Deep learning-based GLOF modelling for hazard assessment and risk management

Glacial Lake Outburst Flood (GLOF) has become a crucial aspect as the increase in the meltdown of glaciers results in the breach of unstable debris dams. Hence, it is essential to understand the nature of the glacial lakes for proper planning and development of the region in the long term. In this paper, a deep learning network is developed for GLOF hazard and risk assessment. The Shepard Convolutional Neural Network Fused Deep Maxout Network (ShCNNFDMN) is developed by fusing the Shepard Convolutional Neural Networks (ShCNN) and the Deep Maxout Network (DMN) based on regression analysis. Here, various data and feature attributes, like geometric properties, location properties, lake-based properties, and global properties are determined from the glacial lake data. Afterthat, hazard assessment is carried out based on these parameters by the ShCNNFDMN. Then, risk assessment is performed based on the hazard levels and the feature attributes. The ShCNNFDMN is analyzed based on metrics, such as Hazard modelling error, Risk prediction error, Mean Average Error (MAE), and R-Squared are found to produce values of 0.462, 0.423, 0.358, and 0.288, respectively. The proposed method is useful in applications, like infrastructure planning, taking preventive and mitigative actions in downstream areas of glacier lakes.