1 Spatial Modelling of the Terrestrial Environment - Georeferencial

Flood Inundation Modelling Using LiDAR and SAR Data 93
be larger than this. This begins to indicate appropriate grid resolutions for use in hydraulic
modelling. In this respect, there is an analogy here between topography parameterization
and large eddy simulation of turbulence. In both, grid resolution distinguishes between those
flow features at the grid scale or above that are captured explicitly, and those at sub-grid
scales, which are more homogeneous and can merely be treated statistically in terms of their
impact on the mean flow field (see discussion on parameterization of sub-grid wetting and
drying models below) or as part of the resistance term. Hence, differently scaled elements
of the total frictional loss can be represented as combinations of these three, very different,
approaches. Nor are these different representations discrete, but rather they inter-grade and
interact.
In terms of the necessary spatial resolution for topographic data, Horritt and Bates (2001b)
have validated various resolution implementations of a raster-based flood inundation model
against hydrometric and satellite SAR inundation data for a 40-km reach of the River Severn
between the gauging stations at Montford Bridge and Buildwas. The topographic data used
to drive these simulations were derived from the 1999 LiDAR survey conducted by the UK
Environment Agency processed using the algorithm described in section 5.2.3 to yield a
10 m raster digital elevation model of the whole reach. Horritt and Bates tested 10, 25, 50,
100, 250, 500 and 1000 m resolution models generated by spatial averaging of the DEM
and found that the optimum calibration was stable with respect to changes in scale when
the model was calibrated against the observed inundated area. Observed and simulated
inundated areas were compared using the measure of fit:
F 〈2〉 = A obs ∩ A mod
A obs ∪ A mod
(1)
Here F 〈2〉 is thus a zero-dimensional global performance measure, where A obs and A mod
represent the sets of pixels observed to be inundated and predicted as inundated respectively.
F 〈2〉 therefore varies between 0 for a model with no overlap between predicted and observed
inundated areas and 100 for a model where these coincide perfectly.
In the case of the River Severn simulations F 〈2〉 reached a maximum of ∼72% at a
resolution of 100 m, after which no improvement was seen with increasing resolution. Projecting
predicted water levels onto a high resolution DEM improved performance further,
and a model resolution of 500 m proved adequate for predicting water levels. Predicted
floodwave travel times were, however, strongly dependent on model resolution, and water
storage in low lying floodplain areas near the channel was identified as an important mechanism
affecting wave propagation velocity as this is strongly influenced by the near channel
micro-topography and hence model resolution. The impact of changing topographic representation
therefore differs depending on what the modeller wishes to predict.
High-resolution topographic data has also stimulated two other developments. First, the
digital raster format of LiDAR has led to the creation of new flood inundation models
which directly integrate with this data format in a GIS-type framework (e.g. FLOODSIM –
Bechteler et al., 1994; GISPLANA – Estrela and Quintas, 1994; and LISFLOOD-FP -Bates
and De Roo, 2000; Horritt and Bates, 2001a and 2001b). These models use channel routing
combined with storage cell or 2D diffusive wave routing on floodplains to simulate dynamic
flooding in a computationally efficient manner. Such codes could also be readily integrated
with socio-economic datasets to provide complete management systems. Second, whilst a
number of algorithms have been developed which correct for dynamic wetting and drying