Determining the Structural Characteristics of Farmland Shelterbelts in a Desert Oasis Using LiDAR

Abstract

The structural analysis of shelterbelts forms the foundation of their planning and management, yet the scientific and effective quantification of shelterbelt structures requires further investigation. This study developed an innovative heterogeneous analytical framework, integrating three key methodologies: the LeWoS algorithm for wood–leaf separation, TreeQSM for structural reconstruction, and 3D alpha-shape spatial quantification, using terrestrial laser scanning (TLS) technology. This framework was applied to three typical farmland shelterbelts in the Ulan Buh Desert oasis, enabling the first precise quantitative characterization of structural components during the leaf-on stage. The results showed the following to be true: (1) The combined three-algorithm method achieved ≥90.774% relative accuracy in extracting structural parameters for all measured traits except leaf surface area. (2) Branch length, diameter, surface area, and volume decreased progressively from first- to fourth-order branches, while branch angles increased with ascending branch order. (3) The trunk, branch, and leaf components exhibited distinct vertical stratification. Trunk volume and surface area decreased linearly with height, while branch and leaf volumes and surface areas followed an inverted U-shaped distribution. (4) Horizontally, both surface area density (Scd) and volume density (Vcd) in each cube unit exhibited pronounced edge effects. Specifically, the Scd and Vcd were greatest between 0.33 and 0.60 times the shelterbelt’s height (H, i.e., mid-canopy). In contrast, the optical porosity (Op) was at a minimum of 0.43 H to 0.67 H, while the volumetric porosity (Vp) was at a minimum at 0.25 H to 0.50 H. (5) The proposed volumetric stratified porosity (Vsp) metric provides a scientific basis for regional farmland shelterbelt management strategies. This three-dimensional structural analytical framework enables precision silviculture, with particular relevance to strengthening ecological barrier efficacy in arid regions.

1. Introduction

Farmland shelterbelts are artificial forest ecosystems established to regulate and improve the structure and function of fragile, disaster-prone agricultural ecosystems. Their ultimate objective is to create or restore sustainable and stable farmland ecosystems, characterized by high productivity and enhanced ecological benefits [1,2,3]. From the standpoint of modern ecology, shelterbelts function as heterogeneous ecotones—zones characterized by ecological stress, edge effects, and biodiversity convergence [4]. Research on their structure aims to elucidate the interrelationships among integrity, stability, and persistence components that define their formation. These insights provide a scientific foundation for shelterbelt design, management, and the sustained delivery of multifunctional benefits, which explains why the farmland shelterbelt structure has long been a focal research area.

The shelterbelt structure refers to the spatial density and distribution patterns of tree trunks, branches, and leaves. It is determined by a combination of multiple factors, including species composition, stand density, vertical stratification, diameter at breast height, tree height, and age [5]. Shelterbelt structural descriptors are categorized into two distinct groups: external structural characteristics (including width, height and cross-sectional geometry), and internal structural characteristics (comprising optical porosity (Op), surface area density (Scd) and volumetric density (Vcd)) [6,7]. Notably, porosity emerges as a holistic parameter that synthetically represents the integrated effects of all other structural components. Porosity is fundamentally defined as the ratio of void space to the space occupied by plant trunks, branches, and leaves [8,9,10]. However, the direct measurement of true porosity remains methodologically challenging, leading to the widespread adoption of optical porosity as a practical representation in most studies [11,12]. For multi-row shelterbelts, the internal three-dimensional architecture cannot be adequately characterized by two-dimensional approaches. To describe the three-dimensional structural characteristics of shelterbelts more scientifically, several researchers have suggested certain parameters, such as stumpage porosity, density, and biomass volume. However, the acquisition of these parameter data requires destructive sampling, making it a complicated process.

In the early stage, shelterbelt structure research was mainly carried out based on ground experiments and statistical methods. Subsequently, remote sensing technology was gradually applied to the monitoring of farmland shelterbelts. However, the inversion of some shelterbelt structure parameters was not mature and complete. As an active remote sensing technology, Light Detection and Ranging (LiDAR) originated in the 1960s and gradually entered China after 2000. Terrestrial laser scanning (TLS), a type of LiDAR, possesses high angular and distance resolution capabilities, enabling it to accurately capture object information. This provides strong technical support for shelterbelt structure investigation [13,14]. These emerging developments are revolutionizing our capacity to observe and monitor structural and functional dynamics in trees and forest ecosystems [15,16,17,18].

