Benchmarking Snow Products under Imperfect Observational Constraints: The Case of ERA5-Land across Canada and Alaska

Abstract. Gridded snow products enable assessment where observations are sparse, but benchmarking their fidelity is complicated because the snow state is imperfectly observed. We formulate benchmarking as an evidence-based inference problem using ERA5-Land Snow Cover Fraction (SCF), snow density (Rho), Snow Depth (SD), and Snow Water Equivalent (SWE) across Canada and Alaska. ERA5-Land and competing products are harmonized to 25 km and evaluated against satellite retrievals and in-situ observations at daily and monthly scales. Fidelity is assessed using goodness-of-fit measures for time-series reconstruction and relative differences in long-term mean, trend, and detrended interannual variability. Fidelity varies with terrain, measurement technique, and observational density. ERA5-Land generally represents SCF and Rho more faithfully than SD and SWE, for which ERA5L consistently exhibits lower time-series fidelity relative to the competing gridded representations. Yet, ERA5-Land can be competitive, or preferred, for long-term mean SWE and trends of SD and SWE. Long-term mean and variability generally covary with time-series fidelity, whereas trend fidelity is weakly related and can exhibit trade-offs. This diagnostic dependence produces diagnostic inconsistency, with consequential outcomes such as reversing the inferred direction of regional SCF change in the Northern Arctic. Incomplete observational support creates distinct limits. For example, in the Marine West Coast Forest, gridded representations cannot be distinguished in their fidelity for SD trend where observations exist, yet yield opposing trends across unconstrained areas. In the ~800,000 km² Southern Arctic, ambiguity arises because observations are absent, while four SWE products support markedly different trend magnitudes and signs. Benchmarking should therefore assess product fidelity alongside the strength of evidence supporting it, evaluate diagnostic sensitivity, and retain unresolved disagreement where observations are absent or cannot discriminate among alternatives.