Forest sample plot vertical layered vegetation coverage estimation method

Abstract

The invention discloses a forest sample plot vertical layered vegetation coverage estimation method, which comprises the following steps of S10, acquiring laser radar point cloud data and optical image data of a target plot, and performing preprocessing; s20, vertically layering the laser radar point cloud data, respectively extracting the uppermost point cloud of each vegetation layer in a visible and sheltered region based on the first echo and the multiple echoes, and calculating the initial projection area of each layer according to the extracted uppermost point cloud; s30, fusing the optical features with the point cloud structure features, constructing a vegetation layer projection area correction model, and correcting the initial projection area of the vegetation layer by using the model to obtain a corrected projection area; and S40, establishing a quantitative conversion model of the two-dimensional FVC and the three-dimensional FVC based on a geometrical optical model and a clearance rate theory. According to the method, the dependence on point cloud classification precision is overcome, the estimation precision of the FVC of the under-forest vegetation, especially the herbaceous layer, is remarkably improved, and the method has the advantages of being clear in physical mechanism and high in adaptability.

Description

Technical Field

[0001] This invention belongs to the field of forest remote sensing monitoring and vegetation parameter inversion technology, specifically involving a method for estimating the vertical stratification of vegetation cover in forest plots by integrating optical remote sensing and lidar data. Background Technology

[0002] Regional vegetation cover (FVC) is a key indicator for measuring the status of land cover and plays an important role in ecological monitoring and climate change research. Traditional methods, mostly based on optical remote sensing data, can only provide comprehensive cover information in a two-dimensional plane and cannot reveal the differences in the vertical stratification structure of forests. However, forest ecosystems exhibit distinct vertical stratification characteristics, including canopy, shrub, and herbaceous layers, each with different ecological functions. Therefore, accurate estimation of vertical FVC is crucial for understanding forest structure and function.

[0003] Airborne lidar (LiDAR) can acquire three-dimensional structural information of forests, providing technical support for vertical stratification FVC estimation. Existing LiDAR-based FVC estimation methods mainly include echo point count method, echo intensity method, point cloud volume method, and area cover method, but these methods have obvious limitations: on the one hand, they are highly dependent on the accuracy of point cloud classification, sensitive to parameter settings, and have large errors in complex forest structures; on the other hand, they mainly focus on the overall or canopy FVC, and are insufficient in representing the understory vegetation such as shrub and herb layers, making it difficult to achieve true vertical stratification estimation.

[0004] Current methods for estimating 3D FVC (Floating Ventilation Cover) based on vertical stratification mainly rely on techniques such as point cloud vertical stratification statistical analysis to express vegetation cover information at different height intervals within the canopy. However, the research focus of these methods remains on the vertical division of the canopy structure, with limited quantitative expression of the structure of understory vegetation such as shrubs and herbs. Particularly in the estimation of herbaceous layer FVC, due to weak point cloud signals, easy confusion with ground echoes near the ground surface, and severe point cloud loss caused by shading from the upper canopy, existing methods struggle to accurately reconstruct the spatial structural characteristics of understory vegetation. This has become the main bottleneck restricting the accuracy of vertical stratification FVC estimation.

[0005] In summary, existing technologies have not yet been able to effectively integrate surface vegetation distribution information from optical remote sensing with the vertical structural advantages of LiDAR, and lack a robust method with a clear physical mechanism that can independently and accurately estimate the FVC of each vertical vegetation layer. Summary of the Invention

[0006] To address the shortcomings of existing methods for estimating vertical vegetation cover in forest plots, such as strong dependence on point cloud classification, insufficient ability to represent understory vegetation structure, and unclear physical mechanisms, this invention provides a method for estimating vertical vegetation cover in forest plots. By fusing two-dimensional vegetation cover information provided by optical remote sensing imagery with three-dimensional structural information described by lidar data, independent estimation of the field-vegetation capacity (FVC) of different vegetation layers is achieved under the constraints of a geometric optics model and gap rate theory, thereby improving the accuracy and stability of forest vertical structure inversion.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for estimating vertical vegetation cover in forest plots includes the following steps: S10: Acquire lidar point cloud data and optical image data of the target location and perform preprocessing; S20: Vertically layer the lidar point cloud data, extract the topmost point cloud of each vegetation layer in the visible and occluded areas based on the first echo and multiple echoes, and calculate the initial projected area of ​​each layer based on the extracted topmost point cloud; S30: The optical features and point cloud structural features are fused to construct a projection area correction model for the understory vegetation layer. The model is then used to correct the initial projection area of ​​the understory vegetation layer to obtain the corrected projection area. S40: Based on the geometric optics model and gap ratio theory, a quantitative conversion model between two-dimensional FVC and three-dimensional FVC is established. Combining the corrected vegetation layer projection area and optical visibility ratio, the three-dimensional FVC of each vertical vegetation layer is independently derived.

