Developing Representative Impact Scenarios From Climate Projection Ensembles, With Application to UKCP18 and EURO‐CORDEX Precipitation.

Calculating impacts from climate projection ensembles can be challenging. A simple approach might consider just the ensemble mean, but this ignores much of the information in the ensemble and does not explore the range of possible impacts. A more thorough approach would consider every ensemble member, but may be computationally unfeasible for many impact models. We investigate the compromise in which we represent the ensemble by the mean and a single deviation from the mean. The deviation from the mean would ideally be representative both of variability in the ensemble, and have a significant impact, according to some impact metric. We compare methods for calculating the deviation from the mean, based on traditional compositing and a statistical method known as Directional Component Analysis (DCA). DCA is based on linearizing the impact metric around the ensemble mean. We illustrate the methods with synthetic examples, and derive new mathematical results that clarify the interpretation of DCA. We then use the methods to derive scenarios from the UKCP18 and EURO‐CORDEX projections of future precipitation in Europe. We find that the worst ensemble member is not robust, but that deviations from the ensemble mean calculated using compositing and DCA are robust. They thus give robust insight into the patterns of change in the ensemble. We conclude that mean and representative deviation methods may be suitable for climate projection users who wish to explore the implications of the uncertainty around the ensemble mean without having to calculate the impacts of every ensemble member. Plain Language Summary: Using the results from climate models to force models that calculate the impacts of climate change can be challenging when ensembles are large. For detailed impacts models, it may not be computationally feasible to use all the data. We investigate ways to reduce the ensemble of different results down to just an ensemble of two patterns of change: one pattern for the mean, and one pattern for variations around the mean. Using these two patterns allows us to estimate the typical impact and the possible range of impacts. The pattern for variations around the mean we call a "representative deviation pattern." An ideal representative deviation pattern would capture the most relevant differences from the mean, where what defines relevance may vary according to the application. We compare simple and traditional methods for deriving the representative deviation pattern with a recently derived statistical method called Directional Component Analysis. We apply all the methods to help better understand the changes in rainfall expected in Europe due to climate change. We find robust results that show that the deviation from the mean with the greatest total rainfall consists of a pattern with more rainfall in most of Europe but less rainfall in the far North. Key Points: We show how to make climate model results easier to use by reducing large ensembles to just the ensemble mean and a single patternWe apply the methods to European rainfall in two climate model ensembles, and gain new insights into the structure of likely changesComparing ways to define the representative deviation, we find that simple compositing and a linear method give similar results [ABSTRACT FROM AUTHOR]