Explainable Artificial Intelligence (XAI) is essential for enhancing the transparency and trustworthiness of deep learning models, particularly in engineering and scientific domains where interpretability is critical. This study proposes a Shapley-Additive-Explanations (SHAP)-based a-posteriori interpretability framework specifically designed for Graph Neural Networks (GNNs) in Computational Fluid Dynamics (CFD)-based airflow modeling, with a focus on optimizing opaque ventilated facades (OVFs). By integrating graph structural dependencies into SHAP computations, the framework ensures that feature attributions accurately reflect hierarchical aerodynamic interactions, addressing the limitations of traditional SHAP methods that assume independent features. To further assess interpretability, a comparative evaluation is conducted against other a-posteriori techniques, including LIME, Integrated Gradients (IG), and DeepSHAP, highlighting SHAP’s superior attribution consistency and stability in airflow simulations. Computational efficiency is enhanced through dimensionality reduction and GPU acceleration, enabling scalable SHAP computations for large-scale CFD-driven design processes. The reliability of SHAP-based attributions is validated by comparing them with high-fidelity CFD simulations, ensuring alignment with fundamental aerodynamic principles. Results demonstrate that the proposed framework significantly improves the interpretability of GNN-based CFD surrogates, providing actionable insights into how OVF design parameters influence airflow dynamics. By bridging the gap between XAI theory and real-world CFD applications, this research establishes a computationally efficient and scientifically grounded approach for AI-driven sustainable building design.