Transforming Habitat Analysis: How Terrestrial Laser Scanning LiDAR Reveals 3D Ecosystem Structure and Function

Abstract

This article provides a comprehensive overview of Terrestrial Laser Scanning (TLS) LiDAR and its transformative role in habitat analysis for researchers and scientists. It covers the foundational principles of TLS technology, explores its advanced methodological applications in creating detailed 3D structural models of forests and other ecosystems, addresses key challenges and optimization strategies in data processing, and presents a comparative analysis of its performance against other sensing technologies. The synthesis aims to equip professionals with the knowledge to leverage TLS for precise environmental monitoring, carbon stock assessment, and biodiversity conservation.

The Fundamentals of TLS LiDAR: Principles and Evolution in Environmental Sensing

Terrestrial Laser Scanning (TLS) is a ground-based, contact-free remote sensing technology that uses light detection and ranging (LiDAR) to capture highly accurate three-dimensional measurements of environments and objects [1] [2]. The core principle involves emitting laser pulses and measuring their return time to construct detailed digital representations of the physical world. TLS concentrates on smaller spatial extents—from tens of centimeters to a couple of kilometers—to achieve extremely high spatial resolution on the scale of millimeters to centimeters, making it an indispensable tool for detailed ecological and habitat research where structural precision is paramount [2] [3].

Unlike airborne systems that map from above, TLS instruments are positioned at ground level, typically on a stationary tripod, allowing them to capture detailed measurements of complex structures such as forest understories, geological features, and river banks from a unique lateral perspective [2] [4]. This positions TLS as a critical technology for creating highly accurate 3D models of small-scale topographic features relevant to habitat characterization [3].

Core Technical Principles of TLS

Measurement Principles and Components

TLS systems operate on several fundamental measurement principles, with time-of-flight being the most common for long-range environmental scanning [1]. The process involves:

• Time-of-Flight Measurement: The scanner emits laser pulses and precisely measures the time difference between the emission of a pulse and the detection of the reflected signal. This time interval, multiplied by the speed of light and divided by two, gives the distance between the scanner and the target surface [1] [3].
• Azimuth and Zenith Angles: Simultaneously, the system records the horizontal (azimuth) and vertical (zenith) angles of each laser emission, enabling the calculation of precise X, Y, Z coordinates for each measured point [2].
• Intensity Recording: Most systems also record the intensity of the return signal, which provides information about the reflective properties of the surface [1].

A complete TLS instrument consists of a laser scanner, a precision navigation system (internal encoders for angle measurement), control software, and often an integrated or mounted digital camera for photo-texturing purposes [5] [1]. Modern systems can acquire hundreds of thousands to over a million points per second, with effective ranges extending up to several hundred meters depending on target reflectivity and atmospheric conditions [1] [2].

Data Output: The Point Cloud

The primary data output from TLS is a point cloud—a dense collection of discrete data points representing the precise three-dimensional coordinates of surfaces within the scanned environment [2] [3]. Each point in the cloud contains:

• Cartesian coordinates (X, Y, Z) defining its spatial position
• Intensity value representing the return signal strength
• Possibly RGB color values if correlated with digital imagery [1]

This point cloud serves as a digital foundation for creating measurable 3D models, digital elevation models (DTMs), and photorealistic representations when integrated with photographic data [1] [2].

TLS vs. Airborne LiDAR: A Comparative Analysis

Platform and Operational Differences

Feature	Terrestrial Laser Scanning (TLS)	Airborne LiDAR
Platform	Ground-based, stationary tripod [6] [2]	Aircraft, helicopter, or drone [6] [7]
Viewing Perspective	Lateral/ground-level, upward-looking [4]	Nadir/overhead, downward-looking [6]
Spatial Coverage	Limited to line-of-sight from scan positions [2]	Large-area coverage from above [6]
Mobility	Static setup; requires multiple positions [6] [1]	Continuous data collection during flight [6]
Typical Range	Up to several hundred meters [1] [2]	Hundreds to thousands of meters [7]

Technical and Application-Based Differences

Characteristic	Terrestrial Laser Scanning (TLS)	Airborne LiDAR
Spatial Resolution	Millimeter to centimeter (e.g., 2mm point spacing) [2]	Centimeter to decimeter [8]
Positional Accuracy	~4mm positional error [2]; comparative studies show majority of UAV LiDAR points within 1.8 inches of TLS [8]	Varies with altitude; generally lower absolute accuracy
Data Collection Focus	Fine-scale structural details, vertical profiles, undersides of features [2] [4]	Broad topographic mapping, canopy surface models [6] [7]
Ideal Applications	Tree architecture, stem curves, ecological plot studies, cliff faces, river banks [2] [4] [9]	Regional mapping, forest canopy height, landscape-scale topography [6] [7]
Limitations	Shadowing/occlusion, limited spatial extent, setup time required [1] [2]	Limited undersory penetration, less structural detail, weather/airspace dependencies [6]

TLS in Habitat Research: Applications and Protocols

Key Research Applications

In environmental and habitat research, TLS has emerged as a transformative technology that provides unprecedented structural detail for ecosystem assessment:

• Forest Ecology and Carbon Monitoring: TLS captures extremely detailed 3D measurements of trees, supporting applications in forest ecology, carbon monitoring, and biodiversity assessment [4]. It enables precise quantification of tree architecture, stem diameters, and biomass through the development of quantitative structure models (QSMs) [4] [9].
• Structural Habitat Characterization: The technology allows researchers to characterize the three-dimensional arrangement of plant components, which influences and responds to environmental changes, playing a key role in regulating light regimes, forest productivity, and physiological processes [4].
• Erosion and Geomorphic Monitoring: TLS enables high-accuracy mapping of surface changes through repeat measurements, making it ideal for monitoring coastal erosion, river bank erosion, landslides, and other geomorphic processes [2] [3] [10].
• Digital Twin Creation: TLS data facilitates the development of "digital twins" or virtual forest approaches that represent maximum structural detail for radiative transfer modeling and functional structural plant modeling (FSPM) [4].

Experimental Protocol: TLS for Forest Habitat Structural Assessment

Phase 1: Pre-Field Planning

• Objective Definition: Clearly define structural parameters of interest (e.g., stem density, canopy volume, leaf area distribution).
• Site Delineation: Establish plot boundaries with permanent corner markers for repeat surveys.
• Scanner Positioning: Plan multiple scan positions to minimize occlusion, ensuring sufficient overlap (typically 30-60%) between adjacent scans [1].

Phase 2: Field Deployment and Data Acquisition

• Equipment Setup: Establish TLS on stable tripod, ensuring instrument leveling at each position.
• Control Target Placement: Position fixed reference targets (minimum of 3-5) visible from multiple scan positions for subsequent registration [1] [2].
• Scan Registration: The process of aligning multiple scans into a unified coordinate system using common targets or natural features [1] [2].
• Scan Parameterization: Configure appropriate angular resolution (point density) based on research questions—typically 1-10 cm at 100m distance for habitat applications [1].
• Ancillary Data Collection: Acquire photographic data for point cloud coloring and collect supporting field measurements (e.g., soil samples, vegetation samples) as needed.

Phase 3: Data Processing and Analysis

• Point Cloud Registration: Align individual scans using software such as Leica Cyclone, leveraging control targets or feature matching algorithms [1] [2].
• Data Cleaning: Remove outliers and noise artifacts, including mixed pixel returns that occur at abrupt elevation changes [2].
• Feature Extraction: Apply algorithms to classify points into ground, vegetation, and structural components, enabling stem detection, canopy modeling, and terrain derivation [4].
• Metric Derivation: Compute habitat structural metrics such as leaf area index, canopy volume, stem density, and complexity indices from the classified point cloud [4].

TLS Habitat Research Workflow

Research Reagent Solutions: Essential Materials for TLS Habitat Studies

Research Component	Function in TLS Habitat Research
High-Precision TLS Instrument	Core data acquisition tool; provides accurate 3D point measurements of habitat structure [2] [4]
Geodetic Control Targets	Enable precise registration of multiple scans into a unified coordinate system [1] [3]
GPS/GNSS Receiver	Provides absolute positioning for georeferencing TLS data in a global reference frame [2] [3]
Digital Camera (High-Res)	Captures photographic data for point cloud colorization and visual interpretation [1]
Specialized Processing Software	Processes raw point clouds, performs registration, classification, and metric extraction [1] [2]

Terrestrial Laser Scanning represents a powerful methodological approach for habitat research, offering unparalleled resolution and structural detail compared to airborne alternatives. Its ground-based perspective provides complementary data to aerial surveys, enabling comprehensive 3D characterization of ecosystems from the soil surface to the canopy. While TLS requires careful planning and processing to overcome limitations such as occlusion, its ability to capture millimeter-to-centimeter scale structural information makes it indispensable for modern ecological studies, particularly those focused on understanding fine-scale habitat structure, monitoring ecosystem changes, and developing predictive models of vegetation dynamics. As computational power and artificial intelligence capabilities continue to advance, TLS is poised to play an increasingly central role in quantifying and monitoring the complex three-dimensional nature of habitats in a changing world.

Terrestrial Laser Scanning (TLS), also referred to as terrestrial LiDAR, has fundamentally transformed data acquisition in forest science by providing unprecedented three-dimensional structural details of forest ecosystems [11] [12]. Unlike airborne systems, TLS instruments are deployed at ground level, capturing intricate measurements of the forest understory and upper canopy with superior geometric accuracy and structural completeness [11]. The technology's application in geomorphology and forest science is a relatively recent advancement, gaining significant momentum from around 2010 onwards [11] [12]. Its adoption was driven by key improvements in three areas: a reduction in the price of instruments, increased speed of point acquisition, and a decrease in the physical size of the devices, making the technology more accessible and field-practical [11].

Table: Fundamental Characteristics of Terrestrial Laser Scanning

Characteristic	Description
Technology Principle	Emits laser pulses to measure distances, recording XYZ coordinates of numerous points to create a 3D "point cloud" [12].
Spatial Resolution	Can be up to 1 mm intervals for short-range scanners, though not practical for all but the smallest areas [12].
Point Acquisition Rate	Modern TLS devices can measure 10^4–10^6 points per second with an accuracy of 10^−1–100 cm [12].
Typical Range	Categorized into short-, medium-, and long-range scanners, with a trade-off between pulse rate and laser energy [12].

Key Technological Milestones and Applications

The evolution of TLS has enabled a progression from simplified structural models towards highly detailed "digital twins" of forest environments [11]. This has empowered researchers to tackle complex questions in several key areas:

• Understanding Tree Architecture: TLS provides detailed 3D measurements of the size and arrangement of a tree's fundamental components, known as tree architecture or morphology [11].
• Radiative Transfer Modeling: TLS facilitates the creation of highly accurate 3D canopy representations. These allow scientists to solve radiative transfer problems using methods like Monte Carlo ray tracing, which is crucial for quantifying canopy photosynthesis and modeling the Earth's radiation budget [11].
• Functional Structural Plant Modeling (FSPM): TLS data offers an effective method for parameterizing FSPMs and directly testing their structural predictions, enabling exploration of ecological and environmental hypotheses [11].
• Quantifying Ecosystem Dynamics: The precision and accuracy of TLS allow for repeat surveys to track topographic and structural changes over time, linking processes and forms to detect environmental change [12].

Table: Evolution of TLS Application in Forest Science

Era	Primary Capabilities	Key Application Areas
Pre-2010s (Early Adoption)	Basic forest structure assessment (tree height, stem diameter) [11].	Initial topographic mapping of landforms [12].
2010s Onwards (Rapid Uptake)	Improved plot-scale forest measurements; estimation of tree metrics and biomass [11].	Hillslope-channel coupling; debris flow monitoring; gravel-bed river and fault surface analysis [12].
Current State (Digital Twins)	High-detail 3D reconstruction via Quantitative Structure Models (QSM); coupling with AI and advanced radiative transfer models [11].	Creation of virtual forests; understanding light regimes and forest productivity; spectral property analysis [11].

Experimental Protocols in Modern TLS Habitat Research

Protocol for Sward Development Monitoring in Grasslands

This protocol outlines the use of TLS to track the development of grassland canopies at a high temporal resolution [12].

1. Objective: To follow the subtle changes in the grassland canopy structure throughout the growing season and relate mean canopy height to community biomass.

2. Materials & Equipment:

• Terrestrial Laser Scanner (capable of sub-2 mm spatial resolution).
• Calibration targets (if required by the system).
• RTK GPS or other surveying equipment for georeferencing.
• Data storage and processing unit.

3. Field Procedure:

• Plot Selection: Identify and mark the grassland plots to be monitored.
• Scanner Setup: Position the TLS at a predetermined location to capture the entire plot. The example study used one perspective scan of 92 plots every second week [12].
• Data Acquisition: Perform scans at regular intervals (e.g., bi-weekly). Ensure consistent scanner position and settings across all time points.
• RGB Imagery: Capture digital photographs of the plots alongside each scan for visual reference [12].

4. Data Processing & Analysis:

• Point Cloud Generation: Register multiple scans (if applicable) to create a comprehensive 3D point cloud of the plot.
• Surface Modeling: The point cloud represents the canopy surface. Derive metrics like mean canopy height from the point cloud data.
• Biomass Calibration: Calibrate the TLS-derived height metrics against physically harvested dry community biomass. Note that calibration precision can be restricted if biomass samples are only taken at peak mowing times and cover a smaller area than the TLS scan [12].

Protocol for Channel-Bed Level and Bathymetric Mapping

This protocol combines TLS with empirical-optical techniques to map underwater topography [12].

1. Objective: To derive detailed channel-bed levels, including submerged areas, in river anabranches.

2. Materials & Equipment:

• Terrestrial Laser Scanner.
• Non-metric vertical aerial camera (e.g., Nikon D90 with fixed lens).
• Acoustic depth sounder (e.g., Sontek S5 RiverSurveyor with integrated RTK GPS).
• Lightweight boat.

3. Field Procedure:

• TLS Survey: Conduct TLS surveys during low-flow conditions to capture exposed and shallow submerged areas [12].
• Aerial Photography: Acquire vertical aerial photographs of the wetted channels immediately following the TLS survey. Use a helicopter and camera set to acquire images automatically every 5 seconds [12].
• Depth Sounding: Right after the aerial photos are taken, collect depth data along transects of the primary anabranches. Use a boat-mounted acoustic system acquiring geo-located depth soundings at a high frequency (e.g., 10 Hz) while the boat is guided downstream on tethers [12].

4. Data Processing & Analysis:

• TLS Point Cloud Processing: Process and merge TLS data to create a digital elevation model (DEM) of the non-submerged areas.
• Depth Model Development: Use the aerial photographs and the acquired depth transect data to develop and validate an empirical-optical model for estimating water depth.
• Data Fusion: Combine the TLS-derived DEM with the optically-derived bathymetric model to create a seamless map of the entire channel-bed.

The Scientist's Toolkit: Essential Research Reagents & Materials

Table: Key Research Reagent Solutions for TLS Field Research

Item / Solution	Function & Application
High-Accuracy TLS Instrument	Core sensor for 3D data acquisition. Selection depends on required range, accuracy, and portability [11] [12].
Calibration Targets	Used for co-registering multiple scan positions into a single, cohesive point cloud [11].
RTK GPS System	Provides precise geolocation for scan positions and ground control points, enabling data integration into geographic coordinate systems [12].
Point Cloud Processing Software	Bespoke software packages are required for managing, analyzing, and visualizing the large volumes of data in a TLS point cloud [12].
Quantitative Structure Model (QSM) Algorithm	Algorithmic tool for enclosing point clouds in topologically-connected, closed volumes to reconstruct detailed tree architecture [11].

Visualization of Workflows

TLS Habitat Research Workflow

From Point Cloud to Forest Model

Terrestrial Laser Scanning (TLS) is a ground-based light detection and ranging (LiDAR) technology that captures the three-dimensional structure of environments with high precision [3]. For researchers conducting habitat research, the ability to accurately measure complex vegetation structures, topography, and ecosystem properties hinges on a fundamental understanding of the core hardware components of a TLS system. The key parameters of laser wavelength, scanning mechanism, and system accuracy directly determine the suitability of the technology for specific research applications, the quality of the collected data, and the validity of the resulting ecological inferences. This application note provides a detailed technical overview of these critical hardware components, framed within the context of terrestrial LiDAR habitat research, to enable scientists to make informed equipment selections and implement robust data collection protocols.

Core Hardware Components and Specifications

Laser Wavelength

The laser wavelength is a primary determinant of how a laser beam interacts with different materials and surfaces, making it a critical consideration for habitat mapping.

Table 1: Common Laser Types and Wavelengths in TLS and Related Technologies

Laser Type	Gain Medium	Common Wavelengths	Typical Operation Mode	Relevance to Habitat Research
Nd:YAG	Solid-state (Crystal)	1064 nm [13]	CW, Pulsed [13]	Standard for topographic TLS; reflects well from vegetation and soil.
Er-glass	Solid-state (Glass)	1530-1560 nm [13]	CW [13]	Used in optical amplifiers; eye-safe wavelengths can be advantageous for field surveys.
Tm:YAG/Ho:YAG	Solid-state (Crystal)	2000-2100 nm [13]	µs, ns [13]	Tissue ablation studies; potential for specialized material identification.
Cr:ZnSe	Solid-state (Crystal)	2200-2800 nm [13]	CW, fs [13]	Spectroscopy applications; can be used for MWIR chemical sensing.
CO₂	Gas	10600 nm [13]	CW, µs [13]	Materials processing; less common for field-based TLS.
Laser Diode (GaN)	Semiconductor	410 nm [13]	CW, ns [13]	Not typical for TLS; used in Blu-ray, represents short-wavelength end.
OPSL (e.g., 561 nm)	Solid-state	561 nm [14]	CW [14]	Common in fluorescence microscopy labs for biological sample analysis.

Most standard topographic TLS systems utilize lasers in the near-infrared (NIR) range (e.g., 900-1100 nm, such as the Nd:YAG at 1064 nm) [13] [15]. These wavelengths are ideal for general habitat research as they reflect well from a variety of surfaces, including leaves, wood, and soil. A key consideration in wavelength selection is eye safety; longer wavelengths are often considered "eye-safe" as they are less focused by the cornea and thus permit higher energy levels to be used in the field, which can be crucial for scanning long distances or through sparse vegetation.

Scanning Mechanisms

The scanning mechanism defines how the laser beam is directed across the scene to capture the 3D point cloud. The mechanism influences the speed, range, and overall reliability of the system.

• Oscillating Mirrors/Moving Optics: This is a common mechanism in many TLS systems. A mirror is rapidly tilted or rotated to deflect the laser beam across the field of view. These systems can achieve very high scan rates (up to millions of points per second) and are suitable for a wide range of distances. However, the presence of moving parts can lead to wear and tear over time, potentially affecting precision [16].
• Rotating Polygon Mirrors: Another mechanical approach uses a multi-faceted mirror that rotates at high speed. Each face of the mirror sweeps the laser beam across a portion of the scene. This design can enable very fast line-scanning and is robust, but may have a more limited vertical field of view compared to other mirror-based systems.
• MEMS (Micro-Electro-Mechanical Systems) Mirrors: A more recent advancement is the use of MEMS, which are tiny, solid-state mirrors etched from silicon. They can be tilted electronically with no large moving parts, leading to a more compact, power-efficient, and potentially more durable scanner. While initially limited in scan angle and power handling, MEMS technology is rapidly advancing and being incorporated into newer TLS systems [16].
• Phased-Array and Solid-State Scanning (Emerging): These are truly solid-state systems with no macroscopic moving parts. They steer the beam by electronically controlling the phase of the light from an array of emitters or by using optical principles. While not yet widespread in high-precision TLS, they represent the future trend toward more robust and compact systems, as seen in their growing adoption in automotive LiDAR [16].

Accuracy Specifications

Accuracy in TLS is not a single value but a combination of several interrelated specifications that define the quality of the measured point cloud. Understanding these is essential for designing a survey and interpreting results.

Table 2: Key TLS Accuracy and Performance Specifications

Parameter	Definition	Typical Range for Modern TLS	Impact on Habitat Data
Range Accuracy	The uncertainty in a single distance measurement.	1-10 mm [12]	Directly affects the precision of tree diameter, canopy height, and micro-topography measurements.
Beam Divergence	The angular spread of the laser beam, determining the spot size at a given distance.	0.1 - 1 mrad [14]	A smaller divergence provides a finer spot size, allowing for better resolution of small branches and fine structural details.
Point Spacing	The angular or spatial separation between consecutive measured points.	Can be < 1 mm at 50 m for short-range scanners [12]	Determines the level of structural detail captured. Denser spacing is needed for complex vegetation like shrubs.
Positional Accuracy	The overall 3D positional error of a point, influenced by GPS, IMU, and angular encoders.	10-100 cm (absolute, without ground control); can be mm-cm with control [3] [15]	Critical for georeferencing and combining multiple scans or aligning with other spatial datasets (e.g., satellite imagery).

It is vital to distinguish between precision (the repeatability of measurements) and accuracy (the closeness to the true value). A scanner can be precise but inaccurate if it has a systematic error. Furthermore, the effective accuracy in a real-world habitat setting is also influenced by environmental conditions (e.g., rain, fog, high ambient light) and target characteristics (e.g., wet vs. dry leaves, dark bark).

Experimental Protocol: TLS for 3D Habitat Structure and Fuel Monitoring

The following protocol, adapted from USDA Forest Service research, provides a detailed methodology for using TLS to monitor habitat structure, with a specific application to pre- and post-fire fuel dynamics [17].

Research Reagent Solutions

Table 3: Essential Materials for TLS Habitat Survey

Item	Function
Terrestrial Laser Scanner	The primary data collection tool that emits laser pulses and measures their return to create a 3D point cloud.
High-Precision GPS Receiver	Provides absolute geographic coordinates for the scanner and ground control points, enabling data georeferencing.
Scan Targets (Spheres/Checkboards)	Used as common, recognizable points in multiple scans to allow for accurate registration (alignment) of individual scans.
Field Laptop/Controller	For operating the scanner, monitoring data collection in real-time, and performing initial data quality checks.
Calibration Equipment	Used to verify and maintain the manufacturer's specified accuracy of the TLS system.
Direct Measurement Tools (e.g., D-tape, Clinometer)	Used to collect ground-truth data (e.g., tree diameter, height, debris size) for calibrating and validating TLS estimates.
Data Processing Software (e.g., CloudCompare, RIEGL software)	Specialized software for registering point clouds, filtering noise, extracting metrics, and analyzing 3D structure.

Step-by-Step Procedure

Step 1: Pre-Field Planning

• Define the research plot boundaries (e.g., 1 ha) and establish a permanent monument at the plot center if repeat surveys are planned.
• Use a high-precision GPS to record the coordinates of the monument and several potential scanner setup locations around the plot perimeter, ensuring good visibility of the interior.
• Strategically place and survey fixed markers (e.g., checkerboard targets on stakes) within the plot. These will serve as tie points for registering multiple scans.

