A theoretical and real world evaluation of two Bayesian techniques for the calibration of variety parameters in a sugarcane crop model

Process based agricultural systems models allow researchers to investigate the interactions between variety, environment and management. The 'Sugar' module in the Agricultural Productions Systems sIMulator (APSIM-Sugar) currently includes definitions for 14 sugarcane varieties, most of which are no longer commercially grown. This study evaluated the use of two Bayesian approaches to calibrate sugarcane varieties in APSIM-Sugar: Generalized Likelihood Uncertainty Estimation (GLUE) and Markov Chain Monte Carlo (MCMC). Both GLUE and MCMC calibrations were able to accurately simulate green biomass and sucrose yield in both a theoretical and real world evaluation. In the theoretical evaluation GLUE and MCMC parameter estimates accurately reflected differences between two pre-defined sugarcane varieties. We found that the MCMC approach can be used to calibrate varieties in APSIM-Sugar based on yield data. With appropriate variety definitions, APSIM-Sugar could be used for early risk assessment of adopting new varieties.