A Crop Growth Model For Predicting Corn (Zea mays L.) Preformance in the Tropics

Abstract

The Crop Environment Resource Synthesis (CERES) maize model was verified, calibrated, and validated on data from a wide range of agroenvironments in the tropics. These agroenvironments ranged from 5 S to 20 N latitude and from sea level to 800 meters above sea level. The model assumed: (i) complete irrigation; (ii) all nutrients at optimum level except nitrogen; (iii) no weeds, pests, and pathogens; and (iv) no wind damage. Adjustments were made only on physiological basis. These adjustments were made to: (i) incorporate soil temperature as a means of computing thermal time up to the tassel initiation stage; (ii) modify maize genotype coefficients based on field data; (iii) raise optimum temperature for photosynthesis; (iv) reflect the effect of minimum temperature instead of mean temperature on grain filling ; (v) reflect the effect of nitrogen deficiency and water stress on grain numbers; and (vi) lower the nitrogen mineralization constant based on mineralogical and chemical properties of the soil. The model was designed to minimize the need for future model calibration when the factors currently not simulated are later incorporated into the model. CEELES maize model predictions for phonological development, kernel weight, kernels per ear, and grain yield were nonsite-specific. The model was sensitive to latitudinal differences, seasonal variation, altitudinal differences, response to nitrogen fertilizer applications and planting density. However, unmeasured environmental and management variables caused considerable differences between simulated and observed values. These variables affected yield predictions and phenological development. The CERES maize model was able to mimic the high sensitivity of maize to temperature and solar radiation. Evaluation of statistical validation techniques indicated that both the R and the Freese statistics required improvements. The R test accepted model predictions which were subjectively "poor" because the field experiment had a large coefficient of variation. The Freese statistics, on the other hand, showed that the CERES maize model was capable of simulating grain yields from 2,500 to 11,200 kg ha”^ with a critical error of approximately 1,200 kg ha“^, in a wide range of agroenvironments, when a model bias to overestimate in yield was taken into account. Phosphorus regression models were developed to determine labile phosphorus, organic phosphorus, buffering capacity, and phosphorus availability index from readily available soil test F methods and soil physical and chemical properties. These models were used to generate input data for the phosphorus simulation model. With the above changes the P model simulated maize grain yields with high accuracy.