Step 2: Field Deployment and Scanning

• Set up the TLS system on a stable tripod at the first pre-surveyed location.
• Power on the system and connect the field controller. Initialize the system's internal GPS and IMU.
• Configure the scan settings. For comprehensive habitat structure, select a high-resolution/high-density scan mode. This will result in a smaller point spacing (e.g., 1 cm at 10 m), capturing finer details of vegetation and surface fuels, albeit with a longer scan duration.
• Initiate the scan. A single scan from one position may take from 5 to 30 minutes depending on the selected density and field of view.
• After the scan is complete, move the TLS to the next pre-surveyed location, ensuring sufficient overlap (recommended >30%) with the previous scan's coverage. Repeat the scanning process.
• Continue until the entire plot has been covered from multiple angles (typically 3-5 positions for a 1 ha forest plot) to minimize occlusion, where objects hide areas behind them from the scanner's view.

Step 3: Ground Truthing and Calibration

• Within the scanned plot, perform traditional field measurements to calibrate the TLS data.
• Measure a subset of trees for Diameter at Breast Height (DBH) and height.
• Conduct surface fuel transects, measuring the diameter and length of woody debris and estimating the height and cover of shrubs.
• These direct measurements are critical for converting the TLS point cloud into quantifiable biophysical parameters like biomass and fuel volume [17].

Step 4: Data Processing and Analysis

• Registration: Import all individual scans into processing software. Use the surveyed targets as common references to precisely align all scans into a single, unified point cloud of the entire research plot.
• Filtering and Classification: Apply algorithms to remove noise (e.g., flying birds, insects) and classify points into different categories: ground, vegetation (further separated into low, mid, and canopy if possible), and dead wood.
• Metric Extraction:
  • Canopy Height Model (CHM): Subtract the digital terrain model (DTM) from the digital surface model (DSM) to derive canopy height.
  • Fuel Volume: Calculate the 3D volume of surface and ladder fuels from the classified point clouds.
  • Biomass Estimation: Develop a regression model between the ground-truthed biomass measurements and TLS-derived metrics (e.g., canopy volume, plant area index) to estimate biomass across the entire plot [17].
• Change Detection (for fire effects): For a pre- and post-fire study, compare the registered 3D models from two different time periods. Quantify changes in fuel biomass, canopy cover, and soil surface elevation due to erosion.

The workflow for this protocol is summarized in the following diagram:

Diagram 1: Workflow for TLS habitat monitoring, showing key phases from planning to metric generation.

Hardware Selection and Validation Workflow

Choosing the right hardware and validating its performance requires a logical decision-making process. The following diagram outlines the key considerations and steps.

Diagram 2: Decision workflow for selecting and validating TLS hardware based on research requirements.

Terrestrial Laser Scanning (TLS) provides a multi-dimensional digital record of habitats, capturing not only structure but also material properties. The primary data outputs include 3D point clouds, which define spatial coordinates (X, Y, Z) for each measured point; intensity, which records the strength of the returned laser signal; and spectral information, which can be derived from multi-wavelength systems. In habitat research, the fusion of these data types enables a comprehensive understanding of ecosystem structure, composition, and function, moving beyond simple geometry to identify species and assess physiological status [18] [11]. The integration of these data streams is crucial for creating detailed "digital twins" of forest environments, which serve as virtual replicas for scientific analysis and modeling [11].

Table 1: Core Data Types from Terrestrial Laser Scanning Systems

Data Type	Description	Key Metrics/Units	Primary Ecological Application
3D Point Cloud	A set of data points in a 3D coordinate system defining the external surface of objects.	Point density (pts/m²), spatial accuracy (m), number of returns.	Tree architecture quantification, biomass estimation, habitat structural complexity [19] [11].
Intensity	The strength of the backscattered laser signal for each point, influenced by surface properties and range.	Unitless digital number (DN), often 8-16 bit; requires calibration for physical interpretation.	Material discrimination (e.g., leaf vs. bark), rough species classification, and condition assessment [20].
Spectral (Multi-/Hyper-spectral)	Data captured at multiple specific wavelengths, revealing material-specific reflectance signatures.	Reflectance values across discrete bands (e.g., 532 nm, 1064 nm, 1550 nm) [18].	Detailed species identification, assessment of plant health and chemistry, substrate characterization [21] [18].

Table 2: Terrestrial LiDAR Scanner Types and Specifications

Scanner Type	Operating Principle	Typical Range	Key Strengths	Example Application in Habitat Research
Phase-Shift	Measures phase difference between emitted and received continuous-wave laser.	Short to medium (e.g., up to ~180m [22])	Very high point acquisition speed, high density.	Detailed understory and plot-level structural mapping [23].
Pulse-Based (Time-of-Flight)	Measures time for a short laser pulse to travel to and from a target.	Long-range (e.g., up to 6000m [22])	Excellent long-range performance, robust in varied conditions.	Large-scale ecosystem monitoring, scanning from few positions [23] [22].

Experimental Protocols for Habitat Research

Protocol 1: Multi-Scale Habitat Structural Assessment using 3D Point Clouds

Objective: To quantitatively assess the 3D structural complexity of a forest habitat at the plot level from a terrestrial point cloud.

Materials:

• Terrestrial Laser Scanner (e.g., RIEGL VZ-series [22])
• Calibration targets (e.g., spheres)
• GNSS receiver (for georeferencing, optional)
• Computer with point cloud processing software (e.g., CloudCompare, R packages such as lidR)

Methodology:

• Site and Scan Position Setup: Establish a systematic grid of scan positions within the study plot to minimize occlusion. A minimum of 5 scan positions per hectare is often effective, though density should be increased in complex vegetation [11].
• Scanner Registration: Place calibration targets in stable positions visible from multiple scan locations to facilitate later co-registration of the individual point clouds.
• Data Acquisition: At each position, execute a full-dome scan, ensuring settings (e.g., angular resolution, quality) are consistent. Record the intensity and, if available, RGB data alongside the 3D coordinates.
• Data Pre-processing:
  • Co-registration: Use the calibration targets to align all individual scans into a single, unified point cloud in a common coordinate system [11].
  • Noise Filtering: Apply statistical outlier removal or noise filters to eliminate erroneous points (e.g., flying pixels).
  • Classification: Use automated algorithms (e.g., Cloth Simulation Function - CSF) to classify points into "ground" and "non-ground" categories.
• Structural Metric Extraction:
  • Plot-Level Metrics: Calculate metrics like Plant Area Index (PAI), gap fraction, and vertical plant profile from the normalized point cloud.
  • Individual Tree Metrics: Isolate individual trees from the point cloud. For each tree, model its structure using a Quantitative Structure Model (QSM) to derive metrics such as Diameter at Breast Height (DBH), tree height, volume, and branch architecture [11].
  • Habitat Complexity: Compute voxel-based metrics or the rumple index to quantify the spatial heterogeneity of the vegetation.

Workflow for 3D point cloud structural analysis.

Protocol 2: Species Classification via Intensity Image Compensation

Objective: To improve object and species recognition accuracy in a point cloud by integrating geometric features with complementary intensity data.

Materials:

• TLS capable of simultaneous 3D and intensity data capture (e.g., Avalanche Photon Diode (APD) array LiDAR [20])
• Computing environment for feature calculation (e.g., MATLAB, Python)

Methodology:

• Data Acquisition: Capture the co-registered 3D point cloud and intensity image of the target scene.
• Local Feature Calculation (3D Geometric):
  • For a keypoint in the point cloud, construct a Local Reference Frame (LRF) to achieve pose invariance [20].
  • Within the keypoint's local spherical neighborhood, calculate the deviation angle between the normal vector of each point and the LRF.
  • Construct a histogram of these deviation angles to represent the local 3D surface geometry.
• Intensity Feature Extraction:
  • From the intensity image corresponding to the point cloud, extract the contour of the target object (e.g., a tree crown or leaf cluster).
  • Calculate the barycenter (center of mass) of the contour.
  • Compute the distance sequence from the barycenter to every point on the contour.
  • Perform a Discrete Fourier Transform (DFT) on this distance sequence. The resulting DFT coefficients serve as a contour shape descriptor that is invariant to rotation [20].
• Feature Fusion and Matching:
  • Fuse the 3D geometric feature histogram and the intensity-based DFT contour feature into a combined feature vector.
  • Match this combined feature vector against a pre-existing model library of known objects/species using a nearest-neighbor or machine learning classifier for recognition.

Fusing 3D geometric and intensity features for classification.

Protocol 3: Multi-Cloud Classification for Complex Habitats using 3DMASC

Objective: To classify points in complex natural environments (e.g., topo-bathymetric zones, vegetated floodplains) by leveraging the distinct samplings of multiple point clouds, such as from a bi-spectral TLS.

Materials:

• Topo-bathymetric LiDAR system or multiple co-registered TLS datasets (e.g., NIR and green lasers [21])
• Computing platform for the 3DMASC (3D point classification with Multiple Attributes, Multiple Scales, and Multiple Clouds) workflow or equivalent.

Methodology:

• Data Input: Use two or more distinct point clouds of the same scene. A prime example is a topo-bathymetric dataset consisting of a near-infrared (NIR) point cloud and a green-wavelength point cloud [21].
• Multi-Scale, Multi-Cloud Feature Extraction:
  • For each point in each cloud, define a spherical neighborhood at multiple spatial scales (e.g., 0.1 m, 0.5 m, 1.0 m).
  • At each scale, calculate a suite of features for both the "native" cloud and the "complementary" cloud within the neighborhood. Features can include:
    • Geometric features: Linearity, planarity, scattering, verticality, etc.
    • Joint-cloud features: Density differences, median intensity difference between clouds, 3D distance between nearest neighbors in the two clouds [21].
  • This process can generate a large feature set (e.g., >80 features).
• Feature Selection and Training:
  • Use a feature selection algorithm to identify the most discriminative features and scales for the target classes (e.g., "water surface," "vegetation," "riverbed").
  • Train a random forest classifier with a limited set of manually labeled training points (~2000 points per class is often sufficient [21]).
• Classification and Validation: Apply the trained model to classify the entire point cloud. Validate the results against a withheld set of ground-truth points.

The Scientist's Toolkit: Essential Research Reagents & Materials

Table 3: Key Research Solutions and Materials for TLS Habitat Studies

Item / Solution	Function & Application in TLS Research
Quantitative Structure Models (QSMs)	Algorithmic models that enclose tree point clouds into topologically-connected, closed volumes (e.g., cylinders). They are used to translate point data into quantifiable tree architecture metrics like volume, biomass, and branching topology [11].
3DMASC Workflow	A classification framework operating directly on multiple point clouds. It is essential for leveraging the spectral and sampling differences in datasets like topo-bathymetric LiDAR to classify complex scenes with vegetated, urban, and submerged objects [21].
Local Reference Frame (LRF)	A local coordinate system constructed at a keypoint in a point cloud. It is a foundational component for creating pose-invariant local feature descriptors, which are crucial for robust object recognition [20].
Discrete Fourier Transform (DFT) Contour Feature	A shape descriptor derived from the contour of an object in an intensity image. It provides a rotation-invariant feature that can be fused with 3D geometric features to enhance the discrimination of objects like different tree species [20].
Random Forest Classifier	A machine learning algorithm that operates by constructing multiple decision trees. It is widely used in point cloud classification due to its high performance, ability to handle high-dimensional feature spaces, and provision of feature importance scores [21].
Monte Carlo Ray Tracing (MCRT)	A simulation technique used in radiative transfer modeling. When applied to highly detailed 3D forest models ("digital twins") derived from TLS, it helps scientists understand how forest structure influences light regimes and canopy scattering processes [11].

Visualization and Analysis Workflows

3DMASC multi-cloud classification workflow.

The integration of 3D, intensity, and spectral data profoundly advances terrestrial LiDAR habitat research. It enables the creation of "digital twins" for simulating ecological processes like radiative transfer [11], improves the accuracy of species classification and structural metrics [20], and allows for the monitoring of complex interfaces such as land-water-vegetation in topo-bathymetric studies [21]. As TLS technology continues to evolve toward greater portability, affordability, and integration with multi-sensor platforms [23] [18] [22], the protocols outlined here will empower researchers to quantitatively interrogate habitat structure and function at an unprecedented level of detail, providing critical insights for ecology and conservation.

From Point Clouds to Ecological Insights: TLS Methodologies for Habitat Assessment

The three-dimensional arrangement of plant components is fundamental for characterizing forest ecosystems, influencing and responding to environmental changes while regulating light regimes, productivity, and physiological processes [11]. Over recent decades, terrestrial laser scanning (TLS), also called terrestrial LiDAR, has revolutionized forest science by providing unprecedented detailed measurements of forest understory and upper canopy structure with superior geometric accuracy compared to other ground-based methods [11]. This technology enables the creation of highly accurate digital replicas of forest ecosystems—virtual forests that serve as critical research tools.

These digital twins represent more than simplified models; they are dynamic, data-integrated virtual copies of physical forest landscapes that enable continuous monitoring, predictive modeling, and adaptive management [24]. For researchers and scientists engaged in habitat research, TLS-derived digital twins provide a transformative approach to studying ecosystem dynamics, quantifying disturbance impacts, and advancing conservation strategies with unprecedented precision.

Application Notes: Key Use Cases and Quantitative Benefits

Terrestrial LiDAR technology has expanded beyond basic forest inventory to enable sophisticated applications across ecological research. The technology captures extremely detailed 3D descriptions of tree and forest structure, facilitating what are termed "digital twin" or virtual forest approaches [11]. These detailed structural descriptions are algorithmically enclosed in topologically-connected, closed volumes known as Quantitative Structure Models (QSMs), which enable precise measurements of individual trees and stand structure [11].

Table 1: Quantitative Applications of TLS in Forest Ecosystem Research

Application Domain	Key Measurable Parameters	Measurement Accuracy/Impact	Research Utility
Radiative Transfer Modeling	Canopy photosynthesis, radiation budget, biophysical feedback [11]	Enables scientific understanding of multi-angular scattering processes [11]	Quantifies forest-climate interactions; models Earth's radiation budget
Functional Structural Plant Modeling (FSPM)	Crown development, growth mechanisms, phenotypic information [11]	Parameterizes FSPMs to simulate structure-environment-physiology interactions [11]	Tests ecological hypotheses; links structure to function in plant resource use
Forest Restoration Monitoring	Biomass accumulation, biodiversity indicators, canopy closure [24]	95-98% accuracy in individual-tree reconstruction [24]	Verifies restoration outcomes; enables automated climate finance via smart contracts
Ecosystem and Fire Effects	Fuel loads, forest structure, ecological features [25]	Captures detailed metrics in <5 minutes per plot [25]	Supports fire risk assessment; quantifies landscape-scale conditions
Cost Efficiency in Restoration	Planting costs, monitoring expenses, verification costs [24]	Reduces per-tree costs from USD $2.00-3.75 to USD $0.11-1.08 [24]	Enables scalable restoration projects; improves financial transparency

The integration of TLS with artificial intelligence has significantly advanced the field. Increasing computational power, alongside the rise of AI, is empowering researchers to tackle more complex questions about forest ecosystem dynamics in a changing world [11]. Modern algorithms, including deep learning approaches for crown delineation and automated pipelines for large-scale tree extraction, are streamlining TLS data processing and enabling more efficient analysis of complex point clouds [11].

For pharmaceutical development professionals studying natural products or environmental impacts on ecosystem services, these virtual forests provide critical insights into medicinal plant architecture, distribution, and abundance under changing environmental conditions. The structural economics spectrum concept embeds tree size and structural diversity within the broader framework of plant resource use, potentially informing drug discovery from plant sources [11].

Experimental Protocols for Terrestrial LiDAR Data Collection and Processing

Field Deployment and Scanning Protocol

The following methodology outlines a standardized approach for TLS data collection in forest ecosystems, optimized for digital twin creation:

• Plot Establishment: Delineate research plots representing the forest heterogeneity. For ecosystem monitoring, the USDA Forest Service protocol utilizes portable, push-button TLS equipment that captures detailed forestry, fuels, and ecological features in <5 minutes per plot [25].

• Scanner Setup and Registration: Deploy TLS instruments at multiple positions within the plot to reduce occlusion and improve structural data completeness. Modern scanning systems increasingly eliminate the requirement for fixed calibration targets for registration, reducing setup time and enabling faster fieldwork workflows [11].

• Data Acquisition: Conduct scans at each predetermined position, ensuring sufficient overlap between scan positions. High-end TLS instruments typically feature high ranging accuracy and long effective range, though more affordable options have become increasingly available [11].

• Environmental Documentation: Record ancillary data including GPS coordinates, sensor specifications, weather conditions, and phenological stage of vegetation to support subsequent data interpretation and modeling.

Data Processing and Digital Twin Creation Pipeline

• Point Cloud Registration: Align individual scans into a unified coordinate system using co-registration methods [11]. This critical step transforms multiple discrete scans into a comprehensive 3D representation of the forest plot.

• AI-Driven Feature Extraction: Apply deep learning approaches for automated crown delineation, stem detection, and ecological feature classification [11]. These methods significantly reduce manual processing time while improving accuracy and repeatability.

• Quantitative Structure Modeling (QSM): Generate topologically-connected, closed volumes that algorithmically enclose point clouds to create mathematically defined tree architectures [11]. QSMs provide the fundamental building blocks for virtual forest construction.

• Radiometric Parameterization: Assign spectral properties to structural components (leaves, bark, soil) to enable radiative transfer modeling [26]. This process transforms structural models into functional virtual forests capable of simulating light interactions.

• Validation and Accuracy Assessment: Compare TLS-derived metrics with traditional field measurements to quantify accuracy and identify potential systematic errors. This step is essential for establishing scientific credibility and quantifying uncertainty in digital twin representations.

Digital Twin Creation Workflow: This diagram illustrates the end-to-end pipeline for creating virtual forests from terrestrial LiDAR data, highlighting the sequential stages from planning to research application.

Technical Framework for Virtual Forest Deployment

The architecture for forest digital twins typically follows a layered approach that integrates multiple technologies:

• Physical Layer: Consists of TLS instruments, drones, and IoT-enabled sensors for in-situ environmental monitoring [24]. This layer captures the raw 3D data and continuous environmental parameters.

• Data Layer: Manages secure and structured transmission of spatiotemporal data, including point clouds and associated metadata [24]. This layer addresses data storage, retrieval, and interoperability challenges.

• Intelligence Layer: Applies AI-driven modeling, simulation, and predictive analytics to forecast biomass, biodiversity, and risk factors [24]. This layer transforms structural data into actionable ecological insights.

• Application Layer: Provides stakeholder dashboards, milestone-based smart contracts, and automated reporting functionalities [24]. This layer delivers research tools and decision support interfaces.

Table 2: Research Reagent Solutions for TLS Ecosystem Studies

Tool Category	Specific Technologies	Research Function	Implementation Considerations
Acquisition Hardware	Terrestrial Laser Scanners, UAV-LiDAR, IoT sensors [24]	Captures 3D ecosystem structure and environmental parameters	Resolution, portability, cost, and operational complexity trade-offs
Processing Algorithms	Deep learning crown delineation, automated tree extraction, co-registration methods [11]	Converts point clouds to ecological metrics and QSMs	Computational demands, accuracy validation, and parameter sensitivity
Modeling Software	Radiative transfer models, FSPM platforms, 3D reconstruction tools [11]	Simulates ecological processes and forest dynamics	Model fidelity, parameter requirements, and computational efficiency
Validation Instruments	Field calipers, hemispherical photography, dendrometers, soil sensors [25]	Ground-truths TLS-derived metrics and model outputs	Measurement precision, labor requirements, and spatial sampling design

Digital Twin System Architecture: This diagram visualizes the four-layer technical framework for implementing forest digital twins, showing how data flows from physical sensors to research applications.

Advanced Implementation Considerations

Integration with Emerging Technologies

The fusion of TLS with complementary technologies enhances virtual forest capabilities:

• Blockchain Integration: When paired with blockchain and smart contracts, digital twins become trust-enabling systems that allow automated payments to be triggered when restoration milestones are digitally verified [24]. This approach reduces transaction costs and fraud risks in research funding and conservation finance.

• AI and Machine Learning: Algorithms analyze extensive biological data to identify patterns and predict ecosystem responses [11]. These capabilities enable more efficient and targeted research approaches, increasing the likelihood of successful interventions.

• Internet of Things (IoT): Complementary sensor networks provide continuous environmental monitoring data that enrich static TLS scans with temporal dynamics, enabling real-time ecosystem assessment [24].

Validation and Uncertainty Quantification

Establishing rigorous validation protocols remains essential for scientific acceptance of virtual forests. Traditional monitoring methods provide reference data for TLS-derived metrics, though they are often limited by data quality and repeatability issues [25]. The integration of multi-scale validation approaches, combining field measurements, aerial photography, and satellite data, creates robust frameworks for quantifying digital twin accuracy and uncertainty.

For pharmaceutical professionals applying these methods to natural product research, the detailed 3D representations of medicinal plant species within their ecosystem context provide unprecedented insights into plant architecture, distribution, and abundance—critical factors for understanding medicinal compound production and availability under changing environmental conditions.

Terrestrial Laser Scanning (TLS), also referred to as terrestrial LiDAR, has emerged as a cornerstone technology for capturing the three-dimensional arrangement of plant components within forest ecosystems. This 3D structure is fundamental for characterizing forests, as it influences and responds to environmental changes, playing a key role in regulating light regimes, forest productivity, and physiological and biophysical processes [11] [4]. Over the past decades, TLS has provided a unique perspective that offers new insights into ecological processes and forest disturbances, while significantly enhancing structural assessments in forest and carbon inventories [11]. Unlike airborne systems, TLS instruments are positioned at ground level, allowing them to capture highly detailed measurements of both the forest understory and the upper canopy with superior geometric accuracy and structural completeness compared to other ground-based methods [11] [4].

The adoption of TLS in forest studies has accelerated from around 2010 onwards, driven by improvements in affordability, instrument speed, and reduced size [11] [4]. Modern TLS systems can rapidly acquire dense point clouds—collections of precise three-dimensional data points—that digitally represent the forest structure. Recent algorithmic advances, including co-registration methods and deep learning approaches for crown delineation, are streamlining TLS data processing and enabling more efficient analysis of these complex point clouds [11]. This technological evolution is expanding the scope of ecological research and transforming how researchers study forest structure and dynamics, particularly in the context of climate change and carbon cycle science.

