Conceptual River Modelling in Support of Uncertainty and Scenario Analyses at Catchment Scale

River and coastal floods are regarded as the worldwide number one natural hazard in terms of economic losses. The frequency and consequences (fatalities, financial costs or environmental impacts) of these flood events are expected to increase in the next decades due to climate change, land use change and population growth. It is in this regard that the European Commission adopted the EU Water Framework Directive and the EU Floods Directive. The former compels EU member states to achieve a good qualitative and quantitative status for all water bodies and to come up with river basin management plans. The latter requires them to identify all areas at risk from flooding and to take adequate measures to reduce this flood risk. To fulfil the requirements set by the Water Framework Directive and the Floods Directive, water managers rely on a combination of hydrologic rainfall runoff and hydrodynamic river models. Both models try to emulate the real world processes as accurate as possible, but are inevitably connected with the existence of errors and uncertainties. Considering that decision making processes are more and more based on these model results, river managers and authorities should recognize and account for these uncertainties.

Handling uncertainties in any modelling practice usually requires a large number of model runs and/or long-term simulations. Conventional full hydrodynamic river models are not suited for this purpose, due to their relatively long simulation times and their limited flexibility in user-interface, which inhibits automatization. Using conceptual surrogate models that mimic the results of the detailed models provides a good alternative to circumvent these deficiencies. The discrete explicit calculation scheme of such conceptual models allows to strongly reduce model simulation times, combined with only limited loss of accuracy, which makes them suitable for any application that requires fast and accurate simulation results. This dissertation demonstrates how the conceptual modelling approach can be used in the framework of the EU Floods Directive and thereby accounting for the uncertainty in the system.

The first part of this manuscript describes a conceptual modelling approach that is suitable for a wide range of applications in water management. River networks are schematized by a number of mutually interconnected storage reservoirs, whereby processes are aggregated both in space and time. Each storage reservoir uses a discrete form of the mass balance equation and all water quantity variables are directly or indirectly related to this storage state. The discrete solver and the lumping of processes enable to strongly reduce simulation times. A variety of model structures, components and associated calibration techniques are available to cover a broad spectrum of possible river topologies and river dynamics. Until now, this existing approach was primarily used to model the free flowing upstream parts of river systems, where no influences are present of backwater effects induced by downstream water level boundaries. Extra model components are, therefore, developed to include tidally influenced reaches in the conceptual modelling approach. River systems can now be modelled in a computationally efficient way on basin scale, to investigate the impact of changing boundary conditions and man-made interactions in the system.

The second and most important part of this dissertation focusses on the role of uncertainty in hydrodynamic river modelling. A first objective in this regard is the quantification and decomposition of river model output uncertainties in its contributing sources. This permits modellers to understand where the model uncertainty is coming from, to recognize the most important source(s) and, hence, also to identify efficient strategies to reduce model uncertainty. Model output uncertainties are quantified with a data-based approach, that can be applied to both water levels and discharges. Such an approach compares model simulation results with historical observations and is very computationally efficient. Effects such as heteroscedasticity, non-normality, skewness etc. are also easily accounted for. Uncertainties on model inputs and model parameters are subsequently propagated through the model to assess their impact at locations of interest. This propagation process can happen relatively fast thanks to the low calculation times of the conceptual model. After propagating all uncertainty sources, the results are statistically analysed to determine the relative importance of each uncertainty source. This is done for both bias and variance components to achieve a better understanding of the total model output uncertainty. The approach is demonstrated on two case studies of the rivers Dender and Zenne, in Belgium, and a number of advices are formulated to improve model performance.

A second objective related to uncertainty is the calibration of hydrodynamic river model parameters. Calibrating these parameters is usually a manual and time-consuming task, or based on default or text-book values. This depends on the experience of the modeller and one can assume that a global optimum of all parameters is hard or even impossible to find within a reasonable timeframe. A more accurate calibration of the parameters would allow to diminish the effect of model parameter uncertainty on the total model output uncertainty. In this dissertation an approach to calibrate hydrodynamic river model parameters is presented. An accurate conceptual model, which allows to incorporate the effect of changing parameter values, is coupled with the SCEM-UA optimization toolbox. Head loss factors, discharge coefficients and Manning roughness coefficients are calibrated by minimizing the errors between model results and observed water levels. Results show that the updated river model is able to produce more accurate results than the original model. The approach also allows to include the variability of parameters in the conceptual model, which is not possible with the hydrodynamic model. The former can therefore outperform the latter.

A third objective related to uncertainty, is to demonstrate the applicability and benefits of the conceptual modelling approach in a scenario analysis. The flood risk in the Dender basin is studied for the current and future climate conditions and a number of scenarios to reduce this flood risk are evaluated. Risks are calculated based on long term simulation results with the conceptual model, whereby the climate change impact on rainfall-runoff and the downstream tidal water level is incorporated. A total of 120 flood risk reduction strategies are considered and a cost-benefit analysis is set up to identify the most efficient strategy in terms of economic and social benefits. In a second step of the analysis, uncertainties on the boundary conditions are also accounted for. The main idea here is to estimate the probability that a given scenario effectively leads to a reduction of the flood risk. The fact is that the results of the preliminary cost-benefit analysis are influenced by the different sources of uncertainty in river modelling. Calculating the aforementioned probability thus provides extra information for the decision making process.