Yield response to nutrient supply across a wide range of conditions

This paper tests PARJIB, a simple but powerful model to analyse and forecast yield responses to nutrients in annual crops. The model was calibrated for maize using data obtained under a very wide range of conditions in New Zealand from 1996 to 1999. The calibration process used a genetic algorithm technique. After a preliminary calibration using about half of the available data and only one cultivar, the model performed well, accounting for 66–73% of the observed yield variation in independent datasets that included five cultivars. When simulating the test datasets, the largest errors were associated with one long-season cultivar. The model was then recalibrated against all available data. That dataset spanned five cultivars, four regions of New Zealand, a wide range of soil and weather conditions, and a range of fertiliser application treatments. Grain yields varied from 4 to 18 t/ha. The root mean square (RMS) error of calibration was 0.92 t/ha (9.3%), and the model accounted for 83% of the observed variation in yield. We checked various model predictions against previously published results, and found good agreement. The model indicates strong interactions between water stress, plant population and the need for nutrients. The PARJIB model has substantial potential to aid growers in fertiliser decisions, and to assist researchers in the analysis and interpretation of field experiments.