Resilient IoT Architectures for Disaster Management: Spatiotemporal Imputation of Seismic Sensor Networks Using Cloud-Native Machine Learning
The deployment of decentralized, low-cost Internet of Things sensor networks has revolutionized emergency situation awareness and disaster management. However, during catastrophic events such as high-magnitude earthquakes, these networks are highly susceptible to nodal failures and intermittent connectivity, resulting in critical data loss when predictive accuracy is most vital. This paper proposes a cloud-native architectural framework utilizing Amazon Web Services to construct a highly resilient, serverless pipeline for disaster telemetry. By deploying asynchronous Python middleware integrated with K-Nearest Neighbors and eXtreme Gradient Boosting regression algorithms, the proposed system ingests high-frequency micro-seismic data, instantaneously identifying offline nodes and applying real-time spatiotemporal imputation to synthesize the missing data. Preliminary architectural evaluations demonstrate that moving data fusion and imputation logic to a distributed cloud infrastructure guarantees continuous emergency situation awareness, providing disaster management coordinators with a deterministic, scalable tool for maintaining early warning systems during severe network degradation.