A method for identifying and estimating the biomass of a natural understory vegetation layer

Abstract

The application discloses a natural undergrowth layer identification and biomass estimation method, and belongs to the technical field of forest inventory. In the prior art, a simple unified height threshold is used for undergrowth layer division, and the height variation of undergrowth is not considered, which is not conducive to accurate identification of the undergrowth layer. The application analyzes a forest canopy point cloud profile, constructs a demarcation point based on a profile statistical function, and automatically identifies the undergrowth layer. A plurality of undergrowth layer LiDAR point cloud features are extracted, and feature variable optimization and modeling are performed to accurately estimate the undergrowth layer biomass. The application considers the vertical and horizontal distribution variation of undergrowth, thereby enhancing the undergrowth layer identification capacity and precision. The application can fundamentally increase the description capacity of the canopy space structure of point cloud data, and further improve the undergrowth layer identification and biomass estimation precision.

Description

Technical Field

[0001] This invention belongs to the field of forestry survey technology, specifically relating to a method for identifying natural forest understory vegetation and estimating its biomass. Background Technology

[0002] The understory vegetation layer (such as shrubs and herbs) is a crucial component of the forest ecosystem, accounting for approximately 6.2% of the total forest biomass. Accurate extraction of the understory vegetation layer is essential for precise forest resource surveys and carbon sequestration. Furthermore, understory vegetation information is vital for analyzing vegetation competition and suitability distribution, gaining a deeper understanding of forest biodiversity and functional structure, and for sustainable forest management, community succession analysis, and research on the entire ecosystem's carbon sequestration processes. Conventional forest resource surveys primarily focus on the tree layer, neglecting the contribution of understory vegetation. Moreover, their parameter extraction mainly relies on field surveys and interpretation of aerial or satellite imagery, which is inefficient and often lacks accuracy, hindering its widespread practical application.

[0003] In recent years, methods for estimating forest understory vegetation structure parameters based on UAV LiDAR point cloud data have been gradually applied in forest resource surveys and ecological applications. UAV LiDAR data can accurately obtain three-dimensional structural information of the forest canopy and its surroundings, including the spatial distribution relationships of canopy and understory vegetation. Furthermore, its high point density helps in the accurate identification and separation of different vegetation layers beneath the tree canopy. Identifying and separating the lower vegetation point cloud can effectively eliminate the influence of the tree canopy point cloud, improving the accuracy of understory vegetation structure parameter estimation.

[0004] Research on estimating forest understory vegetation structure parameters based on point cloud data includes: "Under-canopy UAV laserscanning for accurate forest field measurements," published in Volume 164 of the *ISPRS Journal of Photogrammetry and Remote Sensing* in 2020. This existing technology uses unmanned aerial vehicles (UAVs) to acquire point cloud data of forest understory vegetation, and combines individual tree detection, point cloud clustering, and principal component analysis to estimate structural parameters such as diameter at breast height (DBH) and stem shape curves. "A method for quantifying understory leaf area index in a temperate forest through combining small footprint full-waveform and point cloud lidar data," published in Volume 13 of *Remote Sensing* in 2021. This existing technology uses aircraft to acquire LiDAR point clouds, divides the forest canopy into upper and lower canopies based on vertical canopy porosity, and estimates the leaf area index of the lower canopy accordingly. The paper "Detection of sub-canopy forest structure using airborne LiDAR," published in Volume 244 of *Remote Sensing of Environment* in 2020, utilizes LiDAR point clouds acquired by airborne aircraft and combines them with ground-based measurements of Lorey's average tree height to divide the forest canopy into canopy and sub-canopy layers, subsequently estimating parameters such as understory vegetation volume and number of trees. However, these methods rely on simple, uniform height thresholds for vegetation layer division, failing to consider variations in understory vegetation height, thus hindering accurate identification of the understory vegetation layer. Furthermore, no comprehensive and in-depth analysis of the vertical distribution characteristics of the forest canopy has been found, nor has a method been developed to accurately identify the understory vegetation layer by combining vertical profile distribution functions. Summary of the Invention

[0005] The technical problem solved by this invention is to provide a method that fundamentally enhances the ability to characterize the canopy spatial structure of point cloud data, thereby improving the accuracy of forest understory vegetation layer identification and biomass estimation.

