Towards a More Robust Evaluation of Climate Model and Hydrological Impact Uncertainties

Abstract

AbstractThe uncertainty of climate model projections is recognized as being large. This represents a challenge for decision makers as the simulation spread of a climate model ensemble can be large, and there might even be disagreement on the direction of the climate change signal among the members of the ensemble. This study quantifies changes in the hydrological projection uncertainty due to different approaches used to select a climate model ensemble. The study assesses 16 Euro-CORDEX Regional Climate Models (RCMs) that drive three different conceptualizations of the MIKE-SHE hydrological model for the Ahlergaarde catchment in western Denmark. The skills of the raw and bias-corrected RCMs to simulate historical precipitation are evaluated using sets of nine, six, and three metrics assessing means and extremes in a series of steps, and results in reduction of projection uncertainties. After each step, the overall lowest-performing model is removed from the ensemble and the standard deviation is estimated, only considering the members of the new ensemble. This is performed for nine steps. The uncertainty of raw RCM outputs is reduced the most for river discharge (5 th , 50 th and 95 th percentiles) when using the set of three metrics, which only assess precipitation means and one ‘moderate’ extreme metrics. In contrast, the uncertainty of bias-corrected RCMs is reduced the most when using all nine metrics, which evaluate means, ‘moderate’ extremes and high extremes. Similar results are obtained for groundwater head (GWH). For the last step of the method, the initial standard deviation of the raw outputs decreases up to 38% for GWH and 37% for river discharge. The corresponding decreases when evaluating the bias-corrected outputs are 63% and 42%. For the bias corrected outputs, the approach proposed here reduces the projected hydrological uncertainty and provides a stronger change signal for most of the months. This analysis provides an insight on how different approaches used to select a climate model ensemble affect the uncertainty of the hydrological projections and, in this case, reduce the uncertainty of the future projections.