Detecting Storm Intensity from Satellite Imagery Using Machine Learning

The Deep Learning-based Hurricane Intensity Estimator is an experimental portal that uses machine learning techniques to analyze spatial patterns in infrared satellite imagery in order to predict tropical cyclone intensity.  The developers of the portal used an adapted convolutional neural network (CNN) to train a model based on 200,000+ images from the Naval Research Laboratory Tropical Cyclone Archive which were attributed with wind speed pulled from the NOAA National Hurricane Center (NHC) Best Track Database. More about this methodology is outlined in the IEEE article: Tropical Cyclone Intensity Estimation Using a Deep Convolutional Neural Network.

An alpha release, the portal is intended to be an exploration of the efficiency of using machine learning in lieu of using the Dvorak technique which is currently the standard for estimating tropical cyclone intensity from visible and infrared satellite imagery.  The developers of the portal explain:

Essentially the Dvorak technique uses subjective pattern recognition of features in visible and infrared imagery, and then applies objective rules for relating those patterns to the tropical cyclone intensity.

The success of the Dvorak technique proves that spatial patterns in infrared satellite imagery strongly relate to tropical cyclone intensity. But because parts of the Dvorak technique are subjective, two well-trained analysts can derive different intensity estimates. To help overcome the challenges, we have launched the Deep Learning-based Hurricane Intensity Estimator.

The results of the model are displayed on a web map.  Active storm systems are highlighted on the map.  Clicking on a storm brings up graphical data that plots the wind speed and category predictions made along the storm path along with NHC storm outlook data. There is also the option to click on the download button to retrieve a CSV file with a timestamp, the lat/long of the storm’s location, the wind prediction, and a link to the GOES 16 image from that time period.


See also

Views of Hurricane Florence from space

Like this article and want more?

Enter your email to receive the weekly GIS Lounge newsletter: