Comparison of the wheat simulation models Afrcwheat2, Ceres-wheat and Swheat for non-limiting conditions of crop growth

Abstract

Comparison of crop simulation models is rare but necessary pre-requisite for their use and improvement. The wheat crop simulation models Afrcwheat2, Ceres-Wheat and Swheat were tested against data from five closely monitored experiments on cvs Avalon and Rongotea using weather data from Lincoln, New Zealand for 1984, 1985, and 1986. There was a single sowing date in 1984 and 1985, and three sowing dates in 1986. Sufficient irrigation water and nitrogen fertiliser were applied each year not to limit crop productivity. Afrcwheat2 was the most accurate predictors of development. With this model, average root mean square differences over all growth stages until anthesis were 5.8 days for cv. Avalon and 7.6 days for cv. Rongotea, but 10.4 days and 15.8 days respectively using Ceres-Wheat. Swheat had an average error of 26.4 days for cv. Rongotea. Interception of photosynthetically active radiation (PAR) was modelled with reasonable accuracy for both varieties for all sowings by afrcwheat2, except or an over-estimation of the rate of decline in green leaf area following anthesis. Conversely, Ceres-Wheat often overestimated the proportion of PAR intercepted before anthesis but was usually closer than Afrcwheat2 for the rest of the season for Avalon. Afrcwheat2 described PAR interceptance better than did the other models for cv. Rongotea except for late sowing in 1986 for which no model performed especially well. Total dry matter increase was best predicted by Afrcwheat2 for both varieties. Ceres-Wheat predicted Avalon final yields better than did Afrcwheat2. Averaged over the five crops, the absolute error in prediction grain yield in Avalon by Ceres-Wheat was 12% of that observed whilst that for Afrcwheat2 was 15%. For cv. Rongotea, the Afrcwheat2 error was 14%, but that with Ceres-Wheat was 29%. Swheat predictions were low by an average of 40%. Simulation of other components of grain yield was more satisfactory for cv. Rongotea than for cv. Avalon. For the circumscribed range of non-limiting experimental conditions examined, Afrcwheat2 and Swheat underestimated leaf area during its decline folloing anthesis, and this was reflected in underestimation of biomass production during this phase. The errors in estimation of grain dry mass, grain number and final shoot number were inconsistent with respect to both variety and model. The chief improvement with Swheat is in simulation of crop phenology. Improvements in the estimation of PAR interceptance by Ceres-Wheat and Afrcwheat2 are required at different times during the crop growing cycle.