Reducing uncertainties in hydrological climate change impact research
Vision: Five important knowledge gaps currently limit much of the present effort to understand and project climate change impacts on water resources. The goal of the proposed project is to fill these existing knowledge gaps and strengthen the basis for robust projections of future hydrological climate change impacts by carefully assessing existing approaches and by developing new dynamic and multi-dimensional methods to reduce uncertainties.
Hydrological climate change impact studies rely on complex serial modeling chains (Figure 1) which typically involve several methodological decisions including choices of greenhouse gas emission scenarios as a basis for a global climate model (GCM). GCM simulations are then used as boundary conditions for regional climate models (RCMs). Climate ensemble simulations of such RCMs with resolutions of 10-50 km for Europe can be downloaded from public web portals such as CORDEX1 or ENSEMBLES2. For hydrological impact studies at the catchment scale, RCM-simulated variables such as temperature (T) and precipitation (P) are most commonly used to drive a hydrological model. These RCM variables are, however, often affected by considerable systematic model errors3,4 (also called biases), and consequently require suitable bias correction approaches5. Bias correction is the process of re-scaling climate model outputs to reduce the effects of systematic errors in the climate models and to make the output more suitable as driving force for the hydrological model, which in turn provides simulations of a hydrological component (e.g., streamflow) under varying climate conditions. Bias correction methods typically do not consider the physical reasons of the biases and can introduce inconsistencies between simulated variables (knowledge gap 1). Furthermore, a crucial assumption of common bias correction methods is that the RCM simulation bias is assumed to be invariant over time (i.e. stationary), which is often not the case in reality (knowledge gap 2). Another limitation of the above mentioned type of modeling chain is that the decisions made along the way have a considerable impact on the resulting hydrological projections, because different models/methods have different skill levels6 and therefore cause highly variable results (knowledge gap 3). This does not only apply to the climate models, but also to the hydrological models at the end of the modeling chain. Especially the way of calculating certain hydrological components within the model (e.g. potential evaporation or snow pack) and their dependence on the driving RCM data is a source of large uncertainties (knowledge gap 4). And finally, stationary model parameterization (i.e., fixed model parameter values) can be problematic as it does not allow the hydrological model to respond to physical changes in the catchments caused by climate change or anthropogenic modifications (knowledge gap 5).
- Bridge knowledge gap 1: Develop a higher-dimensional bias correction (i.e., multivariate distribution scaling, MVDS) that is able to maintain physical links among multiple variables to effectively reduce RCM biases and evaluate its suitability for hydrological impacts simulations in Sweden
- Bridge knowledge gap 2: Add a dynamic component to MVDS, so that it will become sensitive to time-variant RCM biases for hydrological impacts simulations in Sweden
- Bridge knowledge gap 3: Investigate the skill levels of different hydrological models under climate conditions unlike those used for model training, while considering different ensemble averaging techniques when examining inter-period transferability
- Bridge knowledge gap 4: Benchmark approaches for deriving potential evaporation conditions (for Swedish climate) based on RCM simulations in a changing climate context
- Bridge knowledge gap 5: Evaluate whether time-varying hydrological model parameters better reflect changes in hydrological processes in a changing climate