Prediction of Soybean Growth and Development Using Artificial Neural Network and Statistical Models

Abstract

To develop predictive models using a simple and effective modeling technique, field experiments were conducted at the Delta Research and Extension Center in Stoneville, Mississippi (33°25′ N, 90°55′ W) under irrigated conditions. The models were constructed using 4 years of field data (1998 to 2001), and validated with a fifth year of data (2002). Potential factors affecting stages of soybean growth and development were considered for developing the models. Affecting factors, such as weeds, insects, diseases, and drought stresses, were controlled optimally to simplify the modeling procedures. In addition, stepwise regression (SR) analysis, artificial neural networks (ANN), and interpolation approaches were used to construct the models. The modelling of soybean growth and development processes was separated into 2 distinct periods: vegetative growth stage (V-stage) and reproductive growth stage (R-stage). The models included 10 V-stages (up to V8) and 8 R-stages. In the V-stage models, planting date (PD) and mean relative time-span from planting to a particular stage were the only significant parameters, whereas in R-stage models, PD and maturity group (MG) were significant. The models obtained accurate predictions when only using PD, MG, and mean relative time-span from planting to a particular stage. The ANN method provides the greatest accuracy in predicting phenological events, indicating that the ANN method can be effectively applied in crop modeling.