Human intervention in rivers - quantifying the uncertainty of hydraulic model predictions
Koen Berends is a PhD student in the research group Marine and Fluvial Systems. His supervisor is prof.dr. S.J.M.H. Hulscher from the Faculty of Engineering Technology.
Human intervention in rivers is increasingly supported by computer model predictions. Due to limited observations of extreme events and the challenges related to predicting the future, it is recommended to quantify the uncertainty of model predictions. However, this is is not done in practice as the computational costs for such computations are often prohibitative. It is therefore unknown how model uncertainty affects model predictions of human intervention, even though human intervention in rivers has a large impact on society and the environment. The aim of this thesis is to improve the understanding of uncertainty surrounding model predictions of effect studies. We approached this as follows.
First, we developed a method to significantly decrease the computational burden of uncertainty quantification (80%-95% in our case studies). Second, we quantified the uncertainty of archetypical spatial interventions and demonstrated that the uncertainty (expressed as the 90% confidence interval) of the effect is between 20% to 40% of the mean effect. So, for an intervention with a average water level decrease of 20 cm, 90% percent of model simulations will be between 17 cm and 23 cm. Next, we studied whether calibration could potentially mitigate the sources of uncertainty. Results showed that this was not the case; in fact, we showed that errors in model assumptions (such as a too simplistic vegetation model) increase uncertainty. Finally we compare model simulations with thirty years of hydraulic observations. Results showed that the uncertainty in observations are too great to discern the effect of human intervention. The accuracy of model simulations could therefore not independently be verified.
The recommended approach going forward is to explicitly quantify and communicate the uncertainty of model predictions. The methods and insights developed in this thesis contribute toward this goal.