Predictive models as decision support systems for grape downy, mildew powdery mildew and ocratoxin A wine contamination in South Italy

According to Sustainable Development Goals, new and more sustainable control strategies are highly desirable. Predictive models can improve crop protection management and rationalize the use of plant protection products. In this work, a three-year evaluation of forecasting models for the onset of primary infections of downy mildew (EPI and DMCast), powdery mildew (Pmaxacc), and the risk of Ochratoxin A (OTA) contamination (OTA-Grapes) was conducted. The reliability of the models was assessed through the application of Bayesian statistics, in comparison with data collected from over 160 table and wine grape vineyards, representative of the mesoclimatic heterogeneity of the Apulian and Calabrian grape growing areas. The limited occurrence of winter rainy events influenced the mechanistic model DMCast, which correctly predicted the absence of downy mildew, but showed a high rate of false negative predictions. Oppositely, the empiric model EPI consistently rightly predicted the onset of disease, although resulting in a high rate of false positive predictions. Based exclusively on temperatures, Pmaxacc frequently predicted the appearance of powdery mildew in areas with high temperatures during the growing season, even when the disease was undetected. The capability of the OTA-Grapes model to predict the risk of OTA contamination was poor under the assayed conditions; however, the numerous data collected will allow a calibration of the model, improving its reliability and may be helpful in developing maps of risk. These results highlight the need for calibration of forecasting models in different grape growing conditions, especially for empirical models.