Grassland above-ground biomass estimation method and system based on coupling of semi-variance function and grazing heterogeneity

Abstract

The invention discloses a grassland above-ground biomass estimation method and system based on semi-variance function coupling grazing heterogeneity, and belongs to the technical field of grassland feature estimation, and the method comprises the steps: obtaining a fixed resolution canopy height model CHM through the processing of collected grassland point cloud data, obtaining CGF and LAI raster data through the CHM, and carrying out the calculation of the grassland above-ground biomass. Fitting a semi-variance function model to obtain spatial heterogeneity characteristics of the grassland, obtaining an optimal inversion scale, screening out key independent variables with stability, and providing a reliable basis for grassland AGB inversion; through combination of CHM, CGF and LAI of different spatial scales and ground actually measured biomass data, inversion is carried out by using a linear regression model, a random forest and a BP neural network to obtain an optimal variable combination and an inversion model, and finally spatial heterogeneity characteristics of gAGB of different spatial scales in different management modes are analyzed, management measures based on local conditions are formulated, and the method is suitable for mass production. The grassland resource utilization efficiency is improved; and the ecological system is protected.

Description

Technical Field

[0001] The present invention relates to the technical field of grassland feature estimation, and more specifically, to a method and system for estimating aboveground biomass of grassland based on semi-variance function coupling grazing heterogeneity. Background Technique

[0002] Grassland is one of the largest terrestrial ecosystems on earth, with important ecological service functions such as regulating climate, conserving water sources, preventing wind and sand, and carbon sequestration. Aboveground biomass (AGB) of grassland is a key indicator for measuring grassland productivity and ecosystem function, and is of great significance for scientific management and sustainable utilization of grassland resources. However, due to the increase in grazing intensity and over-exploitation of grassland resources, the problem of grassland degradation is becoming increasingly serious, and it is particularly important to accurately estimate the aboveground biomass of grassland to achieve grass-livestock balance.

[0003] Traditional methods mainly rely on high-resolution remote sensing data for biomass estimation, but the spatial resolution of these data often has limitations. Although high-resolution remote sensing data can provide a certain degree of ground cover information, its resolution is often insufficient to accurately capture the subtle changes in the grass layer, such as the height difference of grass clumps, local differences in vegetation distribution, and specific areas affected by grazing. This limitation of spatial resolution may lead to the neglect of some key environmental features when evaluating the actual biomass of grazing grasslands, thus affecting the accuracy and reliability of the estimation.

[0004] In addition, existing technologies often neglect the spatial heterogeneity within the grassland when estimating the biomass of grazing grasslands. There may be significant differences in biomass and ecological characteristics in different regions within the grassland, while traditional methods usually rely on the overall NDVI value or a single biomass model for estimation, which results in the neglect of micro-environmental differences, and may thus underestimate the biomass in some regions or overestimate the carrying capacity in other regions. This neglect of spatial heterogeneity not only reduces the accuracy of the estimation, but may also have an adverse impact on the rational utilization and protection of grassland resources.

[0005] More critically, when selecting the inversion scale in existing technologies, it often relies on empirical or trial-and-error methods. Researchers may choose a widely used or verified fixed scale based on past experience, or screen the optimal scale by trying multiple spatial scales and comparing the model results. This method is not only highly subjective and lacks scientific basis, but also the screening process is blind and inefficient. More importantly, it neglects the spatial heterogeneity characteristics of variables and cannot reveal the specific impact of scale changes on variable characteristics and model performance. This uncertainty in scale selection further increases the error and uncertainty of biomass estimation.

[0006] In view of the above problems, the present invention proposes a method for estimating aboveground biomass of grassland based on the coupling of semivariogram function and grazing heterogeneity. This method aims to improve the accuracy and reliability of biomass estimation by combining the semivariogram function to analyze the spatial autocorrelation of grassland biomass and considering the impact of grazing activities on grassland heterogeneity. This method not only overcomes the limitations of the spatial resolution of traditional remote sensing data but also fully considers the spatial heterogeneity within the grassland, providing a new idea and technical means for the accurate estimation of aboveground biomass of grazing grassland.

[0007] Therefore, how to provide a method and system for accurately estimating the aboveground biomass of grassland is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0008] In view of this, the present invention provides a method and system for estimating aboveground biomass of grassland based on the coupling of semivariogram function and grazing heterogeneity, which improves the accuracy and reliability of biomass estimation by combining the semivariogram function to analyze the spatial autocorrelation of grassland biomass and considering the impact of grazing activities on grassland heterogeneity.

[0009] To achieve the above object, the present invention adopts the following technical solutions:

[0010] On the one hand, the present invention provides a method for estimating aboveground biomass of grassland based on the coupling of semivariogram function and grazing heterogeneity, including:

[0011] Obtain the point cloud data of the target grassland and perform preprocessing;

[0012] Construct grassland canopy height model data at different spatial scales based on the preprocessed point cloud data;

[0013] Based on the grassland canopy height model data, respectively obtain the canopy structure parameters of the target grassland, and the canopy structure parameters include: canopy height, grassland canopy gap fraction, and leaf area index;

[0014] Use the semivariogram function to calculate the semivariance values of the grassland canopy structure parameters at each spatial scale respectively, construct fitting models by combining the exponential model, Gaussian model, and spherical model based on the semivariance values respectively, and obtain the optimal fitting model and fitting parameters through accuracy verification;

[0015] Analyze the spatial heterogeneity characteristics of the grassland canopy structure parameters at each spatial scale based on the optimal fitting model and fitting parameters, and obtain the first optimal inversion scale and the first most stable canopy structure parameter according to the spatial heterogeneity characteristics;

[0016] Respectively obtain the measured data of the aboveground biomass of the target grassland;

[0017] Taking the canopy height, grassland canopy gap ratio, and leaf area index as independent variables and the measured data as the dependent variable, inversion models based on linear regression models, random forest models, and BP neural network models are constructed respectively;

[0018] The inversion models of the constructed linear regression model, random forest model, and BP neural network model are verified for accuracy to obtain the optimal inversion model, the second-best inversion scale, and the second-most stable canopy structure parameters;

[0019] According to the second-best inversion scale and the second-most stable canopy structure parameters, the accuracy of the first-best inversion scale and the first-most stable canopy structure parameters is verified;

[0020] The optimal inversion model inverses the measured data to obtain the inversion result of the above-ground biomass of the target grassland;

[0021] The spatial heterogeneity of the inversion result of the above-ground biomass of the target grassland is analyzed to obtain the most sensitive spatial scale of the grassland;

[0022] Based on the inversion result of the above-ground biomass of the target grassland and the most sensitive spatial scale of the grassland, optimized management measures for the sensitive spatial scale of the grassland are determined.

