Fusing daily snow water equivalent from 1980 to 2020 in China using a spatiotemporal XGBoost model

Accurate snow water equivalent (SWE) data are crucial for estimating the water supply capacity of snow and are of profound importance for the communities of hydrology, climatology, and ecology. However, existing SWE products suffer from varying degrees of uncertainty, incompleteness, and inconsistency. This study proposes a novel spatiotemporal fusion model to estimate SWE by combining multisource SWE products (GlobSnow, ERAInterim, ERA5-Land, GLDAS, and CSWE) and spatiotemporal covariates (land cover, temperature, snow -related, geospatial, and temporal features) with the extreme gradient boosting (XGBoost) model. Specifically, the proposed spatiotemporal XGBoost model can encapsulate the advantages of multisource SWE products and address the nonlinear relations between SWE and its influencing variables. Furthermore, it embeds the geospatial and temporal features to capture the heterogeneities of SWE. Evaluation results on daily SWE records from 647 meteorological sites in China show that the proposed XGBoost SWE outperforms existing SWE products and other fusion methods (RF, SVM, RR, SA, and MLR), which demonstrates the effectiveness of the model with R values of 0.77, 0.70 and MAE and RMSE values of 7.54, 8.62 mm and 12.29, 13.73 mm based on cross -validation according to year and site, respectively. Using the proposed model, a dataset of daily SWE from 1980 to 2020 with a spatial resolution of 25 km over China is provided, aiming to deliver accurate SWE data for hydrological research and water resource management.