Predicting fire-induced individual tree mortality at the landscape level using fire intensity and airborne laser scanning data

The prediction of fire-induced tree mortality is important for evaluating potential timber volume losses, replanting costs, and for assessing how mortality will impact long-term yield and carbon dynamics. However, current methods do not provide spatially explicit predictions, limiting the use of this data for proactive forest management, such as thinning and reducing fuel load to reduce tree mortality. In this study we assess whether the incorporation of individual tree inventory data derived from airborne laser scanning and modeled and observed fire intensity data can provide accurate spatially explicit tree mortality predictions. Specifically, tree-level mortality was predicted for over 1.9 million trees within six wildfires in mixed coniferous forest in northwestern Montana, USA, and validated using high resolution imagery for each segmented tree crown. Random forest classification models utilizing observed fire intensity metrics derived from VIIRS observations were the most accurate (overall accuracy: 77.2 %), followed by random forest classification models utilizing modeled fire intensity (64.5–66.1 %) and those that utilized existing logistic regression relationships (55.7–59.0 %). The random forest tree mortality models also produced lower RMSE (6.3–10.2 %) and bias (1.7–3.7 %) compared to the logistic regression approach (RMSE: 38.4 %, bias: −29.6 %) when mortality accuracy was assessed across tree size class. The predictor variable importance quantification showed that fire intensity metrics were more important than species and structural variables in the mortality classification. Ultimately, this study contributes to the remote sensing of fire effects and fire science fields by developing a remote sensing-based methodology for predicting spatially explicit individual tree mortality across large spatial extents.