Geology Reference
In-Depth Information
we would not be able to fully state the reasons
why the model did not provide behavioural simu-
lations over these two events. Such studies need
to be encouraged to benchmark the predictive
capability of our models and improve and refine
their conceptualization of soil erosion processes.
The simulations reported in this chapter took
approximately 5.5 days to run on a high perform-
ance cluster. This process was repeated a number
of times as we learnt from our initial attempts.
Overall the simulations took in the order of five
months to complete on a part-time basis. Such an
overhead is unthinkable outside of a research
arena. However, computers are getting faster - a
study such as this would have been unfeasible
when Quinton carried out his initial uncertainty
evaluation of EUROSEM in 1997 - and we can
expect the time needed for such simulations to
fall. Thus model
developers
should be encouraged
to quantify erosion model performance and be
explicit about the quality of model predictions
against observed data. However, for the land man-
ager or model
user
, deploying the methods we
describe here is not viable. However, this is not to
say that model users should not try to incorporate
uncertainty into their model application. We sug-
gest that, at a minimum, model users should:
●
assess the uncertainty in their measurements
and express them in comparison with model
output;
●
be guided by reports of model sensitivity to
vary as large a number of parameters as can be
afforded in order to generate maximum, mini-
mum and most likely model predictions;
●
decide in advance what makes an acceptable
prediction, and reject and refine the model if it
does not meet these conditions;
●
avoid using models to underpin decision-making
without consideration of uncertainty where at all
possible.
of its parameters and insensitive to many others.
The use of Generalised Likelihood Uncertainty
Estimation (GLUE) illustrated the uncertainty in
model output. It was important to note that meas-
urements of observed data are also not error-free,
and that while we were able to incorporate such
error into the assessment of the hydrological per-
formance, such information was not available for
the sediment dynamics. In general EUROSEM
was able to simulate the hydrology, but the two
events studied required different parameter sets.
Sediment concentrations were simulated an order
of magnitude higher than observed. We suggest
that this may be because EUROSEM does not
simulate well the erosion processes occurring in
low-energy rainfall and overland flow, and that
process descriptions within the model may need
to be revisited. Given the amount of computer
time required for this work, we suggest that this
method may not always be appropriate for land
managers or model users, but we do encourage
them to include model and data uncertainty esti-
mation when evaluating models and to consider
model uncertainty when making decisions about
land management. Although two billion simula-
tions were carried out for this study, considerably
fewer simulations would still provide an effective
framework for evaluating uncertainty when mod-
elling many land management scenarios.
Acknowledgements
This research was part-funded by the UK
Department for Environment, Food and Rural
Affairs (Defra) with the project number PE0120,
the UK Natural Environment Research Council
(NERC), Flood Risk from Extreme Events (FREE)
programme with the grant number NE/E002242/1,
and the UK Research Councils, Rural Economy
and Land Use (RELU) programme with the grant
number RES-229-25-0009-A.
5.6 Conclusion
References
This chapter described an application of EUROSEM
to a small catchment in Devon, UK. The chapter
demonstrates that EUROSEM is sensitive to some
Avery, B.W. (1980) A
Soil Classification for England
and Wales
. Rothamsted Experimental Station,
Harpenden, UK.