[0008] Optionally, preprocessing includes ground point extraction and height normalization of the lidar point cloud, as well as calculating the two-dimensional FVC within each analysis unit; the initial projected area of ​​each vegetation layer is the vegetation cover area calculated within the analysis unit based on the projection of the topmost point cloud of that layer onto the horizontal plane; the analysis unit is a regular grid unit, an image segmentation unit, or other spatial partitioning unit.

[0009] Optionally, in step S20, the extraction of the uppermost point cloud of each vegetation layer includes: First echo point cloud processing: Based on a preset vertical layering height threshold, the first echo point cloud is layered, and the highest point cloud in each layer that is not obscured by the upper vegetation layer is extracted as the visible uppermost point cloud of each layer. Multiple echo point cloud processing: For vegetation layers that were not recorded in the first echo due to shading by upper vegetation, multiple echo point clouds are used to identify and extract the uppermost point cloud in the shading area, which is then used as the uppermost shading point cloud for each layer.

[0010] Optionally, the step of identifying and extracting the occluded uppermost point cloud using multiple echo point clouds specifically includes: Vertical height stratification: Based on a preset height threshold, multiple echo point cloud data are divided into different vegetation layers in the vertical direction; Spatial matching: For multiple echo points in the current vegetation layer, determine whether there is a first echo point in the overlying vegetation layer within a preset horizontal projection distance; if so, mark the multiple echo points as occluded candidate points. Top-level point filtering: Within the preset horizontal neighborhood of the occluded candidate point, determine whether there is a point cloud of the same vegetation layer with a higher vertical height; if not, then determine the multiple echo points as the top-level point cloud of the current vegetation layer in the occluded area.

[0011] Optionally, the forest understory vegetation layer projection area correction model in step S30 is constructed based on optical features and point cloud structure features. The optical features include at least one of vegetation index, spectral texture features, or spectral derived features. The point cloud structure features include at least one of vegetation layer projection area, point cloud transmittance, visible area parameter, or shading area parameter. The forest understory vegetation layer projection area correction model is a regression model, a statistical model, or a machine learning model.

[0012] Optionally, in step S40, the quantitative conversion model is expressed as:

[0013] in, FVC 2D This represents the total vegetation coverage of the target area projected onto a two-dimensional plane. FVC top Indicates the upper canopy layer FVC, FVC mid Indicates the intermediate shrub layer FVC, FVC understory Indicates the lower herbaceous layer FVC, a This represents the visible proportion of the middle layer vegetation that is not obscured by the upper layer vegetation in the vertical projection direction. b This indicates the proportion of lower vegetation that is not obscured by upper vegetation in the vertical projection direction.

[0014] Optionally, in step S40, the FVC calculation formulas for the canopy layer, shrub layer, and herb layer are related to the visibility parameters of the shrub and herb layers. a , b The calculation formulas are as follows:

[0015]

[0016] in,S visible_shrub This refers to the area of ​​vegetation in the uppermost point cloud layer of the shrub layer that is not obscured by the upper canopy. S occluded_shrub_LiDAR This represents the area of ​​vegetation obscured by the upper canopy in the topmost point cloud of the shrub layer, calculated based on LiDAR data. S occluded_shrub_correction This is the compensation amount for the shading area based on model prediction. S visible_grass This refers to the area of ​​vegetation in the uppermost point cloud of the herbaceous layer that is not obscured by the upper canopy. S occluded_grass_LiDAR This represents the area of ​​vegetation obscured by the upper canopy in the top point cloud of the herbaceous layer, calculated based on LiDAR data. S occluded_grass_correction This is the compensation amount for the occlusion area based on model prediction; Formulas for calculating the FVC of the canopy and shrub layers:

[0017]

[0018] For step S40, the herbaceous layer FVC is obtained by reverse calculation using a transformation model:

[0019] in, S canopy This represents the vertical projection area of ​​vegetation on the uppermost point cloud layer of the canopy. S pixel The area per unit pixel.

[0020] Optionally, each vegetation layer is preset as a canopy layer, a shrub layer, and a herbaceous layer; or, each vegetation layer is a more subdivided or coarser hierarchical structure formed by dividing the vertical structure of vegetation based on the preset vertical stratification height threshold.

[0021] Optionally, the optical image data may include multispectral images, hyperspectral images, or visible light images for extracting spectral features or spectral texture features.

[0022] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the program, implements the above-described method for estimating vertical stratified vegetation cover in forest plots.

[0023] The present invention achieves the following beneficial effects through the above technical solution: 1. Avoids dependence on traditional point cloud classification and parameter settings: This invention directly extracts the topmost point cloud of each layer based on height layering and spatial occlusion analysis, without the need to classify ground and vegetation points, and does not rely on sensitive parameters such as voxel size and filtering threshold. This effectively reduces systematic errors in the preprocessing stage and improves the automation level and applicability of the model.

