Geology Reference
In-Depth Information
consideration of the spatial pattern of erosion,
may look impressive (see Peeters
et al
., 2008 for a
recent example), but it does not tell us anything
about how well the model represents reality - far
from it, such modelling exercises mislead the
reader into thinking that the model is better than
it actually is, which does not improve the repre-
sentation of scale within erosion models, or
indeed improve the predictions that are made by
such models. What is needed is an evaluation of
the pattern of erosion as well as the total erosion
that has occurred. We will not be able to evaluate
the ability of erosion models to predict across
scales without datasets that describe the change
in erosion rates across each scale of interest. Some
progress has been made in this respect using nest-
ed-catchment studies (Van Oost
et al
., 2005;
Deasy, 2007); however, very few datasets exist
that permit the evaluation of erosion predictions
for the right spatial reasons.
Consequently, a sound understanding of all the
processes operating in the system is necessary.
Modelling tools are required that provide distrib-
uted output to evaluate the prediction of sediment
sources and sinks within the catchment system,
and furthermore, datasets are also required that
describe the spatial variability of soil erosion
fluxes within a catchment (or plot) in order to
evaluate these models. If we continue simply to
collect data at plot or catchment outlets and
report these data as sediment yields per unit area,
thus not describing the spatial pattern of erosion
with our area of interest, we have little chance of
making erosion predictions for the correct process
(and form) reasons.
Specific issues arise when we seek to take our
understanding of soil erosion, and representation
of processes involved in soil erosion, and apply
them over larger spatial and temporal scales to
answer different questions about the environ-
ment than the models were originally developed
to answer. As most 'questions' posed by soil-
erosion modellers are represented by applications
of models to sites on which they were not origi-
nally parameterized/calibrated (and often sites
that are larger in scale than the datasets that
underpin the model development in the first
place), a common problem is that models are
over-extended or simply applied at scales which
render their output meaningless.
Finally, the representation of changing domi-
nant processes with scale has led some authors
to suggest that multiple scales of erosion model
are required (Kirkby
et al
., 1996; Poesen
et al
.,
1996). Theoretically, coupling plot- or hillslope-
scale models that represent detailed processes
with larger scale models that can predict soil
erosion over appropriately large areas for man-
agement decisions, should yield useful and scale-
sensitive predictions. However, implementation
of such an approach has proven to be very diffi-
cult, as
a priori
knowledge of which processes
dominate at any given modelling scale is not
available (although see the analysis using
discriminant functions by Howard, 1995). More-
over, observed data held at different scales very
rarely exist with which to evaluate such multi-
scale models, although some exceptions are
starting to enter the literature (Parsons
et al
.,
2006a; Deasy
et al
., 2007) and may be useful for
model evaluation in the future, as exemplified by
Wainwright
et al
. (2008c).
6.8 Conclusion
This chapter has highlighted a number of chal-
lenges that environmental modellers face in mak-
ing scaleable predictions of soil erosion. Although
a large number of erosion models exists, very few
have been explicitly developed to make robust
predictions across the range of scales (plot, hills-
lope, catchment and basin) that will make results
useful to stakeholders who are interested in
reducing erosion or mitigating soil loss. In part,
this situation has arisen due to the difficulties of
observing erosion at multiple scales and therefore
understanding how dominant processes change
with scale, in order to represent changing process
dominance within model structures. There has
been an emphasis on using what empirical data
we have collected (which is certainly not insig-
nificant) to underpin the development of increas-
ingly complex numerical models of erosion. This
