Composite carbon sink estimation method based on multi-mode mangrove forest canopy data

Abstract

The invention discloses a composite carbon sink estimation method based on multi-mode mangrove forest canopy data, and belongs to the technical field of carbon sink calculation in ecological environment monitoring, and the method comprises the steps: obtaining laser radar point cloud data, a hyperspectral remote sensing image and ground actual measurement data of a mangrove forest region; preprocessing the laser radar point cloud data, the hyperspectral remote sensing image and the ground actual measurement data to obtain standardized data; structure parameters and biochemical parameters are inversed based on the standardized data; fusing the structure parameters and the biochemical parameters to construct a comprehensive inversion model; the mangrove forest carbon reserve is calculated through the comprehensive inversion model; and verifying the precision of the comprehensive inversion model through ground measured data, and outputting a final carbon sink estimation result.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of carbon sink estimation in ecological environment monitoring, and particularly relates to the technical field of mangrove ecosystem carbon storage estimation based on multi-modal data. BACKGROUND

[0002] As a typical blue carbon ecosystem, mangrove has outstanding carbon fixation and storage capacity. Precise estimation of its carbon storage is of great significance to ecological environment protection and carbon cycle research. However, the spatial distribution of mangrove is complex, the growth environment is harsh, and the community type is diverse. The canopy structure presents obvious vertical stratification characteristics, including tall trees at the top layer, shrubs and young mangrove at the understory layer, and low trees. These factors pose many challenges to the accurate estimation of aboveground vegetation carbon storage of mangrove.

[0003] Existing mangrove carbon sink estimation methods mainly include optical remote sensing detection, laser radar detection, ground measurement, and estimation relying solely on remote sensing data. Optical remote sensing is a passive remote sensing method that relies on the reflection of sunlight by objects to form an image. It is easily affected by light shielding or reflection caused by weather factors, and has poor stability in obtaining canopy structure parameters of mangrove. Although laser radar can collect three-dimensional structure data all-weather, it is difficult to reflect the optical properties and attributes of vegetation due to the limitation of the detection laser beam function, which brings difficulties to tree species identification and leaf biochemical parameter analysis. Ground measurement needs to be carried out in complex environments such as intertidal zone, which not only has safety risks, but also has the problems of time-consuming and high cost. Moreover, human activities may damage the vegetation and roots in the forest, making it difficult to achieve large-scale application.

[0004] When using remote sensing data alone to estimate the biomass of mangrove, only the tall trees at the top layer can be covered, while the biomass of low plants, associated shrubs and new mangrove seedlings at the understory layer is often ignored or cannot be completely calculated. This is because of the limitations of optical remote sensing detection, and the complexity and difficulty of measuring low vegetation types on the ground, which makes it difficult for traditional remote sensing estimation methods to fully cover the vertical structure of mangrove ecosystems. At the same time, existing technologies fail to effectively integrate the advantages of different data sources, making it difficult to accurately obtain the structure parameters and biochemical parameters of different canopy layers of mangrove at the same time. There is also a lack of systematic estimation scheme for full canopy vegetation, making the completeness and accuracy of carbon sink estimation difficult to meet the actual needs. SUMMARY

[0005] The present application aims to provide a composite carbon sink estimation method based on multi-modal mangrove canopy data, to solve the problem of simultaneous accurate acquisition of structure parameters and biochemical parameters of different canopy layers in existing mangrove carbon sink estimation, and accurate calculation of carbon sink.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0007] A composite carbon sink estimation method based on multimodal mangrove canopy data includes the following steps:

[0008] Acquire lidar point cloud data, hyperspectral remote sensing images, and ground-measured data for mangrove areas;

[0009] The lidar point cloud data, hyperspectral remote sensing images, and ground-measured data are preprocessed to obtain standardized data.

[0010] Structural parameters and biochemical parameters were retrieved based on the standardized data.

[0011] A comprehensive inversion model is constructed by integrating the structural and biochemical parameters.

[0012] The carbon storage of mangroves was calculated using the comprehensive inversion model described above.

[0013] The accuracy of the comprehensive inversion model is verified by ground-based measured data, and the final carbon sink estimation result is output.

[0014] In one possible implementation, when acquiring data, the mangrove area is scanned by airborne or drone-based lidar to obtain three-dimensional point cloud data of the canopy; three-dimensional point cloud data of the mangrove understory vegetation is collected by handheld lidar; hyperspectral remote sensing images of a specific band range are acquired; sample plots are set up in the mangrove understory vegetation area, and the plant height and spatial distribution data of the understory vegetation are extracted using handheld lidar. The diameter at breast height of trees is obtained by combining measurement tools, organ samples of different tree species are collected, and carbon content coefficients are obtained through laboratory analysis to obtain ground-based measured data containing vegetation growth information and carbon content.

