Geoscience Reference
In-Depth Information
Such models are also now being used as the basis for the type of distributed hydrological response unit
models noted earlier, such as the Grid to Grid (G2G) model (Moore
et al.
, 2006; Bell
et al.
, 2009).
2.7 Recent Developments: What is the Current State of the Art?
Computers continue to get more powerful. As a result there is absolutely no doubt that distributed
hydrological models will become more detailed, more complex and more closely coupled to geographical
information systems for the input of data and display of results (see, for example, Maidment, 2002;
Vieux, 2004; Refsgaard
et al.
, 2010). This, in one sense, is the current start of the art. There is a question,
however, as to whether this type of development will lead to better hydrological predictions. The answer
to this question is not at all clear. More complexity means more parameters; more parameters mean
more calibration problems; more calibration problems often mean more uncertainty in the predictions,
particularly outside the range of the calibration data.
This still leaves open the possibility that other, parametrically simpler, models may have much to
offer. If the interest is in discharge prediction only, then it would appear that, where calibration data are
available, simple lumped parameter models, such as IHACRES, can provide just as good simulations
as complex physically based models. For distributed predictions, no study has yet demonstrated that a
fully distributed model can do better in predicting the distributed responses in a catchment than a much
simpler distribution function model, such as TOPMODEL, where the assumptions of the simpler model
are reasonably valid (see Franchini and Pacciani (1991) and discussion of Chapter 7).
It is probable that these various categories of rainfall-runoffmodel will start to be subsumed into a single
framework in the next generation of software products. Two important concepts that have been introduced
in the decade since the first edition of this topic will be important in shaping the next generation of models.
These are the Representative Elementary Watershed (REW) concept first introduced by Reggiani
et al.
(1999) and the Models of Everywhere concept introduced by Beven (2006c). These ideas will change the
way that modellers approach the modelling problem and so have been given a discussion of their own in
Chapter 9.
Something that has not changed is that the application of all types of model is limited by the available
data on how hydrological systems work. Models are data constrained because of the strong limitations
of current measurement techniques. However, even if improved measurement techniques lead to a better
understanding of complex flow processes, it appears that it will be necessary for the foreseeable future
to distinguish between models developed for understanding (which describe those processes in detail at
small scales) and models developed for prediction at catchment scales. The former will certainly depend
on detailed (even if statistical) descriptions of the geometry of the flow domain. The latter will not be
able to demand such inputs because they will not be measurable practically or economically at the larger
scales at which predictions are required. Models for prediction will necessarily reflect the types of data
that are readily available.
In Chapter 3, we look at the data available for modelling in more detail. Chapter 4 then takes a look at
modern lumped catchment scale models, Chapter 5 at fully distributed physically based models, Chapter
6 at distribution function and semi-distributed models, and Chapter 7 at model calibration and uncertainty
in the practical application of models.
2.8 Where to Find More on the History and Variety of Rainfall-Runoff
Models
There are now many more sources of information about hydrological modelling than a decade ago, in
the form of written texts, encyclopaedias, e-books, and Internet resources including sources of free and