[0024] 2. The ability to identify understory vegetation was enhanced by reconstructing the spatial structure of the layered point cloud: By integrating the spatial pairing analysis of the first echo and multiple echoes, a strategy for extracting the top layer of point cloud in each layer was constructed. Even in areas occluded by the upper layer of vegetation, the top layer of point cloud representing the middle and lower layers of vegetation can be accurately identified. This breaks through the traditional simple division method based solely on height thresholds and improves the ability to express the structure of understory vegetation.

[0025] 3. Overcoming the bottleneck in forest understory FVC estimation: To address the challenges of low penetration, insufficient sampling, and easy confusion between lidar point clouds and ground points in forest understory areas, an area correction model combining optical features and point cloud structure features is introduced to effectively compensate for occluded areas. Simultaneously, by combining the sensitivity of optical remote sensing data to surface vegetation distribution, a two-dimensional to three-dimensional cover conversion model is constructed based on a geometric optics model and gap rate theory, indirectly achieving reliable estimation of herbaceous layer cover and solving the problem of missing herbaceous layer structure information in existing methods.

[0026] 4. Possesses excellent adaptability, robustness, and scalability: This invention demonstrates high accuracy and strong robustness under different vegetation structures, forest types, and seasonal conditions. Furthermore, this invention is not only applicable to the three-layer division of trees, shrubs, and grasses, but theoretically can be extended to the estimation of continuous profiles at arbitrary vertical height intervals, providing a universal technical framework for the refined and multi-layered characterization of forest vertical structure. Attached Figure Description

[0027] Figure 1 This is a flowchart of the method steps in an embodiment of the present invention; Figure 2 This is a schematic diagram of the field observation plots of BART, HARV, and GRSM used for verification in the embodiments of the present invention; Figure 3 This is a diagram illustrating the conceptual relationship between the two-dimensional FVC and the vertically layered FVC, which are the core concepts of this invention. Figure 4 This is a schematic diagram of the layered vegetation point cloud extraction process of the present invention, with occlusion point cloud extraction taking shrub layer extraction as an example; Figure 5 Spatial distribution map of the stratified FVC results estimated in sample plots of two different forest types (evergreen coniferous forest and deciduous broadleaf forest) in the embodiments of the present invention; Figure 6Spatial distribution map of stratified FVC results estimated in spring, summer and autumn in deciduous broad-leaved forest plots (GRSM sites) using embodiments of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0029] Example: Figure 1 As shown, this invention provides a method for estimating vertical vegetation cover in forest plots, comprising the following steps: S10: Acquire lidar point cloud data and optical image data of the target location and perform preprocessing.

[0030] Acquire airborne LiDAR point cloud data and synchronized optical remote sensing imagery (such as high-resolution multispectral or hyperspectral imagery) for the target area. Preprocess the LiDAR point cloud, including extracting ground points using filtering algorithms (such as cloth simulation filtering) and performing height normalization to eliminate terrain effects. Process the optical imagery, calculating two-dimensional vegetation cover (FVC) within selected analysis units (e.g., a regular grid of 10m × 10m) using methods such as supervised classification.

[0031] S20: Vertically layer the lidar point cloud data, extract the top layer point cloud of each vegetation layer in the visible and occluded areas based on the first echo and multiple echoes, and calculate the initial projected area of ​​each layer based on the extracted top layer point cloud.

[0032] Based on preset vegetation layer height thresholds (e.g., canopy layer > 3m, shrub layer 0.5-3m, herbaceous layer 0.1-0.5m), the LiDAR point cloud is vertically layered. For the first echo point cloud, the highest point in each layer that is not occluded by the upper layer is directly extracted as the uppermost point cloud in the "visible area" of that layer. For understory vegetation not recorded in the first echo due to occlusion, it is supplemented using multiple echo point clouds: for a multiple echo point, if there is a first echo point within a certain horizontal tolerance range above its vertical projection, it is marked as an "occluded candidate point"; if there is no point cloud of the same vegetation layer with a higher vertical height in the horizontal neighborhood of the candidate point, it is determined as the uppermost point cloud in the "occluded area" of that layer. All uppermost point clouds of each layer are summarized, and their projected area on the horizontal plane is calculated as the initial projected area of ​​each layer.

[0033] S30: The optical features and point cloud structural features are fused to construct a vegetation layer projection area correction model. The model is used to correct the initial projection area of ​​the vegetation layer to obtain the corrected projection area.

[0034] Due to the limitations of LiDAR in detecting understory vegetation, the initial projected area may be underestimated. Therefore, a correction model is constructed. The model's input features include optical features such as vegetation indices and spectral texture extracted from optical images, and point cloud structural features such as transmittance and visible area ratio calculated from point clouds. The training objective of the model is to make the predicted vegetation layer area closer to the actual situation. This model can be a machine learning model such as random forest. Using the trained model, the initial projected area of ​​each layer (especially the shrub and herbaceous layers) obtained in step S20 is corrected, and the corrected projected area and the area compensation amount for the corresponding occlusion area (e.g., ...) are output. Figure 4 As shown in the schematic diagram, the corrected projected area and the occlusion area compensation amount output will be used as key input parameters for the layered FVC calculation in the subsequent step S40.

