Performance Analysis of Supervised Remote Sensing Methods for Forest Identification in Pakistan
International Journal for Research in Applied Science and Engineering Technology
In order to overcome the deforestation rate and to track the growth rate of forests in Pakistan, automated monitoring methods need to be developed and adopted. An effective way of monitoring and identifying existing forest conditions is through remote sensing. We can monitor and observe the land cover through satellite. It is a challenging task because during the spring season competing for green fields also appears along with forests which make it difficult to differentiate from these green
... from these green fields like shrubs and bushes. In this paper, supervised classifiers are presented to classify the underlying land cover into different categories including forest. Specifically, a patch image of the Northern Pakistan region is obtained through SPOT-5 (2.5 meters) satellite imagery. Mahalanobis Distance Classifier and Maximum Likelihood classifier is executed on this and end results are compared in this paper. Maximum Likelihood achieved better classification results than Mahalanobis Distance classifier. Overall Accuracy of Maximum likelihood is 97.65% as compared to Mahalanobis distance which has 85.97% overall accuracy. Similarly, Maximum likelihood achieved forest's producer accuracy of 97% with reference to Mahalanobis distance in which we achieved forests Producer accuracy of 83%.