Identification of Hazardous prone lakes using Remote Sensing and ANN applications in Western Himalaya

Abstract The Himalayan glaciers are observed to be declining for the past three to four decades, due to ongoing changes in climate. Among them, a few glaciers have favored the formation of (hazardous prone) lakes, which may burst and can cause severe damage to the downstream communities. Hence, these lakes need to be monitored continuously by identifying them. However, identification of these lakes is quite difficult because of limitations exist in a) accurate identification of pixels of lakes due to mixed pixels, b) existence of cloud cover c) classification error while applying automatic or semi-automatic approaches d) identification of movement of mass down slope due to coarse spatial resolution e) limited field measurements for validation etc. Also, Remote Sensing techniques alone may not be helpful for providing the solution for these types of catastrophic events. Therefore, advanced machine learning techniques such as Artificial Neural Network approach are preferred along with Remote sensing images for assessment of flood prone lakes. In this study, the glacial lakes are interpreted by using NDSI & NDWI approach using the latest satellite data of Landsat -8, and Sentinel -2a. A total of 355 glacial lakes are identified. However, due to above said limitations, manual interpretation is also carried out using high spatial resolution Google Earth pro images, and with this the initial count had increased from 355 to 655 lakes. Then, the lakes were classified into different categories namely Ice- dammed lakes, Moraine dammed lakes, Bed dammed lakes and other lakes. Further, the back propagation method of ANN algorithm is trained using the data from the spatial datasets and found that 18 lakes in the study region are more prone to hazards.