From “Black Boxes” to Clear Guidance: How UNIVERSWATER Empower Sustainable Agriculture with AI
Digital agriculture increasingly relies on data from satellites, drones, IoT sensors, and laboratory analyses to support more sustainable farming practices. These technologies provide unprecedented insight into crops, soils, and water systems, but their true value lies in turning complex data into guidance that farmers can understand, trust, and act upon. Explainable Artificial Intelligence (XAI) is therefore essential to ensure that digital tools genuinely support informed decision-making rather than adding further complexity.
Within the UNIVERSWATER project, advanced decision-support tools are being developed to help farmers make informed choices on fertiliser use, pest activity prediction, and irrigation management. These tools integrate heterogeneous data sources, including Earth Observation imagery, in-situ measurements, and IoT sensors— in order to assess current soil and water conditions and anticipate future risks related to nutrient leaching, pest pressure, and water quality. Rather than operating as opaque “black boxes”, Explainable AI approaches clarify how recommendations are produced at field level, showing which factors, such as soil nutrient status, weather conditions or the quality of water used for irrigation, drive specific suggestions and how alternative actions could influence outcomes.
By linking advanced analytics with environmental domain knowledge, UNIVERSWATER ensures that data-driven insights remain practical, robust, and trustworthy, even under changing climatic conditions. Within this effort, the University of Western Macedonia (UOWM), under the scientific coordination of Prof. Malamati Louta leads the Greek use case in the Small Prespa lake of Greece. The Use Case focuses on advancing explainable AI approaches that make machine learning model outputs transparent and interpretable for decision support. UOWM applies feature-attribution methods, such as SHapley Additive exPlanations (SHAP), to reveal how different data inputs contribute to each recommendation, enabling farmers and stakeholders to understand not only what action is suggested, but also why it is recommended.
This work ensures methodological accuracy and strongly aligns with real farming and water-management needs. The integration of explainability into decision-support tools for fertiliser application and water quality management allows UOWM to align AI-driven recommendations with real farming practices, strengthening trust in digital tools and supporting more sustainable and evidence-based agricultural decision-making.