[0035] S40: Based on the geometric optics model and gap ratio theory, a quantitative conversion model between two-dimensional FVC and three-dimensional FVC is established. Combining the corrected vegetation layer projection area and optical visibility ratio, the three-dimensional FVC of each vertical vegetation layer is independently derived.

[0036] This embodiment uses a typical forest plot in the National Ecological Observatory Network (NEON) as an example to illustrate the specific implementation process of the method described in this invention. To comprehensively verify the applicability and robustness of this method under different forest types and seasonal conditions, three representative plots in NEON (such as...) were selected for the study. Figure 2 As shown): Bartlett Experimental Forest (BART) evergreen coniferous forest plot, used to verify the estimation ability of the method in evergreen forests; Harvard Forest (HARV) deciduous broad-leaved forest plot, used to test the performance of the method under typical deciduous forest conditions; Great Smoky Mountains National Park (GRSM) deciduous broad-leaved forest plot, using observational data from its spring, summer, and autumn seasons to analyze the stability of the method under different phenological conditions.

[0037] like Figure 3As shown, the two-dimensional planar FVC essentially reflects the uppermost unobstructed visible vegetation information within a pixel. Its physical meaning corresponds to the proportion of vegetation intercepted in the first echo of the lidar, i.e., the ratio of vegetation points to the total number of points in the first echo. In the vertical stratification of forests, the two-dimensional FVC integrates the comprehensive contribution of each layer of vegetation to surface shading; the lower layer of shaded vegetation cannot be reflected in the two-dimensional FVC. Each layer's FVC should be distinguished into visible and shaded portions, and after shading correction, the sum of the actual FVCs of each layer should be greater than 1. Therefore, the two-dimensional FVC obtained by inverting remote sensing spectral data can be expressed as a weighted linear combination of the actual FVC of each layer and its spatial visibility proportion. The visible proportion of the lower layer vegetation is estimated through spatial shading analysis of point clouds, thereby achieving reasonable separation and accurate estimation of the contribution of each layer's FVC. The physical conversion relationship between two-dimensional and three-dimensional FVC is expressed as:

[0038] in, FVC 2D This represents the total vegetation coverage of the target area projected onto a two-dimensional plane. FVC top Indicates the upper canopy layer FVC, FVC mid Indicates the intermediate shrub layer FVC, FVC understory Indicates the lower herbaceous layer FVC, a This represents the visible proportion of the middle layer vegetation that is not obscured by the upper layer vegetation in the vertical projection direction. b This indicates the proportion of lower vegetation that is not obscured by upper vegetation in the vertical projection direction.

[0039] The above-described calculation process mainly includes four key steps: data acquisition and preprocessing, extraction of layered vegetation point clouds, construction of a forest understory vegetation area correction model, and calculation of layered FVC. The entire process fully integrates remote sensing spectral features and LiDAR multi-echo structure information, and performs mathematical decomposition of the layered structure based on a physical model, thereby achieving accurate estimation of layered FVC.

[0040] The following is combined Figures 1-6 The solution will be further described in detail with reference to a specific embodiment of the present invention.

[0041] like Figures 1-6 As shown in the figure, this embodiment provides a method for estimating the vertical stratification of vegetation cover in forest plots.

[0042] The preprocessing in step S10 includes ground point extraction and height normalization of the lidar point cloud, as well as calculating the two-dimensional FVC within each analysis unit.

[0043] Specifically, the optical remote sensing image data includes multispectral, hyperspectral, or visible light images that can be used to extract spectral features. This requires acquiring high-resolution UAV RGB imagery (data product number DP3.30010.001, spatial resolution 0.1m), UAV hyperspectral imagery (data product number DP3.30006.001, spatial resolution 1m), and registered airborne LiDAR point cloud data (data product number DP1.30003.001). The UAV RGB imagery is primarily used to generate high-precision two-dimensional FVCs through supervised classification methods. The UAV hyperspectral imagery is used to extract spectral and texture features to construct a forest understory vegetation area correction model. The airborne LiDAR point cloud data is used in conjunction with the remote sensing spectral data for hierarchical FVC estimation.

[0044] The present invention uses a 10m×10m regular grid as the basic analysis unit of FVC. The scale is selected based on the following considerations: (1) it meets the mesoscale observation requirements of forest parameter survey; (2) it matches the LiDAR point cloud density (about 10-20 points / m²) to ensure that each grid contains sufficient statistical samples; (3) it can effectively balance the expression of spatial details and computational efficiency.

[0045] Considering the impact of actual terrain undulations on vegetation height estimation, the ground point elevations must first be extracted. This invention employs a cloth-based simulated filtering algorithm to extract ground points from the LiDAR point cloud. Specific parameter settings are as follows: cloth grid resolution is set to 0.5 m, rigidity parameter is selected as medium to adapt to hilly terrain, iteration step size is set to 0.65, classification distance threshold is 0.03 m, maximum number of iterations is 300, and particle smoothing is disabled. After acquiring the ground points, they are further spatially divided into 10 m × 10 m grid cells, and the average elevation of the ground points within each grid cell is calculated as the ground elevation reference for that pixel.

