Agriculture Reference
In-Depth Information
the pen trial, in which the 'real' (commercial)
world is modelled by replicated small groups
of birds held in pens or cages that, more or
less, reflect 'normal' conditions. This familiar
procedure is undoubtedly a model used to
predict how the real world will behave in dif-
ferent circumstances and some of its charac-
teristics as a model may be noted:
has been that the model parameters can be
updated very quickly, e.g. to accommodate
genetic improvements. A third impact has
been that poultry nutritionists have been less
driven to probe for an understanding of
lower level functions such as metabolic
regulation, biochemical pathways of nutrient
utilization, and energy expenditures in order
to describe different metabolic relationships;
and further to incorporate such information
into mechanistic models that might enable
more general application.
1.
This model has formed the main basis of
applied poultry science and has clearly
been very successful in providing the tech-
nical platforms on which a very successful
industry has grown. The method is simple,
relatively cheap and easy to understand.
The results are easily communicated.
2.
The limitations of the model are acknow-
ledged in a general way; for example, the
extent to which population size or exposure
to disease may influence the predictive ac-
curacy of the model. But the question of
model validation is not considered in a for-
mal way and receives little attention.
3.
The model is clearly empirical, the results
applying only to the combination of circum-
stances that prevailed in the trial. The repeti-
tion of similar trials over time and in differ-
ent places may be justified by the fact that
some fixed effects in the 'real' world change
over time (e.g. bird genotypes) or to take ac-
count of 'local' factors such as country or
broad environmental classifications.
4.
There is a lot of repetition of similar trials.
This is good in the sense that it increases
confidence in the results, but is bad insofar
as it wastes limited research resources.
These remarks seem very pertinent to the
present discussion.
The applications of mathematical mod-
elling in poultry science have been exten-
sive and varied. Conferences like this one,
and elsewhere, and several reviews are tes-
tament to considerable achievement in many
areas of applied poultry science. And yet if
we ask two questions about the impact of
modelling on poultry science there still
seems to be a lot to do. First we may ask
whether modelling is an integral part of ap-
plied poultry science methodology. Second,
whether modelling is fulfilling its potential
to improve commercial decision making. At
the present time it is suggested that the an-
swer to both of these questions has to be no.
In this overview of poultry modelling
the literature on modelling is reviewed using
these two questions as a background agenda.
Types of Poultry Models
France and Thornley (1984) have suggested
a useful
2
×
2
×
2
classification of models:
empirical or mechanistic, deterministic or
stochastic, dynamic or static. While these
are important qualities of the models of inter-
est, a different grouping is used in the pres-
ent discussion:
Commenting on the effect of applied trials
on poultry modelling, Chwalibog and Bald-
win (1995) wrote as follows:
In comparison with nutritionists working
with other domesticated species, poultry
nutritionists have exploited the specific
advantage of being able to run experiments
with large numbers of animals economically.
It is common to find experiments evaluating
responses to numerous treatments which are
well replicated and in which up to
10,000
birds were used. This advantage has had
several major impacts on the practice of
poultry nutrition. One of these has been that
most models developed for feeding systems
are based on response data. A second impact
1.
Models of scientific theories.
2.
Models to extend and increase the value
of pen trials.
3.
Growth curves.
4.
Empirical models of poultry production
systems.
5.
Mechanistic models of poultry produc-
tion systems.
6.
Real-time control models.
