Multitemporal LiDAR improves estimates of fire severity in forested landscapes

Landsat-based fire severity maps have limited ecological resolution, which can hinder assessments of change to specific resources. Therefore, we evaluated the use of pre- and post-fire LiDAR, and combined LiDAR with Landsat-based relative differenced Normalized Burn Ratio (RdNBR) estimates, to increase the accuracy and resolution of basal area mortality estimation. We vertically segmented point clouds and performed model selection on spectral and spatial pre- and post-fire LiDAR metrics and their absolute differences. Our best multitemporal LiDAR model included change in mean intensity values 2–10 m above ground, the sum of proportion of canopy reflection above 10 m, and differences in maximum height. This model significantly reduced root-mean-squared error (RMSE), root-mean-squared prediction error (RMSPE), and bias when compared with models using only RdNBR. Our top combined model integrated RdNBR with LiDAR return proportions <2 m above ground, pre-fire 95% heights and pre-fire return proportions <2 m above ground. This model also significantly reduced RMSE, RMSPE, and bias relative to RdNBR. Our results confirm that three-dimensional spectral and spatial information from multitemporal LiDAR can isolate disturbance effects on specific ecological resources with higher accuracy and ecological resolution than Landsat-based estimates, offering a new frontier in landscape-scale estimates of fire effects.