Dominant sources of prediction uncertainty in wheat crop models vary by simulation outputs

Abstract

Context

Uncertainty analysis, encompassing both prediction uncertainty due to model parameter and model structure, is a key element in crop model-based risk assessment and decision-making. It provides essential insights for risk assessors and decision-makers regarding the accuracy of model predictions. Despite its importance, previous studies have predominantly focused on uncertainties in crop phenology, with limited attention to uncertainties affecting crop yield.

Objective

This study aims to evaluate the uncertainties in various simulation outputs of crop models arising from both model parameter and structure, and to further investigate how parameter uncertainty influences simulated wheat phenology and yield under global warming.

Methods

Using data from the International Heat Stress Genotype Experiment (IHSGE), we employed the MCMC method to evaluate prediction uncertainty due to model parameter, and evaluated prediction uncertainty due to model structure based on results from the AgMIP-Wheat multi-model simulations. The CSM-CERES-Wheat model, a widely applied wheat simulation model, was used in this analysis.

Results

When compared to the prediction uncertainty due to model structure from 30 wheat models in the AgMIP-Wheat simulations, the prediction uncertainty due to model parameter of the CSM-CERES-Wheat model was found to be higher for flowering and maturity simulations. However, both model structure and parameter uncertainties were significant contributors to uncertainties in biomass, yield, and grain number simulations, with model structure uncertainty sometimes exceeding parameter uncertainty. Parameter estimation through the MCMC method significantly enhanced accuracy of CSM-CERES-Wheat model, with the ensemble mean reducing the RRMSE by 2 %-10 % in yield simulations. Uncertainty from both sources increases with higher growing-season temperatures and is projected to rise under global warming.

Conclusions

The study highlights that uncertainty due to model parameter plays a crucial role in the accuracy of crop phenology simulations. Both model structure and parameter uncertainties significantly impact predictions for biomass, yield, and grain number, particularly in regions with higher temperature increases during the growing season. Global warming is expected to intensify parameter uncertainty at most sites.

Implications

This study underscores the importance of accurately quantifying model uncertainty to enhance the reliability of crop model predictions, offering valuable insights for future crop impact assessments under climate change.