Land Cover Classification Using ICESat/GLAS Full Waveform Data

Abstract

In the present study, parameters derived from Ice, Cloud, and land Elevation Satellite/Geoscience Laser Altimeter System (GLAS) full waveform were used for land cover classification in western part of Doon valley, Uttarakhand, India. Three parameters, viz, height, front slope angle (afslope) and canopy return ratio (rCanopy) were extracted from the returned full waveform signals. k-means (KM), partitioning around medoids (PAM), and fuzzy c-means (FCM) with different cluster sizes were used for classifying the land cover types with the help of GLAS-derived parameters. Among the clustering methods, KM performed the best. The overall accuracy (89.41 %) of all methods were quite significant with cluster size three i.e. with three classes forest, mango orchard and other class including agriculture, barren/fallow land, settlement, dry river bed, etc. The accuracy of the PAM (60 %) and the FCM (68.4 %) decreased drastically at four clusters with the separation of agriculture from barren/fallow land. The accuracy of the PAM and the FCM further decreased with increase in the number of clusters whereas KM showed reliable results for all clusters. KM with five clusters was able to distinguish five different land covers, viz, forest, mango orchard, agriculture and barren/fallow land and other class including settlement, dry river bed, etc. with an overall classification accuracy of 72.93 %. The study presents a method for classifying land cover types using GLAS full waveform data.