INTEGRATION OF ARTIFICIAL NEURAL NETWORK AND GEOGRAPHIC INFORMATION SYSTEM FOR AGRICULTURAL YIELD PREDICTION

The main objective of this study was to develop the Artificial Neural Network (ANN) modules for agricultural yield prediction as an extension of the ArcMap software. The Object-Oriented methodology was used for both design and programming. The application coding was done in VB.NET. The ANN modules developed were tested with longan yield prediction in Chiang Mai and Lamphun provinces. The ANN input data are soil group and climate data for the years 2006 - 2008, which relate to longan yield in 2007 and 2008. All data were normalized in the same range of 0-1 to be suitable as the input of the ANN model. The normalized weekly highest, lowest, and average temperature, average sunlight, and rainfall were interpolated. They were then averaged to spatially represent districts in the study area, which corresponded to the longan yield districts. These data were varied with several input variations. The cross validation process was applied to each variation. The optimal parameters including learning rate, number of nodes in the hidden layer, and number of iterations obtained from testing were 0.4, 6, and 3,000 respectively. These parameters were applied for all training and testing processes. The best accuracy achieved is 99%. The ANN modules developed for the ArcMap environment worked well for longan yield prediction with accurate results despite the limitations of the data set.