Snow depth mapping in agricultural areas of Northeast China based on deep learning and multi-temporal Sentinel-1 data

Snow is an essential element in surface climate studies. Snow measurement using remote sensing data gradually has become mainstream with the continuous development of remote sensing technology. Synthetic aperture radar (SAR) has the capability of all-day, all-weather ground observation, and combines high spatial resolution, interferometry, and polarization imaging. C-band SAR images are more sensitive to snow characteristics and are an effective data source for obtaining the spatial distribution of snow in areas with complex terrain. In recent years, the methods of using machine learning and deep learning to invert snow parameters such as snow depth and snow water equivalent have been widely used. In this paper, the study area is the farmland area of Northeast China. The relationship between snow depth and snow parameters is analyzed by a deep learning algorithm and machine learning algorithm using multi-temporal Sentinel-1 Cband SAR data, measured snow depth data, and so on. The inversion model based on the type of farmland subsurface is established to invert the snow depth in the farmland area of northeast China and output the map. The aim is to generate surface-based snow depth inversion results with less error and higher accuracy. The average MAE of the model is 2.06cm, RMSE is 2.96 cm, R² is 0.72, the maximum absolute error is 10.206 cm, and the minimum absolute error is 0.045cm for the farmland mixed group feature screening dataset.