How do growers respond to host resistance? A conditional Gaussian Bayesian network for causal inference of fungicide cost savings

Authors: Hwang, Jae Young; BHATTACHARYYA, SHARMODEEP - Oregon State University; CHATTERJEE, SHIRSHANDU - City University Of New York; MARSH, THOMAS - Washington State University; PEDRO, JOSHUA - City University Of New York; Gent, David

Submitted to: Phytopathology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/24/2025
Publication Date: 12/29/2025
Citation: Hwang, J., Bhattacharyya, S., Chatterjee, S., Marsh, T.L., Pedro, J.F., Gent, D.H. 2025. How do growers respond to host resistance? A conditional Gaussian Bayesian network for causal inference of fungicide cost savings. Phytopathology. 116(2):239-251. https://doi.org/10.1094/PHYTO-06-25-0199-R.
DOI: https://doi.org/10.1094/PHYTO-06-25-0199-R

Interpretive Summary: There are large investments made in developing plant cultivars that are disease resistant. The economic value of such resistance is of central importance for making informed decisions on resource allocation and understanding the economic value of resistant cultivars. In this research we draw upon a well-described data set of the incidence of hop plants with powdery mildew and associated production meta-data as a motivating example that demonstrates the utility of Bayesian networks as a framework for quantifying causal relationships in fungicide use patterns and costs. A Bayesian network provides a directed causal graph that links covariates in a multivariate data setting and can be used to infer and estimate causal effects of a given intervention. We found that cultivars differing in race-specific resistance to powdery mildew influence disease levels in the early stages of epidemics, which have a causal effect on how often and what fungicides growers later apply. The annual cost of fungicides depended not only on the number of applications made but the specific types of fungicides growers selected. For cultivars with resistance to all cognate pathogen strains present, annual costs of fungicides were reduced commensurate with the level of resistance. We find that growers apply a baseline level of fungicide, independent of cultivar resistance. However, fungicide cost savings result from how fungicide inputs differentially scale with the incidence of powdery mildew and the type of fungicides used. Our analyses indicate that for a high value crop, deployment of disease resistance may cause complex and unexpected changes in growers’ fungicide use patterns that may not be obvious in simple experiments.

Technical Abstract: The economic value of cultivars resistance to disease is of great interest, but how growers change fungicide use in response to host resistance may be nuanced. We draw upon a well-described data set of the incidence of hop plants with powdery mildew and associated production meta-data and demonstrate the utility of Bayesian networks as a framework for quantifying causal relationships that change fungicides use and cost in response to host resistance. Bayesian network methods provide a directed causal graph between covariates in a multivariate data setting. Conditional Gaussian Bayesian network models applied to cultivars differing in race-specific resistance to powdery mildew revealed cultivar resistance to powdery mildew influenced disease levels in early spring, which had causal effect on how often and what fungicides growers later applied. Annual costs depended not only on the number of applications made but the specific types of fungicides growers selected. Fungicide costs were little changed on cultivars that possessed race- specific resistance to only one of two extant strains of the pathogen. For cultivars with resistance to both pathogen strains, annual costs of fungicides were reduced commensurate with the level of resistance. Predicted values from the Bayesian networks and simulation indicate that growers apply a baseline level of fungicide, independent of cultivar resistance. Fungicide cost savings result from how fungicide inputs differentially scale with the incidence of powdery mildew and the type of fungicides used. Our analyses indicate that for a high value crop, deployment of disease resistance may cause complex and unexpected changes in growers’ fungicide use patterns that may not be obvious in simplified randomized controlled trials.