Core Principles of Forest Structural Quantification

Fundamental Tree Metrics from TLS Data

The conversion of raw TLS point clouds into ecologically meaningful metrics relies on established processing pipelines and algorithms. The primary measurements include:

• Diameter at Breast Height (DBH): A standard forestry measurement obtained at 1.3 meters above ground level, DBH is crucial for assessing forest structure and calculating wood volume, tree density, and carbon storage [27]. TLS derives DBH by fitting cylinders or other geometric primitives to the point cloud data of tree stems.
• Tree Height (H): The vertical distance from ground level to the highest part of the tree. TLS captures this through detailed 3D measurements that overcome the limitations of traditional tools like clinometers or laser rangefinders [27].
• Crown Dimensions: TLS data enables precise mapping of crown width, depth, volume, and surface area, which are essential for understanding light interception and growth dynamics.
• Stem Form and Taper: The detailed point clouds allow for analysis of stem straightness and how diameter changes along the bole, which improves volume estimations.
• Architectural Complexity: Advanced metrics such as structural complexity indices can be derived from the 3D spatial distribution of points, quantifying habitat heterogeneity and micro-environmental variation.

From Structure to Biomass and Carbon

The pathway from structural metrics to carbon stocks follows established allometric relationships, though TLS offers opportunities for refinement:

Aboveground Biomass (AGB) Estimation: Traditional allometric equations relate DBH and/or height to AGB through power-law functions. TLS enhances this approach by providing more detailed structural data that can be used to develop species-specific models or reduce uncertainty in existing equations. For carbon stock assessment, biomass is converted to carbon content using a carbon fraction factor, typically ranging from 0.47 for perennial trees to 0.413 for palm species [27].

Volumetric Approaches: Advanced TLS processing uses Quantitative Structure Models (QSMs)—algorithmic enclosures of point clouds in topologically-connected, closed volumes—to compute total tree volume directly from the 3D data [11]. When combined with wood density information, these volumetric estimates provide an alternative pathway to biomass estimation that may complement or improve upon allometric methods.

Carbon Sequestration Capacity: Beyond estimating current carbon stocks, TLS data supports modeling of carbon sequestration rates through growth monitoring. Time-series TLS acquisitions can track structural changes in individual trees, enabling direct measurement of growth and carbon accumulation without relying on allometric projections [27].

Experimental Protocols and Methodologies

TLS Data Acquisition Protocol for Forest Plots

Pre-field Planning:

• Plot Establishment: Define plot size (typically 20×20 m to 1 ha) based on research objectives and forest density, ensuring clear corner markers.
• Scan Configuration: Plan scan positions using a systematic grid or optimized pattern to minimize occlusion effects. For a 1-ha plot, 5-12 scan positions are typically required depending on vegetation density [11].
• Equipment Check: Verify TLS instrument calibration, battery levels, and data storage capacity before fieldwork.

Field Deployment:

• Scanner Setup: Position the TLS on a stable tripod at approximately 1.3 m height. Use a leveling plate to ensure horizontal alignment.
• Scan Registration: Deploy targets (for target-based registration) or utilize natural features (for target-free registration) between scan positions to enable subsequent co-registration [11] [4].
• Parameter Settings: Configure scan resolution (point spacing) and quality settings based on research goals. Higher resolution (e.g., 1 cm at 10 m distance) captures finer structural details but increases acquisition time and data volume.
• Data Acquisition: Execute full dome scans at each position, ensuring sufficient overlap between adjacent scans (typically 30% minimum). Record scan position coordinates using GPS if available.

Complementary Data Collection:

• Collect traditional measurements (e.g., DBH with diameter tape, tree height with laser rangefinder) for validation purposes [27].
• Document tree species, health status, and any notable observations.
• For carbon stock studies, collect wood density samples if feasible.

Point Cloud Processing Protocol

Data Pre-processing:

• Registration: Align individual scans into a unified coordinate system using target-based or target-free methods [11]. Registration accuracy should achieve residuals of <1 cm for high-quality structural analysis.
• Colorization: Apply RGB values from co-acquired imagery if available to enhance point cloud interpretability.
• Filtering: Remove noise and outliers through statistical filters while preserving ecological meaningful structure.
• Ground Classification: Identify and classify ground points using algorithms such as Progressive Morphological Filter or Cloth Simulation Filter.

Tree-Level Segmentation:

• Stem Detection: Identify individual stems from the point cloud using clustering algorithms or deep learning approaches [11] [4].
• Crown Delineation: Isolate crown points through region-growing methods, voxel-based partitioning, or contour approaches [11].
• Individual Tree Extraction: Group stem and crown components into individual tree objects, addressing challenges of occluded stems and overlapping crowns.

Metric Extraction:

• DBH Estimation: Fit cylindrical models to stem points at 1.3 m height, implementing quality checks for fit statistics.
• Height Calculation: Compute as the difference between highest crown point and ground level beneath the tree.
• Crown Metrics: Derive crown volume, surface area, and dimensions from the isolated crown points.
• QSM Construction: Reconstruct detailed 3D models of tree architecture using series of connected cylinders or other geometric primitives [11].

Table 1: TLS Data Processing Workflow Stages

Processing Stage	Key Algorithms/Methods	Output	Quality Control Metrics
Scan Registration	Iterative Closest Point (ICP), Feature-based matching	Unified point cloud	Registration error (<1 cm), Point cloud completeness
Ground Classification	Progressive Morphological Filter, Cloth Simulation Filter	Classified ground points	Terrain representation accuracy, Commission/omission errors
Stem Detection	Density-based clustering, Deep learning (e.g., PointNet++)	Individual stem points	Detection rate, False positive rate, Positioning accuracy
DBH Estimation	Cylinder fitting (RANSAC, Hough transform)	DBH values	Cylinder fit quality (R², RMSE), Comparison with manual measures
QSM Reconstruction	TreeQSM, AdTree, SimpleTree	3D tree models	Volume closure, Component connectivity, Branch topology

Biomass and Carbon Stock Calculation Protocol

Allometric Approach:

• Equation Selection: Choose species-specific or generic allometric equations that relate DBH and/or height to aboveground biomass. For example:

For perennial trees [27]:

• Aboveground stem biomass (W_S) = 0.0396 × (D² × H)^0.933
• Aboveground branch biomass (W_B) = 0.00349 × (D² × H)^1.030
• Aboveground leaf biomass (WL) = (28 / (WS + W_B + 0.025)) - 1
• Total aboveground biomass (WT) = WS + WB + WL

For palm species [27]:

• Total aboveground biomass (W_T) = 6.666 + 12.826 × (H)^0.5 × (ln H)

• Application: Input TLS-derived metrics into selected equations to compute individual tree biomass.

Carbon Stock Calculation:

• Convert biomass to carbon stocks using appropriate carbon fraction (CF):
  • Carbon in aboveground biomass (CAGB) = WT × CF × 44/12 [27]
  • Where CF = 0.47 for perennial trees and 0.413 for palm species [27]
  • The factor 44/12 converts carbon to carbon dioxide equivalent

Volumetric Approach:

• Compute tree volume directly from QSMs
• Convert volume to biomass using wood density: Biomass = Volume × Wood Density
• Apply carbon fraction as above to estimate carbon stocks

Uncertainty Quantification:

• Propagate errors from point cloud measurement through to final carbon estimates
• Compare TLS-derived estimates with destructive sampling or traditional inventory data when available

Research Toolkit: Essential Equipment and Software

Table 2: Terrestrial Laser Scanning Research Toolkit

Category	Specific Tools/Options	Key Features/Specifications	Application Context
TLS Hardware	Leica Geosystems AG, Trimble Inc., FARO Technologies, RIEGL Laser Measurement Systems [28]	Varying range, accuracy, speed; Phase-shift (indoor/medium range) vs. Pulse-based (long range) scanners [23]	Permanent plot monitoring, High-precision structural measurement
Mobile Scanning	Handheld/backpack systems (e.g., Zoller + Fröhlich) [28] [23]	SLAM-based positioning, Rapid data acquisition, Reduced occlusion	Large-area surveys, Complex terrain, Understory mapping
Complementary Sensors	Smartphone LiDAR (e.g., iPhone with ATH application) [27]	Portable, low-cost, ~0.897 R² for tree height compared to traditional methods [27]	Rapid assessment, Community science, Educational applications
Field Equipment	Diameter tape, Laser rangefinder, GPS, Calibration targets [27]	Validation measurements, Positioning, Scan registration	Ground truthing, Method validation, Quality assurance
Processing Software	CloudCompare, TreeQSM, 3D Forest, LidR, R packages	Point cloud visualization, Segmentation, Metric extraction, QSM reconstruction	Data processing, Analysis, Model development
Analysis Platforms	Python, R, MATLAB with point cloud libraries	Custom algorithm development, Statistical analysis, Visualization	Advanced method development, Bulk processing, Custom metrics

Workflow Visualization

TLS Forest Structure Assessment Workflow

Biomass and Carbon Estimation Pathways

Advanced Applications and Future Directions

Digital Twins and Forest Modeling

TLS is enabling a transition from simplified forest models toward "digital twins"—virtual representations with maximum structural detail that precisely mirror physical forests in both space and time [11] [4]. This approach provides unprecedented opportunities for understanding forest dynamics:

• Radiative Transfer Modeling: Detailed TLS reconstructions enable sophisticated simulation of light interception and scattering within canopies using methods like Monte Carlo Ray Tracing, advancing understanding of canopy photosynthesis and climate-vegetation feedbacks [11].
• Functional-Structural Plant Models (FSPMs): TLS data supports parameterization and validation of FSPMs that simulate interactions between tree structure, environment, and physiological processes [11] [4].
• Structural Economics Spectrum: The quantitative links between structure and function enabled by TLS are helping embed tree size and structural diversity within the broader framework of plant resource use strategies [11].

Temporal Dynamics and Carbon Monitoring

The use of TLS for monitoring temporal changes represents a frontier in forest carbon science:

• 4D Forest Monitoring: Repeated TLS acquisitions create time series that capture structural dynamics, including growth, mortality, and disturbance impacts, enabling direct measurement of carbon flux without reliance on allometric projections [11] [4].
• Disturbance Impact Quantification: TLS can precisely measure structural damage from windthrow, ice storms, or insect outbreaks, improving estimates of carbon emissions from disturbances.
• Regeneration and Recovery Tracking: High-resolution TLS data can monitor understory development and canopy gap dynamics, informing understanding of forest recovery carbon dynamics.

Technological Integration and Methodology Development

Future advancements in TLS applications will likely focus on:

• Artificial Intelligence Integration: Deep learning approaches are streamlining TLS data processing, enabling automated tree detection, species classification, and structural metric extraction [11] [4].
• Multi-platform LiDAR Fusion: Combining TLS with airborne, mobile, and UAV LiDAR to overcome occlusion limitations and scale measurements from individual trees to landscapes [28] [23].
• Standardization and Intercomparison: Methodological studies are establishing best practices and quantifying uncertainties in TLS-derived forest metrics to support policy-relevant carbon accounting [27].

As TLS technology continues to evolve toward more compact, affordable, and user-friendly systems [28] [23], its applications in forest research and carbon monitoring will expand, ultimately enhancing our understanding of forest ecosystems and their role in the global carbon cycle.

Forest voids—the three-dimensional, unoccupied spaces within forest ecosystems—represent a critical yet under-described component of stand structure. These voids are not merely empty spaces but active elements shaped by vegetation, microclimate, and disturbance regimes, governing essential processes such as light penetration, airflow, and habitat connectivity [29]. Traditional canopy-centric metrics and simplified radiative assumptions have proven insufficient for capturing the complex structural interplay between vegetation and void space. This protocol outlines a novel LiDAR-based framework that identifies, visualizes, and quantifies forest voids directly from terrestrial and mobile laser scanning (TLS/MLS) point clouds, providing a scalable, assumption-light representation of forest architecture with applications in biodiversity monitoring, habitat suitability assessment, and climate-adaptation research [29].

The framework operates by treating voids as the 3D regions between the digital elevation model (DEM) or digital surface model (DSM) where no LiDAR returns are detected, effectively bypassing traditional structural metrics [29]. Across diverse forest sites, void configurations have been shown to reflect underlying stand architecture with remarkable fidelity: structurally heterogeneous forests with multi-layered canopies and irregular stem distributions exhibit diffuse, vertically extensive voids, while structurally uniform stands contain more confined voids largely restricted to lower strata due to diminished understory development [29]. This structural lens on spatial openness provides integrated metrics of overstory and understory attributes, offering fresh insights into ecosystem dynamics and function.

Theoretical Foundation and Void Classification

Defining Forest Void Spaces

Forest voids exist across a continuum of spatial scales and originate from various biological and physical processes. Methodologically, voids are defined as contiguous three-dimensional regions within the forest volume unoccupied by vegetation and below the outermost canopy envelope, as determined by LiDAR point clouds [29]. These spaces can be categorized into three primary classes based on their structural origin and functional attributes:

• Canopy gaps: Vertical voids extending from the ground surface upward, created by treefall, mortality, or disturbance events [30].
• Inter-crown spaces: Lateral voids between adjacent tree crowns or vegetation clusters, influencing light regimes and animal movement pathways [29].
• Sub-canopy voids: Horizontal voids beneath the primary canopy but above the understory, often connected to stand density and management history [29].

The spatial distribution and connectivity of these void types form what can be conceptualized as the "forest void network"—an essential component of habitat complexity that influences numerous ecological processes including seedling recruitment, predator-prey interactions, and microclimate regulation.

Structural Signatures of Forest Types

Void configurations provide distinctive structural signatures for different forest types, serving as quantitative descriptors of ecosystem condition and developmental stage [29]:

Table: Characteristic Void Patterns Across Forest Types

Forest Type	Void Distribution	Vertical Extent	Structural Drivers
Structurally Heterogeneous Forests (Multi-layered canopies, irregular stem distributions)	Diffuse, discontinuous	Extensive vertical development	Gap-phase dynamics, complex succession
Structurally Uniform Stands (Even-aged, single canopy layer)	Concentrated, confined	Primarily lower strata	Limited understory development, management history
Secondary Deciduous Broadleaf Forests	Dynamic, shifting patterns	Variable	Gap formation and closure cycles [30]

Equipment and Computational Requirements

Research Reagent Solutions

Successful implementation of the forest void quantification framework requires specific hardware, software, and data processing tools. The following table details essential components of the research toolkit:

Table: Essential Research Toolkit for Forest Void Quantification

Category	Specific Tool/Platform	Function in Void Analysis
Data Acquisition Hardware	Terrestrial Laser Scanner (TLS)	High-resolution understory and trunk-level data capture [4]
	Mobile Laser Scanner (MLS)	Efficient large-area under-canopy mapping [31]
	Airborne Laser Scanner (ALS)	Canopy surface and topographic modeling [30]
Software & Algorithms	3DForest, TreeSeg	Individual tree extraction and point cloud processing [31]
	CloudCompare	Point cloud visualization and manual segmentation [31]
	Kernel Point Convolutions (KPConv)	Deep learning-based point cloud analysis [32]
Data Requirements	Minimum 300 points/m³	Accurate tree height estimation (RMSE < 1m) [31]
	Minimum 600-700 points/m³	Accurate DBH estimation (RMSE < 1cm) [31]
	Multi-temporal point clouds	Monitoring void dynamics over time [30]

Point Cloud Density Specifications

The accuracy of void characterization is directly influenced by point cloud density, which varies significantly across acquisition methods. Mobile Laser Scanning (MLS) systems have demonstrated that accurate tree height estimation (RMSE < 1m, representing <5% error) requires densities exceeding 300 points/m³, while accurate DBH estimation (RMSE < 1cm, representing <5% error) necessitates higher densities of 600-700 points/m³ [31]. These density thresholds ensure sufficient resolution for distinguishing true void spaces from data artifacts caused by occlusion or undersampling.

Experimental Protocol: Void Quantification Workflow

Data Acquisition and Pre-processing

Field Scanning Procedures

Terrestrial Laser Scanning (TLS) Protocol:

• Establish systematic scan positions with approximately 30-40% overlap to minimize occlusion
• Use fixed calibration targets or natural features for subsequent co-registration
• Employ a minimum of 8-12 scan positions per hectare for comprehensive coverage
• Collect data during leaf-off conditions for deciduous forests to enhance trunk detection
• Document scan positions with GPS coordinates for multi-temporal alignment

Mobile Laser Scanning (MLS) Protocol:

• Utilize backpack or robotic platforms equipped with SLAM technology
• Maintain consistent walking speed (0.5-1.0 m/s) to ensure uniform point density
• Implement overlapping traverse patterns to minimize systematic gaps
• Conduct surveys during optimal weather conditions (no precipitation, minimal wind)

Point Cloud Processing

Data Registration and Filtering:

• Apply iterative closest point (ICP) algorithms for multi-scan alignment
• Implement statistical outlier removal to eliminate noise artifacts
• Classify ground points using progressive morphological filters or cloth simulation
• Generate digital canopy height models (DCHMs) by normalizing elevations to ground level [30]

Quality Control Metrics:

• Verify registration accuracy with mean alignment error <10 cm
• Ensure point cloud density meets minimum thresholds for target applications
• Confirm complete spatial coverage with no significant data voids

Void Identification and Quantification

Void Detection Algorithm

The core void detection process involves several computational steps:

Voxelization and Occupancy Analysis:

• Convert point clouds to voxel arrays with resolution appropriate to research questions (typically 0.25-2.0m) [32]
• Calculate occupancy percentage within each voxel based on LiDAR returns
• Implement density-based relevance (DBR) weighting to address class imbalance in material distribution [32]
• Classify voxels with occupancy below predetermined thresholds as void space

Region Growing and Void Delineation:

• Apply connected components algorithm to identify contiguous void regions
• Implement 3D morphological operations to smooth void boundaries
• Calculate void metrics including volume, surface area, and sphericity index

Void Metric Extraction

Basic Void Metrics:

• Volume (m³): Total three-dimensional space of each void
• Surface Area (m²): Total boundary area between void and vegetation
• Vertical Profile: Height distribution and vertical extent
• Orientation: Primary directional alignment of elongated voids

Spatial Pattern Metrics:

• Void Size Distribution: Frequency distribution of void volumes
• Void Connectivity: Degree of interconnection between adjacent voids
• Spatial Autocorrelation: Clustering or dispersion patterns of voids
• Vertical Stratification: Distribution of voids across canopy layers

Multi-Temporal Void Dynamics Assessment

For monitoring temporal changes in void architecture, implement the following protocol:

Time-Series Alignment:

• Co-register multi-temporal point clouds using permanent reference features
• Apply change detection algorithms to identify void expansion, contraction, fragmentation, or coalescence
• Quantify rates of void turnover using established gap dynamics methodologies [30]

Dynamic Metrics Calculation:

• Void lifespan estimation based on shrinkage rates and initial area [30]
• Transition probabilities between void states (formation, persistence, closure)
• Lateral growth rates of gap-edge trees into void spaces [30]

Data Analysis and Interpretation

Quantitative Void Metrics

The following metrics provide comprehensive characterization of forest void spaces:

Table: Comprehensive Void Metric Framework

Metric Category	Specific Metric	Calculation Method	Ecological Interpretation
Basic Dimensions	Void Volume	Voxel counting × resolution	Resource availability potential
	Void Surface Area	Mesh reconstruction	Vegetation-atmosphere interface area
	Void Depth	Vertical extent from canopy surface	Light penetration capacity
Spatial Distribution	Void Density	Number of voids per unit area	Structural heterogeneity
	Mean Nearest Neighbor Distance	Average distance between void centroids	Void isolation/connectivity
	Void Size Distribution	Power-law fitting to volume frequencies	Disturbance regime characterization
Structural Complexity	Fractal Dimension	3D box-counting algorithm	Structural complexity at multiple scales
	Lacunarity	Measurement of gappiness patterns	Spatial heterogeneity of void distribution
Temporal Dynamics	Void Formation Rate	New voids per unit time per unit area	Disturbance frequency
	Void Persistence	Temporal autocorrelation of void locations	Structural stability
	Vertical Profile Change	Shift in void distribution across strata	Successional stage development

Statistical Analysis and Modeling

Spatial Analysis:

• Implement spatial point pattern analysis (Ripley's K, pair correlation function) for void distributions
• Apply variogram analysis to quantify spatial dependence in void metrics
• Use regression trees to identify environmental drivers of void patterns

Predictive Modeling:

• Develop linear regression models relating initial gap area to shrinkage rates [30]
• Implement survival analysis to predict void lifespan based on structural metrics [30]
• Apply machine learning (Random Forests, KPConv) to classify void types and predict dynamics [32]

Applications and Implementation Notes

Integration with Ecological Research

The void quantification framework supports diverse ecological applications:

Habitat Assessment:

• Correlate void metrics with species presence/absence data for fauna requiring movement corridors
• Map structural connectivity for arboreal and volant species
• Identify critical void thresholds for specialist species with specific microhabitat requirements

Ecosystem Function:

• Relate void configurations to light transmission and understory productivity
• Model influences on wind penetration and gas exchange
• Quantify carbon dynamics associated with gap formation and closure

Methodological Considerations and Limitations

Technical Limitations:

• Occlusion effects in TLS data may lead to underestimation of void connectivity
• MLS systems face challenges in complex terrain with dense understory vegetation [31]
• Voxelization introduces discretization errors, particularly with larger voxel sizes [32]

Implementation Recommendations:

• Combine TLS and MLS approaches to balance detail and coverage
• Validate void detection against field measurements in subset of study area
• Adjust voxel size based on research questions: finer resolutions (0.25-0.5m) for within-canopy analysis, coarser (1-2m) for landscape-scale patterns [32]

This protocol presents a comprehensive framework for quantifying forest void spaces using terrestrial and mobile LiDAR, addressing a critical gap in structural assessment methodology. By moving beyond traditional canopy-centric metrics, the approach provides novel insights into the three-dimensional organization of forest ecosystems and their functional implications. The standardized protocols for data acquisition, processing, and analysis ensure reproducible characterization of void patterns across diverse forest types and conditions. As LiDAR technology becomes increasingly accessible and computational methods continue to advance, this void-centric perspective offers promising avenues for understanding forest dynamics, predicting ecosystem response to environmental change, and informing conservation and management strategies aimed at maintaining structural complexity and biodiversity.

Terrestrial Laser Scanning (TLS), a ground-based Light Detection and Ranging (LiDAR) technology, has established itself as a cornerstone of modern habitat research, creating detailed, three-dimensional representations of forest ecosystems. However, the application of this powerful technology extends far beyond its traditional domain. TLS provides transformative capabilities for capturing the precise geometry and complex fabric of built environments, offering researchers and professionals in heritage science and structural engineering a tool for non-destructive, high-fidelity documentation and analysis [33]. This article details the application notes and experimental protocols for deploying TLS in these cross-disciplinary fields, framed within the methodological context of terrestrial LiDAR habitat research.

Application Note I: Heritage Documentation

In heritage documentation, TLS is pivotal for preserving cultural heritage by capturing the existing conditions of historic structures with millimeter-level accuracy. This process creates a critical "snapshot" of a site's state, which serves as an invaluable resource for preservation, restoration, and educational purposes [33]. The technology is particularly beneficial for sites that are in poor condition, structurally unstable, or pose access challenges, as it allows for meticulous documentation without physical contact [33].

