TreeSpecies PC2DT: Automated Tree Species Modelling from Point Cloud to Digital Twin

Digital twin (DT), a technology of virtual representation of an object or system over its lifecycle, is advancing urban design and planning in Singapore. It enables urban planners to make informed decisions, elevate quality of life, and enhance sustainability efforts. While most DT cities focus on buildings and urban structures, limited attention is given to dynamic living structures such as trees. To generate DT trees, challenges in city-scale automation/maintenance, as well as growth dynamics and species variation of individual trees must be addressed. Currently, trees in resource-constrained DT cities have been limited to simple low-resolution, static models for simulation compatibilities and cost performance.

To address the current limitations, researchers from A*STAR’s Institute of High Performance Computing (IHPC) devised TreeSpecies-PC2DT (Point Clouds to Digital Twins), an automated digital twin tree species modelling workflow from remote sensing data (Fig 1). TreeSpecies-PC2DT leverages non-temporal, limited point cloud scans (Fig 2a) to generate large-scale, lightweight and dynamic tree models representative of their botanical species characteristics (Fig 2b).

Fig 1. TreeSpecies-PC2DT Workflow

Original Point Clouds

Digital Twin Models

Fig 2. Tree plot comparison: (a) original point clouds,

(b) digital twin models

Features

Automatic workflow
Leveraging one snapshot of point cloud data to create temporal models
Large, city-scale tree modelling
Lightweight and dynamic DT trees, responsive to environment stimuli
Species-representative growth process and branching patterns

The Science Behind

TreeSpecies-PC2DT workflow begins by processing the remote sensing data of tree point cloud through a sequence (Fig 3) involving five main modules.

Branch reconstruction derives tree trunk and low-order branches for initial physically measurable parameters.
Species profiling taps on the knowledge-based transfer learning from the big data of synthetic species models to predict species architecture parameters of actual trees.
Tropism transfer detects parameters of actual trees’ growth outlook in response to environment stimuli such as sunlight, gravity, etc.
Constraint optimisation completes each individual profile by solving remaining unknown parameters in the profile, such as age and growth rate, within all growth constraints of each actual tree.
Species growth modelling generates a DT tree species for each actual tree through a set of growth rules based on its complete individual profile (Fig 4).

At the backend, along with existing botany knowledge and field measurements, every newly completed individual profile contributes to refining a collection of species profiles, gradually building up a species library. The species library serves to define the search scope of parameters for known species, as well as to identify new or unknown species.

Fig 3. TreeSpecies-PC2DT data transformation: (a) original point cloud, (b) woody components, (c) reconstructed trunk and low-order branches (red) within original growth space (green), (d) digital twin species (yellow) filling up the original growth space

Fig 4. Clockwise from top left: a DT Tabebuia rosea species individual grows with time

Industry Applications

TreeSpecies-PC2DT generates large-scale, lightweight tree species models that represent their true species characteristics and dynamic botanical architecture.

Possible industry applications include:

Tree safety and health management
Environmental simulations
Landscape and green building designs
Social economic impact studies

IHPC Tech Hub

TreeSpecies-PC2DT: Automated Tree Species Modelling from Point Cloud to Digital Twin

IHPC Tech Hub

TreeSpecies-PC2DT: Automated Tree Species Modelling from Point Cloud to Digital Twin

A*STAR celebrates International Women's Day