What's the use of crop growth models ?
This is a question that is whirling around my head
since Dr. Matthew's started the discussion on
development impacts of agricultural models.
According to my understanding we can devide
crop growth models into three sub-categories:
(1) Empirical models which try to establish statistical
links between crop environment and growth. Though
analyzing cause and effects in plant growth they are
dealing by no means with plant internal and external
mechanisms which constitute its "real" behaviour.
I would therefore like to call it a PHENOMENALISTIC
approach.
(2) Mechanistic models try to overcome the limitations of
empiricism by establishing PROCESS oriented links between
environmental signals/resources and crop growth. This is
a DETERMINISTIC approach.
(3) If understanding it right, neural networks, fuzzy logic or
genetic algorithms are "SELF-LEARNING" models which are
based on selective analyzation of dynamically changing
experience.
Agricultural research in this early century was marked by
empircal studies analyzing the impact of management
practices on crop growth. A good example are the numerous
studies on nitrogen effects on yield which were stimulated by
Liebig's discovery of the "minimum law". It was possible to
undertake these studies using simple field observations
using a pencil, a few sheets of paper and statistical tables.
However, I get more and more the feeling that the time of this
PHENOMENALISTIC approach is over.
Using advanced electronic technology we are now able to
attach all kinds of sensors to the plant which can give us
a much better insight into a plant's "real" behaviour when
being properly used (... by highly skilled technicians): It is
possible to measure ion fluxes controlling stomatal opening,
hormones (think of the concentration !) controlling growth,
short-term changes in xylem nutrition transport and airodynamic
pressure forces acting on the leaf surface using fractional
second intervals, to name only a few examples.
I am very certain that this advanced technology will support
the construction of DETERMINISTIC, process oriented models.
Models are good tools to test our current hypotheses about
a "real" plants behaviour. Statistical methods are hopefully
used to test these hypothesises in field experiments and it
is in this sence that empiricism has its strong legitimacy.
However, when being used to "tweak" these models to
FIT certain environmental conditions, empiricism looses
its scientific soundness. It is in this respect, that I am opposing
the current concept of the CERES component in DSSAT. To
verify this conclusion you may take a look at the following
examples : PTF, RGFILL, RNFAC, WX, EO, EOS, RWU, CNRF,
RNAC etc.
I am aware, that a model FITTED to certain environmental conditions
still needs sound input data (Miguel Calmon from University of
Florida) and I have acknowledged that allready in my first letter.
However, this data will be never perfectably available unless
derived under exactly the same conditions for which the model
was developed. This implies that due its PHENOMENALISTIC nature
CERES type models are none universal applicable. It is therefore
dangerous, to quote Landau (1998) again, to base policy or decision
making on these models. In my eyes this is the most important
quintesence of his paper and not the fact that their input data were
(PARTLY !) derived using a "good" guess (Dr. Jamison) , which will
almost need to me made when applying models to practical conditions.
The fact that especially CERES, but also AFRCWHEAT2 and SIRIUS
failed to explain variation in predicted and observed yield verifies
that these models are only locally applicable.
I don't understand, why yield should be an insufficient measure to
test the success of a model (Dr. Jamieson). Comming back to
Dr. Matthew's question on the impact of ag-models on development
it is JUST THAT parameter which interests the farmer practicing
agriculture in harsh and highly fluctuating environments in the less
developed world most. He wants to know HOW MUCH yield he gets if
he changes his current practice, which was developed "using
selective analyzation of dynamically changing experience" (see above)
for hundreds of years (which makes him reluctant to change his
practice to fast). The success of a model can therefore be only
determined by the accuracy it predicts economic yield and how it
affects the liveleyhood of a farmer. I totally agree with Rodney
Beard's opinion that crop models must be coupled with socio-economic
models. However, this would mean that we not only need to understand
plants, but human behaviour in a world-wide trend to open market
reforms as well. This is an extremely COMPLEX approach, which is
certainly needed, but not realizable so far, as I see it.
Summarizing what has been said above I think that crop models, when
largely based on PHENOMENALISTIC observation, are USELESS in
international development. In fact it is dangerous to apply them,
because people in the less developed world don't have a second
chance if they fail to get any positive impacts from them. However,
to conclude from this that modelling research should be not funded
anymore is complete nonsense in my eyes. We need ag-models out of two
reasons:
(1) To test our current hypothesises on "real" plant growth and
thereby to increase our knowledge.
(2) To provide DECISION SUPPORT SYSTEMS for a change in agricultural
technology (I hesitate to use the word "transfer", because it
arrogantly implies that knowledge is only produced in the first
world, which needs to be transferred to the third world in a one
way single minded fashion using the motto: "Eat it or die !" - There
will be always feedback. Disregarding this fact will lead almost
certainly to a complete failure of a developing project).
The 40.000 children which die of hunger each day drastically forces
us to develop decision tools as fast as we can to avoid these deaths.
They are powerfull tools to assist policy makers and economy to
take certain actions to achieve this goal. Modelling work is
therefore not only a matter of scientific interest, but also of
social engagement.
A cross breed of mechanstic and heuristic approaches may be a good
way to achieve this goal.
Matthias Langensiepen
--------------------------------------------------------------------------
Dr. Matthias Langensiepen
Environmental and Agricultural Engineer
Department of Crop Science
Faculty of Agriculture
University of Kiel
Olshausenstrasse 40
24118 Kiel
Germany
Tel. +49-431-880-4398
FAX +49-431-880-1396
EMAIL [log in to unmask]
--------------------------------------------------------------------------
|
