Sensitivity of Large‐Scale Hydrological Predictions to Precipitation Phase Partitioning Methods Under a Changing Climate

ABSTRACT

Accurate partitioning of precipitation into rain and snow remains a major source of uncertainty in hydrological projections for cold regions. This study quantifies how four precipitation phase‐partitioning formulations influence simulated precipitation phase, snow storage and streamflow response in the Saskatchewan River Basin under historical (1981–2010) and late‐century (2071–2100) climate conditions. Simulations were conducted under naturalised conditions, with all reservoir regulation, irrigation and water‐management abstractions deactivated to isolate the effect of phase‐partitioning choice, using the MESH land‐surface hydroloical model forced with CanRCM4 climate projections spanning 1950 to 2100. Three empirical temperature‐based methods (single‐threshold, linear‐transition and polynomial) and one physically based psychrometric formulation were evaluated using diagnostic variables including total precipitation, snow and rain fractions, annual maximum snow water equivalent (SWE), runoff, peak discharge and streamflow timing metrics. Late‐century projections showed increases in total precipitation of 19%–30% and declines in snowfall fraction of 25%–34% across landforms. Reductions in annual maximum SWE ranged from approximately 5%–13% in the cordillera and foothills to 13%–19% in the plains. Inter‐method spreads in projected SWE were small in colder landforms (< 4 percentage points) but reached 6–9 percentage points in the plains and lowlands, comparable in magnitude to the projected climate‐change signal. Streamflow timing advanced consistently: spring pulse onset date advanced by a median of 27 days, centre‐of‐mass timing by 22 days and time to peak flow by 26 days. Peak discharge increased by a median of 12.6% and annual streamflow volume by 23.9%. Inter‐method differences were smallest for spring onset and centre‐of‐mass timing (1–3 days) and largest for time to peak flow and peak discharge, with mean absolute deviations exceeding 10 days and 14 percentage points respectively under some empirical methods. Snowfall‐fraction biases between empirical and physical methods were spatially structured, with time of emergence occurring as early as 1952 for the single‐threshold and linear‐transition methods and several decades later for the polynomial method. The physically based psychrometric formulation remained consistent across historical and future climates, whereas empirical temperature‐threshold schemes exhibited systematic biases and increasing divergence under stronger warming. These findings highlight the importance of explicitly accounting for precipitation phase‐partitioning uncertainty in climate‐impact assessments for snow‐dominated and mixed‐regime river basins.