Survival analysis and classification methods for forest fire size.

Area of Science:

• Forestry and Environmental Science
• Statistical Modeling
• Wildland Fire Management

Background:

• Wildland fire size distribution is influenced by complex interactions between weather, fuel availability, and suppression efforts.
• Understanding these factors is crucial for effective wildfire management and risk assessment.

Purpose of the Study:

• To apply survival analysis to quantify the impact of weather, fuel type, and initial attack on lightning-caused wildfire sizes in Alberta, Canada.
• To develop a statistical classifier to predict potential fire growth after initial assessment.

Main Methods:

• Utilized survival analysis, specifically the Cox proportional hazards model, to analyze fire size dynamics.
• Developed a logistic regression classifier to predict fire growth, comparing it with alternative models.
• Identified key covariates: fire weather index, fuel type, and initial attack method.

Main Results:

• Fire weather and fuel type were highly significant predictors of fire size, aligning with established fire behavior principles.
• The statistical classifier using logistic regression outperformed alternative models in predicting fire growth.
• Initial attack method did not show statistical significance, suggesting potential reverse causality in resource allocation.

Conclusions:

• Survival analysis provides a robust framework for understanding factors influencing wildland fire size.
• Fire weather and fuel type are critical determinants of fire growth.
• Further research with larger datasets is needed to accurately estimate the unbiased effects of fire suppression strategies.