Emulating climate extreme indices

Abstract

We use simple pattern scaling and time-shift to emulate changes in a set of climate extreme indices under future scenarios, and evaluate the emulators' accuracy. We propose a metric for the error in emulation in the context of initial condition ensembles, to specifically characterize the role of internal variability in the emulation performance. Our metric separates systematic emulation errors from unavoidable discrepancies between emulated and target values due to internal variability. We compute the metricis at grid-point scale, and we show geographically resolved results, or aggregate them at global scale. We demonstrate the use of our error metric in the emulation of a suite of temperature and precipitation extreme indices. We test and compare simple pattern scaling and time-shift using a range of trajectories spanning targets inspired by the Paris agreement -- warming to 1.5C and 2.0C from the pre-industrial baseline -- and two of the longer-established trajectories, RCP4.5 and RCP8.5. With this suite of scenarios we can test the effects on the performance of the size of the temperature gap between emulation origin and target. We find that for most indices emulation the dominant source of discrepancy is internal variability. For at least one index, however, counting exceedances of a high temperature threshold, significant portions of the globally aggregated discrepancy and its regional pattern originate from the systematic emulation error. This error exceeds internal variability of both the target and the emulated quantities in large coherent regions at low latitudes, and the explanation can be found in the differential behavior of temperature distributions across latitudes. The metric also highlights a fundamental difference in the two methods related to the simulation of internal variability, which is dampened significantly by simple pattern scaling. This aspect is of consequence when using these methods for specific applications, where preserving variability for uncertainty quantification is deemed important. With this study we offer our metric as a diagnostic tool, facilitating the formulation of scientific hypotheses on the reasons for the error. In the meantime, we show that for many impact relevant indices by now traditional emulation techniques can be accurate within the variations unavoidably introduced by internal variability, establishing the fundamental condition for using their emulation in impact modeling.