Water resources are still one of the most pressing issues in precision agriculture and will need smart, intelligent solutions to adapt to complicated and ever-changing real-world environmental scenario. This research proposes an Agentic Reinforcement Learning (RL) framework optimized for autonomous smart irrigation and decision-making applications using multi-modal sensor fusion. The system integrates real-time data from soils, climate, and UAV sensors into a single unifying state representation that can operate under real time uncertainties. The proposed model uses an Actor-Critic RL architecture with a hybrid feature-level fusion design to enhance environmental situational awareness and irrigation policy design optimization. Experimental results show the model achieves 29.7% greater savings in water consumption and 17.4% increased yields over baseline models, in addition to being energy efficient. Explainability tools, like SHAP and PCA variance analysis were utilized in order to extract feature influences, establishing transparency and reliability in the framework. The results also suggest the model based on multi-modal fusion significantly improved the R² of prediction accuracy (R² = 0.94) and policy convergence. Our framework represents a major contribution to sustainable Agriculture 4.0 by supporting adaptive, autonomous, data-driven scheduling irrigation frameworks, leading to the eventual scalable pathway towards more resource-efficient farming systems.

Agentic Reinforcement Learning; Multimodal Sensor Fusion; Smart Irrigation Control; Precision Agriculture; Water-Use Efficiency; Explainable AI; Sustainable Agriculture 4.0; Actor–Critic Model; IoT Sensors; UAV Data Integration.