Extraction of Plant Ecological Indicators and Use of Environmental Simulation Methods Based on 3D Plant Growth Models: A Case Study of Wuhan’s Daijia Lake Park ^†

Abstract

The acquisition of plant ecological indicators, such as leaf area index and leaf area density values, typically relies on labor-intensive field sampling and measurements, which are often time-consuming and hinder large-scale application. As different plant ecological indicators are closely related to plants’ geometric characteristics, the development of dynamic correlation and prediction methods for relevant indicators has become an important research topic. However, existing 3D plant models are mainly used for visualization purposes, which cannot accurately reflect the plant’s growth process or geometric characteristics. This study presents a workflow for parametric 3D plant modeling and ecological indicator analysis, integrating dynamic plant modeling, indicator calculation, and microclimate simulation. With the established plant model, a method for calculating and analyzing ecological indicators, including the leaf area index, leaf area density, aboveground biomass, and aboveground carbon storage, was then proposed. A method for exporting the model-generated data into ENVI-met v.5.0 to simulate the microclimate environment was also established. Then, by taking Daijia Lake Park as an example, this study utilized site planting construction drawings and field survey data to perform parametric modeling of 21,685 on-site trees from 65 species at three different growth stages using Blender v.4.0 and The Grove plugin v.10. The generated plant model’s accuracy was then verified using the 3D IoU ratio between the models and on-site scanned point cloud data. Plant ecological indicators at various stages were then extracted and exported to ENVI-met for microclimate analysis. The workflow integrates the simulation of plant growth dynamics and their interactions with environmental factors. It can also be used for scenario-based predictions in planting design and serves as a basis for urban green space monitoring and management.

1. Introduction

1.1. Applicability of 3D Plant Models in Ecological Analysis

Urban green spaces play a vital role in providing ecosystem services such as microclimate regulation [1,2,3], carbon sequestration, emission reduction [4,5], and air purification [6,7]. In order to improve the accuracy of ecosystem service assessments, evaluation methods have shifted from 2D planting-area-based calculations to dynamic simulations with 3D plant models [8,9,10]. Unlike traditional gray infrastructure (e.g., buildings, roads, and drainage systems), plants are living organisms and grow and change continuously, leading to seasonal and annual variations in associated ecological indicators and ecosystem services [11,12]. Plant ecological indicators are closely tied to plants’ geometric characteristics [13], driving the need for dynamic correlation and prediction methods [14,15,16].

Existing ecological indicators of urban green space vegetation, such as leaf area index (LAI), leaf area density (LAD), biomass, and carbon storage, are typically collected and measured through field sampling [17,18], which is often time-consuming and involves a certain degree of destructiveness to the plants, making it inapplicable for the rapid acquisition of data across large areas [19]. At the regional scale, the estimation of plant ecological indicators mainly relies on the analysis of remote sensing images [20]. However, due to the level of data availability of high-resolution remote sensing images, the assessment accuracy is relatively limited and cannot reflect subtle differences in the indicators at the 3D scale [21]. Moreover, different geographic environments have a significant impact on analysis models, often requiring localized adjustments and calibrations [22,23].

The visualization of 3D plant models has been widely used to express plant growth characteristics [24,25,26]. Although some existing design software (e.g., RhinoLands and Vectorworks) and rendering software (e.g., Twinmotion, Enscape, and Lumion) already include various 3D plant models as built-in assets, the geometric features of these models are typically static [27,28], so they are unable to reflect the dynamic changes in plant morphology caused by plant growth and environmental interactions [29]. This results in the individual morphological characteristics of generated plant models being similar and lacking diversity, making it difficult to accurately depict the natural growth characteristics of plants [30,31].

For urban microclimate simulations, numerical modeling and computational fluid dynamics (CFD) are the two key methods for analyzing vegetation–environment interactions [32]. Widely used tools include ANSYS Fluent, PHOENICS, and OpenFOAM. Zhang et al. used ANSYS Fluent to simulate the effects of vegetation spatial layout and airflow on microclimate temperature under nine scenarios [33]. Sun et al. employed PHOENICS to simulate outdoor wind environments, and they identified key areas and established the linkage between campus vegetation distribution, microclimate, and comfort in outdoor spaces [34]. Mirza et al. integrated building temperature effects with the influence of vegetation using OpenFOAM [35]. Most of these tools provide detailed geometric modeling capabilities and emphasize accurate simulation of physical processes [36]. However, these tools are designed for the general physical environment and lack the detailed modeling capacity for urban vegetation. Some specialized environmental analysis and simulation tools, such as the i-Tree tools and ENVI-met, contain the data structure definitions and predefined attributes of different tree species [37]. For example, the Albero module in ENVI-met (Version 5.0 and above) incorporates an L-system algorithm-based modeling module and the physiological and ecological characteristics of certain tree species [38,39]. However, the tree species available are still limited and are typically based on local climatic conditions in the United States and Europe, and lack variations in other regions [40,41]. Existing software for plant simulation often requires extensive manual operations and typically treats each tree as an isolated object from the environment. For example, i-Tree Eco requires the manual input of each tree’s parameters; however, the trees in ENVI-met’s Albero are static when modeled, lacking dynamic growth and interaction. Additionally, most existing tree growth prediction models rely on simple rules, which are not capable of responding to surrounding plants and environmental factors.

