1 Spatial Modelling of the Terrestrial Environment - Georeferencial

80 Spatial Modelling of the Terrestrial Environment
physical basis, it is extremely difficult to measure in the field, is highly variable in space
and time and is rather sensitive as a model parameter. In addition, how frictional effects
vary with scale in terms of ‘effective’ grid scale parameters (Beven, 1989) is also poorly
understood. There are usually, therefore, a wide range of parameter values for any given
problem that could be considered ‘physically acceptable’ and the calibration space for
the model is thus poorly constrained. Typical applications have tended to use one friction
value for the channel, which should control the point at which bankful discharge is
exceeded and water moves onto adjacent floodplain sections, and one for the floodplain,
which should control the floodplain flow velocity and depth. In practice, however, even
this simple spatial disaggregation is probably not warranted by the available data and one
could likely obtain an equally good fit to an external bulk flow hydrograph with a single
calibrated value for boundary friction. In terms of discriminatory power, such data tests
the ability of a calibrated hydraulic model to replicate flow routing behaviour and very
little else. Moreover, as Beven and Binley (1992) point out, it is likely that there will be
many available parameter sets that provide an equally good match to the available data
at a single location, yet the performance of these parameter sets will differ elsewhere
in the distributed model domain and during further events, particularly if these are substantially
different to the calibration event. The predictive uncertainty resulting from such
equifinal behaviour is a significant problem for hydraulic modellers and it is clear that
validation data from available gauging stations cannot discriminate well between different
models and different parameterizations. Neither will non-distributed validation data at the
model external boundary require complex (or even accurate) topographic data to complete
the model specification. In such cases, topographic errors are easily subsumed within the
calibration process and it is impossible to discriminate between topographic, boundary
condition, parameterization or model conceptual errors. Using such validation we are thus
unable to identify sources of error and hence cannot design new schemes or even rigorously
inter-compare those hydraulic models which are available. It is not clear whether
the instrumentation of reaches with multiple gauges is capable of solving this problem,
as it will still be relatively easy to calibrate friction in the sub-reaches between gauges
and we merely reduce the scale of the calibration problem rather than change its essential
basis. However, this possibility is yet to be adequately tested and is deserving of further
research.
The above discussion raises interesting questions concerning the assimilation of spatial
data into distributed models, the relative merits of lumped versus distributed parameterization,
calibration, validation and uncertainty analysis and how errors propagate through
complex non-linear models. However, the approach taken to hydraulic model calibration
and validation also has significant practical consequences for estimation of flood envelopes
and maps of flood risk. These are now statutory requirements in a number of countries,
including the UK, and are based on delimiting the flood extent (ie. a single line on a map)
resulting from a particular discharge, typically the 100-year flow. For most reaches such
high magnitude flows do not occur within the hydrometric record, and even if they do, the
resulting inundation has rarely been mapped in a consistent manner. Flood envelope construction
therefore relies on hydraulic modelling calibrated in the manner outlined above.
Inundation extent is predicted by a single model run at the design discharge, with a single
set of parameter values derived by calibration against a limited number of gauging