Hierarchical UAV-LiDAR Classification of Field Margin Structural Types in Agricultural Landscapes

Field margins substantially contribute to landscape connectivity and ecosystem functioning in agricultural systems, offering key opportunities to enhance biodiversity and ecosystem services. Their structural classification is essential for targeted conservation and management strategies. Currently, detailed characterization of field margin structural variation is limited: traditional field surveys lack reproducibility and scalability, while coarse-resolution remote sensing fails to capture fine-scale structure relevant for ecological assessment. To address this, we developed a hierarchical decision-tree framework based on high-resolution UAV-LiDAR data that automatically classifies field margins into ecologically meaningful structural types, enabling rapid, objective assessment of vegetation structure and ecological potential.

High-resolution UAV-LiDAR point clouds were acquired for four field margins at two study sites in southern Saxony-Anhalt, Germany. We derived four essential pixel-based structural indicators describing (1) vegetation height, (2) vertical stratification across herb, shrub and tree layers, (3) vegetation density/porosity (Pulse Penetration Ratio) and (4) structural homogeneity (dense vegetation fraction). Classification thresholds were defined from metric distributions to maximise separability among field margin types. A hierarchical decision tree with two main pathways (tree-dominant vs. shrub-dominant) classified field margins into five structural types: Tree Row, Compact Hedgerow (Shelterbelt and Hedge subtypes), Complex Woody Mosaic and Open/Degraded Shrub Margin. Classifications were validated internally based on metric-derived thresholds.

Applied to the dataset, the framework successfully distinguished four structural types among its five defined classes using pixel-based metrics. Compact Hedgerow (Shelterbelt) featured tall vegetation (15.1 m), moderate dense canopy fraction (0.46) and relatively low Pulse Penetration Ratio (0.30), suggesting homogeneous structure. Complex Woody Mosaic, despite similar height (17.8 m), showed slightly lower dense fraction (0.40) and Pulse Penetration Ratio (0.34), indicating subtle fragmentation. Open/Degraded Shrub Margin had distinctly lower height (6.2 m), moderate shrub ratio (0.27) and higher Pulse Penetration Ratio (0.36). Compact Hedgerow (Hedge) exhibited shrub dominance (2–5 m ratio 0.40) with highest dense fraction (0.62) and lowest Pulse Penetration Ratio (0.25).

This reproducible and scalable LiDAR-based classification provides a transferable framework for assessing field margin structure independent of species composition and supports targeted management and evidence-based conservation in agricultural landscapes.