Combined use of APSIM and logistic regression models to predict the quality characteristics of maize grain

Most physiology-based crop simulation models do not simulate grain quality dynamics other than protein con-tent. In this study, a simple algorithm was adapted for predicting starch, oil, and protein content of two maize cultivars using four years of experimental data performed in northern Iran. Quality modelling was performed in two steps: (a) the APSIM-Maize model was used to dynamically simulate the phenology and growth of the whole crop and its grain protein and (b) a Three-Parameter Logistic model (3PLM) was adjusted to compute the starch and oil contents of grains. APSIM cultivar-specific parameters related to phenology and growth were selected and manually adjusted to reach satisfactory normalized root means square error (nRMSE) between simulation out-puts and field collected data. Grain dry weight dynamics and starch and oil content of maize cultivars were used to calculate the temporal changes of content during grain filling. Then, the parameters of starch and oil accu-mulation 3PLM were adjusted to fit the experimental data. Results showed that APSIM-Maize performed well in simulating the phenological events (flowering: R2 = 0.88, nRMSE = 2.89; maturity: R2 = 0.92, nRMSE = 3.10), grain yield (R2 = 0.94, nRMSE = 9.86) and protein content (R2 = 0.50, nRMSE = 9.40). In addition, adjusted 3PLM models accurately predicted final starch and oil contents of maize cultivars with nRMSE less than 10% and R2 more than 0.90. These results suggest that the combination of APSIM with a simple three-parameter logistic model could be useful for predicting the protein, starch and oil contents of maize grains.