The workflow involves a detailed TLS survey focused on achieving a high Level of Detail (LoD), which is essential for capturing unique architectural elements. Given the historical significance of these structures, which may not withstand repeated scanning, it is crucial to capture comprehensive data in the initial survey [33].

Quantitative Data and Outputs

TLS data acquisition generates rich datasets that can be processed into several key outputs for heritage management, as summarized in the table below.

Table 1: Primary Data Outputs from TLS Heritage Documentation

Data Output	Description	Primary Application in Heritage
3D Point Clouds	Dense collections of geometric points representing the scanned surface.	Primary record of existing conditions; base for all other derivatives.
Heritage Building Information Modeling (HBIM)	Structured, information-rich 3D model integrating geometric and semantic data.	Preservation planning, damage assessment, restoration design, and management [33].
2D Orthographic Drawings	Measured plans, elevations, and sections generated from the point cloud.	Architectural analysis and traditional archival records [33].
Interactive 3D Environments	Point clouds or meshes integrated into Virtual Reality (VR) or Augmented Reality (AR).	Public education, virtual tourism, and immersive site interpretation [33].

Experimental Protocol: Comprehensive Heritage Site Documentation

A. Project Planning and Pre-Field Preparation

• Research and Objective Definition: Establish the scope, required Level of Detail (LoD), and final deliverables (e.g., HBIM, 2D drawings, conservation report).
• Site Analysis: Conduct a preliminary site visit to identify challenges, scan positions, and necessary safety measures.
• Equipment Selection: Choose a TLS system suitable for the required range and resolution. For example, the FARO Focus3D x330 HDR laser scanner has been effectively used in complex heritage projects, such as the documentation of the historic schooner Equator [34].

B. Field Data Acquisition

• Scanner Setup: Position the scanner on a stable tripod, ensuring a clear line of sight to the target structure. The scan density should be configured to capture intricate architectural details [33].
• Scan Registration: Place registration targets (spheres or checkerboards) in the overlap between scan positions to facilitate the alignment of multiple scans into a single, unified point cloud.
• Data Capture: Execute scans from all planned positions. Supplement TLS data with UAV photogrammetry, as demonstrated in the Equator project, to capture upper sections of structures and provide photorealistic texturing [34].
• Supplementary Data Collection: Employ traditional hand-measuring techniques to capture specific details and validate scan accuracy, as recommended by the U.S. National Park Service's Heritage Documentation Programs [33].

C. Data Processing and Deliverable Generation

• Point Cloud Registration: Use specialized software to align individual scans into a complete 3D model using the registered targets.
• Data Cleaning and Modeling: Remove noise and erroneous data points. Develop HBIM models or surface meshes from the cleaned point cloud.
• Output Production: Generate the final agreed-upon deliverables, such as 2D CAD drawings extracted from the model, or interactive VR/AR experiences.

Figure 1: Workflow for comprehensive heritage site documentation using TLS and integrated technologies.

Application Note II: Structural Engineering and Built Environment

In structural engineering, TLS provides a reality-based modeling (RBM) solution for capturing "as-built" conditions of infrastructure with high precision. This capability is fundamental for structural health monitoring, deformation analysis, renovation planning, and quality control during construction. The technology enables the detection of minute deviations from design models and the monitoring of structural movements over time through periodic scans.

The workflow emphasizes accuracy and the integration of TLS data with engineering software platforms, particularly Building Information Modeling (BIM), to create a digital twin of the structure for analysis and project management.

Quantitative Sensor Data

The selection of a TLS sensor depends on the specific requirements of the engineering project, including required range, accuracy, and operational environment.

Table 2: Comparative Analysis of LiDAR Sensor Specifications for Engineering Applications

Sensor Model	Type/Platform	Estimated Cost (USD)	Key Specifications	Suitable Engineering Applications
FARO Focus3D x330 [34]	Terrestrial Laser Scanner	Market Varies	Range: 0.6m - 330m; Weight: 5.2kg	Large-scale building documentation, infrastructure monitoring, industrial plant modeling.
Velodyne HDL-32E [35]	UAS-borne / Mobile	~$175,000	32 channels;	Rapid topographic surveys, infrastructure inspection (bridges, dams).
Quanergy M8 [35]	UAS-borne / Mobile	~$80,000	8 channels; Lower cost.	Lower-resolution mapping for progress monitoring, stockpile volume measurement.

Experimental Protocol: Structural Deformation Monitoring

A. Baseline Scan and Control Network

• Establish Control Points: Create a network of stable, permanently marked control points around the structure to be monitored. These points must be located on stable ground, unaffected by the structure's potential movement.
• Baseline Scanning: Perform a high-resolution TLS survey of the structure, ensuring all control points are included in the scans. This initial dataset serves as the baseline for all future comparisons.

B. Periodic Monitoring and Analysis

• Repeat Scanning: Conduct subsequent TLS surveys at predetermined intervals (e.g., monthly, quarterly) or following specific events (e.g., earthquakes, major excavations nearby). Use the same scanner and scan positions for consistency.
• Co-Registration and Change Detection: Precisely align the periodic scan data to the baseline model using the stable control points. Software is then used to compute the 3D differences between the point clouds, highlighting areas of deformation or movement.

C. Reporting and Integration

• Deformation Mapping: Generate color-mapped models that visually represent the magnitude and direction of structural movement.
• BIM Integration: Import the deformation data and the accurate as-built point cloud into the project's BIM model to inform structural analysis and maintenance decisions.

Figure 2: Structural deformation monitoring workflow using periodic TLS for change detection.

The Scientist's Toolkit: Essential Research Reagents & Materials

Successful implementation of TLS projects requires a suite of hardware, software, and field equipment.

Table 3: Key Research Reagent Solutions for TLS Applications

Tool Category	Specific Examples	Function & Explanation
TLS Hardware	FARO Focus3D x330 [34]	Captures the primary 3D point cloud data. Selection depends on required range and accuracy.
Supplemental Sensors	UAV (e.g., DJI Mini 3 Pro) [34]	Provides aerial imagery for photogrammetric models, complementing TLS data for hard-to-reach areas.
Registration Targets	Spheres, Checkerboards	Act as reference points to accurately align multiple individual scans into a unified coordinate system.
Processing Software	CAD, BIM, Point Cloud Processing Software	Used for cleaning, modeling, analyzing point clouds, and generating deliverables like HBIM and 2D drawings [33].
Data Storage	High-Capacity Portable Drives	TLS projects generate massive datasets (hundreds of GB), requiring robust storage solutions for raw data and backups [34].

Application Notes

The Role of Advanced Segmentation in TLS LiDAR Habitat Research

Terrestrial Laser Scanning (TLS) has revolutionized forest ecology research by providing highly detailed, three-dimensional point clouds of forest structures. The application of Machine Learning (ML) and Deep Learning (DL) to segment these point clouds into individual tree components represents a critical methodological advancement for quantifying habitat characteristics. This transition from hand-crafted algorithms to data-driven approaches has enabled researchers to overcome long-standing challenges in measuring complex forest environments, particularly in dense canopies with overlapping crowns where traditional methods face significant limitations [36] [4]. These advanced segmentation techniques are transforming our ability to quantify ecosystem dynamics and generate digital twins of forest habitats that support biodiversity monitoring, carbon stock assessment, and climate change research [29] [4].

Performance Comparison of Segmentation Approaches

Table 1: Performance comparison of machine learning and deep learning models for tree structure segmentation.

Model	Model Type	Best F1-Score (Stem)	Optimal Input Features	Point Sampling	Computational Time	Key Advantages
XGBoost	Machine Learning	87.8% [37]	All features (S+G+L) [37]	8192 points [37]	10-47 minutes [37]	Computational efficiency, feature importance scores [37]
PointNet++	Deep Learning	92.1% [37]	Spatial coordinates & normals only [37]	4096 points [37]	49-168 minutes [37]	Superior accuracy, complex pattern recognition [37]
TreeLearn	Deep Learning	Outperformed SegmentAnyTree, ForAINet, TLS2Trees [36]	Training on pre-segmented data [36]	Not specified	Fully automatic pipeline [36]	Less reliant on predefined features, easy to use [36]
PointMLP	Deep Learning	96.94% (species classification OA) [38]	NGFPS sampling [38]	1024-2048 points [38]	Not specified	Robust streamlined solution [38]

Table 2: Tree species classification accuracy comparison across algorithms.

Model	Classification Task	Overall Accuracy	Data Source	Key Findings
PointMLP	4 species classification [38]	96.94% [38]	UAV-LiDAR [38]	Most accurate for species ID [38]
Random Forest	4 species classification [38]	95.62% [38]	UAV-LiDAR [38]	Strong traditional ML approach [38]
SVM	4 species classification [38]	94.89% [38]	UAV-LiDAR [38]	Competitive performance [38]
PointNet++	4 species classification [38]	85.65% [38]	UAV-LiDAR [38]	Lower accuracy than PointMLP [38]
XGBoost	4 species classification in Poland [38]	96% [38]	TLS [38]	Excellent species discrimination [38]

Ecological and Habitat Applications

The integration of ML/DL segmentation with TLS data enables novel approaches to quantifying habitat features that are fundamental to ecological research. The LiDAR-based framework for forest void analysis exemplifies this advancement, treating voids as three-dimensional unoccupied spaces between vegetation that govern light penetration, airflow, and habitat connectivity [29]. Structurally heterogeneous forests with multi-layered canopies exhibit diffuse, vertically extensive voids, while uniform stands contain more confined voids largely restricted to lower strata [29]. This structural lens on spatial openness provides valuable insights for biodiversity monitoring, habitat suitability assessment, and climate-adaptation research [29].

Experimental Protocols

Data Acquisition and Preprocessing Protocol

TLS Data Collection Specifications

• Scanner Configuration: Use a tripod-mounted TLS system (e.g., Riegl VZ-400i) with time-of-flight measurement principle, 1550 nm wavelength, and capability for multiple returns per pulse (up to 8 returns) [39].
• Scan Positioning: Establish 9-12 scan positions per plot arranged evenly around a circular or square plot with approximately 16 m spacing between positions [37] [39].
• Field of View Optimization: Combine standard upright scanner configuration (100° vertical, 360° horizontal FOV) with tilted configuration (360° vertical, 100° horizontal FOV) using a tilt-mount to achieve comprehensive hemispherical coverage [39].
• Georeferencing: Utilize integrated GNSS and IMU sensors for real-time position and orientation estimation, supplemented by 5 ground control points (GCPs) measured with survey-grade GNSS receivers (e.g., Trimble R12i) with horizontal tolerance <3 cm [37] [39].
• Environmental Considerations: Conduct scanning during leaf-off conditions for deciduous forests to maximize stem visibility and structural characterization [39].

Point Cloud Preprocessing Workflow

• Registration: Perform initial alignment using feature point matching between adjacent scans with cloud-to-cloud distance method (registration error <0.02 m), followed by fine alignment using Iterative Closest Point (ICP) algorithm (final error <0.005 m) [37].
• Geometric Correction: Transform registered point clouds into absolute coordinate system using GCPs with root mean square error (RMSE) maintained within 3 cm [37].
• Noise Filtering: Apply distance-based filters to remove outliers and implement voxel-based thinning (1-mm voxelization) to reduce oversampling of nearby surfaces while preserving crown density [39].
• Normalization: Identify ground points using Axelsson's method to create a digital elevation model, then normalize Z-coordinates to represent height above ground [39].

Individual Tree Extraction and Segmentation Protocol

Tree Detection and Delineation

• Stem Identification: Extract a horizontal slice of the point cloud (30 cm thick) between heights of 1.5-2.5 m where stem cross-sections form circular patterns [39].
• Point Filtering: Filter slice points using scanner-derived attributes - exclude points with Reflectance < -2.5 dB or Deviation > 3 to remove branches and retain stem points [39].
• Circle Detection: Apply Hough Circle Transform algorithm with minimum separation of 0.1 m to identify circular clusters representing stem cross-sections when projected onto XY plane [39].
• Validation Metrics: Evaluate detected circles using 'circle completeness' (assesses stem circumference capture) and 'goodness of fit' (quantifies circle-to-point distribution match) to identify false positives [39].
• Seed Point Generation: Use identified tree locations as seed points for bottom-up segmentation process employing 3D Dijkstra Region Growing algorithm to delineate individual trees including stems and branches [39].

Feature Engineering for ML/DL Models

• Spatial Features: X, Y, Z coordinates and normal vectors (S features) [37]
• Geometric Structure Features: Curvature, linearity, planarity, and other geometric descriptors (G features) [37]
• Local Distribution Features: Point density, vertical distribution, and spatial distribution metrics (L features) [37]
• Feature Combinations: Systematically evaluate model performance using S, S+G, S+L, and S+G+L feature combinations to determine optimal input configuration [37]

Model Training and Evaluation Protocol

Data Preparation for Model Training

• Downsampling Strategy: Implement hybrid downsampling combining random sampling and Farthest Point Sampling (FPS) to balance point density and computational efficiency [37]. Evaluate multiple point densities (2048, 4096, 8192 points) to determine optimal sampling level [37].
• Data Partitioning: Split individual tree point clouds into training, validation, and test sets using 6:2:2 ratio [37].
• Data Augmentation: Apply rotation, scaling, and jittering to increase dataset diversity and improve model generalization [36].

Model Implementation and Training

• XGBoost Configuration: Utilize all feature categories (S+G+L) with 8192 points for optimal performance. Monitor feature importance scores to interpret model decisions and identify most predictive features [37].
• PointNet++ Configuration: Implement with spatial coordinates and normals only (S features) with 4096 points for optimal performance. Leverage hierarchical feature learning to automatically capture complex structural patterns [37].
• TreeLearn Implementation: Pre-train on large-scale automatically segmented data (e.g., 6665 trees labeled using Lidar360 software), then fine-tune with manually annotated datasets to substantially improve performance [36].
• Training Regimen: Implement early stopping with patience of 20 epochs, use Adam optimizer with initial learning rate of 0.001, and batch sizes of 16-32 depending on point density [37].

Model Evaluation and Validation

• Performance Metrics: Calculate precision, recall, and F1-score for each tree structure component (stem, crown, ground) using confusion matrices [37].
• Error Analysis: Document systematic missegmentation patterns - XGBoost typically confuses structures near stem-to-ground boundaries and branch junctions, while PointNet++ occasionally missegments complex regions between stems and crowns [37].
• Ecological Validation: Compare derived structural metrics (DBH, tree height, crown volume) with field measurements to ensure ecological relevance and accuracy [39].

Visualization

Experimental Workflow for Tree Structure Segmentation

Method Selection Logic for Segmentation Tasks

The Scientist's Toolkit

Table 3: Essential research reagents and computational tools for advanced tree segmentation.

Tool/Reagent	Specifications/Type	Primary Function	Example Applications
Terrestrial Laser Scanner	Riegl VZ-400i [39] or Leica BLK360 [37]	3D point cloud acquisition with multiple returns	Forest structure digitization, habitat mapping [4] [39]
Georeferencing System	Trimble R12i GNSS receiver [37]	Precise positioning of scan locations	Absolute coordinate transformation, multi-temporal alignment [37]
Registration Software	RiSCAN PRO [39] or Cyclone [37]	Point cloud alignment and co-registration	Multi-scan integration, geometric correction [37] [39]
Tree Segmentation Algorithm	LIS TreeAnalyzer [39] or Custom ML/DL	Individual tree extraction from plot clouds	Stem detection, crown delineation [39]
Machine Learning Framework	XGBoost [37]	Tree structure segmentation using handcrafted features	Stem identification, computational efficiency [37]
Deep Learning Framework	PointNet++ [37] or PointMLP [38]	End-to-end segmentation with automatic feature learning	Complex structure recognition, species classification [37] [38]
Benchmark Datasets	Manually segmented forest plots (156+ trees) [36]	Model training and validation	Algorithm development, performance evaluation [36]
Point Cloud Processing	CloudCompare, LASTools [40]	Data cleaning, filtering, and analysis	Noise removal, feature extraction [40]

Overcoming Practical Challenges: A Guide to Optimizing TLS Data Collection and Processing

In Terrestrial Laser Scanning (TLS), occlusion occurs when parts of the target, such as branches or foliage, block the scanner's line of sight to areas behind them, creating gaps in the resulting point cloud [41]. This happens because laser pulses cannot penetrate solid objects [41]. In the context of LiDAR habitat research, occlusion presents a fundamental challenge for quantifying the three-dimensional arrangement of plant components, which is fundamental for characterizing forest ecosystems [11] [4]. Unmitigated occlusion leads to non-representative sampling, biased structural metrics, and inaccurate estimates of critical ecological parameters such as Leaf Area Density (LAD), biomass, and canopy volume [42]. Therefore, a systematic approach combining strategic scan positioning and robust multi-scan registration is essential for generating complete and accurate 3D representations of habitat structure.

Strategic Scan Positioning

The primary strategy for mitigating occlusion involves capturing the scene from multiple positions to ensure that every element of the vegetation is visible from at least one scanner viewpoint.

Positioning for Individual Trees

For an individual tree, such as an open-grown street tree, scanning from four cardinal directions (North, South, East, West) is recommended to capture the full circumference of the trunk and major branches [41]. Additional scans at different angles and closer distances are often necessary to obtain a detailed view of the upper canopy. A typical protocol involves four to six scan locations spaced at regular intervals around the tree, positioned at distances between 1 and 15 meters from the tree base, depending on the tree's size and structural complexity [41].

Positioning for Forest Plots

For group of trees, such as an urban forest stand or a research plot, a systematic grid-based approach is required [41]. The spacing of this grid is determined by vegetation density:

• For dense forest patches, a 10 x 10 meter grid is effective.
• For relatively open forest areas, a wider 20 x 20 meter grid can be sufficient [41].

This grid ensures comprehensive coverage and minimizes shadowing effects from dense vegetation.

Environmental and Technical Considerations

• Weather Conditions: Scanning must be avoided during rain, snow, or dense moisture to protect sensitive scanner components. Wind speeds should be less than 1 meter per second to prevent branch sway from causing a "ghost effect" in the point cloud [41].
• Scan Resolution and Quality: Higher resolution provides greater detail but increases scanning time and data storage requirements. The optimal setting is a balance between the required level of structural detail and logistical constraints [41].

Multi-Scan Registration Techniques

After data acquisition from multiple positions, the individual point clouds must be aligned into a single, coherent model through a process known as co-registration or registration [41] [43].

Registration Methods

The choice of registration method significantly impacts the accuracy and efficiency of the final model. The table below summarizes the primary methods.

Table 1: Comparison of Point Cloud Registration Methods [43]

Method	Description	Key Requirements	Advantages	Limitations
Target-Based	Uses physical targets (e.g., spheres, checkerboards) placed in the scene as common reference points.	At least four common targets visible between consecutive scan positions [41].	High accuracy; enables automated processing.	Requires field preparation and target placement.
Cloud-to-Cloud	Software algorithm aligns clouds based on common geometric features without targets.	Overlapping areas with distinct geometric features.	Fast; no need for physical targets.	Less accurate on featureless or repetitive structures.
Manual Visual Alignment	User manually aligns point clouds based on visual interpretation.	User expertise and a good understanding of the scene.	Practical when automated methods fail.	Time-consuming and subjective.

Protocol for Target-Based Registration

This method is widely recommended for ecological applications due to its high accuracy [41] [43].

• Place Targets: Before scanning, place retroreflective spherical or checkerboard targets around the area of interest. Ensure they are stable and visible from multiple angles.
• Ensure Visibility: A minimum of four common reference targets must be visible from consecutive scan positions to allow software to automatically detect and align the scans [41].
• Scan the Scene: Perform scans from all planned positions, ensuring targets are captured in each.
• Automated Co-registration: Use specialized software to automatically detect the identical targets and merge the multiple scans into a complete point cloud [41].

Advanced Occlusion Mitigation and Data Analysis

Statistical Correction for Occlusion

Even with optimal scan setups, some occlusion may persist. Advanced statistical methods, such as LAD-kriging, have been developed to address this. This geostatistical approach uses the spatial correlation of the LAD field to improve estimation accuracy in poorly sampled voxels, mitigating the bias caused by occlusion without requiring arbitrary reliability thresholds [42].

Quantitative Data from Registered Point Clouds

Once a complete point cloud is generated, Quantitative Structure Models (QSMs) can be used to derive ecological metrics. These are algorithmic enclosures of point clouds into topologically-connected, closed volumes, enabling precise measurements [11] [4].

Table 2: Key Ecological Parameters Derived from TLS Point Clouds [11] [42]

Parameter	Description	Ecological Significance
Leaf Area Density (LAD)	The density of photosynthetically active vegetation elements per unit volume.	Critical for modeling eco-physiological processes like canopy photosynthesis and transpiration.
Tree Architecture	The 3D size and arrangement of a tree's fundamental components (stems, branches, leaves).	Influences and responds to environmental changes; regulates light regimes and productivity.
Biomass	The total mass of organic material in the tree.	Key for carbon stock inventories and understanding carbon cycling.
Canopy Gap Fraction	The proportion of sky visible through the canopy at a given point.	Informs light availability for understory growth and habitat structure.

Workflow Diagram: Strategic Scanning & Registration

The following diagram visualizes the end-to-end workflow for mitigating occlusion in TLS habitat research.

The Researcher's Toolkit

Table 3: Essential Research Reagents and Equipment for TLS Habitat Studies

Item	Function
Terrestrial Laser Scanner	The core instrument emits laser pulses to capture 3D point clouds of the environment. Modern systems are lighter, faster, and more affordable [11] [41].
Tripod & Leveling Equipment	Provides a stable platform for the scanner. The built-in inclinometer is crucial for leveling to ensure measurement accuracy [41].
Retroreflective Targets (Spheres/Checkerboards)	Act as common reference points (targets) for accurately merging (co-registering) multiple scans into a single model [41] [43].
Specialized Registration Software	Software tools that implement algorithms (e.g., target-based, cloud-to-cloud) to align and combine individual point clouds [43].
High-Capacity Batteries	TLS instruments are power-intensive. Extra batteries are needed for uninterrupted data collection during field campaigns [41].
Quantitative Structure Model (QSM) Algorithms	Computational methods that convert point clouds into topologically-connected, closed volumes for deriving tree metrics like volume and biomass [11].

In terrestrial laser scanning (TLS) for habitat research, the transition from data acquisition to actionable ecological insights is often hindered by significant computational bottlenecks. Effective processing of large-scale point cloud data—encompassing registration, denoising, and downsampling—is a critical prerequisite for accurate habitat modeling and analysis [44] [45]. This article details streamlined protocols and application notes to overcome these bottlenecks, providing researchers with efficient strategies tailored for ecological applications.

Efficient Point Cloud Downsampling

Downsampling reduces data volume while preserving essential geometric features, drastically lowering computational costs for subsequent processing and analysis.

