Comparison of change detection techniques in Chitwan District of Nepal

This study explains the comparison of three change detection techniques employed for Chitwan district in Nepal by using remotely sensed images. Three change detection techniques were looked at, viz , Post classification comparison, image differencing and multitemporal composite and classification from the available images of Landsat TM for three different dates (1988, 1992 &2000). These techniques were analyzed independently, using the conce pt of well known procedures like supervised classification algorithm with maximum likelihood with additional post classification comparison, image differencing between two bands for two dates, and unsupervised classification for the multi temporal data sets as an iterative process. The application of such standard procedures was successfully implemented for detecting change with post classification comparison. The result of the post classification comparison gave thirteen classes with change versus no change out of 80 potential changes between the two dates (1992-2000). Some of these classes are rarely found while others have occurred due to error in earlier classification. Furthermore, how these change maps could also be expressed or visualized by incorporating graphs is also presented. Image differencing was successfully implemented in identifying the areas with respect to radiance values. Out of three techniques employed this method is the most straightforward method. The change map derived from a threshold determined from the histogram can be identified with change classes as ‘Change’ and ‘No Change’. The change map adapted from the multitemporal composite was also successfully generated. The success of such effort depends upon the number of clusters chosen to cover the spectral variation and consequently merging to meaningful change classes. The initial 25 clusters were aggregated to form 7 change classes for the two pairs of imagery (1988 -1992, 1988 -2000).