Accuracy of Northern Hemisphere Snow Water Equivalent Products

Daily continental-domain snow water equivalent (SWE) datasets are required for climate model evaluation, verification of seasonal forecasts, annual updates to climate assessments, and determination of freshwater availability. Numerous spatially continuous, gridded SWE datasets are available, derived primarily from passive microwave (PMW) remote sensing and snow models driven by reanalysis. However, evaluations of their accuracy have often been based on spatially and temporally limited observations and further constrained by differences in representative length scales between typical point observations and coarsely gridded data. We improve upon previous work by evaluating the accuracy of ten gridded SWE products across the Northern Hemisphere using a suite of snow course measurements from Canada, Finland, and Russia conducted over the 2002-2010 period. The evaluation shows that standalone PMW products (NASA AMSR-E v1 and v2) have lower accuracies compared to land surface models driven by reanalysis meteorology (GLDAS-2, ERA-Interim/Land, ERA5, MERRA, MERRA2, CROCUS driven by ERA-Interim meteorology) and those that constrain PMW retrievals with surface observations (GlobSnow v2 and the European Space Agency's recently released CCI SWE v1 which enhances the baseline GlobSnow algorithm). No single product performs best in all regions, rather, multi-product ensembles that exclude the standalone PMW products have better RMSE and correlation than any individual product suggesting that combining remote sensing with modelling and in situ observation may provide a path forward in SWE product development. We further extend this analysis by targeting specific components of the CCI algorithm to assess the degree to which in situ snow depth observations, as opposed to information from the passive microwave measurements, influence estimated SWE. This understanding should aid future development of combined remotely sensed, modelled, and in situ SWE datasets.