Time-resolved moving‐window machine learning modeling for human‐caused wildfire ignitions from 2001 to 2020 in California
Environmental Research Letters · 28.05.2026 · 21 (10) · 104003
DOI: 10.1088/1748-9326/ae692d ↗Kurz gefasst
Fachartikel in der Zeitschrift Environmental Research Letters, online erschienen am 15.05.2026. Verfasst von Zhijiong Cao und 3 weiteren Autor:innen. Die Arbeit nennt 81 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Human-caused wildfire ignitions have become increasingly important drivers of fire activity in California during the 21st century. This study applies a 6 year moving-window XGBoost framework to model monthly wildfire ignition probabilities across California for water years 2001–2020. The model shows strong predictive skill, with average AUC-receiver operating characteristic and AUC-precision–recall gain of 0.84 and 0.96, respectively, and broadly reproduces both the spatial distribution and seasonal cycle of observed ignitions, as well as their interannual fluctuations (Spearman’s ρ = 0.785, p < 0.001). Feature importance analysis identifies road density and maximum temperature as dominant predictors, indicating the growing influence of human infrastructure while confirming the continuing importance of weather-related factors. Further experiments reveal that interannual weather anomalies strongly affect wildfire ignition probability during the wet and early fire seasons, whereas anthropogenic drivers dominate during the peak fire season. The results show that such a temporally adaptive machine-learning framework has the potential to improve ignition prevention strategies in a warming, increasingly human-influenced fire regime.
Zitieren
APA
Cao, Z., Xue, Y., Ye, F., & Gillespie, T. W. (2026). Time-resolved moving‐window machine learning modeling for human‐caused wildfire ignitions from 2001 to 2020 in California. Environmental Research Letters, 21(10), 104003. https://doi.org/10.1088/1748-9326/ae692d
Vancouver
Cao Z, Xue Y, Ye F, Gillespie TW. Time-resolved moving‐window machine learning modeling for human‐caused wildfire ignitions from 2001 to 2020 in California. Environmental Research Letters. 2026;21(10):104003. doi:10.1088/1748-9326/ae692d
BibTeX
@article{Cao2026,
author = {Cao, Zhijiong and Xue, Yongkang and Ye, Fangshu and Gillespie, Thomas W},
title = {Time-resolved moving‐window machine learning modeling for human‐caused wildfire ignitions from 2001 to 2020 in California},
journal = {Environmental Research Letters},
year = {2026},
volume = {21},
number = {10},
pages = {104003},
doi = {10.1088/1748-9326/ae692d}
}
EUVIMED