AI for Good

Chapter 12Mapping Glacial Lakes

—Anthony Ortiz, Kris Sankaran, Finu Shrestha, Tenzing Chogyal Sherpa, and Mir Matin

Executive Summary

Understanding the speed and impact of climate change is critical for mitigation efforts. Particular parts of the world are more susceptible to climate change than others and can act as a barometer of the pace of climate change. Glacial lakes are one of those environments and can represent humanitarian threats: when excessive melting weakens the ice shores of those lakes, floods can occur and cause downstream harm. Here, to better understand and measure climate change, we compared several approaches to mapping glacial lakes in the Hindu Kush Himalayas with the anticipated impact that automated mapping could support risk assessments of Glacial Lake Outburst Floods that pose a risk to communities and infrastructure in valleys below glacial lakes.

Incorporating labels from a 2015 survey using Landsat 7 ETM+ SLC-off imagery to guide segmentation on newer higher-resolution satellite images like Sentinel-2 and Bing Maps imagery, we compared our models to previously used ones. We found that a historically guided version of U-Net and a properly initialized form of morphological snakes most effectively segmented the lakes, each providing between an 8–10 percent intersection over union improvement over existing U-Net segmentation approaches. An error analysis highlighted the strengths and limitations of each method. To make our analyses useful, we designed ...