The shelterbelt structure typically exhibits pronounced spatial heterogeneity. To better quantify this structural variability, researchers have developed the concept of stratified porosity, which accurately characterizes the vertical distribution of porosity within shelterbelts [10]. The structural variation among individual trees leads to diverse stand configurations. Consequently, investigating single-tree architecture forms the foundation for the precision management of shelterbelt systems [19]. The spatial distribution patterns of branches and leaves, along with their biomass composition and allocation structure, constitute critical components in the study of whole-tree architecture. Separating tree’s branches and leaves is often necessary to obtain detailed information about their components from the point cloud data. At present, the methods for the separation of leaf–wood mainly include geometric and radiation intensity methods [20,21]. In recent years, network diagram methods have been increasingly applied to unsupervised leaf–wood separation, which uses tree points as a connected topological network to separate wood or larger branches from leaves through the Dijkstra algorithm [22]. The LeWoS algorithm employs a recursive point cloud segmentation and regularization process to facilitate the separation of leaf–wood. A key advantage of this algorithm is its ability to operate automatically by merely configuring the feature threshold parameter, “Nz_thres”, which is applicable across various types of LiDAR data [23]. The TreeQSM algorithm is utilized to derive parameters of tree branches, with optimization accomplished by adjusting the sensitivity of the specific input parameter “PathDiam” [24]. The 3D alpha-shape algorithm is mainly suitable for calculating the volume and surface area parameters of objects in a disordered point cloud. By creating a sphere with a radius of “α” around three points in the point set and identifying the boundary points, the surface of the entire point cloud is determined. This approach resolves the issue of quantitatively analyzing the components of leaves [25]. However, for wide shelterbelts composed of multi-row trees, further research into the effective quantification of their structures is required.

In this study, three typical shelterbelts in the desert oasis of Ulan Buh were studied. These include Populus alba var. pyramidalis Bunge, Populus popular’s Zhonghua, Populus nigra var. thevestina (Dode) Bean. The point cloud data of the shelterbelts were obtained by LiDAR, and the structural characteristics of the shelterbelts were obtained by combining the LeWoS leaf–wood separation algorithm, the TreeQSM algorithm, and the 3D alpha-shape algorithm. The main objectives of this research were (1) to explore the tree branch and leaf structure within the shelterbelt, (2) to analyze the spatial distribution characteristics of the shelterbelt structure, and (3) to develop quantitative structural parameters for shelterbelt characterization. The results of this study provide a basis for the structural optimization and management of farmland shelterbelts in the Ulan Buh desert oasis.

2. Materials and Methods

2.1. Study Area

The study plot is located in the Desert Forestry Experimental Center of the Chinese Academy of Forestry in Dengkou County, Inner Mongolia, China. The geographical coordinates are 40°17′–40°29′ N, 106°35′–106°59′ E (Figure 1). Strong winds mainly occur in spring, accounting for 47.5% of the year, and cause frequent sandstorms and serious surface erosion in the area. The area is dominated by artificial vegetation, mainly farmland shelterbelts and crops. The main tree species of the farmland shelterbelts are P. alba var. pyramidalis, P. nigra var. thevestina, P. popular’s, Elaeagnus angustifolia Linn, Ulmus pumila L, and Salix matsudana Koidz, among others. Notably, poplars collectively account for over 90% of the tree composition [26,27,28,29].

2.2. Methods

2.2.1. Field Survey

From May to June 2021, three shelterbelts with different configurations were selected, including three typical species: P. alba var. pyramidalis, P. nigra var. thevestina and P. popular’s (

Table 1. Basic information on shelterbelts.

Species	Row	Age (a)	Spacing (m)	Quadrate (m)
Populus alba var. pyramidalis Bunge	5	28	4.5 × 5	20 × 25
Populus popular’s Zhonghua	4	35	4 × 5	20 × 20
Populus nigra var. thevestina (Dode) Bean	4	33	5 × 5	20 × 20

). The quadrate is set in the shelterbelt. This was examined for various parameters. The obtained parameters included tree height, diameter at breast height (DBH), crown width, and under-branch height.

2.2.2. LiDAR Data Acquisition

The shelterbelt point cloud data were acquired using a RIEGL VZ-400i terrestrial laser scanner (TLS) (CHINTERGEO, Beijing, China) featuring a 360° horizontal and 100° vertical scanning range, with a maximum measurement distance of 300 m and a scanning speed of 1 million points per second. To mitigate occlusion effects from dense vegetation, a multi-station scanning approach was employed as shown in Figure 2. The initial scanning station was established at the upper-left corner of the shelterbelt, followed by subsequent stations positioned at 8 m intervals (corresponding to twice the mean tree spacing), with each station scanned for 5 min. Four reflective targets are strategically positioned alongside each other during the scanning process to facilitate point cloud data stitching.

