Prediction of Snow Depth Based on Multi-Source Data and Machine Learning Algorithms

As an important element of the earth's surface, snow cover plays an important role in the global terrestrial ecosystem, climate change, water cycle and energy cycle. Snow depth (SD) provides information on the spatial distribution of snow cover and material energy information. It is also used to study the climatic effects of snow cover, water balance in the basin, snowmelt runoff simulation, and monitoring and evaluation of snow disasters. Snow depth data has become an indispensable basic supporting data in multi-disciplinary research. However, the current snow depth data is relatively poor in completeness and consistency and cannot meet the needs of related scientific research and industry applications, which will bring great confusion to users. This study attempts to use machine learning methods to effectively integrate snow depth data products from multiple sources to obtain a snow depth data set with high consistency in China. This paper chooses the passive microwave remote sensing data (WESTDC), ground-based data (Canadian Meteorological Centre, CMC) and Land surface models (Global Land Data Assimilation System, GLDAS; the NASA Modem-Era Retrospective Analysis for Research and Applications MERRA2; the European Centre for Medium-Range Weather Forecasts Interim Reanalysis, ERA-Interim) as the main SD data source of random forest model, and considering theinfluencing factors (e.g., land cover, snow class, forest cover fraction, surface roughness). The results show that the random forest fusion model method can effectively gather the advantages of each data source, improve the accuracy of snow depth estimation, and reduce the inconsistency between snow depth data from multiple sources. The correlation coefficient between the fused snow depth dataset and the observed snow depth can reach 0.87, and the root mean square error is 5.1 cm. Therefore, the multi-source snow depth fusion using random forest model can improve the accuracy of snow depth estimation.