Evaluation of methods to determine the fraction of attributable risk for a given extreme event

Abstract:

As climate change progresses in response to the rising levels of CO2 the question arises how an observed extreme has been affected by anthropogenic climate change. A measure has been proposed to quantify this influence, called the fraction of attributable risk (FAR). FAR estimates are either model-based or solely based on statistical analyses of observed time series. In both cases untested assumptions underly the methodologies. In model-based estimates, climate models are used to estimate the probability of the event in a world that could have been without anthropogenic climate change with assumptions about the world that could have been given the lack of the detailed state of the climate system in the pre-industrial era. These estimates are affected by model error. In observation-based estimates assumptions are made about the cause of observed changes and the stationarity of statistical relations.

We test the different approaches to calculate the FAR in a perfect model context, where we regard an ensemble of 16 simulations with the EC-Earth climate model under the RCP8.5 scenario as truth. Proposed methods to calculate the FAR are benchmarked against the true FAR value estimated directly from the ensemble.