A method for intelligent coupling analysis and modeling of forest structure and absorption of photosynthetically active radiation

Abstract

The application discloses a forest structure and intelligent coupling analysis and modeling method for absorbing photosynthetically active radiation, and belongs to the technical field of forest ecology remote sensing, structure analysis and computer simulation. The method focuses on fusing airborne and handheld laser radar information, accurately inverting forest structure indexes, covering single-tree level attributes, neighborhood spatial structure indexes, gap structure, stand density SD and slope SL; three-dimensional radiation transmission simulation is carried out; through the construction of a single-tree three-dimensional model library and a digital sample plot, high-precision three-dimensional forest reconstruction and hierarchical APAR simulation are realized, and photosynthetic spatial heterogeneity parameters are extracted; Spearman correlation analysis is carried out based on the structure and APAR parameters, and PCA dimension reduction and KMeans clustering are combined to identify the dominant structure mode and spatial type; based on the clustering result, a typed structure-APAR coupling integrated learning model is constructed, and the regulation mechanism of single-tree attributes and neighborhood structure on APAR is revealed.

Description

Technical Field

[0001] This invention belongs to the field of forest ecological remote sensing, structural analysis and computer simulation technology, and relates to a method for intelligent coupling analysis and modeling of forest structure and photosynthetically active radiation absorption. Background Technology

[0002] Forests are playing an increasingly important role in carbon storage, ecosystem stability, and climate regulation. Plants complete photosynthesis through photosynthetically active radiation (PAR), and the portion of PAR absorbed by vegetation (APAR) is a core factor affecting carbon assimilation efficiency. A complex and close coupling relationship exists between forest structural characteristics (such as tree height, canopy width, gaps, and spatial patterns) and the spatial distribution of APAR. Although existing studies have explored the regulatory mechanisms of structure on light energy absorption, a systematic and refined method for characterizing the spatial distribution of photosynthetic radiation in natural forests with complex structures and significant topographic relief is still lacking. Existing methods mostly rely on small-scale, local ground observations (such as photosynthetic radiometer measurements), and the extraction accuracy of structural parameters is low and the systematization is insufficient, making it difficult to support high-precision, unified structure-radiation coupling modeling. Furthermore, traditional statistical or two-dimensional approximate models cannot accurately characterize the complex relationship between structural heterogeneity and light distribution patterns. Summary of the Invention

[0003] This invention addresses the problems of existing technologies by proposing an intelligent coupling analysis and modeling method for forest structure and APAR (Absorbed Photosynthetically Active Radiation). It fully utilizes the advantages of multi-source data fusion, combines 3D modeling, radiative transfer simulation, and intelligent algorithms, and systematically and meticulously analyzes the coupling mechanism between forest structure and APAR, providing efficient and reliable technical support for quantitative and mechanistic research on forest ecosystems.

[0004] A method for intelligent coupling analysis and modeling of forest structure and absorption of photosynthetically active radiation includes the following steps:

[0005] Step 1: Focus on integrating airborne and handheld lidar information to accurately retrieve forest structure indicators, covering individual tree attributes (diameter at breast height (DBH), tree height (H), crown width (CW), crown area (CA), and crown volume (CV), neighborhood spatial structure indicators (spatial density (DE), size ratio (U), forest layer index (S), openness (OP), competition index (Cl), nearness to nature (NNLI), and dominance (LCDI), forest gap structure (density (GD) and average area (GAM), stand density (SD), and slope (SL).

[0006] Step 2: Conduct three-dimensional radiative transfer simulation. By constructing a single-tree three-dimensional model library and digital sample plots, high-precision three-dimensional forest reconstruction and layered APAR simulation are achieved, and spatial heterogeneity parameters of absorbed photosynthetic radiation are extracted.

[0007] Step 3: Perform Spearman correlation analysis based on structure and APAR parameters, and combine PCA dimensionality reduction and KMeans clustering to identify dominant structural patterns and spatial types.

[0008] Step 4: Construct a fractal structure-APAR coupled ensemble learning model based on the clustering results to reveal the regulatory mechanism of individual trees and neighborhood structures on APAR.

