Multi-Crop <i>Sclerotinia sclerotiorum</i> Apothecia Prediction Models for Irrigated Environments Are Improved by On-Site Weather Monitoring and Supervised Machine Learning

<i>Sclerotinia sclerotiorum</i> causes Sclerotinia stem rot, or white mold, on multiple economically important crops in Michigan. Soybean farmers and crop consultants in the Midwestern U.S. currently use <i>S. sclerotiorum</i> apothecia prediction models to inform fungicide application timing to optimize disease control and economic return. However, current models have not been validated for use in dry bean or potato and do not account for the effects of irrigation on apothecia development. To improve <i>S. sclerotiorum</i> apothecia prediction, on-site weather data were collected and used to generate new binomial logistic regression (LR) and supervised machine learning (ML) models for irrigated soybean, dry bean and potato fields. The ML algorithms investigated included decision trees, random forests and support vectors machines. Decision tree classification models outperformed LR and other ML models, achieving 77% accuracy on testing data. Accuracy increased to 89% when on-site weather data were included, indicating that on-site weather monitoring may be required to reliably predict apothecia presence in irrigated environments. Feature importance analysis identified row shading (the distance the plant canopy extends into the row) as critical for prediction accuracy. The minimum row shading required to trigger apothecia development varied slightly between crop types and row spacings, from 0.15 to 0.21m. Apothecia density peaked when soil temperatures were 21.51°C and volumetric water content were 11.43% and 19.58%. Additionally, a rapid increase in apothecia presence was observed after canopy closure reached 87%. Future model testing and validation will be required prior to deployment as a decision aid for farmers and crop consultants.