Page 288 - Contributed Paper Session (CPS) - Volume 2

CPS1857 Nicholas J. et al.
3. Result
3.1 The Training Process
The training data are acquired from the Landsat 8 satellite imagery and
consist of pixels of the impervious, green, water, and soil land. We obtain the
data by looking through the coordinates of the location of each land types
using the Google Earth. The pixels of the location of each land types are
located in the Landsat 8 images by cropping the longitude and latitude
coordinates of the coresponding area. The cropped images of each land types
will be used as the training data (input). The regression model using the limited
dependent variable will be estimated based on the robust subsample n. The
model estimation cannot be interpreted as the change in response y given
one unit increased of because of ∈ {1,0}. The outcome of the training

process is four regression models of four land types, i.e. for impervious,
for green, for water, and for soil land.

= −0.2879 + 0.0012 + 0.0105 – 0.0081 + 598.2345 (13)
1
3
2
− 400.127
4
5
= 0.7005 + 0.0109 – 0.0112 – 0.0013 – 38.0647 (14)
2
3
1
+ 25.9631
4
5
= −0.0496 + 0.0002 – 0.0002 + 0.0003 – 332.0284 (15)
2
3
1
+ 221.7376
4
5
= 0.595 – 0.0172 + 0.0068 + 0.0082 – 262.6767 + (16)
3
4
2
1
175.4784
5
The outlier detection using the Cook’s distance method shows that the
distance is below the 0.05 critical of F table value (FTable = 2.2151). With that
result, there are no outliers detected by the LMS regression (Fig. 2). It shows
that the LMS regression is a simple method to perform a high performance
computation for minimizing the outliers.

Figure 2. The Scatter Plot of Cook’s Distance For Each Land Types
3.2 The Mapping Process
The LMS Regression is a potential method for mapping the land
classification and land-use change. Assumming that , , … , are the pixels
1

2
of the Jabodetabek area, we will mapping the land classification using the four
regression models that are already generated before on each pixel , =

1, 2, … , . Each pixel is classified into one of the four classes based on the
277 | I S I W S C 2 0 1 9