Improving river management by estimating model uncertainty


Persons involved
Ir. K.D. Berends (PhD cadidate)
Prof. dr. S.J.M.H. Hulscher (Promoter)
Dr. J. J. Warmink (daily supervisor)

Funding of the project

Other stakeholders
HKV Consultants
Waterbouwkundig Laboratorium (Vlaanderen)


Decisions in river management are often supported by complex physics-based numerical prediction models. While these models are built to be as accurate as possible, there has been growing awareness that uncertainty quantification can decrease the tendency towards conservative choices in engineering design. Furthermore, open communication and better understanding of the limitations of such models might increase credibility of its predictions.

In this research we deal with two problems of uncertainty analysis for hydraulic river models. In the first place, complex models often are too computationally demanding for Monte Carlo methods of uncertainty quantification. Secondly, existing uncertainty quantification methods are well established for systems that exhibit relatively little change. However, river systems can change quickly over a relatively short time span – both from natural processes as human intervention. Therefore, models with established accuracy on historical records might no longer be (as) valid for their intended purpose. For both probabilistic as deterministic use of models this might introduce a bias in their prediction.

The aim of this research is to study the drivers that govern applicability of calibrated models and use this knowledge to develop a method for efficient uncertainty estimation for models applied in river engineering.