The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
1380
subjects is straightforward. Thanks to the excellent visual
quality of the image and the effectivity of displaying the
satellite image even at 1:2000 scale. An open source GIS
software called fGIS was first used to delineate the areas
occupied and controlled by the informal settlements. The choice
of the software was mainly due to its availability at the time of
mapping.
One particular challenge in mapping informal settlements is the
existence of houses partially or totally under large trees.
Significant portion of the campus is vegetated particularly the
northwestern part called the “arboretum” which is considered to
be the last remaining forest in the city. Mapping informal
settlements in this situation required some innovations
involving actual field boundary demarcation and the aid of
positioning devices to plot the peripheries of the community. A
handheld Garmin Global Positioning System (GPS) attached to
the pocket computer provided instantaneous positioning
information and markings on the satellite image which was
useful not only in documentation of the paths that were taken by
the field crew but also as a navigational aid.
Fieldworks ensued to validate and update the mapping result
with the aid of GPS. Stratified random sampling was employed
to verify samples from all informal settlement clusters. A high
95% identification accuracy was achieved. The
misidentification is attributed to the one year temporal
difference of the data and the actual fieldwork, during which
eviction occured as a result of the continuous effort of the
campus administration to recover the control of the land. Base
from the QuickBird satellite image, around 16% of the total 493
hectares UP Campus or roughly 79 hectares may be labeled as
informal settlements. This 2004 UP Campus informal
settlements map will now serve as the baseline data for
monitoring further encroachment.
After the establishment of the UP Campus informal settlements
map, the individual houses were digitized to prepare for the
other phase of this research which is to estimate the population
of informal settlement communities. In the high resolution
satellite image, the semi-formal houses with an average surface
area of 30 square meters are large enough to be easily
delineated. Problem is apparent only in mapping the roofs of
each slum type houses. The very cramped area even as small as
an unimaginable 10 square meters is very difficult to visually
delineate especially when they are placed so closed with each
other forming a seemingly one large continuous roof. Jain (2008)
experienced the same in extracting information in old
developments as well as in informal settlements where dwelling
size is considerably small and building are placed adjacent to
each other.
6. RESULT AND ANALYSIS
6.1 Regression Analysis
A fieldwork was conducted to gather sample data for the
regression analysis. A stratified random sampling method was
employed to assure complete representation of the total
population throughout the image. A total of 160 samples were
collected bearing the identity number, number of residents in
each house, informal settlement category whether slum or semi-
formal, roof-derived surface area of the house, and the type
indicating whether single or multilevel. The field data gathering
is a challenging task. There is security to consider and there is
the hesitation of residents to share informations for fear that
anything they say might be used against them.
The sample data was processed using the Grid and Theme
Regression, a program made by Jeff Jenness in the AVENUE
programming language, an extension of the Arc View 3.2
software. The initial plot of the raw data in Figure 3 showed low
correlation of the two variables in all polynomial order with the
third order producing the highest. The summary of correlation
is shown in Table 1. This may be attributed to the undivided
type of data (slum and semi-formal) and the existence of
multilevel houses showing large number of residents with very
small area, producing an inconsistent trend or random
occurrence.
Scatterplot is linked w rth table 'fagdata 1 _regress_1 .dbf...
Selecting features from one will automatically select features from the other.
Model = B0 + B 1 ’'[Area]
R-Squared = 0.0007, Adjusted R-Squared = -0.0057
Figure 3. First Order Regression Plot
1 st Order
2 nd Order
3 rd Order
R 2
0.000689
0.002756
0.013938
Table 1. Summary of Correlation
Having multilevel floors among the houses appears to be the
biggest problem in this endeavor to establish an effective way of
estimating population. This exposes the limitations of the
satellite image being not able to recognize this particular
natural phenomena. However, this should not hinder us from
formulating solutions to discover and make something useful.
According to Rindfuss and Stem (1998), each data source has
its imperfections, but combining sources with different
limitations might provide a better picture of the entire
phenomenon. In this way, remote sensing even with its
imperfections, can make a contribution to social scientific
measurements by improving on some measures and cross
checking others.
Modifying the data entries by removing the multilevel houses
and separating the sample data of the two informal settlement
types, the number of samples was reduced from 160 to over a
hundred. For the two informal settlement types, slums and semi-
formal, the same processing using the Grid and Theme
Regression in AVENUE has been repeated. The correlation of
the variables for both type has dramatically increased in all
polynomial order with the third order producing the