces a map of
GDP per km
2
. The LandScan population density dataset is then divided into this
map of GDP to produce a map of wealth and poverty or GDP per capita of
Guatemala at 1 km
2
resolution (Fig.
17.9
). This kind of analysis is merely
exploratory; it has not been validated in any way and probably suffers from
problems like the ecological fallacy and the modifiable areal unit problem
(MAUP). The Ecological fallacy and MAUP are problems associated with inter-
preting the relationships between variables as their formal representations are
