Causal-Driven Spatial-Temporal Modeling for Enhanced Glacial Lake Outburst Flood Prediction

Abstract

Communities nearby are at serious risk from Glacial Lake Outburst Floods (GLOFs), which are becoming more frequent as a result of climate change-induced accelerated glacier retreat. To save lives and livelihoods, GLOF forecasting is crucial. The goal of this study is to use cutting-edge deep learning methods to enhance GLOF prediction. It uses models such as Convolutional Long Short-Term Memory (ConvLSTM) and Long Short-Term Memory (LSTM) to forecast the primary cause, Glacial Lake Outburst. A thorough GLOF dataset is also used to test models that incorporate Granger’s causality with ConvLSTM and LSTM. ConvLSTM networks, designed for spatiotemporal data, capture the relationships between glacial lake behaviour and climatic factors. Granger’s causality enhances input selection, identifies important predictors, and facilitates the combination of GLOF probability estimation and lake evolution forecasting. Together, these methods support a strong early warning system for improved preparedness for disasters