Multi-scale remote sensing-based extraction and freeze–thaw dynamics of Qinghai Lake ice

Qinghai Lake, China's largest inland saline lake, has freeze–thaw dynamics crucial for climate change and ecological monitoring. This study extracted lake ice using multi-scale remote sensing data (MODIS, Planet, UAV) and analyzed freeze–thaw responses to meteorological factors. Comparing spectral threshold and Support Vector Machine (SVM) classification, SVM achieved superior accuracy (>95% overall) across all data, especially with low-resolution MODIS (250 m). Scale effect analysis via resolution downscaling (3 m Planet to 250 m MODIS-equivalent resolution) reveals that SVM maintains high precision with <5% ice area deviation at coarse resolutions, validating MODIS utility for large-scale monitoring despite mixed pixel challenges. Spatially, driven by local bathymetric gradients and geomorphic conditions, Qinghai Lake ice initiates freezing in shallow nearshore zones, exhibiting edge-to-center and east-to-west expansion patterns during freezing, while melting begins in the west and rapidly propagates eastward, with wind-driven breakup accelerating melt in the shallow eastern region. The freeze–thaw process exhibits a 3-day lag to air temperature and is modulated by wind speed/direction; extreme wind events alter patterns. These findings underscore SVM's robustness for climate-driven lake ice phenology studies on the Tibetan Plateau, offering critical insights for ecological monitoring and hydrological modeling under global warming.