[0006] Technical Solution: To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0007] This invention relates to a method for identifying the understory vegetation layer and estimating its biomass in natural forests based on UAV LiDAR point clouds. Specifically, it involves performing flight strip stitching, denoising, ground point identification, and interpolation to generate a digital terrain model from UAV LiDAR point clouds. Then, based on a three-dimensional voxel framework, the LiDAR point cloud is segmented, and multi-scale profile functions are fitted to the point cloud within each voxel using Gaussian density distribution, multimodal Weibull distribution, and Gamma mixture distribution functions. Finally, an automatic boundary point identification (DPI) method is constructed, combining profile functions and curve shape indices to accurately identify the upper-layer trees and lower-layer vegetation point clouds in natural forests, and a lower-layer vegetation biomass estimation model is constructed based on the features of the lower-layer vegetation point cloud. The method includes the following steps:

[0008] S1: Collect lidar data using a drone platform;

[0009] S2: Remote sensing data preprocessing;

[0010] S3: Construct a forest point cloud vertical profile function model;

[0011] S4: Vertical stratification of forest;

[0012] S5: Extract feature variables from the LiDAR point cloud of the lower vegetation layer;

[0013] S6: Screen characteristic variables through correlation and collinearity analysis;

[0014] S7: Using the measured forest understory biomass as the dependent variable and the selected characteristic variables as independent variables, a multiple linear regression model is established.

[0015] Furthermore, in step S1, survey plots are set up on the ground, and the tree species, height, diameter at breast height (DBH), and ground diameter of the lower vegetation in the plots are recorded. The aboveground biomass of the lower vegetation is calculated according to the allometric growth equation of the corresponding tree species.

[0016] Furthermore, in step S2, the single-strip LiDAR data of the UAV is first matched and stitched together, and noise points in the point cloud are identified and removed based on the neighborhood search algorithm. At the same time, an improved progressive triangulation filtering algorithm is used to identify ground points in the point cloud. By calculating the average height of the laser points in each pixel, a digital elevation model is generated.

[0017] Furthermore, the implementation of step S3 includes the following steps:

[0018] S31: The forest point cloud is divided into several vertical voxels using a voxel framework, and the point cloud in each voxel is height normalized using a DEM.

[0019] S32: Use the Gaussian probability density function to fit the vertical profile of the point cloud in each voxel, and construct the Weibull distribution function with the corresponding number of peaks based on the number of peaks in the vertical profile.

[0020] S33: For the point cloud within the first wavelength range of the Weibull distribution function, construct a fine point cloud vertical profile function model within each voxel using the Gamma mixing distribution function.

[0021] Furthermore, in step S4, the first local minimum point and the last local maximum point in the vertical profile function model of the point cloud within each voxel are used as the boundary points between the understory vegetation layer and the ground, and between the upper trees and the understory vegetation layer, respectively, to perform stratification of the upper trees, lower vegetation and ground into three layers.

[0022] Furthermore, in step S5, for the identified and separated lower vegetation point cloud, n point cloud structure feature variables are extracted, mainly including: canopy height distribution percentile; percentage of points above each percentile height in the total point cloud; average and maximum height; coefficient of variation of canopy point cloud distribution; volume percentage of each canopy structure category, including four canopy structure categories: open layer, light-transmitting layer, low-light layer, and closed layer, and the volume percentage of each canopy structure category.

[0023] Furthermore, in step S5, the control factor k is first obtained by calculating the correlation matrix through principal component analysis as a collinearity index to screen characteristic variables with collinearity less than 30. Then, the characteristic variables with a correlation with the biomass of the lower vegetation layer are screened by correlation analysis.

[0024] Furthermore, in step S6, the control factor k is first obtained by calculating the correlation matrix through principal component analysis as a collinearity index to screen characteristic variables with collinearity less than 30. Then, the characteristic variables with a correlation greater than 0.8 with the underlying vegetation biomass are screened through correlation analysis.

[0025] Furthermore, in step S7, stepwise regression is used to select the variables to enter the model and to select the best model.

[0026] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0027] 1) Previous methods have been based on a single fixed height threshold to divide the forest understory vegetation layer. This invention, however, automatically identifies the forest understory vegetation layer by deeply analyzing the forest canopy point cloud profile and then constructing a boundary point identification method based on profile statistical functions. Because this invention considers the vertical and horizontal distribution variations of forest understory vegetation, it enhances the ability and accuracy of forest understory vegetation layer identification.