[0023] Preferably, the point cloud data is preprocessed, including: point cloud coordinate transformation, flight strip splicing, data block division, denoising processing, and filtering the denoised point cloud data based on ground point filtering to separate non-ground points and ground points.

[0024] Preferably, grassland canopy height model data at different spatial scales is constructed based on the preprocessed point cloud data, including:

[0025] Select the minimum value of multiple echoes within the ground points as the surface position, and use inverse distance weighting to interpolate the discrete points in the ground points to obtain digital elevation model data;

[0026] Select the maximum value of multiple echo points in the non-ground data unit as the pixel value, and use the inverse distance weighting method to interpolate the discrete points in the non-ground points to obtain digital surface model data;

[0027] Based on the digital elevation model data and the digital surface model data, canopy height model data is obtained.

[0028] Preferably, based on the grassland canopy height model data, the canopy structure parameters of the target grassland are obtained respectively, including:

[0029] Calculate the canopy height using the grassland canopy height model data;

[0030] Calculate the vegetation canopy density based on the two-dimensional grassland canopy height model data, and obtain the grassland canopy gap ratio based on the vegetation canopy density:

[0031]

[0032] Among them, ∑pixe[i,j] veg is the number of vegetation pixels in the window, and ∑pixe[i,j] is the total number of pixels in the window;

[0033] Calculate the leaf area index within the window based on the Beer-Lambert law:

[0034]

[0035] Among them, Ω(θ) is the aggregation index, θ is the zenith angle of the incident light, G(θ) is the extinction coefficient, P(θ) is the gap probability of observing the zenith angle θ, and cosθ is the optical path length of the homogeneous vegetation layer considering the random distribution of leaves.

[0036] Preferably, use the semivariogram function to calculate the semivariance values of the grassland canopy structure parameters at each spatial scale, and perform fitting based on the semivariance values in combination with the exponential model, Gaussian model, and spherical model. Obtain the optimal fitting model and fitting parameters through accuracy verification, including:

[0037] Use the semivariogram function to calculate the semivariance values of the grassland canopy structure parameters at each spatial scale as:

[0038]

[0039] Among them, h is the distance, representing the interval between two locations, N(h) is the number of point pairs within the distance h, Z(x i ) is the vegetation parameter data value at location x i ; Z(x i +h) is the vegetation parameter data value at location x i +h at a distance h;

[0040] Construct fitting models based on the semivariance values in combination with the exponential model, Gaussian model, and spherical model respectively:

[0041]

[0042] Among them, γ Exp (d) is the exponential model, γ Gau (d) is the Gaussian model, γ Sph (d) is the spherical model, C 0 is the nugget effect; C is the partial sill; C 0 +C is the sill; a is the range; d is the distance of the observation point in spatial analysis.

[0043] Preferably, taking the canopy height, grassland canopy gap ratio, and leaf area index as independent variables and the measured data as the dependent variable, inversion models based on linear regression models, random forest models, and BP neural network models are constructed respectively:

[0044] The inversion model constructed based on the linear regression model is:

[0045] y = a 1 x 1 + a 2 x 2 + a 3 x 3 + … + a k x k + a 0

[0046] a 1 、a 2 、a 3 … a k are the regression coefficients of the LRM fitting equation, a 0 is the constant term of the fitting equation, x 1 , x 2 , x 3 … x k are independent variables;

[0047] When using a random forest to invert the aboveground biomass of the target grassland, samples are randomly selected to form each decision tree, and then the results of all decision trees are averaged to obtain the inversion result of the aboveground biomass of the target grassland. The inversion model constructed based on the random forest model is:

[0048]

[0049] where T i (x) represents the individual prediction result of the i-th decision tree for the sample x, and m is the total number of decision trees in the random forest;

[0050] A BP neural network consists of an input layer, an output layer, and several hidden layers, and a multi-layer feedforward neural network is trained through forward propagation and backpropagation algorithms; in the forward propagation stage, the input data P 1 , P 2 , …, P n pass through the weights ω 1 , ω 2 , …, ω nPerforms a weighted sum of the weights \(w\) and the bias value \(b\) of the neurons, and performs a non - linear transformation through the activation function \(\sigma\), transmitting the signal from the input layer to the hidden layer and then to the output layer; compares the biomass of the output layer with the expected value. If the error does not meet the requirements, adjusts the weights and biases of each layer through the back - propagation algorithm until the error reaches the expected value or the set number of training times.

[0051] Preferably, obtaining the optimal fitting model and the optimal inversion model through accuracy verification includes: respectively using the coefficient of determination, the sum of squared residuals, and the root mean square error to evaluate and test the model accuracy.

