ALTERNATE APPROACH TO IMPROVE KERNEL NUMBER CALCULATION IN CERES–MAIZ

CERES–Maize is a process–oriented model that has been widely used to predict growth, development, and yield of a maize crop as affected by environment, genotypes, and management. The model has performed well in predicting yield on a large scale (field, county, region) but has had difficulty in predicting the range of variation in kernel numbers found within a field. This study was conducted to compare two approaches to describing the relationship between average intercepted photosynthetically active radiation (IPAR) around silking and the potential number of kernels set per plant (KN). IPAR was averaged during a critical thermal time period between 250 growing degree–days (GDD) before and 100 GDD after silking. Equations describing a double–curve and a line–cutoff were incorporated into the CERES–Maize model (Generic CERES version 3.1) to predict potential number of kernels per plant. The potential number of kernels calculated from this average IPAR was reduced using the water or nitrogen stress factors calculated by the model. An optimization algorithm (Simulated Annealing) was used to calibrate the parameters of the equations, using 105 treatments containing field–measured KN. The new relationships were tested with 134 independent treatments from Iowa. Results indicated that the modified model with double–curve function gave much better predictions of kernel numbers than both the original model and the modified model with line–cutoff, which consistently underpredicted kernel numbers per plant. The mean squared deviation between predicted and measured kernel numbers of the modified model with double–curve function was reduced to 34% of the mean squared deviation obtained with the unmodified model. The modified model requires no additional inputs.