Snow depth estimation in Northeast China based on space-borne scatterometer data and ML model with optimal features

Real-time snow depth (SD) monitoring is crucial for water resource management, weather forecasting, and environmental research. Current SD estimation methods primarily rely on passive microwave or synthetic aperture radar (SAR) technologies. As an active microwave sensor, the HaiYang-2B (HY-2B) spaceborne scatterometer with a Ku-band is used for wind field assessment and polar sea ice detection. This study explores its application to SD estimation in Northeast China. Multiple machine learning (ML) models, including support vector regression (SVR), k-nearest neighbors (KNN), XGBoost, and random forest (RF), were deployed and contrasted for SD estimation. The multiple correlation coefficient feature selection (MCCFS) algorithm proposed in this study can effectively eliminate redundant features by retaining the features that have the most significant impact on SD estimation. Furthermore, feature ablation experiments were conducted to obtain the optimal features. In comparison to the public SD product and the ground-based SD measurements, the experimental results using the RF model with optimal features demonstrate superior SD estimation performance, yielding a root mean square error (RMSE) of 3.91 cm, mean absolute error (MAE) of 2.27 cm, and an R2 of 0.80. This study indicates that spaceborne scatterometer data hold promise for SD estimation in snow monitoring applications.