[0052] On the other hand, the present invention provides a grassland above - ground biomass estimation system based on the coupling of the semivariogram and grazing heterogeneity, including:

[0053] A data acquisition unit for acquiring the point cloud data of the target grassland and performing pre - processing;

[0054] A canopy structure parameter calculation unit for constructing grassland canopy height model data at different spatial scales based on the pre - processed point cloud data, and respectively obtaining the canopy structure parameters of the target grassland based on the grassland canopy height model data. The canopy structure parameters include: canopy height, grassland canopy gap ratio, and leaf area index;

[0055] A fitting unit for respectively calculating the semivariance values of the grassland canopy structure parameters at each spatial scale using the semivariogram, constructing fitting models by respectively combining the exponential model, the Gaussian model, and the spherical model based on the semivariance values, and obtaining the optimal fitting model and fitting parameters through accuracy verification;

[0056] A parameter screening unit for analyzing the spatial heterogeneity characteristics of the grassland canopy structure parameters at each spatial scale based on the optimal fitting model and fitting parameters, and obtaining the first optimal inversion scale and the first most stable canopy structure parameters according to the spatial heterogeneity characteristics;

[0057] A measurement unit for respectively obtaining the measured data of the above - ground biomass of the target grassland;

[0058] A model construction unit for respectively constructing inversion models based on linear regression models, random forest models, and BP neural network models, with the canopy height, grassland canopy gap ratio, and leaf area index as independent variables and the measured data as the dependent variable;

[0059] A model selection unit for performing accuracy verification on the constructed inversion models of the linear regression model, random forest model, and BP neural network model, and obtaining the optimal inversion model, the second best inversion scale, and the second most stable canopy structure parameters;

[0060] A verification unit, configured to verify the accuracy of the first optimal inversion scale and the first most stable canopy structure parameters according to the second optimal inversion scale and the second most stable canopy structure parameters;

[0061] An inversion unit, configured to perform inversion on the measured data by using an optimal inversion model to obtain an inversion result of the aboveground biomass of the target grassland;

[0062] An analysis unit, configured to perform spatial heterogeneity analysis on the inversion result of the aboveground biomass of the target grassland to obtain the most sensitive spatial scale of the grassland;

[0063] A management unit, configured to determine optimized management measures for the sensitive spatial scale of the grassland based on the inversion result of the aboveground biomass of the target grassland and the most sensitive spatial scale of the grassland.

[0064] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method and a system for estimating the aboveground biomass of a grassland based on a semi-variance function coupling grazing heterogeneity. The present invention uses lidar to capture three-dimensional structure information of the grassland vegetation canopy in the enclosed grazing and free grazing coupling grassland, overcoming the deficiency of traditional optical remote sensing in lacking three-dimensional structure information of the community when obtaining surface vegetation coverage. By setting sampling steps at different spatial scales, canopy height grid data at the centimeter-level spatial scale is obtained, and by setting different grass height thresholds and window sizes, canopy gap ratio grid data at the corresponding scale is generated. Then, according to the Beer-Lambert law, the canopy gap ratio grid data of the grassland is brought into the model formula to generate the leaf area index grid data of the grazing grassland at the corresponding scale. The exponential model, spherical model, and Gaussian model of the semi-variance function are used to fit and analyze the spatial heterogeneity characteristics of the canopy structure parameters of the enclosed and grazing management grasslands at each scale, obtain the optimal inversion scale in these two scenarios, and screen out the most stable canopy structure parameters. Then, a model of machine learning is used to construct the aboveground biomass of the grazing grassland at different scales to obtain a modeling result, verify the conclusion that spatial heterogeneity analysis helps to improve the modeling accuracy, and obtain the optimal model and variable combination. Then, spatial heterogeneity analysis is performed on the biomass inversion results of different management scenarios at each scale to reveal the impact of different management methods on the biomass inversion results and the grassland ecosystem, providing scientific support for optimizing management measures, evaluating ecological functions, etc. Finally, it can be found that when the optimal inversion scale and the most stable variables obtained through the spatial heterogeneity analysis are used for the aboveground biomass of the grassland in the enclosed and grazing scenarios, the modeling accuracy is the highest, and when the worst inversion scale and unstable variable combination are used, the modeling accuracy is the worst. Description of the Drawings

[0065] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the provided drawings.

[0066] Figure 1 It is a flowchart of the evaluation method of the present invention;

[0067] Figure 2 It is a structural schematic diagram provided by the present invention. Detailed implementation manners

[0068] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0069] The aboveground biomass of grazing grassland (grasslandAbovegroundBiomass, gAGB) is an important indicator reflecting the growth and development of vegetation, and serves as basic data for research such as grassland-livestock balance, grassland carbon cycle, grassland biodiversity, and grassland ecosystem services.

[0070] The Canopy Height Model (CHM) is usually the difference between the vegetation canopy height and the aboveground surface data of the soil, and is a basic indicator for measuring the vertical structure of vegetation, which helps to understand and evaluate the state and evolution of the grazing grassland ecosystem. The CHM model of the grassland canopy designed in the present invention is obtained by subtracting the Digital Elevation Model (DEM) from the Digital Surface Model (DSM) established by Unmanned Aerial Vehicle Light Detection And Ranging (LiDAR) data. In particular, the Unmanned Aerial Vehicle LiDAR technology used in the research can obtain large-scale continuous high-precision and high-density three-dimensional point cloud data of the grazing grassland, which helps to obtain high-resolution canopy height model raster data of the grazing grassland in this experiment.

[0071] The Canopy Gap Fraction (CGF) refers to the probability that a beam of light shines on the grass canopy along a certain direction without being blocked by the grass canopy. It describes the proportion of the space in the grass canopy that is not covered by plant bodies, reflecting the density and structural characteristics of grass vegetation cover. The grass canopy gap fraction CGF designed in the present invention first calculates the canopy closure using two-dimensional CHM raster data. By counting the number of CHM raster data pixels higher than the grass height threshold within a custom window, the number of grass vegetation and the total number of window pixels are obtained to obtain the canopy closure. Since the sum of the canopy closure and the gap fraction is 1, the grass canopy gap fraction CGF raster data can be obtained.

[0072] The Leaf Area Index (LAI) is an important state variable of the grassland ecosystem and is usually used to describe the density, growth status, and ecological functions of vegetation. The present invention designs and generates the grassland leaf area index LAI raster data based on the Beer-Lambert law through the grass canopy gap fraction CGF raster data.

