Analyzing precipitation projections: A comparison of different approaches to climate model evaluation

摘要: [1] Complexity and resolution of global climate models are steadily increasing, yet the uncertainty their projections remains large, particularly for precipitation. Given impacts precipitation changes have on ecosystems, there is a need to reduce projection by assessing performance models. A common way evaluating consider maps errors against observations range variables. However, depending purpose, feature-based metrics defined regional scale one variable may be more suitable identify most accurate We compare three different ways ranking CMIP3 models: in broad variables, field precipitation, features modeled areas where pronounced future expected. The same analysis performed temperature potential differences between multimodel mean found outperform all single field-based rankings but performs only averagely ranking. Selecting best each metric reduces absolute spread projections. If anomalies considered, model reduced few regions, while can increased others. also demonstrate that attribution lack agreement physics misleading. Agreement similarly poor within ensemble members model, indicating robust trends attributed partly low signal-to-noise ratio.