2.2.3. Measurement of Structural Parameters

Fifteen trees from the three shelterbelts were selected for the accuracy assessment of the TLS parameter extraction and algorithm tuning. Five trees were selected for each tree species. Branch angles were measured using a protractor below 0.20 H (where H represents tree height), with photographic documentation for verification. After falling, the trunk was separated from the branches and leaves, and their respective parameters were measured. The surface area and volume of branches and trunks were calculated using cylindrical formulas. Leaf surface area was measured with a CI-202 leaf area scanner (LI-COR Biosciences, Beijing, China), while leaf volume was determined through the water displacement method. Branch diameters were measured using vernier calipers, while branch lengths were determined with measuring tapes.

2.3. Data Processing

2.3.1. Shelterbelt Point Cloud Data Pretreatment

Point cloud data pretreatment mainly includes stitching, cropping, denoising, thinning, filtering, and normalization. The RiSCANPro software (http://www.ilidar.com/, accessed on 12 August 2021), compatible with the TLS, was utilized for point cloud data stitching and cropping. Point cloud data denoising and ground point separation were carried out in CloudCompare v2.12 software (https://github.com/cloudcompare/cloudcompare, accessed on 1 May 2022). The Statistical Outlier Removal algorithm was employed for noise removal. The Cloth Simulation Filtering method was used to eliminate ground points [30].

2.3.2. Single Tree Segmentation and Parameter Extraction

For the accurate segmentation of single trees, the manual segmentation of point cloud data was performed. The parameters of tree height, DBH, crown width and under-branch height were extracted by Lidar360 v6.0 software (https://www.lidar360.com/, accessed on 7 May 2022). Due to the different modeling principles of wood and leaves, the estimated structural parameters of the whole tree model are high, and so it is necessary to separate wood and leaves. This study combines the LeWoS wood–leaf separation algorithm (https://github.com/DiWangLiDAR/LeWoS, accessed on 1 June 2022), the TreeQSM 2.4 algorithm of branch modeling (https://github.com/InverseTampere/TreeQSM, accessed on 20 June 2022), which is a leaf modeling algorithm based on the 3D alpha-shape for tree branches and leaf structure parameters.

2.3.3. Algorithm Optimization

Firstly, we set different Nz_thres and PathDiam parameter combinations using LeWoS and TreeQSM algorithms. Early researchers made a lot of attempts to define a series of values for each input parameter and select a group of optimal parameters. The typical value of Nz_thres falls within the range of 0.1 to 0.2. The value range for PathDiam is from 0.02 to 0.09 [31,32]. The leaf structure parameters were extracted using the 3D alpha-shape algorithm. A sphere has a radius of “α”, which is a value set between 0.0048 and 0.0053 [33,34]. In this study, nine destructive samplings were used for sensitivity analysis. When this involved three parameters, six destructive samplings were tested. The optimal parameters were used as input parameters for other samples. The optimal values of Nz_thres, PathDiam and α parameters are set as follows: The optimal parameter combination of P. alba var. pyramidalis is 0.145, 0.04 and 0.005, that of P. popular is 0.14, 0.03 and 0.0051, and that of the P. nigra var. thevestina is 0.145, 0.03 and 0.005.

2.3.4. Tree Branching Feature and Leaf Parameters Acquisition

In order to further analyze the structural characteristics of the shelterbelt, this study used the optimized TreeQSM algorithm to extract the first- to fourth-level branches of trees. The length, diameter, angle, ratio of branch diameter (RBD) and bifurcation ratio (BR) of branches were extracted. The calculation formula for the RBD is as follows:

$$RB{D}_i=D_{i+1}/D_i$$

(1)

where RBD_i denotes the ratio of branch diameter at the i level, D_i+1 denotes the diameter of the branch at the i + 1 level, and D_i denotes the diameter of the branch at the i level. The BR calculation formula is as follows:

$$B{R}_i=N_{i+1}/N_i$$

(2)

where BR_i denotes the bifurcation ratio at the i level, N_i+1 denotes the number of the branch at the i + 1 level, and N_i denotes the number of the branch at the i level.

The 3D alpha-shape algorithm was employed to extract leaf surface area and volume structural parameters.

