Estimating Hurricane Wind Speeds with Machine Learning

Better estimates of hurricane wind speeds can lead to better decisions around evacuations and general hurricane response planning, saving both lives and property. Hurricane windspeed estimates are currently made using the manual Dvorak technique. The National Hurricane Center releases them every three to six hours. Artificial intelligence (AI) experts with the IMPACT team at NASA’s Marshall Space Flight Center and Development Seed created the Deep Learning-Based Hurricane Intensity Estimator to automate this process.

This process takes infrared imagery as input and runs a machine learning algorithm to estimate the windspeed. These estimates are more reliable than human classifiers and can be run every 15 minutes using near real-time images from a GOES-16 WMS server. At this frequency, the model is able to see short-term changes to wind speeds like those that occur during eyewall replacement.

The underlying infrastructure makes use of AWS to provide a stable production system that is easy to deploy and maintain. The stack is deployed via AWS CloudFormation and leverages the existing open source NASA project called Cumulus for many out-of-the-box features, which orchestrate earth observation workflows.

The AWS Technology Behind Cumulus

Cumulus is a cloud-based framework for data ingest, archiving, distribution, and management of earth observation data. In this deployment, Amazon CloudWatch rules kick off a new estimation process at the desired frequency. The machine learning algorithm runs on an Amazon Elastic Container Service (Amazon ECS) cluster, and the results are stored in Amazon Relational Database Service (Amazon RDS). They are then served to the front-end application via the API gateway. Each process is run as a step function for easier debugging and logging.

“This is an excellent example of how combining free and open data with the latest in machine learning technology and cloud processing makes complex information actionable. NASA’s Cumulus software infrastructure and the cloud platform open up new opportunities for researchers, developers, and the general public,” said Kevin Murphy, Earth Science Data Systems Program Executive, NASA.

Get Started Forecasting and Planning for Natural Disasters

Explore the site or the open API and see how the continued development of machine learning-driven platforms can give us more tools for forecasting and planning for natural disaster scenarios.