Predicting the Unpredictable: IIT Guwahati Identifies 492 High-Risk Sites for New Glacial Lakes

Author : Yasaswini Sampathkumar

Published: 27 January 2026 Category: Research and Partnerships Office: Department of Civil Engineering

As climate change accelerates glacial melting, water accumulates in mountain depressions to form new lakes. When these lakes burst—an event known as a Glacial Lake Outburst Flood (GLOF)—the results are catastrophic.

While previous models focused almost entirely on weather and temperature, the IIT Guwahati team, led by Prof. Ajay Dashora, realized that the "bowl-shaped" topography of the mountains is just as important. Their research proves that land structures such as valleys, slopes, and existing lake basins are a primary predictor of where these hazards will appear.

A smarter forecast: the Bayesian advantage

The team used high-resolution satellite imagery and digital elevation models to test various predictive methods. They found that a Bayesian Neural Network (BNN) was the most reliable because it accounts for the natural unpredictability of mountain environments.

The model identified 492 potential new lake sites in the Eastern Himalayas by analyzing:

• Retreating Glaciers: Areas where ice is rapidly pulling back.

• Cirques and Basins: Natural hollows that act as catchments for meltwater.

• Surrounding Slopes: Identifying gentle gradients where water can safely pool versus steep areas prone to sudden overflow.

Safeguarding Infrastructure and Lives

This framework is more than a scientific study; it is a practical tool for "climate-resilient" planning. By pinpointing these 492 sites, authorities can now make informed decisions on:

• Early Warning Systems: Installing sensors at the most vulnerable new lake sites.

• Infrastructure Safety: Ensuring new roads, bridges, and hydropower projects are built away from high-risk flood paths.

• Water Management: Tracking how water availability will shift as glaciers move further up the mountains.

"Our framework can guide safer locations for settlements and hydropower," says Prof. Dashora. "It offers a practical tool for reducing risks to communities across the Himalayas and other glaciated regions globally."

Impact at a Glance: The Predictive Framework

Feature	New BNN Framework	Traditional Models
Primary Focus	Topography + Climate	Climate/Temperature only
Sites Identified	492 High-Risk Locations	Generalised regions
Reliability	High (captures terrain uncertainty)	Limited (overlooks landscape features)
Application	GLOF Early Warning & Infrastructure	General Climate Research

The findings, published in Scientific Reports, were co-authored by Ms Anushka Vashistha (IIT Guwahati) and Dr Afroz Ahmad Shah (Universiti of Brunei Darussalam). The team now plans to automate data preparation and include field-based validation to scale this model for global mountain ranges.

DOI: https://doi.org/10.1038/s41598-025-17401-7