Assessing uncertainty of soybean yield response to seeding rates in on-farm experiments using Bayesian posterior passing technique

Abstract

Understanding the optimum seeding rate for soybeans is crucial to maximizing the revenue of farmers amidst rising seed costs. On-farm experimentation (OFE) is often performed over several years to gather information about the uncertainties of yield response to different seeding rates. This study aimed to testify the potential of the posterior passing technique under the Bayesian approach by incorporating the results from preceding OFE trials as the prior information of the following year's trials to reduce the uncertainty of optimum seeding rate input. OFE trials were conducted in Gifu, Japan, over two growing seasons. A Gaussian process model was used to evaluate the impact of the seeding rate on yield while accounting for spatial variations in the fields. Two types of prior distributions were tested, including noninformative (no prior knowledge) and informative (based on previous OFE trials) priors. Model established using informative priors could improve predictive performance and reduce uncertainty in yield response for subsequent trials. However, the utilization of posterior passing also needs to be cautious, as prior distribution with small variance may lead to unreliable results to the following yield response. In the current results, providing a single general optimum seeding rate is impractical, as each model contribute to a different prescription. Nonetheless, as the OFE framework is a continuous learning process, integrating the trial results with posterior passing technique offers a promising way to improve confidence in determining optimum seeding rates if there are more available datasets.