Abstract
Timely flood response requires combining complementary but widely heterogeneous Earth Observation (EO) data that strongly differ in spatial support, temporal availability, and uncertainty. We present an event-driven multimodal fusion framework that integrates satellite flood data, Unmanned Aerial Vehicle (UAV) imagery, geosocial media analytics, and hydrodynamic predictions to provide accurate updated estimates of the flood state. It provides two outputs: a Flood Probability Map (FPM) that reflects the current flood status, implemented as a Bayesian occupancy grid updated asynchronously as new data arrive; and an Object Presence Map (OPM) that tracks persons and vehicles using Kalman filtering. The FPM fuses satellite observations and broad-scale hydrodynamic predictions with UAV-derived flood and geosocial measurements to produce a consistent probabilistic flood-extent estimate. The OPM produces smooth, temporally coherent trajectories that remain consistent with the map resolution under intermittent UAV-based detections. The framework is demonstrated on the July 2021 Ahr Valley flood in Germany using UAV and satellite EO data complemented with hydrodynamic predictions and geosocial data. The framework supports continuous event-driven situational awareness, enabling joint reasoning about inundation, exposure, and accessibility during flood response. It operates efficiently online, transforms all inputs into a common probabilistic representation, and accommodates different spatial and temporal resolutions, supporting extensibility to additional sources.