[0009] The advantages of this invention are: a structure-radiative intelligent coupling analysis and modeling method that integrates multi-source lidar point clouds, multi-source spectral data, 3D modeling, and three-dimensional radiative transfer. This method first achieves high-precision automatic extraction of structural parameters from the individual tree to the stand scale, including diameter at breast height (DBH), tree height, crown width, crown area, crown volume, and multi-dimensional structural indicators such as spatial density, vertical and horizontal structure, competition index, and forest gap. Based on complete point clouds and structural attributes, it automatically matches the optimal three-dimensional tree model and reconstructs a digital forest plot, combining this with a three-dimensional radiative transfer model to finely simulate the distribution characteristics of APAR in vertical and horizontal spaces. The entire process supports large-sample, batch-based automatic modeling, effectively reducing reliance on manual intervention and traditional ground observations, and achieving an automatic closed loop from acquisition and processing to simulation analysis of structure-radiative information.

[0010] In the structure-APAR coupling modeling stage, this invention integrates intelligent algorithms such as Spearman correlation analysis, PCA dimensionality reduction, KMeans clustering, and Bagging ensemble learning to systematically reveal the regulatory mechanism of structural factors on APAR. The model interpretation results not only identify key parameters for structure-regulated APAR but also improve the model's prediction accuracy, robustness, and interpretability. This method demonstrates excellent adaptability and scalability in complex natural forest environments. Its high-precision structure and light energy response modeling capabilities not only provide an efficient and reliable tool for quantitative analysis of forest light environment and research on ecological mechanisms but also offer important technical support for refined forest management, possessing broad prospects for promotion and application. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. As shown in the figures:

[0012] Figure 1 This is the overall technical roadmap for the intelligent coupling analysis and modeling of forest structure and photosynthetically active radiation absorption of the present invention.

[0013] Figure 2This is a flowchart illustrating the construction of neighborhood spatial structure units and the calculation of spatial structure indicators in this invention.

[0014] Figure 3 The results of forest gap structure extraction in the experimental area of ​​this invention (0.1m resolution).

[0015] Figure 4(a) shows the results of digital plot reconstruction according to the present invention.

[0016] Figure 4(b) Simulation results of multi-time uplink radiative transfer of the present invention (seven time points).

[0017] Figure 5(a) shows the statistics and distribution of cumulative APAR (TAPAR) of different trees in the sample plot.

[0018] Figure 5(b) shows the cumulative APAR (TAPAR) at different tree heights of 3457 trees in all sample plots, which shows a trend of first increasing and then decreasing.

[0019] Figure 6 The results of the correlation analysis between the APAR stratification characteristic parameters and the structural parameters of a single tree are presented.

[0020] Figure 7(a) shows one of the results of the correlation analysis between APAR parameters and structural parameters at the individual tree and plot levels.

[0021] Figure 7(b) shows the second result of the correlation analysis between APAR parameters and structural parameters at the individual tree and plot levels.

[0022] Figure 7(c) shows the third result of the correlation analysis between APAR parameters and structural parameters at the individual tree and plot levels.

[0023] Figure 7(d) shows the fourth result of the correlation analysis between APAR parameters and structural parameters at the individual tree and plot levels.

[0024] Figure 8 Principal component loading plot (load distribution of 12 structural variables on the first two principal components (PC1, PC2)).

[0025] Figure 9 The distribution of KMeans clustering results in the principal component space (normalized mean of the three cluster structure variables).

[0026] Figure 10(a1) Performance evaluation of three structural subtypes of the Bagging model (spatial residual hexagonal aggregation thermogram).

[0027] Figure 10(a2) Performance evaluation of three structural subtypes of the Bagging model (spatial residual hexagonal aggregation thermogram).

[0028] Figure 10(a3) Performance evaluation of three structural subtypes of the Bagging model (spatial residual hexagonal aggregation thermogram).

[0029] Figure 10(b1) Performance evaluation of the three structural classification Bagging models (scatter plot of predicted values ​​and measured TAPAR values, showing the model fitting effect).

[0030] Figure 10(b2) Performance evaluation of the three structural classification Bagging models (scatter plot of predicted values ​​and measured TAPAR values ​​to show the model fitting effect).

[0031] Figure 10(b3) Performance evaluation of the three structural classification Bagging models (scatter plot of predicted values ​​and measured TAPAR values ​​to show the model fitting effect).

[0032] Figure 10(c1) Performance evaluation of the three structural classification Bagging models (model learning curves).

[0033] Figure 10(c2) Performance evaluation of the three structural classification Bagging models (model learning curves).

