30 Based on these reasons, logistic regression model in this research is combined with
CLUE-S framework to predict future land use change in the area.
2.4.2.3. Land use probability maps based on logistic regression
The final results of the logistic regression model are used to create probability map of each land use, including urban built-up area, agricultural area, mix
vegetation, ponds area and river. The value of probability of each land use type is in the range between 0 and 1. The probability maps can be seen in Appendix 7,
which show that the darker of the color, have the higher the probability of land use to change. The range of probability water area is 0.870
– 0.984, grassland area is 0 – 0.273, estate is 0 – 0.929, settlement is constant is 0.5, and forest area is 0.002 –
0.966. The interesting from probability of settlement where the constant is 0.5, that’s mean every pixel in a whole study area has the same probability to change to
settlement see Appendix 7.
2.4.2.4. Model Validation
In order to achieve a valid land use simulation, the validity of this model is examined by using the observed land use map year 2009, which has been created
from image interpretation. Land use map year 2009 resulted from the simulation is compared to the observed land use map year 2009 to measure the number of equal
grid cells in both of maps and the similarity of land use pattern. Similar with the method done by Zhu et. al. 2009, this approach results a overall accuracy which
indicate the degree of similarity between those two maps and Kappa accuracy that depicts the degree of pattern similarity. According to Landis and Koch 1977,
Kappa accuracy is useful to calculate the agreement of two maps. It is stated that Kappa values 80 is categorized as fit, 60
– 80 is high agreement, 40 – 60 is moderate and 40 is poor. Therefore Kappa accuracy is used in this
research to measure pattern similarity. After comparing both maps, the results show that overall accuracy and Kappa
accuracy for land use simulation year 2009 is 90.83 and 86.00 or categorized as fit. It indicates that the driving factors have a good capability to explain land use
31 pattern in study area and it can be used to predict future land use pattern.
Furthermore, related to the uncertainty of land use pattern in the future, 90.83 overall accuracy and 86.00 Kappa accuracy values show that all of driving factors
capable to reduce the uncertainty because they capable to describe the land use behavior that will shape the future land use condition. The results are confirmed to
the statement of Pontius and Neeti 2009 where high agreement resulted from validation process indicates that the processes of land use change during the
calculation are stable trough the interval of validation and suitable to be used in simulation process. The result of Kappa measurement can be seen in the table
below.
Table 11. Results of KAPPA Analysis for Comparison between Land use map
between interpretation and simulation for year 2009.
Land Use User
Total User
Accuracy W
G E
S F
Pr o
d u
c e
r W
374 374
1.000 100.000
G 6971
832 76
1375 9254
0.753 75.330
E
830 42608
1419 315
45172 0.943
94.324
S 172
1407 12222
17 13818
0.884 88.450
F 1280
325 102
18593 20300
0.916 91.591
Total 374
9253 45172
13819 20300
88918
Producer Accuracy
1 0.75
0.94 0.88
0.92 100
75.34 94.32
88.44 91.60
Overall Accuracy 90.83
Kappa Accuracy 86.00
Figure 7. Land use map based on interpretation and simulation result year 2009