Environmental Engineering Reference
In-Depth Information
large-scale impacts, and then back again to the
small scale where the (derived) data on vulnera-
bility are plotted on the vulnerability map. Since
we do not have the means available to do this
tracking through the catchment itself, particularly
for design floods, we have to resort to using a
catchment model of suitable form.
One step that can be taken to investigate this
problem is to track sensitivity as it propagates
through the modelling process. This is possible
using a technique called reverse (or adjoint) algo-
rithmic differentiations (Griewank 2000; Hascoet
and Pascual 2004), which can be applied to any
model. It can also, as is needed here, be applied to
an entiremodelling process, for example fromfield
plot measurements to flood frequency curve pre-
diction, provided the process is fully specified as
an algorithm (i.e. a set of unambiguous sequential
instructions). Figure 2.8 shows amap of sensitivity
derived using an adjoint version of the physically
based spatially distributed catchment rainfall-
runoff model SHETRAN, created during a trial of
this technique (Ewen, personal communication
2009; SHETRAN 2009). For a single storm
a catchment is an interconnected dynamic sys-
tem, so the effect of any given intervention is
likely to vary from storm to storm because of
changes in the spatial pattern and intensity
of rainfall, changes in interactions and timings of
flood waves in the drainage network, and changes
in any number of other things that control runoff
and routing at different scales (e.g. Burt and Slat-
tery 1996). Conceivably, an intervention that is
thought to be beneficial on initial inspection, or
following short-term monitoring, might actually
increase the flood hazard downstream (e.g. it
might increase the 50-year return period flood).
Essentially, vulnerability maps show the sen-
sitivity of large-scale impacts to changes in
small-scale properties. It is therefore important
to return to the broad question of how detailed
knowledge and understanding gained at a small
scale finally ends up affecting predictions made
for impacts at a large scale. To create a good
vulnerability map, it is necessary to track the
propagation of information all the way from
small-scale information (e.g. behaviour seen on
field plots), through the catchment and up to the
Fig. 2.8
Vulnerabilitymap forDunsop catchment (26 km
2
) createdusing adjointmodelling, showing sensitivity of flood
peak flow to change in Strickler coefficient. (See the colour version of this figure in Colour Plate section).
