Improved modelling of mountain snowpacks with spatially distributed precipitation bias correction derived from historical reanalysis

Abstract. Accurate estimates of snow water equivalent (SWE) are essential for effective water management in regions dependent on seasonal snowmelt. However, significant biases and high uncertainty in mountain precipitation data products pose significant challenges. This study leverages a SWE reanalysis framework and historical dataset to derive factors that can downscale and bias-correct mountain precipitation in a real-time modelling context. We evaluate through hindcast modelling how different versions of this precipitation bias correction affect errors in 1 April SWE estimates within a representative snow-dominated watershed in the Western U.S. We also evaluate how the additional assimilation of fractional snow-covered area (fSCA) or snow depth observations during the accumulation season impact the 1 April SWE estimates. Results show that spatially distributed historically informed precipitation bias correction significantly improves SWE estimates, reducing the normalized root mean square difference (NRMSD) by 58 %, increasing the correlation (R) by 43 %, and decreasing mean difference (MD) by 88 %. The primary strength of this bias correction method lies in capturing the spatial distribution of precipitation bias rather than its interannual variability. Assimilating snow depth observations further reduces errors both at the watershed scale (NRMSD less by 46 %) and pixel level in most years, while accumulation season fSCA assimilation is not generally useful. We demonstrate the value of these methods for streamflow forecasts: bias-corrected precipitation improves the correlation between daily simulated snowmelt and observed streamflow by 31–39 % and reduces bias in predicted April–July runoff volumes by 46–52 %. This study highlights how historical SWE reanalysis datasets can be leveraged and applied in a real-time context by informing precipitation bias correction.