2.3.5. Shelterbelt Parameter Calculation

Optical porosity (Op) is the ratio of the vertical porosity S₀ of the shelterbelt to its total area S [10]. In this study, a digital camera is used to capture the whole shelterbelt photo, and Adobe Photoshop 2022 (Adobe Systems, San Jose, CA, USA, https://www.adobe.com/products/photoshop.html, accessed on 5 March 2023) is used for monochrome processing to distinguish the shelterbelt part from the transparent part, and to calculate the proportion of pixels. The Op is determined by the Formula (3):

$$Op={\displaystyle \frac{S_0}{S}}$$

(3)

Volume porosity (Vp) refers to the ratio of the pore volume to the total volume of the shelterbelt [35]. The expression is shown as (4), where V₀ represents the volume occupied by leaves and branches, and V represents the total volume of the shelterbelt (its length multiplied by its width and height) (Figure 3a).

$$Vp=1-{\displaystyle \frac{V_0}{V}}$$

(4)

The variable Vp was standardized in the present study. The existing literature has presented various methods for indicator standardization, including the min–max scaling method, the indication method, and the Z-score transformation method. The min–max scaling method retains the relationships between data values and applies to various distributions [36]. The min–max scaling formula is expressed as follows:

$$S_{ij}={\displaystyle \frac{x_{ij}-min(x_i)}{max(x_i)-min(x_i)}}$$

(5)

where S_ij and x_ij represent, respectively, the standardized and original value for the indicator i (i = 1, 2,…, n), referring to the sample case j (j = 1, 2,…, m), and max(x_i) and min(x_i) denote the maximum and minimum value among all the samples for the indicator i, respectively.

Stratification porosity (Sp) is used to define the distribution of pores of a shelterbelt, calculated vertically, i.e., cutting a shelterbelt into many slices in the vertical direction and computing the stratification porosity. This study employed a volumetric stratification porosity (Vsp) measurement method with 1 m vertical intervals [37].

Scd refers to the surface area of the tree branches and trunks in each cube unit of the shelterbelt space. Vcd refers to the volume of the tree trunks and branches in each cube unit of the shelterbelt space [38]. In this study, the calculation formula of surface area density and volume density of shelterbelts is as follows:

$$S_{cd}={\displaystyle \frac{1}{△x△y△z}}{\displaystyle {\sum }_{i=1}^3{S}_i}$$

(6)

$$V_{cd}={\displaystyle \frac{1}{△x△y△z}}{\displaystyle {\sum }_{i=1}^3{V}_i}$$

(7)

where Δx, Δy, and Δz are cell resolutions in the direction of the shelterbelt width (x), length (y), and height (z) (Figure 3b). Δx is defined by the rows spacing within the shelterbelt; Δy is defined by the spacing of trees within the shelterbelt. i = 1, 2 and 3, respectively, represent tree trunks, branches, and leaves in each cell. S_i and V_i represent the total surface area and total volume of each component of the tree in a particular cell, respectively.

2.3.6. Accuracy Evaluation

Bias (deviation), RMSE (root mean square of error), rBias (relative deviation), and rRMSE (relative root mean square of error) were used to evaluate the accuracy of the obtained DBH, tree height, under-branch height, crown width, branch structure, and leaf structure.

$$Bias={\displaystyle \frac{1}{n}}{\displaystyle {\sum }_{i=1}^n(y_i-{\widehat{y}}_i})$$

(8)

$$rBias={\displaystyle \frac{Bias}{\overline{{\widehat{y}}_i}}}\times 100\%$$

(9)

$$RMSE=\sqrt{{\displaystyle \frac{1}{n}}{\displaystyle {\sum }_{i=1}^n{(y_i-{\widehat{y}}_i)}^2}}$$

(10)

$$rRMSE={\displaystyle \frac{RMSE}{{\overline{\widehat{y}}}_i}}\times 100\%$$

(11)

where $y_i$ is the calculated value and ${\widehat{y}}_i$ is the measured value.

3. Results

3.1. Single Tree Structure

3.1.1. Integral Structure of Single Tree

We extracted the data on the tree height, DBH, height under branches, and crown width of individual trees in the quadrat (

Table 2. Integral structure of single tree.

Species	Height (m)	DBH (m)	Crown Width (m)	Under-Branch Height (m)	Surface Area (m2)	Volume (m3)
P. alba var. pyramidalis	20.113	0.331	3.232	2.171	113.991	1.411
P. popular’s	16.365	0.242	4.673	0.912	88.014	1.049
P. nigra var. thevestina	20.241	0.290	2.911	1.384	87.219	1.068

). P. nigra var. thevestina had the highest average tree height of 20.241 m. P. alba var. pyramidalis had the largest average DBH, and P. popular’s had the smallest. The order of the tree crown widths among tree species was P. popular’s > P. alba var. pyramidalis > P. nigra var. thevestina. Based on the calculations for individual trees in the sampled area, P. alba var. pyramidalis had the largest surface area and volume.

3.1.2. Tree Branch Structure

Branch architecture distribution analysis relied exclusively on TreeQSM-derived metrics. Figure 4a shows that the diameter of each branch decreases with an increase in branch level. On the whole, P. alba var. pyramidalis exhibited the largest mean branch diameters across all orders: 0.113 m (first), 0.061 m (second), 0.029 m (third), and 0.011 m (fourth). In contrast, P. nigra var. thevestina showed the smallest mean branch diameters. The length distribution patterns are consistent with those of branch diameters (Figure 4b), with P. alba var. pyramidalis demonstrating the most extensive length variation in the first branches (0.226–10.457 m).