The DFPS Downsampling Protocol

The DFPS (Efficient Downsampling Algorithm for Global Feature Preservation) algorithm is designed for large-scale point cloud data. It combines an adaptive multi-level grid partitioning mechanism with a multithreaded parallel computing architecture, achieving significant efficiency gains without requiring GPU acceleration [46].

Experimental Protocol:

• Input Raw Data: Load the raw point cloud data and set a target sampling rate (e.g., 12.5%, 3.125%).
• Initial Partitioning: Divide the entire point cloud space into eight equal segments.
• Adaptive Threshold Check: For each segment, check if the number of points (N0) exceeds the minimum threshold (Nmin, often set to 256). If N0 ≤ Nmin, proceed to grid merging and recombination.
• Hierarchical Grid Partitioning: If N0 > Nmin, initiate the adaptive hierarchical grid partitioning. This step uses a first-round farthest-point sampling to dynamically adjust weights for local detail preservation, which in turn reduces the computational load for the second-round sampling.
• Iterative Farthest-Point Sampling: Perform iterative sampling within the partitioned grids. The process is controlled by a parameter β, which allows manual calibration of the weight assigned to preserving local details, catering to different research needs [46].

Table 1: Performance Benchmark of DFPS vs. Traditional FPS

Sampling Rate	Traditional FPS Processing Time	DFPS Processing Time	Efficiency Gain
12.5%	~161,665 s	~71.64 s	>2200 times
3.125%	Not Reported	Not Reported	~10,000 times

The following workflow diagram illustrates the DFPS downsampling process:

DFPS Downsampling Workflow

Robust Point Cloud Denoising

Denoising removes spurious points caused by sensor limitations or environmental factors, which is crucial for accurate geometric analysis of habitats, such as measuring tree bark texture or ground surface roughness.

Robust Bilateral Filtering with Improved Normals

Bilateral filtering is a common denoising technique, but its performance is highly dependent on the accuracy of estimated point normals. A novel method improves upon this by refining normal estimation, particularly for edge points [47].

Experimental Protocol:

• Initial Normal Estimation: Use Principal Component Analysis (PCA) to compute the initial normals for all points in the cloud.
• Point Classification: Categorize points into planar and edge types by fitting a three-dimensional sphere to the local neighborhood.
• Normal Refinement: Apply Iterative Weighted PCA exclusively to edge points. This step uses a robust estimator to compute neighbor weights, significantly improving normal direction accuracy at sharp features.
• Bilateral Filtering: Execute bilateral filtering using the improved normals. The spatial and range filters in the bilateral filter can also use a robust estimator instead of a standard Gaussian function for greater resilience to noise [47].

Performance Metrics: The performance of this denoising protocol can be evaluated using:

• Mean Square Angular Error (MSAE) and Angular Error Distribution (AE): For assessing the accuracy of normal estimation.
• Mean Square Error (MSE) and Signal-to-Noise Ratio (SNR): For assessing the final denoising performance [47].

Table 2: Denoising and Normal Estimation Methods Comparison

Method Category	Example Methods	Key Principle	Reported Advantage
Traditional Denoising	Robust Bilateral Filtering [47]	Improved normal estimation via PCA & classification	Most accurate normals, smallest MSE in study
Deep Learning Denoising	PointFilter [48]	Learns per-point displacement vectors	Bilateral projection loss for sharp edges
Deep Learning Denoising	DMRDenoise [48]	Downsampling-Upsampling strategy	Learns distinctive representations from data
Normal Estimation	Jet, VCM [47]	Variants of local surface fitting	Outperformed by proposed robust method

Point Cloud Registration for Multi-View Scans

Registration aligns multiple point clouds from different scanner positions into a unified coordinate system, which is fundamental for creating complete habitat models.

Coarse-to-Fine Registration Framework

A common and effective strategy for point cloud registration involves a two-stage process: coarse registration followed by fine registration [44] [45].

Experimental Protocol:

• Data Preprocessing:
  • Equal Density Dilution: Process mobile and terrestrial point clouds to achieve a more uniform point density, mitigating differences in data acquisition methods [49].
  • Primitive Extraction: Reduce scene complexity by extracting stable, artificial ground objects (e.g., road curbs, building corners) or natural features with distinct geometry to serve as registration primitives [49].
• Coarse Registration:
  • Keypoint Detection: Use a multi-scale keypoint extraction method, often based on octree voxel indices and principal curvature attributes, to identify distinctive feature points on the registration primitives [49].
  • Feature Matching: Find corresponding keypoints between the two point clouds (the "source" and "target"). Algorithms like 4-Points Congruent Sets (4PCS) are commonly used for this stage to obtain an initial, rough alignment [49].
• Fine Registration:
  • Iterative Closest Point (ICP): Refine the coarse alignment using the ICP algorithm or one of its many variants. ICP iteratively minimizes the distance between points in the source cloud and their closest points in the target cloud, resulting in a highly precise transformation [45] [49].
  • Keypoint Constraint: To improve ICP's performance in complex urban or natural habitats, constraints from the previously identified keypoints can be integrated to guide the convergence and avoid local minima [49].

The logical flow of the core registration process is shown below:

Coarse-to-Fine Registration Logic

Table 3: Registration Techniques for TLS Point Clouds

Registration Stage	Technique	Description	Considerations for Habitat Research
Coarse	4-Points Congruent Sets (4PCS) [44]	Finds approximate alignment using wide-base congruent sets	Works with low overlap; sensitive to repetitive vegetation structures.
Coarse	Keypoint-based (e.g., RoPS) [45]	Extracts and matches salient local features	Feature detection can be challenging on organic, complex surfaces like foliage.
Fine	Iterative Closest Point (ICP) [45] [49]	Precise alignment by minimizing point-to-point distances	High computational cost; requires good initial position from coarse stage.
Fine	Keypoint-Constrained ICP [49]	ICP guided by known feature correspondences	Increases robustness and accuracy in complex scenes (e.g., forests).

The Scientist's Toolkit

Table 4: Essential Research Reagent Solutions for Point Cloud Processing

Category / 'Reagent'	Function in Protocol	Exemplar Tools / Methods
Downsampling Algorithms	Reduces data volume for manageable processing while preserving global features.	DFPS (For efficiency), FPS (For feature preservation) [46]
Normal Estimation Methods	Estimates surface orientation, critical for denoising and feature detection.	Weighted PCA, Jet, VCM [47]
Denoising Filters	Removes sensor noise and outliers to improve geometric fidelity.	Robust Bilateral Filtering, Moving Least Squares (MLS) [47] [48]
Registration Primitives	Stable features used as anchors for aligning different scans.	Artificial Ground Objects, Planar Patches, Keypoints [49]
Benchmark Datasets	Provides standardized data for objective method evaluation and comparison.	WHU-TLS Dataset, ETH Zurich TLS Scenes [44]
Data Labeling Platforms	Enables annotation for training deep learning models (e.g., for segmentation).	Encord, Labelbox, Scale AI [50]

The analysis of Terrestrial Laser Scanning (TLS) LiDAR data is fundamental to advancing habitat research, providing unprecedented three-dimensional structural details of ecosystems [11]. A critical step in this analysis is segmentation, the process of partitioning point clouds into meaningful regions or objects, which enables the quantification of habitat features [51]. Researchers face a fundamental choice in their analytical pipeline: selecting between traditional Machine Learning (ML) algorithms and more complex Deep Learning (DL) models. This selection entails a direct trade-off between computational efficiency and predictive accuracy, a balance that dictates the feasibility and scope of research projects. These application notes provide a structured comparison and detailed protocols to guide this decision-making process, framed specifically within the context of terrestrial LiDAR habitat research.

Core Segmentation Concepts and TLS Relevance

Segmentation Types in Point Cloud Analysis

For TLS data, segmentation techniques can be categorized by their objective, each serving a distinct purpose in habitat analysis [51]:

• Semantic Segmentation: Assigns a class label (e.g., "leaf," "branch," "ground") to every point in the cloud. This is crucial for calculating biophysical parameters like leaf area index or distinguishing vegetation from bare earth [11] [52].
• Instance Segmentation: Identifies and delineates individual objects within a class, such as separating one tree from another in a dense stand. This is essential for tree censuses and studying competition [51].
• Panoptic Segmentation: A unified approach that combines semantic and instance segmentation, providing a comprehensive scene understanding by classifying all points and identifying all object instances [51].

The Role of Segmentation in TLS Habitat Research

TLS captures extremely detailed 3D measurements of forest ecosystems, enabling the creation of highly accurate "digital twins" for research [11]. Segmentation is the key that unlocks this data, allowing researchers to:

• Quantify Tree Architecture: Understand the 3D size and arrangement of tree components, which influences light regimes and productivity [11].
• Assess Habitat Availability: Map structural habitats for species of conservation concern, such as identifying early successional forest nesting sites for Golden-winged Warblers [52].
• Drive Ecological Modeling: Provide structural parameters for Functional Structural Plant Models (FSPMs) and radiative transfer models, bridging the gap between structure and function [11].

Comparative Analysis: Machine Learning vs. Deep Learning

The following table summarizes the core characteristics of ML and DL approaches for segmenting TLS-derived point clouds.

Table 1: Comparative Analysis of ML and DL Segmentation Models for TLS Data

Aspect	Machine Learning (ML) Models	Deep Learning (DL) Models
Typical Algorithms	K-Means, DBSCAN, Hierarchical Clustering, Support Vector Machines (SVMs), Random Forests [53]	Convolutional Neural Networks (CNNs), U-Net, PointNet++, Graph Convolutional Networks (GCNs) [53] [51] [54]
Computational Efficiency	Generally high efficiency. Lower hardware requirements; can often run on powerful CPUs. Training and inference are less computationally intensive [53] [55].	Generally low efficiency. Requires high-performance GPUs with substantial VRAM (e.g., NVIDIA RTX 3080/3090+). Training is highly resource-intensive, though inference can be optimized [53] [55].
Typical Accuracy	Moderate to Good. Effective for well-defined tasks based on handcrafted features (e.g., height, density). Accuracy can plateau with complex, heterogeneous structures [53] [54].	High to State-of-the-Art. Excels at learning complex, hierarchical features directly from data. Achieves superior performance on tasks like fine-scale branch and leaf classification [53] [51].
Data Dependencies	Lower volume requirements. Performance relies heavily on quality feature engineering and domain expertise [53].	Requires large datasets. Performance is dependent on the quality, quantity, and diversity of annotated training data [53] [54].
Interpretability	High. Models like decision trees are often more transparent, and the role of handcrafted features is clear [53].	Low (Black-Box). The internal workings and basis for predictions are complex and difficult to interpret [53].
Ideal TLS Use Cases	- Initial data exploration and preprocessing- Segmentation based on geometric rules (e.g., ground classification)- Projects with limited data or computational resources [53] [52]	- Complex scene understanding (e.g., full tree architecture)- Fine-scale instance segmentation (e.g., individual leaves, fruits)- Large-scale, high-throughput analysis [11] [51]

The workflow for selecting and applying a segmentation model involves a series of logical steps, from data preparation to final model deployment, as illustrated below.

Experimental Protocols

Protocol 1: Habitat Mapping with Traditional Machine Learning

This protocol is adapted from methodologies used in studies like McNeil et al. (2023) to identify potential wildlife habitat using LiDAR-derived structural metrics [52].

1. Objective: To segment a TLS point cloud of a forest area into distinct structural classes (e.g., ground, understory, mature trees) to map potential habitat for a target species.

2. Materials & Data:

• Input Data: TLS point cloud from a single or multiple registered scans [11].
• Software: Libraries such as Scikit-Learn for ML algorithms, and point cloud processing tools (e.g., Open3D, CloudCompare).

3. Step-by-Step Methodology:

• Step 1: Preprocessing. Clean the point cloud by applying noise removal and downsampling to ensure manageable data size.
• Step 2: Feature Extraction. Compute handcrafted geometric features for each point or a segmented region. Key features include:
  • Height above ground.
  • Point density within a specified sphere.
  • Surface normals and their variation.
  • Eigenvalue-based features (linearity, planarity, sphericity).
• Step 3: Algorithm Selection and Training.
  • For unsupervised tasks (e.g., broad structural classes), use K-Means or DBSCAN clustering on the extracted features [53].
  • For supervised tasks (e.g., classifying "suitable" vs. "unsuitable" habitat), use Random Forest or Support Vector Machines (SVMs). The model is trained on a labeled subset of the data [53] [52].
• Step 4: Segmentation & Validation.
  • Apply the trained model to the entire point cloud.
  • Validate results against manually annotated reference plots or field survey data. Metrics include overall accuracy and Cohen's Kappa [52].

4. Expected Outcomes:

• A labeled point cloud where each point is assigned a structural class.
• Quantified area and spatial distribution of each habitat class.

Protocol 2: Instance Segmentation of Trees Using Deep Learning

This protocol outlines a process for detailed individual tree segmentation, a task essential for creating Quantitative Structure Models (QSMs) [11].

1. Objective: To perform instance segmentation on a TLS point cloud to identify and separate individual trees, including their major structural components.

2. Materials & Data:

• Input Data: A large, curated dataset of TLS point clouds with instance labels for trees and potentially branches/trunks.
• Hardware: High-performance GPU (e.g., NVIDIA RTX 3090 with 24GB VRAM)[ccitation:6].
• Software: Deep learning frameworks like PyTorch or TensorFlow, and libraries for 3D deep learning (e.g., PyTorch3D, TorchPoints3D).

3. Step-by-Step Methodology:

• Step 1: Data Preparation.
  • Partition the dataset into training, validation, and test sets.
  • Apply data augmentation techniques such as random rotation, scaling, and jittering to improve model generalization [53].
• Step 2: Model Selection & Configuration.
  • Select a DL architecture designed for point clouds, such as PointNet++ or a Graph Convolutional Network (GCN) [51].
  • Configure the model's output for instance segmentation, typically involving a dual output for semantic class and instance ID.
• Step 3: Model Training.
  • Use a supervised learning approach with the annotated data.
  • Employ a loss function that combines semantic classification loss (e.g., cross-entropy) and instance discrimination loss.
  • Utilize an optimizer (e.g., Adam) and techniques like learning rate scheduling [53].
• Step 4: Inference and Post-processing.
  • Apply the trained model to new, unseen TLS scans.
  • Use post-processing techniques like non-maximum suppression to refine instance masks.

4. Expected Outcomes:

• A point cloud where each point is assigned to a specific tree instance.
• Derived metrics for each tree, such as height, stem diameter, and crown volume, which can be used to construct QSMs [11].

The Scientist's Toolkit: Essential Research Reagents & Materials

This table details key resources required for implementing the segmentation protocols described above.

Table 2: Essential Research Toolkit for TLS LiDAR Segmentation

Category	Item / Solution	Function & Application Notes
Hardware	High-End Workstation	Runs ML/DL models and processes large point clouds. Requires a powerful CPU (e.g., Intel i9/AMD Ryzen 9), ample RAM (32GB+), and a high-VRAM GPU (e.g., NVIDIA RTX 3080/3090) [55].
Hardware	Terrestrial Laser Scanner	Acquires the raw 3D point cloud data. Modern TLS systems are lighter, faster, and more efficient, reducing fieldwork bottlenecks [11].
Software & Libraries	Scikit-Learn	Provides robust implementations of traditional ML algorithms like K-Means, DBSCAN, and Random Forests for prototyping and analysis [53].
Software & Libraries	PyTorch / TensorFlow	Core open-source frameworks for developing and training deep learning models, including custom architectures for point cloud segmentation [53] [51].
Software & Libraries	Open3D / PCL	Libraries for 3D data processing. Used for point cloud visualization, preprocessing (denoising, downsampling), and basic geometric operations.
Data Resources	Custom Annotated Datasets	High-quality, finely annotated point clouds are critical for supervised DL. Services like BasicAI can provide expert data annotation to overcome this bottleneck [51].
Data Resources	Benchmark Datasets (e.g., FTW)	While often for aerial data, benchmarks like "Fields of The World" (FTW) provide examples of large-scale, multi-domain datasets for developing generalizable models [56].
Methodological Frameworks	Quantitative Structure Models (QSMs)	Algorithmic enclosure of point clouds into topologically-connected volumes. Used to derive biomass and other ecological metrics from segmented tree data [11].

The following diagram visualizes the structured workflow of the deep learning experimental protocol (Protocol 2), from data preparation to final model application.

The selection between machine learning and deep learning for TLS LiDAR segmentation is not a question of which is universally better, but which is more appropriate for a given research context. Traditional ML algorithms offer a robust, interpretable, and computationally efficient path for projects with limited data, well-defined structural features, or a need for high transparency. In contrast, deep learning models provide superior accuracy and automation for complex segmentation tasks like fine-scale instance segmentation, at the cost of significant data, computation, and reduced interpretability.

Future developments in weakly-supervised learning [54], more efficient model architectures, and the use of synthetic data [53] will help mitigate some challenges of DL. By aligning project goals, available resources, and analytical requirements with the strengths and limitations of each paradigm as outlined in these protocols, habitat researchers can strategically leverage TLS technology to advance our understanding of ecosystem structure and function.

Terrestrial Laser Scanning (TLS) has emerged as a transformative technology for heritage documentation, enabling the capture of highly detailed, accurate, and measurable 3D data for conservation, research, and education. This application note synthesizes guidance from real-world case studies and established conservation practices, providing researchers with structured methodologies for deploying TLS in heritage contexts. The protocols outlined herein balance technological capabilities with the practical demands of field research, ensuring that data collection meets the stringent requirements for preservation science while remaining feasible for operational implementation. By establishing best practices for planning, data acquisition, processing, and analysis, this document serves as an essential resource for researchers integrating TLS into heritage conservation workflows.

Heritage documentation (HD) serves as a critical repository of cultural, historical, and architectural knowledge, providing invaluable data for preservation, restoration, and educational purposes [33]. As defined by Stylianidis, HD is a continuous process that enables the monitoring, maintenance, and understanding needed for conservation through the supply of appropriate and timely information [33]. Within this framework, Terrestrial Laser Scanning (TLS), a ground-based form of Light Detection and Ranging (LiDAR) technology, utilizes laser sensors to collect comprehensive point clouds that depict physical objects with exceptional accuracy [33]. This non-contact method is particularly valuable for documenting fragile or structurally compromised heritage sites where physical contact may cause damage [33].

The transition from traditional documentation methods to TLS represents a paradigm shift in conservation practice. Traditional HD methods, like measured drawings, large-scale photography, and written reports, while reliable, face limitations in terms of labor intensity, time consumption, and the precision needed in modern conservation contexts [33]. TLS addresses these limitations by enabling the rapid and precise collection of dense point clouds, capturing the geometry and fabric of scanned environments in intricate detail [33]. The rich datasets generated by TLS can be transformed into various outputs, including Heritage Building Information Models (HBIM), 2D drawings, and interactive 3D environments for virtual reality (VR) and augmented reality (AR) applications, thereby enhancing access to and engagement with cultural heritage sites [33].

Technical Foundation of Terrestrial LiDAR

Core Operating Principles

TLS systems operate on the principle of Time-of-Flight (ToF) measurement, calculating distance by measuring the time interval between the emission of a laser pulse and the detection of its reflected signal [57]. The fundamental distance equation is:

d = t~ToF~ · c / 2

where d is the distance to the target, t~ToF~ is the measured time-of-flight, and c is the speed of light [57]. As laser pulses are emitted toward surfaces, some photons reflect off objects such as architectural elements or vegetation, while others continue until they hit subsequent surfaces or are fully absorbed [58]. This behavior enables the recording of multiple returns from a single laser pulse, providing crucial information about structure and density [58].

LiDAR Data Outputs and Characteristics

The primary data output from TLS surveys is a 3D point cloud, where each point possesses specific attributes that define its position and characteristics [58]. The table below summarizes the fundamental attributes of LiDAR point cloud data:

Table 1: Fundamental LiDAR Point Cloud Data Attributes

Attribute	Description	Research Application
X, Y, Z Coordinates	Precise spatial positioning of each point [58].	Geometric measurement and spatial analysis.
Intensity	Amount of light energy returned to the sensor [58].	Material classification and surface characterization.
Return Number	Sequence of the return for a given pulse (e.g., 1st, 2nd, 3rd) [58].	Vertical structure analysis and penetration assessment.
Classification	Label assigning point to a class (e.g., ground, vegetation, building) [58].	Feature extraction and object identification.

LiDAR systems are typically categorized as either discrete return or full waveform. Discrete return LiDAR systems record individual points for peaks in the returned energy signal, while full waveform LiDAR systems record the complete distribution of the returned energy, capturing more detailed information about the interaction between laser light and objects [58]. The standard file format for storing LiDAR point cloud data is the LAS format, with the compressed LAZ format also being widely used to reduce file sizes [58].

Field Deployment Protocols

Pre-Fieldwork Planning and Reconnaissance

Effective TLS deployment begins with comprehensive pre-fieldwork planning. This critical phase ensures that data collection meets project objectives while optimizing resource allocation.

• Project Objective Definition: Clearly define the Level of Detail (LoD) required for the specific heritage site. For historic structures with unique architectural elements, high-resolution scans are often necessary, despite requiring more time and resources [33]. Consider the final deliverables (e.g., 2D drawings, HBIM, 3D visualizations) as these requirements directly influence scanning parameters and station placement [33].
• Site Reconnaissance: Conduct a preliminary site visit to identify potential obstacles, structural conditions, and accessibility constraints. Document lighting conditions and identify areas requiring special attention, such as complex geometries, fragile elements, or occlusion-prone zones [33].
• Scan Planning: Develop a scanning network strategy that ensures complete coverage while minimizing occlusions. Plan for sufficient overlap between scan positions (typically 25-30%) to facilitate effective registration during data processing. For complex structures, this may require a higher density of scan positions [33].

Essential Research Reagent Solutions

The table below details essential equipment and software required for professional TLS heritage documentation campaigns:

Table 2: Research Reagent Solutions for TLS Heritage Documentation

Tool Category	Specific Examples	Function & Application
TLS Hardware	Long-range, narrow-FoV sensors; High-resolution panoramic scanners [33] [57].	Captures dense, accurate 3D point clouds of heritage structures and sites.
Registration Targets	Spheres, checkerboard targets [33].	Provides common reference points for aligning multiple scans into a unified coordinate system.
Positioning Systems	GPS receivers [33] [58].	Georeferences scan data for integration with other geospatial datasets.
Data Processing Software	CloudStation, LasTools [59] [58].	Filters, classifies, visualizes, and models raw point cloud data.
Supplementary Documentation	Digital cameras, field notebooks [33].	Captures visual context and supports interpretation of geometric data.

