Geoscience Reference
In-Depth Information
Scale is an issue in HRU modelling. The type of HRU representation used to predict runoff generation
might be expected to vary both with the hydrological environment and with the spatial scale at which
the HRU elements are defined (see also Chapter 9). Each HRU is generally considered as homogeneous
in its parameter values and response so that, for example, if surface runoff is calculated to occur, it will
do so over the whole HRU. The HRUs are also often treated independently with no explicit routing of
downslope surface or subsurface flows between HRU elements, only routing of runoff to the nearest
channel. This assumption of independence of position in the catchment is, in fact, necessary if HRUs
with similar characteristics are to be grouped together.
One advantage of the HRU approach is that the calculated responses can be mapped back into space
using visualisation routines in a GIS so that this can, in principle at least, provide information for a spatial
evaluation of the predictions. The major disadvantage is the way in which each HRU is considered to
be spatially homogeneous, an assumption that must become poorer as the HRU gets larger. At very
large scales, new methods may be needed, as in the macroscale hydrological models that are discussed
in Chapter 8. At the current time, we have no theory for predicting the appropriate model structure or
parameter values at one scale, given information at another scale. Indeed, it has been argued that we will
never have such a theory and that we will need to resort to models that are essentially scale dependent
(Beven, 1995, 1996, 2006b). There is an alternate view, expressed by Vinogradov
et al.
(2010) that we
should only use models that work well with common parameter values at all scales and that, if it is shown
that they do not, then there is something wrong with the model structure. There is work in progress to
show that their DHMS model can be successfully applied in this way across a range of catchments and
climatic regions (though, as we see in Chapter 7, the declaration of “success” in this respect depends on
how the simulations are evaluated).
In applying HRU models, I still feel that the scale dependence of the parameter values used in rep-
resenting each unit will generally need to be considered (see also Chapter 9 for more discussion of the
scale issue). Because there are many HRUs and several (or many) parameters are needed for each HRU,
it is not easy to calibrate parameters by an optimisation process (though see Section B6.2.5). In this, such
models face similar problems to fully distributed physically based models. The GIS may store soil type
and vegetation type but the information about model parameters for each classification may be highly
uncertain and may not be independent (e.g. rooting depth of a certain vegetation type may depend on
soil type or the hydraulic characteristics or erosivity of a soil type might depend on the type of land use).
The real hydrological response of an HRU may depend on the heterogeneity within the element, which
might not be well represented by homogeneous “effective” parameter values. This has to be an important
limitation on this type of model structure but, as discussed in Chapter 5, it is essentially a limitation of
all types of model given the limitations of our knowledge of how to represent the detailed variability of
hydrological systems. Again, it suggests that the predictions of such models should be associated with
some estimate of uncertainty.
6.7 Some Comments on the HRU Approach
With some exceptions, the semi-distributed HRU models that use conceptual storage representations as
a basis for their runoff predictions are the direct continuation of the lumped conceptual or explicit soil
moisture accounting models of the past. There are many such models that have taken conceptualisations
developed at the catchment scale to smaller spatial units, taking advantage of all the modern possibilities
of GIS overlays. In some cases, for example the SWAT model of Box 6.2, the pedigree of the conceptual-
isation from earlier models is quite explicit. The exceptions are those few studies that have started anew
by allowing that the process representations required in different areas of the landscape might be quite
different. The idea then is that if an assessment of the dominant processes can be made on the basis of