[0015] In one possible implementation, during preprocessing, the lidar point cloud data is filtered and topographically corrected to construct a canopy height model; the hyperspectral remote sensing image is subjected to strip stitching, atmospheric correction, geometric fine correction, and band dimensionality reduction; effective bands of the hyperspectral image are selected based on spectral index correlation, and the spectral reflectance curves of plants within the sample plots are recorded; the lidar point cloud data, hyperspectral remote sensing image, and ground-measured data are coordinate unified and spatially registered to obtain standardized data.

[0016] In one possible implementation, when inverting structural parameters, individual trees are segmented from the standardized canopy 3D point cloud data to extract key vegetation structure parameters; statistical analysis and spatial data interpolation are performed on the standardized understory 3D point cloud data to establish an understory vegetation spatial data model, extract understory structural feature parameters, and obtain structural parameters including the top layer and the understory.

[0017] In one possible implementation, when retrieving biochemical parameters, spectral indices related to vegetation biochemistry are calculated based on the standardized hyperspectral remote sensing image; mangrove species are classified using machine learning methods, and leaf chemical parameters are retrieved by combining the spectral indices to obtain biochemical parameters that include species type and leaf chemical properties.

[0018] In one possible implementation, when constructing the comprehensive inversion model, the key dimensions of vegetation growth data in the structural parameters and the spectral index in the biochemical parameters are input into a fusion modeling algorithm. Corresponding regression coefficients are set for each species to establish a top-level plant carbon storage inversion model. The structural characteristic parameters and diameter parameters of the understory are substituted into an improved allometric growth equation, and corresponding regression coefficients are set for each understory species to establish an understory plant carbon storage inversion model. The top-level and understory plant carbon storage inversion models are integrated to obtain a comprehensive inversion model covering the entire canopy.

[0019] In one possible implementation, the fusion modeling algorithm includes random forest regression, XGBoost regression, or a neural network-based nonlinear regression model. When establishing the top-level plant carbon storage inversion model, tree height and diameter at breast height are used as structural input dimensions, and vegetation biochemically related spectral indices are used as biochemical input dimensions. The model is constructed by fitting regression coefficients to calculate the carbon storage per unit area of ​​the top-level plants.

[0020] In one possible implementation, when establishing a carbon storage inversion model for understory vegetation, the height quantile, point cloud coverage, number of non-empty voxels per unit area, and leaf height diversity index are extracted from the structural characteristic parameters of the understory. These parameters are then combined with the ground diameter parameters of the understory vegetation and substituted into an improved log-linear allometric growth equation. The model is then constructed by fitting regression coefficients to calculate the carbon storage per unit area of ​​understory vegetation.

[0021] In one possible implementation, when calculating mangrove carbon storage, the comprehensive inversion model is used to calculate the carbon storage per unit area of ​​top-layer plants and understory plants respectively; combined with the distribution area of ​​each species in the mangrove area, the total carbon storage of top-layer plants and the total carbon storage of understory plants are calculated respectively; the two are superimposed to obtain the aboveground carbon storage of mangroves, and the total carbon storage of mangroves is calculated by combining the fixed ratio between underground biomass and aboveground carbon storage.

[0022] In one possible implementation, when verifying accuracy, the actual carbon storage is calculated by combining the allometric growth equation with the measured biomass and carbon content coefficients on the ground; the root mean square error, mean absolute error, and coefficient of determination are used to compare the actual carbon storage with the carbon storage predicted by the model; the model fitting effect is evaluated based on the index results, and if the coefficient of determination is close to the preset threshold, the verification is completed; otherwise, the model parameters are adjusted and retrained until the model meets the accuracy requirements.

[0023] Compared with existing technologies, the advantages of this invention are as follows: lidar data can provide three-dimensional structural information of mangroves, hyperspectral remote sensing images can capture the biochemical characteristics of species, and ground-based measured data can be used for parameter calibration. The combination of the above data makes up for the limitations of a single data source. Compared with existing technologies that rely solely on optical remote sensing and are easily affected by weather, and lidar is difficult to reflect the optical properties of vegetation, this invention achieves the simultaneous acquisition of structural parameters and biochemical parameters through the complementarity of different data sources, making the basic data for carbon sink estimation more comprehensive.

