Uncertainty analysis of a 2D sediment transport model: an example of the Lower River Salzach

Abstract

In sediment transport modeling, several sources of uncertainty exist that impinge on the variability of model results. Therefore, it is essential to conduct an uncertainty analysis to quantify the impact of these uncertainties, to detect regions of enhanced sensitivity and subsequently to determine a range of possible model outcomes. The first-order second moment method with numerical differentiation is applied to assess the uncertainties of a 2D sediment transport model Hydro_FT-2D at the Lower River Salzach. In comparison to other methods, the first-order second moment method has benefits in terms of total time requirement since it uses considerably less simulation runs to determine model uncertainty. In total, eight uncertain parameters are investigated including both model and river specific parameters. For this purpose, only 2n + 1 simulation runs are necessary leading to a total of 17 simulations. The results are evaluated against a reference simulation regarding bed elevation changes, bed load transport rates, grain size distribution, and total riverbed evolution volume. The results of the total riverbed evolution volume indicate a large influence of the investigated river specific parameters roughness of river channel (k st ), grain roughness (k s ), and bed load input rate of the upstream River Saalach (QS SAAL). Among the model specific parameters, the critical Shields parameter (θ crit) and the scaling factor of Meyer-Peter and Mueller equation (MPM) have a significant effect on the model results. Moreover, a spatial evaluation of the maximum and minimum parameter-specific deviation from the reference indicates sensitive areas in regions with poor descriptive data as well as in close vicinity to weirs, ramps, and lateral inflows. In these areas, the model predictions are subject to a high degree of uncertainty and have to be taken with caution. The applied first-order second moment method with numerical differentiation is a powerful method to identify sensitive areas within the numerical model and to gain knowledge on both uncertain model and river specific parameters. Based on the results, the variability of model outputs can be evaluated and assessed with respect to the uncertainty in the input parameters and can thus contribute to a deeper understanding of the model behavior, which is highly beneficial for long-term morphodynamic studies. The method is found to be applicable for sediment transport models especially in an applied engineering context and for long-term simulation runs due to the simplicity of implementation as well as the reasonable total time requirement.