Explainable Spatiotemporal Multi-Sensor Fusion for Urban Environmental Sensing

Urban environmental sensing is a critical component of smart city infrastructure, yet current deep learning solutions remain fragmented. A primary motivation for this study is the observation that temporal modeling, multi-sensor fusion, and explainability are often developed independently, which limits predictive reliability and restricts real-world deployment. To address this lack of integration, a novel conceptual framework named ST-MFXAI (SpatioTemporal Multi-modal Fusion with Explainable AI) is proposed. This architecture unifies graph-based spatial modeling, temporal learning, and attention-driven fusion with embedded interpretability. Significant findings from the analysis highlight that current limitations stem from the absence of unified frameworks capable of jointly addressing heterogeneous data dynamics and model transparency. By bridging the gap between complex modeling and interpretative clarity, this study contributes a comprehensive synthesis of the literature and introduces a deployment-oriented perspective. The proposed framework and review find direct applications in real-time air quality monitoring, urban resource management, and the development of resilient, transparent sensing systems for future smart city environments.