In-Situ Data Acquisition Workflow

The following workflow diagram illustrates the sequential protocol for field data acquisition:

Field Data Acquisition Protocol

• Equipment Setup and Calibration: Position the TLS instrument on a stable tripod on level ground. Power on the system and allow it to initialize according to manufacturer specifications. Verify that all sensors, including the integrated GPS and Inertial Measurement Unit (IMU), are functioning properly [58].
• Target Placement for Registration: Strategically place registration targets (e.g., spheres, checkerboards) throughout the scan area. Ensure targets are visible from multiple scan positions and are distributed to create a robust network for alignment. Avoid placing targets in positions where they might be moved during the scanning process [33].
• Scanner Parameter Configuration: Adjust scanning parameters based on the required Level of Detail (LoD) and environmental conditions. Key parameters include:
  • Scan Resolution: Determines the point spacing and density. Higher resolution captures more detail but increases scan time and data volume [33].
  • Scan Quality: Influences the number of measurements averaged per point. Higher quality settings reduce noise but extend acquisition time [33].
  • Range Settings: Optimize for the specific distances to targets within the heritage site [57].
• Scan Execution and Verification: Execute the scan according to the predefined network strategy. After each scan, perform an initial quality assessment by reviewing the point cloud preview to check for major occlusions, coverage gaps, or movement artifacts [33].
• Supplementary Data Collection: Capture high-resolution photographic imagery for colorizing point clouds. Document field conditions, observations, and any anomalies in a field notebook to assist during data processing and interpretation [33].

Data Processing and Analysis Framework

Post-Processing Workflow

Raw TLS data requires substantial processing to transform it into usable information. The following diagram outlines the sequential post-processing workflow:

Data Processing Protocol

• Data Filtering and Cleaning: Remove erroneous points (noise, outliers) resulting from sensor inaccuracies or reflective/transparent surfaces that scatter lasers unpredictably [59]. Apply density-based filtering to identify and remove points that are too sparse or dense, improving point cloud clarity and reducing data volume without sacrificing essential detail [59].
• Scan Registration and Alignment: Align individual scans from multiple positions into a unified coordinate system using the previously placed registration targets [33]. Modern software often incorporates cloud-to-cloud registration algorithms to refine alignment beyond target-based methods, improving overall accuracy [33].
• Point Cloud Classification: Assign semantic labels to points (e.g., "ground," "vegetation," "building," "architectural detail") [59]. This process involves feature extraction, segmentation, and clustering of the point cloud based on spatial and geometric similarities [59]. Automated classification algorithms, including deep learning approaches, are increasingly used for this task [4].
• Surface Modeling and 3D Model Creation: Convert point clouds into continuous surfaces through triangulation (mesh generation) to create more visually interpretable 3D models [59]. This step is crucial for generating outputs like Digital Elevation Models (DEMs) and Digital Surface Models (DSMs) or for preparing models for HBIM and VR applications [59].

Deliverables and Output Generation

Processed TLS data serves as the foundation for various analytical outputs and digital products essential for heritage research and conservation:

Table 3: TLS Data Outputs for Heritage Research

Output Type	Description	Heritage Application
3D Point Cloud	Primary, measurable dataset of XYZ coordinates [33].	As-built condition recording; deformation analysis.
Heritage BIM (HBIM)	Structured, information-rich 3D model with semantic data [33].	Conservation management; structural analysis; change monitoring.
2D Measured Drawings	Plan, section, and elevation drawings derived from point clouds [33].	Architectural documentation; conservation planning.
Digital Elevation Model (DEM)	3D representation of a terrain's surface [59].	Site topography analysis; drainage planning.
Orthographic Images	Scaled, distortion-corrected images from point cloud data [59].	Façade analysis; texture mapping.
Interactive 3D Environment	VR/AR experiences integrating scanned data [33].	Public education; virtual tourism; remote expert analysis.

Case Study Integration and Best Practice Synthesis

Synthesis of Expert Guidance

Analysis of real-world heritage documentation projects reveals several critical success factors for TLS deployment. The U.S. National Park Service's Heritage Documentation Programs (HDP) now uses TLS on nearly every project to produce 2D and 3D measured drawings, while strongly emphasizing the importance of supplementing digital data with traditional hand-measuring techniques for verification and capturing details that may be missed by scanning [33]. This hybrid approach ensures both comprehensive coverage and data integrity.

A practice-based guide to TLS for heritage documentation emphasizes the need for holistic and practical guidance that leverages the technical strengths of TLS while addressing the evolving needs of heritage conservation [33]. This includes optimizing the use of TLS to enhance the quality, accuracy, efficiency, and completeness of data collection efforts for preservation, analysis, interpretation, and education [33].

Error and Limitation Management

TLS applications are subject to specific limitations that researchers must anticipate and manage:

Table 4: Common TLS Challenges and Mitigation Strategies

Challenge	Impact on Data	Mitigation Strategy
Occlusions	Data shadows or gaps behind objects [33].	Strategic scan network planning with multiple viewpoints.
Varying Reflectivity	Data drop-out or noise on dark/absorbing or shiny/specular surfaces [57].	Adjust scan settings; use multiple scan modes; apply supplemental techniques.
Environmental Factors	Reduced data quality; inaccurate measurements [57].	Schedule fieldwork during favorable conditions; use weather-appropriate equipment.
Registration Errors	Misalignment between scans; reduced overall accuracy [33].	Use stable, well-distributed targets; verify registration with check points.
Large Data Volumes	Processing and storage bottlenecks [4].	Implement efficient data management pipelines; use multi-resolution approaches.

Advanced processing techniques are helping to overcome these challenges. For example, artificial intelligence and deep learning approaches are increasingly being applied for tasks such as crown delineation in forest environments and automated pipelines for large-scale feature extraction, which can be adapted for complex architectural elements [4]. Furthermore, the integration of TLS with complementary technologies like photogrammetry, mobile mapping systems, and geophysical prospection creates robust multi-sensor approaches that overcome the limitations of any single method [33].

Terrestrial Laser Scanning represents a fundamental advancement in the methodological toolkit for heritage science, providing unprecedented capabilities for capturing the physical reality of cultural heritage sites in measurable digital form. The fieldwork best practices and processing protocols outlined in this document provide a framework for researchers to design and implement TLS documentation campaigns that yield scientifically valid, preservation-quality data. As TLS technology continues to evolve, becoming more accessible and integrated with analytical platforms like HBIM and AI-driven processing tools, its role in documenting, analyzing, and preserving our shared cultural heritage will only expand. By adhering to these structured methodologies, researchers can ensure that their work not only meets current professional standards but also creates a durable digital record for future generations.

Table 1: Reported Accuracy of Terrestrial Laser Scanning (TLS) Data from Selected Studies

Study / Context	Reported Vertical RMSE	Reported Horizontal RMSE	Key Influencing Factors
General Quantitative Assessment [60]	0.12 m (Average Discrepancy)	~0.50 m	Biases in geo-positioning system; random short-period variation
Topographic Mapping [12]	0.10 - 0.15 m	Not Specified	Ground point density; instrument accuracy
Mountainous Terrain with Dense Forest [60]	0.15 m - 0.62 m	Not Specified	Ground point density (0.89 to 0.09 points/m²); ground filtering efficacy
Watershed Area DEMs [60]	0.75 m	Not Specified	Interpolation method (e.g., Inverse Distance Weighting, Nearest Neighbor)
Coastal Geomorphology [60]	0.25 m (after offset compensation)	0.50 m (in vegetated areas)	Vegetation cover; terrain topography; post-processing correction
Archaeological Site DTM [60]	0.50 m (Average Accuracy)	Not Specified	Data ground density (15 points/m²); geospatial registration with control points

Experimental Protocols

Protocol for High-Accuracy TLS Data Collection for Habitat Monitoring

Objective: To collect high-resolution, high-accuracy 3D point cloud data of a habitat site for the purpose of detecting subtle changes over time (e.g., erosion, vegetation growth, micro-topography).

Materials:

• Terrestrial Laser Scanner (e.g., models from Riegl, Leica)
• Survey tripod
• GPS-equipped base station and rover (for Real Time Kinematic - RTK)
• High-visibility scan targets (minimum 3)
• Field notebook and data storage media
• Calibration data for all equipment

Procedure:

• Pre-Field Planning:
  • Define the scan extent and required point density. For detecting fine-scale habitat changes, a density of >100 points/m² is often targeted [12].
  • Establish the locations for permanent ground control points (GCPs) around the study site.

• Field Setup and Scanning:

  • Set up the TLS on a stable tripod in a location that provides a comprehensive view of the area of interest.
  • Distribute scan targets within the scene such that they are visible from multiple scanner positions. Pre-measure the targets using the RTK-GPS to establish highly accurate global coordinates (X, Y, Z) for each target [3].
  • Perform the initial scan, ensuring all parameters (resolution, quality) are set according to the project's data quality plan.
  • Move the scanner to a new position, ensuring significant overlap (e.g., >30%) with the previous scan. Ensure a minimum of three common targets are visible from each new setup.
  • Repeat the scanning process until the entire site is covered from multiple perspectives to minimize occlusion [3].

• Data Management and Post-Processing:

  • Transfer all scan data and GPS measurements to a secure storage system.
  • Use specialized software (e.g., TerraScan, CloudCompare, LasTools) to register all individual scans into a single, unified point cloud using the common targets as reference points [60] [12].
  • Apply any necessary georeferencing transformations to place the final point cloud into its correct real-world coordinate system.
  • Classify the point cloud to separate ground points from vegetation and other non-ground points [60].
  • Generate derivative products such as Digital Terrain Models (DTMs) or canopy height models as required for analysis.

Protocol for TLS Data Collection for Vegetation Structure Analysis

Objective: To non-destructively measure vegetation canopy height, density, and structure for ecological research [58].

Materials:

• TLS system
• Calibration panels (if intensity values are used)
• Optional: RGB camera integrated with or used alongside the TLS

Procedure:

• Site and Scanner Configuration:
  • Select scan positions that optimize the coverage of the vegetation canopy. In dense habitats, multiple scan positions beneath the canopy may be necessary.
  • Configure the scanner to capture multiple returns per pulse to better characterize the vertical profile of the vegetation [58].

• Data Collection:

  • Conduct scans at a high point density (e.g., >50 points/m²) to resolve fine structural details like small branches and leaves.
  • If possible, collect data during leaf-off seasons for deciduous habitats to improve ground detection, or at multiple times throughout the growing season to monitor phenological changes [60].

• Data Processing and Analysis:

  • Register multiple scans if used.
  • Classify ground points accurately, as this is the foundation for calculating canopy height.
  • Model the ground surface from the classified ground points.
  • Normalize the point cloud by subtracting the ground elevation from the Z-value of each non-ground point to derive height above ground.
  • Calculate metrics of interest, such as:
    • Canopy Height Model (CHM): The height of the highest return within a given raster cell.
    • Canopy Cover: The proportion of returns classified as vegetation above a certain height threshold.
    • Vertical Distribution of Leaf Area: Analyzed by examining the density of returns at different height intervals.

Workflow Visualization

TLS Data Acquisition and Management Workflow

TLS for Habitat Change Detection Logic

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 2: Key Research Reagent Solutions for Terrestrial LiDAR Habitat Research

Item / Solution	Function / Purpose	Technical Specifications / Notes
Terrestrial Laser Scanner	Core instrument for acquiring 3D point cloud data; emits laser pulses and measures return time to calculate distance [3].	Short-, medium-, or long-range variants; accuracy of 10⁻¹–100 cm; acquisition rate of 10⁴–10⁶ points/sec [12].
Ground Control Points (GCPs)	Physical markers providing absolute positional reference; critical for co-registering multiple scans and time-series data [3].	High-visibility targets; surveyed with high-precision GPS (e.g., RTK) to establish known XYZ coordinates.
RTK-GPS System	Provides high-accuracy georeferencing for GCPs and scanner positions; integrates TLS data into a real-world coordinate system [60].	Typical accuracy: 1-3 cm; essential for change detection over time and integrating with other geospatial data.
Point Cloud Processing Software	Computational environment for managing, registering, classifying, and analyzing large, high-resolution point clouds [12].	Examples: TerraScan, LasTools, CloudCompare; used for filtering, DTM generation, and metric extraction.
Classification Algorithm	Digital reagent for segregating raw point cloud into meaningful classes (e.g., ground, vegetation, buildings) [60] [58].	Often integrated into software; accuracy is paramount for deriving correct ecological metrics (e.g., canopy height).

Benchmarking TLS Performance: Validation, Accuracy, and Comparative Analysis with Other Technologies

Terrestrial Laser Scanning (TLS) has emerged as a powerful tool for capturing high-resolution, three-dimensional data in habitat research. By emitting laser pulses and measuring their return, TLS creates dense point clouds that digitally represent the physical environment [33] [11]. However, the accuracy and reliability of metrics derived from these point clouds must be rigorously validated against traditional field measurements, a process known as ground truthing. This process is critical for ensuring that TLS data can be confidently used in ecological modeling, forest inventory, and structural analysis [61] [62]. This document provides detailed application notes and protocols for the validation of TLS-derived metrics, framed within the context of LiDAR habitat research.

TLS Validation Workflow

The following diagram illustrates the comprehensive workflow for validating TLS-derived metrics against traditional field measurements, integrating feedback loops for continuous refinement.

Quantitative Validation Data from Peer-Reviewed Studies

Table 1: Validation of TLS for Tree Structural Attribute Estimation

Metric	Traditional Method	TLS Performance	Study Context	Citation
Tree Height	Manual hypsometer	TLS reliable for trees <15-20m; challenges with tall trees in dense stands	Boreal forest, 1174 trees	[62]
Stem Volume	Destructive harvesting/Allometric equations	High correlation with manual measurements (r=0.95)	Grapevine height estimation	[63]
Branch Volume	QSM from manual wood separation	Accuracy depends on segmentation: KPConv (OA: 98%), DBSCAN (OA: 92%)	Southern pine trees	[64]
Above-Ground Biomass	Direct measurement	Enabled via QSM reconstruction	Forest carbon estimation	[61]

Table 2: TLS Accuracy in Structural and Infrastructure Applications

Application	Reference Method	TLS Accuracy	Conditions/Limitations	Citation
Retaining Wall Inspection	Total Station	RMSE: 0.065 cm	Control surfaces	[65]
Pavement Distress Detection	Visual inspection	14 distress types detected	Within 10m with proper sampling	[65]
Crack Detection	Physical measurement	0.125 cm cracks detected		[65]
Grapevine Volume	Manual measurement	Strong correlation (r>0.83, p<0.001)	UAV comparison study	[63]

Detailed Experimental Protocols

Protocol for Validating Tree Height Measurements

Objective: To validate TLS-derived tree height measurements against traditional field measurements.

Materials Required:

• Terrestrial Laser Scanner (e.g., RIEGL VZ400i)
• Traditional hypsometer or clinometer
• GNSS receiver (differential RTK) for georeferencing [64]
• Field tags for tree identification
• Tape measure

Methodology:

• Site Selection: Establish sample plots representative of the forest conditions. Ensure a minimum of 4 scan positions per plot to reduce occlusion [64].
• TLS Data Acquisition:
  • Set scanner angular resolution to 0.02° in both azimuth and zenith [64]
  • Use a 360° horizontal field of view and vertical field of view between -40° and 60°
  • Ensure overlap between scan positions for effective registration
• Field Measurement:
  • Measure tree height using a hypsometer following standard field protocols
  • Record species, DBH, and crown class for each tree [62]
• Data Processing:
  • Register point clouds using software such as RiSCAN Pro [64]
  • Remove noise using Statistical Outlier Removal filter in CloudCompare [64]
  • Manually segment individual trees from the point cloud
  • Extract tree height as the maximum Z-value in the tree point cloud
• Validation:
  • Calculate correlation coefficients (r), R², and RMSE between TLS and field measurements
  • Perform significance testing using F-test of overall significance [63]
  • Analyze bias patterns related to tree height, crown class, or species [62]

Protocol for Validating Structural Models (QSMs)

Objective: To validate Quantitative Structure Models (QSMs) derived from TLS data against traditional measurements.

Materials Required:

• TLS system with high ranging accuracy
• Field calipers for diameter measurements
• Tree climbing equipment for destructive sampling (if applicable)
• Segmentation software (e.g., TLSep, Graph, DBSCAN, KPConv) [64]

Methodology:

• Data Collection: Follow TLS acquisition protocol as in 4.1
• Wood-Leaf Segmentation:
  • Apply multiple segmentation algorithms (KPConv, DBSCAN, etc.)
  • KPConv requires training but offers high accuracy (OA: 98%) [64]
  • DBSCAN provides balance between performance and efficiency without training data
• QSM Reconstruction:
  • Use algorithms such as TreeQSM, AdQSM, aRchi, or SimpleForest [64]
  • Input manually segmented wood points for baseline comparison
  • Parameterize models based on trunk diameter estimates
• Traditional Measurement:
  • For destructive validation: harvest trees and physically measure branch dimensions
  • For non-destructive validation: use detailed manual measurements of accessible branches
• Metric Validation:
  • Compare QSM-derived volume estimates with manual measurements
  • Validate branch architecture and topology
  • Assess the impact of segmentation accuracy on QSM reliability [64]

The Scientist's Toolkit: Essential Research Reagents and Equipment

Table 3: Key Equipment for TLS Validation Studies

Category	Item	Specification/Function	Application Notes
Scanning Hardware	Terrestrial Laser Scanner	RIEGL VZ400i, Leica ScanStation P50	Select based on required range and accuracy [65]
Field Validation Tools	Differential GNSS RTK Receiver	Provides georeferencing accuracy	Essential for global accuracy assessment [65]
	Total Station	High-precision angular and distance measurement	Serves as ground truth for infrastructure studies [65]
Data Processing Software	Point Cloud Processing	RiSCAN Pro, CloudCompare	Registration, filtering, and analysis [64]
	Segmentation Algorithms	KPConv, DBSCAN, Graph, TLSep	Separate leaf and wood components; KPConv shows highest accuracy [64]
	QSM Reconstruction	TreeQSM, AdQSM, aRchi, SimpleForest	Generate quantitative structure models [64] [61]
Ancillary Equipment	Ground Control Targets	1×1 m targets for registration	Assist with data alignment and georeferencing [63]

Analysis of Validation Results and Common Pitfalls

Research indicates that TLS validation outcomes are context-dependent. In forest studies, field measurements tend to overestimate heights of tall trees, particularly those in codominant crown classes [62]. TLS-based tree height estimates have proven robust across varied stand conditions, with reliability increasing for taller trees [62]. The primary challenge in TLS measurements stems from occlusion effects, which may lead to incomplete crown representation, especially for tall trees [62].

For structural metrics beyond basic dimensions, the accuracy of derived models depends heavily on preprocessing steps. In particular, the segmentation of leaf and wood components directly impacts the quality of Quantitative Structure Models (QSMs), with misclassification potentially causing unrealistic branch structures and overestimation of volume and biomass [64]. Studies on southern pines demonstrated that selection of segmentation algorithms involves trade-offs: while KPConv achieved 98% overall accuracy, DBSCAN offered a favorable balance between performance and efficiency without requiring training data [64].

In infrastructure applications, TLS has demonstrated high precision, with sub-millimeter accuracy achievable under controlled conditions [65]. However, validation studies must account for environmental factors such as vegetation coverage, which can significantly impact measurement accuracy [65].

Terrestrial Laser Scanning (TLS) and Unmanned Aerial Vehicle LiDAR (UAV LiDAR) represent two pivotal technologies in the domain of habitat research, enabling high-fidelity three-dimensional data acquisition. These non-contact measurement systems facilitate the detailed characterization of structural habitats, which is fundamental for ecological studies, biomass estimation, and conservation planning. TLS, a ground-based system, captures the environment from a static tripod, producing exceptionally detailed point clouds of vertical surfaces and understory components [33] [66]. Conversely, UAV LiDAR, an airborne system, surveys from above, providing rapid coverage of upper canopies and extensive areas [8] [67]. Within a habitat research framework, the choice between these technologies involves critical trade-offs between accuracy, efficiency, and the capability to capture specific structural elements. This article provides a comparative analysis of their accuracy, outlines detailed operational protocols, and defines their suitability for various research applications, serving as a guide for scientists undertaking precise environmental mapping.

Quantitative Accuracy Comparison

The performance of TLS and UAV LiDAR varies significantly across different structural metrics and environmental contexts. The following tables summarize key accuracy findings from recent studies.

Table 1: Comparative Accuracy of TLS and UAV LiDAR for General Mapping

Metric	TLS Performance	UAV LiDAR Performance	Context / Notes
Absolute Accuracy	Millimeter to sub-millimeter level [68] [66]	~3 cm with RTK/GCPs; typically 1-5 cm in practice [69] [70]	Accuracy is distance-dependent for TLS.
Relative Accuracy (vs. TLS)	N/A (Reference)	~80% of points within 1.8 inches (~4.6 cm); ~60-65% within 1.2 inches (~3 cm) [8] [70]	Comparison of UAV LiDAR point cloud to TLS reference.
Point Cloud Density	Very high (e.g., 59.2M to 316M points per site) [70]	Moderate (e.g., 4.7M to 8.8M points per site) [70]	Higher density from TLS is due to proximity and static scanning.

Table 2: Performance in Forest Structural Parameter Estimation

Parameter	TLS Performance	UAV LiDAR Performance	Context / Notes
Tree Height	Reliable for all canopy layers [67]	Consistent underestimation, especially in dense, multi-layered stands (R² < 0.2) [67]	UAV pulses often fail to penetrate fully to the ground in complex forests.
Diameter at Breast Height (DBH)	High accuracy (R² up to 0.98) [67]	Not reliably measurable [67]	UAV LiDAR has limited ability to capture lower stem sections due to occlusion.
Canopy & Understory Mapping	Dominates below canopy; captures ~93% of interior crown volume [67]	Primarily delineates the outer canopy surface [67] [71]	Structural complexity is a major driver of UAV performance.

Detailed Experimental Protocols

To ensure the collection of high-quality, research-grade data, standardized protocols for both TLS and UAV LiDAR are essential. The following sections detail the methodologies for site establishment, data acquisition, and processing.

Pre-Field Planning and Site Establishment

• Objective Definition: Clearly define the primary structural metrics of interest (e.g., tree DBH, canopy height model, understory density) as this influences platform choice and scan/flight planning.
• Hybrid Approach Assessment: Determine if a hybrid TLS/UAV approach is warranted. This is recommended for structurally complex habitats like multi-layered forests, where UAV LiDAR captures the upper canopy and TLS provides critical sub-canopy and stem data [67] [72].
• Control Network Establishment:
  • Equipment: Use a high-precision GNSS receiver (e.g., Emlid Reach 2) [70].
  • Procedure: Establish a network of permanent ground control points (GCPs) across the study area. These should be clearly identifiable targets (e.g., checkerboard patterns).
  • Accuracy: Record GCP coordinates with centimeter-level accuracy using Post-Processed Kinematic (PPK) or Real-Time Kinematic (RTK) methods. These points are critical for minimizing error during cloud-to-cloud registration of multiple TLS scans and for providing absolute accuracy to the UAV LiDAR point cloud [8].

