Agriculture Reference
In-Depth Information
include the ensiling of feeds (Buckmaster
et al
.,
1989), suitable days for field operations (Rotz
and Harrigan, 2005) and animal production
(Rotz
et al
., 2005). This level of evaluation may
include statistical comparisons, but often these
are more qualitative comparisons of how well
model predictions represent the actual farm
components. The constraint to performing a
more formal validation is normally the lack of
well-defined and comprehensive measured data.
Sometimes model evaluation involves
newly developed and well-defined relation-
ships for incorporation into the larger model.
Examples from IFSM include hay drying and
ammonia volatilization. To fulfil a need for simu-
lating the field curing of hay, an empirical model
based upon scientific understanding was devel-
oped to predict drying rate as a function of soil
and weather conditions (Rotz and Chen, 1985).
The large amount of field-drying data collected
allowed a statistical comparison of measured
and predicted drying rates, and hay moisture
contents through time. After refining a set of
theoretical relationships for predicting ammo-
nia volatilization, the new model was shown to
predict measured ammonia emission rates more
accurately then other previous models (Montes
et al
., 2009).
Model verification
Model evaluation must always begin with verifi-
cation. Verification is the process of confirming
that the model is functioning properly, i.e. that
there are no errors in the implementation of the
model. A model is verified by stepping through
each line or section of code within a model to
make sure that it is mathematically and logically
correct. Many minor and easily corrected coding
errors, that would otherwise go unnoticed, can
be uncovered with this close evaluation. If left
uncorrected, these errors can greatly affect
model performance and the use of computer pro-
cessing and memory resources. With most soft-
ware development platforms used today, this type
of verification is easily done using debugging
options. This level of verification requires consid-
erable time though, and it may be overlooked in
the desire to complete the model. This verifica-
tion process is critical. If the model is not func-
tioning properly at this level, other forms of
model evaluation and application are irrelevant.
Component evaluation
The next step in model evaluation is the evalua-
tion of individual model components. A complex
model such as that representing a farm system
consists of many components. These compo-
nents can include complex system models
themselves representing crop growth and devel-
opment, crop harvest and storage, animal feed
intake and performance, and manure-handling
systems. As component models such as these are
developed and integrated into the larger system
model, they must be shown to represent their
component properly, both independently and
within the full model. For example, as existing
crop models were added to the IFSM, they were
evaluated to determine how well they repre-
sented crop production on farms. Although
these crop models were developed and evaluated
by others, further evaluation was done to com-
pare predicted crop yields with reported yields
from farms over several years of weather to
determine not only how well the model predicted
farm yields but also the variation in yield from
year to year (Rotz
et al
., 2002a). Other examples
of this level of evaluation of IFSM components
Farm-scale evaluation
Evaluation at the whole farm level is necessary
to assure proper representation of the real-world
farm system. For this level of evaluation, more
general data, often readily obtained from oper-
ating farms, are most appropriate for compari-
son. Such data include crop yields, feeds
produced, feeds bought and sold to maintain the
herd and the animal products produced and
sold from the farm. If a model has been properly
verified and evaluated at the component level,
an agreement of predicted and actual data at
the farm scale can further support an accurate
representation of the farm system. Farm-scale
evaluation has been used to support the use of
the IFSM for evaluating P accumulation on
dairy farms in a sensitive watershed in New York
(Rotz
et al
., 2002b), organic dairy production in
Pennsylvania (Rotz
et al
., 2007) and confine-
ment fed and grazing based dairy production in
Georgia (Belflower
et al
., 2012).