[0034] Figure 10(c3) Performance evaluation of the three structural classification Bagging models (model learning curves).

[0035] Figure 11 This is a feature importance analysis based on SHAP values ​​and a contribution pattern diagram for each structure clustering. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] Example 1: As Figure 1 , Figure 2 , Figure 3 Figure 4(a), Figure 4(b), Figure 5(a), Figure 5(b) Figure 6 Figures 7(a), 7(b), 7(c), and 7(d) Figure 8 , Figure 9 Figures 10(a1), 10(a2), 10(a3), 10(b1), 10(b2), 10(b3), 10(c1), 10(c2), and 10(c3) and Figure 11As shown, a smart coupling analysis and modeling method for forest structure and photosynthetically active radiation absorption (APAR) is proposed, overcoming the limitations of traditional methods such as single structural indicators, low accuracy of illumination simulation, and insufficient modeling interpretability. This method integrates lidar point clouds, multispectral imagery, and hyperspectral reflectivity parameters, combined with 3D reconstruction and radiative transfer models, to achieve quantitative analysis of the entire process from point cloud extraction to structural parameter extraction, 3D model construction, refined characterization of APAR spatial heterogeneity, and structural indicators and light energy response. This provides solid technical support for forest light environment modeling, structural optimization management, and ecological mechanism research.

[0038] A method for intelligent coupling analysis and modeling of forest structure and absorption of photosynthetically active radiation includes the following steps:

[0039] Step 1: Focus on integrating airborne and handheld lidar information to accurately retrieve forest structure indicators, covering individual tree attributes (diameter at breast height (DBH), tree height (H), crown width (CW), crown area (CA), and crown volume (CV), neighborhood spatial structure indicators (spatial density (DE), size ratio (U), forest layer index (S), openness (OP), competition index (Cl), nearness to nature (NNLI), and dominance (LCDI), forest gap structure (density (GD) and average area (GAM), stand density (SD), and slope (SL), etc.);

[0040] Step 2: Conduct three-dimensional radiative transfer simulation. By constructing a single-tree three-dimensional model library and digital sample plots, high-precision three-dimensional forest reconstruction and layered APAR simulation are achieved, and spatial heterogeneity parameters of absorbed photosynthetic radiation are extracted.

[0041] Step 3: Perform Spearman correlation analysis based on structure and APAR parameters, and combine PCA dimensionality reduction and KMeans clustering to identify dominant structural patterns and spatial types;

[0042] Step 4: Construct a fractal structure-APAR coupled ensemble learning model based on the clustering results to reveal the regulatory mechanism of individual trees and neighborhood structures on APAR.

[0043] like Figure 1 As shown, multi-source data is imported: airborne radar data, handheld radar data, airborne multispectral data, and ground hyperspectral data. Through data fusion and preprocessing, a unified analytical foundation is constructed.

[0044] The structural parameter extraction section includes two types of information: forest structural parameter extraction and sample plot structural parameter extraction. For the former, individual tree attribute parameters are extracted first, and the spatial structural parameters of individual trees are calculated by normalizing the spatial coordinates of trees and constructing forest spatial units. For the latter, the slope and gap parameters are extracted using the canopy height model and digital elevation model, and the stand density parameters are extracted based on the individual tree attribute parameters.

[0045] The extraction of APAR parameters at the individual tree level includes: First, digital plot reconstruction is performed. Based on individual tree branch reconstruction, LAI adjustment, and 3D leaf addition, a tree model library is built, and then the digital plot reconstruction is achieved through a model matching algorithm. Next, a three-dimensional radiative transfer simulation is performed. The constructed 3D scene is imported into the model, spectral parameters are imported into the model, and finally, simulation environment parameters are set. The simulation is then run and the four final individual tree vertical APAR parameters (Gini-APAR, CV-APAR, Rel-MeanLayer, and Rel-MaxLayer) are extracted.

[0046] Subsequently, the extracted individual tree attribute parameters, individual tree spatial structure parameters, individual tree APAR accumulation parameters, and individual tree stratified accumulation parameters were analyzed for correlation. The extracted slope, forest gap parameters, and stand density parameters were also analyzed for correlation with the average APAR accumulation parameters of trees in the sample plot.

