LiDAR-based assessment of treetop dieback to detect asian longhorned beetle damage in poplar forests

Effective forest management depends on accurately monitoring changes in forest health. The Asian longhorned beetle (ALB) has caused extensive mortality in broadleaf trees worldwide. ALB typically manifests as a distinctive treetop-dieback phenotype that progresses downward in damaged trees. Although LiDAR is widely used for plant-stress detection, two challenges persist: (1) non-specific structural responses that hinder the identification of damage-specific phenotypes and (2) limited transferability of structural metrics across age- and size-heterogeneous stands. We address these challenges with a within-tree ratio-based framework that targets the treetop-dieback phenotype with internal normalization. Two poplar plots (young, old) were selected to represent age- and size-related heterogeneity. Based on field evidence (oviposition pits, frass holes, exit holes), each tree was labeled as healthy, lightly damaged, or severely damaged. For each tree, we vertically segmented the point cloud at 50% and 80% of total height and computed crown volume (V), point density (PD), and leaf area index (LAI) of each segment. We then derived upper-to-lower ratio metrics, using the lower crown as an internal normalizer, to capture top-down dieback while normalizing size- and age-related heterogeneity. We also defined a family of combination ratios that aggregate ratios at 50% and 80% heights. Using Linear Discriminant Analysis (LDA), we evaluated separability of each metric under two strategies: (1) a 70/30 random split by trees and (2) a plot-transfer test (train: Stand young ; test: Stand old ). The results indicated that the proposed metrics, Summary Ratio of Volume (SRV = V u p p e r 50 % V l o w e r 50 % + V u p p e r 20 % V l o w e r 80 % ), achieved the highest overall accuracy (OA): 76.23% in random split, and 70.58% in plot-transfer, significantly outperforming other proposed metrics and existing LiDAR indices. By coupling phenotype-focused features with internal normalization, the approach enables precise detection of treetop dieback and improves transferability across age and size heterogeneity.