Time series analysis of regional climate model performance

Abstract

Four regional climate models (RegCM2, MM5/BATS, MM5/SHEELS, and MM5/OSU) were intercompared on a fairly small domain covering a relatively homogenous area in Kansas, United States, including the First International Satellite Land Surface Climatology Project (SLSCP) Field Experiment (FIFE) site. The models were integrated for a 2‐year period covering 1987 and 1988. The model results are evaluated against data collected during this time period at the Konza Prairie Long‐Term Ecological Research (LTER) site as well as over the summer observation periods of FIFE. The models all captured the proper qualitative behavior of the interannual variability, though the magnitudes varied considerably between models. They also found it particularly difficult to reproduce observed changes in the variance of surface variables. No model performed consistently better, with each model displaying particular strengths and weaknesses of its own. RegCM2 could be improved by including an ice phase in the cloud microphysics parameterization. MM5/BATS and MM5/SHEELS need revision of the formulation of stability dependence of the surface drag coefficients, including the coupling to the wind field, as well as using a total soil depth more representative of the area. MM5/OSU simulates too much resistance to evapotranspiration and fails to close the energy budget. All of the models overestimate runoff and evapotranspiration during winter, creating a dry anomaly which persists throughout the following summer. Development and verification of parameterizations involved in coupling the land surface and atmospheric components of these models together is at least as important as the development and verification of each component individually.