1.2. Problem Statement

In current urban and landscape design practice, existing 3D plant models are primarily used for visualization purposes. The modeling processes are typically optimized for visual effects such as 3D rendering and live animation [42,43]. Due to the geometric complexity of plant branches and leaves, the plant modeling process is often oversimplified, resulting in model components, such as branches and leaves, which cannot accurately reflect the plant’s growth process or geometric characteristics [44,45]. The detailed plant models are often represented as a “proxy” during the modeling process, with their geometries only loaded and rendered at the final visualization stage [46]. These limitations have restricted their further application for plant growth simulation and analysis. Therefore, current 3D plant modeling methods have not been widely applied in the research and practice of ecological analysis and monitoring [47], and there is a high potential demand for the application of detailed plant models in the extraction and analysis of ecological indicators [48,49].

Unlike static gray infrastructure models, authentic 3D plant modeling requires a modeling method to be established that can accurately reflect the geometries and attributes of different plants, their dynamic growth processes, and interactions with the environment [50,51,52,53]. With recent advances in parametric algorithms and generative modeling, plant model generation based on real morphological features has emerged as a new research trend [51,54,55].

Three-dimensional plant models are commonly constructed using data-driven reconstruction based on field measurements or rule-based modeling derived from plant geometry and growth processes. The reconstruction of 3D plants utilizes images or point cloud data to extract tree structure parameters for 3D reconstruction. Havel et al. employed a neural network on a single plant image to extract tree geometries, identify species, and predict geometric values for branch reconstruction [56]. Guo et al. used multi-view plant depth images to generate a 3D point cloud and proposed a procedural modeling approach with the generated point clouds [57]. Münzinger et al. reconstructed individual tree crowns from LiDAR point clouds using segmentation and parametric algorithms, enabling their integration into semantic 3D city models for urban digital twins [58]. However, these tree reconstruction methods rely on high-quality images and point cloud data, along with significant survey work, which limits their use in large-scale applications [59]. Parametric modeling methods, such as the L-system algorithm and its variations, have been widely used for plant growth simulation, including branching orders and leaf morphology with predefined parameters. Combined with algorithms simulating the interactions between plant physiological processes and the environment, plant morphology and structure can be controlled to express the 3D geometric characteristics of plants [60,61,62]. Petrenko et al. proposed a 3Gmap L-System grammar for generating complete plant models, including stems, stamens, petals, and leaves [63]. Ding et al. optimized L-system geometric parameters using an improved particle swarm optimization (PSO) algorithm and simulated plant forms at different growth stages [64]. Liu et al. utilized growth model data to link the tree crown geometries with IFS fractals, providing an effective method for tree growth simulation [65]. The rules are primarily defined based on the plant’s growth changes and structural variations. However, parametrically generated trees are mainly used for visualization purposes and have not been tested in plant indicator calculations or environmental simulations.

1.3. Project Goals

For the generation of dynamic plant models and the calculation of related ecological indicators, this study employed field sampling and parametric modeling methods to generate 3D dynamic growth models of 21,685 trees in an urban park with Blender v.4.0 and “The Grove” plugin v.10 [66]. Blender is an open-source software developed by the Dutch animation studio NeoGeo (Amsterdam, The Netherlands), using Python as its built-in scripting language [67]. “The Grove” is a plant growth simulation plugin developed by F12, which adopts a parametric method to generate trees by controlling various internal factors (growth rings, branching patterns, branch height, curvature, deflection, etc.) and external factors (lighting, wind, obstacles, seasonal changes, etc.). Each individual plant can be generated as a unique geometry influenced by growth processes and external environmental conditions [68].

The Intersection over Union (IoU) is a commonly used metric for evaluating the overlap of 2D geometries. It is the ratio of the area of the intersection between the predicted region and the ground truth region compared to the area of their union. To evaluate the accuracy of the generated 3D plant model, by referring to the calculation method of 2D IoU, this study developed a 3D spatial IoU calculation method through 3D spatial slicing [69]. The 3D spatial IoU between the generated plant models and those obtained from field-based LiDAR scanning was used to verify the accuracy of the generated models.

With 3D models established that can reflect plant morphology and growth processes, the acquisition of ecological indicators can be converted into a geometric calculation process based on 3D models. This study further examined methods for the extraction of ecological indicators and microclimate simulation of plant growth models. Based on the parametric models at different growth stages, a voxel sampling method was proposed to calculate the four ecological indicators (LAI, LAD, aboveground biomass, and aboveground carbon storage). The LAD and the plant distribution data were then imported to ENVI-met v.5.0 for further analysis, which enables more accurate microclimate and environmental simulation results.

2. Materials and Methods

2.1. Data Acquisition and Processing

This study selected Daijia Lake Park as a case study. Located in Qingshan District, Wuhan City, Hubei Province, it is Wuhan’s first urban park built on a brownfield. The park’s construction started in early 2013, with planting completed in late 2013. The park covers an area of 50.67 hm², with about 42.00 hm² of green space and 4.70 hm² of water areas. There are a total of 21,685 trees across 65 species planted in the park, resulting in a green coverage rate of 91.8% [70]. A highway overpass goes across the park in the northeastern area, resulting in a linear shading area for the vegetation in the park.

