Seed Initialization

The density and distribution of seeds are controlled by two parameters:

• Step Size: Defines the horizontal spacing between seeds distributed across the forest floor.
• Seed Height: Defines the vertical offset of the seeds above ground level.

A smaller Step Size results in more seeds and potentially finer segmentation, but also increases the risk of over-segmentation (splitting individual trees into multiple segments).

In general, denser understory vegetation requires smaller seed spacing. Typical values range from 0.5 m in dense understory to 2.0 m in forests with little or no understory.

In a standard workflow, the user provides Known Tree Positions from the Tree Trunk and Diameter Detection tool. These positions are treated as primary seeds, along with available attributes such as stem diameter.

Maximum Gap Size

The Maximum Gap Size is one of the key parameters controlling individual tree segmentation. It defines the maximum distance between points that the algorithm considers connected.

This parameter helps bridge gaps in the point cloud caused by incomplete scanning or occlusions. In forest environments, capturing a complete representation of all branches is practically impossible, particularly in the upper canopy.

By allowing connections across small gaps, the algorithm can reconstruct more complete tree structures.

Cost Weighting

The Cost Weighting parameter controls the relative cost of vertical and horizontal connections during segmentation.

It is a dimensionless parameter:

• 1: Neutral weighting without a directional preference.
• Values greater than 1: Reduce the relative cost of vertical growth compared to lateral growth.

Increasing this value introduces a preference for upward connections, encouraging vertical rather than lateral expansion.

Higher values (e.g., 3–4) are generally suitable for forests with tall trees and relatively narrow crowns.

For trees with broad crowns and extensive lateral branching, a value of 1 is usually recommended.

The parameter should be adjusted according to the expected tree architecture to achieve stable and realistic segmentation results.

Pre-Segmentation Filtering – Intensity and Deviation

Similar to stem detection, individual tree segmentation can benefit from distinguishing solid surfaces from noisy or porous structures.

The algorithm uses this distinction to identify the main tree structure while treating foliage and other less reliable points separately.

This separation is controlled by two parameters:

• Minimum Intensity (or RIEGL Reflectance)
• Maximum Deviation (if available)

Points meeting the filtering criteria are treated as potential solid surface points (red), while the remaining points are considered less reliable for structural connections (gray).

Foliage Labelling Distance

Although the segmentation algorithm primarily uses solid surface points to identify tree structures, foliage and other excluded points can still be assigned to the final tree segments.

The Foliage Labelling Distance defines the maximum distance within which these points can be assigned to an already segmented tree structure.

After identifying the core tree structure, nearby foliage or noisy points are assigned to the corresponding tree segment if they fall within this distance.

The parameter influences the completeness of the final segmentation:

• Smaller values:
  • Include only points close to the identified tree structure.
  • Reduce the risk of including outliers or neighboring vegetation.
  • May leave parts of the upper canopy unsegmented (e.g., labels 0 or -1).
• Larger values:
  • Include more foliage and porous structures.
  • Provide a more complete representation of the canopy.
  • Increase the risk of assigning outliers or neighboring vegetation to the wrong tree.

Minimum Tree Height

The Minimum Tree Height parameter allows us to exclude trees below a specified height from the final segmentation results.

For example, we can:

• Exclude small understory trees below 3 m.
• Focus on taller trees within the stand.
• Apply a minimum height threshold according to the objectives of the analysis.

This helps restrict the final results to trees relevant to the intended application.

Minimum Point Density

The Minimum Point Density defines the minimum local point density required for structural connections during segmentation.

Regions with insufficient point density are excluded from the connection process, reducing the risk of linking structures based on sparse or unreliable observations.

Crown Metrics Resolution

After segmentation, each tree is assigned basic metrics such as tree height, crown area, and crown radius.

The Crown Metrics Resolution controls the spatial resolution used to estimate crown area.

Tree crowns contain both actual gaps in the canopy and apparent gaps caused by incomplete scanning or low point density. At very fine resolutions, these gaps may strongly influence the estimated crown area.

However, in many applications, we are interested in the outer crown envelope rather than individual gaps within the canopy.

Larger resolution values (e.g., > 0.1 m) increasingly reduce the influence of small internal gaps, resulting in a smoother representation of the crown area.

Typical starting values are:

• TLS (Terrestrial Laser Scanning): 0.1 m
• ULS (Uncrewed Laser Scanning): 0.2 m
• ALS (Airborne Laser Scanning): 0.5 m

The appropriate resolution depends on point density, data completeness, and the desired level of detail.

Iterations

For most datasets, the Iterations parameter can be left at 1.

Additional iterations are mainly intended for very tall trees captured by ground-based scanning systems, such as TLS. In these datasets, larger vertical gaps may occur in the upper canopy, preventing the algorithm from establishing continuous connections from the ground to the tree top.

As a result, segmentation may stop before reaching the upper crown.

To address this, additional iterations progressively increase the allowed vertical Maximum Gap Size, while keeping the horizontal gap size unchanged.

For example, with an initial Maximum Gap Size of 0.5 m:

• 1st iteration: Vertical gap size = 0.5 m
• 2nd iteration: Vertical gap size = 1.0 m
• 3rd iteration: Vertical gap size = 2.0 m

This allows the algorithm to bridge increasingly large vertical gaps while maintaining the original horizontal connection threshold.