A simple raster-based model for flood inundation simulation

Abstract

In this paper the development of a new model for simulating flood inundation is outlined. The model is designed to operate with high-resolution raster Digital Elevation Models, which are becoming increasingly available for many lowland floodplain rivers and is based on what we hypothesise to be the simplest possible process representation capable of simulating dynamic flood inundation. This consists of a one-dimensional kinematic wave approximation for channel flow solved using an explicit finite difference scheme and a two-dimensional diffusion wave representation of floodplain flow. The model is applied to a 35 km reach of the River Meuse in The Netherlands using only published data sources and used to simulate a large flood event that occurred in January 1995. This event was chosen as air photo and Synthetic Aperture Radar (SAR) data for flood inundation extent are available to enable rigorous validation of the developed model. 100, 50 and 25 m resolution models were constructed and compared to two other inundation prediction techniques: a planar approximation to the free surface and a relatively coarse resolution two-dimensional finite element scheme. The model developed in this paper outperforms both the simpler and more complex process representations, with the best fit simulation correctly predicting 81.9% of inundated and non-inundated areas. This compares with 69.5% for the best fit planar surface and 63.8% for the best fit finite element code. However, when applied solely to the 7 km of river below the upstream gauging station at Borgharen the planar model performs almost as well (83.7% correct) as the raster model (85.5% correct). This is due to the proximity of the gauge, which acts as a control point for construction of the planar surface and the fact that here low-lying areas of the floodplain are hydraulically connected to the channel. Importantly though it is impossible to generalise such application rules and thus we cannot specify a priori where the planar approximation will work. Simulations also indicate that, for this event at least, dynamic effects are relatively unimportant for prediction of peak inundation. Lastly, consideration of errors in typically available gauging station and inundation extent data shows the raster-based model to be close to the current prediction limit for this class of problem.