[0024] To address the issue of existing technologies neglecting carbon storage in the understory vegetation, this invention uses handheld lidar to collect 3D point cloud data of low-lying vegetation, shrubs, and newly sprouted mangrove seedlings in the understory. Combined with statistical analysis and spatial interpolation, a dedicated model is established to fully cover the vertical structure of both the mangrove canopy and the understory. This full-canopy-coverage modeling approach overcomes the limitation of traditional remote sensing estimation, which can only cover tall trees, ensuring that carbon storage calculations do not omit key components and better reflect the actual structural characteristics of the mangrove ecosystem. Simultaneously, ground-based measurement plots are preferentially selected in the understory vegetation area, reducing the impact of terrain undulations and background interference on data collection and facilitating the acquisition of accurate structural parameters and complete leaf samples.

[0025] In terms of model construction, this invention utilizes machine learning methods for species classification and parameter inversion. It integrates structural and biochemical parameters through a fusion modeling algorithm. Compared to traditional allometric growth equations that rely solely on tree height and diameter at breast height (DBH), this invention incorporates more key factors influencing carbon storage, such as chlorophyll, water content, and lignin, resulting in richer model inputs. Separate carbon storage inversion models are established for the different growth characteristics of top-layer and understory vegetation, with dedicated regression coefficients, achieving accurate classification calculations and avoiding the limitations of single models in adapting to different vegetation types.

[0026] This invention calculates actual carbon storage by measuring biomass and carbon content coefficients, and verifies accuracy using multiple indicators, ensuring the reliability and stability of the model and overcoming the estimation bias caused by the lack of measured verification in some existing technologies. Furthermore, the model parameters can be adjusted according to the species types and vertical structure information of mangroves in different regions, adapting to different application scenarios without reconstructing the overall framework. It is easy to operate and highly scalable. In addition, relying on real-time remote sensing data and the retrieved model parameters, carbon sequestration can be continuously monitored remotely, further expanding the application value of the technology. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic flowchart of the composite carbon sink estimation method according to an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0030] Example:

[0031] It should be noted that the terms "comprising" and "having" and any variations thereof in the embodiments of the present invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0032] This invention provides a composite carbon sink estimation method based on multimodal mangrove canopy data, comprising the following steps:

[0033] Step 110: Acquire lidar point cloud data, hyperspectral remote sensing images, and ground measurement data for the mangrove area.

[0034] Specifically, lidar point cloud data is a set of three-dimensional spatial points obtained by lidar scanning, which can be canopy point clouds collected by airborne lidar or forest understory point clouds collected by handheld lidar; hyperspectral remote sensing images are remote sensing images containing multiple continuous bands, which can be images with a band range of 400-2500nm; ground-measured data are data related to vegetation growth and carbon content collected in the field, which can be tree height, diameter at breast height, carbon content coefficient, etc.

[0035] The data acquisition process involved scanning mangrove areas with airborne or drone-based lidar to obtain three-dimensional point cloud data of the canopy; collecting three-dimensional point cloud data of the mangrove understory vegetation using handheld lidar; acquiring hyperspectral remote sensing images within specific spectral bands; establishing sample plots in the mangrove understory vegetation area; extracting plant height and spatial distribution data of the understory vegetation using handheld lidar; obtaining tree diameter at breast height (DBH) using measurement tools; collecting organ samples from different tree species; and obtaining carbon content coefficients through laboratory analysis to obtain ground-based measured data containing vegetation growth information and carbon content.

[0036] Specifically, the canopy 3D point cloud data is a set of 3D points reflecting the spatial structure of the top-layer trees in mangroves, which can be millions of point cloud data obtained by UAV lidar scanning; the understory 3D point cloud data is a set of 3D points reflecting the spatial morphology of understory shrubs and low vegetation, which can be hundreds of thousands of point cloud data collected by handheld lidar; the specific band range is the spectral range suitable for vegetation biochemical parameter inversion, which can be 400-2500nm; the measuring tool is a device used to measure the diameter at breast height of trees, which can be calipers; the organ sample is a sample of leaves, branches, or roots of trees, which can be a leaf sample of mangrove trees; the carbon content coefficient is the proportion of carbon storage per unit of biomass, which can be 0.48 as determined by an elemental analyzer.