[0073] Spatial heterogeneity refers to the non-uniformity or difference in a certain attribute or phenomenon (such as vegetation, biomass, soil properties, etc.) at different positions in the geographical space. It reflects the complexity and diversity of the spatial distribution of ecosystems, landscapes, or other geographical entities under the action of natural and human activities. Commonly used measurement methods include statistical methods (standard deviation, coefficient of variation, and spatial autocorrelation), geostatistical methods (i.e., semivariogram function), landscape metrics (patch density, boundary length, etc.), and spatial analysis models (spatial regression, random forest, etc.). The present invention uses the semivariogram function in geostatistics to calculate the semivariance values of the grass canopy structure parameters (CHM, CGF, and LAI) at different spatial scales, and obtains the corresponding fitting parameters by fitting exponential models, Gaussian models, and spherical models. Based on the model fitting results, the spatial heterogeneity characteristics of the grass canopy structure parameters at different spatial scales are analyzed to determine the optimal spatial scale suitable for inversion, and key independent variables with stability are screened out, promoting the improvement of the grassland biomass inversion accuracy and expanding its application potential in grassland ecosystem management.

[0074] The method for estimating aboveground biomass of grassland based on semi-variance function coupling grazing heterogeneity disclosed by the present invention has important theoretical and practical significance in obtaining high-precision canopy structure parameters of fenced and unfenced grazing coupled grassland using unmanned aerial vehicle lidar technology and obtaining high-precision aboveground biomass of grazing grassland using spatial heterogeneity. By processing the unmanned aerial vehicle LiDAR point cloud data, a large number of echo points of data units are obtained, and low-value data and high-value data in the data units are respectively selected. The inverse distance weighted (IDW) method is used to obtain the continuously varying regular grid digital elevation model (DEM) data and digital surface model (DSM). The difference operation is performed between the two, and the canopy height model (CHM) with a fixed resolution is obtained through coordinate matching. The CGF is obtained based on the two-dimensional CHM raster data. Then, based on the CGF, the LAI raster data is obtained using the Beer-Lambert law. The semi-variance function model is fitted to analyze the spatial heterogeneity characteristics of each vegetation parameter under different grazing management modes, obtain the optimal inversion scale, and screen out the key independent variables with stability, providing a reliable basis for the subsequent inversion of grassland AGB. To obtain the optimal variable combination and inversion model, the study combines CHM, CGF, and LAI at different spatial scales with the ground measured biomass data, and uses the linear regression model (LRM), random forest (RF), and backpropagation neural network (BPNN) to perform gAGB inversion. Finally, the spatial heterogeneity characteristics of gAGB at different spatial scales under different management modes are analyzed, and management measures are formulated according to local conditions to improve the utilization efficiency of grassland resources and protect the ecosystem.

[0075] An embodiment of the present invention discloses a method for estimating aboveground biomass of grassland based on semi-variance function coupling grazing heterogeneity, as Figure 1 shown, including:

[0076] Obtain the point cloud data of the target grassland and perform preprocessing. Specifically, the unmanned aerial vehicle Lidar is used to emit laser pulses to the grazing grassland in the study area, which are captured by the receiver of the lidar device. The distance is measured by measuring the time of the laser beam reflected from the target. A large number of single points are generated during the scanning process, and each point represents the position where the laser contacts the object. The data contains three-dimensional coordinates (longitude, latitude, and elevation) and other attributes such as intensity, forming point cloud data and storing it in the storage format las1.2 of standard LiDAR. Python 3.9 is used to implement the preprocessing of the las1.2 data, including point cloud coordinate conversion, strip stitching, and data chunking, etc. Then, the abnormal points are removed through denoising, and the denoised point cloud data is filtered and classified based on the CSF ground point filtering to separate the non-ground points and ground points. This point cloud data is used as the original point cloud data.

[0077] LiDAR is a remote sensing technology that obtains surface information by emitting laser pulses and measuring their return time. The LiDAR device emits short laser pulses into the target area (usually the ground or vegetation). The laser pulses travel through the air and encounter objects (such as trees, buildings, or the ground). Some of the laser pulses are reflected back, and the LiDAR device receives these returned laser signals. By measuring the time difference between the emission and reception of the laser pulses, the distance traveled by the laser is calculated. The basic principle of laser ranging can be expressed by the following formula:

[0078]

[0079] Where: D is the target distance in meters, c is the speed of light, approximately 3×10^8 m / s, and t is the round-trip time of the laser pulse in seconds.

[0080] The equation for the intensity of the reflected signal is as follows:

[0081] I = I 0 ×R×e -kD #(2

[0082] Where I is the intensity of the received signal, which is the intensity of the laser signal reflected from the target object by the LiDAR device. I 0 is the intensity of the emitted signal, which is the intensity of the laser pulse initially emitted by the laser device. R is the reflectivity, representing the ability of the target object to reflect the laser. The reflectivity is usually related to the material and surface characteristics of the object. A higher reflectivity means the object can reflect more laser signals. k is the attenuation coefficient, indicating the energy loss of the laser during propagation due to absorption, scattering, etc. by air or other media. It may be related to the properties of the medium (such as density, humidity, etc.). D is the distance traveled by the laser, that is, the distance from the emission source to the target object. This formula describes that as the distance D increases, the intensity I of the received signal will exponentially decrease due to the presence of the attenuation coefficient k. It also shows the important influence of the reflectivity R of the target object on the intensity of the received signal. If the reflectivity of the target object is high, even at a relatively long distance, a relatively strong signal may still be received; conversely, if the reflectivity is low or the distance is too far, the signal attenuation will be very obvious.

[0083] In the embodiments of the present invention, the lidar device used is the DJI Zenmuse L2, which has good reflected signal intensity and strong penetration ability especially in complex environments. Its maximum detection range can still maintain effective data acquisition under low reflectivity conditions (250 meters at 10% reflectivity). This means that clear signal echoes can be obtained even in grassland environments with more vegetation or lower reflectivity. For the acquisition of the three-dimensional point cloud structure of grasslands, the high echo acquisition rate of the Zenmuse L2 (up to 1,200,000 points / second in the multi-echo mode) and the high-precision inertial navigation system (with a pitch accuracy of 0.025° after post-processing) enable it to generate high-precision three-dimensional point cloud data. Moreover, it supports five echo types and can capture different levels of ground and vegetation information, making it very suitable for multi-layered surfaces like grasslands. Therefore, the Zenmuse L2 performs excellently in the acquisition of three-dimensional point cloud data of grazing grasslands and can provide high-precision grassland structure information, including grassland canopy height, gap fraction, and leaf area index.

