Wu et al. (2026) Prediction of Forest Fire Occurrence Risk in Heilongjiang Province Under Future Climate Change

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Forests
• Year: 2026
• Date: 2026-03-26
• Authors: Zechuan Wu, Houchen Li, Mingze Li, Xintai Ma, Yuan Zhou, Yuan Tian, Ying Quan, Jianyang Liu
• DOI: 10.3390/f17040414

Research Groups

Not provided in the paper text.

Short Summary

This study quantified the multi-source drivers of forest fire occurrence in Heilongjiang Province and developed a long-term fire risk forecast using a Deep Neural Network with Residual Connections (ResDNN), which achieved 85.6% accuracy and was applied with CMIP6 projections to map future fire probability from 2030 to 2070.

Objective

• To systematically assess the relative contributions of multi-source factors (topography, vegetation, and meteorological conditions) to forest fire occurrence in Heilongjiang Province.
• To compare the predictive performance of Deep Neural Network with Residual Connections (ResDNN), Artificial Neural Network (ANN), and Support Vector Machine (SVM) models for forest fire prediction.
• To conduct long-term (2030-2070) spatiotemporal forest fire risk forecasting under future climate scenarios using the best-performing model in conjunction with CMIP6 climate projections.

Study Configuration

• Spatial Scale: Forested regions of Heilongjiang Province.
• Temporal Scale: Historical fire records for model training and validation; 2030 to 2070 for future fire risk projections.

Methodology and Data

• Models used: Deep Neural Network with Residual Connections (ResDNN), Artificial Neural Network (ANN), Support Vector Machine (SVM). CMIP6 multi-model climate projections.
• Data sources: Historical fire records, topographic data, vegetation data, meteorological conditions data, CMIP6 climate projections.

Main Results

• The Deep Neural Network with Residual Connections (ResDNN) model achieved the highest predictive accuracy (85.6%) for forest fire occurrence, outperforming Artificial Neural Network (ANN) and Support Vector Machine (SVM).
• ResDNN demonstrated superior capability in capturing complex feature interactions due to its robust nonlinear mapping.
• Long-term spatiotemporal probability maps of forest fire occurrence from 2030 to 2070 were generated using the ResDNN model combined with CMIP6 climate projections, providing an intuitive representation of future fire-risk trajectories under climate change.

Contributions

• Systematic assessment of multi-source factors (topography, vegetation, meteorological conditions) influencing forest fire occurrence in Heilongjiang Province.
• Comparative analysis of ResDNN, ANN, and SVM models for fire prediction, establishing ResDNN as the most accurate model for the region.
• Development and application of a robust methodology for long-term (2030-2070) spatiotemporal forest fire risk forecasting under climate change using ResDNN and CMIP6 projections.
• Provides scientific support for regional fire prevention, monitoring, early-warning systems, and forest management strategies in the context of climate change.

Funding

Not provided in the paper text.

Citation

@article{Wu2026Prediction,
  author = {Wu, Zechuan and Li, Houchen and Li, Mingze and Ma, Xintai and Zhou, Yuan and Tian, Yuan and Quan, Ying and Liu, Jianyang},
  title = {Prediction of Forest Fire Occurrence Risk in Heilongjiang Province Under Future Climate Change},
  journal = {Forests},
  year = {2026},
  doi = {10.3390/f17040414},
  url = {https://doi.org/10.3390/f17040414}
}