Geoscience Reference
In-Depth Information
the SWAT HRUs (Figure B6.2.1), as far as possible keeping each field as a separate HRU. This
resulted in three sub-basins and 18 HRUs. Although this is a very small catchment, the model
was used with a daily time step. In an initial test of the hydrological performance, baseflow was
removed using the Arnold and Allen (1999) filter, potential evapotranspiration was estimated
using the Hargreaves equation, and a manual calibration of seven parameters was performed,
including the curve number, available water capacity, soil evaporation compensation factor
and hydraulic conductivity of the top soil.
Reasonable results were obtained but, by comparison with measurements at other sites in
the region, predicted surface runoff seemed to be overestimated and loss to evapotranspiration
underestimated. The latter appeared to be related to problems with the crop growth model,
which is based on accumulated heat in the growing season. The model was modified so that
plant heat units were calculated separately for each HRU outside of the model, while crop
parameters, such as maximum leaf area index, canopy height and rooting depth, were specified
based on published information. The model was then recalibrated by running best estimate
and upper and lower perturbed values of the most sensitive parameters. The revised model
was subject to a split sample evaluation. The results from the validation period are shown
in Figure B6.2.2. The proportion of surface runoff predicted was felt to be more reasonable,
but the model still tends to underpredict peak flow values. This has been reported in many
SWAT model applications, but here might partly be induced by the discretisation problems of
using a daily time step in simulating a small catchment. The authors suggest that if the curve
numbers were updated based on the water content of the upper layer rather than the whole
soil profile (in a way that might also reflect surface effects such as crusting), it might lead to
further improvements in the predictions.
Figure B6.2.2
Predictions of streamflow for the Colworth catchment using the revised ArcView
SWAT2000 model: validation period (after Kannan
et al.
, 2007, with kind permission of Elsevier).
B6.2.7 Some Comments on SWAT
Until the introduction of the sub-daily time step options in SWAT, it had always been presented
as a long term yield model rather than a hydrograph prediction model (Arnold
et al.
, 1998;
Peterson and Hamlett, 1998). This is, however, somewhat at odds with the requirement of
getting the partitioning of the runoff processes correct when these processes are drivers for the
other components predicted by the full model. The results of Figure B6.2.2 are not untypical
of those obtained with the SWAT model; the predictions are good in places and not so good in
others. It is difficult to know howmuch belief to put into the partitioning of the runoff processes
since there is rarely any independent information with which to evaluate the predictions.
It is, however, understandable why the SWAT model has been so widely used given the range
of facilities it provides, the ability to freely download the model, the links to GIS databases,
the way it can reflect catchment characteristics and land management strategies, and the