[0028] 2) Previous methods did not employ UAV LiDAR technology, which can record in detail the spatial distribution of materials inside the forest canopy. Therefore, this invention can fundamentally increase the ability of point cloud data to characterize the spatial structure of the canopy, thereby improving the accuracy of forest understory vegetation layer identification and biomass estimation.

[0029] 3) Based on the identification and separation of the forest understory vegetation layer, this invention comprehensively and deeply extracts multiple sets of LiDAR point cloud features of the forest understory vegetation layer, and performs feature variable optimization and modeling, thereby estimating the biomass of the forest understory vegetation layer with high accuracy. Furthermore, this invention not only facilitates the identification and separation of the forest understory vegetation layer, but also eliminates interference from the upper canopy trees from a mechanistic perspective, which is beneficial for the analysis and modeling of forest understory vegetation. It is also easy to transfer the method (i.e., it can be applied to different forest types in different regions).

[0030] 4) The verification results show that, by using the present invention to estimate the aboveground biomass of vegetation in natural forests, the relative root mean square error is improved by more than 10% compared with other similar remote sensing methods for estimating aboveground biomass. Attached Figure Description

[0031] Figure 1 This is a regional lidar point cloud data map after decomposition using a voxel framework;

[0032] Figure 2 It is a point cloud image of each voxel after DEM height normalization;

[0033] Figure 3 This is a schematic diagram of the vertical distribution profile function model of point cloud within a volume element;

[0034] Figure 4 This is a schematic diagram of the vertical stratification of a forest;

[0035] Figure 5 It is a point cloud map of trees, bamboo and ground surface identified by vertical layering within the voxel;

[0036] Figure 6 This is a rendering of the forest canopy layer, bamboo layer, and ground layer in the area identified by this invention;

[0037] Figure 7 This is a correlation diagram between the characteristic variables and structural parameters of the bamboo dot cloud under the forest in this invention. Detailed Implementation

[0038] The present invention will be further illustrated below with reference to specific embodiments. These embodiments are implemented based on the technical solutions of the present invention, and it should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0039] Example 1

[0040] The present invention provides a method for identifying natural forest understory vegetation and estimating its biomass. This method involves stitching flight strips, denoising, identifying ground points, and interpolating UAV LiDAR point clouds to generate digital terrain models. The method segments the LiDAR point clouds based on a three-dimensional voxel frame and constructs a multi-scale profile function model of the point cloud within the voxel by combining Gaussian density distribution, multi-peak Weibull distribution, and Gamma mixture distribution function.

[0041] A boundary point automatic identification (DPI) method is constructed, which combines profile model and curve shape index to accurately identify the point clouds of upper trees and lower vegetation in natural forests, and constructs a lower vegetation biomass estimation model based on the features of the lower vegetation point cloud.

[0042] Specifically, steps S1 to S7 are included:

[0043] S1: Collect lidar data using a drone platform.

[0044] Survey plots were set up on the ground, and the tree species, height, diameter at breast height (DBH), and ground diameter of the lower vegetation in the plots were recorded. The aboveground biomass of the lower vegetation was calculated according to the allometric growth equation of the corresponding tree species.

[0045] S2: Remote sensing data preprocessing.

[0046] Existing technologies were used for remote sensing data preprocessing. First, strip matching and stitching were performed on the single-strip LiDAR data from the UAV, and noise points in the point cloud were identified and removed based on a neighborhood search algorithm (Liu et al., 2018). At the same time, an improved progressive triangulation filtering algorithm was used to identify ground points in the point cloud, and a digital elevation model (DEM) (spatial resolution of 0.1m) was generated by calculating the average height of the laser points in each pixel.

[0047] S3: Construct a forest point cloud vertical profile function model, the details of which are as follows:

[0048] S31: First, the forest point cloud is divided into several vertical voxels using a voxel framework, and the point cloud in each voxel is height normalized using a DEM with a spatial resolution of 0.1m.

[0049] S32: Then, the Gaussian probability density function is used to fit the vertical profile of the point cloud in each voxel, and the Weibull distribution function with the corresponding number of peaks is constructed according to the number of peaks in the vertical profile.

