The management and control of water systems is often based on the results of hydraulic or morphological models, for example. Although based on physical principles, these models contain several uncertainties, even if these models are well calibrated. The relevance of uncertainties and the effect they may have in practise on decision-making processes is becoming increasingly more recognised. It is often difficult, however, to obtain accurate and quantitative estimates for these uncertainties, especially in the case of complex, computational intensive numerical models. Using a case study, a specific aspect of this problem has been addressed, dealing with the uncertainties in model predictions due to uncertainties in the model's input(s) or external forcings.
In this case study, a 2D morphological model based on Delft3D was applied to a reach of the Upper Rhine. The discharges at the upstream boundary were considered as an uncertain input for the model and the effect of this uncertainty on the river bed was estimated. This was done using a Monte Carlo (MC) method. The uncertainties in the upstream discharge were described through the use of a statistical model, and a set of 100 random discharge series of 3 years was generated. The morphological model was evaluated for each of these synthetic discharges.
A standard MC method, as applied in this case study, has the disadvantage that quite a large number of simulations is required to obtain sufficiently accurate estimates for the uncertainty in the model results. Alternative sampling methods have been developed to improve the efficiency of standard MCmethods, and in the present case the performance of a so-called Quasi Monte Carlo (QMC) method was examined. The results of preliminary tests are promising and on this basis it may well be feasible to refine and improve the present uncertainty analysis significantly. More generally, such QMC techniques may provide important new opportunities for uncertainty analyses to other complex and computationally intensive numerical models. |
