Geoscience Reference
In-Depth Information
quality components, in the form of the US EPA's Hydrological Simulation Program - Fortran (HSPF,
Donigian
et al.
, 1995). Models of this type, called explicit soil moisture accounting (
ESMA) models
by
O'Connell (1991), varied in the number of storage elements used, the functions controlling the exchanges,
and consequently in the number and type of parameters required. The Stanford Watershed Model had up
to 35 parameters, although it was suggested that many of these could be fixed on the basis of the physical
characteristics of the catchment and only a much smaller number needed to be calibrated.
In the early years of digital computing, there was a tendency for every hydrologist with access to a
computer (there were no personal computers then) to build his own variant of this type of model. It was
not, after all, a difficult programming exercise. It was one of the ways that I learned how to program
computers as an undergraduate by writing a model to try to “hindcast” runoff generation on Exmoor
during the Lynmouth flood event. This model was written in the Algol programming language, in 1971,
stored on punched cards and run on the Bristol University Elliot 503 computer with 16 kB of memory,
with all output to paper from a lineprinter. This is an indication of how rapidly the resources available to
the modeller have changed in the last four decades.
Most of these models had sufficient number of parameters and flexibility to be able to produce a
reasonable fit to the rainfall-runoff data after some calibration. Indeed it was all too easy to add more and
more components (and more associated parameters) for different processes. The potential for a confusing
plethora of models was soon recognised, and Dawdy and O'Donnell (1965) tried to define a relatively
simple “generic” model structure, with just a few parameters (Figure 2.8). This did not, however, stop a
continued expansion in the number of models of this type published in the hydrological literature (see,
for example, the review by Fleming, 1975). A number of examples that are still in current use include the
HSPF, SSARR and Sacramento models from the USA, the HBV model from Sweden, the Tank model
from Japan, the UBC model from Canada, and the RORB and AWBM models from Australia (Singh,
1995; Singh and Frevert, 2005; Boughton, 2011).
A comparison of different models of this type reveals the subjectivity involved in defining a particular
model structure, albeit that there is often similarity in some components. An example in current use,
variously known as the Xinanjiang model, Arno model or Variable Infiltration Capacity (VIC) model, is
described in Box 2.2. This model is of interest in that, although it can be classed as an ESMA-type model,
the surface runoff generation component can also be interpreted in terms of a distribution function of
catchment characteristics (see Section 2.6). It has also been implemented as a macroscale hydrological
model or
land surface parameterisation
in some global climate models.
Functionally, there are also similarities between the modern IHACRES package, mentioned in Section
2.3, and ESMA-type models, since in both cases the runoff generation and runoff routing components
are based on storage elements. The difference lies in the modern approach of trying to find the simplest
model structure supported by the data (see the discussions by Jakeman and Hornberger, 1993, and Young,
2000, 2003) and in not necessarily fixing the model structure beforehand. Instead, an analysis of the data
should be allowed to suggest what the appropriate structure should be, as in the data-based mechanistic
approach of Young and Beven (1994) (see Chapter 4).
Providing some data are available to calibrate parameter values, the results from even simple ESMA
models can be quite acceptable, both in modelling discharges (Figure 2.9) and in soil moisture deficit
modelling (Figure 2.10). The performance demonstrated in Figure 2.10 is particularly impressive if it is
remembered that 1976 was one of the driest summers on record in the UK. A number of comparisons
of ESMA models have been published, although the limitations both of model structures and of input
and output data means that it has not generally been possible to conclude that one model consistently
performs better than another after the model parameters have been calibrated (see, for example, the studies
of Franchini and Pacciani, 1991; Chiew
et al.
, 1993; and Editjatno
et al.
, 1999). More recently, software
has been developed that allows different model components to be put together to allow appropriate model
structures to be found for particular applications. This software includes the USGS Precipitation-Runoff
Modelling System (PMRS) of Leavesley
et al.
(2002), the Imperial College Rainfall-Runoff Modelling
