On the Verification and Comparison of Extreme Rainfall Indices from Climate Models

Abstract

The interpretation of model precipitation output (e.g., as a gridpoint estimate versus as an areal mean) has a large impact on the evaluation and comparison of simulated daily extreme rainfall indices from climate models. It is first argued that interpretation as a gridpoint estimate (i.e., corresponding to station data) is incorrect. The impacts of this interpretation versus the areal mean interpretation in the context of rainfall extremes are then illustrated. A high-resolution (0.25° × 0.25° grid) daily observed precipitation dataset for the United States [from Climate Prediction Center (CPC)] is used as idealized perfect model gridded data. Both 30-yr return levels of daily precipitation (P30) and a simple daily intensity index are substantially reduced in these data when estimated at coarser resolution compared to the estimation at finer resolution. The reduction of P30 averaged over the conterminous United States is about 9%, 15%, 28%, 33%, and 43% when the data were first interpolated to 0... Abstract The interpretation of model precipitation output (e.g., as a gridpoint estimate versus as an areal mean) has a large impact on the evaluation and comparison of simulated daily extreme rainfall indices from climate models. It is first argued that interpretation as a gridpoint estimate (i.e., corresponding to station data) is incorrect. The impacts of this interpretation versus the areal mean interpretation in the context of rainfall extremes are then illustrated. A high-resolution (0.25° × 0.25° grid) daily observed precipitation dataset for the United States [from Climate Prediction Center (CPC)] is used as idealized perfect model gridded data. Both 30-yr return levels of daily precipitation (P30) and a simple daily intensity index are substantially reduced in these data when estimated at coarser resolution compared to the estimation at finer resolution. The reduction of P30 averaged over the conterminous United States is about 9%, 15%, 28%, 33%, and 43% when the data were first interpolated to 0...