2.1.1. Data Acquisition

The study selected three typical plant growth stages for the analysis: the initial growth stage at the park’s planting completion date in 2013, the current growth stage at the site’s survey date in 2023 (10 years later), and the predicted mature stage of the plants in 2033 (20 years after the initial growth stage). The three stages were selected to represent the temporal changes in the growth of park trees.

The basic data for constructing models at different growth stages includes the park’s planting design and construction drawings, plant species lists, geometric features of different tree species obtained through field surveys, and allometric equations for different tree species [71], as shown in

Table 1. Data sources and applications.

Data	Time	Purpose	Application Stage
Park’s Planting Design Drawings	2013	To obtain planting locations and tree species.	Initial stage
Plant Species Lists	2013	To obtain a list of plants and initial tree dimensions.	Initial stage
Google Earth Online Remote Sensing Imagery	2013	To validate the average crown width of different tree species on the site.	Initial stage
Field Surveys	2023	To obtain twig geometries (twig length, leaf width and length, and petiole length)	Growing stage
Allometric Equations Data for Different Tree Species	--	To predict tree growth dimensions (DBH, tree height, and crown width) with equations corresponding to similar climate zones	Growing stage Mature stage

.

The plant species, planting locations, and their initial geometric sizes were obtained from the planting design drawings of Daijia Lake Park, which included a total of 21,685 trees across 65 species. The initial dimensions of the plants were extracted based on the specifications listed in the park’s plant species lists. To ensure that the geometric features of the generated plant models remained consistent with the actual morphology, the study utilized field survey data on the twigs of different tree species to construct corresponding geometry models.

2.1.2. Site Segmentation

To reduce computational load and ensure scalability across various simulation scales, this study established a tile-based simulation method by dividing the site into multiple rectangular tiles, performing growth simulations on each tile separately, and then stitching them together to generate a complete site model. Specifically, the park was divided into 64 tiles with a grid size of 100 × 100 m. For each tile, parametric plant modeling was performed using a unified set of simulation parameters (Figure 1). To address the absence of interactions between plants within and outside the tile cropping boundary, this study uniformly expanded each tile’s simulation boundary outward by 5 m, resulting in a 110 × 110 m square tile that retained neighboring plants outside the site boundary, allowing them to interact dynamically with the plants within the grid. After generating the plant models on each tile, they were then trimmed into a series of 100 × 100 m grid tiles again to remove peripheral reference plants outside the boundaries. The generated models were then stitched together based on the corresponding locations of each tile to generate a complete 3D vegetation model of the park.

2.2. Plant Growth Model Generation

2.2.1. Geometric Features Acquisition and Prediction of Different Tree Species

To predict growth changes in different tree species, this study conducted simulations using allometric equations for various tree species. The allometric equations utilize tree age (x) as the primary variable, which is substituted into a set of equations to calculate DBH, tree height, and crown width, thereby estimating the spatial dimensions of plants at various growth stages [72]. Since there is a lack of publicly available growth data for relevant tree species in China, this study utilized allometric equations created by the U.S. Forest Service, based on tree surveys from multiple cities in the United States [71]. To ensure the applicability of the growth equations, based on the geographical location of Daijia Lake Park (29°58′–31°22′ N, 113°41′–115°05′ E), tree species equation data from counties in the subtropical monsoon and monsoon humid climate zones of the United States that are similar to the site were selected from the database for the growth prediction of different plant species.

The study obtained data on DBH, tree height, and crown width at planting from the plant species lists in the planting design documents and verified the locations and crown widths of the trees using historical remote sensing images from Google Earth Pro 7.3 in 2013. The age data of trees provided by plant suppliers served as the initial ages for each species and were then incorporated into allometric equations to predict tree dimensions at various growth stages (Figure 2).

2.2.2. Individual Plant Model Generation

In the modeling process, each plant was divided into two main components: the branches and a set of attached twigs. A twig is a 3D model that represents the latest flush of growth, i.e., a small branch with leaves [68]. Using Blender v.4.0 and “The Grove” plugin, this study conducted parametric modeling of individual plants of different tree species, with twigs and branches modeled separately. The twig model for a particular tree species is regarded as a static geometric mesh. For twig models of various tree species, the geometric features are obtained through field measurements and modeling in Blender.

The branch models were developed based on the morphological features of various tree species. The interactive growth and branching process was manipulated parametrically by adjusting the growth rate and environmental conditions. Trees were categorized into three types according to their branching patterns: monopodial, sympodial, and whorled branching. The three variables in an allometric equation—tree height, crown width, and sub-branch height—were established as geometric constraints for each growth stage of the trees. Twig models were then attached at the branching points of the branch model. By adjusting the relevant “Twig” parameters in “The Grove” plugin, the growth behavior and density of twigs could be controlled. Blender’s Geometry Nodes were used to further refine the tree morphology, spatial proportions, and structural details (Figure 3). Based on the branching characteristics of different tree species, parameters such as growth, turning, thickening, and bending were used to simulate their growth process, ultimately generating branch growth models for 65 tree species (Figure 4).

