Geology Reference
In-Depth Information
4
Dealing with Uncertainty
in Erosion Model Predictions
K.J. BEVEN
1
AND R.E. BRAZIER
2
1
Lancaster Environment Centre, University of Lancaster, Lancaster, UK; GeoCentrum,
Uppsala University, Uppsala, Sweden; ECHO/ISTE, EPFL, Lausanne, Switzerland
2
School of Geography, University of Exeter, Exeter, UK
4.1
Why Worry About Uncertainty
in Erosion Models?
power now available, and the way in which it has
made spatial datasets and geographical informa-
tion systems more accessible. This tendency had
earlier been seen in the development of distrib-
uted hydrological models, for example the many
descendants of the Blueprint described in the
seminal paper of Freeze and Harlan (1969); indeed,
one of the reasons used in justifying the develop-
ment of distributed models was that they would
provide spatial predictions of runoff amounts and
velocities that could be used in the prediction of
sediment transport and other water quality vari-
ables (e.g. Freeze, 1978; Beven & O'Connell, 1982;
Beven, 1985; Refsgaard & Abbott, 1996). Examples
of such process-based models include WEPP
(Nearing
et al
., 2005) and EUROSEM (Morgan
et al
., 1993; Quinton, 1994). Soil erosion compo-
nents also appear in other distributed models
such as SHETRAN (Ewen
et al
., 2000; Lukey
et al
., 2000).
Any distributed process-based model of soil
erosion and sediment transport is necessarily
dependent upon some underlying distributed
hydrological model. But the limitations of dis-
tributed hydrological models became recognised
soon after they started to become more widely
available (e.g. Beven, 1989, 1996b; Grayson
et al
.,
1992). There are, indeed, important philosophical
issues about the possibility of developing general
process-based models of environmental systems,
as discussed in Beven (2001b, 2002a, 2009). The
fact is that it is very difficult to be secure about
the equations that are used to describe complex
It is probably still the case that the most widely
used erosion model in practice today is the
Universal Soil Loss Equation (USLE; Wischmeier &
Smith, 1960, 1978) and its revisions and variants,
either as a stand-alone predictor or as a compo-
nent of other models (e.g. the SWAT model;
Arnold
et al
., 1998; Gassman
et al
., 2007). The
USLE is an empirical relationship, originally
derived from a limited dataset from erosion plot
studies at 49 sites in the US, but later applied all
over the world (Nearing
et al
., 2005). It has this in
common with the Soil Conservation Service
(SCS) runoff generation model of Mockus (1949)
with which it has often been coupled (see Beven,
2001a, for further discussion of this model). In
scientific terms, these equations make no real
attempt to represent the
processes
involved in
runoff generation or sediment mobilization,
transport and deposition, and consequently are
intellectually somewhat dissatisfying even if we
could be secure in their predictions - but they
also can be dissatisfying in that respect as well, as
will be seen below.
There has therefore been a tendency to develop
process-based erosion models, a tendency that has
been reinforced by the much greater computer
