Research on GNSS-R Snow Depth Inversion Based on Deep Learning Method

Snow cover plays a vital role in meteorology, hydrology, atmospheric circulation and ecosystem, and is closely related to human life. Research on how to accurately obtain snow depth and monitor its change, as well as the correlation between the change of snow depth and global warming or sea level rise, is the focus of current research. For a long time, the method of combining passive microwave remote sensing data with on-site snow cover data is often used to invert snow depth. However, in the case of sparse weather stations, the inversion accuracy of the method is limited, and the high-quality application of the inversion results is also hindered in specific areas. Ground-based Global Navigation Satellite System Reflection Measurement (GNSS-R) is currently a potential method for monitoring changes in snow depth, with features such as high precision and real-time observation. In this paper, we combined satellite observations to estimate snow depth and trained and tested data obtained in Alaska in 2017 using a random forest network model. The results show that the GNSS-R snow depth inversion method combined with random forest network model is more accurate than the conventional single snow depth inversion calculation. The network model is superior to other neural network models, and can effectively combine GNSS-R method to obtain the snow depth inversion results with high precision, and the RMSE is reduced by 17.02cm compared with other methods. By analyzing the relationship between the inversion of snow depth and air temperature, it is found that the change of snow depth in January and May of 2017 in Alaska is closely related to air temperature. The results of this paper prove that GNSS-R snow depth inversion technology based on deep learning method has wide application value and prospect although the space coverage is limited.