PyroAdapt lifts California wildfire prediction, boosting extreme-fire recall by 39.7 points

UIUC-CS · hf · 2026-09-29

A UIUC team introduces PyroAdapt, a pretrain–retrieve–rank framework for adapting wildfire occurrence models to spatial heterogeneity and temporal distribution shift. It retrieves historically similar locations and fine-tunes a pretrained model with risk-ranking objectives conditioned on terrain, ecoregion embeddings, and local fire rates.

Key results over California (666 grid cells)

Original post →

More from Research

Research channel →