Using single-date airborne LiDAR features to map tree mortality timing in central european forests

Mapping tree mortality timing is essential for understanding forest disturbance dynamics and supporting management decisions under changing climatic conditions, particularly in regions affected by bark beetle outbreaks and increased wildfire risk. This study proposes a framework to estimate tree mortality timing using single-date airborne LiDAR data combined with an area-based approach. The method was applied in Bohemian Switzerland National Park (Czech Republic), where extensive Norway spruce mortality occurred between 2018 and 2021. Geometric and radiometric LiDAR features were extracted and used in three classification scenarios: (i) individual mortality years (2018–2021), (ii) grouped mortality classes (2018–2019 vs. 2020–2021), and (iii) contrasting mortality years (2018 vs. 2021). Spatial dependence was accounted for using spatial-block cross-validation, and Global Random Forest and Spatially Localized Random Forest models were compared, showing only minor differences in performance. Classification accuracy increased with the temporal contrast between classes. Scenario (i) showed limited performance (OA = 0.50), reflecting the difficulty of distinguishing adjacent years due to gradual structural and radiometric changes. Grouping mortality years (scenario ii) improved results (OA = 0.72), while the highest accuracy was achieved with scenario (iii), when contrasting temporally distant mortality years (OA = 0.81). Feature importance and ALE analysis indicated that geometric variables captured subtle structural differences, whereas radiometric features provided stronger discrimination for more distinct mortality conditions. Although precise annual dating from a single acquisition remains challenging, the proposed framework enables scalable mapping of mortality timing across large areas, supporting fire risk assessment and forest management.