Incorporating prolificacy into CERES-maize prediction of kernel number

South Africa produces an average of 6.7 t of maize (Zea mays L.) annually, with a standard deviation of 2.6 t. This large variation indicates that maize is produced under environmental conditions that change markedly from one season to another. The variation in yield is closely correlated with kernel number per plant. Therefore, simulating this variation in maize yield, depends on accurate simulation of kernel number. We tested whether kernel number simulation by CERES-Maize, which generally estimated kernel number with low accuracy, could be improved by considering kernel set on secondary and tiller ears. The CERES-Maize model v3.0 was recalibrated with data from trials of six row widths over three seasons and a plant population trial of five cultivars at four plant populations in one season. Kernel numbers for apical, secondary, and tiller ears were simulated separately. By including the number of days from silking to the commencement of rapid kernel growth, it was possible to improve the simulation of kernel number. The original function for kernel number simulation had low sensitivity to water stress from silking to the commencement of rapid kernel growth. With the modifications made to CERES Maize v3.0, this sensitivity was increased. The D-index (model accuracy index) for kernel number simulation was increased from 0.11 to a value of 0.87. The accuracy in kernel number simulation improved for both calibration and the independent data sets, due to improved water stress sensitivity.