Constrained Earth system models show a stronger reduction in future Northern Hemisphere snowmelt water

Although Earth system models (ESMs) tend to overestimate historical land surface warming, they also overestimate snow amounts in the Northern Hemisphere. By combining ground-based datasets and ESMs, we find that this paradoxical phenomenon is predominantly driven by an overestimation of light snowfall frequency. Using spatially distributed emergent constraints, we show that this paradox persists in mid- (2041–2060) and long-term (2081–2100) projections, affecting more than half of the Northern Hemisphere’s land surface. ESMs underestimate the frequency of freezing days by 12–19% and overestimate snow water equivalent by 28–34%. Constrained projections indicate that the raw ESM outputs overestimate future Northern Hemisphere snowmelt water by 12–16% across 53–60% of the Northern Hemisphere’s land surface. This snowmelt water overprediction implies that the amount of water available in the future for agriculture, industry, ecosystems and domestic use may be lower than unadjusted ESM projections suggest.