[0037] Under suitable weather conditions, airborne or drone-based lidar is used to conduct a full-coverage scan of the target mangrove area, acquiring three-dimensional point cloud data of the canopy. Operators then use handheld lidar to enter the mangrove understory, scanning the vegetation area by area and collecting three-dimensional point cloud data of the understory. Hyperspectral remote sensing images covering the target area and within a wavelength range of 400-2500 nanometers are acquired using hyperspectral remote sensing satellites or airborne hyperspectral equipment. Multiple sample plots are established in areas with a canopy height of 2-4 meters in the understory. Handheld lidar is used to extract data on plant height and spatial distribution within the sample plots. Tree diameter at breast height (DBH) is measured using calipers, and leaf and branch samples from different tree species are collected. These samples are sent to a laboratory where carbon content coefficients are determined using an elemental analyzer. All data are then integrated to obtain ground-based measured data.

[0038] Step 120: Preprocess the lidar point cloud data, hyperspectral remote sensing image and ground measurement data to obtain standardized data.

[0039] Specifically, standardized data is a collection of data with a unified format and consistent coordinates after preprocessing. It can be an integrated data set of registered point clouds, corrected images, and measured data.

[0040] During preprocessing, the lidar point cloud data is filtered and topographically corrected to construct a canopy height model; the hyperspectral remote sensing image is subjected to strip stitching, atmospheric correction, geometric fine correction, and band dimensionality reduction; effective bands of the hyperspectral image are selected based on spectral index correlation, and the spectral reflectance curves of plants within the sample plots are recorded; the lidar point cloud data, hyperspectral remote sensing image, and ground-measured data are coordinate unified and spatially registered to obtain standardized data.

[0041] Specifically, filtering is a method for removing noise points from point cloud data, which can be a statistical filtering algorithm; terrain correction is a method for eliminating the influence of terrain undulations on vegetation height, which can be a correction method based on a digital elevation model; canopy height model is a two-dimensional raster model reflecting canopy height distribution, which can be a raster model with a resolution of 1 meter; band dimensionality reduction is a method for reducing redundant bands in hyperspectral images, which can be a principal component analysis method; effective bands are spectral bands that are highly correlated with vegetation biochemical parameters, which can be bands such as 750nm and 550nm; spectral reflectance curves are curves showing the reflectance changes of vegetation in different bands, which can be the spectral reflectance curves of mangrove leaves; coordinate unification is an operation that transforms different data sources to the same coordinate system, which can be a transformation to the WGS84 coordinate system; spatial registration is an operation that aligns the spatial positions of different data sources, which can be a registration method based on ground control points.

[0042] Point cloud processing software was used to statistically filter the lidar point cloud data to remove noise points. Then, a digital elevation model (DEM) was used for terrain correction to eliminate the influence of terrain undulations. A canopy height model was constructed based on the corrected point cloud data. Remote sensing image processing software was used to strip-stitch hyperspectral remote sensing images to eliminate image stitching gaps. An atmospheric radiative transfer model was used for atmospheric correction to remove atmospheric effects. Geometric precision correction was performed using ground control points to correct spatial position deviations. Principal component analysis was used for band dimensionality reduction. Vegetation spectral indices were calculated, and effective bands were selected based on the correlation between spectral indices and biochemical parameters. The spectral reflectance curves of plants within the sample plots were recorded simultaneously. All data sources were converted to the same coordinate system, and spatial registration was performed using ground control points to ensure accurate spatial alignment of different data sources, ultimately yielding standardized data.

[0043] Step 130: Based on the standardized data, invert the structural parameters and biochemical parameters respectively.

[0044] Specifically, structural parameters are parameters that reflect the spatial morphology of vegetation, such as tree height, diameter at breast height, and crown width; biochemical parameters are parameters that reflect the chemical properties of vegetation, such as chlorophyll content, water content, and species type.

[0045] In the process of inverting structural parameters, individual trees are segmented from the standardized canopy 3D point cloud data to extract key vegetation structure parameters; statistical analysis and spatial data interpolation are performed on the standardized understory 3D point cloud data to establish a spatial data model of understory vegetation, extract understory structural feature parameters, and obtain structural parameters including the top layer and the understory.

[0046] Specifically, tree segmentation is the operation of dividing the canopy point cloud into individual tree point clouds, which can be a watershed segmentation algorithm based on the canopy height model; key parameters of vegetation structure are important parameters reflecting the morphology of top-level trees, such as tree height, diameter at breast height (DBH), crown width, and canopy density; statistical analysis is the operation of calculating statistical characteristics of understory point cloud data, such as calculating the height mean and standard deviation; spatial data interpolation is the operation of extrapolating the understory data of the entire area from quadrat point cloud data, which can be a Kriging interpolation algorithm; the understory vegetation spatial data model is a model reflecting the spatial distribution and morphology of understory vegetation, which can be a three-dimensional raster model; and understory structural characteristic parameters are parameters reflecting the morphology of understory vegetation, such as height quantiles, point cloud coverage, and non-empty voxels per unit area.

