Geospatial Today

Better flood inundation models with RS

D. C. Mason, M. S. Horritt
P.D. Bates and N.M. Hunter

Flooding amounts to one third of all economic losses due to natural hazards in the world. Global warming and increased building activity on flood plains are leading to increasing instances of flooding. Better flood models are the need of the hour. Two dimensional hydraulic models are currently at the forefront of research into river flood inundation prediction. Remote sensing data like SAR and LiDAR data are proving to be rich sources for parameterisation and validation of these models. This is also aiding in extending the rural flood extent prediction to the urban environment.

Globally, flooding causes about one half of all fatalities and one third of all economic losses due to natural hazards. It is also on the increase, due partly to the effects of global warming and partly due to increased building on floodplains. The UK Hadley Centre estimates that in the UK, river floods that were previously 1 in 100 year events will become 1 in 10 year events over the next century. In the UK, it is a standard practice to employ a flood risk map to identify the risk at a particular location. There are two main end-users for such maps. One is the UK Environment Agency (EA), which provides web-based flood risk maps for the general public and maintains the existing network of river flood defences. Many defence works are in urgent need of maintenance, and it is important that the areas most at risk of flooding are identified in order that the associated flood defence works can be prioritised and completed. The other main end-user is the insurance industry, which uses the maps to set insurance premia for properties depending on their flood risk. Computerised models of river flood flow are currently used to predict maps of flood inundation extent. These have a number of limitations, and there is a need to improve current maps by developing better flood models.

Two dimensional hydraulic models are currently at the forefront of research into river flood inundation prediction. These models often adopt a 2-D finite element approach, solving the shallow water equations at each node of an irregular mesh covering the channel and floodplain. Each node of the mesh must be assigned a topographic height and a bottom friction factor. During a model run, the time-evolving water depths and flow velocities at each node are calculated, given the input flow rate to the reach and any other boundary conditions.

The two dimensional nature of these models require spatially distributed 2-D data for their parameterisation and validation. Until recently, development of these models has been hampered by lack of suitable data. As regards data for parameterisation, there has been inadequate floodplain topography data. Inundation depends on topographic features with length scales <~50m horizontal and <~0.2m vertical. UK Ordnance Survey maps provide inadequate coverage, being limited to 5m contours. The specification of flow resistance also remains a significant problem. In lowland floodplains at inundation depths <1m, flow resistance is probably dominated by vegetation. Typically flow resistance currently has to be left as a free parameter in the model, with a single bottom friction factor being specified for the whole of the floodplain and a different friction factor specified for the channel. As regards data for validation, routinely collected hydraulic data consist solely of bulk flow measures, namely discharge rates and water depths. The spacing of river gauges is quite sparse (10 –20 km), so that they usually provide validation data only at the external boundaries of the model. It is easy to calibrate models against such data and therefore they have a limited value in reducing calibration uncertainty.

Remote sensors carried on satellites and aircraft are now proving to be a rich source of spatially distributed data for model parameterisation and validation. We have been using satellite and airborne Synthetic Aperture Radar (SAR) and airborne scanning laser altimetry (LiDAR) data to improve flood models.

Firstly, we have been using inundation extent measured from SAR imagery to validate the modelled flood extent. SAR has the advantage of being day-night and all-weather, important for the storm conditions that often prevail during flooding, as a visible band satellite cannot see through cloud. Generally flood water will appear darker than adjacent land in a SAR image. An image segmentation algorithm to determine flood extent in a SAR image has been developed, which uses a statistical active contour model or ‘snake’ to distinguish flood water from other classes (Figure 1). We estimate that satellite SAR can predict the true inundated area to an accuracy of 85-90%, the main error being due to unflooded wet vegetation giving similar returns to water (Horritt et al., 2001).