Analysis of Land Cover Change and Misregistration Error in North District of Sikkim, India

要旨

Satellite remote sensing provides an important basis for studying routine land cover dynamics, especially in difficult and inaccessible terrain with complex physiography. Land cover change and patch analysis was done for the hilly terrain of North District, Sikkim state between the period of January 1997 and November 2002. 10-land cover classes were derived using unsupervised classification technique and change analysis was done following post-classification comparison method. Majority of the land cover changes were attributed to seasonal snow cover, agricultural activities and deforestation. Analysis of mean patch size and mean patch shape indicates spatial heterogeneity of the landscape. The mean patch shape of forest cover was observed to be higher than rest of the land cover categories. Patch size distribution revealed that snow cover, dense forests and forest blanks occur with higher degree of contiguity. In North district, the altitude ranges between ca 350m and 8588m (height of Mount Kanchenjunga) in a small area of about 4295 km2, where image-to-image misregistration is imminent in absence of an accurate digital terrain model. This became the basis for calculating the misregistration bias in land cover changes. An estimation using PAR (perimeter area ratio) as an index revealed that highly dispersed and adjacent patches have higher misregistration bias and vice versa. The land cover change accuracy has dropped to a maximum of 11% within 1.04 (~1) pixel of misregistration as revealed by the study. This method of PAR provides one of the simplest and definite methodologies in calculating misregistration bias in post-classification based change detection exercise.