Geology Reference
In-Depth Information
A topography-based relative index of erosion and
deposition explained between 32% and 56% of
the variation in the soil Cu inventories. Finally,
Zhang
et al
. (2008) used rare earth elements suc-
cessfully as tracers to compare the detachment
predicted by WEPP with the observed rill erosion
on a series of plots.
(v)
Remote sensing images
. An obvious source of
spatial erosion information is remote sensing
imagery. In a review, Vrieling (2006) discussed the
application of remote sensing to derive informa-
tion on erosion and erosion-controlling factors.
Direct information on erosion features (mostly
rill systems and gullies) is only possible with
high-resolution images (such as IKONOS,
Quickbird and Worldview with resolutions of
0.2-1 m). It is possible to obtain information on
controlling factors such as land cover, topogra-
phy, soils and tillage practices, using various sen-
sors. Using images for vegetation and crop-derived
variables is fairly standard for deriving NDVI type
indices. Multispectral or hyperspectral informa-
tion is needed to derive soil properties (Shepherd &
Walsh, 2002). LIDAR is increasingly used for
altitude mapping and provides very detailed infor-
mation on topography, which also enables gully
mapping to be carried out (James
et al
., 2007).
Vrieling
et al
. (2008) employed a multiscale
approach on a 100 km
2
grazing area in Brazil. They
used low-resolution (250 m) high-frequency
MODIS images to establish the annual period of
likely erosion risk based on absence of land cover.
A better estimate for that period was obtained
using several ASTER images with 10 m resolu-
tion, and finally erosion features were detected
using very high-resolution (0.6 m) Quickbird
imagery.
in spite of the fact that almost all models are
spatially distributed, the success of a model is
mostly expressed as the ability to predict dis-
charge and soil loss lumped in space and time,
with prediction of hydrographs as a second
method. Occasionally simulated patterns are
compared with observed patterns of, for instance,
137
Cs or soil loss per field in a catchment, but
annual or seasonal totals are clear favourites. The
real value of spatial modelling appears to be the
use of the spatial output for purposes such as pre-
dicting the effect of land-use changes or conserva-
tion measures (Walling
et al
., 2003, Jetten
et al
.,
2003), but the spatial output is rarely used for
verification, except for the few studies mentioned
above. Also interesting to note is the lack of vali-
dation results; while calibration fortunately is
often done, at least in scientific literature, inde-
pendent validation is far less common. Calibration
is usually done when a model is used in a new
area or for a new set of circumstances (e.g. effects
of forest fire), for which adaptations to the model
are needed. The authors then want to prove that
their adaptations make the model perform better.
This makes generalizations in model perform-
ance difficult because the same model is used in
different variations, and also hybrids between
models are created.
The more complex the area that is modelled,
the more the uncertainty increases. One of the
main influences on the predictive quality of ero-
sion models seems to be the changing spatial pat-
terns of sources and sinks, and their connectivity.
Generally the larger the area, the more sources
and sinks are included. A study of soil loss from
small plots deals mainly with sources of sedi-
ment and does not have to consider connectivity
issues, whereas a catchment study will have to
include information on source areas such as
crusted fields and sinks, valley floors and obstruc-
tions, as well as their connectivity. The soil loss
from the catchment measured at the outlet is
often only a small part of the amount of detached
sediment. This may cause large uncertainties in
the predictions, which is demonstrated by vari-
ous studies using Monte Carlo type simulations.
It may be, however, that by regarding only lumped
3.6 Conclusions
This overview of calibration of models on differ-
ent spatial and temporal scales shows that,
although the objectives are different for each
scale, the calibration procedures used are very
similar. First of all, calibration is almost always
done at all scales, which is encouraging. However,