[0047] Next, we will conduct in-depth relationship mining, and perform PCA dimensionality reduction and K-means clustering based on individual tree attribute parameters and individual tree spatial structure parameters.

[0048] Finally, based on the reduced individual tree attribute parameters, individual tree spatial structure parameters, and individual tree APAR accumulation parameters, a fractal individual tree structure-APAR Bagging integrated learning model is constructed to achieve high-precision modeling and quantitative analysis of the structure-radiation coupling mechanism.

[0049] like Figure 2 As shown, using 12 structural indicators at the single-tree scale as input features, principal component analysis (PCA dimensionality reduction algorithm) was used to compress and remove redundancy from the structural variables, retaining the main variation information. Then, K-means clustering algorithm was applied to perform cluster analysis on the dimensionality-reduced data, generating dimensionality-reduced clustering results (e.g., A, B, C), thereby revealing the spatial heterogeneity patterns and classification characteristics of single-tree structures.

[0050] After entering the integrated modeling stage, an independent Bagging integrated model is built for each structural fractal.

[0051] The dimensionality-reduced data within each fractal will be repeatedly sampled using the Bootstrap method to generate multiple sample subsets (Bootstrap resampled samples A, B, C), which will be used to train multiple base learners (M1, M2, M3) to capture local patterns in the sample structure from different perspectives.

[0052] Subsequently, the outputs of all basic learners within this subtype are integrated and summarized to obtain the ensemble models (ensemble model 1, ensemble model 2, and ensemble model 3) for this structural type.

[0053] Example 2: As Figure 1 , Figure 2 , Figure 3 Figure 4(a), Figure 4(b), Figure 5(a), Figure 5

[0054] (b) Figure 6 Figures 7(a), 7(b), 7(c), and 7(d) Figure 8 , Figure 9 Figure 10

[0055] (a1), Figure 10(a2), Figure 10(a3), Figure 10(b1), Figure 10(b2), Figure 10(b3), Figure 10(c1), Figure 10(c2), Figure 10(c3) and Figure 11 As shown, a smart coupling analysis and modeling method for forest structure and photosynthetically active radiation absorption includes the following steps:

[0056] Step 1.1: Dynamic modeling and multidimensional parameter analysis of forest stand structure integrating neighborhood topological relationships:

[0057] The point clouds of airborne lidar (ALS) and handheld lidar (HLS) are integrated and spatial registration, noise filtering and height normalization are performed in sequence to achieve precise segmentation at the single-tree level.

[0058] Based on single-tree point clouds, basic parameters such as diameter at breast height (DBH), tree height (H), canopy feature parameters (CW, CA, CV), and latitude and longitude are extracted using point cloud editing software (such as Lidar360).

[0059] Further extraction of neighborhood spatial structure parameters: First, the tree coordinates are normalized to a certain area range to facilitate the construction of digital plots; then, based on the normalized coordinates, Delaunay triangulation is performed, and the central tree and its four nearest neighbors are searched according to the "1+4" principle to construct spatial structure units, and seven spatial structure indicators are calculated.

[0060] Forest gap density (GD) was extracted based on individual tree counts in each sample plot, forest gap mean area (GAM) was extracted based on the canopy height model (CHM), slope (SL) was derived from the digital elevation model (DEM), and stand density (SD) was derived from tree counts in each sample plot.

[0061] The definitions and calculation formulas for the seven spatial structure indicators are as follows.

[0062] (1) Spatial density index (DE)

[0063] This index describes the horizontal distribution and density of trees in a neighborhood space. Its calculation formula is:

[0064]

[0065] Where, r ir is the maximum distance between the central tree and its four neighboring trees in the i-th unit. max The maximum nearest neighbor distance observed across all cells.

[0066] (2) Forest Layer Index (S)

[0067] This metric is used to characterize the hierarchical complexity of vertical structures within a neighborhood, and the calculation formula is as follows:

[0068]

[0069] Where n represents the number of neighboring trees in the unit, S ij For discrete variables: if the central tree and its neighboring trees are not in the same forest layer, S ij =1; otherwise S ij =0. The forest layer is divided using a 2-meter vertical height interval.

[0070] (3) Openness (OP)

[0071] This indicator is used to quantify the significance of canopy shading and canopy gaps. Its calculation formula is:

[0072]

[0073] The variable n represents the number of neighboring trees within a neighborhood structural unit, and t ij For discrete variables: when the horizontal distance between the central tree and its neighboring trees is greater than the absolute value of their height difference, t ij =1; otherwise t ij =0.

