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

Abstract

Meeting the EU targets to halve chemical pesticide use by 2030 necessitates European farmers to adopt Integrated Pest Management principles as the standard. Decision support systems are valuable tools to meet this target and rely on individual disease models to identify conducive conditions to fungal infections. These models are often proprietary assets of digital startups and agrochemical companies, leading to a lack of transparency for farmers and a bias towards chemical solutions over sustainable practices. We present octoPus, the first free digital tool designed to support the control of primary infections of grapevine downy mildew, and we evaluate its performance and behavior on a wide set of environmental conditions and agricultural contexts. We implemented eight models from scientific articles (Rule310, Laore, EPI, IPI, DMcast, UCSC, Misfits, Magarey, the tentacles), and evaluated them across Italian grapevine areas from 2001 to 2020. Model outputs were integrated with phenology and susceptibility models (the eyes), which were calibrated using data from regional extension services bulletins. The simulated infections serve as predictors in a Random Forest algorithm (brain) that elaborates an overall risk level (very low to very high). The Llama large language model is used to generate user-supportive messages (the mouth). octoPus is released as an open-source software, which reads weather data, executes the models, and presents outputs in natural language and symbolic syntax. Our results showed reasonable accuracy in simulating grapevine phenology (RMSE = 9-10 days) and seasonal risk (RMSE ≈ 0.75). The infection models consistently identified a moisture and thermal north-south suitability gradient in Italy and accurately detected years with low or high downy mildew pressure. However, the models displayed significant differences in the number and dynamics of simulated infections, with two distinct patterns within the ensemble. Meeting the EU targets to halve chemical pesticide use by 2030 necessitates European farmers to adopt Integrated Pest Management principles as the standard. Decision support systems are valuable tools to meet this target and rely on individual disease models to identify conducive conditions to fungal infections. These models are often proprietary assets of digital startups and agrochemical companies, leading to a lack of transparency for farmers and a bias towards chemical solutions over sustainable practices. We present octoPus, the first free digital tool designed to support the control of primary infections of grapevine downy mildew, and we evaluate its performance and behavior on a wide set of environmental conditions and agricultural contexts. We implemented eight models from scientific articles (Rule310, Laore, EPI, IPI, DMcast, UCSC, Misfits, Magarey, the "tentacles"), and evaluated them across Italian grapevine areas from 2001 to 2020. Model outputs were integrated with phenology and susceptibility models (the "eyes"), which were calibrated using data from regional extension services' bulletins. The simulated infections serve as predictors in a Random Forest algorithm ("brain") that elaborates an overall risk level (very low to very high). The Llama large language model is used to generate user-supportive messages (the "mouth"). octoPus is released as an open-source software, which reads weather data, executes the models, and presents outputs in natural language and symbolic syntax. Our results showed reasonable accuracy in simulating grapevine phenology (RMSE = 9-10 days) and seasonal risk (RMSE ≈ 0.75). The infection models consistently identified a moisture and thermal north-south suitability gradient in Italy and accurately detected years with low or high downy mildew pressure. However, the models displayed significant differences in the number and dynamics of simulated infections, with two distinct patterns within the ensemble. By developing and releasing the first free and open-source tool to support the control of grapevine downy mildew, we address a critical gap in the availability and transparency of decision support systems for European farmers. Unlike proprietary models that often lack transparency and may favor agribusiness' logic, octoPus provides a comprehensive and accessible alternative that promotes Integrated Pest Management practices. We propose the adoption of octoPus by plant health authorities to identify areas for performance refinement and capabilities expansion.