Wildfire Spread Modeling & the Shrubwise Data Challenge
UC San Diego / SDSC · Societal Computing and Innovation Lab
LiDAR-derived ground truth and independent verification for a community machine-learning challenge on shrub classification, alongside work on reproducible fire-behavior model implementations.
- Role
- Graduate Student Researcher; ground-truth production, independent verification, and team mentorship
- Dates
- Feb 2026 – Present
- Advisor
- Dr. İlkay Altıntaş
- Status
- AGU 2026 abstract accepted; workshop paper under review
Overview
The Wildfire Commons Shrubwise Data Challenge asked teams to build machine learning pipelines that label shrubs from aerial imagery. The results depend on the quality of the ground truth used for scoring, and on the verification that follows.
My contribution
- Produced the LiDAR-derived ground truth the challenge was evaluated against.
- Mentored the winning team from result validation through to an accepted AGU contribution.
- Independently verified the winning team's NAIP-imagery shrub-classification pipeline, confirming their reported results and method quality against the challenge's evaluation criteria.
I am also a contributing author on related work from the lab on reproducible fire behavior model implementations and standardized AI-ready datasets for operational wildfire spread modeling.
Output
- Karabas S, Ersus B, Pearce A, Lee L, Ramonetti Vega P, Altintas I, Floca M, Stirm C. “Wildfire Commons Shrubwise Data Challenge: Outcomes from a designed machine learning pipeline to label shrubs.” Abstract accepted, AGU Fall Meeting 2026, Wildfire Commons Session (NH056).
- Floca M, …, Pearce A, …, Altintas I. “Advancing Operational Wildfire Spread Modeling through Reproducible Fire Behavior Model Implementations and Standardized AI-Ready Datasets.” Under review, IEEE BigData 2026, Workshop on Trustworthy AI Pipelines (WS#60).