Laurence C. Smith (Brown University, United States); Thomas Howard (Brown University, United States); Cassie Stuurman (NASA JPL, United States)

Accurately monitoring the seasonal freeze-thaw state of lakes and rivers is essential for understanding water resources and climate feedback in cold regions. However, observing lake ice transitions across remote high-latitude regions remains challenging, emphasizing the need for reliable satellite-based approaches. The Surface Water and Ocean Topography (SWOT) satellite mission provides high-resolution measurements of water surface elevations for lakes and rivers across the globe, offering new opportunities to enhance our understanding of freshwater dynamics. In high-latitude regions, interpreting SWOT’s measurements requires distinguishing between open-water and ice-covered water surfaces, yet SWOT lacks a reliable native method for this discrimination. This study evaluates the capability of SWOT data (i.e., backscatter, water surface elevation, and incidence angle) to effectively differentiate between ice-covered and open-water lakes in Alaska. We show that machine learning classifiers (e.g., random forest, gradient boosting) accurately distinguish ice-covered from open-water lakes, with backscatter emerging as the most informative predictor. Our findings affirm SWOT's potential to detect lake ice cover and extend its utility beyond water level monitoring. More broadly, this work highlights the value of integrating new satellite observations with machine learning to advance the monitoring of Earth’s changing freshwater systems.