Abstract: Projecting the impacts of future climate change is critical for water managers planning infrastructure resilience strategies. Global earth system model (ESM) output is generally too coarse to provide the detailed, localized information needed for impact assessment, and the computational cost of running higher-resolution global simulations over sufficiently long periods is prohibitive. Downscaling methods span a wide range of computational cost, and statistical approaches in particular offer a relatively inexpensive way to refine climate model output into a more actionable form. Here, we assess a broad suite of statistical and simplified dynamic downscaling approaches for nine regions across the coterminous United States (CONUS).

We develop a suite of metrics to evaluate these methods, including a novel assessment of method performance under dominant large-scale meteorological conditions identified using a k-means weather-typing algorithm along with traditionally used measures of precipitation and temperature, such as seasonal means and spatial variability, and representations of teleconnection patterns associated with El Niño–Southern Oscillation (ENSO). We also include multiple observational datasets to account for observational uncertainty. This framework enables uncertainty quantification associated with climate model forcing, downscaling method choice, and observational dataset selection on a per metric basis. Downscaled dataset subselection using our evaluation can substantially alter the magnitude of the projected change signal, but rarely alters the sign of change when compared with the ensemble average. Further, by quantifying regional downscaling method performance, we can provide actionable guidance to decision makers on the most appropriate downscaling dataset or method for a given application.