Multi-hazard early warning systems increasingly rely on the technical core of modern disaster risk reduction. They utilise heterogeneous data streams, including earth observations, in-situ sensor networks, process-based models, AI and data-driven approaches, and increasingly citizen-reported and computer-vision-derived observations. These datasets need to be transformed into forecasts that are both scientifically robust and operationally actionable, ranging from minutes to hours for rapid-onset hazards, such as floods, landslides, and cyclones, to months and seasons for slow-onset hazards such as drought. Moreover, natural hazards increasingly occur concurrently, consecutively, or in combination, creating multi-hazards and cascading risks. Operational early warning systems, however, often focus on single hazards. In addition, warning thresholds and dissemination pathways frequently fail to translate warnings into effective community and institutional response, creating a persistent know-do gap that cannot be closed through additional sensing and technical sophistication alone.

This session welcomes contributions across the full arc of multi-hazard early warning, including: (i) AI/ML and process-based modelling approaches for multi-hazard forecasting and nowcasting (e.g., floods, droughts, landslides, and cascading hazards); (ii) data fusion architectures combining satellite EO, real-time sensor networks, and computer vision for hazard detection; (iii) participatory and citizen-science methodologies, local and traditional knowledge integration, and including living labs and serious games for co-creating warnings directly with at-risk communities; (iv) dashboard and decision-support system design that translate model outputs into actionable, trusted, inclusive, and often multilingual alerts for operational use; and (v) validation and lessons from deployed or piloted systems at the interface between technical performance and community uptake.

We particularly encourage submissions presenting multi-hazard forecasting approaches, warning thresholds, and deployed or piloted systems, including honest lessons on where technical sophistication and last-mile trust have, and have not, come together to support effective action.