Dynamic downscaling of 22-year CFS winter seasonal hindcasts with the UCLA-ETA regional climate model over the United States

Abstract

This study evaluates the UCLA-ETA regional model’s dynamic downscaling ability to improve the National Center for Environmental Prediction Climate Forecast System (NCEP CFS), winter season predictions over the contiguous United States (US). Spatial distributions and temporal variations of seasonal and monthly precipitation are the main focus. A multi-member ensemble means of 22 winters from 1982 through 2004 are included in the study. CFS over-predicts the precipitation in eastern and western US by as much as 45 and 90 % on average compared to observations, respectively. Dynamic downscaling improves the precipitation hindcasts across the domain, except in the southern States, by substantially reducing the excessive precipitation produced by the CFS. Average precipitation root-mean-square error for CFS and UCLA-ETA are 1.5 and 0.9 mm day−1, respectively. In addition, downscaling improves the simulation of spatial distribution of snow water equivalent and land surface heat fluxes. Despite these large improvements, the UCLA-ETA’s ability to improve the inter-annual and intra-seasonal precipitation variability is not clear, probably because of the imposed CFS’ lateral boundary conditions. Preliminary analysis of the cause for the large precipitation differences between the models reveals that the CFS appears to underestimate the moisture flux convergence despite producing excessive precipitation amounts. Additionally, the comparison of modeled monthly surface sensible and latent heat fluxes with Global Land Data Assimilation System land data set shows that the CFS incorrectly partitioned most of surface energy into evaporation, unlike the UCLA-ETA. These findings suggest that the downscaling improvements are mostly due to a better representation of land-surface processes by the UCLA-ETA. Sensitivity tests also reveal that higher-resolution topography only played a secondary role in the dynamic downscaling improvement.