Computer models used to predict sedimentation are computationally very demanding and sensitive to uncertainty in model parameters. In this publication, we explore an alternative multifidelity approach to estimate model output uncertainty for computationally expensive but accurate morphodynamic models using computationally fast but innacurate models. The approach is tested on a siltation study of an estuarine navigation approach channel.
Probability density function of the low-fidelity model (red) and ten random realisations of the high-fidelity model (gray). Source: Berends et al. 2019.
Finding and implications to practice
We found that multifidelity approach is promising to approximate the uncertainty of detailed morphodynamic models. Compared to a full Monte Carlo analysis, the computational effort can be reduced by over 80%, while the precision of the approximation is explicitly computed. For practice, this opens up the possibility to approximate the uncertainty of computer model output on standard computer resources (i.e., without the need for high-performance computing).
Berends, K.D., Scheel, F., Warmink, J.J., de Boer, W.P., Ranasinghe, R., Hulscher, S.J.M.H., 2019. Towards efficient uncertainty quantification with high-resolution morphodynamic models: A multifidelity approach applied to channel sedimentation. Coastal Engineering 152, 103520. https://doi.org/10.1016/j.coastaleng.2019.103520
Model or tool access
Models and datasets are available on request. An implementation of the method described in the publication is available from https://github.com/kdberends/coral
Uncertainty quantification of flood mitigation predictions and implications for decision making
Efficient uncertainty quantification for impact analysis of human interventions in rivers
Last modified: 17/07/2019