Testing the Performance of Two Maize Simulation Models with a Range of Cultivars of Maize (Zea mays) in Diverse Environments

Maize production is increasing in importance in Australia, and has potential for substantial further expansion. Additional production areas and/or more intensive use of existing production areas will be needed. Simulation models offer the capacity to rapidly assess the suitability of a range of genotypes and phenotypes, and to predict yield and yield reliability over a range of environmental conditions. However, they must be validated and be sufficiently robust to provide reliable predictions. The performance of two maize simulation models, a complex mechanistic one, AUSIM-Maize, and a simpler one, the Muchow - Sinclair model, was evaluated against experimental data from field trials at Gatton, South East Queensland and Katherine, Northern Territory. AUSIM-Maize predicts phenological and canopy development, total dry matter and grain yield. The Muchow - Sinclair model concentrates on total dry matter and grain yield. Sensitivity analysis indicated that the output of the models was most affected by the values used for the duration of the basic vegetative period, photoperiod sensitivity and leaf initiation rate (in AUSIM - Maize), radiation use efficiency, leaf appearance rate (in both models) and one coefficient that affects leaf area senescence (in the Muchow - Sinclair model). AUSIM - Maize consistently overpredicted the time from emergence to tassel initiation (especially with short season cultivars, and when environmental conditions favoured rapid plant development to TI), silking and physiological maturity. Leaf number was consistently overpredicted by AUSIM - Maize. Neither model predicted total dry matter or grain yield satisfactorily over the range in the experimental data, though each tended to be more accurate than the other on one measure of model performance (regression or root mean square deviation). Both provided sound predictions within a limited range of conditions and genotypes that resulted in relatively short crop durations, but were inaccurate when the data extended over a greater range of environmental conditions and genotypes. Several areas of the models where modification is needed to improve predictions and to make the models more generally applicable are identified