The Challenge of Ensemble Subset Selection for Climate Change Impacts Assessments

Climate model projections are increasingly used for climate adaptation planning and impact assessments. However, while many climate models and sets of projections exist, most impact assessments and planning activities will use only a subset of projections for varying reasons. There are several potential approaches for selecting global climate models to use including analytic methods based on historical skill and future spread or culling based on equilibrium climate sensitivity. However, many users also make use of downscaled climate projections to provide localized guidance on climatic changes and their potential impacts. Often times these projections are created with statistical downscaling procedures. Statistical downscaling corrects the biases of GCMs while incorporating local climatic effects (such as topography) to produce localized estimates of potential changes. What influence does statistical downscaling have on the challenge of ensemble subset selection? This talk will use downscaled projections of precipitation to discuss the challenge of statistical downscaling in ensemble subset selection. This analysis will focus on precipitation in the United States Southern Great Plains using a CMIP5 GCM ensemble and an ensemble of the same models downscaled using Localized Canonical Analogs (LOCA). An analytic approach to ensemble subset selection shows that one would select different models to use based on whether the ensemble is statistically downscaled. We then use an example to show how statistical downscaling alters the individual members of an ensemble distribution to enough of a degree that analytic ensemble subset selection approaches can select different models for use. We conclude with a discussion of how the traits of statistical downscaling relate to other subset selection methods (such as culling by equilibrium climate sensitivity) and future research directions.