For the site terrain modeling, contours and elevations from the site’s topographic design drawings were used to create a 3D terrain model in Blender. This terrain model served as the foundation for attaching plant models. Using the planting positions from the park’s planting design and construction drawings as references, individual plant models were created at the initial stage and placed at the corresponding planting locations within the 64 tiles for the simulation.

2.2.3. Environmental Factors Simulation

Plant growth is affected by various factors, including responses to environmental conditions and interactions between plants [73]. This study simulated the effects of three main influential factors: phototropic growth toward sunlight, competition among trees, and physical obstructions on plant growth (Figure 5). By controlling growth duration parameters, the dynamic growth process of each plant was modeled, resulting in on-site plant models for both the growing and mature stages.

For the simulation of plant growth reactions to sunlight, the distribution and growth patterns of plant branches under corresponding lighting conditions were simulated by adjusting the parameters, including “Leaf Area,” “Reduce,” and “Alongside,” in The Grove plugin’s built-in shading module.

For the simulation of plant growth competitions, factors such as plant branching structure, branch density, branch survival rate, and environmental impact were simulated to model the competitive relationships between plants by adjusting related parameters in the plugin, including “Favor,” “Drop,” “Add,” and “Shade.”

For the simulation of building obstructions on plant growth, relevant parameters in the plugin were adjusted, including growth suppression caused by shaded areas adjacent to the overpasses and buildings, as well as phototropic growth bending in response to light availability.

In the plant growth simulation process, species-specific allometric equations were used to predict the geometric dimensions of each tree at both the growing and mature stages, which served as spatial constraints for dynamic simulations. Vegetation growth processes were simulated for each of the 64 tiles (Figure 6).

2.3. Accuracy Verification

To verify the geometric accuracy of the generated vegetation model, this study proposed a method to calculate IoU in 3D space. This metric was used to measure the overlap percentage between the generated plant models and actual plant models obtained through LiDAR scanning by calculating the ratio of the overlapping volume between the predicted model boundary and the actual model boundary to the union of the two models (Equation (1)).

$$IoU={\displaystyle \frac{S_1\cap S_2}{S_1\cup S_2}}$$

(1)

In Equation (1), $S_1$ represents the measured data; $S_2$ represents the target data; $S_1\cap S_2$ represents the overlapping volume of the two regions; and $S_1\cup S_2$ represents their combined total volume. The IoU value ranges from 0 to 1, with higher values indicating a greater degree of overlap and thus higher model accuracy.

2.3.1. Field Sampling

This study validated the modeling accuracy by comparing the generated models with point cloud data scanned on-site using a handheld LiDAR scanner. Three 100 m × 100 m plots (Figure 7) with different vegetation communities were selected to validate the accuracy of the generated models. Plot 1 consisted of a broadleaf mixed forest; Plot 2 was a conifer–broadleaf mixed forest with highly variable canopy structures; and Plot 3 was dominated by tall conifers. The three plots contained 18 tree species.

A handheld laser scanner (GeoSLAM ZEB Revo RT, manufactured by Scanner (Beijing) Science and Technology Co., Ltd., Beijing, China) was used to collect the point cloud data of Plot 1 in July 2023. Data for Plot 2 and Plot 3 were collected in August 2025. The corresponding plant growth models for each site were generated for later validation. The obtained point cloud data was then imported into CloudCompare v.2.13 for calibration, stitching, and segmentation to create a complete point cloud model of the sample site.

2.3.2. Verification Process

To evaluate the 3D IoU of the sample site model and ground truth, we employed a slicing method by segmenting the plant model at equal intervals along the horizontal plane. This created multilayered voxel slices, and we calculated the intersection and union of each slice pair to obtain the overall result [74]. The slicing method also enabled the validation of the model accuracy at a certain height.

For the 3D IoU calculation, the voxel sampling size should be set to balance the need for modeling accuracy and computational efficiency. The voxel sampling size and the IoU value are positively correlated in the test (Supplementary Materials Table S5), with smaller sizes better able to capture the geometry of plant leaves and branches. Since plant branching and growth always involve some randomness, having a too-small sampling size is irrelevant. Additionally, for ecological indicator calculations at the leaf-cluster level, the randomness in distribution can produce statistically even results and has a limited impact on the outcomes. Therefore, in this study, a voxel resolution of 0.1 m³ was used for the IoU calculation, maintaining consistency with the sampling interval used for ecological indicator assessments.

Using CloudCompare, the point cloud model was down-sampled to a 0.1 m equal-interval voxel grid to serve as the ground-truth dataset. In addition to the overall IoU calculation of the model, the IoU variations across different heights were also calculated by model slicing. Both the ground-truth and the model-generated point cloud datasets were segmented at equal intervals along the z-axis, with a 1 m interval. This process produced 10 sets of IoU values corresponding to different slices at various vertical heights. The ratio of the accumulated intersection and union values was then calculated to obtain the final IoU result.