[0050] The Gaussian probability density function is:

[0051]

[0052] In the formula, h is a highly random variable, and α and β are the mean and standard deviation, respectively.

[0053] The multimodal Weibull distribution function is:

[0054]

[0055] In the formula, h is a highly random variable, and α and β are the shape parameter and the scaling parameter, respectively.

[0056] S33: Finally, for the point cloud within the first wavelength range of the Weibull distribution function, a detailed vertical profile function model of the point cloud within each voxel is constructed using the Gamma mixing distribution function:

[0057]

[0058] In the formula, h is a highly random variable, and α and β are both shape parameters.

[0059] S4: Forest vertical stratification, details are as follows:

[0060] The first local minimum and the last local maximum point in the vertical profile function model of the point cloud within each voxel are used as the boundary points between the understory vegetation layer and the ground, and between the upper trees and the understory vegetation layer, respectively, to perform stratification of the upper trees, lower vegetation, and ground. This enables the identification and separation of the upper trees and understory vegetation layers throughout the entire area.

[0061] S5: Extract feature variables from the LiDAR point cloud of the lower vegetation layer.

[0062] For the identified and separated lower vegetation point cloud, 16 point cloud structural feature variables were extracted, mainly including: canopy height distribution percentile (25%). th 50 th 75 th 95 th Point cloud quantity at each percentage height 10 th 30 th 50 th 70 th 90 th The percentage of points above d1, d3, d5, d7, d9 in the total point cloud; the average and maximum heights (h) mean ,h max ); Coefficient of variation (H) of canopy point cloud distribution cv ); the volume percentage of each canopy structure category, including the four canopy structure categories of open layer, light-transmitting layer, low-light layer and closed layer, and the percentage of volume occupied by each canopy structure category (i.e., O); g E,C g ,O).

[0063] S6: Screen characteristic variables through correlation and collinearity analysis.

[0064] First, principal component analysis is used to calculate the correlation matrix to obtain the control factor k (i.e., the ratio of the square root of the largest eigenvalue to the smallest eigenvalue) as the collinearity index, so as to screen eigenvariates with collinearity less than 30. Then, the correlation between eigenvariates and the biomass of the underlying vegetation is screened by correlation analysis.

[0065] S7: Using the measured forest understory biomass as the dependent variable and the selected characteristic variables as independent variables, a multiple linear regression model is established.

[0066] The stepwise regression method is used to select the variables to enter the model, and the Akaike Information Criterion (AIC) is used to select the best model, that is, the model with the lowest AIC is the best model.

[0067] This invention uses the coefficient of determination (R²) 2 The root mean square error (RMSE) and relative root mean square error (rRMSE) are used to evaluate the fit of the regression model and the accuracy of the estimation.

[0068]

[0069]

[0070] In the formula, x i This represents the measured value of aboveground biomass of forest understory vegetation. This represents the measured average aboveground biomass of forest understory vegetation. The model estimate represents the aboveground biomass of forest understory vegetation; n represents the number of sample plots; and i represents a specific sample plot.

[0071] Example 2

[0072] The study area is located in the Foping Giant Panda National Nature Reserve in Shaanxi Province (107°47′39.59″E, 33°38′53.55″N), covering an area of ​​approximately 290 km². 2The elevation ranges from 980 to 2900 meters. The forest type in the study area is natural forest, which can be further subdivided into two stands: one dominated by coniferous species and the other by broad-leaved species. The main coniferous and broad-leaved deciduous species include Pinus tabuliformis, Tsuga chinensis, Quercus variabilis, and Quercus aliena var. acutiserrata. The understory vegetation mainly consists of Fargesia qinlingensis and Arundinaria fargesii. Fargesia qinlingensis is mainly distributed above 1700 meters in altitude, while Arundinaria fargesii is mainly distributed below 1900 meters. These two types of bamboo, as a major food source for giant pandas, together with the upper trees, constitute a complex natural forest ecosystem.

