A federated explainable artificial intelligence architecture for multimodal disaster intelligence and response coordination

Abstract

Natural disasters demand rapid synthesis of fragmented, heterogeneous data often distributed across independently governed organizations. This paper introduces DisasterAI, a federated, cloud-native, and multi-modal AI architecture for disaster intelligence. Unlike centralized analytics platforms, DisasterAI emphasizes federated data ingestion, modular adapters, and the alignment of derived intelligence—enabling cross-agency coordination without requiring raw data consolidation. The architecture integrates satellite imagery (optical and SAR), IoT sensor streams, public alerts, and contextual civic data into a unified, explainable decision-support pipeline. DisasterAI addresses critical challenges in disaster intelligence, including spatial-temporal alignment of asynchronous data, modality-aware fusion, uncertainty representation, and fairness-aware decision-making. Designed to support the entire disaster lifecycle—from early warning and forecast-informed risk analysis to impact assessment and response prioritization—it operates within a human-in-the-loop decision-support context. A scenario-driven evaluation illustrates how DisasterAI enhances coordination effectiveness, improves prioritization consistency, and maintains robustness under partial data availability. By focusing on architectural behavior and operational integration rather than algorithmic benchmarking, this work positions federated, explainable AI systems as a viable foundation for scalable and reliable disaster-response frameworks in complex, data-constrained environments.