Geoscience Reference
In-Depth Information
the vertical. Any application of the SHE model will therefore require the specification of thousands of
parameter values. The parameter values required are effective values at the grid element scale, which
may not be the same as values that might be measured locally. There is also the potential to specify
fully distributed precipitation and meteorological data across the model grid elements, if the data are
available. The predictions are, however, dependent on the grid scale used. A study by Refsgaard (1997),
using the SHE model, is one of the few studies to have looked at the effect of the grid scale on the model
predictions. This study, of the Karup catchment in Denmark, compared predictions using a finest grid
of 500 m, to those with degraded grids of 1000 m, 2000 m and 4000 m. His conclusion was that it might
still be possible to obtain reasonable simulations of catchment discharge above 1000 m but that it would
require recalibration of parameters and possibly reformulation of some model components. Refsgaard
infers that not much improvement in accuracy would be gained by using finer scale grids than 500 m
but this conclusion may be conditional on the nature of the Karup catchment which is dominated by
groundwater flows. Xevi
et al.
(1997) and Vazquez
et al.
(2002) have also demonstrated that the results
of the SHE model are sensitive to grid size. It follows then that the effective parameter values to get good
predictions of the variables of interest, such as catchment discharge, should also be expected to vary with
grid size. The same effect should also be expected with other model fomulations.
The different SHE development teams have implemented impressive pre- and post-processing pack-
ages for preparing model applications and visualising the distributed predictions, including graphical
animations of the predicted responses. The distributed predictions of the SHE model have also allowed
other model components to be developed within the most recent versions, which are now being devel-
oped independently by the original partners. The UK version, SHETRAN, now based within the Water
Resource Systems Research Unit at the University of Newcastle, has added contaminant and sediment
transport components (Bathurst
et al.
, 1995, 2004; Ewen
et al.
2000). The DHI version, MIKE SHE, has
also added a contaminant transport component (Refsgaard and Storm, 1995). In both cases, the predictions
of contaminant transport are based on the advection-dispersion equation. Both DHI and the University of
Newcastle now have versions of SHE which make fully 3-D solutions for the unsaturated-saturated flow
domain. MIKE SHE has also added options to use a simple groundwater store where a fully susubsurface
solution is not justified and to predict a preferential recharge to the saturated zone as a simple proportion
of the infiltration rate (Refsgaard and Storm, 1995). Such modifications undermine the way in which
models purport to be “physically based”.
There have been other models based on grid elements, now largely superseded by the more general
modelling packages such as InHM, tRIBS and HydroGeoSphere. The fully 3-D models of Binley
et al.
(1989a, 1989b) and Paniconi and Wood (1993) use a grid-based spatial discretisation. The ANSWERS
model (see for example, Beasley
et al.
, 1980; Silburn and Connolly, 1995; Connolly
et al.
, 1997), which
has its origins in one of the very first fully distributed grid-based models of Huggins and Monke (1968),
essentially considers only an infiltration excess runoff generation mechanism, using the Green-Ampt
infiltration equation (see Box 5.2) to predict excess rainfall on each grid element. The runoff generated is
then routed towards the stream channel in the direction of steepest descent from each grid element. The
CASC2D model of Doe
et al.
(1996) and Downer
et al.
(2002) is similar in that it also uses a Green-
Ampt infiltration equation, but it uses a 2-D diffusion wave approximation to model overland flow on the
hillslopes and a 1-D diffusion wave model for the channel reaches. CASC2D was later extended to include
more subsurface flow processes as the Gridded Surface/Subsurface Hydrologic Analysis (GSSHA) model
(Downer and Ogden, 2004; Downer
et al.
, 2005). The 3-D version of HILLFLOW of Bronstert and Plate
(1997) is a grid-based model, with the interesting option of modelling the Richards equation using the
fuzzy logic
methodology of Bardossy
et al.
(1995). HILLFLOW also has a 2-D option for modelling
individual hillslope elements in a way similar to the models discussed in Section 5.2.3 and a 1-D version
for individual soil profiles. All the HILLFLOW versions have a component for modelling preferential
flow in macropores, at the expense of introducing additional parameters. Bronstert (1999) provides a
review of experience in using HILLFLOW in a variety of applications. Models that have been linked to
