Geoscience Reference
In-Depth Information
B6.2.3 Enumerating the Assumptions of SWAT
As recommended elsewhere in this topic, the user of a model should always aim to list and
evaluate the assumptions made by a modelling package but in the case of SWAT there is a
certain difficulty in listing all the assumptions (especially if going beyond the rainfall-runoff
components to all the other dependent functions). The list depends on the specific options
chosen (and possibly calibrated) for a specific application.
Two critical assumptions are, however, worth making explicit:
A1 Each HRU within a subcatchment is assumed to respond homogeneously. Thus if an HRU
produces surface runoff, it does so over 100% of its area, regardless of scale. Any effects
of subcatchment heterogeneity (and processes such as runon infiltration) are therefore
assumed to be treated implicitly by the use of effective parameter values.
A2 SWAT is intended only to predict runoff above some “baseflow” (which implies that some
baseflow separation is required for model evaluation, e.g. Arnold and Allen, 1999). This
then means that even the water balance assumption can be relaxed for particular appli-
cations, by allowing losses to deeper groundwater which then make no contribution to
the predicted flow.
As discussed in Box 6.3, the original derivation of the SCS-CN method for predicting fast
runoff was made at the small watershed scale and would therefore have allowed for hetero-
geneity implicitly. It is not clear, however, how far the default estimates of curve numbers
and Green-Ampt infiltration parameters for different soils do so. Lyon
et al.
(2004) and Easton
et al.
(2008) have used the reconceptualisation of the curve number approach suggested by
Steenhuis
et al.
(1995) as a way of representing variable source areas within SWAT. They use the
TOPMODEL topographic index as a way of defining the HRUs for the SWAT model. However,
most applications still make the homogeneous response assumption (A1).
B6.2.4 Additional Components Available in SWAT
The intended use of SWAT for agricultural management means that many other component
processes have been added to the hydrological components: crop and forest growth routines;
fertiliser, manure and pesticide additions; tillage operations and grazing; erosion and trans-
port of sediments (using a Modified Universal Soil Loss equation), nutrients, pesticides, and
pathogens; in-stream processes affecting water quality; and management strategies including
irrigation and conservation practices. Many of these components require the specification of
time schedules for the operations and all require the specification of additional parameter sets.
B6.2.5 Calibration and Uncertainty in the SWAT Model
Any application of SWAT requires the specification of a large number of parameter values.
Whittaker
et al.
(2010), for example, report on an application to the Blue River basin in Okla-
homa with 55 sub-basins and 193 HRUs each of which requires many parameters depending
on the number of components included. One of the attractive features of SWAT is that it can be
downloaded with a database of default parameter values on the basis that the parameters are
physically based
and therefore can be defined with some realistic range. The default parameters
can be used to provide predictions of catchments for which no observational data are available
for calibration (e.g. Srinivasan
et al.
, 2010). This, of course, assumes that parameter values do
not need to compensate for model structural errors, heterogeneities within HRUs and scale
effects. So it is perhaps not surprising that the great majority of applications include some form
of manual or automatic calibration of selected model parameters and, more recently, attempts
to account for uncertainty in the estimated parameter values.
It is always best to concentrate calibration effort on those parameter values to which the
model outputs are most sensitive and there have been a number of sensitivity analyses of
