Understanding the Cascade: Removing GCM Biases Improves Dynamically Downscaled Climate Projections
The more practitioners need to implement bias correction (BC) of climate information, the less useful BC becomes. The role of BC in dynamical downscaling has never been explored using an ensemble of global climate models (GCMs) given their computational expense. For the first time, we implement a simple monthly delta method approach to the boundary conditions of carefully chosen Coupled Model Intercomparison Project Phase 6 (CMIP6) GCMs to explore the broad effects of bias correction on downscaled temperature, precipitation, and snow over the recent historical past. We compare these experiments to non-bias-corrected twin experiments from the Western United States Dynamically Downscaled Dataset (WUS-D3) to assign physical explainability of BC effects in downscaling. In the process, we find that CMIP6 GCMs share surprisingly similar biases across the region, which lead to similar bias outcomes in downscaling.
A mean BC of the GCM boundary conditions leads to a less biased and more constricted spread of regional climate model-simulated snow and precipitation compared to in situ measurements from the SNOw TELemetry (SNOTEL) network. Locally, non-biased corrected experiments can be >50% (>100%) wet-biased in the annual climate mean for precipitation (snow). Further, BC, applied to means only, also improves the climatology of extreme precipitation. Finally, fundamental meteorological arguments can be used to interpret the effects of BC in improving the downscaled hydroclimate.
Pending a robust evaluation of the potential for BC to distort climate trends, bias correcting the boundary conditions of carefully selected GCMs using simple BC techniques should be encouraged in dynamical downscaling.