Explainable AI for Smart Greenhouse Control: Interpretability of Temporal Fusion Transformer in the Internet of Robotic Things

Black box AI systems in agriculture now have transparency tools: The study demonstrates that complex AI models controlling greenhouse equipment can be made interpretable using explanation techniques (LIME and SHAP), addressing a critical gap where automated farming decisions previously operated without visible reasoning—important for farmers and regulators who need to understand why systems make specific choices.

Multiple sensor inputs have varying importance for automation decisions: The research reveals that different environmental sensors (temperature, humidity, CO2, light, and outdoor climate) contribute differently to actuator control decisions. This finding helps farmers understand which measurements most influence their greenhouse's automated responses, enabling them to prioritize sensor maintenance and calibration accordingly.

High accuracy alone isn't sufficient for trustworthy AI in farming: Although the model achieved 95% test accuracy, the paper emphasizes that accuracy metrics alone are insufficient for smart farming applications. The explicit focus on explainability alongside performance highlights that farmers and regulators need to understand how decisions are made, not just that they work—a principle applicable across responsible AI implementation.

Explainability supports regulatory compliance and adaptive management: By making AI decision-making transparent, the approach enables compliance with farming regulations and allows farmers to fine-tune systems based on interpretable insights rather than treating automation as a fixed black box, supporting both accountability and continuous improvement in resource efficiency.