A global climate model ensemble for downscaled monthly climate normals over North America

Abstract

Use of downscaled global climate model projections is expanding rapidly as climate change vulnerability assessments and adaptation planning become mainstream in many sectors. Many climate change impact analyses use climate model projections downscaled at very high spatial resolution (~1 km) but very low temporal resolution (20- to 30-year normals). These applications have model selection priorities that are distinct from analyses at high temporal resolution. Here, we select a 13-model ensemble and an 8-model subset designed for robust change-factor downscaling of monthly climate normals, and describe their attributes in North America. All models are selected from the Coupled Model Intercomparison Project Phase 6 (CMIP6) archives. The 13-model ensemble is representative of the distribution of equilibrium climate sensitivity, grid resolution, and transient regional climate changes in the CMIP6 generation. The 8-model subset is consistent with the IPCC's recent assessment of the very likely range of Earth's equilibrium climate sensitivity. Our results emphasize several principles for selection and use of downscaled climate ensembles: (a) the ensemble must be observationally constrained to be meaningful; (b) analysis of multiple models is essential as the ensemble mean alone can be misleading; (c) small (<8-member) ensembles should be region-specific and used with caution; (d) higher grid resolution is not necessarily better; and (e) multiple simulations of each model/scenario combination are necessary to represent precipitation uncertainty. Although we have focused our documentation on North America, our model selection uses primarily global criteria and is applicable to downscaling climate normals in other continents. Downscaled projections for the selected models are available in ClimateNA (http://climatena.ca/). An accompanying web application (https://bcgov-env.shinyapps.io/cmip6-NA/) provides tools for further model selection and visualization of the ensemble.