Geoscience Reference
In-Depth Information
This history of attempts to model R-5, one of the most intensively studied in the world, was summarised
as a “never ending story” by Loague
et al.
(2000). The story did not, however, end there and has continued
(Loague and VanderKwaak, 2002, 2004; Loague
et al.
, 2005; Heppner and Loague, 2007; Heppner
et al.
,
2007). The role of time variant infiltration parameters, the roads and deeper subsurface flows in improving
the predictions of discharge at the site have all been examined. The understanding of the hydrology at this
site has changed from an assumption that the response is dominated by surface runoff, to one in which the
subsurface plays a much more important role. Thus the original attempt to determine model parameter
values by an extensive (and expensive) programme of infiltration measurements had only limited success.
Determining parameter values by measurements deeper into the soil is, of course, much more difficult and
even more expensive. This has implications for the application of detailed physically based distributed
models more generally (remembering that R-5 is a catchment of only 9.6 ha).
Similar stories of using this type of model to help understanding of hillslope and catchment responses
have been reported from elsewhere. InHM has also been applied to the Coos Bay site in Oregon (Ebel
et al.
,
2007, 2008; Ebel and Loague, 2008), while the TOUGH2 code has been applied to the Panola hillslope in
Georgia (James
et al.
, 2010). Both of these sites have been the subject of intensive field experiments and
would have made interesting case studies here. They are both worth investigating further. The Coos Bay
site is particularly interesting in that it was a very steep hillslope hollow that eventually failed as a shallow
landslide during a high volume storm. The failure then revealed that deeper subsurface flows through
fractured bedrock, concentrating at two locations, might have contributed to the excess pressures that
triggered the shallow landslide (Montgomery
et al.
, 2002). InHM has been extended to allow sediment
transport simulations (Heppner
et al.
, 2007; Ran
et al.
2007).
5.8 Good Practice in the Application of Distributed Models
Validation
of distributed models is an issue that has received a great deal of recent attention in the
field of groundwater modelling following a number of studies in which predictions of groundwater
behaviour were not borne out by subsequent experience (see Konikow and Bredehoeft, 1992). Some of this
discussion has, in fact, suggested that validation is not an appropriate term to use in this context, since no
model approximation can be expected to be a valid representation of a complex reality (e.g. Oreskes
et al.
,
1994). Model evaluation has been suggested as a better term. Because distributed models make distributed
predictions, there is a lot of potential for evaluating not only the predictions of discharge at a catchment
outlet, but also the internal state variables, such as water table levels, soil moisture levels and channel
flows at different points on the network. A decade ago, when the first edition of this topic was written, there
had been relatively few attempts to validate the predictions of distributed models (Bathurst, 1986; Parkin
et al.
, 1996; Refsgaard, 1997). Now there have been a few more, such as those discussed in the case studies
(Bathurst
et al.
, 2004; Loague and VanderKwaak, 2002, 2004; Ebel
et al.
2007; James
et al.
, 2010).
The lack of evaluation with respect to internal
state variables
is clearly partly due to the expense of
collecting widespread measurements of such internal state variables. Hence, nearly all of these detailed
evaluations have been made on small, intensively studied catchment areas for research purposes. An inter-
esting development in this respect has been the construction in 2005 of the 6 ha artificial “Chicken Creek”
catchment on an old open-cast-mine site in Germany. In constructing the catchment, significant effort
was made to ensure that the base of the catchment was water tight and that the soil that was put in place
was homogeneous (as far as possible given the amount of material that had to be moved), free of macro-
pores and of known textural characteristics. The first modelling studies of this catchment have now been
reported, as a competition between models using
a priori
parameter estimates (Hollander
et al.
, 2009).
The results of this comparative study have some interesting implications for good practice in the appli-
cation of distributed models. It was found that the more complex models based on continuum physics (both
CATFLOW and HYDRUS2D were included in the comparison) did not perform any better in predicting
