AI
PyroAdapt: Adapting Wildfire Prediction under Spatial Heterogeneity and Temporal Shift
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 27/09/2026, 20:00:00 UTC
Briefing created: 29/09/2026, 04:19:39 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 29/09/2026, 04:18:24 UTC.
Briefings
PyroAdapt is a pretrain-retrieve-rank framework designed to adapt wildfire prediction models to spatial heterogeneity and temporal distribution shifts. Evaluated on California using 666 grid cells, the method improves daily average precision from 21.62% to 24.35-24.57% and Top5% recall from 18.70% to 22.11-22.79%. Under a fixed detection budget of 34 cells, selective ranking captures 344 additional positive cell-days. For fires in the top 5%/10%/20% of dry matter consumption, recall increases by 39.70/28.18/20.50 percentage points respectively. Rolling evaluations over Yosemite confirm gains persist under temporal distribution shift.
Key points
- PyroAdapt uses a pretrain-retrieve-rank framework to adapt wildfire prediction models to spatial heterogeneity and temporal distribution shifts.
- Selective ranking under a fixed daily detection budget of 34 cells captures 344 additional positive cell-days compared to direct ranking.
Context and limitations
Results are limited to California grid cells; temporal distribution shift was observed in rolling evaluations over Yosemite.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 29/09/2026, 04:19:39 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeicpj3itqqey7f7zswthrku64pisk5w75f3eil6nhrmevgbaawzpjq