[0046] Based on high-resolution UAV RGB imagery, a high-precision spatial distribution map of vegetation types, including trees, shrubs, and grasslands, was obtained using a random forest supervised classification algorithm combined with visual interpretation. Subsequently, the proportion of vegetation pixels within each grid cell was statistically analyzed on a 10 m × 10 m scale, serving as a reference value for the pixel's two-dimensional field-capacity (FVC).

[0047] The extraction of the uppermost point cloud of each vegetation layer in step S20 includes: First echo point cloud processing: Based on a preset vertical layer height threshold, the vertical vegetation layers are preset to be canopy layer, shrub layer and herb layer; or, the vertical vegetation layers are a more subdivided or coarser hierarchical structure formed by dividing the vertical structure of vegetation based on the preset height threshold. The first echo point cloud is layered, and the highest point cloud in each layer that is not obscured by the upper layer of vegetation is extracted as the uppermost visible point cloud in each layer. Multiple echo point cloud processing: For vegetation layers that were not recorded in the first echo due to shading by upper vegetation, multiple echo point clouds are used to identify and extract the uppermost point cloud in the shading area, which is used as the uppermost shading point cloud of each layer. The method utilizes multiple echo point clouds to identify and extract the occluded topmost point cloud layer, specifically including: Vertical height stratification: Based on a preset height threshold, multiple echo point cloud data are divided into different vegetation layers in the vertical direction; Spatial matching: For multiple echo points in the current vegetation layer, determine whether there is a first echo point in the overlying vegetation layer within a preset horizontal projection distance; if so, mark the multiple echo points as occluded candidate points. Top-level point filtering: Within the preset horizontal neighborhood of the occluded candidate point, determine whether there is a point cloud of the same vegetation layer with a higher vertical height; if not, then determine the multiple echo points as the top-level point cloud of the current vegetation layer in the occluded area.

[0048] In step S20, the initial projected area of ​​each vegetation layer is the vegetation coverage area calculated within the analysis unit based on the projection of the topmost point cloud of that layer onto the horizontal plane; the analysis unit is a regular grid unit, an image segmentation unit, or other spatial division unit.

[0049] Specifically, to ensure a unified benchmark for FVC calculation across all vegetation layers, this method uniformly extracts the topmost point cloud of each layer for statistical analysis. Vegetation layer FVC calculations should be based on the topmost point cloud within the corresponding layer. Except for the canopy layer, which only considers the first echo point from the unshaded area, the intermediate and lower vegetation layers must include the topmost point cloud from both the unshaded and shaded regions.

[0050] For the extraction of unoccluded point clouds, the point cloud of the first echo is decomposed according to height to extract the canopy layer ( H>H mid ), shrub layer ( H understory <H≤H mid ) and herbaceous layer ( H land <H≤H understory The first echo point cloud, that is, the part that is not occluded, is the upper point cloud.

[0051] Among them, the height of the boundary between the canopy layer and the shrub layer ( H mid The depth is set at 3-4 m, representing the boundary between the shrub and herb layers. H understory The thresholds are set to 0.3-0.5 m, and these thresholds need to be set based on the height of each vegetation layer in the sample plot forest.

[0052] For the point cloud in the occluded area, multiple echo point clouds from the intermediate and lower vegetation layers are used to identify the uppermost point cloud layer of each layer through spatial matching analysis. For each multiple echo point in the intermediate and lower layers, a search is conducted to determine if a first echo vegetation point from a higher layer exists directly above it. If such a point exists and there are no higher-level point clouds at the same location (within the horizontal tolerance range) above it, then that point is considered the uppermost point cloud layer in the occluded area. A spatial matching tolerance ε needs to be set. This tolerance parameter must ensure that the density of the solved point cloud in the occluded area is equal to that in the unoccluded area, thus decomposing the uppermost point cloud layer of the occluded area. ε is generally set to 1-3 times the average horizontal spacing of the point clouds.

[0053] After extracting the topmost point cloud layer for each layer, this invention employs a concave hull algorithm to calculate the ground projection area of ​​the point cloud to characterize the spatial cover features of each vegetation layer on the ground surface. Within a unit pixel, due to spatial discontinuities in vegetation distribution (such as gaps between tree canopies or patches of understory vegetation), multiple independent concave hull polygons may be formed. This invention calculates and sums the total projected area of ​​these polygons as the effective cover area of ​​that vegetation layer, serving as input data for subsequent understory layer area correction models and layered FVC calculations. This embodiment uses the α-shape concave hull algorithm, where the α parameter is set to 1.5 times the average spacing of the point cloud to accurately extract vegetation patch boundaries.

