An automated method for glacial lake mapping in High Mountain Asia using Landsat 8 imagery

Abstract

Glacial lakes in the High Mountain Asia (HMA) are sensitive to global warming and can result in much more severe flood disasters than some largesized lakes. An accurate and robust method for the extraction of glacial lakes is critical to effective management of these natural water resources. Conventional methods often have limitations in terms of low spectral contrast and heterogeneous backgrounds in an image. This study presents a robust and automated method for the yearly mapping of glacial lake over a large scale, which took advantage of the complementarity between the modified normalized difference water index (MNDWI) and the nonlocal active contour model, required only local homogeneity in reflectance features of lake. The cloud computing approach with the Google Earth Engine (GEE) platform was used to process the intensive amount of Landsat 8 images from 2015 (344 path/rows and approximately 7504 scenes). The experimental results were validated by very high resolution images from Chinese GaoFen-1 (GF-1) panchromatic multi-spectral (PMS) and appeared a general good agreement. This is the first time that information regarding the spatial distribution of glacial lakes over the HMA has been derived automatically within quite a short period of time. By integrating it with the relevant indices, it can also be applied to detect other land cover types such as snow or vegetation with improved accuracy.