Abstract
The AMAPVox R package allows users to read, manipulate, and write voxel spaces generated by the AMAPVox LiDAR point cloud voxelisation software. It aims to estimate leaf area through several theoretical and numerical approaches, providing a standardized programmatic environment for handling complex 3D datasets. (Vincent et al., 2017; 2021).
Introduction
LiDAR (Light Detection and Ranging) has revolutionized forestry and ecology by providing high-resolution 3D maps of vegetation. However, raw point clouds are essentially collections of coordinates; to extract meaningful ecological metrics, researchers often use "voxelisation"—dividing the 3D space into a grid of cubes (voxels). While the standalone AMAPVox software handles the initial voxelisation, the AMAPVox R package provides a robust toolkit for post-processing, cleaning, and analyzing these datasets at scale.
Methodological Framework
The package acts as a specialized extension for handling VoxelSpace objects. It addresses common challenges in 3D data processing, such as noise reduction and spatial interpolation. Key methodological features include:
- Noise Reduction: The
butterfly()function identifies isolated non-empty voxels (termed "butterflies") that likely represent sensor noise rather than actual vegetation. - Spatial Interpolation: The
fillNA()function uses a user-defined radius to fill missing values based on neighboring voxels, ensuring the neighborhood is isotropic. - Layer Extraction: Functions like
canopy()andground()allow researchers to extract specific layers based on hit thresholds, effectively "slicing" the forest structure.
As shown in the parameter documentation, the hit.min parameter is a critical threshold for determining the minimum number of echoes required to consider a voxel as containing vegetation, helping to filter out low-confidence signals.
Ecological Modeling and Leaf Angle Distribution
A primary goal of AMAPVox is the estimation of Plant Area Density (PAD) and Plant Area Index (PAI). Unlike simple point-counting methods, the package implements estimators that account for beam attenuation and transmittance.
Accurate leaf area estimation requires modeling how leaves are oriented in space. The computeG() function supports various theoretical Leaf Angle Distribution (LAD) models, including spherical, planophile, and erectophile distributions. This flexibility is essential because different vegetation types (e.g., grasses vs. broadleaf trees) require distinct mathematical treatments.
The package supports a wide range of probability density functions (PDFs) and parameters, allowing for precise G-projection coefficient estimation based on the specific incident beam inclination (theta) and leaf geometry.
Key Insights and Performance
- Efficiency: By using a
data.tablebackend, the package handles millions of voxels with minimal memory overhead. - Software Orchestration: Uniquely, the R package can manage the standalone AMAPVox Java software itself—checking for versions, installing them, and running batch processing directly from the R console.
- Visualization: Integrated 3D plotting via the
rglpackage allows for interactive inspection of sampling intensity and PAD distribution. - Vertical Profiling: Users can generate vertical PAI profiles with a single command, a task that previously required complex manual geometry.
Limitations and Future Work
The underlying AMAPVox engine is Java-based, which can create environment conflicts (e.g., requiring Java 8 for older versions). Additionally, while the R package is highly efficient for post-processing, the initial ray-tracing and voxelisation still rely on the standalone AMAPVox software. Future iterations may aim for a more integrated, R-native voxelisation engine to streamline the workflow further.
Summary updated