Evaluating the APSIM crop simulation model for prediction of lucerne dry matter yield in Canterbury

Abstract

Yield data collected from an irrigated lucerne (Medicago sativa) crop at Lincoln University from 1997/98 to 2000/01 were used to calibrate the APSIM-Iucerne simulation model. Throughout each regrowth cycle within each season, the lucerne was sampled at 7 - 14 day intervals for dry matter accumulation.

The original APSIM-Iucerne model settings were used to calculate yield predictions for the lucerne crop and these were compared with the field observations using a root mean squared deviation (RMSD) calculation. With the exception of the first regrowth cycle within a year, sampling dates within all regrowth cycles achieved an RMSD of more than 60% and it was above 140% in the third and sixth regrowth cycles. This meant that the model was over predicting total annual yield of 3 to 13 t DM/ha.

APSIM-Iucerne variables, including frosting, the amount of herbage removal at defoliation, cardinal temperatures for thermal time accumulation, stem populations and radiation use efficiency were all used to alter the model's yield prediction. By modifying these, RMSDs of the final five regrowth cycles within each season were improved markedly, ranging from 12% in the fifth regrowth cycle to 33% in the sixth regrowth cycle. A limitation in the model meant that after implementing changes based on published data, the accuracy of prediction in the first spring regrowth cycle each season was adversely affected. As a result, the RMSD increased to 79% for that period. Annual yield predictions were improved vastly, from ~ 10 t DM/ha per year above observed to within ~ 3 t DM/ha.

Validation of the changes made to the model was made by using a dataset gathered from 2001/02 - 2003/04 at the same experimental site at Lincoln University. The modified model generated a distinct pattern of standard bias for each season, with yield prediction in the first 3 regrowth cycles under-predicted and the final 3 regrowth cycles over predicted. It is proposed that the inability of the model to account for storage and remobilisation of carbohydrate in the autumn and spring regrowth cycles respectively is a major contributor to the pattern of yield prediction.

Lucerne grown at Lincoln University in a dryland experiment at the same time as the calibration data was accurately modelled. The RMSDs for the final five regrowth cycles ranged from 15 to 35%. The first regrowth cycle in spring was poorly modelled, with an RMSD of 79%.

The current study showed that APSIM-Iucerne was able to predict yield production with an acceptable level of accuracy once modified to account for the latest physiological understanding. However, the study also exposed large gaps in the current body of work, with the biggest unknowns in the area of carbohydrate storage and remobilisation and in the precise environmental factors that lead to crop frosting.