GRACE-1

Inspiration

Emergency managers need faster, more accurate impact assessments. Existing tools are slow and manual.

What it does

Submit a flood or wildfire scenario. GRACE-1 runs a physics-based simulation and returns structured impact data — structures, population, infrastructure — on an interactive map with an AI assistant for navigation and methodology questions.

How we built it

AWS serverless pipeline (API Gateway → Step Functions → ECS) with SFINCS for flood and Rothermel for wildfire. Live data from USGS, 3DEP, NLCD, and Census. React + Leaflet frontend, Claude Haiku via Bedrock for chat. Deployed via SAM and CloudFront.

Challenges we ran into

Coordinating VPC networking, ECS task isolation, and CORS across API Gateway and S3 presigned URLs. Several issues only surfaced after reading raw access logs.

Accomplishments that we're proud of

End-to-end flood assessment for the 2007 Chehalis event in under 5 minutes, pulling live geospatial data from six sources.

What we learned

Ambiguous errors need a diagnostic pass before a fix attempt. Integration failures tend to occur at layer boundaries, not inside the components themselves.

What's next for GRACE-1

Seismic and drought engines, multi-hazard coupling, and probabilistic scenario ensembles for regional planning.

Built With

• amazon-web-services