SmartOlivar aims to bring predictive AI to olive farms

The SmartOlivar operational group has started a project applying artificial intelligence and data analysis to olive-growing management. The initiative is coordinated by UPA and includes San Vicente de Mogón Cooperative, Xymbot and the Tecnalia Foundation, with UPA Castilla-La Mancha and UPA Aragón also participating. Its connection to Castilla-La Mancha is not merely a geographic reference: the regional organisation will act as a link with producers, and the project aims to validate the solution on real farms.

The planned architecture combines a predictive-prescriptive system, a mobile application, a dashboard, a digital assistant and low-cost sensors. The system is intended to estimate harvest yield and turn that estimate into personalised recommendations for each farm. From an engineering perspective, this is not simply a model-training exercise. It requires integrating heterogeneous data sources, presenting information through an understandable interface and checking that a recommendation can become a useful agronomic decision.

The project will use farm information and assess the agronomic and economic impact of its solutions. That validation stage is essential. A reasonable prediction on a historical dataset does not guarantee similar performance on plots with different varieties, soils, water availability, management practices or weather conditions. Trials on real farms can measure not only model accuracy, but also whether farmers can interpret the output, whether the recommendation arrives at the right time, and whether sensor and connectivity costs are affordable.

SmartOlivar will run from 2026 to 2029, with total funding of 585,224.54 euros, 80% co-financed by the European Agricultural Fund for Rural Development and Spain’s Ministry of Agriculture. UPA describes the project as particularly focused on small and medium-sized farms. That priority adds a technical requirement: the platform must work with incomplete data, affordable devices and different levels of digital experience. A system that only works with expensive instrumentation or constant connectivity would not address the gap it is meant to reduce.

For software teams, the project offers a lesson that applies to other sectors: useful AI is built as a sociotechnical system. The model needs data with known provenance, quality metrics, change traceability and a clear way to communicate uncertainty. The application should distinguish a prediction from a recommendation, retain the context of the decision and let the user review it. There must also be a feedback loop for correcting errors and comparing outcomes with established practice.

The announcement does not describe a finished production tool or report operational results. It announces a development and validation programme. Its interest lies precisely there: success will depend less on the AI label than on data quality, field evaluation, economic accessibility and the ability to turn a statistical output into a decision that farmers can understand.