[0073] Fifty-two square plots (20×20m) were set up within the study area. 2 (This refers to a survey conducted on bamboo, specifically arrow bamboo, in the Qinling Mountains, with 22 plots and 30 plots of woody bamboo.) The coordinates of the plot center points were determined using differential GPS, which receives Qianxun differential signals for positioning, achieving an accuracy better than 0.5m. The species, height, diameter at breast height (DBH), and ground diameter of the understory vegetation in each plot were recorded, and the aboveground biomass of the understory vegetation was calculated based on the allometric growth equations for the corresponding tree species. The plot-scale understory bamboo structure parameters summarized from the ground survey are shown in Table 1.

[0074] Table 1 Summary of measured forest understory vegetation characteristics in the sample plots

[0075]

[0076] LiDAR data was collected using a drone platform. First, the single-strip LiDAR data from the drone was strip-matched and stitched together, and noise points in the point cloud were identified and removed based on a neighborhood search algorithm. Simultaneously, an improved progressive triangulation filtering algorithm was used to identify ground points in the point cloud. By calculating the average height of the laser points within each pixel, a digital elevation model (DEM) (spatial resolution of 0.1m) was generated.

[0077] A vertical profile function model of forest point clouds was constructed based on Gaussian probability density function, multimodal Weibull distribution function, and Gamma mixture distribution function. First, a voxel framework was used to segment the forest point cloud into several vertical voxels, and the point cloud within each voxel was height-normalized using a DEM with a spatial resolution of 0.1m. Then, a vertical profile was fitted to the point cloud within each voxel, and a Weibull distribution function with the corresponding number of peaks was constructed based on the number of peaks in the vertical profile. Finally, for the point cloud within the first wavelength range of the Weibull distribution function, a refined vertical profile function model of the point cloud within each voxel was constructed using the Gamma mixture distribution function. The first local minimum point and the last local maximum point in the vertical profile function model of the point cloud within each voxel were used as the boundary points between the understory vegetation layer and the ground, and between the upper trees and the understory vegetation layer, respectively, to perform three-layer stratification of the upper trees, lower vegetation, and ground. This enables the identification and separation of the upper trees and understory vegetation layers throughout the entire area.

[0078] For the identified and separated lower vegetation point cloud, 16 point cloud structural feature variables were extracted, mainly including: canopy height distribution percentile (25%). th 50 th 75 th 95 th Point cloud quantity at each percentage height 10 th 30 th 50 th 70 th 90 th The percentage of points above d1, d3, d5, d7, d9 in the total point cloud; the average and maximum heights (h) mean ,h max ); Coefficient of variation (H) of canopy point cloud distribution cv ); the volume percentage of each canopy structure category, including the four canopy structure categories of open layer, light-transmitting layer, low-light layer and closed layer, and the percentage of volume occupied by each canopy structure category (i.e., O); g E,C g ,O).

[0079] Feature variables were screened using correlation and collinearity analysis. First, principal component analysis was used to calculate the correlation matrix to obtain the control factor k (i.e., the ratio of the square root of the largest eigenvalue to the smallest eigenvalue) as a collinearity index to screen feature variables with collinearity below 30. Then, correlation analysis was used to screen feature variables with a correlation higher than 0.8 with the biomass of the underlying vegetation.

[0080] Using measured forest understory biomass as the dependent variable and selected characteristic variables as independent variables, a multiple linear regression model was established. Stepwise regression was used to select variables for the model, and the Akaike Information Criterion (AIC) was used to select the optimal model; that is, the model with the lowest AIC was the best model. This invention uses the coefficient of determination (R²). 2 The root mean square error (RMSE) and relative root mean square error (rRMSE) are used to evaluate the fit of the regression model and the accuracy of the estimation.

[0081]

[0082]

[0083] In the formula, x i This represents the measured value of aboveground biomass of forest understory vegetation. This represents the measured average aboveground biomass of forest understory vegetation. The model estimate represents the aboveground biomass of forest understory vegetation; n represents the number of sample plots; and i represents a specific sample plot.

[0084] Table 2 shows the aboveground biomass estimation model and prediction accuracy of bamboo understory. The correlations between various characteristic variables and structural parameters of bamboo understory point clouds are shown in Table 2. Figure 2 .

[0085] Table 2. Forest understory aboveground biomass estimation model and model prediction accuracy

[0086] aboveground biomass <![CDATA[exp(-0.52+3.34lnh mean +0.42lnE)×1.081]]> <![CDATA[0.85 *** ]]> 8.63 mg / ha 29.41 3.43

[0087] Note: h mean E represents the average height of the bamboo spot clouds under the forest cover within the sample plot, and E represents the percentage of the light-transmitting space volume within the canopy of the bamboo spot clouds under the forest cover.

