Reply on RC3

Abstract

In light of the recent climate warming, monitoring of lake ice in Arctic and sub-Arctic regions is becoming increasingly important. Many shallow arctic lakes and ponds of thermokarst origin freeze to bed in the winter months, maintaining the underlying permafrost in its frozen state. However, as air temperatures rise and precipitation increases, less lakes are expected to develop bedfast ice. In this work, we propose a novel temporal deep learning approach to lake ice regime mapping from synthetic aperture radar (SAR) and employ it to study lake ice dynamics in the Old Crow Flats (OCF), Yukon, Canada over the 1993 to 2021 period. We utilized a combination of Sentinel-1, ERS-1 and 2, and RADARSAT-1 to create an extensive annotated dataset of SAR time-series labeled as either bedfast ice, floating ice, or land, used to train a temporal convolutional neural network (TempCNN). The trained TempCNN, in turn, allowed to automatically map lake ice regimes. The classified maps aligned well with the available field measurements and ice thickness simulations obtained with a thermodynamic lake ice model. Reaching a mean overall classification accuracy of 95 %, the TempCNN was determined to be suitable for automated lake ice regime classification. The fraction of bedfast ice in the OCF increased by 11 % over the 29-year period of analysis. Findings suggest that the OCF lake ice dynamics is dominated by lake drainage events, brought on by thermokarst processes accelerated by climate warming, as well as fluctuations in water level and winter snowfall. Catastrophic drainage, and lowered water levels cause surface water area and lake depth to decrease and lake ice to often transition from floating to bedfast ice, while a reduction in snowfall allows for the growth of thicker ice.