Characterizing Ice Phenology Variability of Small Lakes in the Cascade Mountain Range Using Daily Ice Presence Probabilities from Satellite Remote Sensing

Abstract

Lake ice is a reliable indicator of climate variability and change. This research investigates annual and spatial variations in lake ice phenology of small lakes (<2 km2) in the Cascade Mountain Range (U.S.A.) using high-temporal frequency remote-sensing data and a probabilistic model. Data from the Harmonized Landsat-Sentinel-2 (HLS-2) project were analyzed to detect the presence of lake ice for 2872 small lakes and characterize ice phenology during each of the 10 ice seasons (2014–2023). A multiple logistic regression model (Accuracy ~0.80 and AUC 0.88 for cross-validation) was used to predict the probability of ice for every non-imaged day during each ice season. From the time-continuous combination of HLS-2 observations and modelled ice presence, we extracted three ice phenology metrics: ice-on date, ice-off date, and total number of ice days (TID), and we analyzed their variabilities across ice seasons and across the Cascade Range. Both ice-on date and ice-off date distributions exhibited high inter- and intra-season temporal variability. All median TID fell between 100 and 150 days. Our research quantifies the variability of a large sampling of small alpine lakes in a climate-sensitive region and demonstrates the efficacy of extracting ice phenology metrics from daily ice presence probabilities based on remote-sensing data.