Identification of Unstable Glacier Flow in the Western Tibetan Plateau and Karakoram Using Machine Learning

Abstract A novel method was developed for rapidly and accurately identifying unstable glacier flows, including pulse and surge events. The identification method was based on machine learning and high‐frequency glacier velocities derived from short time interval Sentinel‐1 Synthetic Aperture Radar (SAR) data. The method was built and tested using 32 glaciers in the eastern Pamir Plateau and applied it to 13 glaciers in western Karakoram. The test results showed that the success rate of identifying speedup events was more than 90% compared to manual interpretation results derived using velocity change maps and DEM differencing. The application results showed that there were some neglected pulse events, including two pulses on the Kukuar Glacier in 2017 and 2019, one pulse on the Karambar Glacier in 2017, and one pulse on the Ghulkin Glacier in 2016. Moreover, a complete surge event with multiple scattered pulses on the Shispare Glacier in 2018–2019 was also detected. The speedup events of the Shispare Glacier indicated the possibility of mutual transformation between pulse and surge events due to the influence of glacier basal roughness. In all unstable flows occurring within the test and application areas, pulse events could be observed on the nonsurging glaciers during the active phase and quiescent phase of the surge‐type glaciers, which might suggest the randomness of pulse events. Machine learning method made it possible to identify glacier speedup events in large areas with high efficiency and low manpower.