Rainy Days
1st Runner-Up · NASA Space AppsNASA Space Apps 2025 (Toronto) 1st Runner-Up — predicting cloud-seeding potential from NASA MODIS satellite data with a multi-output regressor.
- Python
- scikit-learn
- NASA MODIS
- Geospatial
- Docker
- Flask
Rainy Days was our entry to the NASA Space Apps Challenge 2025 (Toronto), where it placed 1st Runner-Up. The project predicts cloud-seeding potential across a geographic grid using NASA satellite data, packaged as a deployed web app.
What we built
- Mined and processed NASA MODIS satellite data at 0.25° grid resolution, engineering geospatial and atmospheric features across multiple meteorological targets.
- Trained and evaluated a MultiOutput Regressor that reached 75.2% accuracy on unseen test data, covering the full pipeline from raw satellite ingestion through evaluation.
- Shipped it as a containerized web stack — an nginx reverse proxy with TLS, a Node.js SPA front end, and a Python API — orchestrated with Docker Compose and deployed live at rainy-days.earth.
Role
A team project built under hackathon time pressure. My focus was the data and modeling pipeline — turning raw MODIS grids into engineered features and a trained multi-target model — alongside getting the stack deployed behind HTTPS.
Built in ~48 hours for NASA Space Apps. Code lives on the team’s repo, DeltaTecs/Rainy-Days.