In terms of average branch angle size, the order of the tree species was P. alba var. pyramidalis (51.131°) > P. popular’s (48.209°) > P. nigra var. thevestina (47.277°). The branch angle of all trees increased gradually from the first to the fourth grade, and that of P. popular’s demonstrated the greatest increase (Δ9.733°, from 42.341° to 52.074°) compared to P. nigra var. thevestina, which showed minimal variation (Δ1.701°, from 46.591° to 48.292°). The patterns of variation in both branch surface area and tree volume with branch level were consistent (Figure 5), that is, the branch surface area and volume decreased gradually from the first to the fourth grade. The mean values of both branch surface area and volume across all orders consistently exhibited the following species ranking: P. alba var. pyramidalis > P. popular’s > P. nigra var. thevestina.

With an increase in the branch level, the number of branches increased first and then decreased. The average BR values of the three species were in the order P. alba var. pyramidalis > P. nigra var. thevestina > P. popular’s. The BR_2:1 was highest in P. nigra var. thevestina (3.279), while the BR_3:2 was highest in P. alba var. Pyramidalis (0.981). P. popular’s had the highestRBD at all levels.

3.1.3. Tree Leaf Structure

As shown in Figure 6, P. nigra var. thevestina had the widest variation in leaf surface area (LS) (6.871–56.954 m²) and volume (LV) (0.006–0.055 m³), while P. alba var. pyramidalis demonstrated more concentrated distributions (18.443–52.421 m² and 0.010–0.032 m³, respectively). The average leaf surface area and volume of the trees were in the order of P. popular’s > P. alba var. pyramidalis > P. nigra var. thevestina.

3.2. Shelterbelt Structure

3.2.1. Integral Structure of Shelterbelt

Table 3. Two-dimensional structural parameters of the shelterbelt.

Species	Height (m)	Width (m)	Length (m)	Op
P. alba var. pyramidalis	20.113	25.120	450.340	0.449
P. popular’s	16.365	20.041	196.511	0.424
P. nigra var. thevestina	20.241	18.114	390.328	0.492

lists the two-dimensional structural parameters of the shelterbelts, among which P. nigra var. thevestina had the lowest Op (0.424), followed by P. alba var. pyramidalis. P. nigra var. thevestina had the highest Op. On comparing the density of shelterbelts (

Table 4. Three-dimensional structure parameters of the shelterbelt.

Species	Surface Area (m2)	Volume (m3)	Scd (m2·m−3)	Vcd (m3·m−3)	Vp
P. alba var. pyramidalis	44,225.921	445.254	0.205	0.002	0.532
P. popular’s	15,789.813	154.843	0.252	0.003	0.490
P. nigra var. thevestina	33,068.122	314.342	0.235	0.002	0.528

), the Scd of P. popular’s was the largest (0.252 m²·m⁻³), while the Scd of P. alba var. pyramidalis was the smallest (0.205 m²·m⁻³). Vcd changed consistently with Scd. However, changes between volume porosity (Vp) and Op were inconsistent. The P. alba var. pyramidalis shelterbelt had the highest Vp.

3.2.2. Spatial Distributions of Trunk, Branch, and Leaf in the Shelterbelts

The vertical distribution of the volume and surface area of tree trunks, branches, and leaves in the three shelterbelts is presented in Figure 7 and Figure 8. The volume of the trunk decreases gradually with an increase in height, and the decline rate slows down gradually as height increases. The volume of branches and leaves showed a similar inverted U-shaped distribution with height.

Branch volume was generally higher at medium heights, and the value decreased sharply as height increased. The height segments with maximum branch volumes for the three shelterbelts were P. alba var. pyramidalis (0.40 H–0.50 H), P. popular’s (0.55 H–0.65 H), and P. nigra var. thevestina (0.40 H–0.60 H). For the P. popular’s shelterbelt, branch volume decreased the fastest after reaching the maximum value. The leaves were mostly concentrated in the middle and lower part of the shelterbelt (0.44 H–0.60 H). The leaves of P. alba var. pyramidalis and P. nigra var. thevestina were mostly concentrated at the medium and low heights of the shelterbelt (0.44 H–0.60 H). P. popular’s shelterbelt leaves were mostly concentrated at upper heights (0.75 H–0.82 H).

