Assessing and Reducing the Uncertainty of Predictions from Hydraulic and Hydrologic Models
The goal of this project is to develop a methodology and associated software to assess parameter uncertainty and its impact on the predictions of hydrologic, hydraulic, and sediment transport models. Reclamation relies on computer models in a wide range of activities from analysis of dam safety to prediction of bank erosion. These models are typically used to predict the behavior of a system for conditions that cannot be directly observed.
Unfortunately, model predictions always contain some degree of error. A key source of error lies in the values of the model parameters, which are constants that control the model's behavior. The parameter values are usually determined by calibrating the model output to reproduce some available observations. Unfortunately, multiple sets of parameter values can often reproduce such observations, and these parameter sets may produce very different predictions for the unobserved conditions. This Science and Technology (S&T) Program research project will help identify such cases.
Need and Benefit
Model uncertainty can add significant costs to Reclamation projects. In general, neither the conditions under which a project must function nor the impacts of a project on the surrounding environment can be known with certainty in advance of the project's implementation. Consequently, projects are planned and designed using scenarios that fall at the conservative end of the range of plausible outcomes. This procedure results in the over-design of hydraulic structures and the over-estimation of project impacts. If uncertainty in the predictions of models can be reduced, then the range of possible outcomes would also be reduced, which would ultimately increase the efficiency of project designs.
The uncertainty of a model prediction is essentially derived from three sources:
* Input data (e.g. riverbed material)
* Model formulation and implementation (e.g. the methods used to predict sediment movement given a flow)
* Future conditions (e.g. future river flows)
These three sources of uncertainty are often blended and not quantified. The methods and tools developed under this research project will help to quantify the uncertainty induced in model predictions caused by each of these factors individually.
The software tools developed by this research project would benefit three general activities related to hydrologic and hydraulic modeling:
* Data collection: The software would allow a user to determine observation strategies that reduce the uncertainty in the model parameters, which would make the model predictions more reliable. The tools will help identify the most important input parameters and input data for accurate predictions. For example, if future sediment deposition needs to be predicted at a flood prone area, the tools will help identify the data is most important to reduce the uncertainty of that prediction. The tools would quantify the error in the prediction caused by an error in the bed material collection. As another example, the tools would help quantify the error in the prediction caused by errors in the assumed sediment transport formula.
* Computer model improvement: Nearly all of these models use simplified representations of the actual hydrologic and hydraulic processes. The significance of these approximations depends on the particular application of the model. The software will allow a user to assess whether the mathematical structure of the model is adequate for the particular application being considered. By identifying the circumstances where the model structure is more or less certain, improvements can be made that specifically benefit Reclamation applications. For example, if one is trying to predict river reaches where deposition will occur, the tools developed under this research project will help identify which reaches can be simulated accurately and in which reaches the model is deficient.
* Reclamation and its project partners' decisionmaking: As described above, numerical models are commonly used to guide decisionmakers by providing forecasts for expected conditions and/or worst-case conditions. These forecasts are typically derived from engineering judgment rather than a systematic analysis of the possible outcomes. The more formalized procedure developed in this project would help ensure that the actual range of possible outcomes and their likelihood of occurrence are understood in advance. For example, a distributed watershed model is sometimes used to predict flows into a reservoir. The uncertainty of these predictions is sometimes not known. This research proposal would develop the methodology necessary to assess that uncertainty.
Independent Peer Review
The following documents were reviewed by qualified Bureau of Reclamation employees. The findings were determined to be achieved using valid means.
A METHOD FOR ASSESSING IMPACTS OF PARAMETER UNCERTAINTY IN SEDIMENT TRANSPORT MODELING APPLICATIONS (final, PDF,
By Ms. Morgan Ruark
Report completed on January 17, 2012