Integrating ICESat-2 photon-counting lidar with multispectral imagery for bathymetry retrieval and water storage estimation in Bosten Lake

Study region: Bosten Lake, largest freshwater lake in Xinjiang. Study focus: In arid regions, lake water storage is fundamental to ecological stability and oasis sustainability. Bosten Lake lacks sufficient research in bathymetric inversion methods and water storage quantification. This study denoised ICESat-2 (ATL03) data using DBSCAN, performed refraction correction, and identified bathymetry retrieval variables through Landsat 8/Sentinel-2 and ICESat-2 correlation analysis. On this basis, empirical and Random Forest models for bathymetry inversion were developed, evaluated their accuracy with in-situ measurements and track profiles, and generated the lake’s underwater topography and calculated its inherent water storage. New hydrological insight for the region: The results show that the Landsat 8 Dual-Band Logarithmic Ratio Model performed the highest overall accuracy, with the LnB7/LnB3 model performing best (MAE = 0.89 m, RMSE = 1.27 m, 20.35% average photon consistency within ± 0.50 m of predicted depth). Landsat 8-based models outperformed Sentinel-2 counterparts, and quadratic polynomial structure further improved Landsat-based inversion accuracy. Landsat 8 model errors were generally concentrated (uncertainty increased without prior constraints), while Sentinel-2 errors clustered in shallow nearshore areas. Bosten Lake's inherent water storage was estimated at 6.55 × 10⁸ m³ , with max/mean depths of 11.28 m and 7.94 m. This study provides data support for hydrological parameters extraction in Bosten Lake and contributes to underwater topography and water storage estimation research.