How well do global reanalyses estimate snow density?

Snow density is critical to the accurate simulation of energy exchange between the atmosphere and the underlying soil. Along with snow water equivalent (SWE) it is a key constraint on snow depth (HS). The snow density formulations used in global reanalyses have been developed and validated using detailed observational datasets but often from very limited locations. Unlike SWE and HS from global climate reanalyses which have been comprehensively evaluated using in situ observations, to our knowledge there are no similar hemispheric-scale assessments of snow density from these products. To address this gap, we use snow course observations from the NorSWE dataset to simultaneously evaluate SWE, SD, and snow density in five reanalysis products (ERA5, ERA5-Land, GLDAS2.1 Noah, JRA-3Q, MERRA2). We consider snow across the Northern Hemisphere over a range of snow classes and assess its seasonal evolution across the snow onset, peak and melt periods.Results show a large spread in snow density both in terms of its spatial pattern and average magnitude. Products that can reasonably estimate SWE and/or HS do not necessarily have accurate snow density representations and vice versa. For example, MERRA2 ranks in the middle of the assessed products in terms of SWE and HS skill but its snow densities are poorly correlated with observations, whereas GLDAS 2.1 (Noah 3.6) has some of the largest SWE and HS errors, but some of the smallest snow density errors. Inaccuracies in SWE, HS and density can compensate for each other in different ways and these relationships vary between reanalyses. By examining snow density alongside SWE and HS, we aim to diagnose the principal sources of error in reanalysis snow estimates as stemming from errors in the snow model and its structural implementation within the reanalysis or from biases in meteorological forcing.