Model-based Climate Management during Winter Period in Mediterranean Greenhouses G. Dimokas and C. Kittas University of Thessaly, School of Agricultural Sciences, Department of Agriculture, Crop Production and Rural Environment, Fytokou St., N. Ionia, GR-38446, Magnesia, Greece, M. Tchamitchian Unité de Bioclimatologie, I.N.R.A., Domaine Saint-Paul, Site Agroparc, Avignon Cedex 9, France, Air Energy Balance Climate Model Mass Balance Cover Energy Balance Crop Energy Balance Soil Energy Balance Heating System Balance Discusion The model will be designed to describe growth of an indeterminate tomato variety under the outside climate conditions. Timing and quantity of tomato fruit yield will be affected by climate conditions in the greenhouse controlled by heating and windows opening. The model will describe the effects of these management measures through their influence on total dry matter accumulation and distribution. Inputs of the Climate Model Air Temperature Relative Humidity Solar Radiation Wind Speed “Central Engine” Optimization Criteria Optimised Control Strategy The rate of organ apparition Energetic cost of Botrytis Management Energetic cost of the strategy Dry matter increase Model Development The combined model will include the physical and the biological model and it will simulate the interacting evolution of the crop and of the greenhouse environment, in response to control and outside weather. The combined model will include a simulation process manager (“central engine”) where the physical and the biological model will communicate. The physical model will be able to run not on an autonomous way and bring to the “central engine” of the combined model integrated values for the air temperature, the relative and absolute humidity, the solar radiation, the condensation rate at the crop and the cover inside the greenhouse in an hourly loop. The biological model will be fed by the hourly climate integrated values necessary for the dry matter production as a The “central engine” will control the growth rate of the plant organs, by increasing or decreasing the inside temperature. Manageable Options of the Combined Model According to the goal that have been arranged at the start of the model description, the model will be able to manage different actions as the growth rates of the crop, the intensity of the heating or the ventilation and the adding energy that will be necessary to avoid the infection of Botrytis cinerea. To address this problem, a reference control strategy will be used to generate reference growth and development and the cost values of the heating and avoid the infection. These actions according a grower’s criterion will be introduced to the “central engine” with a different aim each time. Introduction Models are powerful tools to test hypotheses, to synthesize knowledge, to describe and understand complex systems and to compare different scenarios. Models may be used in decision support systems, greenhouse climate control and prediction and planning of production (Marcelis et. al., 1998). Horticulture needs crop models for a large range of applications, including yield forecast, policy analysis and management (Gary et al., 1998). However, only few models simulate both greenhouse climate and crop growth (Gijzen et. al., 1998). In practice, in the Mediterranean region, two main temperature set points are used: one for the night and one for the day. The later can sometimes be automatically modulated according the outside conditions like radiation. Although only two set points are used, the greenhouse climate does not remain constant, especially during daytime, due to the evolution of outside conditions. Furthermore, it has been shown that crops can withstand deviations from optimal conditions for a while and recover later, provided that these episodes do not last more than a few days (Heuvelink, 1989). Physical Model Outputs of the Climate Model. Integration of the 4 sub-model algorithms will lead the user of the “central engine” to more manageable values like air temperature, air humidity, crop and cover condensation rate and crop transpiration. Air humidity will be the result of the calculation function of crop transpiration, ventilation, condensation on cover and the crop, taking into account the exchange coefficient at the cover and the crop. Air temperature will be raised from ventilation rate and convective heat flux from the greenhouse cover, the crop and the heating system Covers condensation rate will be proportional to the differences between the absolute vapour pressure of the inside air and the vapour pressure at saturation at the inside surface of the cladding. Crops condensation rate will be proportional to the differences between the absolute vapour pressure of the inside air and the vapour pressure at saturation at the surface of the crop. Transpiration will be calculated using the “big-leaf” approach. The combined model needs calibration and validation. Integration of hourly variables for daily temperature calculation Growth respiration Number of organs Demand for assimilates for each organ Age class for each organ Daily initialization Amount of carbohydrates for the pool Rate of development for each organ Ratio source / sink Calculation the percentage of the appearance Confirm the weather data from Climate Model Organs development rate Maintenance respiration Gross photosynthesis Biological Model Calibration Hourly “Loop” Daily “Loop” Model Structure Requirements Optimisation Method Simulation based optimisation, namely reinforcement learning, will be applied to the combined model that will be constructed. Reinforcement learning methods have scarcely been used on agricultural problems (Garcia, 1999) and not to greenhouse climate optimization although they seem promising. This optimisation method allows searching for the best management strategy for a range of possible future climates to which the greenhouse may be submitted. Observations will be repeated under different climates and strategies to build a matrix mapping a wide range of state-action pairs to their outcome value as simulated by the model. Calibration Reinforcement Learning Inputs of the Biological Model Air Temperature Relative Humidity Solar Radiation CO 2 Concentration Ventilation Heating Acknowledgments This paper is part of the 03ED526 research project, implemented within the framework of the “Reinforcement Program of Human Research Manpower” (PENED) and co-financed by National and Community Funds (25% from the Greek Ministry of Development-General Secretariat of Research and Technology and 75% from E.U.-European Social Fund).