A glasshouse environment is a complex assemblage of interacting physical and biological systems, many of which are inherently nonlinear dynamic processes, with considerable uncertainty about both their nature and their interconnections.
It is surprising, therefore, that stochastic dynamic models are the exception rather than the rule in research on the modelling of glasshouse systems.
One reason for this anomaly lies in the very successful history of physical science over the last century: modelling in deterministic terms has permeated scientific endeavour over this period and has led to a pattern of scientific investigation which is heavily reductionist in nature.
Such deterministic reductionism appears to be guided by a belief that physical systems, such as a glasshouse, can be described very well, if not exactly, by deterministic mathematical equations based on well known scientific laws, provided only that sufficient detail can be included to describe all the physical processes that are perceived to be important by the scientists involved.
This leads inexorably to large, nonlinear models which reflect the scientist's perception that the environment is an exceedingly complex dynamic system.
A recent paper (Young et al, 1995) rejects this kind of deterministic reductionism as the major approach to environmental modelling, while recognising the value of simulation models and reductionist thinking in the early stages of the overall modelling process.
More importantly, however, it stresses the importance of explicitly acknowledging the basic uncertainty that is essential to any characterisation of the physical and biological processes in environmental modelling, and argues for the greater utilisation of data-based statistical methods.
In particular, it demonstrates how such methods can be used both to expose the dynamic simplicity that underlies the superficial complexity of reductionist simulation models and to identify similar low order models on the basis of experimental data obtained from the real system.
We believe that this kind of Data-Based Mechanistic (DBM) modelling philosophy (Young and Minchin, 1991; Young and Lees, 1992; and Young and Beven, 1994) is particularly relevant to horticultural engineering and is essential if the practical utility of mathematical models in horticultural modelling and control is to be fully realised.
Of course, we are not suggesting that simulation models based on deterministic reductionism do not have an important role to play in such studies; indeed, the practical example used in this paper exploits such a model.
Rather we are emphasising the need for a DBM phase in the analysis which introduces an objective element in the modelling process and avoids unwarranted complexity in the final models that are to be used in the scientific evaluation of the system and control systems design.
With this objective in mind, the present paper first outlines this DBM approach to modelling and then shows how it has been exploited in the modelling and control of the micro-climate of a horticultural glasshouse.