- Role
- Finalist
- Region
- Bay Area
- Pronouns
- she/her/hers
Jasmine Wang
ea adipisicing consectetur ex in sit enim duis consequat exercitation officia sit fugiat eiusmod ex cupidatat in laborum dolore amet magna sunt culpa ea pariatur consectetur enim ex consequat laborum laborum irure exercitation culpa non amet
Project
Wildfire Prediction and Detection: Custom Data Pipeline, Transformer Model, and Web App integration
Project #7205
The recent LA wildfires and Canada’s record-breaking 2023/2024 fire seasons highlight the urgent need for improved wildfire prediction and detection. Existing industrial systems rely on outdated models that fail to address the severity of today’s wildfires. Despite the need for updates, research progress is hindered by two key challenges: the lack of scalable wildfire datasets to support modern ML models, and the need to adapt these models to wildfire-specific characteristics. To address this, a custom data pipeline was built to generate cost-effective, scalable wildfire datasets. A transformer model was applied to capture contextual patterns, and a custom designed loss function to target critical wildfire metrics(e.g. reducing fire misses). Compared to traditional models, my detection model improves recall by 11%, and prediction shows a 61% boost. To ensure practical value, a web app was built that demonstrates the real-world application of the models—helping communities prepare, act earlier, and stay safer.
- Challenge
- Environment and Climate Change
- Category
- Intermediate
- Type
- Innovation