A methodology for the validation of uncertain flood inundation models

A rationale for the comparison of uncertain spatial predictions of flood extent with observed data is presented. The accuracy of model predictions is quantified through use of the reliability diagram and the associated root mean square error between uncertain predicted inundation and observed flooded proportion, and the precision measured through use of an entropy-like measure. The use of these two criteria is tested for the calibration and validation of a simple cellular flow model on a reach of the river Severn, UK, for which satellite imagery of flood extent for two events is available. The two measures give an insight into the differences between the calibration and validation processes and the effectiveness of data from different satellite sensors in constraining uncertain model parameters, and allow uncertainty and precision in model predictions to be quantified in a meaningful way.