[0054] The forest understory vegetation layer projection area correction model in step S30 is constructed based on optical features and point cloud structure features. The optical features include at least one of vegetation index, spectral texture features, or spectral derived features. The point cloud structure features include at least one of vegetation layer projection area, point cloud transmittance, visible area parameter, or shading area parameter. The forest understory vegetation layer projection area correction model is a regression model, a statistical model, or a machine learning model.

[0055] Specifically, airborne LiDAR is easily affected by canopy shading in forest understory areas, resulting in insufficient sampling of point clouds in the shrub and herbaceous layers, thus affecting the accuracy of area estimation. To compensate for this lack of structural information, this invention introduces joint spectral and LiDAR structural features to construct a random forest model for predicting and correcting forest understory vegetation area. The model features include three categories: spectral indices, spectral texture features, and LiDAR structural parameters, totaling 18 feature variables (see Table 1). Among them, spectral features can capture the multiple scattering and transmission effects of photons within the vegetation canopy, compensating for the lack of understory information caused by LiDAR shading; spectral texture features characterize the spatial heterogeneity of vegetation distribution, assisting in identifying the multi-layered structure of vegetation; and LiDAR structural features quantitatively describe the spatial penetration ability and visibility ratio of point clouds in each layer, providing geometric and physical constraints for estimating the area of ​​shading areas. Based on the above features, using measured data of shrub and herbaceous layer areas as dependent variables, two independent random forest regression models are constructed to achieve accurate prediction and correction of the area of ​​each layer.

[0056] In the modeling process, variables are first selected based on the importance ranking of random forest features, and the 6–10 features with the highest contribution are chosen to participate in the final model training to improve the model's robustness and reduce redundant input. After training, the resulting model is used to generate the predicted area of ​​the shrub and herbaceous layers for each grid cell, and then corrected by combining the vegetation area estimated by LiDAR for each layer. Specifically, for grid cells with a penetration rate of less than 50% in the understory vegetation point cloud, an occlusion effect exists. In this case, it is necessary to determine whether the difference between the model's predicted area and the visible area of ​​the LiDAR point cloud (i.e., the predicted occlusion area) is greater than the occlusion vegetation area extracted from the LiDAR point cloud. If the predicted occlusion area is greater than the occlusion vegetation area extracted from the LiDAR point cloud, the excess is considered as the compensation amount for occlusion vegetation area not detected by LiDAR; conversely, if the predicted occlusion area is less than or equal to the occlusion vegetation area extracted from the LiDAR point cloud, the model prediction is considered consistent with the point cloud result, and the compensation amount for occlusion vegetation area is 0. This discrimination step allows for appropriate compensation of vegetation cover in the obscured area while maintaining physical plausibility, thereby improving the accuracy of understory area estimation.

[0057] Considering the strong spatial heterogeneity of forest understory vegetation at small scales, both model training and prediction were performed at a spatial resolution of 1m to more finely characterize the response relationship between spectral and structural features. All prediction results were summed up by area weight within a 10m regular grid to maintain consistency with the hierarchical FVC estimation unit.

[0058] Table 1 Input features and calculation methods of the forest understory vegetation area correction model

[0059] Abbreviations: NDVI (Normalized Difference Vegetation Index), EVI (Enhanced Vegetation Index), NDRE (Normalized Difference Red-Edge Index), NIRv (Near-Infrared Reflectance of Vegetation), GLCM (Gray-Level Co-occurrence Matrix), and TPI (Topographic Position Index).

[0060] In step S40, based on the principles of geometric optics, the two-dimensional FVC can be considered as a weighted sum of the three-dimensional FVC of each layer and its visible proportion (e.g., ...). Figure 3 As shown, the vertical layering structure and occlusion relationship are illustrated (this relationship determines the visibility ratio of each layer). The quantitative conversion model between two-dimensional and three-dimensional FVC is expressed as follows:

[0061] in, FVC 2D This represents the total vegetation coverage of the target area projected onto a two-dimensional plane. FVC top Indicates the upper canopy layer FVC, FVC mid Indicates the intermediate shrub layer FVC, FVC understory Indicates the lower herbaceous layer FVC, a This represents the visible proportion of the middle layer vegetation that is not obscured by the upper layer vegetation in the vertical projection direction. b This indicates the proportion of lower vegetation that is not obscured by upper vegetation in the vertical projection direction.

[0062] Based on the vegetation area of ​​each vegetation layer in the forest decomposed above, the formulas for calculating the FVC of the canopy layer, shrub layer, and herb layer, and the formulas for calculating the visibility parameters a and b of the shrub and herb layers are as follows:

[0063]

[0064] in, S visible_shrub This refers to the area of ​​vegetation in the uppermost point cloud layer of the shrub layer that is not obscured by the upper canopy. S occluded_shrub_LiDARThis represents the area of ​​vegetation obscured by the upper canopy in the topmost point cloud of the shrub layer, calculated based on LiDAR data. S occluded_shrub_correction This is the compensation amount for the shading area based on model prediction. S visible_grass This refers to the area of ​​vegetation in the uppermost point cloud of the herbaceous layer that is not obscured by the upper canopy. S occluded_grass_LiDAR This represents the area of ​​vegetation obscured by the upper canopy in the top point cloud of the herbaceous layer, calculated based on LiDAR data. S occluded_grass_correction This is the compensation amount for the occlusion area based on model prediction.

