Rivers are continuously changing, both naturally and through human intervention. We use models to understand and predict the effects of such changes. These predictions are affected by uncertainty about system behavior and parameter values under unseen conditions. Ideally, this uncertainty is explicitly taken into account. However, uncertainty quantification methods are computationally highly demanding. In this project, we study alternative cost-effective methods to estimate the uncertainty of model output in changing systems. These methods are then applied to typical case studies of changing river systems. Our objective is to significantly reduce the computational cost of uncertainty estimation in practice.
Management phase: Evaluation & Adjustment | Intervention Planning |
Management goals: Hydrodynamic understanding |
We study models of nonstationary systems. We specifically look at systems that are or will be changed due to human intervention. The aim of this study is twofold:
- To develop a cost-effective method for uncertainty estimation of model output
- To study the benefit of calibration to reduce model uncertainty in changing systems
Temporal scale: Recent evolution (1-10years) |
- Calibration is generally performed to increase the performance of river models – and decrease uncertainty of their predictions. As a first step to deeper understanding about how model performance and model uncertainty behave under change, we carried out sensitivity of model performance to systematic change by human intervention for an idealized case study.
- Modeling a stream-scale experimental flume under various flow conditions to better understand model performance.
- Use of physics-based emulators for computationally less expensive uncertainty analysis.
Data-collection methods: Numerical modeling | Process-based modeling |
Main progress and next steps
- For our idealised case study, the sensitivity of models to systematic change by human interventions suggests several mechanics that lead to additional uncertainty and decreasing model performance. These mechanics can lead to spurious expectations of accuracy for traditional impact assessment and should therefore be taken into account when communicating model predictions. A deeper analysis of these mechanics is currently being carried out by modeling a stream-scale experimental flume under various conditions. Early results demonstrate the limitations imposed on predictive power by calibration and highlight the necessity of effective validation.
- Next, we will continue to pursue the use of physics-based emulation for river systems. Advances are already being made in the generation of such an emulator, to be followed by applied uncertainty analysis at a later stage. However, we expect that the potential use of the emulator will be constrained by similar challenges from the sensitivity of models performance – thus highlighting the common thread in this project. Early results of using an emulator for uncertainty estimation of sedimentation in an estuarine system suggest a hopeful outlook for combining emulators in combination with a transfer function.
Key study areas: Ijssel river (The Netherlands) | Nederrijn-Lek river (The Netherlands) | Waal river (The Netherlands)
Last modified: 17/06/2018
prof. dr. Suzanne Hulscher
Efficient uncertainty quantification for impact analysis of human interventions in rivers
14/06/2018 by Koen Berends et al.
Contains: Data repository access Journal publication Modeling tool upon request