Detection of weekly buried meltwater in Greenland Ice Sheet with Sentinel-1 imagery

Recent studies show that buried meltwater (BM) beneath the Greenland Ice Sheet (GrIS) can persist through winter, influencing the englacial hydrological system and thermal regime. However, previous research has primarily focused on annual distributions, lacking insight into its seasonal evolution. This study introduces a deep learning network, UNet-BM, to obtain weekly BM distributions across the GrIS from 2016 to 2020. However, little is known about the weekly BM distributions and its dynamic variations. Validation results demonstrate its effectiveness, achieving an overall accuracy of 99.87%, precision of 0.83, and a kappa score of 0.77, demonstrating superior performance compared to previous studies. The results reveal an average of 713 BM bodies per year, covering 523.94 km², predominantly located in marginal areas at elevations of 800–1700 m. Over 80% of BM bodies are concentrated in the Center-West, North-West, and South-West basins. The reduction rates of BM range from 4.8–15.5 km² per day. Approximately 27.22% of BM bodies disappear due to freezing or drainage, with higher disappearance rates in the northern GrIS (52.33% in the North and 37.2% in the Northeast). These findings enhance our understanding of BM dynamics and demonstrate the value of deep learning for monitoring englacial hydrology.