1 Spatial Modelling of the Terrestrial Environment - Georeferencial

96 Spatial Modelling of the Terrestrial Environment
5.3.3 Automatic Mesh Generation
Two-dimensional structured and unstructured grids form the basis of many flood models.
The automatic mesh generator of Horritt (1998) requires the meandering position of the river
channel banks and some (arbitrary) contour to describe the extent of the model domain.
Rather than digitizing this information from maps, the LiDAR segmentation system of
Cobby et al. (2001) described above can be used to automatically identify any river channel
and provide the chainage description of its banks in a format that can be used by the mesh
generator. Furthermore, the map of LiDAR topographic slope is used to define the domain
extent contour, so that the domain is narrower where steep slopes bound the channel than
where the floodplain is flat and wide. The number of elements in the domain is further
reduced by using larger elements away from regions of high process gradient, for example,
at the highly curved sections of a river (Horritt, 2000b). Figure 5.3 gives an example of a
two-dimensional finite-element mesh generated in this manner. Such methodologies could
also be developed for one-dimensional models and one can envisage a situation where
model discretizations can be generated at least semi-automatically from LiDAR data. In
the case of finite-element techniques, the mesh generation process could even incorporate
the vegetation pattern (hedge lines, woodland) information data from the LiDAR and CASI
surveys into the mesh. Thus, elements would follow the boundaries of such features and
enable spatially correct and well discretized friction fields to be specified.
5.3.4 Model Calibration and Validation Studies
Very little direct validation of inundation models against inundation extent data has been
conducted (or at least formally reported in peer-reviewed journals), and such schemes are
typically validated only indirectly against hydrometric data. As noted above, given the
relatively sparse hydrometric data available, this may be an insufficient test of an inundation
model. Recently, the first comprehensive trials of hydraulic (2D raster storage cell and 2D
FE) models against consistent inundation datasets have been conducted and published
(Bates and De Roo, 2000; Horritt and Bates, 2001a and b; Horritt and Bates, 2002). These
used satellite SAR images of flooding on the Rivers Meuse (Netherlands), Thames (UK)
and Severn (UK) and showed that it was possible to replicate these single, rather inaccurate
inundation images with relatively simple 2D raster storage cell models such as LISFLOOD-
FP (see Figure 5.4). The maximum predictive ability for all models calculated using the
F 〈2〉 performance measure given in equation (1) was in the range 70%–85% of inundated
area predicted correctly and therefore approached the 85%–90% accuracy of inundation
extent derived from currently available satellite SAR sources. More recent research using
the LISFLOOD-FP to model a single RADARSAT image of the October 1998 flood on
the River Severn, UK, has shown that, because of these errors, current satellite-derived
inundation images are no better than hydrometric data at constraining the predictions of
hydraulic models over some reaches.
As we note in section 5.2.1, for particular reaches two flooding images do exist and an
essential next step is to attempt model validation using these, rather than single image data.
This will allow split sample calibration and validation using inundation extent information
and allow calibration portability to be assessed. Hence, we should be able to ascertain the