Robustness and uncertainties in future hydrological regimes of Swiss catchments

Projections of discharge under future climate are impaired by uncertainties arising from different sources: the emission scenarios, the climate models and internal climate variability, the post-processing and sampling of the climate projections, and the hydrological models structure and parameterization. This study aims to investigate the contribution of these sources to the final uncertainty of discharge projections. To this end we performed analyses of variance (ANOVA) in six catchments expected to react differently to climate change and representative of the typical Swiss discharge regimes. We used climate projections of the CH2011 dataset obtained from the Center for Climate Systems Modeling (C2SM). This dataset consists of two types of projections, both relying on the delta change technique. The delta coefficients were determined either in a deterministic way by spectral smoothing of 10 GCM-RCM chains or in a probabilistic way by a Bayesian multi-model approach combining 20 runs until 2050, and then 14 runs until 2099. In addition to the climate projections for emission scenario A1B chosen for the ENSEMBLES project, the CH2011 team generated simulations for the scenarios A2 and RCP3PD using pattern scaling. This enabled us to address the influence of the uncertainty in greenhouse gas emissions on discharge projections. We ran hydrological simulations using three conceptual models: HBV, PREVAH and WaSiM. HBV and PREVAH rely on a similar reservoir structure, while WaSiM uses the more process-oriented Richards-equation approach. PREVAH and WaSiM rely on a higher level of spatial discretization than the lumped HBV model. The use of the three different models allowed evaluation of the sensitivity of discharge projections to the hydrological model complexity and structure. Further, we compared two ways to determine PREVAH parameter values: via calibration and via regionalization. Simulations were run for the periods 2020-2049, 2045-2074 and 2070-2099 to assess the variation of the different sources of uncertainty over time. We combined the elements of the chain in a factorial way, leading to an ensemble of 72 simulations for each catchment and future period. Robust changes in discharge regimes emerge from our ensemble of runs: (i) drier summers projected by all the model runs, irrespectively of the emission scenario, climate model, post-processing or the hydrological model, for all but one catchment (ii) earlier snow-ice melt in both alpine catchments and (iii) larger winter flows in most catchments. Overall, the variance captured by our setting is best explained by the uncertainties stemming from the climate models and the internal climate variability. The contribution of the uncertainty in emissions increases over time and is large by the end of the century. The complexity and parameterization of the hydrological models have a comparatively small influence on the projections for the non-glacierized catchments. Our results suggest that, even if greenhouse gas concentration is constrained to the lowest level (RCP3PD), climate change signal is expected to emerge from natural discharge variability for all catchments by the end of the century. However, limiting emissions to RCP3PD levels could reduce the largest impacts on regime (in summer and winter) by approximately a factor two in comparison to the impacts projected for scenario A2 for 2070-2099.