In the horizontal direction, the overall volume proportion of each component was in the order of trunk > branches > leaves. The proportions of trunks and branches in the P. alba var. pyramidalis shelterbelt were the largest among the three species, at 68.989% and 29.641%, respectively. The proportion of leaves was the largest in the P. popular’s shelterbelt (7.971%). The surface area contribution of each component was in the order leaf > branch > trunk. The P. alba var. pyramidalis shelterbelt had the largest proportions of trunk and branch surface area, at 13.203% and 36.725%, respectively. The P. popular’s shelterbelt had the largest proportion of leaf surface area (79.691%).

3.2.3. Spatial Distributions of Scd and Vcd in the Shelterbelts

As shown in Figure 9, in the vertical direction, the maximum Vcd and Scd were mostly located at medium and low heights of the shelterbelt (0.33 H–0.60 H). The Scd of the P. alba var. pyramidalis shelterbelt was greatest in the range 0.40 H to 0.60 H, that of the P. popular’s shelterbelt was greatest in the range 0.33 H to 0.60 H, and that of the P. nigra var. thevestina shelterbelt was greatest in the range of 0.40 H to 0.60 H (Figure 9a).

In the horizontal direction, the average Scd of the five-row P. alba var. pyramidalis shelterbelt, running from the first row to the fifth row, showed a trend of first decreasing and then increasing. Similarly, the maximum Vcd appeared in the first row (0.004 m³·m⁻³) and the fifth row (0.004 m³·m⁻³) on both sides of the shelterbelt. The minimum Vcd appeared in the third row (0.002 m³·m⁻³), with an obvious edge effect. The average Scd (0.397 m²·m⁻³–0.537 m²·m⁻³) and Vcd (0.001 m³·m⁻³–0.004 m³·m⁻³) of the P. popular’s shelterbelt increased gradually from the first row to the fourth row, and the difference in growth between the two marginal rows was large. The average Scd of the four-row P. nigra var. thevestina shelterbelt was higher in the outer two rows (0.244 m²·m⁻³–0.324 m²·m⁻³) than in the middle two rows (0.173 m²·m⁻³–0.213 m²·m⁻³), while the Vcd increased gradually.

3.2.4. Spatial Distributions of Op and Vp in the Shelterbelts

The results demonstrate heterogeneity in both vertical and horizontal distributions of Op (Figure 10a) and Vp across the shelterbelts. Along the horizontal dimension, variations in Op along the belt length were insignificant at heights below 0.20 H. Within the 0.20 H–0.80 H range, porosity increased markedly in central sections: P. alba var. pyramidalis (6.20 H–9.00 H), P. popular’s (0.50 H–2.00 H) and P. nigra var. thevestina (6.55 H–10.00 H). Above 0.80 H, P. alba var. pyramidalis exhibited pronounced porosity increases in both central (1.85 H–10.93 H) and distal (11.30 H–13.87 H) sections, whereas P. nigra var. thevestina displayed gradual increases along its length and P. popular’s maintained relatively stable values.

The spatial variation in Op in shelterbelts was predominantly vertical. With increasing height, Op initially decreased then increased. Op exceeded 0.800 at z < 0.10 H and 0.700 at z > 0.90 H. Minimum values occurred at P. alba var. pyramidalis (0.181 at 0.67 H), P. popular’s (0.117 at 0.59 H), and P. nigra var. thevestina (0.274 at 0.43 H), all located at medium heights.

Variations in shelterbelt Vp in x, y, and z are presented in Figure 10. Across the shelterbelt width (x, Figure 10b), P. alba var. pyramidalis exhibited lower values in edge rows (0.00 H–0.30 H and 0.65 H–0.90 H). P. popular’s showed reduced levels at 0.30 H–0.50 H and 0.85 H–0.90 H, whereas P. nigra var. thevestina demonstrated minimal measurements in 0–0.15 H zones but higher readings at 0.20 H–0.45 H. Along the shelterbelt length (y, Figure 10c), P. alba var. pyramidalis displayed elevated porosity in central sections (6.01 H–7.98 H and 10.93 H–13.02 H), P. popular’s maintained depressed edge levels (±1.50 H), and P. nigra var. thevestina showed relatively stable distribution.

The change in vertical distributions of Vp was significantly different from that of Op. The value of P. alba var. pyramidalis ranged from 0.062 to 0.714 (minimum 0.45 H), that of P. popular’s ranged from 0.001 to 0.662 (minimum 0.50 H), and that of P. nigra var. thevestina ranged from 0.013 to 0.679 (minimum 0.25 H), with the lowest values seen at mid–low heights.

4. Discussion

4.1. Feasibility Analysis of the Methodology

The proposed LeWoS-TreeQSM-3D alpha-shape integrated method demonstrates outstanding performance in quantifying structural parameters of shelterbelts. Multispecies validation (

Table 5. Integral parameter extraction accuracy.

