Thermal unit is often used as the main driving force in crop simulation models. However, simulation models built with this approach often do not lead to a satisfactory accuracy of prediction when it regards to soybean; mainly due to strong photoperiod influence on soybean and complicated interactions between photoperiod and temperature. This study tried to simulate and predict soybean phenological growth using calendar-day based approach. Field experiments were conducted at the Delta Research and Extension Center, Stoneville, Mississippi, USA. Five year (1998-2002) field data were used with 24 sowing dates from maturity groups (MG) III to MG VI soybean varieties. Three methods, artificial neural network (ANN), k- nearest neighbor (kNN) and regression were used to construct prediction models. Vegetative and reproductive growth stages were modeled separately. Results indicated that calendar-based prediction model in soybean growth calculation is a feasible approach. All three methods achieved the acceptable prediction accuracy. On average, prediction errors of ANN, kNN and Regression methods were 3.6, 2.8 and 3.6 days for vegetative stage and 4.4, 3.5 and 4.7 days for reproductive stages, respectively.

Grain Legumes; Plant Breeding; Plant Production; Soybean

Soybeans; Phenology; Prediction; Model; Regression; United States Of America; Artificial Neural Network; Simulation

Acp; North America

United States

American-Eurasian Journal of Agricultural and Environmental Sciences