Impact of bias adjustment strategy on ensemble projections of hydrological extremes

Abstract

Hydrological climate change impact studies typically rely on hydrological projections generated by hydrological models driven with bias-adjusted climate simulations. Such hydrological projections are influenced by internal climate variability, which can mask the emergence of robust climate trends. To account for internal variability in climate projections, single-model initial-condition large ensembles (SMILEs) can be employed. SMILEs are generated by running a single global/regional climate model many times with slightly perturbed initial conditions. However, it remains challenging to select an appropriate bias adjustment strategy for SMILEs used in hydrological impact studies because of the relative importance of inter-variable dependence and the preservation of both climate variability and the change signal. To facilitate such selection, we here compare different bias adjustment strategies applied to SMILEs and their effect on hydrological impact assessments. Specifically, we investigate how climate and hydrological extremes change for 87 catchments in the Swiss Alps when using (a) univariate vs. bivariate, (b) ensemble vs. individual-member, and (c) change-preserving vs. non-change-preserving bias adjustment methods. To do so, we adjust the biases of a 50-member SMILE with the different adjustment methods and drive a hydrological model to simulate and project high and low flows. Our comparison shows (1) no clear benefits from using bivariate instead of univariate bias adjustment methods when the SMILE already efficiently simulates the dependence between temperature and precipitation and (2) that the choice of using ensemble vs. individual-member and change-preserving vs. non-change-preserving bias adjustments leads to large differences in the values of signal robustness indicators, including temperature, precipitation and streamflow signal-to-noise ratios and streamflow and precipitation time-of-emergence. These influences need to be considered when selecting an appropriate bias adjustment strategy for a given application. Based on our comparison, we generally recommend to apply change-preserving and ensemble bias adjustment methods in future hydrological impact studies using SMILEs. Further research is needed to improve bias adjustment methods that preserve both the signal and the variability of ensemble climate projections.