The 3D spatial IoU calculation results of the 0.1 m voxel models for Plot 1, Plot 2, and Plot 3 are 61%, 57%, and 59%, respectively, indicating that the generated models could match the actual 3D vegetation spatial distribution at a certain level and can reflect the growth conditions of the trees to a certain extent (Figure 8). The slicing also reveals that the lower parts of the generated plant model are more accurate than the upper parts, suggesting that the tree growth simulation at the top level requires further refinement to enhance accuracy. The differences in canopy structure also correlate with the modeling accuracy. The model parts with more complex combinations of tree species have lower accuracy, while relatively simple plant communities show higher accuracy across the model.

By extracting individual tree data from the point cloud, a total of 90 trees across 8 species were obtained. A comparative analysis of each tree’s geometric indicators (tree height, crown width, and DBH) between the growth equation and the on-site observations was performed. The results show that tree height and crown width are more accurate than DBH, and the relationship between tree height and crown width shows relatively consistent deviation patterns. The RMSE range for tree height is 0.581–6.201, while for the crown diameter it is 0.767–5.075. Among the surveyed species, the allometric equations are more accurate for small-to-medium-sized trees such as Ligustrum lucidum and Celtis sinensis, whereas larger trees like Metasequoia glyptostroboides and Pinus elliottii have less accurate results, with predicted growth rates generally overestimated.

2.4. Ecological Indicators Extraction and Analysis

Based on the generated 3D plant model, corresponding calculation methods are proposed for four ecological indicators: LAI, LAD, aboveground biomass, and aboveground carbon storage. LAI and LAD are geometric indicators and can be directly calculated from the model, while aboveground biomass and aboveground carbon storage are estimated using both the geometries and the biological attributes of corresponding tree species. By analyzing the ecological indicators across different plant growth stages, we can also conduct predictive analysis of plant growth, which could provide a method for long-term ecological performance assessment of the plant community.

2.4.1. Model Data Processing

The LAI and LAD of plants were calculated using the total leaf models of the plants, while aboveground biomass and aboveground carbon storage were determined from all the aboveground plant models. For the calculation, models of each tree species were grouped separately, and the resulting leaves and branches were categorized and labeled for analysis.

To calculate leaf areas and branch volumes, different segments of the 3D plant models were down-sampled into an evenly spaced point cloud using Houdini v.19.0. Each point in the model represents either a unit area (for leaves) or a unit volume (for branches). The down-sampling approach for calculating these metrics, based on the geometric features of plant leaves and branches, enables quick estimation of relevant measurements. For plant leaves, a surface point sampling method was used. Because leaves are usually thinner than the 0.10 m sampling interval, the leaf surface area can be determined by the number of sampling points on the leaf surface. For branches and trunks, spatial points are sampled at 0.05 m intervals to create a voxel model, which is then used to estimate the corresponding spatial volume (Figure 9). This method is generally suitable for calculating leaf area and branch volume across various tree species, and by adjusting the sampling interval, it can accommodate different accuracy requirements.

A smaller sampling distance produced results closer to the LAI of the actual model but required more computational power. To evaluate how different sampling point distances affect the LAI calculation accuracy, we performed a comparative analysis using a plant community with three trees on a 1 m × 1 m grid. Using the LAI calculation with a sampling distance of 0.02 m as the reference (since the number of point samples at 0.01 m resolution exceeded the software limit), we created 17 test datasets at various sampling distances. The RMSE analysis indicates that bias increased with larger sampling intervals, while a 0.10 m sampling distance may offer a good balance between accuracy and performance (

Table 2. RMSE evaluation results at different sampling point distances.

Distance (cm)	0.03	0.04	0.05	0.06	0.07	0.08	0.09	0.10	0.11	0.12	0.13	0.14	0.15	0.20	0.30	0.40	0.50
RMSE	0.020	0.034	0.035	0.051	0.059	0.072	0.083	0.089	0.102	0.120	0.144	0.149	0.176	0.144	0.149	0.176	0.245

).

2.4.2. LAI Calculation

LAI refers to the total leaf area per unit of the ground area. It is a common indicator of plant growth status. Using 3D models of plants, LAI calculations can be converted into the total area represented by points per unit area in the leaf model (Equation (2)).

$$LAI={\displaystyle \frac{N\times d^2}{a}},$$

(2)

In Equation (2), $N$ represents the total number of sample points within a unit planar grid area. $d^2$ represents the spacing between adjacent sample points, corresponding to the sum of the leaf area, and $a$ represents the area of the unit grid, measured in m².

This study used point cloud models of leaves at various stages, with a sampling interval of 0.1 m between adjacent points and a 1 m × 1 m unit for LAI 2D mapping. The LAI value for each unit area was determined by the total number of points in each grid, with each point representing a leaf area value of 0.01 m².

2.4.3. LAD Calculation

LAD refers to the total leaf area of plants within a unit volume at a certain height (m²/m³). It is a parameter used to characterize the canopy structure of the plant communities. Based on the 3D model of plants, the calculation of LAD can be converted to the total leaf area within a unit volume (Equation (3)).

$$LAD={\displaystyle \frac{N\times d^2}{v}},$$

(3)

In Equation (3), $N$ represents the total number of point cloud points within a unit volume. $d^2$ represents the squared spacing between points, corresponding to the sum of the leaf area, with units of m². And $v$ represents the volume of the statistical unit voxel, measured in m³. In this study, $d$ is set to 0.1 m and $v$ is set to 1 m³.

