Dynamical Downscaling Improves Upon Gridded Precipitation Products in the Sierra Nevada, California

Uncertainties in gridded and regional climate estimates of precipitation are large at high elevations, where observations are sparse and spatial variability is substantial. We explore these uncertainties for water year 2008 across California’s Sierra Nevada in 10 datasets: 6 regional climate downscalings generated using the weather research and forecasting (WRF) model at convection-permitting resolution with differing lateral boundary conditions and microphysical parameterizations, and four gauge-based, interpolation-gridded precipitation datasets. Precipitation from these 10 datasets is evaluated against 95 snow pillows and a precipitation dataset inferred from stream gauges using a Bayesian inference method. During water year 2008, the gridded datasets tend to underestimate frozen precipitation on the windward slope of the Sierra Nevada, particularly in the vicinity of Yosemite National Park. The WRF simulations with single-moment microphysics tend to overestimate precipitation throughout much of the region, whereas the WRF simulations with double-moment microphysics tend to better agree with both the snow pillows and inferred precipitation estimates, although they somewhat overestimate the windward/leeside precipitation contrast in the northern Sierra Nevada. WRF simulations, in particular those with single-moment microphysics, better distinguish spatial patterns of wet-versus-dry pillows and watersheds over the water year than the gridded estimates. Our results suggest treating gauge-based datasets as ‘truth’ may give a misleading representation of model accuracy, since these gauge-based datasets often have issues of their own.