Estimation of snow depth from AMSR-2 based on an AutoML method over the Qinghai-Tibet Plateau

Abstract

Snow depth is a crucial parameter for describing the spatiotemporal variations of snow cover, and passive microwave snow depth products (10–25 km) are widely used for monitoring snow depth changes. However, as one of the three major snow-covered regions in China, the Qinghai-Tibet Plateau has complex terrain and rapid changes in snow cover with strong spatial heterogeneity, making it difficult for coarse-resolution snow depth products to accurately describe its spatiotemporal characteristics. This study proposes a high spatial resolution (500 m) snow depth estimation method based on the Advanced Microwave Scanning Radiometer 2 (AMSR-2) brightness temperature data and Automated Machine Learning. Firstly, using Pearson correlation coefficients, 19 key factors influencing snow depth, including AMSR-2 brightness temperature, slope, and surface roughness, were selected as input data (independent variables) for Automated Machine Learning. Meanwhile, passive microwave downscaled snow depth data and ground-based snow depth measurements were introduced as dependent variables for Automated Machine Learning. The Automated Machine Learning model was then trained separately for four different types of snow cover surfaces (forest, grassland, water, and bare land). Finally, through ten-fold cross-validation, the optimal machine learning model for snow depth estimation under each type of underlying surface coverage was selected, thus generating sequential snow depth datasets for the ten-year snow cover period of the Qinghai-Tibet Plateau from 2012 to 2021. Results show that (1) the estimated snow depth values well with ground-based observations, yielding a coefficient of determination (R2) of 0.71 and a root mean square error (RMSE) of 3.64 cm, indicating high estimation accuracy. (2) Snow depth estimation demonstrates the highest accuracy in unused land (CatBoost, R2=0.82), followed by grassland (CatBoost, R2=0.77, RMSE=3.11 cm), water (ET, R2=0.75, RMSE=2.20 cm), and forest (XGBoost, R2=0.71, RMSE=3.30 cm). (3) A comparison with snow cover extent derived from Landsat-8 optical imagery reveals that the estimated snow depth spatial distribution is consistent with snow cover extent, providing reliable data for monitoring snow cover changes in mountainous regions.