[0084] Construct grassland canopy height model data at different spatial scales based on the preprocessed point cloud data; the generation process is to use a program written in Python 3.9 to generate a centimeter-level digital elevation model DEM and a digital surface model DSM, perform a difference operation on the two, and obtain the fine-resolution canopy height model CHM data through coordinate matching.

[0085] When calculating the DEM and DSM, it is necessary to set the spatial sampling scale, which are set to 0.05m, 0.1m, 0.15m, 0.2m, and 0.25m respectively to generate the DEM and DSM at five scales. In order to subsequently generate the CHM, CGF, LAI, and gAGB data at five scales.

[0086] Based on the grassland canopy height model data, obtain the canopy structure parameters of the target grassland respectively. The canopy structure parameters include: canopy height, grassland canopy gap fraction, and leaf area index. When calculating the grassland canopy gap fraction, it is necessary to set the height threshold of the grass, that is, the CHM pixels higher than the threshold are grass, and the rest are the ground. Calculate by setting the spatial sampling intervals of 0.05m, 0.1m, 0.15m, 0.2m, and 0.25m in turn, and set the window to 3×3, 5×5, 7×7, and 9×9 at the same time. Study the gap fraction and window at different spatial scales for the subsequent calculation of the leaf area index, and obtain the optimal grassland canopy gap fraction and leaf area index raster data at different scales through comparison and analysis.

[0087] Use the semi-variance function to calculate the semi-variance values of the grassland canopy structure parameters at each spatial scale respectively, construct fitting models based on the semi-variance values combined with the exponential model, Gaussian model, and spherical model respectively, and obtain the optimal fitting model and fitting parameters through accuracy verification;

[0088] Analyze the spatial heterogeneity characteristics of grassland canopy structure parameters at each spatial scale based on the optimal fitting model and fitting parameters, and obtain the first optimal inversion scale and the first most stable canopy structure parameters according to the spatial heterogeneity characteristics;

[0089] Respectively obtain the measured data of the above-ground biomass of the target grassland;

[0090] Taking canopy height, grassland canopy gap fraction, and leaf area index as independent variables and the measured data as the dependent variable, construct inversion models based on linear regression models, random forest models, and BP neural network models respectively;

[0091] Verify the accuracy of the constructed inversion models of linear regression models, random forest models, and BP neural network models to obtain the optimal inversion model, the second best inversion scale, and the second most stable canopy structure parameters;

[0092] Verify the accuracy of the first best inversion scale and the first most stable canopy structure parameters according to the second best inversion scale and the second most stable canopy structure parameters;

[0093] The optimal inversion model inverses the measured data to obtain the inversion result of the above-ground biomass of the target grassland;

[0094] Conduct spatial heterogeneity analysis on the inversion result of the above-ground biomass of the target grassland to obtain the most sensitive spatial scale of the grassland;

[0095] Determine the optimized management measures for the grassland sensitive spatial scale based on the inversion result of the above-ground biomass of the target grassland and the most sensitive spatial scale of the grassland.

[0096] Furthermore, preprocess the point cloud data, including: point cloud coordinate transformation, flight strip stitching, data chunking, denoising processing, and filtering the denoised point cloud data based on ground point filtering to separate non-ground points and ground points.

[0097] Even further, construct grassland canopy height model data at different spatial scales based on the preprocessed point cloud data, including:

[0098] Select the low-value data of multiple echoes within the ground point data unit as the surface position, and use Inverse Distance Weighted (IDW) to interpolate the unevenly distributed three-dimensional discrete points in the ground points to obtain continuous and variable regular grid Digital Elevation Model (DEM) data, which represents a finite sequence of three-dimensional vectors of the surface of a specific area in functional form, such as Equation (3).

[0099] V i =(X i ,Y i,M i )(i = 1, 2, 3, …, n)#(3)

[0100]

[0101] Among them, the vector V in equation (3) i is the set of elevation values of the grassland canopy, which is the Digital Elevation Model (DEM) data; (X i , Y i ) are the plane coordinates, and M i is the elevation value of this coordinate point. Z in equation (4) represents the predicted value of the known point value Z i after IDW interpolation, and p represents the power. When p is generally selected from 0.5 to 3, the interpolation effect is better. d in equation (5) i represents the distance between the prediction point (x, y) and the known point (x i , y i ).

[0102] Select the maximum value among multiple echo points of non-ground data units as its pixel value, and use the Inverse Distance Weighting (IDW) method to interpolate discrete numerical values to generate a Digital Surface Model (DSM) with a fixed resolution, as shown in equation (6).

[0103] N i = (X i , Y i , W i )(i = 1, 2, 3, …, n)#(6)

[0104] The vector N in equation (6) i is the set of height values of all surface objects including grassland vegetation, that is, the Digital Surface Model data. The variables (X i , Y i ) are the plane coordinates, and W i is the grassland canopy elevation value.

[0105] Performing a difference operation between DSM and DEM, i.e., (N i - V i ) and through coordinate matching, the Canopy Height Model (CHM) data with a fixed resolution can be obtained. The formula is as shown in (7).

[0106] CHM = DSM - DEM#(7)

[0107] In another embodiment, based on the grassland canopy height model data, the canopy structure parameters of the target grassland are respectively obtained, including:

[0108] Calculating the canopy height using the grassland canopy height model data;

[0109] Calculate the vegetation canopy density based on the two-dimensional grassland canopy height model data. Start indexing from the upper left corner of the CHM raster data. In the order from left to right and from top to bottom, take each indexed grid as the central pixel point of the window. For the positions where the window has no value, use the mirror method for filling. By counting the number of CHM raster data pixels higher than the grassland height threshold within this window, obtain the number of vegetation and soil pixels of the grassland, and thus obtain the canopy density. Since the gap fraction and the canopy density add up to 1, the formula for calculating the gap fraction is as follows:

[0110]

[0111] Among them, if the pixel value pixe[i,j] at the coordinate (i,j) in the window is greater than the user-defined grassland height threshold, then pixe[i,j] is a vegetation point in this window and is defined as pixe[i,j] veg Therefore, ∑pixe[i,j] in the above formula veg is the number of vegetation pixels in this window, and ∑pixe[i,j] is the total number of pixels in this window.

