The impact of uncertainty in satellite data on the assessment of flood inundation models

The performance of flood inundation models is often assessed using satellite observed data; however, these data have inherent uncertainty. In this study we determine the patterns of uncertainty in an ERS-2 SAR image of flooding on the River Dee, UK and, using LISFLOOD-FP, evaluate how this uncertainty can influence the assessment of flood inundation model performance. The flood outline is intersected with high resolution LiDAR topographic data to extract water levels at the flood margin, and to estimate patterns of uncertainty the gauged water levels are used to create a reference water surface slope for comparison with the satellite-derived water levels. We find the residuals between the satellite data points and the reference line to be spatially clustered. A new method of evaluating model performance is developed to test the impact of this spatial dependency on model calibration. This method uses multiple random subsamples of the water surface elevation points that have no significant spatial dependency; tested for using Moran's I. LISFLOOD-FP is then calibrated using conventional binary pattern matching and water elevation comparison both with and without spatial dependency. It is shown that model calibration carried out using pattern matching is negatively influenced by spatial dependency in the data. By contrast, calibration using water elevations produces realistic calibrated optimum friction parameters even when spatial dependency is present. Accounting for spatial dependency reduces the estimated modelled error and gives an identical result to calibration using spatially dependent data; it also has the advantage of being a statistically robust assessment of model performance in which we can have more confidence. Further, by using the variations found in the subsamples of the observed data it is possible to assess how the noisiness in these data affects our understanding of flood risk. This has highlighted the requirement for a probabilistic treatment of observed data, and using multiple subsamples is one way of achieving this. (C) 2011 Elsevier B.V. All rights reserved.