Parameter estimation and uncertainty analysis of a large-scale crop model for paddy rice: Application of a Bayesian approach

Abstract

A Bayesian approach, the Markov Chain Monte Carlo (MCMC) technique, was applied to a newly developed large-scale crop model for paddy rice to optimize a new set of regional-specific parameters and quantify the uncertainty of yield estimation associated with model parameters. The developed large-scale model is process-based and up-scaled from a conventional field-scale model to meet the intended spatial-scale of the large-scale model to the typical grid size of high-resolution climate models. The domain of the large-scale model covers all of Japan, but the crop simulation is conducted for each local governmental area in Japan. The MCMC technique exhibits powerful capability to optimize multiple parameters in a nonlinear and fairly complex model. The application of the Bayesian approach is useful to quantify the uncertainty of model parameters in a comprehensive manner when researchers on crop modeling analyze the uncertainty of yield estimation associated with model parameters under given observations. A sensitivity analysis of the large-scale model was conducted with the obtained posterior distribution of parameters and warming conditions that have never been experienced before to demonstrate the change in the uncertainty of yield estimation associated with the uncertainty of parameters of the large-scale model. The uncertainty of yield estimation under warming conditions was larger than that obtained under climate conditions that have been experienced before. This raises a concern that the uncertainty of impact assessment on crop yield may increase if future climate projections are fed to crop models with parameters optimized under current climate conditions.