Reliability analysis of river bed simulation models
Numerical simulations of river bottom evolution can be used for evaluating river engineering concepts and efficient operation of the natural and manmade waterways. The requirements for the precision of such simulations as well as to their ability to represent the real behaviour of the river bed morphodynamics are very high due to the large impact of civil water engineering to nature and society. In morphodynamic numerical models model parameters, initial conditions and input data can be uncertain due to the natural variability, the deficient description of the physical processes in the model and the imprecision of the model parameters. The propagation of these uncertainties can have serious implications in the reliability of the simulation results. Therefore, it is necessary to identify the various sources of uncertainty and to quantify their contributions to the variance of the model result. We present a novel approach for reliability analysis of coupled morphodynamic - hydrodynamic simulations of river bed evolution. It consist of sensitivity analysis of the simulation results to the variation of input parameters with linear FORM, preliminary estimation of non-linear effects and final determination of large confidence limits in substantially non-linear problems. For this purpose, an adaptive design of experiments (DoE) and a Monte-Carlo-style method accelerated by RBF metamodeling of bulky simulation results have been developed. A realistic application case is used to demonstrate efficiency of the overall approach. Two artificial flood events have been simulated along 10 km of river Danube, while 13 model parameters have been varied applying normal and distorted initial distributions. 68%, 95% and 99.7% confidence intervals for the resulting river bottom evolution have been determined, and the non-linear distortions of the resulting distribution have been demonstrated.