Uncertainty in canola phenology modelling induced by cultivar parameterization and its impact on simulated yield

Abstract

Accurate modelling of crop phenology is essential for evaluating how crops respond to future environmental and management changes. However, cultivar parameters are often estimated based on limited data and using a trial-and-error method, leading to uncertainties in simulated phenology and subsequent crop yield. In this paper we evaluated the ability of the APSIM-Canola model to simulate canola phenology and the impact of uncertainty in phenology modelling on simulated grain yield.We constrained the APSIM-Canola model to experimental data to derive the parameters controlling canola flowering and maturity dates using a Bayesian optimisation method, by minimising the RMSE between simulated and recorded pre- and post-flowering durations. The dataset covered observations for 10 cultivars, 35 site-years, with maximum of seven sowing dates each year from four canola growing regions in China.Our results demonstrated that multiple combinations of parameters could lead to the same simulation accuracy of canola phenology (equifinality) due to insufficient information/understanding to separate vernalisation and photoperiod sensitive phases. This could potentially lead to incorrect cultivar characterisation and wrong yield simulations. Our results further showed that the critical photoperiod below which canola phenological development slows down is likely to be 20h instead of the 16.3h currently used in the APSIM model. With this correction, the model was able to accurately simulate canola phenology across environments, and the impact of equifinality on simulated yield was small. Cultivar differences in terms of phenology could be accurately described by only three parameters in APSIM, i.e., vernalisation sensitivity, photoperiod sensitivity, and thermal time required for grain-filling period.