Uncertainty-informed selection of CMIP6 Earth System Model subsets for use in multisectoral and impact models

Abstract. Earth System Models (ESMs) are heavily used to provide inputs to impact and multisectoral dynamic models. Therefore, representing the full range of model uncertainty, scenario uncertainty, and interannual variability that ensembles of ESMs capture, is critical to the exploration of the future co-evolution of the integrated human-Earth system. The pre-eminent source of these ensembles has been the Coupled Model Intercomparison Project (CMIP). With more modeling centers participating in each new CMIP phase, the size of the ESM archive is rapidly increasing, which can be intractable for impact modelers to effectively utilize due to computational constraints and the challenges of analyzing large datasets. In this work, we present a method to select a subset of the latest phase, CMIP6, models for use as inputs to a sectoral impact or multisectoral models, while still representing the range of model uncertainty, scenario uncertainty, and interannual variability of the full CMIP6 ESM results. This method is intended to help human-relevant impact and multisectoral modelers select climate information from the CMIP archive efficiently. This is particularly critical for large ensemble experiments of multisectoral dynamic models that may be varying additional features beyond climate inputs in a factorial design, thus putting constraints on the number of climate simulations that can be used. We focus on temperature and precipitation outputs of ESMs, as these are two of the most used variables among impact models and many other key input variables for impacts are at least correlated with one or both of temperature and precipitation (e.g. relative humidity). Besides preserving the multi-model ensemble variance characteristics, we prioritize selecting ESMs in the subset that preserve the very likely distribution of equilibrium climate sensitivity values as assessed by the latest IPCC report. This approach could be applied to other output variables of ESMs and, when combined with emulators, offers a flexible framework for designing more efficient experiments on human-relevant climate impacts. It can also provide greater insight into the properties of existing ESMs and the method may be informative for future experiment planning across ESMs.