Natural hazards and disasters disrupt lives and livelihoods worldwide, with lasting societal and environmental consequences. Emerging Earth observation assets—governmental and commercial constellations of optical, SAR, and hyperspectral sensors—now deliver hazard-relevant data at unprecedented spatial, spectral, and temporal resolution. Meanwhile, advances in machine learning, from physics-informed neural networks to geospatial foundation models, are transforming how we detect, forecast, and attribute hazards and their impacts. This session brings these two revolutions into conversation to accelerate hazard and disaster science.

We invite contributions pairing AI/ML methods with novel or underutilized Earth observation datasets across extreme weather (heatwaves, floods, droughts, hurricanes), landslides, earthquakes, wildfires, and compound hazards. Topics include foundation and generative models; multi-sensor, multi-modal data fusion; near-real-time detection and nowcasting; probabilistic forecasting and uncertainty quantification; damage assessment; and rapid-response products. We especially welcome work on rare-event generalization, benchmark datasets, open science, explainability, and equity.

• B - Biogeosciences
• EP - Earth and Planetary Surface Processes
• NH - Natural Hazards
• SY - Science and Society