Glacier lakes detection utilizing remote sensing integration with satellite imagery and advanced deep learning method

Abstract

The Himalayan glaciers are extremely susceptible to global climate change, leading to substantial glacial retreat, the creation and expansion of glacial lakes, and a rise in GLOFs. These alterations have changed river flow patterns and moved glaciers' borders, resulting in significant socioeconomic damage. Accurately monitoring glacial lakes is essential for managing GLOF events and evaluating the effects of climate change on the cryosphere. This study utilizes a Deep Learning-based U-net technique to extract glacial lakes from Landsat-8 satellite imagery by propagating characteristics and minimizing information loss. The method improves the importance given to glacial lakes, reduces the influence of low contrast, and handles different pixel categories. We applied this methodology to the Chandra-Bhaga basin, Himachal Pradesh, located in NW Indian Himalaya, and successfully extracted 107 glacial lakes. The U-net model attains an accuracy of 97.32%, precision of 95.98%, recall of 95.23%, MSE 0.0043, Kappa Coefficient 97.43% and an IoU of 97.45% during validation with high-resolution photos from Google Earth and a digital elevation model. The suggested approach could be beneficial for precise and effective monitoring of glacial lakes in different areas, assisting in the management of natural disasters and offering vital information on the effects of climate change on the cryosphere.