Releasing the octoPus, an open-source digital tool to promote Integrated Pest Management

• The octoPus is a digital tool to support farmers in grapevine downy mildew control. • Eight disease models were combined in the first multi-model ensemble in plant pathology. • Grapevine phenology and host susceptibility have been derived with specific models. • Simulated outcomes were post-processed by Random Forest and Large Language Models. • Calibration and validation supports the use of the octoPus to reduce fungicide use. Decision support systems are primary tools to meet EU targets to halve pesticide use, as they advise farmers to use chemical sprays only when weather conditions are conducive to plant pathogens. Although many alternative disease models are available from scientific literature, current decision support systems are often proprietary and based on individual modelling approaches. To overcome this limitation, we developed the octoPus , the first open-source modelling tool designed to enhance the control of primary infections of grapevine downy mildew. Eight disease models have been reimplemented from scientific articles (the "tentacles"), and their outputs integrated with a grapevine phenology and susceptibility model (the "eyes"). Simulated infections served as predictors in a Random Forest algorithm ("brain") to compute an overall risk level. All model outputs have been prompted to a large language model (Llama), which has been used to generate user-supportive messages according to Integrated Pest Management principles (the "mouth"). We evaluated the behavior of the eight disease models through gridded simulations across the whole Italian grapevine areas from 2001 to 2020. Reference data to calibrate grapevine phenology (RMSE = 9-10 days) and seasonal risk of downy mildew infection (RMSE ≈ 0.75) have been extracted from official plant protection bulletins at provincial level (NUTS-3). The disease models displayed significant differences in the number and dynamics of simulated infections, with two distinct patterns within the ensemble. However, they consistently identified a moisture and thermal north-south suitability gradient in Italy, accurately detecting grapevine seasons with low or high downy mildew pressure. The octoPus addresses a critical gap in the availability and transparency of decision support systems for extension services and farmers, who can freely use it to support plant protection strategies to control grapevine downy mildew.