Bayesian estimation of flood inundation probabilities as conditioned on event inundation maps.

The generalized likelihood uncertainty estimation (GLUE) methodology is applied to the problem of predicting the spatially distributed, time-varying probabilities of inundation of all points on a floodplain. Advantage is taken of the relative independence of different effective conveyance parameters to minimize the simulations required. Probability estimates are based on conditioning predictions of Monte Carlo realizations of a distributed quasi-two-dimensional flood routing model using maps of maximum inundation and aerial photographs of flooding in the area. The methodology allows posterior distributions of conveyance parameters to be estimated and maps of inundation potential probabilities to be drawn up for flood events of different magnitudes. The results suggest that combining information from different magnitude events should be done with care, as the distributions of effective parameter values may vary with event magnitude. The value of accurate topographic information that is consistent with mapped inundation is also highlighted. The methodology can be used to obtain dynamic probabilities of floodplain inundation in real time forecasting.