This study used point cloud models with a sampling interval of 0.1 m between adjacent points. The point cloud models of leaves were sliced into 1 m³ voxel grids, and the number of points in each voxel grid was counted to obtain the LAD value represented by each voxel unit.

2.4.4. Aboveground Biomass and Aboveground Carbon Storage Calculation

Aboveground biomass and aboveground carbon storage indicate the aboveground mass and carbon sequestration capacity of plants. Using 3D plant models, the corresponding leaf area and branch volume data were obtained.

The aboveground biomass was calculated by summing the biomass of leaf and branch components using biomass conversion coefficients specific to each tree species (Figure 10). This study used data on Conversion Factors for Leaf Area to Biomass and Wood Density Values [75,76] from the U.S. Forest Service’s i-Tree database to determine the leaf biomass and branch biomass of various tree species and to estimate the biomass of plants at different growth stages (Equation (4)).

$$\begin{gathered} B_{Total}={\sum }_{i=1}^k(B_{leaf, i}+B_{branch, i}), \\ {B}_{leaf}={\sum }_{i=1}^k{\sum }_{i=1}^n{(S}_i\times a_i)\times {10}^{-6}, \\ {B}_{branch}={\sum }_{i=1}^k{\sum }_{i=1}^m{(V}_i\times b_i), \end{gathered}$$

(4)

In Equation (4), $k$ represents the total number of tree species; $B_{Total}$, $B_{leaf}$, and $B_{branch}$ represent the total aboveground biomass of trees; and the leaf and branch biomass of tree species is represent by $i$, with all units given in tons. $S_i$ represents the leaf area of different tree species, measured in m²; $n$ represents the total number of points for the leaves of different tree species, with each sample point corresponding to a leaf area of 0.1 m × 0.1 m; and $a_i$ represents the leaf area compared to the biomass conversion coefficient for tree species $i$. $V_i$ represents the branch volume of different tree species, measured in m³; $m$ represents the number of points for the branches and trunks of various tree species, with each sample point corresponding to a spatial volume of 0.05 m × 0.05 m × 0.05 m; and $i$ represents the wood density of tree species $i$.

The aboveground carbon storage of the trees was calculated by multiplying the aboveground biomass and the species-specific carbon content rate (Equation (5)).

$$C_{Aboveground carbon storage}={\sum }_{i=1}^k{(B}_i\times {CF}_i),$$

(5)

In Equation (5), $k$ represents the number of tree species; $C_{Aboveground carbon storage}$ represents the total aboveground carbon storage of each tree species in the park, measured in tons. $B_i$ represents the aboveground biomass of tree species $i$ within the regional area, measured in tons; ${CF}_i$ represents the carbon content rate of tree species $i$.

2.5. ENVI-met Simulation

ENVI-met is a 3D modeling software used to simulate and analyze urban microclimates by modeling the interactions between urban surfaces, vegetation, buildings, and the atmosphere [77]. We developed a method to export 3D plant modeling and LAD data to ENVI-met for further simulation, and we used the Fangcaoyuan Garden in Daijia Lake Park as a case study to examine how plant communities affect the park’s microclimatic conditions.

2.5.1. ENVI-met Plant Modeling

For the built-in vegetation models in ENVI-met, the LAD values of generic trees are usually constant or simply vary with distance from the tree’s geometric center. The L-system plant samples include a number of species-specific geometric details. However, the species coverage remains limited, and the models are static, lacking growth control based on environmental factors, which constrains the simulation accuracy. Based on the 3D plant models generated and the 1 m³ resolution LAD model, a data exchange method was developed for ENVI-met simulation. The grid cells in the “Albero” module of ENVI-met were set to dimensions of 1 m × 1 m × 1 m corresponding to the generated LAD model [78]. Using the parameters of the modeled vegetation—such as species type, geometric parameters, and LAD values for each plant in different coordinate grid directions—individual plants were generated, resulting in a site-specific plant database (Figure 11 and Figure 12).

The sample area covers three growth stages for seven plant species: Celtis sinensis, Cedrus deodara, Osmanthus fragrans, Styphnolobium japonicum, Platanus orientalis, Magnolia grandiflora, and Lagerstroemia indica.

2.5.2. ENVI-met Site Modeling

To minimize the boundary impact on the simulation results, the simulation area was expanded by three cells to encompass the surrounding environment, resulting in a final simulation area of 50 m × 50 m × 25 m with a resolution of dx = dy = dz = 1 m. The landscape modeling elements in ENVI-met included buildings, overpasses, roads, and green spaces. Site data, such as contour lines, tree planting locations, building layout, and overpass height, is obtained from the landscape design drawings of Daijia Lake Park. Using the “Spaces” module in ENVI-met, terrain, building information, and vegetation elements were added to construct a 3D model of the sample site (Figure 13).

2.5.3. ENVI-met Parameter Settings

Wuhan experiences abundant solar radiation and sustained high temperatures during summer. The date parameter for the simulation was set to August to reflect the plant’s microclimate regulation effects. Local weather data were obtained from EnergyPLUS [79]. We used the meteorological data from 13 August 2023, for the simulation (

Table 3. Parameter settings of ENVI-met.