[0065] Based on the point clouds extracted from each vertical layer of the forest, the FVC of the canopy layer and the shrub layer can be calculated:

[0066]

[0067] in, S canopy This represents the vertical projection area of ​​vegetation on the uppermost point cloud layer of the canopy. S pixel The area per unit pixel.

[0068] Since LiDAR has difficulty effectively detecting herbaceous layers obscured by dense vegetation, and it is also difficult to separate ground point clouds from herbaceous point clouds, the underlying vegetation cover can ultimately be derived based on the physical conversion relationship between 2D and 3D FVC. FVC understory We can obtain:

[0069] in, FVC 2D It can be directly measured through field surveys; in the absence of actual measurement data, it can also be estimated by inversion from remote sensing optical images.

[0070] Through the above steps, this invention ultimately achieves independent and accurate estimation of the free vegetative capacities (FVCs) of the canopy, shrub, and herbaceous layers in forest plots, and generates corresponding spatial distribution maps of vertically stratified FVCs (see reference). Figure 5 and Figure 6 This provides reliable data support for monitoring the three-dimensional structure of forests and studying their ecological functions.

[0071] Based on the above specific implementation process, the forest vertical stratification FVC estimation method provided by this invention has the following characteristics and advantages: (1) The physical mechanism is clear and the principle is reliable: The method establishes a quantitative conversion model between two-dimensional optical FVC and three-dimensional vertically layered FVC based on the geometric optical model and the gap ratio theory. The process has a clear physical basis and the calculation results are highly interpretable.

[0072] (2) Simplified process and strong robustness: The method directly extracts the top point cloud of each vegetation layer through height threshold and spatial occlusion analysis, avoiding the dependence on complex point cloud classification (such as ground / vegetation separation) and sensitive filtering parameters, simplifying the preprocessing process, and improving the automation level and robustness of the algorithm in different scenarios.

[0073] (3) High estimation accuracy and outstanding ability to express understory vegetation: By reconstructing the top point cloud of each layer and introducing the area correction model with spectral-structural joint features, the insufficient sampling of lidar in the understory area is effectively compensated, the estimation accuracy of FVC of shrub and herb layer is improved, and the key problem of weak expression of understory structure by traditional methods is solved.

[0074] (4) Good adaptability and practicality: The method framework has good inclusiveness to data sources. Although this embodiment takes UAV hyperspectral imagery as an example, the optical constraint data can also come from various commonly used remote sensing platforms such as satellite multispectral imagery, which is convenient for promotion and application under different observation scales and data conditions.

[0075] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method for estimating vertical stratified vegetation cover in forest plots.

[0076] Specifically, the electronic device includes a computer program stored in a memory, a memory for storing data, and a communication interface for transmitting and acquiring data via a wireless or wired interface; the execution steps are as follows: 1. Acquire LiDAR point cloud data and optical image data of the target location through the communication interface, and call the preprocessing program to perform ground point extraction and height normalization processing on the LiDAR point cloud, as well as calculate the two-dimensional FVC within each analysis unit.

[0077] 2. Vertically layer the lidar point cloud data, extract the top layer point cloud of each vegetation layer in the visible and occluded areas based on the first echo and multiple echoes, and calculate the initial projected area of ​​each layer based on the extracted top layer point cloud.

[0078] 3. The optical features and point cloud structural features are fused to construct a vegetation layer projection area correction model. The initial projection area of ​​the vegetation layer is corrected using the model to obtain the corrected projection area.

[0079] 4. Based on the geometric optics model and gap ratio theory, a quantitative conversion model between two-dimensional FVC and three-dimensional FVC is established. Combining the corrected vegetation layer projection area and optical visibility ratio, the three-dimensional FVC of each vertical vegetation layer is independently derived.

[0080] It should be noted that the embodiments of this invention are intended to illustrate the technical solutions of this invention, and not to limit its scope of application. For those skilled in the art, any reasonable modifications or adaptive adjustments to form or implementation details made based on the core concepts and working principles disclosed in this invention, without departing from the essence of this invention, should be considered to fall within the protection scope of this invention. The actual protection scope of this invention should be determined by the scope defined in the appended claims.

Claims

1. A method for estimating vegetation cover in vertically stratified forest plots, characterized in that, Includes the following steps: S10: Acquire lidar point cloud data and optical image data of the target location and perform preprocessing; S20: Vertically layer the lidar point cloud data, extract the topmost point cloud of each vegetation layer in the visible and occluded areas based on the first echo and multiple echoes, and calculate the initial projected area of ​​each layer based on the extracted topmost point cloud. S30: The optical features and point cloud structural features are fused to construct a projection area correction model for the understory vegetation layer. The model is then used to correct the initial projection area of ​​the understory vegetation layer to obtain the corrected projection area. S40: Based on the geometric optics model and gap ratio theory, a quantitative conversion model between two-dimensional FVC and three-dimensional FVC is established. Combining the corrected vegetation layer projection area and optical visibility ratio, the three-dimensional FVC of each vertical vegetation layer is independently derived.

