Evaluating climate model simulations of precipitation: methods, problems and performance

Climate system modelling has been used extensively to investigate the role of human activities in causing global change. Model evaluation assesses the ability of the models used to simulate current climate. This article reviews the methodology of model evaluation with examples from recent studies involving precipitation. This crucial element of climate is difficult to model since the majority of precipitation occurs at scales less than that of the gridboxes of the highest resolution models. Detailed and reliable evaluation requires investigation of interannual variability as well as of climatological means on a variety of spatial scales. This sort of detailed analysis requires time-series of observed global precipitation at monthly time-steps or less. No single currently available global dataset of precipitation fulfils all the requirements for model evaluation, making the comparison of modelled global precipitation fields with 'reality' difficult. A number of recent precipitation evaluation projects are reviewed and a hierarchy of evaluation methods is provided based on spatial and temporal scale and whether or not tests for statistical significance are applied. Most studies to date have not tested for statistical significance, although when models improve with higher resolution and better physical parameterizations, statistical significance testing of differences will become increasingly more essential. The problems of evaluating modelled precipitation are being tackled by international projects such as the Global Precipitation Climatology Project, the WetNet Precipitation Intercomparison Projects and the Atmospheric Model Intercomparison Project. The results of evaluation studies to date emphasize that model simulations of future changes to the magnitude, timing and spatial pattern of global precipitation be viewed as scenarios and not as predictions.