WP 5: Process Uncertainty

Careful consideration and use of the terms ‘risk’ and ‘uncertainty’ is required when defining the objectives and scope of work. Text under original project description of work (DOW) mixes these terms. A clearer short summary would read:

“The objective of this work package is to identify and emphasise the uncertainty associated with the various components of the flood prediction process; namely breach formation, flood routing and sediment transport. The effect that uncertainty in each of these predictions has on the overall flood prediction process will be demonstrated through application to a real or virtual case study”.

Thus, the focus of this work under IMPACT will be to:

1. Investigate uncertainty within modelling predictions for predicting breach formation, flood propagation and sediment transport
   
   
2. Demonstrate how uncertainty within each of these modelling approaches may contribute towards overall uncertainty within the prediction of specific conditions (such as flood water level at a specific location)
   
   
3. Consider the implications of uncertainty in specific flood conditions (such as water level, time of flood arrival etc.) for end users of the information (such as emergency planners).

The scope of work under IMPACT does not allow for an investigation of uncertainty in the impact of flooding or in the assessment and management of flood risk. The assessment of modelling uncertainty provides essential information upon which a later assessment of the uncertainty in risk may be developed through further research.

The programme of work under IMPACT has deviated significantly from its original outline programme (DOW) since the issues involved are more complex than originally envisaged. In order to achieve the objectives outlined above, it is now necessary to focus on some specific stages of analysis, and in doing so to clearly define limits as to the extent and nature of analysis.

The proposed approach for analysis is as follows:

1. Analysis of individual model sensitivity to parameter variation will be undertaken. This may include either simple analysis of the effect of varying individual parameters and / or the effect of varying parameter probability distribution functions (pdf) – this where models can undertake Monte Carlo analysis and parameters can be represented in this way.
   
   This analysis will permit an understanding of model performance in relation to modelling parameters. It will demonstrate the relative importance of various parameters to modelling results.
   
   The extent to which the analysis is undertaken will be left to the (expert) judgement of the modellers such that all parameters which are likely to have significant influence on modelling predictions are analysed. Note that the choice of pdf is also open to expert judgement, hence sensitivity to assumed pdf may also be examined.
   
   
2. When analysing parameter variation [1 above] consideration should also be given to the difference between model parameter uncertainty and model physics uncertainty. That is, how well does the model represent the physical process? This issue is far from simple and, in effect, is the process through which model development is undertaken. Conclusions for model improvement will be made and where possible, (subjective) estimates as to how this might reduce uncertainty made.
   
   
3. Demonstrating the combined effect of model uncertainty will be undertaken through application of models to the Tous and / or Lake Ha! Ha! case studies.
   
   A probabilistic approach to combining uncertainties will not be undertaken. A less rigorous, more subjective but equally more practicable approach (within the IMPACT framework) will be taken whereby central, upper and lower scenarios will be estimated using the various models. This will require:
   
   

   a)	Clear identification of a parameter that is being investigated (e.g. water level at a specific location; time of flood wave arrival etc)
   b)	Expert judgement as to how individual modelling results might combine to give central, upper and lower estimates

   
   Where time permits, further combinations may be assessed to highlight combined model behaviour in more detail.

4. Consideration will be given to the combined modelling results to assess how potential uncertainty may influence use by end users. Prior to undertaking steps 1-3 above, it will be important to review who the key end users are, and how they might use the modelling results. This will influence the selection of modelling results under steps 1-3.

The following points were noted during discussion at the 3rd IMPACT workshop:

• Clarification of the scope of work on this topic under IMPACT is needed. Perceptions both within and external to the project vary significantly.
  This statement should clarify the scope of work. Details will also be posted to the web and the DOW updated in January 2004.
  
  
• Discussion upon methods of analysis focussed around two approaches; a more rigorous mathematical approach versus a less rigorous subjective approach.
  
  The work required undertaking a rigorous mathematical approach is extensive and requires specialist skills that IMPACT team members do not necessarily have. The scope of work proposed under IMPACT and the budget available does not allow for this. It was concluded that a more subjective approach would a) fit with the original scope and budget of IMPACT and b) meet the requirements for many end users. [In addition, the knowledge of pdfs for the so called ‘rigorous’ analysis does not exist. The ‘rigorous’ analysis therefore requires subjective assumptions that may be disguised in the presentation of the results which could be misleading as to the true level of knowledge.]
  
  
• The subjective approach means selection (at minimum) of scenarios representing an upper, lower and central estimate. Selection of conditions to create these scenarios will require expert judgement (hence subjectivity). Parameters and what may be considered upper, lower and central will vary from model to model, and modeller to modeller.
  This approach does allow different modellers (breach modellers particularly) to compare predictions for upper, lower and central estimates. Variation may be a function of judgement and / or uncertainty in the model representation of the physical processes.
  
  However, it is important to recognise that scenarios (i.e. selected parameter values or distributions) need to be credible. They are not necessarily limits or bounds of all possible outcomes and they have no probabilistic interpretation (e.g. confidence limits). By ‘credible’, it is suggested that parameter values selected have or may have occurred or may realistically be expected to occur.
  
  
• It is only realistic to consider uncertainty for a specific parameter at a specific location. No attempt will be made to assess uncertainty in multiple parameters simultaneously for the entire modelled area.
  
  
• The focus on analysing model uncertainty means that issues such as whether breach occurs or not, will not be considered.

Step1: Identification of end users and overall modelling parameters
HR to initiate. Other partners to contribute.
Identify likely end users for modelling outputs. Identify probable parameters for end user use.
Conclude which parameter(s) will be studied in combined modelling analysis
[End Jan 2004]

Step 2: Data set for sensitivity analysis
Theme Leaders for WP2, 3 & 4 to select data sets (case study or laboratory data) for sensitivity analysis of breach, flood propagation and sediment transport models.
Theme leaders to confirm parameter(s) for comparison during analysis (based upon 1 above)
[End Feb 2004]

Step3: Undertake parameter sensitivity analysis
Theme leaders and other partners to undertake sensitivity analysis for their respective models leading to a prediction of upper, lower and central estimates.
Theme leaders to also review differences between results where more than one model has been analysed for a specific function (e.g. multiple breach model predictions)
Initial conclusions on modelling uncertainty produced
[End Mar 2004]

Step4: Combined uncertainty analysis
Following review of Lake Ha! Ha! data, decision to be made as to final case study or studies data.
Agreement on extent of run scenarios.
Analysis of case study data
Theme leader analysis of combined results
[to be undertaken as part of overall case study application – mid summer 2004]

Step5: Implications for End Users
HR to initiate. Other partners to contribute.
Discussion leading to key points and observations for presentation at final workshop.
[Autumn 2004]

back to Work Programme

back to Top