Answers are grounded in this paper and cite their sources.
Forest fires are a major ecological disturbance that threatens biodiversity, carbon storage, and human safety, and quantifying their response to climate change is critical for adaptive management. In this study, we integrated 19,632 historical fire records (2000–2024) across 985 Chinese counties with meteorological, forest, geographic, and socioeconomic covariates to develop multi-target machine learning models and to project county-level fire probability under four SSP scenarios (SSP1, SSP2, SSP3, and SSP5) for 2030–2060. For heterogeneous distributions of these fire-related outcomes, we adopted a task-specific heterogeneous ensemble framework in which fire occurrence, record count, burned area, and fire duration were each assigned a model, with selection based on temporally separated validation and outcome-appropriate metrics. The satellite-enhanced XGBoost burned-area model and LightGBM fire-duration model achieved R2 values of 0.767 and 0.696, respectively, whereas the count model showed limited point-prediction performance in the panel data. SHAP analysis identified seasonal cyclic variables, historical fire legacy, and geographic coordinates as the most influential predictors, indicating that the fitted model relied more strongly on seasonal rhythm and long-term fire-probability information than on transient meteorological fluctuations. These represent associative predictive contributions rather than causal relationships. Our projections showed a consistent decline in national mean fire probability by 13–30% from 2030 to 2060 across all scenarios, with SSP3 consistently yielding the highest fire probability. The spatial pattern remained stable, with southwestern mountainous areas persisting as high-risk hotspots, and March remained the peak fire month across all scenario–year combinations. Within the specified adjustment model, sensitivity analysis showed positive associations of temperature and population change with projected fire probability and a negative association of rainfall change, while baseline fire probability contributed most to cross-county variation. These findings provide scientific support for regionally differentiated, climate-adaptive forest fire management strategies, with priority resources allocated to persistent high-risk counties and the spring fire season.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: