Modeling Delayed Mortality of Fire-Damaged Pines in Korea

This study aimed to develop and compare prognostic models (logistic regression, Cox proportional hazards, and random forest) to assess delayed mortality in pine (Pinus densiflora) following fires in Korea. Data from a 72-month monitoring of 734 trees across four fire areas were used, accounting for 19 variables: tree size, fire severity, multispectral indices, topography, and bioclimatic variables. Key predictors of mortality included diameter at breast height (DBH), bark scorch index (BSI), delta normalized burn ratio (dNBR), slope, topographic wetness index (TWI), precipitation of the warmest quarter, temperature seasonality and isothermality. The key results indicate that the random forest model was the most effective (AUC = 0.924; sensitivity = 0.892) in identifying trees at high risk of mortality. These results suggest that nonlinear approaches are effective for predicting delayed mortality in fire-damaged pines and can support rapid decision-making in post-fire forest management and restoration under increasing wildfire risk.