Soil Erosion and Conservation - Environmental Modelling: Finding Simplicity in Complexity - page 354

of detailed sensitivity analyses and calibration of the
model to the large database of natural runoff-plot
information used to develop the USLE and RUSLE.
Without the tie between model and database, and
without knowledge of the sensitive input variables so as
to know where to focus efforts, turning a complex model
such as WEPP into a useful conservation tool would not
be possible. Thus, in a sense, even though WEPP routines
are process-based descriptors of various components
of the erosional system, ultimately the model must be
empirically based on the same type of data as was used
to develop the USLE and RUSLE, along with additional
experimental data collected specifically for WEPP.

annual values of erosion, however, it provides estimates of
off-slope sediment delivery in addition to estimates of on-
slope soil loss. The RUSLE can also provide estimates of
certain auxiliary systemvariables, such as residue amounts
and crop yields. The WEPP model provides a massive
array of system information to the user, if such infor-
mation is desired. The model predicts both on-site soil
loss and off-site sediment delivery, including ephemeral
gully erosion, which neither USLE nor RUSLE attempts
to predict. Sediment-delivery information includes not
just the amount of sediment yield, but the particle-size
distribution information for that sediment, which can be
important in terms of chemical transport by sediment.
The WEPP also provides a detailed description of the
spatial and temporal distributions of soil loss, deposition,
and sediment yields, both along the hillslopes and across
the watershed. Auxiliary system information fromWEPP
is enormous, and is available on a daily basis. Information
includes soil-water content with depth, surface residue
amounts and coverage in both rill and interrill areas sep-
arately, buried residue and root masses, canopy cover
and leaf area index, evapo-transpiration rates, soil surface
roughness, soil bulk density, changes inhydraulic conduc-
tivities of the soil surface layer, changes in soil erodibility
with consolidation and surface sealing, crop biomass and
yields, subsurface interflow of water, tile drainage, and
surface runoff amounts and peak rates, among others.
The USLE, RUSLE and WEPP (or other process-based
models) constitute a complementary suite of models to
be chosen to meet the specific user need. To illustrate
this idea, we will take a look at recent applications of the
USLE and WEPP to address the question of the potential
impact of climate change on erosion rates in the United
States. As we will see, we are able to use the USLE to
provide certain information that WEPP simply cannot
provide because of the restrictions of model complexity,
and we are able to use the WEPP model in way where
only the complex model interactions will provide us the
information we want regarding system response.
In the first study we used the RUSLE R-factor to esti-
mate the potential changes during the next century for
rainfall erosivity across the whole of the United States,
southern Canada, and northern Mexico. In this case, we
do not want to become embroiled in the subtle differences
between effects of various soils, slopes, cropping systems,
and other system variables. Instead, we are looking for the
primary effects over regions. With the USLE and RUSLE
we can do this, because RUSLE uses an R-factor that was
derived from a wide array of plot conditions, and it is not
interdependent with the other system variables. Statistical

22.3 The contributions of modelling

The accuracy of the three models introduced above has
been tested using measured soil loss data form plots. We
mentioned above the study by Risse et al . (1993) using
the USLE and 1700 plot-years of data, and the study of

Zhang et al . (1996) of the WEPP model using data from
4124 storm events. The data of Risse et al . (1993) was
also applied to the RULSE model with very similar levels
of accuracy as obtained with the USLE (Rapp, 1994).
These three models all produced essentially equivalent
levels of accuracy for prediction of soil loss, and the
level was somewhat less than the level of fit obtained
with the 'best-case' replicate plot-model discussed above.
The results suggest that we have approached with these
models the maximum level of possible soil-loss accuracy
for ungauged, uncalibrated sites.
This result does not imply, however, that the threemod-
els are equivalent in usage. RUSLE has certain advantages
over the USLE because its database and internal relation-
ships have been expanded beyond that of the USLE for
particular applications such as rangelands in the west-
ern United States and no-till cropped lands in the eastern
United States. The data comparisons reported in the three
studies above included no rangeland data and very little
no-till data, so these advantages were not apparent from
those studies. The USLE may have advantages in other
applications. In areas where data are few, or computations
need to be kept simple, the USLE has distinct advantages
over both RUSLE and WEPP.
Another category of differences between the models is
the type of information provided, rather than the accuracy
of the information. The USLE provides essentially only
average annual soil loss over the area of the field expe-
riencing net loss. The RUSLE also provides only average

+