[0074] (4) Size ratio (U)

[0075] This index reflects the degree of size differentiation among trees in a neighborhood. Its calculation formula is as follows:

[0076]

[0077] The variable n represents the number of neighboring trees within a neighborhood structural unit, and K ij For discrete variables: when the diameter at breast height (DBH) of neighboring trees is smaller than that of the central tree, K ij =0; otherwise K ij =1.

[0078] (5) Hegyi Competition Index (CI)

[0079] This indicator measures the intensity of competition between neighboring trees and the target tree. Its calculation formula is:

[0080]

[0081] Where, d i and dj These are the diameters at breast height (DBH) of the target tree and neighboring trees, respectively. ij It is the distance between them.

[0082] (6) Spatial Dominance Index (LCDI)

[0083] This index reflects the spatial dominance of the canopy of a target tree within its neighborhood. Its calculation formula is as follows:

[0084]

[0085] Where CAc is the crown area of ​​the target tree, and CAn j It is the crown area of ​​the j-th neighboring tree.

[0086] (7) Near-Naturalness Index (NNLI)

[0087] This indicator is used to quantify how close the spatial pattern of trees is to their near-natural state. Its calculation formula is:

[0088]

[0089] NNLI i =1-2|Wi-0.5| (8)

[0090] Among them, W i Z represents the angle weight value. ij Z is a counting variable. The standard angle is defined as 72° (i.e., 360° divided by 4 neighboring trees). If the angle between the target tree and a neighboring tree is less than the set standard, then Z... ij =1; otherwise Z ij =0. NNLI i The higher the value, the closer the spatial distribution of nearby trees is to a near-natural pattern; conversely, the lower the value, the more it deviates from the natural pattern.

[0091] It also includes steps for three-dimensional radiative transfer simulation and single-tree APAR parameter extraction:

[0092] High-quality, complete tree point clouds were selected, and branch and leaf separation was performed using a combination of algorithms and manual methods. Branch and trunk modeling and automatic addition of 3D leaves were then implemented. The leaf area was fixed (adjusted according to tree species), and the number of leaves was controlled based on the LAI parameters extracted from the sample plots. The LAI calculation formula is as follows:

[0093]

[0094] LAI = 0.2227e 3.6566MATVI2 (10)

[0095] In the formula, MTVI2 is the vegetation index, R is the reflectance at the corresponding wavelength, and e is the natural constant (Euler number).

[0096] Finally, a complete 3D tree model library was constructed. Based on model matching algorithms, the best-matching model was selected according to attributes such as tree height, diameter at breast height (DBH), crown width, crown area, and crown volume for digital forest scene layout. After the sample plots were digitized, sensor, atmospheric, photosynthetic radiation, and stand spectral parameters (forest land, leaves, bark, shrubs, etc.) were set, and simulation type and computer performance parameters were configured. After the simulation, the APAR parameters of individual trees in each sample plot were extracted.

[0097] It also includes structural-radial coupling analysis and modeling steps:

[0098] For each tree, the cumulative APAR at vertical height (TAPAR) was extracted with a step size of 2m, and the average APAR at the plot scale (MAPAR) was calculated. Spearman correlation analysis was used at the plot scale to evaluate the relationship between MAPAR and slope (SL), mean gap area (GAM), gap density (GD), and stand density (SD). At the tree-scale, the responses of TAPAR and four vertical APAR distribution indices (Gini-APAR, CV-APAR, Rel-MeanLayer, Rel-MaxLayer) and twelve structural attributes (DBH, H, CW, CA, CV, DE, S, U, OP, Cl, NNLI, LCDI) were analyzed.

[0099] Based on this, the steps to construct a Bagging ensemble regression model include:

[0100] Step (1): Use PCA to reduce dimensionality and select representative variables, eliminating redundancy;

[0101] Step (2): K-means clustering optimizes variable classification;

[0102] Step (3): Bootstrap resampling generates a training subset;

[0103] Step (4): Train multiple base learners;

[0104] Step (5): Integrate predictions to form the final model;

[0105] Step (6): Achieve high-precision prediction of single-tree TAPAR and improve robustness and generalization ability (see...) Figure 2 ).

