Assessment of NARCCAP model in simulating rainfall extremes using a spatially constrained regionalization method

Abstract

ABSTRACTCapturing the intensity and return period of extreme rainfall events in the historic record and projecting them into the future are essential to managing, planning, and designing infrastructure. In this study, we assess the performance of the combination of four General Circulation Models (GCMs) and six Regional Climate Models (RCMs) that comprise the North American Regional Climate Change Assessment Program (NARCCAP) and evaluate their performance in simulating rainfall extremes in the continental United States. We adopt a regionalization method to objectively delineate 12 regions in the continental United States with relatively homogenous annual maximum 24‐h rainfall patterns from the North American Regional Reanalysis (NARR) data set. We then compare the Intensity–Duration–Frequency (IDF) curves generated from control simulations of NARCCAP models with those from NARR in each of these regions. We find significant spatial variability of model performance. The models perform reasonably well in many parts of the country, but poorly in the southeastern United States. The GCM providing boundary conditions strongly influences results – output from those RCMs driven by the Community Climate System Model (CCSM) and Canadian Global Climate Model version 3 (CGCM3) matched the NARR data best. Performance of individual RCMs also varied, often in response to nudging, wherein the regional model is constrained by the GCM fields. We also measure changes in bias‐corrected IDF curves generated from NARCCAP projections of the future. In most regions, most models project intensified 24‐h rainfall events in the future (exceptions include some model‐projected decreases in southern California, the extreme north‐central US, Florida, and the Texas Plains). This study provides a valuable means of assessing NARCCAP models' performance in simulating rainfall extremes at the regional scale and understanding how the GCMs, RCMs, and spatial variability affect model performance.