A Unified Deep Learning Architecture Combining Earth Observation Satellites, Dynamic IoT Sensors, and Historical Climate Data for Real-Time Wildfire and Flood Prediction

• Ada John

  ladoke Akintola university of technology

  Author

Wildfires and floods represent escalating threats under anthropogenic climate change, yet existing prediction systems remain fragmented, relying on unimodal data streams that fail to capture the coupled non-linear dynamics of these hazards. This study addresses this gap by designing, implementing, and validating a unified deep learning architecture that integrates Earth observation satellite imagery, dynamic Internet of Things (IoT) sensor telemetry, and historical climate records for concurrent wildfire and flood prediction. The research employs a design-based methodology combining retrospective analysis of multi-source environmental datasets (2015–2025) with prospective simulation across fire-prone and flood-vulnerable regions. The proposed architecture achieves 89.4% overall prediction accuracy, with a 6.2-hour mean lead time for flood events and 12.8-hour lead time for wildfire ignition risk, substantially outperforming unimodal baselines. Feature importance analysis identifies soil moisture, land surface temperature, and antecedent precipitation as dominant predictors. The framework demonstrates operational viability through edge-compatible inference and provides a replicable blueprint for integrated multi-hazard early warning systems. These findings advance both disaster science and practical emergency management by establishing a scalable, data-driven foundation for climate resilience infrastructure.

• pdf

Copyright (c) 2026 Ada John (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.