[0106] A multi-source data-driven intelligent coupling analysis and modeling method for forest structure-absorbed photosynthetically active radiation (APAR) is proposed. This method achieves accurate extraction of multi-scale structural parameters, high-fidelity 3D forest stand modeling, and radiative transfer simulation, featuring high accuracy, high adaptability, and interpretability. By constructing a structure-light energy response model through genotypic ensemble learning, key structural factors are identified, and the APAR regulation mechanism is quantified. Applicable to natural forests, complex terrains, and diverse forest types, this method can be widely used in forest light environment modeling, ecological process simulation, structural optimization management, and remote sensing product interpretation.

[0107] Example 3: As Figure 1 , Figure 2 , Figure 3 Figure 4(a), Figure 4(b), Figure 5(a), Figure 5(b) Figure 6 Figures 7(a), 7(b), 7(c), and 7(d) Figure 8 , Figure 9 Figures 10(a1), 10(a2), 10(a3), 10(b1), 10(b2), 10(b3), 10(c1), 10(c2), and 10(c3) and Figure 11 As shown, a smart coupling analysis and modeling method for forest structure and photosynthetically effective radiation absorption is presented, with an example of structure-radiation coupling analysis of a fixed sample plot in Qilian Mountain National Forest Park, Qinghai Province.

[0108] 1. Multi-source data acquisition and processing

[0109] Using 144 adjacent plots of pure Qinghai spruce forest at the Qilian Mountain Sigou monitoring station in Qinghai Province as examples, this embodiment integrates airborne and handheld lidar data. After resampling, denoising, and normalization, high-quality point clouds are generated. Point clouds of each tree are extracted through individual tree segmentation to obtain attributes such as diameter at breast height (DBH), tree height, canopy characteristics, and spatial location. Complete point clouds are selected, and branch and leaf separation is completed using a combination of algorithms and manual correction to construct a high-quality point cloud database. Using an airborne multispectral sensor, remote sensing images at noon are simultaneously captured, covering four key spectral channels: green, red, red-edge, and near-infrared (used to calculate the leaf area index (LAI) of each plot). The hyperspectral reflectance of each forest group is measured using an RS-8800 hyperspectral analyzer, and the needle transmittance is calculated based on the physiological parameter PROSPECT-D model.

[0110] 2. Construction of spatial networks and extraction of structural parameters for various sample areas

[0111] Based on individual tree attributes and spatial coordinates, the tree coordinates were normalized to a 20m × 20m range, and then Delaunay triangulation was performed. The four nearest neighboring trees were selected for each tree to construct a spatial structure unit, and seven spatial structure indices (DE, OP, S, U, Cl, LCDl, NNLI) were calculated. Slope (SL) and gap structure parameters (GD, GAM) were calculated using DEM and CHM data (0.1m resolution) extracted by airborne lidar (see...). Figure 3 Stand density (SD) is obtained from tree counts.

[0112] 3. Digital plot reconstruction and three-dimensional radiative transfer simulation

[0113] Using DEM data acquired by airborne LiDAR and shapefiles of 144 20m×20m plots, the terrain was segmented and imported into the LESS 3D radiative transfer model to generate digital terrain. Tree branches and trunks were reconstructed based on a complete tree point cloud library, and leaves were added based on the LAI of each plot, thus reconstructing a complete tree model and forming a tree model library. Based on the individual tree attributes of each plot, the optimal 3D tree model was automatically matched from the model library using the Euclidean distance similarity function to achieve accurate replacement and arrangement according to coordinates. After the digital plots were constructed, the reconstruction results were subjected to individual tree layer rendering and APAR simulation using the 3D radiative transfer model. The model's Python SDK was used to simulate the 144 plots in batches. The simulation results are shown using a typical plot in the core area (number 2511) as an example (see Figure 4(a) and Figure 4(b)).

