Layer-explicit forest understory measurement: Assessing the capability of UAV LiDAR to extract understory vegetation structure

Abstract

Forest understory vegetation is an important component of forests and plays an irreplaceable role in ecological applications such as forest regeneration, succession, fire behavior, and understory economy. Unlike the overstory, which has well-established observational frameworks and structural metrics, the understory lacks explicit delineation and corresponding quantitative descriptors. This gap hampers its structural characterization and integration into forest inventories. Although airborne LiDAR has been used to retrieve understory structure, most efforts rely on spatiotemporally consistent field plots, and the underlying error sources remain unclear, limiting robustness of understory retrieval methods. To address these gaps, we propose an LiDAR-based measurement framework that explicitly defines understory vegetation as a spatially continuous layer. Using ground-based laser scanning (GLS, including terrestrial (TLS) and mobile laser scanning (MLS)) as reference, the capability of unmanned aerial vehicle laser scanning (ULS) to quantify understory structure was systematically assessed in seven sites across China. By combining semantic and individual tree segmentation, we extracted understory layer and quantified understory structure using a complementary set of four metrics including height, cover, density, and complexity. We further analyzed the factors influencing ULS estimation errors based on pulse-thinning experiments. The results showed that ULS-derived structural metrics exhibited strong agreement with GLS (R² = 0.73 ∼ 0.88). Cross-sectional ULS point cloud profiles exhibited clear separation between understory and overstory layers, with no gaps or striping artifacts. Sensitivity experiments demonstrated that penetration capability, quantified by the maximum number of returns per pulse, was the dominant source of uncertainty, determining whether the understory layer could be effectively detected, while point density governed structural detail once detected. Environmental factor analysis indicated that increasing canopy cover amplified errors, whereas greater canopy complexity enhanced penetration and reduced errors. We recommend ULS systems capable of recording ≥ 3 returns with point densities ≥ 500 pts/m² to balance accuracy and efficiency. This study provides a new perspective for the quantitative assessment of understory structure and offers methodological and practical insights for future research.