Context and purpose of the study Plasmopara viticola, the causal agent of downy mildew (DM), and Uncinula necator, the causal agent of powdery mildew (PM), are two of the main phytopathogenic microorganisms causing major economic losses in the primary sector, especially in the wine sector, by wilting bunches and leaves with a consequent decrease in the photosynthetic rate of the plant and in the annual yield. Currently, the most widespread methods for planning spraying are based on the 3-10 rule, which states that the first application should take place when: (i) the air temperature is greater than 10°C; (ii) shoots are equal or greater than 10 cm; and (iii) a minimum of 10 mm rainfall within 24–48 hours has occurred, or at the beginning of the bud break with periodic applications according to the manufacturer’s instructions. These rules are applied to prevent possible infectious events that may occur while new tissues are forming on the vine, which are more susceptible to infection. In addition, establishing a starting point for spraying is crucial, as the pathogen can complete the infection cycle in one to two weeks depending on environmental conditions. However, this approach is not completely effective, as the chemical compound can be washed off the leaves, photo-oxidized, applied at higher doses than necessary, negatively affecting the biodiversity of the agroecosystem, or in discordance with the life cycle of the pathogen. Therefore, the aim of the VitiGEOSS disease early warning service focuses on the application of Artificial Intelligence models to predict the appearance of diseases in the vineyard and consequently apply fungicide products at the right time and dose, minimizing crop losses and the use of pesticides and water. Material and methods A total of six study plots located in three countries of the European Union were used: Quinta do Bomfim (Portugal), L'Aranyó (Spain) and Mirabella Eclano (Italy). Disease monitoring was carried out from March to October 2021 and 2022, with field visits every 7 days to measure the percentage of incidence and severity of infection on leaves. To analyze these data, eight different Machine and Deep Learning models were evaluated to classify the degree of infection and provide treatment recommendations using climatic features and phenological change events in the plant. Results The three study regions showed significant climatic differences. On one hand, the best prediction algorithm was the one based on conditional probability obtaining a precision metric of 90% for DM and 79% for PM, respectively. On the other hand, a comparative analysis showed that the incorporation of plant phenological stages in the model increased the accuracy rate up to 9%, so it would be interesting to consider the effect of other physiological aspects of the plant for future analyses. Finally, it should be noted that model recommendations reduce water consumption by 21% on average. In any case, it is advisable to continue collecting data, as two production seasons can lead to overfitting issues, and to incorporate climatological and phenological predictions to be able to develop short- and medium-term warnings.