[0112] After obtaining the vegetation canopy gap fraction, calculate the LAI value within this window based on the Beer-Lambert law. Assign this LAI value to the position of the central pixel of the window, and then the LAI raster data with the same resolution as the input CHM can be obtained. The calculation formula is as follows:

[0113]

[0114] Among them, Ω(θ) is the aggregation index, θ is the zenith angle of the incident light, G(θ) is the extinction coefficient, P(θ) is the gap probability for observing the zenith angle θ, and cos(θ) is the optical path length of the homogeneous vegetation layer considering the random distribution of leaves. In the existing reference studies, when θ = 57.5°, the influence of the leaf inclination angle of the canopy on the extinction coefficient is the smallest. At this time, when G(θ) is based on the spherical model, its value is 0.5, and the aggregation index Ω is set to the constant 1.

[0115] In another embodiment, the semivariogram is a measure of the sample space variation of a regionalized variable at a separation distance, which can be used to describe the spatial characteristics of the grassland vegetation community. It is usually represented by a graph of the semivariogram value of the vegetation parameter versus the distance, which helps to describe the covariance structure of the geographical data at different distances and can help determine the spatial correlation between data points. Therefore, use the semivariogram to calculate the semivariance values of the grassland canopy structure parameters at each spatial scale respectively. Based on the semivariance values, fit with the exponential model, Gaussian model, and spherical model. Through accuracy verification, obtain the optimal fitting model and fitting parameters, including:

[0116] The semi-variance values of the grassland canopy structure parameters at each spatial scale are calculated using the semi-variance function as follows:

[0117]

[0118] Among them, γ(h) represents the semi-variance function, which is a correlation value related to the distance h. h is the distance, representing the interval between two locations. N(h) is the number of point pairs within the distance h. Z(x i ) is the vegetation parameter data value at location x i . Z(x i +h) is the vegetation parameter data value at location x i +h at a distance h. The semi-variance function values of the canopy structure parameters of the enclosed grassland and the grazing grassland at different scales are calculated using this formula, and the biomass obtained by inversion later is prepared for fitting the semi-variance function model in the next step.

[0119] The semi-variance function requires a fitting model to describe the spatial variability of the grassland vegetation indicators in the study area. Commonly used theoretical models include the Exponential model, Gaussian model, and Spherical model. Fitting models are constructed based on the semi-variance values by combining the Exponential model, Gaussian model, and Spherical model respectively:

[0120]

[0121] Among them, γ Exp (d) is the Exponential model, γ Gau (d) is the Gaussian model, γ Sph (d) is the Spherical model. Among them, C 0 is the Nugget, representing the influence of random factors on the indicator; C is the Partial Sill, which is the part of the variation that can be explained by the spatial process after removing the Nugget effect C 0 , that is, the uncorrelated variation caused by the inherent randomness of the observed data or measurement errors; C 0 +C is the Sill. When the distance h between a pair of samples increases, the corresponding variogram value generally also increases. When the variogram value reaches the Sill C 0 +C, it no longer rises; a is the Range, in meters, which is the distance between samples when the function reaches the Sill C 0 +C, that is, the range of action of the spatial factor variation characteristics; d is the distance between observation points in spatial analysis, in meters.

[0122] Based on the above principles, the model was used to fit the canopy structure parameters CHM, CGF, LAI of enclosed grasslands and grazing grasslands at different scales and the retrieved gAGB to obtain the fitting parameters. By analyzing the fitting parameters of the optimal model, the optimal inversion scale was obtained and the most stable canopy structure parameters were screened out, significantly improving the modeling accuracy. At the same time, it revealed how the canopy structure parameters affect the biomass distribution under different management methods and scales. At the same time, analyzing the spatial heterogeneity characteristics of the biomass of enclosed grasslands and grazing grasslands under different grazing management scenarios at different spatial scales helps to reveal the impact of different management methods on the grassland ecosystem, and can also provide scientific support for optimizing management measures, improving the accuracy of biomass inversion, and evaluating ecological functions.

[0123] Furthermore, taking the canopy height, grassland canopy gap fraction, and leaf area index as independent variables and the measured data as the dependent variable, inversion models based on linear regression models, random forest models, and BP neural network models were constructed respectively:

[0124] The inversion model constructed based on the linear regression model is:

[0125] y = a 1 x 1 + a 2 x 2 + a 3 x 3 + … + a k x k + a 0 #(14)

[0126] a 1 、a 2 、a 3 … a k are the regression coefficients of the LRM fitting equation, a 0 is the constant term of the fitting equation, x 1 , x 2 , x 3 … x k are the independent variables;

[0127] When using the random forest to invert the aboveground biomass of the target grassland, samples are randomly selected to form each decision tree, and then the results of all decision trees are averaged to obtain the inversion result of the aboveground biomass of the target grassland. The inversion model constructed based on the random forest model is:

[0128]

[0129] where, T i (x) represents the individual prediction result of the i-th decision tree for the sample x, and m is the total number of decision trees in the random forest;

[0130] The BP neural network consists of an input layer, an output layer, and several hidden layers, and trains a multi-layer feedforward neural network through forward propagation and backpropagation algorithms; in the forward propagation stage, the input data P 1 , P 2 , …, P n (CHM, CGF, LAI) undergoes weighted summation through weights ω 1 , ω 2 , …, ω n and the bias value b of the neuron, and through the activation function σ for non-linear transformation, the signal is transmitted from the input layer to the hidden layer and then to the output layer; the biomass of the output layer is compared with the expected value. If the error does not meet the requirements, the weights and biases of each layer are adjusted through the backpropagation algorithm until the error reaches the expectation or the set number of training times.

[0131] Specifically, the optimal fitting model and the optimal inversion model obtained through accuracy verification include: using the coefficient of determination, the sum of squared residuals, and the root mean square error to evaluate and test the model accuracy respectively.