[0047] A watershed segmentation algorithm based on a canopy height model was employed to segment standardized 3D canopy point cloud data into individual tree segments, dividing the canopy point cloud into independent point clouds for each tree. For each tree's point cloud data, key vegetation structure parameters such as tree height, diameter at breast height (DBH), crown width, and canopy density were extracted. Statistical analysis was performed on the standardized understory 3D point cloud data, calculating statistical characteristics such as height mean and standard deviation. Kriging interpolation was used for spatial data interpolation, and a spatial data model of understory vegetation was established based on the interpolation results. Understory structural characteristic parameters, such as height quantiles, point cloud coverage, number of non-empty voxels per unit area, and leaf height diversity index, were extracted from this model. The key vegetation structure parameters of the top layer were integrated with the structural characteristic parameters of the understory to obtain complete structural parameters.

[0048] For example, structural parameters may include: ① Tree height (H): obtained by the difference between the highest point of each tree in the point cloud and the ground point: Where: H: tree height; 𝑍 max : The Z-value of the highest point in the point cloud of this tree; 𝑍 ground ① Ground elevation at this location; ② Plant diameter at breast height (DBH): The industry-standard value of the diameter at breast height (DBH) is adopted as the plant's width at 1.3m above the ground. Point cloud data at 1.3m above the ground is extracted from the laser point cloud data, and the DBH is calculated as D=2R using the least squares circle fitting objective function. , where (x,y) are planar coordinates, and the center (a,b) and radius 𝑅 are obtained by fitting.

[0049] In the process of retrieving biochemical parameters, spectral indices related to vegetation biochemistry are calculated based on the standardized hyperspectral remote sensing images. Mangrove species are classified using machine learning methods, and leaf chemical parameters are retrieved by combining the spectral indices to obtain biochemical parameters that include species type and leaf chemical properties.

[0050] Specifically, spectral indices are mathematical indicators constructed based on reflectance in different wavelength bands, such as chlorophyll index, nitrogen difference index, and water difference index; machine learning methods are algorithms used for species classification, such as support vector machines and random forests; leaf chemical parameters are parameters reflecting the biochemical composition of leaves, such as chlorophyll content, water content, lignin content, and nitrogen content; species types are different plant species in mangroves, such as *Lepidium apetalum* and *Kandelia candel*; and leaf chemical properties are the biochemical characteristics of leaves, reflected by leaf chemical parameters.

[0051] Based on standardized hyperspectral remote sensing imagery, spectral indices related to vegetation biochemistry, such as chlorophyll index, nitrogen difference index, and water difference index, are calculated according to preset formulas. Support vector machine (SVM) algorithm is selected as the machine learning classification method. The band reflectance of the hyperspectral imagery and the calculated spectral indices are used as input features, combined with ground-measured species information as labels, to train a species classification model. The model is then used to classify species in the mangrove area, resulting in a species type distribution map. Using spectral indices as input variables and ground-measured leaf chemical parameters as output variables, an inversion model is established. This model is used to invert leaf chemical parameters such as chlorophyll content, water content, and lignin content. Species types are integrated with their corresponding leaf chemical parameters to form complete biochemical parameters.

[0052] For example, biochemical parameters may include: ① chlorophyll index ②Lignin: ③ Moisture content: In the above formula, R λ The value represents the surface reflectance at wavelength λ (unit: nm). The log in the logarithmic term is the natural logarithm. For different sensors (such as Sentinel-2 and airborne hyperspectral), the nearest neighbor center band can be used to replace the corresponding λ.

[0053] Step 140: Construct a comprehensive inversion model by integrating the structural parameters and biochemical parameters.

[0054] Specifically, the integrated inversion model is a carbon storage calculation model that integrates structural and biochemical parameters, and can be a coupled model that includes the top layer and the understory.

[0055] In constructing the comprehensive inversion model, the key dimensions of vegetation growth data in the structural parameters and the spectral index in the biochemical parameters are input into the fusion modeling algorithm. Corresponding regression coefficients are set for each species to establish a top-level plant carbon storage inversion model. The structural characteristic parameters and diameter parameters of the understory are substituted into the improved allometric growth equation, and corresponding regression coefficients are set for each understory species to establish an understory plant carbon storage inversion model. The top-level and understory plant carbon storage inversion models are integrated to obtain a comprehensive inversion model covering the entire canopy.