Structural Parameters	Species	Evaluation Indicators
		Bias (m)	rBias (%)	RMSE (m)	rRMSE (%)
Tree Height	P. alba var. pyramidalis	−1.177	−5.874	1.498	7.471
DBH		−0.011	−3.401	0.014	4.360
Crown width		−0.077	−2.544	0.222	7.314
Under-branch height		0.053	2.621	0.098	4.880
Tree volume		0.005	1.172	0.078	16.517
Tree surface area		−0.796	8.190	1.052	10.830
Tree Height	P. popular’s	−0.455	−3.041	0.775	5.183
DBH		−0.021	−7.978	0.030	11.494
Crown width		−0.239	−4.946	0.4005	8.306
Under-branch height		−0.007	−0.732	0.057	5.416
Tree volume		0.002	0.697	0.004	1.802
Tree surface area		−0.078	1.393	0.127	2.278
Tree Height	P. nigra var. thevestina	−0.755	−3.669	1.005	4.883
DBH		−0.007	−2.260	0.222	7.019
Crown width		−0.051	−1.628	0.136	4.327
Under-branch height		0.140	8.779	0.183	11.483
Tree volume		0.009	2.750	0.015	4.676
Tree surface area		−0.009	0.152	0.088	1.484

) confirms the method’s high reliability in extracting key parameters, including tree height (with a minimal bias of −0.455 m for P. popular’s), DBH (showing the highest accuracy among whole-tree parameters), and tree volume and surface area (all exceeding 91.810% relative accuracy). These findings align with previous studies by Jin et al. [39], reporting high extraction accuracy for tree volumes. Hildebrandt et al. [40] carried out a three-dimensional reconstruction of trees, divided the reconstructed model into several parts according to a certain length, and calculated the volume of each part, and they observed an error within 1% between the obtained result and the measured value.

The further validation of branch parameters (

Table 6. Branch parameter extraction accuracy.

Structural Parameters	Species	Evaluation Indicators
		Bias (m)	rBias (%)	RMSE (m)	rRMSE (%)
Branch length	P. alba var. pyramidalis	0.140	5.812	0.206	8.550
Branch diameter		0.002	2.614	0.001	1.832
Branch angle		−1.815	−4.003	2.687	5.906
Branch volume		0.006	4.122	0.008	5.554
Branch surface area		0.199	1.084	1.153	6.272
Leaf volume		0.001	2.862	0.002	15.246
Leaf surface area		0.820	2.941	2.967	10.642
Branch length	P. popular’s	0.057	4.316	0.102	7.575
Branch diameter		0.001	1.843	0.002	2.087
Branch angle		−0.656	−1.500	2.114	4.707
Branch volume		0.005	4.165	0.008	7.150
Branch surface area		0.035	0.225	0.692	4.442
Leaf volume		−0.001	−4.692	0.005	26.325
Leaf surface area		1.597	2.989	3.917	9.016
Branch length	P. nigra var. thevestina	0.148	9.226	0.225	15.457
Branch diameter		0.003	2.059	0.006	4.814
Branch angle		1.067	2.612	1.926	4.715
Branch volume		0.003	2.513	0.007	6.080
Branch surface area		−0.393	−2.765	1.038	7.294
Leaf volume		−0.001	−0.946	0.001	10.911
Leaf surface area		−2.629	−23.133	3.306	29.087

) demonstrated that P. popular’s exhibited the highest extraction accuracy for branch length and diameter, with mean rRMSE values of 7.575% and 2.087%, respectively. In contrast, P. nigra var. thevestina showed the lowest relative accuracy in branch length (90.774%). This observation aligns with Lau et al. [41], who achieved 95.0% reconstruction accuracy for branches exceeding 30 cm in diameter in their sample of 279 modeled branches. The branch angle error analysis between measurements and extracted values revealed that P. alba var. pyramidalis exhibited the maximum rBias (−4.003%) (Table 6). The lowest relative accuracy of the leaf surface area was 76.870%. This may result from overlapping adjacent canopies leading to incomplete crown segmentation, ultimately affecting the extraction accuracy of individual crown and leaf surface area parameters. As demonstrated by Béland et al. [42], three primary factors influence leaf parameter estimation in TLS measurements: (1) wood–leaf point cloud separation, (2) leaf orientation effects, and (3) mutual occlusion between point clouds. Collectively, these results confirm the method’s high accuracy in measuring tree height, DBH, and branch architecture parameters, demonstrating its strong applicability for the three-dimensional structural quantification of shelterbelts.