Parameter Name	Parameter Values
Simulated date	13 August 2023
Time period	7:00 a.m.–14:00 p.m.
Temperature (°C)	20–36 °C
Wind speed (m/s)	1 m/s
Wind direction	225°
Humidity (%)	67%–87%

). The time was set from 7:00 to 14:00 to represent changes in the thermal environment. Using a pedestrian height of 1.5 m as a reference, the analyzed microclimate parameters included mean radiant temperature, humidity, and wind speed, and their numerical variations were evaluated across the three growth stages within the study area. In this study, the results mainly aimed to compare the relative differences and trends in microclimate caused by plant growth, which may differ from actual conditions.

3. Results

3.1. Plant Models at Different Growth Stages

Using a generative model of 21,685 trees in Daijia Lake Park, the growth process of the plants was simulated to produce plant community models for the three stages, as shown in Figure 14. The sample sites and their corresponding plant growth statuses during the same period are shown in Figure 15.

3.2. Plant Ecological Indicators Calculation Results

3.2.1. LAI Calculation Results

The 1 m resolution LAI map of the site was generated based on the converted 0.1 m³ point cloud of the leaves (Figure 16). As the trees continued to grow, the LAI values of the plants gradually increased at each stage. The LAI value range in the study area was 0–3.53 (Figure 17). No LAI anomalies were observed, as the generated plant model corresponded with the actual planting conditions, and no instances of plant geometry overlap were detected.

To validate the accuracy of the LAI extracted from the plant growth models, on-site LAI measurements were performed for Plot 2 and Plot 3 using hemispherical photography. Field surveys took place on 22 August 2025, with tree canopies photographed using a 180° fisheye lens camera. Five sampling points were randomly chosen within each plot, and their latitude and longitude coordinates were recorded. At each sampling point, five hemispherical photographs were taken at a height of 0.8 m. The photos were imported into Hemiview canopy analysis software to determine canopy gap positions, sizes, density, and distribution, which were then used to calculate the LAI. The average values from these measurements were used as the LAI for each sampling point, resulting in a total of 10 valid data points across the two plots.

As related research indicates, optical measurement instruments do not account for the non-random distribution and clumping of leaves. The values achieved are recognized as effective LAI, which are lower than the actual LAI [80]. Therefore, the actual LAI was calculated using the empirical correction formula proposed by Guo for the urban green spaces in Wuhan [81].

$${LAI}_a=2.28\times {LAI}_e^{1.14},$$

(6)

In Equation (6), ${LAI}_a$ represents a calibrated leaf area index, and ${LAI}_e$ represents the effective leaf area index.

The 5 m resolution LAI dataset generated from the plant model in 2025 was used for comparative analysis (Figure 18). The comparison between the field-measured data and the model-predicted results (

Table 4. Comparison of LAI values.

	Plot 2					Plot 3
	2-1	2-2	2-3	2-4	2-5	3-1	3-2	3-3	3-4	3-5
Effective LAI	1.424	1.919	0.554	0.624	1.674	1.598	0.842	1.014	0.193	0.613
LAI	3.413	4.792	1.163	1.333	4.104	3.891	1.874	2.316	0.350	1.304
Model-predicted LAI	2.872	3.642	1.089	0.800	3.102	3.186	0.807	2.162	0.125	1.059

) showed a mean deviation of −0.570, indicating that the LAI values calculated from the growth models were slightly lower than the on-site LAI.

3.2.2. LAD Calculation Results

The LAD spatial distribution model with a resolution of 1 m³ was created using the 0.1 m³ point cloud of the leaves. Using Fangcaoyuan Garden as an example, the LAD distribution at three stages is shown in Figure 19, with a range of 0–3.45 m²/m³.

3.2.3. Aboveground Biomass and Aboveground Carbon Storage Results

As the plants grew, their aboveground biomass and aboveground carbon storage increased steadily (

Table 5. Aboveground biomass and aboveground carbon storage of trees at three growth stages.

Growth Stages	Geometric Properties of Tree Models
	Number of Leaf Point Clouds	Number of Branch Point Clouds	Total Leaf Area (m2)	Total Branch Volume (m3)
Initial Stage	86,935,676	3,449,963	869,356.76	431.25
Growing Stage	762,472,591	34,827,755	7,624,725.91	4353.47
Mature Stage	1,312,715,284	112,958,322	13,127,152.84	14,119.79
	Total aboveground biomass and carbon storage
	Leaf Biomass (t)	Branch Biomass (t)	Total biomass (t)	Carbon Storage (t)
Initial Stage	70.58	230.78	301.35	147.71
Growing Stage	619.91	2031.24	2651.15	1297.13
Mature Stage	1078.80	6513.90	7592.70	3728.40

). The total biomass of trees during the growth and mature stages in the park was 8.80 times and 25.20 times that of the initial stage, respectively, while the carbon storage of the two stages was 8.78 times and 25.24 times higher than the initial stage, respectively. By comparing the ecological indicators of the 65 tree species in Daijia Lake Park, it can be seen that Cedrus deodara, Photinia serratifolia, and Metasequoia glyptostroboides have relatively high canopy coverage, significantly contributing to carbon sequestration (

Table 6. Aboveground carbon storage of sample tree species at three growth stages.

