Geology Reference
In-Depth Information
(a)
250
200
150
100
50
0
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
Year
(b)
12
10
8
6
4
2
0
1992
1994
1993
1995
1996
1997
Year
Fig. 4.7
Observed and predicted annual soil loss from the (a) Holly Springs plots, US, and (b) the Woburn Erosion
Reference Experimental plots, UK (kg m
−1
). Observed data are shown as diamonds, with error bars describing range
of replicate values, dotted lines describe minimum and maximum predictive uncertainty, and the solid line
describes the median of model predictions (after Brazier
et al
., 2000).
consideration of uncertainty, otherwise the cred-
ibility of erosion modelling will suffer.
parameter values relative to estimates derived
directly from field measurements; and, of course,
the representation of erosion, transport and depo-
sition processes in the model.
If erosion models fail in this way, it is worth
emphasizing that this is an opportunity to recon-
sider modelling strategies to make them more
realistic as hypotheses about how the sediment
system functions (e.g. Beven, 2002a,b, 2008, 2009).
What then needs to be reconsidered? Are the
input data used to drive the model commensu-
rate or accurate enough? Are the evaluation
observations commensurate or accurate enough?
Do the nature of the failures give an indication of
deficiencies in the model process representa-
tions? Are model structures developed under a
limited range of applications being applied too
widely elsewhere? These types of questions
4.7 The Future
Whenever erosion models are compared with
observations, it is generally found that even when
allowing for uncertainties in model parameters,
there are often observations that cannot be
matched by the model predictions. This suggests
that modelling results are subject to important
epistemic uncertainties that have not been taken
into account. The most important uncertainties
in this respect might be: commensurability and
bias uncertainties in model inputs; commensura-
bility in the observable variables relative to what
is being predicted; commensurability in effective