4.2. The Significance of Stratified Porosity

Shelterbelt structures exhibit the heterogeneous spatial distribution patterns of their components (branches, trunks, leaves, etc.), demonstrating the limitations of integral parameter characterization. Traditional optical porosity utilizes a single value that cannot capture three-dimensional pore architecture heterogeneity. Notably, shelterbelts with distinct configurations may show identical optical porosity values but differ significantly in ventilation performance [43]. Analysis of the spatial distribution patterns of Op and Vp further reveals pronounced variations along the width dimension (x-axis), which cannot be neglected. Empirical studies demonstrate that the product of width and area density (W·Ad) serves as the dominant predictor of windbreak efficiency [7]. The non-uniform distribution of pores makes it difficult to accurately characterize micro-environmental variations, such as turbulence in wind shadow zones, using optical porosity [44,45].

Stratified porosity analysis reveals distinct vertical structural patterns in shelterbelts, typically showing higher porosity in trunk layers and lower values in lower-to-middle canopies. Wuyts et al. [46] and Lalic et al. [47] demonstrated that sparse trunk layers minimize airflow resistance, sustaining higher wind speeds, whereas dense canopy regions enhance drag forces and reduce wind velocity. Research has demonstrated that shelterbelts with non-uniform density distributions exhibit superior windbreak efficiency in peak-density zones compared to homogeneous structures, highlighting the critical role of internal structural heterogeneity in defensive performance [48,49]. Stratified porosity analysis is scientifically valuable for evaluating farmland shelterbelt ecological functions. It precisely characterizes spatial structural heterogeneity and its mechanistic effects on key processes like windbreak and light penetration.

4.3. Shelterbelt Planning and Management Guidelines

Stratified porosity analysis effectively characterizes shelterbelt pore structure spatiotemporal variations, as demonstrated in this study. The edge rows exhibit significantly reduced porosity due to dense foliage, consequently decreasing light’s availability for adjacent crops. Previous research has established that shelterbelt porosity directly influences inter-row light transmittance [50]. Specifically, photosynthetically active radiation (PAR) in sheltered zones is markedly lower than that in open areas. Field studies have reported 22%–27% yield reductions in edge-row corn fields adjacent to poplar shelterbelts due to shading effects [51]. Computer simulations revealed that shelterbelts in New Zealand kiwifruit orchards can reduce canopy irradiance by up to 40%, with shading intensity showing a significant positive correlation with shelterbelt leaf area index [52]. These findings suggest that pruning interventions should be implemented when edge porosity decreases to optimize both the shelterbelt structure.

The degradation of trees within shelterbelts can trigger a cascade of ecological consequences. As demonstrated by Holzwarth et al. [53] and Buettel et al. [54], tree degradation significantly compromises wood structural integrity, increasing the risks of stem breakage and wind throw, leading to gap formation. These forest gaps substantially alter microclimatic conditions, including light intensity, temperature, and soil moisture. Empirical evidence shows that ecological restoration of degraded farmland shelterbelts can effectively reduce porosity and attenuate wind velocity [55]. Consequently, when monitoring detects abnormal increases in internal shelterbelt porosity, immediate replanting measures are necessary to maintain structural continuity.

The disruption of shelterbelt continuity induces pronounced wind erosion effects. When gaps or broken rows occur between shelterbelt segments, these discontinuous areas create wind acceleration channels, leading to airflow convergence and enhanced turbulence at breakpoints. Studies demonstrate that these structural discontinuities significantly increase local wind energy transport efficiency, causing the abrupt acceleration of field wind speeds and consequently exacerbating wind erosion risks [55,56]. The study employed wooden rectangular barrier models to systematically measure three-dimensional velocity fields across both horizontal and vertical planes, revealing airflow acceleration at gap locations [57]. These findings underscore that restoring continuity through the replacement of degraded tree species is critical for maintaining windbreak ecological functions.

5. Conclusions

This study utilized LiDAR technology to analyze three typical shelterbelts in the Ulan Buh Desert oasis. A point cloud-based method integrating the LeWoS leaf–wood separation algorithm, tree QSM algorithm, and 3D alpha-shape algorithm was developed to enhance the efficiency and accuracy of structural parameter extraction. The results indicate that extraction relative accuracy reached 90.774% for all parameters, except leaf surface area. In branch architecture analysis, P. alba var. pyramidalis exhibited significantly greater branch length, diameter, and branching ratio compared to the other two species. Spatial analysis revealed distinct horizontal edge effects, while vertical profiles indicated higher density distributions in mid-to-lower canopy layers (0.33 H–0.60 H), contradicting the conventional homogeneous canopy assumption. A quantitative assessment of shelterbelt density and porosity led to the proposed use of stratified structural parameters to characterize porosity features, enabling the development of optimized management strategies for regional farmland shelterbelts. Future research will employ computer simulations to analyze how stratified porosity governs windbreak and shading effects, ultimately helping to develop an intelligent system for real-time structural optimization.