UAV LiDAR Data Acquisition Protocol

• Equipment:
  • Platform: DJI Matrice 300 or similar enterprise UAV.
  • Sensor: DJI Zenmuse L1 or equivalent integrated LiDAR system.
  • Supporting Gear: GNSS base station for PPK/RTK correction.
• Flight Planning:
  • Altitude: Conduct missions at approximately 175 feet (~53 m) AGL [8].
  • Velocity: Maintain a slow flight speed (e.g., 4 mph) to ensure high point density [8].
  • Overlap: A sidelap of 20-30% is often sufficient for LiDAR, compared to the 70/80% required for photogrammetry [69].
  • Conditions: Flights can be conducted in most weather, but avoid heavy rain or fog, which can scatter laser pulses [69].
• Execution:
  • Perform pre-flight checks, ensuring the LiDAR sensor is calibrated.
  • Execute the automated flight plan. A typical site survey takes 15-25 minutes of flight time, though the entire process may take 2-2.5 hours [70].

Terrestrial Laser Scanning (TLS) Data Acquisition Protocol

• Equipment:
  • Scanner: FARO Premium S350 or similar panoramic phase-based or time-of-flight scanner.
  • Supporting Gear: Tripod, calibration targets (if required by the scanner model).
• Scan Planning:
  • Scanner Placement: Plan multiple scan positions to minimize occlusions and ensure line-of-sight to all key habitat structures (e.g., tree stems, rocks). Positions should be spaced so that targets are captured from at least two different viewpoints.
  • Resolution: Set the scanning resolution and quality to achieve the required point density. A single scan can take 1.5 to 7 minutes [69].
• Execution:
  • Set up the scanner on a stable tripod at the first position.
  • If using targets for registration, place them in stable locations visible from multiple scan positions.
  • Perform the scan. A typical day of fieldwork may involve 64-150 individual scans [69].
  • Move the scanner to the next pre-planned position and repeat. Surveying a single site with TLS may take 2.25 to 4.5 hours [70].

Data Processing and Analysis Workflow

• UAV LiDAR Processing:
  • Trajectory Calculation: Use PPK processing with data from the UAV and the base station to refine the aircraft's trajectory.
  • Point Cloud Generation: Leverage the manufacturer's software (e.g., DJI Terra) to generate a georeferenced point cloud from the LiDAR data and refined trajectory.
• TLS Processing:
  • Registration: Use software (e.g., CloudCompare, proprietary scanner software) to align individual scans into a single, registered point cloud. This can be done via target-based or cloud-to-cloud registration methods, with the established GCPs used to minimize error and georeference the data [8] [33].
• Data Fusion and Analysis (For hybrid approaches):
  • Co-alignment: Align the TLS and UAV LiDAR point clouds into a common coordinate system using the shared GCPs or cloud-to-cloud registration on overlapping areas [8] [72].
  • Metric Extraction: Use specialized software (e.g., Computree, 3D Forest) to extract habitat metrics. For trees, this includes segmentation of individual trees, height extraction from the UAV data, and DBH/stem curve measurement from the TLS data [67] [4].

Data Acquisition Workflow

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 3: Key Equipment for TLS and UAV LiDAR Habitat Research

Item	Function	Example Specifications/Models
Terrestrial Laser Scanner	Captures high-resolution 3D point clouds from ground positions.	FARO Premium S350; typical range: 100-300m; accuracy: mm-level [70] [66].
UAV LiDAR System	Rapid, aerial acquisition of 3D geometry over large areas.	DJI Matrice 300 with Zenmuse L1 sensor; accuracy: ~3 cm with PPK [69] [70].
Geodetic GNSS Receiver	Provides centimeter-accurate positioning for Ground Control Points (GCPs).	Emlid Reach 2 or similar; supports PPK/RTK processing [70].
Scan Registration Targets	Used as reference points to align multiple TLS scans into a single model.	Checkerboard or spherical targets; not always required with modern cloud-to-cloud registration [68].
Point Cloud Processing Software	Used for registration, analysis, and metric extraction from 3D data.	CloudCompare (open-source), FARO SCENE, Pix4D, proprietary vendor software [33] [72].

The choice between TLS and UAV LiDAR is not a matter of superiority but of application-specific suitability.

• Choose Terrestrial Laser Scanning (TLS) when your research requires the highest possible accuracy for individual objects and structural elements. This includes measuring tree DBH and stem curves [67], monitoring structural deformations at millimeter scales [66], documenting complex architectural features in heritage sites [33], and conducting detailed understory and forest interior modeling [67] [4]. TLS is the preferred tool for small, complex, or inaccessible sites where ultimate detail is paramount.

• Choose UAV LiDAR when the project involves mapping extensive areas efficiently, capturing the top surfaces of tall objects, or ensuring personnel safety in hazardous terrain [8] [70] [72]. It is ideal for creating canopy height models [67], conducting large-scale topographic surveys, and mapping areas where ground access is limited. Its speed and coverage make it suitable for projects requiring rapid turnaround.

For comprehensive habitat research, particularly in structurally complex environments like multi-layered forests, a hybrid approach that leverages both technologies is highly recommended [67] [72]. This strategy combines the above-canopy perspective of UAV LiDAR with the sub-canopy structural detail of TLS, enabling the creation of a complete, multi-layered digital twin of the ecosystem [4]. By following the detailed protocols and understanding the trade-offs outlined in this article, researchers can effectively deploy these powerful technologies to advance habitat science.

Accurate individual tree structure segmentation from Terrestrial Laser Scanning (TLS) point cloud data is a foundational task in modern forestry research, enabling non-destructive estimation of biomass, carbon sequestration capacity, and detailed morphological analysis [37]. Selecting an appropriate segmentation model is critical for generating reliable ecological data. This application note provides a comparative performance analysis of two prominent approaches: XGBoost, a leading machine learning (ML) model, and PointNet++, a representative deep learning (DL) architecture. We present structured benchmark results, detailed experimental protocols, and a research toolkit to guide researchers in implementing these methods for terrestrial LiDAR habitat studies.

Performance Benchmarks & Quantitative Analysis

A direct comparative study under standardized conditions evaluated the stem segmentation performance of XGBoost and PointNet++ using identical input features and data preprocessing steps [37]. The models were tested with different input feature combinations and point densities to provide a comprehensive performance profile.

Table 1: Stem Segmentation F1-Scores (%) by Input Feature Configuration and Point Density [37]

Input Feature Configuration	Model	2048 Points	4096 Points	8192 Points
Spatial Coordinates & Normals (S)	XGBoost	84.5	85.2	85.9
	PointNet++	90.8	92.1	91.5
S + Geometric Features (S+G)	XGBoost	86.1	86.7	87.2
	PointNet++	90.1	91.4	91.0
S + Local Distribution Features (S+L)	XGBoost	86.9	87.5	88.1
	PointNet++	89.8	91.0	90.6
All Features (S+G+L)	XGBoost	87.2	87.6	87.8
	PointNet++	89.5	90.9	90.5

Table 2: Overall Model Performance and Computational Characteristics [37]

Characteristic	XGBoost	PointNet++
Highest Achieved F1-Score	87.8%	92.1%
Optimal Input Features	All Features (S+G+L)	Spatial Coordinates & Normals (S)
Optimal Point Density	8192 points	4096 points
Processing Time (for 8192 points)	47 minutes	168 minutes
Key Strength	Computational efficiency, Feature importance interpretation	Segmentation accuracy, Handling complex structures
Common Missegmentation	Stem-to-ground boundaries, Branch junctions	Complex stem-to-crown regions

Detailed Experimental Protocols

Data Acquisition and Preprocessing

Equipment: Use a survey-grade terrestrial laser scanner (e.g., Leica BLK360 used in the benchmark study) [37]. For high-precision georeferencing, a differential GNSS receiver (e.g., Trimble R12i) is recommended. Scanning Protocol: Establish circular plots with an 11.3 m radius. Employ a minimum of nine scan positions per plot—one at the center, four equidistant points on the plot perimeter, and four at the corners of a surrounding 16 m x 16 m square—to minimize occlusion [37]. Install five Ground Control Point (GCP) targets for accurate registration. Point Cloud Registration and Processing:

• Perform initial alignment in software such as Register360 Plus using cloud-to-cloud distance methods.
• Apply fine alignment using the Iterative Closest Point (ICP) algorithm in Cyclone or CloudCompare, constraining the final registration error to below 0.005 m [37].
• Georeference the registered point cloud using the measured GCP coordinates, maintaining a Root Mean Square Error (RMSE) of transformation under 3 cm.
• Manually or semi-automatically extract individual trees from the plot-level point cloud.

Feature Engineering and Data Preparation

For XGBoost and other traditional ML models, manual feature engineering is a critical step. The benchmark study utilized 17 input features, categorized as follows [37]:

• Spatial Coordinates and Normals (S): 3D coordinates (X, Y, Z) and the corresponding normal vector components (Nx, Ny, Nz).
• Geometric Structure Features (G): This includes multi-scale point curvature, linearity, planarity, and omnivariance, which describe the local 3D shape surrounding each point.
• Local Distribution Features (L): Features such as point density and the local Z-range within a defined neighborhood.

For deep learning approaches like PointNet++, the input can be as simple as the raw 3D coordinates and normals, as the network learns relevant features automatically [37] [73].

Model Training and Evaluation

Data Splitting: Divide the dataset of individual tree point clouds into training, validation, and test sets with a standard ratio of 6:2:2 [37]. Downsampling: Implement a hybrid downsampling strategy combining random sampling and Farthest Point Sampling (FPS) to standardize the number of points per tree (e.g., 2048, 4096, 8192) while preserving structural integrity [37]. Model Configuration and Training:

• XGBoost: Train the model on the engineered features. Utilize the model's built-in functions to calculate and analyze feature importance post-training.
• PointNet++: Implement a standard architecture with hierarchical feature learning. Train the model using the raw point clouds (with or without normal vectors) and a cross-entropy loss function. Performance Evaluation: Quantify segmentation accuracy using standard metrics derived from confusion matrices: Precision, Recall, and F1-score for each tree structure class (stem, crown, ground) [37].

Workflow Visualization

Tree Segmentation Model Benchmarking Workflow

The Scientist's Toolkit

Table 3: Essential Research Reagents and Solutions for TLS Tree Segmentation

Tool Category	Specific Tool / Software	Function in Research
Data Acquisition	Terrestrial Laser Scanner (e.g., Leica BLK360, RIEGL VZ-400i) [37] [64]	Captures high-resolution 3D point cloud data of the forest plot.
	Differential GNSS Receiver (e.g., Trimble R12i) [37]	Provides centimeter-accurate georeferencing for scan positions.
Data Preprocessing	CloudCompare, Cyclone, RiSCAN Pro [37] [64]	Performs point cloud registration, georeferencing, noise filtering, and manual editing.
ML/DL Frameworks	XGBoost Library [37] [74]	Provides an optimized implementation of the Gradient Boosting framework for tree segmentation.
	PyTorch or TensorFlow with PointNet++ Implementation [37] [73]	Offers the deep learning ecosystem and specific architecture for point-based semantic segmentation.
Segmentation Algorithms	TreeQSM, SimpleForest [64]	Reconstructs Quantitative Structure Models (QSMs) from segmented wood points for volume and biomass estimation.
Evaluation Metrics	Precision, Recall, F1-Score [37] [64]	Standard metrics for quantitatively evaluating segmentation accuracy against manual annotations.

This application note synthesizes performance benchmarks and methodologies for segmenting individual tree structures from TLS data. The analysis reveals a clear trade-off: PointNet++ achieves higher peak accuracy and is better suited for complex structural analysis with minimal feature engineering, while XGBoost offers superior computational efficiency and model interpretability, making it ideal for large-area inventories or resource-constrained environments. The optimal model choice depends on the specific research objectives, required accuracy, and available computational resources. Researchers are encouraged to adopt the provided protocols to ensure consistent, reproducible, and high-quality results in their terrestrial LiDAR habitat studies.

Accurate skeletal reconstruction from Terrestrial Laser Scanning (TLS) LiDAR data is foundational for advanced habitat research, enabling non-destructive analysis of complex vegetative structures. These skeleton extraction algorithms serve as critical pipelines for generating Quantitative Structure Models (QSMs), which quantify vital ecological attributes like carbon sequestration capacity and above-ground biomass (AGB) [75]. The fidelity of these models hinges on an algorithm's ability to preserve two core properties: the topological integrity, which ensures correct branch connectivity and hierarchy, and detail retention, which captures precise geometrical attributes like branch diameter, length, and inclination [76]. This document provides standardized application notes and experimental protocols for the rigorous evaluation of skeleton extraction algorithms, framed within the context of TLS LiDAR habitat research.

Quantitative Evaluation Metrics

The performance of skeleton extraction algorithms can be quantified using a suite of metrics that assess geometric accuracy, topological correctness, and computational efficiency. The following tables summarize key metrics and reported performance ranges from recent literature.

Table 1: Core Metrics for Geometric Accuracy and Detail Retention

Metric	Description	Ideal Value	Reported Performance
Mean Absolute Error (MAE) of Inclination Angles	Average absolute difference between extracted and ground-truthed branch angles [77].	0°	7° [77]
Root Mean Squed Error (RMSE) of Inclination Angles	Root average of squared differences in branch angles; penalizes larger errors [77].	0°	11.7° [77]
Percentage of Points with Low Error	Proportion of sample points where IA assessment error is below a threshold (e.g., 15°) [77].	100%	>86% [77]
Average/RSME of Skeleton Offset	Average and root mean squared error of the offset distance between extracted and reference skeletons [77].	0 m	Avg. <0.011m, RMSE <0.019m [77]
IoU Buffer Ratio (IBR)	Measures the overlap between the 3D buffer of extracted and actual line structures [78].	100%	90.8% - 94.2% [78]

Table 2: Core Metrics for Topological Integrity and Overall Performance

Metric	Description	Ideal Value	Reported Performance
F-Score	Harmonic mean of precision and recall, evaluating the completeness and correctness of the extracted skeleton structure [78].	1	0.89 - 0.92 [78]
Precision	Proportion of extracted skeleton points/nodes that correspond to true branch structures [78].	1	N/A
Recall	Proportion of actual branch structures that are successfully captured by the extracted skeleton [78].	1	N/A
Overall Accuracy (OA) of Wood-Leaf Separation	Accuracy of separating wood and leaf points, a critical pre-processing step [79].	100%	86% - 98% [79] [64]
Computational Efficiency	Time and memory consumption during processing [76].	Application-dependent	Varies by algorithm and point cloud size [76]

Experimental Protocols for Algorithm Evaluation

Protocol 1: Pre-processing and Wood-Leaf Separation

Objective: To ensure a standardized quality of input data and evaluate the critical first step of separating wood from leaf points, which directly impacts skeleton quality [79] [64].

Materials: TLS point cloud data from single trees or forest plots, computing workstation with software (e.g., CloudCompare, Python PCL library).

Procedure:

• Data Acquisition & Pre-processing: Acquire TLS data using a standardized protocol with multiple scan positions to minimize occlusion [64]. Register individual scans into a single coordinate system. Apply a Statistical Outlier Removal (SOR) filter (e.g., in CloudCompare) with parameters like 10 nearest neighbors and a standard deviation multiplier of 3 to reduce noise [64].
• Manual Reference Creation: Manually segment a subset of the pre-processed point clouds into "leaf" and "wood" components to create a high-fidelity ground truth dataset [64].
• Algorithm Application: Apply the wood-leaf separation algorithms under evaluation (e.g., KPConv, DBSCAN, Graph-based methods) to the pre-processed data [79] [64].
• Accuracy Assessment: Compare the algorithm's output against the manual reference. Calculate Overall Accuracy (OA) and F-Score to quantify performance [64].

Protocol 2: Assessing Topological Integrity

Objective: To quantitatively and qualitatively verify that the extracted skeleton correctly represents the natural branching connectivity and hierarchy without logical errors [76].

Materials: Wood-classified point cloud, skeleton extraction software (e.g., AdTree, TreeQSM, SimpleForest), ground truth data (e.g., physically mapped tree or digital twin).

Procedure:

• Skeleton Extraction: Run the skeleton extraction algorithms on the wood point cloud.
• Topological Error Identification: Visually inspect the skeleton in 3D modeling software and programmatically check for two primary topological errors:
  • Error A (Child from Multiple Parents): Identify any branch nodes that are incorrectly connected to two or more parent branches, violating the tree-like hierarchical structure [76].
  • Error B (Cross-Connection): Identify any instance where a section of one branch is incorrectly connected to a different, unrelated branch [76].
• Graph-Based Analysis: Represent the skeleton as a graph. Use depth-first search (DFS) or breadth-first search (BFS) to traverse the graph from the root and flag nodes with multiple incoming edges (indicating Error A).
• Metric Calculation: For a quantitative measure, calculate the F-Score of the skeleton structure against a known-good reference model [78].

Protocol 3: Evaluating Geometric Detail Retention

Objective: To measure the accuracy of the extracted skeleton in capturing the physical dimensions and spatial orientation of branches.

Materials: Wood-classified point cloud, extracted skeleton, reference measurements (e.g., from manual calipers, total station, or high-resolution photogrammetric model).

Procedure:

• Reference Data Collection: For a set of sample branches, measure key geometric attributes: diameter at breast height (DBH), branch inclination angles using a clinometer, and branch lengths using a tape measure [77].
• Skeleton Attribute Derivation: From the extracted skeleton and its associated fitted cylinders (in QSMs), derive the corresponding branch diameters, lengths, and angles [77].
• Error Calculation: For each sample branch, calculate the difference between the derived and measured values.
• Statistical Analysis: Compute aggregate statistics across all samples: Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) for inclination angles and branch diameters [77]. Calculate the IoU Buffer Ratio (IBR) to evaluate the overall spatial overlap of the skeleton with the input point cloud [78].

Workflow Visualization

Skeleton Evaluation Workflow

Topological Assessment Logic

The Scientist's Toolkit

Table 3: Essential Research Reagents and Computational Tools

Tool/Solution	Function	Application in Protocol
Terrestrial Laser Scanner (TLS)	Captures high-density 3D point clouds of the target habitat or tree structure.	Data acquisition for all protocols. Key for creating the input point cloud [23].
RIEGL VZ-400i	A specific TLS model known for high accuracy and used in foundational studies [76] [64].	Provides millimeter-level point clouds for creating high-fidelity ground truth data [64].
CloudCompare (Open Source)	Software for 3D point cloud and mesh processing, including registration, filtering, and basic analysis.	Pre-processing, Statistical Outlier Removal (SOR), and manual segmentation [64].
TreeQSM (MATLAB)	A widely used, patch-based algorithm for constructing Quantitative Structure Models (QSMs) [64].	Skeleton extraction and cylinder-fitting for geometric attribute derivation [75].
AdQSM / SimpleForest	Skeleton-based QSM algorithms that first extract a tree skeleton before model fitting [64] [75].	Alternative methods for skeleton extraction; useful for comparative performance analysis [75].
KPConv (Deep Learning)	A deep learning-based method for point cloud semantic segmentation (wood-leaf separation) [64].	Achieving high-accuracy (>95%) wood-leaf separation for large-scale applications [64].
DBSCAN (Algorithm)	A density-based clustering algorithm for spatial data, does not require training data [79] [64].	Wood-leaf separation offering a favorable trade-off between performance and computational efficiency [64].
C++/Python with PCL	Programming languages and libraries (Point Cloud Library) for custom algorithm development [76].	Implementing custom evaluation scripts, graph analysis, and metric calculations [76].

Integrating multi-platform remote sensing data has emerged as a pivotal framework for advancing habitat modeling, particularly within terrestrial laser scanning (TLS) LiDAR habitat research. This approach enables a comprehensive digital representation of ecosystems by combining the complementary strengths of various sensing modalities [80]. TLS provides exceptionally detailed, millimeter-to-centimeter accuracy structural data of the understory and lower canopy from a ground-based perspective [11] [81]. However, this ground-level view suffers from occlusion effects, particularly in dense vegetation, which limits its ability to fully characterize the upper canopy and overall tree architecture [81].

Airborne Laser Scanning (ALS) and photogrammetry effectively compensate for these limitations with their above-canopy perspective, providing continuous coverage of canopy topography and broader landscape context [81] [71]. The synergy created by fusing these platforms enables researchers to create detailed, three-dimensional habitat representations that would be impossible with any single platform, supporting applications from carbon stock assessment to biodiversity monitoring and conservation planning [11] [71]. This protocol outlines the methodologies, workflows, and analytical frameworks for effectively integrating these complementary technologies to advance habitat research.

Multi-Platform Data Characteristics and Complementarity

Platform-Specific Capabilities and Limitations

Table 1: Comparative analysis of remote sensing platforms for habitat modeling

Platform	Spatial Perspective	Key Strengths	Inherent Limitations	Ideal Habitat Applications
Terrestrial Laser Scanning (TLS)	Ground-up	Sub-cm measurement accuracy [81]; Detailed stem & understory structure [11]; High point density (>12x MLS, >300x ALS) [81]	Severe upper canopy occlusion [81]; Limited spatial coverage; Labor-intensive deployment	Tree architecture modeling [11]; Species classification [71]; Biomass estimation; Forest inventory metrics
Airborne Laser Scanning (ALS)	Top-down	Broad area coverage; Continuous canopy height models; Efficient landscape-scale sampling	Limited sub-canopy penetration [81]; Lower point density; Coarser structural detail	Landscape-scale habitat mapping; Canopy topography; Biomass extrapolation; Regional carbon stocks
UAS Photogrammetry/LiDAR	Above-canopy (flexible)	Centimeter-resolution canopy data [71]; Flexible deployment; Rapid acquisition	Limited penetration in closed canopies [71]; Regulatory restrictions in urban areas [81]	High-resolution canopy structure; Species-specific crown mapping [71]; Small-area monitoring
Mobile Laser Scanning (MLS)	Ground-level mobile	Rapid data collection along transects; Reduced occlusion compared to TLS [81]	Lower accuracy than TLS; Motion distortion; Limited by terrain accessibility	Linear habitat corridors; Urban tree inventories [81]; Infrastructure-adjacent monitoring

Quantitative Performance Metrics

Table 2: Accuracy assessment of structural metrics across platforms (adapted from urban tree inventory study) [81]

Structural Metric	TLS Performance	MLS Performance	ALS Performance	Optimal Acquisition Conditions
DBH	High accuracy (RMSE: 0.033-0.036m) [81]	High accuracy (comparable to TLS) [81]	Not applicable (occluded stems)	Leaf-off conditions for both TLS and MLS [81]
Tree Height	Underestimation due to canopy occlusion [81]	Moderate accuracy (occlusion in upper canopy) [81]	High accuracy (minimal occlusion) [81]	ALS or UAS platforms; Leaf-off MLS
Crown Volume	Limited to lower crown	Moderate in leaf-off (CCC: 0.85 in leaf-on) [81]	Good coverage from above	Multi-platform fusion required for complete assessment
Point Density	~12× MLS, ~300× ALS [81]	Intermediate density	Lowest density but broadest coverage	TLS for detail, ALS for context

Pre-Processing and Data Integration Framework

Geometric and Radiometric Pre-Processing

Effective multi-platform data fusion begins with comprehensive pre-processing to ensure spatial alignment and radiometric consistency across datasets [80]. Geometric corrections address spatial distortions arising from sensor characteristics, platform motion, and terrain effects [80]. Observer-related distortions (systematic sensor errors) are generally predictable and correctable through calibration, while observed-related distortions (atmospheric effects, terrain variability) require more sophisticated, often dynamic correction models [80].

