YIELD PREDICTION AND GROWTH MODE CHARACTERIZATION OF GREENHOUSE TOMATOES WITH NEURAL NETWORKS AND FUZZY LOGIC

Despite the technological advances implemented in greenhouse crop production, greenhouse operation relies of? human expertise to decide on the optimum values of each environmental control parameter. Most importantly, the selected values are determined by human observation of the crop responses. Greenhouse tomatoes often show a pattern of cycling between reproductive and vegetative growth modes. The growth mode is a practical visual characterization of the source-sink relationships of the plants resulting from the greenhouse environment (aerial and root zone). Experienced greenhouse tomato growers assess the growth mode based on morphological observations, including quantitative (length, diameter, elongation rates) and qualitative (shape and color) features of the plant head, stems, flowers, trusses, and leaves. Data front greenhouse environments and crop records front an experimental production in Tucson, Arizona, and from a large-scale commercial operation in Marfa, Texas, were used for modeling the growth mode of tomato plants with fuzzy logic. Data from the commercial operation were used to model weekly fluctuations of harvest rate, fruit size, and fruit developing time with dynamic neural networks (NN). The NN models accurately predicted weekly and seasonal fluctuations of the fruit-related parameters, having coefficients of determination (R-2) of 0.92, 0.76, and 0.88, respectively, for harvest rate, fruit fresh weight, and fruit developing time, when compared with a dataset used for independent validation. The fuzzy modeling of growth mode allowed discrimination of the reproductive and balanced growth modes in the experimental system, and modeling of the seasonal growth mode variation in the commercial application. Both modeling results might be applicable to commercial operations for making decisions on greenhouse climate control and overall crop management practices.