[0056] Specifically, key dimension data of vegetation growth are important structural data affecting the growth of top-level plants, such as tree height and diameter at breast height (DBH); fusion modeling algorithm is a modeling method that integrates multi-dimensional data, such as the XGBoost regression algorithm; regression coefficient is a parameter in the model that characterizes the degree of influence of variables, such as the tree height regression coefficient of 0.8 for Kandelia candel; top-level plant carbon storage inversion model is a mathematical model for calculating the carbon storage of top-level trees, which can be a linear regression model that includes structural and biochemical variables; understory structural characteristic parameters are parameters that reflect the morphology of understory vegetation, such as height quantiles and point cloud coverage; ground diameter parameter is the diameter data of understory vegetation at the ground; improved allometric growth equation is an optimized equation relating biomass and morphological parameters, which can be a log-linear equation that introduces multiple structural parameters; understory plant carbon storage inversion model is a mathematical model for calculating the carbon storage of understory vegetation, which can be a model based on the improved allometric growth equation; comprehensive inversion model is a full-canopy carbon storage calculation model that integrates top-level and understory models, which can be a coupled model that includes multiple species-specific sub-models.

[0057] Key dimensions of vegetation growth, such as tree height and diameter at breast height (DBH), were selected from structural parameters, and spectral indices were extracted from biochemical parameters. Using these two types of data as input variables, XGBoost regression was chosen as the fusion modeling algorithm. For each top-layer species in mangroves, corresponding regression coefficients were fitted based on ground-measured data to establish a top-layer plant carbon storage inversion sub-model for each species. All sub-models were then integrated to form a top-layer plant carbon storage inversion model. Understory structural characteristic parameters, such as height quantiles and point cloud coverage, were extracted from structural parameters. Ground diameter parameters of understory vegetation were collected. These two types of parameters were substituted into an improved log-linear allometric growth equation. For each understory species, corresponding regression coefficients were fitted based on ground-measured data to establish a understory plant carbon storage inversion sub-model for each species. All sub-models were then integrated to form a understory plant carbon storage inversion model. The top-layer and understory plant carbon storage inversion models were coupled, clarifying the model calling logic and data connection methods to obtain a comprehensive inversion model covering the entire canopy.

[0058] Furthermore, the fusion modeling algorithm includes random forest regression, XGBoost regression, or a nonlinear regression model based on neural networks; when establishing the top-level plant carbon storage inversion model, tree height and diameter at breast height are used as structural input dimensions, and vegetation biochemically related spectral indices are used as biochemical input dimensions. The model is constructed by fitting regression coefficients to calculate the carbon storage per unit area of ​​the top-level plants.

[0059] Specifically, random forest regression is an ensemble regression algorithm based on multiple decision trees; XGBoost regression is a gradient boosting regression algorithm; a neural network-based nonlinear regression model is a model that fits nonlinear relationships through a neural network, and can be a BP neural network model with one hidden layer; the structural input dimension is a variable in the model that reflects the vegetation structure, such as tree height and diameter at breast height; the biochemical input dimension is a variable in the model that reflects the biochemical characteristics of vegetation, such as chlorophyll index and water difference index; and carbon storage per unit area is the carbon storage of vegetation per unit area.

[0060] XGBoost regression was chosen as the fusion modeling algorithm, with parameters such as a learning rate of 0.1 and a maximum tree depth of 5. The input dimensions of the model were determined: tree height and diameter at breast height (DBH) were selected as structural input dimensions, and chlorophyll index, nitrogen difference index, and water difference index were selected as biochemical input dimensions. Ground-based measured data from multiple sample plots were collected, including tree height, DBH, spectral indices, and corresponding carbon storage per unit area. The data were divided into training and validation sets. The XGBoost regression model was trained using the training set data, fitting the regression coefficients corresponding to each input dimension. The model parameters were adjusted using the validation set data to optimize model performance. After training, the tree height, DBH, and corresponding spectral indices of any top-layer vegetation were input into the model to calculate the carbon storage per unit area of ​​that vegetation, thus completing the construction of the top-layer plant carbon storage inversion model.

