Climate Model Benchmarking for CMIP7 A CMIP Task Team

The goal of the Coupled Model Intercomparison Project (CMIP) is to better understand past, present, and future climate changes in a multi-model context. Based on the outcomes of the phase 6 of CMIP (CMIP6) Community Survey, the CMIP panel is seeking to identify ways to increase the project's scientific and societal relevance, improve accessibility, and widen participation. To achieve these goals, a number of Task Teams were established to support the design, scope, and definition of the next phase of CMIP and the evolution of CMIP infrastructure and future operationalization. An important prerequisite for providing reliable climate information from climate and Earth system models is to understand their capabilities and limitations. Thus, systematically and comprehensively evaluating the models with the best available observations and reanalysis data is essential. For CMIP7 new evaluation challenges stemming from models with higher resolution and enhanced complexity need to be rigorously addressed. These challenges are both technical (e.g., memory limits, increasingly unstructured and regional grids), and scientific. In particular, innovative diagnostics, including the support of machine learning-based analysis of CMIP simulations, must be developed. The Climate Model Benchmarking Task Team aims to provide a vision and concrete guidance for establishing a systematic, open, and rapid performance assessment of the expected large number of models participating in CMIP7, including a variety of informative diagnostics and performance metrics. The goal is to fully integrate evaluation tools into the CMIP publication workflow, and their diagnostic outputs published alongside the model output on the ESGF, ideally displayed through an easily accessible website. To accomplish this, existing evaluation tools need to be further developed and applied to historical and other CMIP7 simulations. We expect to produce an increasingly systematic characterization of the models which, compared with early phases of CMIP, will more quickly and openly identify the strengths and weaknesses of simulation results. This will also reveal whether long-standing model errors remain evident in newer models and will assist modelling groups in improving their models. This framework will be designed to readily incorporate updates, including new observations and a multitude of additional diagnostics and metrics as they become available from the research community, and will be developed as fully open-source software with high documentation standards.