24th MEB Conference - Budapest, 24.04.2026

Efficient water resource management is paramount for sustainable agriculture amidst increasing global population and climate variability. This paper establishes a novel theoretical framework for an AI-driven Decision Support System (DSS) specifically designed to enhance precision irrigation practices. The primary aim of this investigation is to leverage this framework to develop a deep learning AI model capable of accurately predicting and precisely detecting water stress sections within crops of interest, thereby enabling highly targeted and efficient water applications. The proposed framework integrates multiple heterogeneous data sources to construct a comprehensive spatio-temporal understanding of crop water status. It includes Earth Observation (EO) data from Sentinel-2 B. Future research will focus on this advanced AI model's rigorous development, training, and validation.satellites, specifically utilizing vegetation indices such as Normalized Difference Vegetation Index (NDVI) for assessing vegetation health and Normalized Difference Moisture Index (NDMI) for soil water content. Complementing this, high-resolution in-situ measurements are collected by IoT sensors (e.g., IoT-NPK for soil moisture, NPK levels, temperature, and pH) mounted on mobile robot platforms like PlatypOUs, providing essential ground truth validation. Furthermore, meteorological data, i.e., precipitation, air, and soil humidity, is integrated to provide crucial environmental context and predictive insights. This paper outlines a methodology for developing a Recurrent Neural Network (RNN) architecture based on a U-Net topology that will effectively encode features from these integrated data streams. The model incorporates multiple convolution layers for efficient spatial feature extraction, Long Short-Term Memory (LSTM) layers to capture temporal dependencies, and attention layers to focus on the most critical features for prediction. The ultimate output is a newly generated image representing the predicted spatial distribution of water stress across the field of interest, allowing pixel-based classification for targeted irrigation recommendations. This foundational investigation, including initial data analysis and feature engineering, paves the way towards optimized water use, significantly improving agricultural productivity and enhancing resource conservation.