Coastal communities face mounting threats from extreme weather events like hurricanes, yet current prediction systems lack granular risk assessment capabilities for critical infrastructure

The National Oceanic and Atmospheric Administration awarded Silurian AI Inc. $189,149 under the NOAA Small Business Innovation Research (SBIR) Program on September 1, 2026.

The Phase I project develops post-trained foundation models to enhance coastal hazard forecasting and hurricane prediction. Silurian AI will create a validation framework, build AI-ready multimodal datasets, and implement post-training strategies for the GFT foundation model to predict hurricane trajectories, intensity, precipitation, and flooding with superior accuracy and spatial resolution compared to existing operational models. The anticipated deliverable is a post-trained foundation model capable of generating dynamic, sub-kilometer-level risk assessment maps that incorporate human infrastructure and population density, along with documented strategies for scalable model adaptation.

Performance occurs in Kirkland, Washington, with a period of performance through February 26, 2027. The SBIR program provides federal assistance to US-based small business concerns in proof of concept, research and development, and commercialization phases. Phase I grants are normally limited to six months; proposals are evaluated competitively on scientific merit and technical approach, innovation, commercial and societal impacts, and investigator qualifications.

This sbir phase i project will advance coastal hazard forecasting by leveraging recent breakthroughs in earth system modeling using state-of-the-art earth system foundation models. Building on our prior success with models like climax and aurora, we propose to enhance our latest foundation model, GFT, through innovative post-training, so that these models can predict hurricane trajectories and impacts with superior accuracy and resolution compared to numerical approaches, while operating orders of magnitude faster. This enables rapid generation of calibrated ensemble forecasts to produce dynamic, high-resolution risk maps that evolve in real-time. As part of the topic extreme events and cascading hazards, phase i of this project focuses on developing a robust validation framework, creating AI-ready multimodal datasets for these tasks, and implementing and evaluating post-training strategies. We aim to demonstrate significant improvements in prediction accuracy, spatial resolution, and lead time compared to existing operational models, specifically targeting the accurate prediction of hurricane intensity, precipitation, and flooding in coastal areas. Anticipated results include a post-trained foundation model capable of generating dynamic, sub-kilometer-level risk assessment maps that incorporate human infrastructure and population density, and documented strategies for scalable model adaptation. Successful completion of this project will establish the feasibility of using post-trained foundation models for enhanced coastal hazard prediction, paving the way for a commercially viable, turnkey solution for customized risk assessment. This technology will empower emergency managers, port authorities, reinsurance companies, and other stakeholders with unprecedented precision in understanding and mitigating coastal risks, ultimately enhancing community resilience and supporting the sustainable development of coastal regions. This technology advances NOAA's strategic priority of building resilient coastal communities while providing new AI-based tools in the operational management of extreme events.