Diel observations suggest earlier snowmelt-driven streamflow than land surface modeling

Earlier streamflow from mountain headwaters challenges water supply and ecosystems and is expected to continue under climate change. There are few observations capable of separating rain and snowmelt contributions to streamflow, thus physically-based models are typically used for diagnosis and prediction. We develop a new method to detect whether snowmelt occurs in a day using the correspondence of incoming solar radiation and diel streamflow changes. We calculate the day when the 20^th percentile of snowmelt days occur (DOS₂₀) each year across 31 watersheds in Western US. Results show that historic DOS₂₀ varies from mid-January to early June. Warmer sites have more intermittent melt that occurs 7 days earlier per ° C. Mean annual DOS₂₀ highly correlates with the date of 25% and 50% annual streamflow volume (DOQ₂₅ and DOQ₅₀), suggesting that a one-day earlier DOS₂₀ results in a one-day earlier DOQ₂₅ and 0.7-day earlier DOQ₅₀. Prediction of future DOS₂₀ (under RCP 8.5, late 21^st century) using a space-for-time substitution shows that on average DOS₂₀ will occur 11 days earlier per 1°C of warming. In contrast, a land-surface model (Noah-MP) coupled with a climate model (WRF) that reasonably simulate historic DOS₂₀ (bias -23 days), shows less sensitivity in DOS₂₀ to climate change (7 days/°C). Moreover, space-for-time based projections of DOQ₂₅ and DOQ₅₀ are twice as sensitive to earlier streamflow (-60 and -41 days, respectively) than NoahMP-WRF. Given the importance of predicting streamflow timing, we suggest re-examining physically based models with auxiliary observations, such as diel streamflow, to improve the fidelity of state-of-the-art hydrological predictions to mitigate climate change impacts.