[0114] 4. Multi-scale structural-radiative coupling analysis and construction of a single-tree-level structural-APAR ensemble learning model

[0115] Four vertical distribution indices of APAR (Gini-APAR, CV-APAR, Rel-MeanLayer, Rel-MaxLayer) and cumulative APAR per tree (TAPAR) were extracted from 3457 trees. Spearman correlation analysis was performed with 12 structural indices per tree (DBH, H, CW, CA, CV, DE, OP, S, Cl, U, LCDI, NNLI) (see [link to analysis]). Figure 6 Figures 7(a), 7(b), 7(c), and 7(d) are shown. The average APAR (MAPAR) of trees in each plot was extracted and correlated with four plot structure indicators (SL, SD, GAM, GD) (see Figures 7(a), 7(b), 7(c), and 7(d)). Based on the 12 structure indicators, PCA dimensionality reduction and KMeans clustering were performed to simplify the structure and divide the population (see Figures 7(a), 7(b), 7(c), and 7(d)). Figure 8 , Figure 9Combining dimensionality reduction and three-class clustering results, a single-tree-level structure-APAR coupled ensemble learning model was constructed. The training results are detailed in Figure 10. Figure 11 This study reveals the regulatory mechanism of forest structure on APAR.

[0116] The results demonstrate that the intelligent coupling analysis and modeling method for forest structure and photosynthetically active radiation (APAR) proposed in this invention exhibits high adaptability and precise, efficient analytical capabilities. This method significantly reduces the cost of manual intervention, improves data processing and modeling efficiency, and possesses good adaptability and scalability, making it suitable for diverse forest types and complex terrains. The constructed intelligent coupling model effectively reveals the regulatory mechanism of forest structure on APAR, providing strong technical support for quantitative research on forest ecosystems and showcasing broad application and promotion prospects. This method integrates multi-source lidar point clouds, multispectral imagery, and hyperspectral reflectance parameters, combined with three-dimensional reconstruction and radiative transfer models, to achieve quantitative analysis of the entire process from structural parameter extraction to refined APAR characterization. It has significant guiding significance for practical forest structure regulation to optimize the light environment and fully meets the needs of refined forestry management at the grassroots level.

[0117] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent coupling analysis and modeling of forest structure and photosynthetically active radiation absorption, characterized in that, Includes the following steps: Step 1: Focus on integrating airborne and handheld lidar information to accurately retrieve forest structure indicators, including single-tree attributes such as diameter at breast height (DBH), tree height (H), crown width (CW), crown area (CA) and crown volume (CV), and neighborhood spatial structure indicators such as spatial density (DE), size ratio (U), forest layer index (S), openness (OP), competition index (CI), nearness to nature (NNLI), dominance (LCDI), gap structure density (GD) and average area (GAM), stand density (SD), and slope (SL). Step 2: Conduct three-dimensional radiative transfer simulation. By constructing a single-tree three-dimensional model library and digital sample plots, high-precision three-dimensional forest reconstruction and hierarchical APAR simulation are achieved, and photosynthetic spatial heterogeneity parameters are extracted. Step 3: Perform Spearman correlation analysis based on structure and APAR parameters, and combine PCA dimensionality reduction and KMeans clustering to identify dominant structural patterns and spatial types; Step 4: Construct a fractal structure-APAR coupled ensemble learning model based on clustering results to reveal the regulatory mechanism of individual trees and neighborhood structures on APAR. It also includes the following steps: Dynamic modeling and multidimensional parameter analysis of forest stand structure based on coupled neighborhood topology. By fusing point clouds from airborne and handheld LiDAR systems (HLS), spatial registration, noise filtering, and height normalization are sequentially performed to achieve precise segmentation at the tree level. Based on single-tree point clouds, point cloud editing software was used to extract basic parameters such as diameter at breast height (DBH), tree height, canopy feature parameters CW, CA, CV, and latitude and longitude. Further extraction of neighborhood spatial structure parameters: First, the tree coordinates were normalized to a certain area range for digital plot construction; then, Delaunay triangulation was performed based on the normalized coordinates, and the central tree and its four nearest neighbors were searched according to the 1+4 principle to construct spatial structure units, and seven spatial structure indices were calculated. Forest gap density (GD) and forest gap mean area (GAM) were extracted based on the canopy height model (CHM), slope (SL) was derived from the digital elevation model (DEM), and stand density (SD) was derived from tree counts in each sample plot. The definitions and calculation formulas for the seven spatial structure indicators are as follows: (1) Spatial density index DE This index describes the horizontal distribution and density of trees in a neighborhood space, and its calculation formula is as follows: (1) Where, r i r is the maximum distance between the central tree and its four neighboring trees in the i-th unit. max The maximum nearest neighbor distance observed across all cells. (2) Forest layer index S This metric is used to characterize the hierarchical complexity of vertical structures within a neighborhood, and the calculation formula is as follows: (2) Where n represents the number of neighboring trees in the unit, S ij For discrete variables: if the central tree and its neighboring trees are not in the same forest layer, S ij =1; otherwise S ij =0, the forest layer is divided using a 2-meter vertical height interval. (3) Openness OP This indicator is used to quantify the significance of canopy shading and canopy gaps, and its calculation formula is as follows: (3) The variable n represents the number of neighboring trees within a neighborhood structural unit, and t ij For discrete variables: when the horizontal distance between the central tree and its neighboring trees is greater than the absolute value of their height difference, t ij =1; otherwise t ij =0, (4) The ratio of the size U This index reflects the degree of size differentiation among trees in a neighborhood, and its calculation formula is as follows: (4) The variable n represents the number of neighboring trees within a neighborhood structural unit, and K ij For discrete variables: when the diameter at breast height (DBH) of neighboring trees is smaller than that of the central tree, K ij =0; otherwise K ij =1, (5) Hegyi Competition Index (CI) This indicator measures the intensity of competition between neighboring trees and the target tree, and its calculation formula is as follows: (5) in, and These are the diameter at breast height (DBH) of the target tree and the neighboring trees, respectively. It is the distance between them. (6) Spatial dominance index LCDI This index reflects the spatial dominance of the target tree's canopy within its neighborhood, and its calculation formula is as follows: (6) Where CAc is the crown area of ​​the target tree. It is the crown area of ​​the j-th neighboring tree. (7) Near-Naturalness Index (NNLI) This indicator is used to quantify the degree to which the spatial pattern of trees is close to its natural state, and its calculation formula is as follows: (7) (8) in, Indicates the angle weight value. It is a counting variable, with a standard angle defined as 72°. If the angle between the target tree and a neighboring tree is less than the set standard, then... = 1; otherwise = 0, The higher the value, the closer the spatial distribution of nearby trees is to the natural pattern; conversely, the lower the value, the more it deviates from the natural pattern.

