Geology Reference
In-Depth Information
these errors remain, and are still virtually unquan-
tifiable (Morgan, 2005). Attempting to predict a
value which is subject to an unknown amount of
error is like aiming at a moving target. However,
model users need to take account of the fact that
it is unrealistic to expect a model to predict a value
with greater certainty than the likely variability
around the measured value (Nearing, 2000, 2006).
A third issue related to uncertainty arises
directly from the different approaches to valida-
tion. Just because the levels of uncertainty
associated with a model are known from a generic
evaluation, it does not mean that the same levels
will be achieved when the model is applied to a
specific locality. Most erosion models are highly
sensitive to variations in rainfall, a parameter that
can be measured reasonably accurately. A generic
analysis of uncertainty is likely to incorporate a
wide range in rainfall, with annual totals between
<100 and >3000 mm or storm totals between
1-3 mm and >400 mm. Within a given field or
small catchment, the rainfall is likely to vary by
much smaller amounts and uncertainties in
prediction associated with rainfall will accordingly
be small. In contrast, parameters that may con-
tribute moderate to small amounts to uncertain-
ties in generic predictions may be extremely
important for predictions over small areas.
Arguably the main problem with uncertainty
analysis at present is that an enormous number
of computer simulations need to be carried out
which, although they may reduce the level of
variability around a prediction, may not make
the prediction a closer fit to the observed value
(Quinton, 1997). It seems unlikely that a user
will undertake uncertainty analysis, but see
Chapter 5 for an example of what can be
achieved. Whether or not they carry out the
analysis, users need to be aware of the level of
uncertainty associated with predictions from a
chosen model. It is therefore incumbent on
model developers that they perform uncertainty
analysis as part of the information they provide
about their model in User Guides. Chapter 4
provides a more detailed discussion of the vari-
ous types of uncertainty and the approaches that
can be used to deal with them.
2.9 Some Practicalities
By concentrating on the problems associated with
developing and validating models, the impression
may be gained that models are either of limited
value or simply too difficult to use. This, how-
ever, is far from the case. Erosion is one of the
leading environmental problems worldwide, and
controlling it requires a range of activities from
the development of policies at national and inter-
national level to the design and implementation
of conservation measures at the field, road bank
or construction site level. Erosion control is rele-
vant not just to the fate of sediment removed
from the landscape, but also to reducing the asso-
ciated risks of pollution from chemicals adsorbed
to the sediment and increased flooding brought
about by the reduction in the capacity of river
channels and reservoirs as a result of sediment
deposition. Policies, strategies and the design of
soil protection works must all be based on sound
scientific data. It takes far too long and is too
expensive to obtain the necessary data from field
measurement, so models are vital for filling that
gap, provided that they too are based on good sci-
ence and good data.
Enormous advances have been made in the
development of models over the last 20 or more
years, and there is now a large number from
which to choose. Since these models range in
complexity from simple black-box types to
detailed process-based ones and operate at a whole
range of temporal and spatial scales, it is impor-
tant that the user defines carefully what is needed
to address a particular problem. The model user
needs to think along the same lines as the model
developer. He or she must start by deciding what
data are needed to deal with the problem, over
what temporal and spatial scales and covering
what processes. Thus it helps if the user develops
a conceptual framework of what is required. Since
this framework is essentially a conceptual model,
this initial step should make it easier to research
the available models and find ones with a similar
conceptual base. It is important at this stage that
the user is not misled into believing that detailed
process-based models that yield numerical
