FloodSOS: An Integrated AI-Enabled Flood Early Warning,Real-Time SOS, and Rescue Coordination System

Abstract

Floods require not only accurate early warning but also resilient field communication and rescue coor dination. This paper presents FloodSOS, an integrated disaster-response platform that couples AI-based f lood forecasting with real-time SOS ingestion, shelter recommendation, flood-aware routing, and operational dashboards. FloodSOS combines multi-source terrain, hydrological, infrastructure, population, storm, and weather data from DEM, OpenStreetMap, Open-Meteo, Visual Crossing, Japan Meteorological Agency tracks, and Google Earth Engine labels. The AI core uses LightGBM for commune-scale 7-day flood forecasting and real-time point-level risk estimation, PyTorch MLP for SOS urgency scoring, and an OSMnx/NetworkX routing engine that penalizes or blocks inundated road segments. A stateless FastAPI microservice layer interoperates with an Express.js/MongoDB backend, while a Flutter mobile client enables GPS tracking, voice-based SOS submission, map visualization, and offline caching for shelters and hotline data. Because the current study formalizes a documented prototype rather than a completed field trial, the evaluation empha sizes prototype-level comparative analysis against warning-only, navigation-only, and cloud-only emergency systems. The results show that FloodSOS provides broader operational coverage and greater resilience under degraded connectivity than conventional single-function solutions.