Modelling Catchment and Fluvial Processes and their Interactions - Environmental Modelling: Finding Simplicity in Complexity - page 183

of physically based models is their application in ungauged
catchments, though we will see later that gaps in param-
eterization data and process knowledge create model
uncertainty and thus the need for these physically based
models to be calibrated against gauging-station data.

and topography, there are many hydrologically impor-
tant properties of catchments, notably those of the soils
and the surface, that cannot be measured remotely. The
measurement of these properties requires intensive field
campaigns, which even so cannot provide the kind of
spatial coverage required to justify their distribution in
models. For example most hydrological models are highly
sensitive to the (saturated) hydraulic conductivity of soils
( K sat ) - see Davis et al . (1999); Chappell et al . (1998);
Michaelides and Wainwright (2002) - which is notori-
ously difficult to measure for volumes greater than a
few hundred cm 3 , particularly because of the presence of
macropores. It is fair to say that there is a significant mis-
match in the sophistication of our physically based models
and the sophistication of the data collection technologies
used to parameterize them. Moreover, as highlighted in
chapter 1, modelling is rather inexpensive compared with
fieldwork and is also perhaps more glamorous, more com-
fortable and more suited to producing publications - see
the discussion of Klemes (1997: 48) - so the gap between
the models and the data to fill them widens ...
It is the imbalance between the model sophistication
and the availability of data at appropriate scales (as well as
our incomplete understanding and thus mathematization
of the processes themselves), which means that even
the most sophisticated models rarely perform well in a
predictive capacity. Empirical models tend to be better
predictors. Thus, a process of parameter calibration is
often exercised on physically based distributed models
to ensure agreement of predicted versus observed runoff.
This process, of course, compromises the physical realism
of the model and thus its ability to explain as well as
to predict. Since explanation is why physically based
distributed models exist, this compromise is a serious one.
If it were not for explanation then an equally predictive
empirical model would always be the best model because
of its parsimony.
An interesting corollary emerges in the use of inverse
modelling - the use of a model structure together with
measured data to estimate what the physical parameters
would be that produced the measured outputs (see Kunst-
mann et al . (2006) for a catchment example and Chapter 8
for more detail). Notwithstanding issues of equifinality
(see below), such an approach must assume that the model
structure is correct, whereas we know that it can only ever
be a simplification of reality. Thus, here as elsewhere,
models and data are inextricably linked (chapter 1).
Some authors have argued that calibration can
render physically based distributed models closer to
overparameterized, conceptual, lumped models (Beven,

11.2.5 Physicallybasedmodels
Since it was first 'blueprinted' by Freeze and Harlan in
1969, distributed, physically based modelling has become
very widespread, on the assumption that a spatially vari-
able physical system is inherently more realistic than
a lumped statistical one. This assumption is likely to be
true but must be considered within the context of spatially
distributed models being themselves often crude simpli-
fications of any spatial variability that does exist in real
catchments. Remote sensing has gone some way towards
improving the observability, at least indirectly, of surface
properties at the catchment scale but subsurface proper-
ties are still largely unobservable at any scale other than
the point or line transect. Examples of current distributed,
physically based models include the SHE model (Systeme
Hydrologique Europeen: see Abbott et al ., 1986) and the
MIKE-SHE and SHETRAN descendants of it (Bathurst
et al ., 1995; Refsgaard and Storm, 1995), the IHDM model
(Institute of Hydrology Distributed Model; e.g. Calver and
Wood, 1995), the CSIRO TOPOG model (e.g. Vertessy
et al ., 1993), Thales (Grayson et al ., 1992a), and WEC-C
(Croton and Barry, 2001), SWAT (Arnold et al ., 1998) and
WaterWorld (Mulligan et al ., 2010a, Chapter 20) amongst
others. Physically based models should be derived deduc-
tively from established physical principles and produce
results that are consistent with observations (Beven,
2002). In reality they are often one of these but rarely both.
According to Ward and Robinson (2000), SHE was
developed jointly by the UK Institute of Hydrology (IH),
the Danish Hydraulic Institute (DHI) and the Societe
Grenoblois d' Etude et d'Applications Hydrauliques
(SOGREAH). It was specifically designed to address the
impact of human activities on catchment processes. This
type of scenario analysis is very difficult to address with
empirical or conceptual models but is the main focus of
most physically based models. In SHE, a finite difference
(grid-based) approach is used in three dimensions with
up to 30 horizontal layers in each cell. Surface and
groundwater flow is two-dimensional whilst flow in the
unsaturated zone is one-dimensional. The model has
been widely applied and continuously updated since 1986.
Whilst remote sensing provides some high-quality
spatial datasets for properties such as vegetation cover