| Summary: | Glacial lakes are highly sensitive indicators of global climate change. The accurate extraction of glacial lakes from image data is of great importance to the evaluation of the hydrological environment in high mountain regions. This paper demonstrates a systematic approach—TSCV (threshold and simplified C-V)—that integrates the advantages of the threshold segmentation method and the active contour model to improve the effect of glacial lake extraction, by overcoming the problems involved in the efficient extraction of small, overlooked glacial lakes and in the removal of some mountain shadows. Three typical areas of glacial lake development in High Asia were selected for glacial lake extraction from Landsat-8 imagery and the results were then compared with those obtained using the threshold segmentation method and the fuzzy clustering method (FCM). The values of the kappa coefficient (KC) and average extraction accuracy (AEA) shows that the TSCV (KC = 0.895, AEA = 0.739) performed superior than other two methods, especially for the small glacial lakes, but also that the threshold segmentation method (KC = 0.869, AEA = 0.631) requires the least time and that the FCM (KC = 871, AEA = 0.663) has the lowest commission error. Our findings lead to a new approach to improve the accuracy of glacial lake extraction in situation where a large proportion of the lakes are small and with a potential for automated glacial lake mapping in rugged mountain areas at a large scale.
|
|---|