Chapter 36 - GLOF Early Warning System: Computational challenges and solutions

Abstract

Due to climate change and rapid anthropogenic activities, the frequency of flash floods has become a major challenge. Glacier Lake Outburst Floods (GLOF) is a source of flash floods, where an unstable moraine dam collapses and the huge volume of water stored behind them are released downstream in a very short time. This calls for an early warning system to monitor the vulnerable glacier lakes and provide robust tools to the disaster managers to plan mitigation, thus saving life and property. Critical parameters to be considered are forecast lead time, robust and stable computation of water flow, automatic data intake and output with minimum human interference, user-friendly interface, and de-cluttered data visualization for effective decision-making. The GLOF Early Warning System (EWS) was envisaged to create and manage a comprehensive GIS database of glacier lakes in Sikkim, study their growth over time using remote sensing data, asses their vulnerability to GLOF, and classify and prioritize them based on scientific criteria. Deployment of real-time automatic water level monitoring sensors at vulnerable lakes, transmitting the sensor data to base station through satellite communication, a GLOF simulation model, and spatial decision support system make the GLOF EWS first of its kind in India. The GLOF simulation model predicts the flood height and arrival time at villages downstream when GLOF occurs. GIS-based Spatial Decision Support System helps the disaster managers to plan evacuation and assess impact. The system is built on open source technologies to reduce development and upgrade costs. The scalable simulation model is capable of handling multiple lake-burst scenarios. As a pilot study, two vulnerable, moraine-dammed glacial lakes were selected for real-time GLOF monitoring. Parallel processing capability enhances forecast lead times. The simulation model was tested on 15 Nodes of HPC system, with a flood forecast lead time of 4.5 h, which is significant in such situations. With more number of nodes, the lead time can be improved.