Comparisons of different methods for debris covered glacier classification

摘要

The paper outlines comparisons between different methods for the mapping of debris covered glaciers. The supervised classification method like Maximum Likelihood Classifier (MLC) has been tested using different data set for deriving the glacier area. Along with MLC the semi-automated method like Hierarchical Knowledge Based Classifier (HKBC) has also been used here. All the results were tested for accuracy, processing time and complexities. The results were also tested against the manually digitized boundary. The results suggests that the MLC when used with other ancillary data like geo-morphometric parameters and temperature image takes slightly more time than HKBC due to some to higher amount of post processing time but the output is satisfactory (89 % overall accuracy). Results show that the time taken in different classifications is significantly different which ranges from 1-2 hours in MLC to 5-10 hours in manual digitization. Depending on the classification method, some to large amount of post processing is always required to achieve the crisp glacial boundary. Classical classifier like maximum likelihood classification is less time consuming but the time taken in post-processing is higher than HKBC. Another factor which is important for a better accuracy is the prior knowledge of glacier terrain. In knowledge based classification method, it is required initially to establish crisp rules which are later used during classification, without this per-classification exercise the accuracy may significantly decrease. This is a time consuming procedure (2-3 hours in this case) but a minimal amount of post-processing is required. Thermal and geo-morphometric data when used synergistically, classified glacier boundaries are more crisp and accurate.