Downscaling Global Climate Simulations to Regional Scales: Statistical Downscaling versus Dynamical Downscaling

Abstract

Downscaling is a fundamental procedure in the assessment of the future climate change impact at regional and watershed scales. Hence, it is important to investigate the spatial variability of the climate conditions that are constructed by various downscaling methods to assess whether each method can properly model the climate conditions at various spatial scales. To gain some insight into this issue with respect to statistical versus dynamical downscaling approaches, an assessment of the precipitation variability from a popular statistical downscaling method [bias correction with spatial disaggregation (BCSD)] and a dynamical downscaling method [by MM5: Fifth-Generation National Center for Atmospheric Research (NCAR)/Penn State Mesoscale Model] was performed. This assessment is based on the historical NCAR/National Center for Environmental Prediction (NCEP) reanalysis data, a community climate system model version 3 (CCSM3) global climate model (GCM) control run for the 1950–1999 period, and the CCSM3 GCM A1B emission scenario simulations for a projection period. Two spatial characteristics are investigated: (1) the normalized standard deviation (NSD), and (2) the precipitation change over the northern California region. The results of this investigation show that BCSD-based NSD and local precipitation change values do not show realistic spatial variation. Instead, they show interpolated spatial patterns from the coarse grid data set of two-degree resolution for both historic and projection periods. Meanwhile, MM5-based NSD and local precipitation change values show realistic spatial characteristics of precipitation variability for each month (December and July), for each season [December-January-February (DJF) and June-July-August (JJA)], and for the annual values, due to the heterogeneity in the northern California study region’s land characteristics. Both PRISM (Parameter-elevation Regressions on Independent Slopes Model) and MM5-simulated precipitation values have similar spatial structures, and they describe well how the NSD values and the precipitation field change locally in a spatially diverse pattern over the study region. Hence, BCSD procedure and MM5 simulations show significant differences for the 100-year average precipitation change. Opposite wet and dry trends between the precipitation estimated by the BCSD method and by MM5 are found in many places (especially in high-elevation areas). The BCSD method has limitations in projecting future precipitation values. As such, it is questionable whether the BCSD method is suitable for the assessment of the impact of future climate change at regional, watershed, and local scales as the future climate will evolve in time and space as a nonlinear system with land–atmosphere feedbacks.