Glacier mass balance estimation is important for understanding glacier responses to climate change and for assessing mountain water resources. Data-driven methods are widely used, but their cross-regional transferability remains unclear, especially in High Mountain Asia (HMA), where observations are limited. This study develops a unified framework to compare 16 machine learning and deep learning models across the European Alps and HMA. A degree-day-based monthly decomposition scheme is used to generate physically constrained monthly mass balance estimates. These are used as intermediate supervision signals. All models are trained at the monthly scale, and the outputs are aggregated to annual values for evaluation against observations. In transfer experiments, models are trained on Alpine data and tested in HMA. In joint-training experiments, different proportions of HMA samples are gradually added to the training set to assess the role of target-region information. Results show that machine learning models outperform deep learning models in cross-regional settings. Random Forest and K-Nearest Neighbors remain relatively stable under limited HMA data, while deep learning models are more sensitive to distribution shifts. Adding a small amount of HMA data improves annual prediction performance, highlighting the value of region-specific information. Overall, this study provides guidance for modeling glacier mass balance in data-scarce regions.
Under the background of global climate warming, glaciers in the Three-River Headwaters Region, as a crucial component of the “Asian Water Tower,” exert profound influences on regional water resource security and ecological stability through their mass balance variations. Due to the scarcity of in situ observations caused by the harsh high-altitude environment, long-term monitoring based on remote sensing techniques is urgently required. In this study, the geodetic method was employed, using the SRTM-C DEM acquired in 2000 as the reference, and recent glacier surface DEMs were generated from high-resolution ZiYuan-3 tri-stereo imagery obtained during 2024–2025. Through refined DEM co-registration, differencing, and systematic error corrections, the glacier mass balance in the Three-River Headwaters Region from 2000 to 2025 was systematically estimated. The results indicate that the glaciers in the study area exhibited an overall negative mass balance during the study period, with significant spatial heterogeneity. Among the sub-regions, the Lancang River source region experienced the most pronounced mass loss (−0.70 ± 0.07 m w.e. yr−1), whereas the Yellow River source region showed the lowest mass loss (−0.37 ± 0.09 m w.e. yr−1). Compared with earlier studies, glacier mass loss has accelerated in recent years and is closely associated with regional climatic characteristics. This study provides a scientific basis for understanding glacier changes and their hydrological and ecological impacts in the Three-River Headwaters Region.