[0132] RSS is the sum of the squares of the differences between the observed values and the model predicted values, which can measure the size of the model prediction error. The smaller RSS is, the closer the predicted value of the model is to the actual value, and the better the model is. R 2 represents the proportion of the model to explain the data variability, which is used to measure the goodness of fit of the model to the data. The value of R 2 is usually between 0 and 1. The closer the value is to 1, the higher the variability explained by the model, and the better the model fits. RMSE is the square root of the average of the squares of the differences between the measured values and the model predicted values, which is an absolute scale to measure the model prediction error. The lower RMSE is, the smaller the prediction error of the model, and the better the prediction performance of the model. The formulas are as follows respectively:

[0133]

[0134] In Equation (16), n represents the number of samples, is the average value of the sample observation data, where y i and represent the observed value and the predicted value of the i-th sample point respectively. In Equation (17), n represents the number of data point pairs, γ i is the actual semi-variance value of the i-th observation point, is the model fitting semi-variance value of this point. In Equation (18), n represents the total number of sampling points of the grassland above-ground biomass; y i represents the measured value of the i-th sampling point; represents the i-th model predicted value.

[0135] On the other hand, the present invention provides a grassland aboveground biomass estimation system based on the coupling of semi-variance function and grazing heterogeneity, as Figure 2 shown, including:

[0136] A data acquisition unit for acquiring point cloud data of the target grassland and performing preprocessing;

[0137] A canopy structure parameter calculation unit for constructing grassland canopy height model data at different spatial scales based on the preprocessed point cloud data, and respectively obtaining the canopy structure parameters of the target grassland based on the grassland canopy height model data. The canopy structure parameters include: canopy height, grassland canopy gap ratio, and leaf area index;

[0138] A fitting unit for respectively calculating the semi-variance values of the grassland canopy structure parameters at each spatial scale by using the semi-variance function, constructing fitting models by combining the exponential model, Gaussian model, and spherical model based on the semi-variance values, and obtaining the optimal fitting model and fitting parameters through accuracy verification;

[0139] A parameter screening unit for analyzing the spatial heterogeneity characteristics of the grassland canopy structure parameters at each spatial scale based on the optimal fitting model and fitting parameters, and obtaining the first optimal inversion scale and the first most stable canopy structure parameter according to the spatial heterogeneity characteristics;

[0140] A measurement unit for respectively obtaining the measured data of the aboveground biomass of the target grassland;

[0141] A model construction unit for constructing inversion models based on linear regression models, random forest models, and BP neural network models respectively with canopy height, grassland canopy gap ratio, and leaf area index as independent variables and the measured data as the dependent variable;

[0142] A model selection unit for performing accuracy verification on the constructed inversion models of the linear regression model, random forest model, and BP neural network model, and obtaining the optimal inversion model, the second best inversion scale, and the second most stable canopy structure parameter;

[0143] A verification unit for verifying the accuracy of the first best inversion scale and the first most stable canopy structure parameter according to the second best inversion scale and the second most stable canopy structure parameter;

[0144] An inversion unit for inverting the measured data by the optimal inversion model to obtain the inversion result of the aboveground biomass of the target grassland;

[0145] An analysis unit for performing spatial heterogeneity analysis on the inversion result of the aboveground biomass of the target grassland to obtain the most sensitive spatial scale of the grassland;

[0146] A management unit is configured to determine optimized management measures for the grassland sensitive spatial scale based on the inversion result of the target grassland aboveground biomass and the most sensitive spatial scale of the grassland.

[0147] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section.

[0148] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for estimating aboveground biomass of grassland based on semivariogram function coupled with grazing heterogeneity, characterized in that: include: Obtain the point cloud data of the target grassland and perform preprocessing; Based on the preprocessed point cloud data, the grassland canopy height model data at different spatial scales were constructed; Based on the grassland canopy height model data, canopy structure parameters of the target grassland are respectively obtained, wherein the canopy structure parameters include: canopy height, grassland canopy gap ratio and leaf area index; The semivariance function is used to calculate the semivariance values ​​of the grassland canopy structure parameters at each spatial scale, and fitting models are constructed based on the semivariance values ​​combined with the exponential model, Gaussian model and spherical model, and the optimal fitting model and fitting parameters are obtained through precision verification; Analyzing the spatial heterogeneity characteristics of grassland canopy structural parameters at various spatial scales based on the optimal fitting model and fitting parameters, and obtaining a first optimal inversion scale and a first most stable canopy structural parameter according to the spatial heterogeneity characteristics; respectively obtaining measured data of aboveground biomass of the target grassland; Taking the canopy height, grassland canopy gap ratio and leaf area index as independent variables and the measured data as dependent variables, inversion models based on linear regression model, random forest model and BP neural network model are constructed respectively; Verify the accuracy of the constructed inversion models of the linear regression model, random forest model and BP neural network model to obtain the optimal inversion model, the second best inversion scale and the second most stable canopy structure parameter; Verifying the accuracy of the first optimal inversion scale and the first most stable canopy structure parameter according to the second optimal inversion scale and the second most stable canopy structure parameter; The optimal inversion model inverts the measured data to obtain the inversion result of the aboveground biomass of the target grassland; Conduct spatial heterogeneity analysis on the aboveground biomass inversion results of the target grassland to obtain the most sensitive spatial scale of the grassland; Based on the inversion results of the aboveground biomass of the target grassland and the most sensitive spatial scale of the grassland, optimal management measures for the grassland sensitive spatial scale are determined.

2. The method for estimating aboveground biomass of grassland based on semivariogram function coupled grazing heterogeneity according to claim 1, characterized in that: The point cloud data is preprocessed, including: point cloud coordinate conversion, flight strip stitching, data segmentation, denoising, and filtering the denoised point cloud data based on ground point filtering to separate non-ground points and ground points.

3. The method for estimating aboveground biomass of grassland based on semivariogram function coupled grazing heterogeneity according to claim 2 is characterized in that: Based on the preprocessed point cloud data, the grassland canopy height model data at different spatial scales are constructed, including: The minimum value of multiple echoes in the ground point is selected as the surface position, and the discrete points in the ground point are interpolated using the inverse distance weight to obtain the digital elevation model data; The maximum value among multiple echo points of the non-ground data unit is selected as the pixel value, and the discrete points in the non-ground point are interpolated using the inverse distance weighted method to obtain the digital surface model data; Canopy height model data is obtained based on the digital elevation model data and the digital surface model data.

