A Large-Scale Snow Depth Retrieval Method for Alaska Based on Point-Surface Fusion and Random Forest Model

Abstract

Accurate snow depth (SD) monitoring is crucial for understanding climate change and managing water resources. However, due to the sparse distribution of meteorological stations and the limited accuracy of passive microwave remote sensing data, the retrieval accuracy of large-scale snow depth in regions with complex terrain and variable climate conditions has faced significant challenges. To address this issue, this paper proposes a large-scale snow depth retrieval method based on point-surface fusion technology with the Random Forest (RF) model. The method integrates ground-based snow depth measurements with passive microwave brightness temperature data using the RF algorithm and incorporates geographic coordinates, elevation, brightness temperature, brightness temperature gradient differences, and time variables for each grid cell in Alaska, which significantly improves the accuracy and spatial resolution of the large-scale snow depth retrieval. Five-fold cross-validation results show the model exhibits excellent fitting performance (R $^{2} =0.9627$ , MAE =4.6 cm, RMSE =10.08 cm), particularly demonstrating strong robustness in sparse meteorological stations. The results indicate that the proposed method effectively captures the spatiotemporal variations in snow depth across Alaska from 2008 to 2016, providing valuable technical support for snow depth monitoring and climate change research in cold regions.