What it does
Allows images to be run against pre-trained Azure ML model and weather prediction database to enhance tornado and lightning detection from still images.
Stormie is intended as a prototype for "shallow storm" detection - these are high velocity but underreported meteorological events that are too small to be seen by radar or satellite; or are too short to alert any numerical systems.
The intention of creating Stormie is to provide better models to detect shallow storms and create more microclimate research.
How I built it
Method: Stormie lets an anonymous user add image data to an interactive climate model. The registered image is added to pre-trained database and sent to the Azure's ML model to calculate the chances that the given image is indeed a shallow storm.
Challenges I ran into
Accomplishments that I'm proud of
- Training an ML model using some data we collected from various sources - Making the Azure API query - Creating a responsive climate model and map that updates - Transitioning from a Flask backend to Firebase hosting
What I learned
- Flashy documentation != straightforward to use - Azure is cool!
What's next for Stormie
- Bringing the interface into the 2000s :( - New ML model trained with more images - Ability to identify shallow storms from publicly available data automatically (street cameras, news footage) - Heatmap of shallow storm activity - integrate with existing weather data sources - Real time updates of storm activity