Green AI for Sustainable Agriculture: An Explainable-AI (XAI) Perspective on Energy-Efficient Crop Intelligence

In modern agriculture precision forecast crop yields, diagnose plant disease and schedule irrigation systems rely on machine learning (ML) and deep learning (DL). Cost of training and deploying these models is weighed their accuracy. Two research threads have far developed largely in parallel. Green AI, which treats computational and energy efficiency as a first-class evaluation criterion alongside accuracy and Explainable AI(XAI) uses technology like Shapley additive explanations(SHAP), LIME and Grad-CAM to open the “black box” models so that farmers , and policymakers can act on their predictions.This paper synthesises recent literature (2017–2026) at the intersection of these threads as they apply to agriculture, covering crop-yield prediction, plant-disease detection, and soil- and irrigation-modelling studies that report either energy-efficiency measures or explainability analyses, or both. The synthesis identifies a specific and consistent gap: while explainability is now common in high-performing agricultural ML studies, almost none of them report the energy or carbon cost of training the underlying models, and the one purpose-built Green-AI agriculture framework identified in the review explicitly leaves explainability as future work rather than integrating it. Building on this synthesis, the paper proposes a unified Green-XAI methodological framework for agricultural ML that couples resource-appropriate model families – tree-based ensembles such as Random Forest, XGBoost and LightGBM, alongside a compact CNN architecture for imagery – with SHAP-based interpretability and continuous energy/carbon instrumentation across the training and inference lifecycle. The paper details the proposed dataset scope, the governing model equations, the SHAP computation, and an accuracy-versus-energy evaluation protocol, and positions the framework against a comparative synthesis of the performance figures reported by prior studies