Koen Daniël Berends

Efficient uncertainty quantification for impact analysis of human interventions in rivers

Human interventions to optimise river functions are often contentious, disruptive, and expensive. To analyse the expected impact of an intervention before implementation, decision makers rely on computations with complex physics-based hydraulic models. The outcome of these models is known to be sensitive to uncertain input parameters, but long model runtimes render full probabilistic assessment infeasible with standard computer resources. In this paper we propose an alternative, efficient method for uncertainty quantification for impact analysis that significantly reduces the required number of model runs by using a subsample of a full Monte Carlo ensemble to establish a probabilistic relationship between pre- and post-intervention model outcome. The efficiency of the method depends on the number of interventions, the initial Monte Carlo ensemble size and the desired level of accuracy. For the cases presented here, the computational cost was decreased by 65%.

Calibration and model performance decay in prediciton of the effect of human intervention in river systems

River managers intervene in the natural system to increase flood safety, inland shipping or biodiversity by dredging, sediment nourishments or other river engineering works. Complex hydraulic numerical models often form the basis of such decision making. Experience has shown that parameters of these complex models require calibration to increase performance to acceptable levels. However, we hypothesize that implementation of interventions will generally lead to decay in model performance. We used an experimental set-up to quantify this decline in performance due to modelling of levee relocation. Results show performance decay for all calibrated parametersets and an increasing uncertainty as to the predicted effect of the intervention. In closing, we argue that continued model testing and a better understanding of the relationship between reference and modified models might lead to improved model use for river management. Such sustainable model management might be a necessity in sustainable river management.