[0061] Furthermore, when establishing the understory plant carbon storage inversion model, the height quantile, point cloud coverage, non-empty voxel number per unit area, and leaf height diversity index were extracted from the structural characteristic parameters of the understory. Combined with the ground diameter parameters of the understory vegetation, these parameters were substituted into the improved log-linear allometric growth equation, and the model was constructed by fitting regression coefficients to calculate the carbon storage per unit area of ​​the understory plants.

[0062] Specifically, height quantile is the quantile value of the height of the understory vegetation, which can be the 95th quantile; point cloud coverage is the coverage ratio of point clouds in the understory vegetation, which can be 80%; non-empty voxels per unit area is the number of voxels with vegetation points per unit area, which can be 50 non-empty voxels per square meter; leaf height diversity index is an indicator reflecting the diversity of leaf distribution at different heights, which can be a leaf height diversity index of 1.2; ground diameter parameter is the diameter of the understory vegetation at the ground, which can be 1.5 cm; the improved log-linear allometric growth equation is an allometric growth equation that introduces multiple structural parameters, which can be a logarithmic equation that includes parameters such as height quantile and point cloud coverage; and carbon storage per unit area is the carbon storage per unit area of ​​the understory vegetation.

[0063] The following parameters were extracted from the structural characteristics of the understory: the 95th quantile of height, point cloud coverage, non-empty voxels per unit area, and leaf height diversity index. Ground diameter (DBD) parameters of the understory vegetation were measured using measurement tools, and DBD data for each species were recorded. These four structural characteristics and DBD parameters were used as input variables and substituted into an improved log-linear allometric growth equation. Ground-measured carbon storage per unit area of ​​the understory vegetation was collected, and the input variables were matched with the measured carbon storage data. The regression coefficients in the equation were fitted using the least squares method. The fitted model was validated, and the regression coefficients were adjusted to optimize model accuracy. After model construction, the carbon storage per unit area of ​​any understory vegetation can be calculated by inputting its four structural characteristics and DBD parameters.

[0064] For example, ① the top-level plant carbon storage inversion model can be expressed as: ;

[0065] in: Let be the carbon storage per unit area (i=1, 2,...,n) of top-layer mangrove plant species i in a mangrove forest, 𝛽 𝑖 is the regression coefficient (each species has a set of corresponding regression coefficients), and n is the number of species in the top layer of mangroves in a mangrove forest.

[0066] ②The inversion model for carbon storage in forest understory can be expressed as:

[0067] Among them, h P95 95th percentile in height Zi represents the point cloud height; the average height h mean , σ h Standard deviation PP: Point cloud coverage h0 is the set height threshold; FHD: leaf height diversity index. This is used to divide the height layer into n intervals and calculate the point ratio n. 𝑖 Nv / A: Non-empty voxels per unit area. The volume of a quadrat is divided into voxels, and the number of voxels that fall into the quadrat is counted to reflect the vegetation fill rate. A is the total number of points in the 3D point cloud.

[0068] Taking the logarithm, we get:

[0069] in: It is the carbon storage per unit area of ​​mangrove plant species j under a mangrove forest (j=1, 2, ... , m). is the regression coefficient, h is the understory plant height fitted based on the handheld lidar of the quadrat, and m is the number of understory mangrove plant species in a mangrove forest.

[0070] ③ The comprehensive inversion model can be expressed as:

[0071] +

[0072] in, This is a predicted value for the total carbon storage of a mangrove forest. It is the area of ​​the top layer mangrove species n in a mangrove forest. It is the area of ​​a mangrove forest with m species in the understory, containing i top-level species and j understory species.

[0073] Step 150: Calculate the carbon storage of mangroves using the integrated inversion model.

[0074] In calculating mangrove carbon storage, the comprehensive inversion model is used to calculate the carbon storage per unit area of ​​top-layer plants and understory plants respectively; combined with the distribution area of ​​each species in the mangrove area, the total carbon storage of top-layer plants and the total carbon storage of understory plants are calculated respectively; the two are superimposed to obtain the aboveground carbon storage of mangroves, and the total carbon storage of mangroves is calculated by combining the fixed ratio between underground biomass and aboveground carbon storage.

[0075] Specifically, the total carbon storage of top-layer plants is the sum of the carbon storage of all top-layer plants in the mangrove forest; the total carbon storage of understory plants is the sum of the carbon storage of all understory plants in the mangrove forest; the distribution area is the area occupied by each species in the mangrove forest area; the aboveground carbon storage is the sum of the carbon storage of the aboveground parts of the mangrove forest; the underground biomass is the biomass of the underground parts of the mangrove forest; the fixed ratio is the ratio of underground biomass to aboveground carbon storage, which can be 40% of the aboveground carbon storage; and the total carbon storage is the sum of the carbon storage of the aboveground and underground parts of the mangrove forest.

