Normalized claim
Prototype time: 4 weeks decrease
“It took just four weeks from spinning up the environment to the final set of initial experiments.”
Met Office automates maritime Shipping Forecast text generation with Amazon Nova Foundation Models
Use case typePredictive decision supportUpdated Jun 8, 2026
The Met Office built a prototype to turn raw weather data into readable maritime forecasts using Amazon Nova Foundation Models. The project explores automated text generation for the Shipping Forecast and broader weather services to improve scalability and consistency.
- Organization
- Met Office
- Industry
- Public Sector
- Location
- United Kingdom
- Published
- June 2026
Reported outcomes
2,920 hours/year
meteorologist hours spent annually on marine warningsTime & speed
4 weeksprototype time+62%LLM accuracy for complete forecast generation+52%VLM accuracy for data-to-text conversion+83%gale warning accuracy
Strategic outcomes
Scale & capacityScalable pattern for text generationNew product / capabilityAutomated maritime forecast text generation
Why do we believe this?Outcome claims, sources, and evidence checks
Normalized claim
LLM accuracy for complete forecast generation: 62% increase
“demonstrated 62 percent LLM accuracy for complete forecast generation”
Normalized claim
VLM accuracy for data-to-text conversion: 52% increase
“52 percent VLM accuracy in data to text conversion”
Normalized claim
Gale warning accuracy: 83% increase
“it saw 83 percent gale warning accuracy compared with forecasters’ observations”
Normalized claim
Meteorologist hours spent annually on marine warnings: 2,920 hours/year decrease
“expert meteorologists spend approximately 2,920 hours a year”
Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks
- Customer
- Met Office
- Provider
- AWS
- Maturity
- Exploring
- Linked source
- AWS Customer Story
Meteorologists spend thousands of hours each year manually evaluating weather data and condensing it into strict broadcast text
Customer identity supportedSource describes one deploymentMaturity supported
Primary read
Use case focus
Showing 3 of 3
- 1Forecast generation automation
- 2Multimodal weather AI
- 3Public sector decision support
- Meteorologists spend thousands of hours each year manually evaluating weather data and condensing it into strict broadcast text.
- The organization needed a way to augment and automate textual forecast generation while maintaining high accuracy and consistency.
- It wanted to explore whether AI could free experts to focus on higher-value analysis and quality assurance.
- The Met Office collaborated with the AWS Specialist Prototyping Team to build a prototype using Amazon Nova Foundation Models via Amazon Bedrock.
- Technical experiments included an LLM data-to-text workflow and a VLM approach for meteorological data.
- The team used maritime-specific prompt engineering and fine-tuned a vision model with distributed training to process gridded weather data.
Technologies
- The prototype was completed in about four weeks.
- Benchmarking showed 62% LLM accuracy for complete forecast generation, 52% VLM accuracy for data-to-text conversion and 83% gale warning accuracy.
- The project demonstrated a scalable pattern that could be extended across nearly 300 products and services that transform grid data into text.
Architecture
Prototype for maritime forecast generation built with Amazon Nova Foundation Models via Amazon Bedrock, including LLM and VLM experiments, maritime-specific prompt engineering, and fine-tuning for vision capabilities with distributed training.
Implementation partners1
Sources & evidence1
Evidence: Medium65/100Evidence strength
- Customer explicitly identified
- Deployment status explicitly supported
- Primary source available
- Quantified outcome available
- Technical implementation details available
Type: Customer StoryPublished: Jun 8, 2026Publisher: AWSEvidence: PrimaryConfidence: High
AI-generated summary. Verify important details with the linked sources before relying on this case.
Explore related AI use cases
Customer evidence
Provider evidence
Market context
AI patterns
Browse the catalog
Was this useful?
Community
Comments
Loading comments...