Critical pre-processing steps include:

• Sensor co-registration: Establishing precise relative orientation between platforms using invariant features (e.g., building corners, permanent structures)
• Geometric correction: Applying platform-specific distortion models, including radial and tangential lens distortions for photogrammetry, and scan angle corrections for LiDAR
• Point cloud registration: Using Iterative Closest Point (ICP) or feature-based algorithms to align separate scans into a unified coordinate system
• Radiometric normalization: Correcting for atmospheric effects, illumination differences, and sensor response variations across acquisition dates

Data Fusion Methodologies

Multi-platform data integration occurs at three primary levels, each with distinct applications in habitat modeling:

• Data-Level Fusion: Direct combination of point clouds from multiple sensors after precise co-registration, creating a comprehensive 3D representation that leverages the complementary perspectives of each platform [80].

• Feature-Level Fusion: Extraction of modality-specific features followed by integration in a shared feature space. This approach particularly benefits species classification, where TLS captures detailed trunk and understory characteristics while ALS/UAS data provide crown architecture metrics [71].

• Decision-Level Fusion: Independent analysis of each data stream with subsequent integration of results through voting schemes, probability averaging, or other consensus mechanisms, preserving the unique strengths of each platform while providing robust final classifications [80].

Experimental Protocols for Multi-Platform Habitat Assessment

Field Deployment and Data Acquisition

Protocol 1: Integrated TLS and ALS Data Collection for Forest Structural Assessment

Objective: To characterize vertical forest structure and composition through synchronized multi-platform data acquisition.

Materials:

• Terrestrial Laser Scanner (phase-shift for detail, pulse-based for range)
• Airborne LiDAR system or UAS-based LiDAR
• Differential GPS for georeferencing
• Field targets for co-registration (minimum 5 per hectare)
• Spectral calibration panels (for optical sensors)

Methodology:

• Pre-field Planning:
  • Establish systematic scan positions using a tessellation strategy (minimum 3-5 scans per hectare for TLS)
  • Identify and mark permanent ground control points visible to both terrestrial and airborne platforms
  • Schedule ALS/UAS acquisitions within 2 weeks of TLS deployment to minimize temporal changes

• TLS Deployment:

  • Position scanner to maximize stem visibility while minimizing occlusion
  • For single-scan TLS (efficient for inventory), place scanner at plot center [71]
  • For multi-scan TLS (high completeness), deploy in a grid pattern with 30-40% overlap
  • Scan in both leaf-on and leaf-off conditions where phenology is ecologically relevant [81]

• ALS/UAS Coordination:

  • Fly ALS with sufficient overlap (minimum 30% side lap) to minimize data gaps
  • For UAS deployment, maintain consistent altitude for uniform point density
  • Acquire synchronized optical imagery for photogrammetric reconstruction if using camera sensors

• Field Validation:

  • Collect traditional inventory metrics (DBH, species, height) for algorithm validation
  • Tag and map individual trees for subsequent individual-tree analysis
  • Document phenological stage, weather conditions, and any disturbance events

Data Processing Workflow

Protocol 2: Multi-Platform Point Cloud Integration and Feature Extraction

Objective: To create a unified structural model of habitat through automated point cloud processing.

Processing Environment:

• High-performance computing system with GPU acceleration
• Point cloud processing software (CloudCompare, LASTools, FUSION)
• Custom scripts for feature extraction (Python, R)

Methodology:

• Pre-processing Sequence:
  • Import raw point clouds from all platforms
  • Apply sensor-specific calibration corrections
  • Remove noise and outliers through statistical filtering
  • Georeference all data to a common coordinate system

• Data Integration:

  • Co-register platforms using invariant features and ground control points
  • Harmonize point densities through strategic subsampling if needed
  • Classify points into ground, vegetation, and structure classes

• Structural Feature Extraction:

  • Segment individual trees using a multi-scale approach [71]
  • Extract platform-specific structural metrics:
    • From TLS: Stem diameter, trunk taper, understory vegetation density [11]
    • From ALS: Canopy height, crown volume, canopy roughness [81]
    • From UAS: Fine-scale crown architecture, leaf area density [71]
  • Compute integrative metrics such as vertical complexity index and canopy cover

Figure 1: Multi-platform data fusion workflow for comprehensive habitat modeling

Advanced Analytical Applications

Species Classification Through Modality-Specific Feature Design

Protocol 3: Multi-Modal Species Identification in Complex Hardwood Forests

Objective: To accurately classify tree species by leveraging complementary structural information from TLS and UAS/ALS platforms.

Rationale: TLS and aerial platforms capture different aspects of tree architecture that vary by species. For instance, oaks and sugar maples exhibit distinct profile shapes detectable through TLS measurements of canopy width, while UAS LiDAR better captures canopy density features, especially when understory is occluded [71].

Methodology:

• Modality-Specific Feature Design:
  • TLS-derived features: Stem curvature, branch insertion angles, bark texture, understory leaf density
  • UAS/ALS-derived features: Crown shape, top-surface rugosity, canopy volume profile, gap fraction

• Feature Fusion and Classification:

  • Employ machine learning classifiers (Random Forest, XGBoost, Neural Networks)
  • Train separate models for each platform followed by decision-level fusion
  • Alternatively, combine all features into a unified feature space for integrated classification

• Validation:

  • Compare automated classification against manual species identification
  • Assess relative contribution of each platform to classification accuracy
  • Evaluate performance across structural types and size classes

Figure 2: Species classification workflow using modality-specific features from TLS and aerial platforms

Digital Twin Development for Habitat Modeling

Protocol 4: Creating Virtual Forest Models Through Quantitative Structure Modeling

Objective: To develop structurally accurate 3D forest representations ("digital twins") that support ecological simulation and forecasting.

Conceptual Framework: Digital twins represent a shift from simplified abstractions to highly detailed digital replicas of physical systems, enabling more realistic simulation of ecological processes [11]. In forest ecology, this involves creating precise 3D representations of individual trees and their spatial arrangements.

Methodology:

• Individual Tree Reconstruction:
  • Develop Quantitative Structure Models (QSMs) that algorithmically enclose point clouds in topologically-connected volumes [11]
  • Combine TLS-derived trunk and branch architecture with ALS-derived crown extent
  • Apply allometric relationships to fill occluded regions where necessary

• Process Integration:

  • Incorporate radiative transfer models to simulate light regimes within the canopy [11]
  • Integrate functional-structural plant models (FSPMs) to simulate growth dynamics [11]
  • Couple with hydrological and microclimate models for comprehensive habitat assessment

• Validation and Refinement:

  • Compare virtual model outputs with empirical measurements (e.g., sap flow, litterfall)
  • Use Bayesian calibration to refine parameter estimates
  • Implement iterative improvement cycles as additional field data becomes available

The Scientist's Toolkit: Essential Research Solutions

Table 3: Critical hardware, software, and analytical tools for multi-platform habitat research

Tool Category	Specific Solutions	Function in Research	Implementation Considerations
Acquisition Hardware	Phase-shift TLS (e.g., Z+F, Faro); Pulse-based TLS (e.g., RIEGL)	High-accuracy 3D data capture; Phase-shift for detail, pulse-based for range [23]	Balance portability vs. accuracy; Consider scan speed & field deployment requirements
Platform Positioning	Differential GPS; Inertial Measurement Units (IMU)	Precise georeferencing of multi-platform data; Sensor orientation tracking	Achieve centimeter-level accuracy for effective data fusion
Software Platforms	CloudCompare; LASTools; FUSION; PyVista	Point cloud visualization, processing, and metric extraction	Open-source options reduce barriers; Custom scripting often required
Analytical Frameworks	Machine Learning (Random Forest, CNN); Quantitative Structure Models (QSMs)	Species classification [71]; 3D tree reconstruction [11]	Transfer learning adapts models across sites; QSMs require high-quality point clouds
Fusion Algorithms	Iterative Closest Point (ICP); Feature-based registration; Voxel-based methods	Multi-platform data alignment; Integrated metric calculation	Account for spatial resolution differences; Preserve unique information from each platform

Implementation Challenges and Future Directions

While multi-platform integration offers transformative potential for habitat modeling, several challenges require careful consideration:

Data Volume and Computational Demands: The fusion of high-density TLS with landscape-scale ALS generates massive datasets requiring significant computational resources and efficient processing pipelines [11] [80]. Cloud-based computing and advanced data structures (e.g., octrees) are increasingly essential for managing these data volumes.

Occlusion and Data Gaps: Despite multi-platform integration, occlusion remains a fundamental challenge, particularly in dense vegetation [81]. Strategic acquisition protocols (e.g., leaf-on/leaf-off campaigns) and advanced gap-filling algorithms using QSMs can mitigate these limitations [11].

Automation and Scalability: Current processing workflows often require significant manual intervention, limiting scalability. The integration of artificial intelligence and deep learning approaches shows promise for automating feature extraction, species classification, and data fusion processes [82] [71].

Future Outlook: Emerging technologies including UAV-based TLS, miniaturized sensors, and real-time processing capabilities will further enhance multi-platform integration. The convergence of these technologies with advanced deep learning frameworks promises to unlock new dimensions in habitat characterization and ecosystem monitoring [82] [80].

Conclusion

Terrestrial Laser Scanning has unequivocally transformed habitat analysis by providing an unprecedented, quantitative view of ecosystem structure in three dimensions. It bridges a critical gap between traditional field surveys and broader-scale remote sensing, enabling the creation of 'digital twins' that enhance our understanding of ecological processes, carbon sequestration, and biodiversity. Future directions point toward the seamless integration of multi-platform LiDAR data, the increasing empowerment of artificial intelligence for automated analysis, and the development of more accessible hardware. For researchers, this progression promises not only richer datasets but also fundamentally new ways to monitor, model, and respond to environmental change in a rapidly evolving world, with profound implications for climate policy and conservation strategy.

References

• 1. Terrestrial Laser Scanning [https://www.gim-international.com/content/article/terrestrial-laser-scanning-2]
• 2. Terrestrial Laser Scanners (TLS) - Southampton Generic Blogs [https://generic.wordpress.soton.ac.uk/tda/terrestrial-laser-scanning/terrestrial-laser-scanners-tls/]
• 3. Terrestrial Laser Scanning [https://www.earthscope.org/what-is/tls/]
• 4. Expanding forest research with terrestrial LiDAR technology [https://pmc.ncbi.nlm.nih.gov/articles/PMC12501281/]
• 5. Laser Scanning: Lidar Vs Laser - Differences [https://pinnacleinfotech.com/lidar-vs-laser/]
• 6. Airborne versus Terrestrial: which one is the right LiDAR for your needs? - ACI Corporation English [https://acicorporation.com/blog/2023/02/22/airborne-versus-terrestrial-which-one-is-the-right-lidar-for-your-needs/]
• 7. Lidar Applications in the Modern World [https://www.fugro.com/news/long-reads/2025/lidar-applications-in-the-modern-world-fugro-longread]
• 8. 2025-01-8694: A Comparison of UAV LiDAR and Terrestrial Laser Scanner Accuracies - Technical Paper [https://saemobilus.sae.org/papers/a-comparison-uav-lidar-terrestrial-laser-scanner-accuracies-2025-01-8694]
• 9. Accuracy comparison of terrestrial and airborne laser scanning and manual measurements for stem curve-based growth measurements of individual trees - ScienceDirect [https://www.sciencedirect.com/science/article/pii/S2666017224000099]
• 10. LiDAR & Environment: advancing environmental research ... [https://www.yellowscan.com/industries/environmental-research/]
• 11. Expanding forest research with terrestrial LiDAR technology [https://www.nature.com/articles/s41467-025-63946-6]
• 12. Terrestrial Laser Scanning - an overview [https://www.sciencedirect.com/topics/earth-and-planetary-sciences/terrestrial-laser-scanning]
• 13. Types of Lasers [https://www.newport.com/n/laser-types]
• 14. Multi-Wavelength Laser [https://integratedoptics.com/wavelength-combiners/ara-laser-combiner%3B-free-space-405nm-488nm-561nm-e-638nm]
• 15. Use of Lidar for coastal habitat mapping [https://www.coastalwiki.org/wiki/Use_of_Lidar_for_coastal_habitat_mapping]
• 16. LiDAR Laser Scanners: Industry Trends in 2025 [https://www.foxtechrobotics.com/a-news-lidar-laser-scanners-industry-trends-in-2025.html]
• 17. Researchers share new 3D scanning protocol for monitoring ... [https://research.fs.usda.gov/srs/articles/laser-focused-future-researchers-share-new-3d-scanning-protocol-monitoring-fire]
• 18. Spectral imaging and spectral LIDAR systems [https://www.degruyterbrill.com/document/doi/10.1515/nanoph-2020-0625/html?lang=en&srsltid=AfmBOooRGHZfUPzDrpScWXkB_5vHoP2QL-yGisreFhZaBdw0QAQfy5Pr]
• 19. Terrestrial laser scanning in forest ecology: Expanding the ... [https://www.sciencedirect.com/science/article/pii/S0034425720304752]
• 20. LiDAR Point Cloud Object Recognition Method via Intensity ... [https://www.mdpi.com/2079-9292/12/9/2087]
• 21. 3DMASC: Accessible, explainable 3D point clouds ... [https://www.sciencedirect.com/science/article/abs/pii/S0924271623003337]
• 22. Innovation in Every Pulse: RIEGL's New LiDAR Technologies [https://www.riegl.com/en-southamerica/news-insights/article/innovation-in-every-pulse-riegls-new-lidar-technologies]
• 23. 2025 to 2035: Terrestrial Laser Scanning Market Outlook [https://www.marketsandmarkets.com/Market-Reports/terrestrial-laser-scanning-market-3652955.html]
• 24. A Digital Twin Architecture for Forest Restoration [https://www.mdpi.com/1999-5903/17/9/421]
• 25. Terrestrial 3D Laser Scanning for Ecosystem and Fire Effects ... [https://research.fs.usda.gov/treesearch/67919]
• 26. Realistic virtual forests for understanding ... [https://www.sciencedirect.com/science/article/pii/S0924271625002576]
• 27. Smartphone LiDAR for urban forest carbon assessment [https://pmc.ncbi.nlm.nih.gov/articles/PMC12273488/]
• 28. Terrestrial Laser Scanning Market Outlook 2025-2034 [https://www.researchandmarkets.com/reports/6186467/terrestrial-laser-scanning-market-outlook?srsltid=AfmBOopQZOa0qeM3fDOAiyvDO6o7-7rWkuNLH78IOc8LxB-HcRRFBLIP]
• 29. A LiDAR-based framework for visualizing and quantifying ... [https://www.sciencedirect.com/science/article/pii/S1569843225004923]
• 30. Analysis and Prediction of Gap Dynamics in a Secondary ... [https://www.mdpi.com/2072-4292/13/1/100]
• 31. Mobile Laser Scanning in Forest Inventories [https://pmc.ncbi.nlm.nih.gov/articles/PMC12473678/]
• 32. Deep Imbalanced Multi-Target Regression: 3D Point Cloud ... [https://arxiv.org/html/2511.12740v1]
• 33. Developing a Practice-Based Guide to Terrestrial Laser ... [https://www.mdpi.com/2571-9408/8/8/313]
• 34. Terrestrial Laser Scanning and UAV Photogrammetry in ... [https://www.sciencedirect.com/science/article/abs/pii/S2212054825000256]
• 35. Quantitative Comparison of UAS-Borne LiDAR Systems for ... [https://www.mdpi.com/1424-8220/20/16/4453]
• 36. TreeLearn: A deep learning method for segmenting ... [https://www.sciencedirect.com/science/article/pii/S1574954124004308]
• 37. Comparative Analysis of Machine Learning and Deep ... [https://www.mdpi.com/2072-4292/17/13/2245]
• 38. Efficient Tree Species Classification Using Machine and ... [https://www.frontiersin.org/journals/forests-and-global-change/articles/10.3389/ffgc.2025.1431603/full]
• 39. Terrestrial laser scanning point clouds and tree attributes ... [https://www.silvafennica.fi/article/24066]
• 40. Tree Analysis Using Lidar [https://www.yungo.com/newsroom/tree-analysis-using-lidar]
• 41. Urban Forest Monitoring with Terrestrial Laser Scanning [https://www.aces.edu/blog/topics/forestry/urban-forest-monitoring-with-terrestrial-laser-scanning/]
• 42. Mitigating occlusion effects in Leaf Area Density estimates ... [https://www.sciencedirect.com/science/article/pii/S0034425720302066]
• 43. Point Cloud Registration: Methods and Applications in 3D ... [https://www.jt3dscan.com/post/point-clouds-and-3d-registration-methods-for-accurate-real-world-capture]
• 44. Registration of large-scale terrestrial laser scanner point clouds: A review and benchmark - ScienceDirect [https://www.sciencedirect.com/science/article/abs/pii/S0924271620300836]
• 45. Registration of Laser Scanning Point Clouds: A Review [https://www.mdpi.com/1424-8220/18/5/1641]
• 46. DFPS: An Efficient Downsampling Algorithm Designed for ... [https://www.mdpi.com/1424-8220/25/14/4279]
• 47. Novel method for robust bilateral filtering point cloud ... [https://www.sciencedirect.com/science/article/pii/S1110016825005952]
• 48. A Survey of Deep Learning-based Point Cloud Denoising [https://arxiv.org/html/2508.17011v1]
• 49. Research on a Point Cloud Registration Method of Mobile Laser ... [https://link.springer.com/content/pdf/10.1007/s12205-022-0366-0.pdf]
• 50. 7 Best Data Labeling Platforms for 3D & LiDAR in 2025 ... [https://medium.com/@encord/7-best-data-labeling-platforms-for-3d-lidar-in-2025-ranked-compared-8c078172192b]
• 51. A Guide to 3D LiDAR Point Cloud Segmentation for AI ... [https://www.basic.ai/blog-post/3d-point-cloud-segmentation-guide]
• 52. Using aerial LiDAR to assess regional availability of ... [https://www.sciencedirect.com/science/article/abs/pii/S0378112723002360]
• 53. Segmentation Machine Learning: Best Methods in 2025 Explained | Label Your Data [https://labelyourdata.com/articles/segmentation-machine-learning]
• 54. An overview of industrial image segmentation using deep learning models [https://www.oaepublish.com/articles/ir.2025.09]
• 55. Hardware Requirements for Machine Learning [https://www.geeksforgeeks.org/machine-learning/hardware-requirements-for-machine-learning/]
• 56. Fields of The World: A Machine Learning Benchmark ... [https://ojs.aaai.org/index.php/AAAI/article/view/35034]
• 57. Experimental Study Regarding Long Range LiDAR ... [https://www.mdpi.com/1424-8220/22/15/5731]
• 58. The Basics of LiDAR - Light Detection and Ranging [https://www.neonscience.org/resources/learning-hub/tutorials/lidar-basics]
• 59. LiDAR Data Processing: Deliverables,and Key Steps [https://www.yellowscan.com/knowledge/lidar-data-processing-importance-deliverables-and-key-steps/]
• 60. A Quantitative Assessment of LIDAR Data Accuracy [https://www.mdpi.com/2072-4292/15/2/442]
• 61. Terrestrial laser scanning: a new standard of forest measuring ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8557362/]
• 62. Is field-measured tree height as reliable as believed [https://www.sciencedirect.com/science/article/pii/S0924271618303046]
• 63. Comparative Analysis of TLS and UAV Sensors for Estimation ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11360376/]
• 64. Combined impact of semantic segmentation and ... [https://www.nature.com/articles/s41598-025-09681-w]
• 65. Comparative Analysis of Drone-Lidar and Traditional TLS ... [https://www.mdpi.com/2076-3417/15/22/12008]
• 66. A review of terrestrial laser scanning (TLS) [https://www.sciencedirect.com/science/article/abs/pii/S0263224123012484]
• 67. A Comparative Analysis of UAV LiDAR and Mobile Laser ... [https://www.mdpi.com/1999-4907/16/9/1481]
• 68. Performance Evaluation of Terrestrial Laser Scanners [https://pmc.ncbi.nlm.nih.gov/articles/PMC8314184/]
• 69. Drone survey vs. laser scan comparison 2025 - Airteam [https://www.airteam.ai/en/blog/drohnenvermessung-vs-laserscan-dachaufmass]
• 70. 2025-10-24 - Are LiDAR Drones Worth It? [https://lightpointdata.com/20251024-are-lidar-drones-worth-it]
• 71. Modality-specific feature design for species classification in ... [https://www.frontiersin.org/journals/forests-and-global-change/articles/10.3389/ffgc.2025.1500178/full]
• 72. Terrestrial Laser Scanning vs. Drone Surveys [https://www.linkedin.com/pulse/terrestrial-laser-scanning-vs-drone-surveys-xtoje]
• 73. Pointwise deep learning for leaf-wood segmentation of ... [https://www.sciencedirect.com/science/article/abs/pii/S0924271625002497]
• 74. Leakage Detection in Subway Tunnels Using 3D Point Cloud ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12299163/]
• 75. A Unified Graph-based Framework for Scalable 3D Tree ... [https://arxiv.org/html/2506.15577v1]
• 76. A Novel Algorithm Based on Geometric Characteristics for ... [https://www.mdpi.com/1999-4907/13/10/1534]
• 77. Assessing inclination angles of tree branches from ... [https://www.sciencedirect.com/science/article/pii/S0303243421002968]
• 78. Geometry and Topology Preservable Line Structure ... [https://www.mdpi.com/2072-4292/17/17/3033]
• 79. Extraction of tree branch skeletons from terrestrial LiDAR ... [https://www.sciencedirect.com/science/article/pii/S1574954124005028]
• 80. A Survey of Multimodal Data Fusion in Earth Observation ... [https://www.preprints.org/manuscript/202510.0743/v1]
• 81. Evaluating airborne, mobile and terrestrial laser scanning ... [https://www.sciencedirect.com/science/article/abs/pii/S1618866724002267]
• 82. Multi-platform and Multi-modal Remote Sensing Data ... [https://www.mdpi.com/journal/remotesensing/special_issues/T8P00J14Q5]