Early detection of emergent high-mountain lakes using Sentinel-1

The mountain cryosphere is rapidly changing in the 21st century. Glacier mass loss is accelerating, permafrost is thawing, and precipitation falls increasingly as rain. These are fundamental changes to the hydrological functioning of high mountain catchments, including throughput and storage. There has been focused attention on the progressive development of large glacial lakes, due to their high visibility and impacts in the case of an outburst flood. Satellite-based monitoring of such sites is a relatively straightforward task, yet effective evaluation and early warning of hazards remains challenging, as it must take into account site specific considerations, the possible chains of geomorphic processes, and implementation challenges for technical, logistical, and community ownership aspects.  In recent years, however, outburst floods have increasingly occurred originating not from known large proglacial lakes, but from glacial and periglacial lakes that have developed on subseasonal timescales of weeks to months. The development of water bodies over these short timescales is often entirely unnoticed before a flood; as a result, these events have caused significant damage and fatalities, as evidenced in at least 8 distinct events in 2025. Early detection of these emergent water bodies is essential and requires a rapid-repeat operational tool that is reliable and insensitive to meteorological conditions.

In this work we develop a framework for rapid identification of anomalous water bodies using Sentinel-1 GRD data in Google Earth Engine. Sentinel-1 is frequently used in inundation mapping efforts due to the low radar backscatter of water, its cloud penetration, and short temporal revisit (6 days for 1A and 1B combined). However, the SAR data present specific challenges: distinct ascending and descending view geometries, difficulties in geocoding due to mountain terrain and glacier thinning, an uncertain backscatter threshold for lake detection, and confounding signals of wet snow, soil moisture, or seasonal vegetation development. Rather than attempting to map lakes directly, we overcome these challenges by instead identifying domains with an anomalous signal that is also consistent with water. Specifically, we determine the anomaly of the latest backscatter data to historic backscatter phenology in each pixel, expressed as a composite z-score for the ascending and descending orbits, and mapped only in domains where radar backscatter and topographic slope are both low. 

We implement this approach in our tool THAW (Transient Hydrologic Anomalies Weekly) and demonstrate its operational utility by examining the detections that would have occurred preceding the Purepu (Tibet/Nepal) and Rawoshan (Pakistan) events in 2025. In both cases, THAW identified an anomalous signal likely to indicate surface water weeks and months prior to the lakes’ drainage. The approach is not sensitive to seasonally wet snow, as it accounts for the location’s typical seasonality, but highlights the early seasonal melt of Himalayan snowpacks in 2025. Our tests at Purepu identified the growth of another lake at Nyanang Phu (Tibet), enabling an early in-situ assessment by authorities. We find this framework to complement operational lake monitoring workflows by highlighting selected domains of rapid change for expert evaluation.