Vulnerability based flood risk evaluation

Introduction

This case study arises from work on uncertainty in predicting flood inundation carried out at Lancaster University as part of the FRMRC project. It has been published as Pappenberger et al. (2006a). It demonstrates the use of the GLUE methodology in conditioning the LISFLOOD-FP 2D hydraulic model using both guaged water levels and historical inundation maps obtained by air-borne photogrammetry and SAR satellite images on the River Alzette in Luxembourg (Figure 1).

Figure 1: Outline of the study site, part of the flood plain of the Alzette in Luxembourg. Points and subdomains for model evaluation are indicated.

One of the interesting aspects of the study is that it was found that none of the Monte Carlo model runs carried out within GLUE was behavioural on all the evaluation measures used. All of the models tried could be rejected on one or more measures. There is then an issue about what is important in making the predictions. Predictions will be required most in areas that, for one reason or another, have the highest vulnerability. This might depend on economic value (infrastructure on the flood plain), it might depend on potential for loss of life, it might depend on critical routes for evacuation (in which case time of day might also be important). The paper shows how different sets of behavioural models might be used for different types of prediction. It represents an example of a strategy for learning about places outlined in Beven (2007).

Case study: details

Some 28000 Monte Carlo realisations of the LISFLOOD-FP model were made, varying channel and flood plain roughness, errors in bed topography and errors in the input discharge hydrolograph. Both global and local (points and 1km reaches) model evaluations were made within the GLUE framework. In each case, because the uncertainty inherent in some of the observed data (e.g. Figure 2), fuzzy model evaluations were used within the “limits of acceptability” approach outlined in Beven (2006). The details of the fuzzy evaluation are given in Figure 3. Fuzzy evaluations for assessing flood inundation models have also been used previously within the GLUE methodology in papers by Aronica et al. (1998); Bates et al. (2004), Pappenberger et al (2005, 2006b)

Figure 2: SAR image during the January 2003 flood on the Alzette River, Luxembourg

Figure 3: Details of fuzzy membership function used in vulnerability weighted evaluation of inundation model predictions in Pappenberger et al. (2006a).

Results

No model runs were found that satisfied acceptability criteria for all the evaluation measures used. It was found that some of the evaluation measures were poorly correlated. Thus, different evaluation measures, weighted by different measures of vulnerability, were used to provide sets of behavioural models for different types of prediction. This gave differences in the estimates of flood risk in different parts of the flood plain (Figure 4).

Figure 4: Differences in estimated flood risk predicted using different local evaluation measures relative to using global measure. A. weighted by road km. B. weighted by numbers of buildings.

Comment

Flood inundation prediction is an example of where there are many different sources of uncertainty that interact nonlinearly with model structural errors. It is therefore perhaps not too surprising that it might be difficult to find models that are acceptable everywhere and on all measures when evaluated against field observations. GLUE is a useful methodology in recognizing model failures in that it does not attempt to compensate for model deficiencies by using an explicit representation of the errors themselves (though it ca do as an additional model component in the analysis if required), since the structure of the errors is often complex and non-stationary. That means, however, that we might choose to use local evaluations for particular types of prediction (here taking different types of vulnerability into account) and not worry too much about models that do not produce good results in places where it is not important. There is, of course, always a danger of over-fitting to specific requirements, such that the quality of the predictions is degraded when applied to other events with different characteristics. This can only be addressed, however, when new data become available at which point the weights on individual models can be updated (see the discussion of Beven, 2007).

References

• Aronica, G, Hankin, B.G., Beven, K.J., 1998, Uncertainty and equifinality in calibrating distributed roughness coefficients in a flood propagation model with limited data, Advances in Water Resources, 22(4), 349-365.

• Bates, P. D., Horritt, M. S., Aronica, G. and Beven, K J, 2004, Bayesian updating of flood inundation likelihoods conditioned on flood extent data, Hydrological Processes, 18, 3347-3370.

• Beven, K. (2006). "A manifesto for the equifinality thesis." Journal of Hydrology 320(1-2): 18-36.

• Beven, K J, 2007, Working towards integrated environmental models of everywhere: uncertainty, data, and modelling as a learning process. Hydrology and Earth System Science, 11(1), 460-467.

• Pappenberger, F., Beven, K., Horritt, M., Blazkova, S., 2005, Uncertainty in the calibration of effective roughness parameters in HEC-RAS using inundation and downstream level observations, Journal of Hydrology, 302, 46-69.

• Pappenberger, F., Beven, K.J., Frodsham, K., Romanovicz, R. and Matgen, P., 2006a. Grasping the unavoidable subjectivity in calibration of flood inundation models: a vulnerability weighted approach. Journal of Hydrology, 333, 275-287.

• Pappenberger, F., Frodsham, K., Beven, K J, Romanovicz, R. and Matgen, P., 2006b. Fuzzy set approach to calibrating distributed flood inundation models using remote sensing observations. Hydrology and Earth System Sciences, 10,1-14.