2. The intelligent coupling analysis and modeling method for forest structure and photosynthetically active radiation absorption as described in claim 1, characterized in that, It also includes steps for three-dimensional radiative transfer simulation and single-tree APAR parameter extraction: High-quality, complete tree point clouds were selected. Branch and leaf separation was performed using a combination of algorithms and manual methods. Branch and trunk modeling and automatic addition of 3D leaves were then carried out. The leaf area was fixed, and the number was adjusted based on the LAI parameters extracted from the sample plots. The LAI calculation formula is as follows: (9) (10) In the formula, MTVI2 is the vegetation index, R is the reflectance at the corresponding wavelength, and e is a natural constant. Finally, a complete 3D tree model library was constructed. Based on the model matching algorithm, the most matching model was selected according to the tree height, diameter at breast height, crown width, crown area, and crown volume attributes to arrange the digital forest scene. After the sample plots were digitized, sensor, atmospheric, photosynthetic radiation, and stand spectral parameters were set, and simulation type and computer performance parameters were configured. After the simulation was completed, the APAR parameters of individual trees in each sample plot were extracted.

3. The intelligent coupling analysis and modeling method for forest structure and photosynthetically active radiation absorption according to claim 1, characterized in that, It also includes structural-radial coupling analysis and modeling steps: For each tree, the cumulative APAR (Total APAR) at vertical height was extracted with a step size of 2 m. The average APAR at the plot scale was calculated. Spearman correlation analysis was used at the plot scale to evaluate the relationship between MAPAR and slope (SL), mean gap area (GAM), gap density (GD), and stand density (SD). At the individual tree scale, the responses of TAPAR and the four vertical APAR distribution indices (Gini-APAR, CV-APAR, Rel-MeanLayer, Rel-MaxLayer) and the twelve structural attributes (DBH, H, CW, CA, CV, DE, S, U, OP, CI, NNLI, LCDI) were analyzed. Based on this, the steps to construct a Bagging ensemble regression model include: Step (1): Use PCA to reduce dimensionality and select representative variables, eliminating redundancy; Step (2): K-means clustering optimizes variable classification; Step (3): Bootstrap resampling generates a training subset; Step (4): Train multiple base learners; Step (5): Integrate predictions to form the final model; Step (6): Achieve high-precision prediction of single-tree TAPAR and improve robustness and generalization ability.

Images