Due to their construction, conceptual hydrological models typically exhibit many parameter uncertainties with considerable variance, leading to significant uncertainty in model predictions (e.g., predictions of flood or drought events). The propagation of the resulting high-dimensional joint parameter uncertainty is a challenging task both mathematically and in terms of the computational demands. In this work, we employ a non-intrusive polynomial chaos expansion to model uncertainty in five or more input parameters that characterize high-flow conditions. We rely on sparse grid (SG) strategies to compute the expansion's coefficients while keeping the necessary model runs small. To keep the black-box property of the combination technique while focusing on regions of interest adaptively, we rely on a recently proposed spatially adaptive SG combination technique with a dimension-wise refinement algorithm. Due to the runtime of the model, parallel execution and parallel post-processing of the UQ analysis are crucial in our solution. On the example of specific hydrological models used for flood forecasting, we show that our results can give an insight into the parameters' stochastic importance and provide authorities with a reliable uncertainty band over the flow predictions in a reasonable time. Our work bridges the gap between earlier theoretical work on UQ and more complex real-world problems.
«
Due to their construction, conceptual hydrological models typically exhibit many parameter uncertainties with considerable variance, leading to significant uncertainty in model predictions (e.g., predictions of flood or drought events). The propagation of the resulting high-dimensional joint parameter uncertainty is a challenging task both mathematically and in terms of the computational demands. In this work, we employ a non-intrusive polynomial chaos expansion to model uncertainty in five or m...
»