[0088] The verification results show that, compared with other similar remote sensing methods, the determination coefficient error of estimating the aboveground biomass of natural forest understory using the present invention is improved by 10% or more.

[0089] Table 3 Comparison of the accuracy of aboveground biomass estimation using different methods

[0090]

[0091] This invention can be applied to fields such as fine-scale forest resource surveys and forest carbon sequestration measurement. It is beneficial to gain a deeper understanding of the relationship between vegetation growth and spatial distribution within forests and the environment, and to further grasp the growth and competition patterns of different vegetation layers in the vertical direction within the forest ecosystem. It is of great significance for the sustainable management of forests, community succession analysis, and research on carbon sequestration processes in the entire ecosystem.

[0092] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying and estimating the biomass of a natural understory vegetation layer, characterized by: This study analyzes forest canopy point cloud profiles, constructs boundary points based on profile statistical functions, and automatically identifies the understory vegetation layer. Multiple sets of LiDAR point cloud features of the understory vegetation layer are extracted, and feature variables are optimized and modeled to accurately estimate the understory vegetation biomass. The steps include: S1: Collect lidar data using a drone platform; S2: Remote sensing data preprocessing; S3: Construct a forest point cloud vertical profile function model; including the following steps: S31: The forest point cloud is divided into several vertical voxels using a voxel framework, and the point cloud in each voxel is height normalized using a DEM. S32: Use the Gaussian probability density function to fit the vertical profile of the point cloud in each voxel, and construct the Weibull distribution function with the corresponding number of peaks based on the number of peaks in the vertical profile. S33: For the point cloud within the first wavelength range of the Weibull distribution function, construct a fine point cloud vertical profile function model within each voxel using the Gamma mixing distribution function; S4: Vertical stratification of forest; The first local minimum point and the last local maximum point in the vertical profile function model of the point cloud within each voxel are used as the boundary points between the understory vegetation layer and the ground, and between the upper trees and the understory vegetation layer, respectively, to perform stratification of the upper trees, lower vegetation and ground into three layers. S5: Extract feature variables from the LiDAR point cloud of the lower vegetation layer; S6: Screen characteristic variables through correlation and collinearity analysis; S7: Using the measured forest understory biomass as the dependent variable and the selected characteristic variables as independent variables, a multiple linear regression model is established.

2. The method of identifying and estimating the biomass of the natural understory vegetation layer according to claim 1, characterized in that: In step S1, survey plots are set up on the ground, and the tree species, height, diameter at breast height (DBH), and ground diameter of the lower vegetation in the plots are recorded. The aboveground biomass of the lower vegetation is calculated according to the allometric growth equation of the corresponding tree species.

3. The method for identifying natural forest understory vegetation and estimating its biomass according to claim 2, characterized in that: In step S2, the single-strip LiDAR data of the UAV is first matched and stitched together, and noise points in the point cloud are identified and removed based on the neighborhood search algorithm. At the same time, ground points in the point cloud are identified, and a digital elevation model is generated by calculating the average height of the laser points in each pixel.

4. The method for identifying natural forest understory vegetation and estimating its biomass according to claim 1, characterized in that: In step S5, for the identified and separated lower vegetation point cloud, n point cloud structure feature variables are extracted, including: canopy height distribution percentile; percentage of points above each percentile height in the total point cloud; average and maximum height; coefficient of variation of canopy point cloud distribution; volume percentage of each canopy structure category, including four canopy structure categories: open layer, light-transmitting layer, low-light layer, and closed layer, and the volume percentage of each canopy structure category.

5. The method for identifying natural forest understory vegetation and estimating its biomass according to claim 1, characterized in that: In step S6, the control factors are first obtained by calculating the correlation matrix through principal component analysis. k As a collinearity index, feature variables with collinearity below 30 are screened, and then feature variables with a correlation higher than 0.8 with the underlying vegetation biomass are screened by correlation.

6. The method for identifying natural forest understory vegetation and estimating its biomass according to claim 1, characterized in that: In step S7, the stepwise regression method is used to select the variables to enter the model and select the best model.

Images