Multi-Source Remote Sensing-Based Reconstruction of Glacier Mass Changes in Southeastern Tibet Since the 21st Century

Abstract. Glaciers are a crucial freshwater resource and a key indicator of climate change. However, tracking annual changes in glacier inventories remains a significant challenge due to persistent cloud cover and seasonal snow accumulation. The increasing availability of satellite data, particularly from the Sentinel series, has greatly enhanced glacier monitoring capabilities. In this study, we developed an ensemble learning-based random forest classifier using data from Landsat, Sentinel-1, Sentinel-2, and NASADEM to automatically delineate glacier extents in southeastern Tibet from 2016 to 2022, achieving the first annual-resolution glacier inventory in the region. To extend the time series to 2000, we manually constructed glacier inventories for 2000, 2005, 2010, and 2015 by integrating a three-year dataset centered on each target year, addressing the limitations posed by the absence of early Sentinel data. Our results reveal a consistent decline in glacier area, from 7898.61 ± 652.15 km² in 2000 to 6317.13 ± 592.57 km² in 2022, with an average annual loss of 85.03 ± 7.60 km²/y. Notably, the retreat rate accelerated after 2010, increasing from 57.72 ± 16.81 km²/y (2000–2010) to 97.72 ± 17.67 km²/y (2010–2022). By integrating satellite altimetry data, we calculated the glacier mass balance using dynamically updated glacier areas, resulting in an annual mass loss 6.20 ± 1.16 Gt/y. Correlation analysis between glacier thickness and area changes showed a strong positive relationship (R² = 0.89, p < 0.001). This study provides a novel approach to high-temporal-resolution glacier assessments by incorporating annual dynamic glacier areas into mass balance calculations. The improved accuracy of these estimations offers a refined understanding of cryosphere changes in southeastern Tibet, underscoring the urgency of monitoring glacier dynamics in response to climate change.