Identification and correction of snow depth bias in ERA5 datasets over Central Europe using machine learning

Abstract. Accurate estimation of snow depth is a crucial problem from both meteorological and hydrological points of view. Global and regional reanalyses still struggle to address it, mostly because the scale of snow spatial heterogeneity is widely beyond current resolutions of the databases. In the study, snow depth estimation from Copernicus reanalyses ERA5 and ERA5-Land are compared and evaluated against point measurements in Poland, Czech Republic and Slovakia in winter seasons 2001/2002–2020/2021. Additionally, a Random Forests (RF) model is developed to reduce identified errors based on various environmental variables and parameters derived from the reanalyses and a digital elevation model. As mountains are main snow water reservoirs for Central Europe, the model is then used to spatially downscale snow depth over a fine-scaled subdomain in mountainous terrain.

For both reanalyses, the deviations are relatively small in flat or gently rolling terrain (below 500 m a.s.l.). ERA5 (0.25°) outperforms ERA5-Land (0.1°) due to the presence of data assimilation. Since only synop measurements are assimilated, errors are the lowest for these stations, however, lower-ranked stations are also affected. In more complex terrain, both reanalyses exhibit an underestimation of snow that increases with elevation. In this area, ERA5-Land is slightly less biased due to its higher resolution and the fact that observations from mountainous sites are often masked out from the data assimilation in ERA5. The proposed RF model improves accuracy of estimation by around 48% with respect to the best reanalysis. The results of spatial downscaling certainly provide added value to the problem of snow estimation in complex terrain. Although they cannot be considered entirely valid and reliable since not all factors determining spatial variability of snow at such resolution are taken into account, they might be useful for future studies concerning, e.g., climatological variability of snow with respect to altitudinal zonation.