Share

Feedback

Projects

Get an overview about the project outputs and related knowledge

F1) Estimating model uncertainty

Start: 03/2015
End: 12/2019
Status: Active

Contact details

Koen Berends

Deltares

The St. Andries bend along the Dutch river Waal is a bottleneck during high water levels where we studied the model uncertainty due to twelve possible river interventions. The figure indicates the location and an example water level profile to visualize the effects. (Source photo: River intervention explorer preview).

Project output

Tool for efficient uncertainty estimation of water levels due to the effect of river interventions.

Challenge

Rivers are continuously changing, both naturally and through interventions to for example mitigate flood or high water levels. Project advisors generally use numerical models to predict the maximum decrease in water levels due to river interventions such as side channels or floodplain vegetation changes. Water level predictions are generally affected by the limited understanding of the resistance for the water to flow in the river and its floodplain, which also affects the choices in model parameters. To deal with these unknowns, the variation or uncertainty of model results is often based on all possible combinations for the model parameters. However, this is computationally demanding and therefore not feasible for every-day practice. Depending on how much is the reduction in the water level at a critical location (see photo), a larger or smaller variation in model results becomes acceptable. Is model uncertainty so large that model results are not useful, or is it so small that it can be ignored? We reduce the computational time of estimating model uncertainty to get an indication of the variation in the predicted water levels.

Key goals: Fundamental understanding

Overview of the innovative components.

Innovative components

We provide new insights for the uncertainty estimation by (see Figure):

  • Considering as model parameters uncertainties in the resistance of the water to flow (channel roughness), the floodplain (vegetation roughness) as well as possible classification errors in the floodplain land cover .
  • Developing an innovative open source tool (CORAL or Correlated Output Regression Analysis) to reduce the computational time of uncertainty estimations based on the correlation of water levels before and after a given intervention.
  • Using the above tool to calculate and compare the water level reduction due to the simple (1D) or more detailed (2D) modelling of river interventions. We further proposed to compare the river interventions based on their relative uncertainty.
  • For experiments in a river channel, we used experimental measurements to estimate the uncertainty of input parameters into a more complex model (3D) for the deposition of suspended sediment in the channel (channel siltation).

For whom and where?

  • For modelers who want to quantify model uncertainty in their designs;
  • For decision makers who want to understand how model uncertainty influences design choices.

Data-collection methods: Numerical modeling Process-based modeling

Temporal scale: 1-10years

Application and findings

We applied the CORAL tool to estimate the uncertainty of both 1D and 2D models  calculations for flood reduction.  As an example, we compared six typical river interventions at a critical high water levels location such as the St Andries bend in the river Waal. Furthermore, we applied CORAL to approximate the uncertainty of channel siltation in a 3D morphological model. We found that CORAL is an effective method to approximate uncertainty in situations where traditional approaches are not feasible or not desirable.

For river intervention design, we focused on the effect of uncertainty in roughness parameters. We found that the uncertainty may be expressed as a percentage (termed ‘relative uncertainty’) relating the range of possible reduction in the water level to the expected reduction. Therefore, we calculated the relative uncertainty for typical river interventions. The comparison shows that floodplain removal is by far the most uncertain intervention. A reduction in relative uncertainty could be achieved by minimizing the amount of change to existing floodplains while optimizing the expected flood level decrease.

Status for day-to-day practice

Tool and examples are available to incorporate model uncertainty in both channel experiments and the design of typical river interventions in the Dutch rivers.


Key locations where the study took place.

Key locations: Rhine River (NL) Waal River (NL)

Spatial scale: River section

Next steps

Test the relative uncertainty as a decision-making criterion to compare multiple river interventions and extend our approach to other sources of uncertainty.

Last modified: 20/06/2019

Explore the contact details to get to know more about the researchers, the supervisory team and the organizations that contribute to this project.

Main researcher

Koen Berends

Deltares

Supervisory team

Jord Warmink

prof. dr. Suzanne Hulscher

University of Twente

Contributing partners

As soon as available, explore the storyline to get to know more about the main methods or prototype tools that were developed within this project.

Explore the output details for available publications to get a glance of the innovative components and implications to practice as well as the links to supporting datasets.

Project outputs

Towards efficient uncertainty quantification with high-resolution morphodynamic models

An alternative approach to lower the computational cost of uncertainty analysis. Application to an idealized example for channel siltation in a harbour.

30/06/2019 by Koen Berends et al.

View details View publication

Contains: Model or tool access Publication

Uncertainty quantification of flood mitigation predictions and implications for decision making

For six typical interventions along the river Waal, we show that the choice between interventions can be different when relative uncertainty is taken into account.

14/11/2018 by Koen Berends et al.

View details View publication

Contains: Dataset access

Efficient uncertainty quantification for impact analysis of human interventions in rivers

A novel method is presented to estimate model uncertainty with a reduced number of model evaluations.

14/06/2018 by Koen Berends et al.

View details View publication

Contains: Dataset access Model or tool upon request

Take a look to the dissemination efforts and application experiences which are available in the news items and blogs.

Events

25/05/2019

Knowledge sharing meeting in Witteveen+Bos, Rotterdam

How can we apply the researchers’ findings to improve our designs and environmental impact assessments? How can we share lessons learned from RiverCare within our team of engineers and consultants?

View event

All events

Videos

RiverCare meeting room

01/11/2016

View video

All videos

Anything to ask or share?

About us

We would like to learn from your experiences and questions to take our knowledge further into practice in the Netherlands and abroad. Your feedback will help us to find out about your interests and how useful the information provided was to you.