Agriculture Reference
In-Depth Information
CO
2
emissions
Fixed N and C
Harvest
Respired
CO
2
Feed sold
CO
2
emissions
Storage
Crops
CO
2
,N
2
O and
NH
3
emissions
Grazing
Soil
Deposition
Purchased feed,
animals and
bedding
Animals
Purchased
fertilizer
Manure
Runoff and
leaching loss
of N, P and C
CO
2
and CH
4
emissions
Milk and animals
sold
Exported manure
CO
2
, CH
4
,N
2
O and
NH
3
emissions
Fig. 10.1.
The Integrated Farm System Model simulates all major farm processes, tracking the flow of
resources and nutrients to quantify the performance, environmental impact and economics of the
production system.
Model Evaluation
amounts of diverse data to quantify all major
aspects of the farm. Such data are not available
due to the labour required and the cost for meas-
urement and recording at this scale.
Although a formal validation of a farm
model is not possible, model evaluation is still
important and necessary (Oreskes, 1998; Web-
ster and McKay, 2003). For farm-scale models,
there are three levels of evaluation needed. They
are model verification, component evaluation
and farm-level evaluation. By satisfactorily com-
pleting each of these phases of evaluation, the
model developer and the user can develop confi-
dence in the ability of the model to represent
actual farm production systems. When the
model adequately represents the actual system,
sensitivity analysis provides valuable feedback
on the environmental and production effects of
changes to the management system. Similarly,
model uncertainty provides a measure of risk to
the environmental and production aspects of
the farm as a result of unknown or uncontrolla-
ble changes in the natural system.
Farm models, like all models that answer ques-
tions about the natural world, are developed to
represent the physical, biological, chemical and/
or financial processes of a complex, unbounded
and highly dynamic system. Evaluating how
well the model represents the real system is
always an important aspect of model develop-
ment and application. This evaluation process is
often referred to as model validation. However,
the term 'validation' must then be defined as
the ability of the model to perform satis-
factorily within its intended domain and use
(e.g. Refsgaard, 2000). Such a definition does
not lend itself to formal, statistical comparisons
of model predicted and measured data or to
a consistent measure for comparison across
system representations or study sites. The
many parameters and interacting processes
along with the biological uncertainty of those
processes make this type of validation infeasible.
Additionally, statistical validation requires large