4. The method for estimating aboveground biomass of grassland based on semivariogram function coupled grazing heterogeneity according to claim 1, characterized in that: Based on the grassland canopy height model data, canopy structure parameters of the target grassland are respectively obtained, including: Calculating the canopy height using the grassland canopy height model data; The vegetation density is calculated based on the two-dimensional grassland canopy height model data, and the grassland canopy gap rate is obtained based on the vegetation density: where ∑pixe[i,j] veg is the number of vegetation pixels in the window, ∑pixe[i,j] is the total number of pixels in the window; Calculate the leaf area index within the window based on the Beer-Lambert law: Where Ω(θ) is the aggregation index, θ is the zenith angle of the incident light, G(θ) is the extinction coefficient, P(θ) is the gap probability of observing the zenith angle θ, and cosθ is the optical path length of the uniform vegetation layer considering the random distribution of leaves.

5. The method for estimating aboveground biomass of grassland based on semivariogram function coupled grazing heterogeneity according to claim 1, characterized in that: The semivariance function is used to calculate the semivariance values ​​of the grassland canopy structure parameters at each spatial scale. The semivariance values ​​are combined with the exponential model, Gaussian model and spherical model for fitting. The optimal fitting model and fitting parameters are obtained through precision verification, including: The semivariance function is used to calculate the semivariance values ​​of grassland canopy structure parameters at each spatial scale: Where h is the distance between two locations, N(h) is the number of point pairs within the distance h, and Z(x i ) is located at location x i The vegetation parameter data value at Z(x i +h) is the point x at distance h i +h vegetation parameter data value; Based on the semivariance values, exponential model, Gaussian model and spherical model are combined to build fitting models: Among them, γ Exp (d) is the exponential model, γ Gau (d) is the Gaussian model, γ Sph (d) is the spherical model, C0 is the nugget value; C is the partial base range; C0+C is the sill value; a is the range; d is the distance of the observation point in the spatial analysis.

6. The method for estimating aboveground biomass of grassland based on semivariogram function coupled grazing heterogeneity according to claim 1, characterized in that: With the canopy height, grassland canopy gap ratio and leaf area index as independent variables and the measured data as dependent variables, inversion models based on linear regression model, random forest model and BP neural network model were constructed respectively: The inversion model constructed based on the linear regression model is: y=a1x1+a2x2+a3x3+…+a k x k +a0 a1, a2, a3…a k is the regression coefficient of the LRM fitting equation, a0 is the constant term of the fitting equation, x1, x2, x3…x k is the independent variable; When using random forests to invert the aboveground biomass of the target grassland, samples are randomly selected to form each decision tree, and the results of all decision trees are averaged to obtain the aboveground biomass inversion result of the target grassland. The inversion model constructed based on the random forest model is: Among them, T i (x) represents the individual prediction result of the i-th decision tree for sample x, and m is the total number of decision trees in the random forest; BP neural network consists of input layer, output layer and several hidden layers. It trains multi-layer feedforward neural network through forward propagation and back propagation algorithm. In the forward propagation stage, input data P1, P2, ..., P n After weights ω1, ω2, …, ω n The signal is transmitted from the input layer to the hidden layer and then to the output layer through the activation function σ, and the weighted sum of the neuron bias value b is taken. The signal is compared with the expected value at the output layer. If the error does not meet the requirements, the weights and biases of each layer are adjusted through the back propagation algorithm until the error reaches the expected value or the set number of training times is reached.

7. The method for estimating aboveground biomass of grassland based on semivariogram function coupled grazing heterogeneity according to claim 1, characterized in that: The optimal fitting model and the optimal inversion model are obtained through accuracy verification, including: using the determination coefficient, residual sum of squares and root mean square error to evaluate and test the model accuracy respectively.

8. A grassland aboveground biomass estimation system based on semivariogram function coupled with grazing heterogeneity, characterized in that: include: A data acquisition unit, used for acquiring point cloud data of the target grassland and performing preprocessing; A canopy structure parameter calculation unit is used to construct grassland canopy height model data at different spatial scales based on the preprocessed point cloud data, and obtain the canopy structure parameters of the target grassland based on the grassland canopy height model data, wherein the canopy structure parameters include: canopy height, grassland canopy gap ratio and leaf area index; A fitting unit is used to calculate the semivariance values ​​of the grassland canopy structure parameters at each spatial scale using a semivariance function, and to construct fitting models based on the semivariance values ​​in combination with an exponential model, a Gaussian model, and a spherical model, and to obtain the optimal fitting model and fitting parameters through precision verification; A parameter screening unit, for analyzing the spatial heterogeneity characteristics of grassland canopy structure parameters at various spatial scales based on the optimal fitting model and fitting parameters, and obtaining a first optimal inversion scale and a first most stable canopy structure parameter according to the spatial heterogeneity characteristics; A measuring unit, used to respectively obtain measured data of aboveground biomass of the target grassland; A model building unit, used to build inversion models based on a linear regression model, a random forest model and a BP neural network model respectively, with the canopy height, grassland canopy gap ratio and leaf area index as independent variables and the measured data as dependent variables; A model selection unit is used to verify the accuracy of the constructed inversion models of the linear regression model, random forest model and BP neural network model, and obtain the optimal inversion model, the second best inversion scale and the second most stable canopy structure parameter; A verification unit, configured to verify the accuracy of the first optimal inversion scale and the first most stable canopy structure parameter according to the second optimal inversion scale and the second most stable canopy structure parameter; An inversion unit is used to invert the measured data using an optimal inversion model to obtain an inversion result of the aboveground biomass of the target grassland; An analysis unit, used for performing spatial heterogeneity analysis on the aboveground biomass inversion results of the target grassland to obtain the most sensitive spatial scale of the grassland; A management unit is used to determine the optimization management measures of the sensitive spatial scale of the grassland based on the inversion result of the aboveground biomass of the target grassland and the most sensitive spatial scale of the grassland.

Images