Rice Yield Prediction Based on the Soil Chemical Properties Using Neural Network Model

Precision agriculture attempts to improve cropping efficiency by variable application of crop treatments such as fertilizers and pesticides, within field on a point-by-point basis. Therefore, a more complete understanding of the relationships between yield and soil properties is of critical importance in precision agriculture. In this study, the functional relationships between measured soil properties and rice yield were investigated. A supervised back-propagation neural network model was employed to relate soil chemical properties and rice yields on a point-by point basis, within individual site-years. As a results, a positive correlation was found between practical yields and predicted yields in 1999, 2000, 2001, and 2002 are 0.916, 0.879, 0.800 and 0.789, respectively. The results showed that significant overfitting for yields with only the soil chemical properties occurred so that more of environmental factors, such as climatological data, variety, cultivation method etc., would be required to predict the yield more accurately.

인공신경망 모형을 이용하여 토양 화학성으로 벼 수확량 예측