Multimodal AI fusion for infrastructure resilience: real-time urban analytics framework aligned with SDG-9

Insufficient human capacity to manage flood risk, limited technical support, weak integrated planning processes, and institutional distortions further exacerbate these challenges. In this paper, we propose a multimodal AI fusion framework combining the power of Long-Short Term Memory (LSTM) and Graph Neural Networks (GNN) to model both temporal dynamics and spatial dependencies within streams of urban data. The architecture also includes a dynamic Resilience Scoring Index (RSI) that enables online anomaly detection and situational-awareness-based decision-making. Edge-AI processing units power instant sensor data intake, and decision dashboards deliver understandable city insights to make life easier for you. The method was thoroughly evaluated in three different cities: Singapore (rich in data), Chennai (with a paucity of data), and Rotterdam (resilience modeled) as a benchmark to understand the generalizability of the approach. The results consistently show that the LSTM+GNN hybrid model performs better than ARIMA, Random Forest, and unimodal deep networks, with a statistically significant improvement in F1 score (<i>p</i> < 0.05), and incurs only marginal performance degradation under noisy and incomplete data scenarios. Our work contributes to Sustainable Development Goal 9 (SDG-9) by creating scalable, evidence-based solutions for infrastructure planning and disaster risk reduction, providing a replicable framework for global smart city resilience initiatives.