Improved model-based decision support by modeling cotton variability and using evolutionary algorithms

The GOSSYM-COMAX production model has been used successfully by cotton producers in the USA since 1989. Since the release of this model, production techniques have evolved while, in many ways the model has remained static and is difficult to use in the new production environment. In this research the basic biological responses and representations of GOSSYM were enhanced. A new light interception model was developed, integrated, and tested. This new approach provided a foundation for developing a plant population model as an alternative to the current average plant model. The inclusion of plant variability was realized through techniques derived from queuing theory. In addition, an emergence model was developed and added to GOSSYM. Furthermore, the decision support capability of the model was improved. An Evolutionary Algorithm (EA) was proposed as an alternative approach to COMAX. A comparison between irrigation schedules produced by COMAX and the EA indicates that this latter approach produces better schedules, which increases the profitability of the cotton crop. Finally, a new interface based on virtual imaging was developed and integrated to the system. This new interface displays 'virtual plants' representing the crop and its variability. Hopefully, these changes will make it easier for farmers to use and accept the system and its recommendations. (Resume d'auteur)