acquisition on 16 May 2006 under the threshold value I as the result with the higher percentage than others. The swamps area distribution using image
acquisition on 16 May 2006 had smaller area proportion than image acquisition on 15 April 2000. It’s caused by the acquisition date of the image influencing the
temperature and water distribution in whole area of interest. It implies to the large area of swamp distribution. Swamps distribution obtained from image acquisition
on 15 April 2000 under threshold II analyzing have the area around 6,389 km
2
and swamps area distribution obtained from image acquisition on 16 May 2006 under
threshold I had the area is around 5,493 km
2
.
4.2 Accuracy Assessment
Accuracy assessment was done by comparing the number of pixels each threshold results classified map with number of pixels for swamps area based on
RBI reference map. The reference map for accuracy assessment can be seen in Figure 26.
Figure 26 Reference map in swamps determination
The most problem occurred in the swamps determination is in the image classification processing. In this step, the initial cluster gathering was undertaken
using unsupervised classification. The resulting clusters were later edited, assigned names and used in final supervised classification of the image as a base
50
reference for each spectral. Training area was formed by using visual interpretation of the image referring to the each spectral from unsupervised
classification. Some land cover types have the similar spectral, but based on field knowledge and visual interpretation is different land cover types. In image
acquisition on 15 April 2000, one constraint in image classification is haze. It caused the spectral value of two different land cover type almost similar for
example paddy field and shrub have the similar spectral value therefore in visual interpretation it makes confusion among others. Likewise in image acquisition on
16 May 2006, settlement and bare land have the similar spectral value. For the next processing, the classification results would be reclassified into 2 type of land
cover that are the swamps and not swamps area. Misclassification in the image classification process can caused the error in determining the swamps area in
whole area of interest. Table 20 summarizes the accuracy assessment results for swamps area
distribution and clearly shows that for image acquisition on 15 April 2000, the overall accuracy under the threshold value I have the value of 72.1, under the
threshold II have the value of 73.5 and under the threshold III have the value of 72.6. For image acquisition on 16 May 2006, the overall accuracy under the
threshold I have the value of 63.5, under the threshold II have the value of 60.9 and under the threshold III have the value 25.6.
According to Appendix 2 can be seen the user’s accuracy and producer’s accuracy for each satellite imagery. For image acquisition on 15 April 2000, under
threshold I producer’s accuracy of swamps area 85.1 is higher than user’s accuracy 81.7. Similar to threshold II, swamps area had the producer’s
accuracy 87.5 higher than user’s accuracy 81.6 and the same trend with the threshold III with the producer’s accuracy of 85.1 and user’s accuracy of 82.2.
It indicates that swamps area was good classified using this method. For image acquisition on 16 May 2006, under threshold I producer’s accuracy of swamps
area 73.6 is less than user’s accuracy 80.4. Similar to threshold II, swamps area had the producer’s accuracy 68.8 less than user’s accuracy 80.7 and
the same trend with the threshold III with the producer’s accuracy of 13.9 and
51
user’s accuracy of 76.0. It indicates that the misclassifications occurred in swamps and not swamps area from this method.
For the next section, the analyzing and discussion is only from image with the higher overall accuracy than others that are for image acquisition on 15 April
2000, the threshold II with LST value range of 23
o
C - 33
o
C and NDWI value range of -0.35 – 0.1. Meanwhile for image acquisition on 16 May 2006 is the
threshold I with LST value range of 23
o
C - 33
o
C and NDWI value range of -0.43 – 0.
Table 20 Accuracies of determination results under each threshold. Image
Acquisition Threshold
Producer’s Accuracy
User’s Accuracy
Overall Accuracy
15 April 2000 I 85.2 81.7 72.1
II 87.5 81.6 73.5 III 85.1 82.2 72.6
16 May 2006 I 73.6 80.4 63.5
II 68.8 80.7 60.9 III 13.9 76.0 25.6
4.3 Suitability Analysis