2. The method according to claim 1, characterized in that, The preprocessing includes ground point extraction and height normalization of the lidar point cloud, as well as calculation of the two-dimensional FVC within each analysis unit; the initial projected area of ​​each vegetation layer is the vegetation coverage area calculated within the analysis unit based on the projection of the topmost point cloud of that layer onto the horizontal plane; the analysis unit is a regular grid unit, an image segmentation unit, or other spatial division unit.

3. The method according to claim 1, characterized in that, The extraction of the uppermost point cloud of each vegetation layer in step S20 includes: First echo point cloud processing: Based on a preset vertical layering height threshold, the first echo point cloud is layered, and the highest point cloud in each layer that is not obscured by the upper vegetation layer is extracted as the top visible point cloud in each layer. Multiple echo point cloud processing: For vegetation layers that were not recorded in the first echo due to shading by upper vegetation, multiple echo point clouds are used to identify and extract the uppermost point cloud in the shading area, which is then used as the uppermost shading point cloud for each layer.

4. The method according to claim 3, characterized in that, The method of identifying and extracting the occluded uppermost point cloud using multiple echo point clouds specifically includes: Vertical height stratification: Based on a preset height threshold, multiple echo point cloud data are divided into different vegetation layers in the vertical direction; Spatial matching: For multiple echo points in the current vegetation layer, determine whether there is a first echo point in the overlying vegetation layer within a preset horizontal projection distance; if so, mark the multiple echo points as occluded candidate points. Top-level point filtering: Within the preset horizontal neighborhood of the occluded candidate point, determine whether there is a point cloud of the same vegetation layer with a higher vertical height; if not, then determine the multiple echo points as the top-level point cloud of the current vegetation layer in the occluded area.

5. The method according to claim 1, characterized in that, The forest understory vegetation layer projection area correction model in step S30 is constructed based on optical features and point cloud structure features. The optical features include at least one of vegetation index, spectral texture features, or spectral derived features. The point cloud structure features include at least one of vegetation layer projection area, point cloud transmittance, visible area parameter, or shading area parameter. The forest understory vegetation layer projection area correction model is a regression model, a statistical model, or a machine learning model.

6. The method according to claim 1, characterized in that, In step S40, the quantitative conversion model is expressed as follows: in, FVC 2D This represents the total vegetation coverage of the target area projected onto a two-dimensional plane. FVC top Indicates the upper canopy layer FVC, FVC mid Indicates the intermediate shrub layer FVC, FVC understory Indicates the lower herbaceous layer FVC, a This represents the visible proportion of the middle layer vegetation that is not obscured by the upper layer vegetation in the vertical projection direction. b This indicates the proportion of lower vegetation that is not obscured by upper vegetation in the vertical projection direction.

7. The method according to claim 6, characterized in that, In step S40, the FVC calculation formulas for the canopy layer, shrub layer, and herb layer are related to the visibility parameters of the shrub and herb layers. a , b The calculation formulas are as follows: in, S visible_shrub This refers to the area of ​​vegetation in the uppermost point cloud layer of the shrub layer that is not obscured by the upper canopy. S occluded_shrub_LiDAR This represents the area of ​​vegetation obscured by the upper canopy in the topmost point cloud of the shrub layer, calculated based on LiDAR data. S occluded_shrub_correction This is the compensation amount for the shading area based on model prediction. S visible_grass This refers to the area of ​​vegetation in the uppermost point cloud of the herbaceous layer that is not obscured by the upper canopy. S occluded_grass_LiDAR This represents the area of ​​vegetation obscured by the upper canopy in the top point cloud of the herbaceous layer, calculated based on LiDAR data. S occluded_grass_correction This is the compensation amount for the occlusion area based on model prediction; Formulas for calculating the FVC of the canopy and shrub layers: For step S40, the herbaceous layer FVC is obtained by reverse calculation using a transformation model: in, S canopy This represents the vertical projection area of ​​vegetation on the uppermost point cloud layer of the canopy. S pixel The area per unit pixel.

8. The method according to claim 3, characterized in that, The vegetation layers are preset to be a canopy layer, a shrub layer, and a herb layer; or, the vegetation layers are a more subdivided or coarser hierarchical structure formed by dividing the vertical structure of vegetation based on the preset vertical stratification height threshold.

9. The method according to claim 1, characterized in that, The optical image data includes multispectral images, hyperspectral images, or visible light images that can be used to extract spectral features.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for estimating the vertical stratified vegetation cover of forest plots as described in any one of claims 1 to 9.

Images