Modelling plant growth : importance of identifiability in developing simple models for control

(eng) The development of models for plant growth typically serves one of several purposes: (1) to improve our understanding of plant growth, (2) for yield prediction, and (3) for decision making in agriculture or related fields. Most state of the art models of plant growth have been developed for the first of these applications, and are therefore typically quite complex, aiming to synthesize our current knowledge of physiology. However, the development of simpler models for decision making in agriculture, or more specifically for process control, is crucial in order to improve the current operation of greenhouses, or for the implementation of more complex plant growth applications – such as life support in space. The MELiSSA project (Micro-Ecological Life Support System Alternative) is a European Space Agency project which aims to develop a regenerative life support system for long term manned space missions. As part of this initiative, a higher plant compartment will be designed to produce food for the crew, regenerate the atmosphere and contribute to the recycling of some wastes. Dynamic plant models of plant growth are needed for the prediction of the life support contributions and the enhanced control of the plant growth chamber environment. Within this context, this thesis provides a comprehensive discussion of the development of reliable dynamic models of plant growth, considering topics ranging from experiment design to model development and testing. We took an approach which focused on capturing the main dynamics of plant growth (i.e. the time evolution of important variables) while maintaining model simplicity and the identifiability of model parameters. We concentrated on these model attributes because they are important in prediction and control applications and are rarely valued in plant growth modelling work. This approach was applied for the development of models of red beet, lettuce and Arabidopsis thaliana, and represents a significant step forward in the development of plant models for prediction and control applications.