Species	Cedrus deodara	Photinia serratifolia	Metasequoia glyptostroboides	Styphnolobium japonicum	Ligustrum lucidum
Carbon content rate	0.4963	0.4901	0.5083	0.4901	0.4502
Initial stage (t)	16.81	38.32	12.96	3.40	8.17
Growing stage (t)	146.11	276.16	244.53	46.06	54.83
Mature stage (t)	950.75	523.36	542.70	203.82	150.05
Total carbon storage (t)	1113.67	837.84	800.18	253.29	213.05

).

To quantify the uncertainty in the total aboveground biomass and carbon storage of different species, a Monte-Carlo simulation was conducted to compare the generated models with the on-site scanned models, focusing on the uncertainty caused by deviations in allometric equations that lead to morphological differences (Figure 20). The results showed that increasing the growth years has a significant impact on precision.

3.3. ENVI-met Simulation Results

The microclimate simulation results for the three growth stages are shown in Figure 21. The type of vegetation and planting configuration significantly affected the on-site temperature, airflow, and wind speed. As vegetation height and canopy density increased with plant growth, there was a decreasing trend in the mean radiant temperature and wind speed across the area, along with an increase in humidity. This indicates that the vegetation in the park has a substantial impact on the local microclimate.

4. Discussion

The method established in this study for obtaining plant ecological indicators and simulating microclimate environments based on generated 3D plant growth models can be used for monitoring and assessing the ecosystem services of plant communities. Unlike static tree assets from ENVI-met or other resources, growth-aware modeling can be adapted to local site conditions and generate site-specific vegetation models for analysis. Compared to methods based on field surveys or remote sensing calculations that derive various ecological indices, the proposed method can improve the convenience and accuracy of obtaining the indicators. It can be applied to different research scales and enables predictions of changes in plant growth. By adjusting the planting pattern and composition of tree species, this study can be used to assess the ecosystem service performance under various planting design scenarios, which could provide a basis for selecting plants and optimizing design layouts.

The accuracy of the extracted plant ecological indicators is directly related to the model’s authenticity. Since plant growth is influenced by a complex combination of environmental factors, the current types of factors considered in plant growth simulations are still limited, and further adjustments to the modeling method are still needed. For plant growth simulation, there remains a lack of systematic research on how relevant indicators affect plant growth.

The method introduced in this study allows for continuous simulation of plant growth. However, because of the high resource demands of simulating and visualizing 21,685 plants across the entire park, only three growth stages were chosen for this simulation. Additional growth stage simulations with shorter intervals can be conducted later based on specific application needs.

Due to the differences in climate and environmental conditions, it is necessary to conduct long-term monitoring of the local environment, climate, and ecological changes for different tree species. Additionally, the effects of urban tree management and maintenance on plant growth still need more discussion. To improve computational efficiency in extracting ecological indicators, this study used a point cloud down-sampling method that involved some simplifications in the indicator estimation process. To determine physiological characteristics in plant models, this study used empirical parameter values from related research without accounting for variations caused by different growth environments for the same tree species. The accuracy of the calculations can be improved through field surveys and site sampling in future studies.

Natural plants include both the aboveground parts and underground roots. Indicators of growth in aboveground parts are relatively easy to measure, while observing and monitoring underground root systems require specialized equipment [82]. Due to limitations in observation and simulation techniques, this study mainly focused on the aboveground parts for modeling and indicator calculation, without considering the underground roots. Future research should explore plant root growth using root observation tools. Additionally, current plant growth simulations mainly emphasize the geometry and growth changes in trees, without accounting for seasonal variations. There is also a lack of research on the growth simulation of stemless plants like shrubs and ground cover, which are also part of plant communities.

For long-term monitoring of the ecological environment and plant growth conditions, environmental sensors and IoT devices can be further integrated to provide continuous data calibration and feedback. Due to the complexity of plant branching morphology, the level of detail (LOD) achievable in large-scale simulations of plant community growth remains limited. When modeling plants and simulating their growth at various scales, there are specific differences in data precision and accuracy requirements. Balancing the realistic representation of ecological characteristics in plant digital models with the convenience of generative modeling still requires further study. The feasibility of adapting to different performance demands by establishing multi-level LOD representation models [83,84] will require further testing and analysis.

5. Conclusions

This study focused on Daijia Lake Park in Wuhan, China, as a case study and created 3D models of urban park vegetation that respond to environmental factors using Blender and “The Grove” plugin. These models can be used for plant visualization and ecological simulation analysis. A method for calculating the 3D spatial IoU metric was proposed to evaluate the accuracy of the generated models. In terms of the indicator calculation, based on the generated 3D plant model, calculation methods were developed for four ecological indicators: leaf area index, leaf area density, aboveground biomass, and aboveground carbon storage. An analysis process was also established by importing the 3D plant model into ENVI-met for more detailed microclimate analysis. This approach has a certain scalability and can be applied to tree growth prediction, ecological analysis, and the comparison of design scenarios in urban green space planning, design, and management.