Scientific question
How can lava-fountain regions be identified consistently in video so that eruption parameters can be extracted quantitatively from field footage?
Why it matters
The pipeline is intended to enable quantitative extraction of eruption parameters from volcanic video recorded under variable lighting, viewing geometry, and field conditions.
Data and materials
- Kīlauea lava-fountain video, including footage from Episodes 49 and 50
- Video metadata, camera settings, and viewing-condition notes
- Observational field notes
- Manually generated masks and selected frames
Methods
- U-Net segmentation models
- Frame selection and mask generation
- Metadata tracking and model-ready dataset organization
- Field videography and documentation
My contribution
- Developing the computer-vision pipeline
- Trained U-Net segmentation models to identify lava-fountain regions across changing lighting, viewing geometry, and field conditions
- Built a Python-based labeling and dataset-preparation system
- Collected field footage for Kīlauea Episodes 49 and 50
- Documented camera settings, viewing conditions, and observational notes
Current results
U-Net models have been trained to identify lava-fountain regions across changing lighting, viewing geometry, and field conditions.
Research outputs
Associated publications and software are in preparation.