[0076] The carbon storage per unit area of ​​each top-layer plant species was calculated by using the top-layer plant carbon storage inversion sub-model of the integrated inversion model, inputting the structural and biochemical parameters of each top-layer species. The carbon storage per unit area of ​​each understory plant species was calculated by using the understory plant carbon storage inversion sub-model, inputting the structural and diameter parameters of each understory species. The distribution area of ​​each species in the mangrove area was obtained through remote sensing image interpretation or field surveys. The carbon storage per unit area of ​​each species was multiplied by its distribution area, and the sum was obtained to obtain the total carbon storage of the top-layer plants and the total carbon storage of the understory plants. The total carbon storage of the top-layer and understory plants was superimposed to obtain the aboveground carbon storage of the mangroves. The underground biomass was calculated according to a fixed ratio, with underground biomass being 40% of the aboveground carbon storage. The aboveground carbon storage and underground biomass were added together to obtain the total carbon storage of the mangroves.

[0077] For example, ① Total aboveground biomass (TC) = aboveground biomass of mangrove top cover (TC) High )+Tree aboveground biomass (TC) Low );

[0078] ② Total aboveground vegetation carbon storage = aboveground biomass of mangrove top cover (TC) High *Top-layer plant carbon content coefficient + Understory aboveground biomass (TC) Low * Carbon content coefficient of understory vegetation;

[0079] ③ Total carbon storage = Aboveground carbon storage + Belowground carbon storage = 1.4 * Aboveground carbon storage (Setting: Belowground biomass: B) BLOW =40% * ).

[0080] For example, setting This represents the actual value of the total carbon storage of plants within the quadrat area, where k is the quadrat number and the total number of quadrats is K. This value is determined by combining the allometric growth equation with vegetation carbon content, and the specific formula is as follows:

[0081] ① Equation for calculating mangrove tree biomass:

[0082] General equation for calculating the biomass of mangrove top-plant species i: ,in These are the fitting coefficients (each species corresponds to its own set of fitting coefficients), where H is the tree height (unit: m) and D is the diameter at breast height (unit: cm). Wood density (unit: g / cm³), B High Biomass (unit: kg).

[0083] General equation for calculating biomass of mangrove understory shrubs and low-growing plants:

[0084] General equation for calculating biomass of shrub species j: ;

[0085] Where α is the δ fitting coefficient (each species has its own set of fitting coefficients), L is the canopy depth of the shrub (cm), and h l Where W is the height of the shrub (cm), and W is the area of ​​the elliptical canopy (cm). m² ), B Low Shrub biomass (kg).

[0086] W = π × (W1 / 2) × (W2 / 2), where W1 is the widest crown width passing through the center of the plant's canopy, and W2 is the crown width perpendicular to W1.

[0087] ② The carbon storage in this example is calculated using the following formula: Carbon storage = Biomass × Carbon content coefficient.

[0088] Therefore, the carbon storage of a single tree i plant within the sample plot is... :

[0089] Carbon storage of a single shrub plant in the quadrat: Where β It is the carbon content coefficient of arborescent plants, γ It is the carbon content coefficient of shrub plants, determined using an elemental analyzer.

[0090] Carbon content of aboveground plants in the sample plot: p represents the number of trees in the quadrat, q represents the number of shrubs in the quadrat, i and j represent the species types of trees and shrubs, and k represents the quadrat number.

[0091] ③ In the mangrove ecosystem, the total carbon content of trees is:

[0092] In the formula, l is the quadrat number (K ​​quadrats in total), w is the tree species type (i species in total), and y is the number of individuals of species i in quadrat l (y = 1, 2, ..., n). l,i If species i is not present in quadrat l, then n i,l =0), Area w Let w be the area of ​​species w in the ecosystem, and area be the area of ​​a quadrat.

[0093] In mangrove ecosystems, the total carbon content of shrubs is:

[0094] In the formula, z is the quadrat number (K ​​quadrats in total), x is the shrub species type (j species in total), and v is the number of individuals of species j in quadrat z (v=1,2,…,n). z,j If species j is not present in quadrat z, then n z,j =0), Areax Let x be the area of ​​species x in the ecosystem, and area be the quadrat area.

[0095] Total aboveground biomass of mangrove ecosystems .

[0096] Step 160: Verify the accuracy of the integrated inversion model using ground-based measured data, and output the final carbon sink estimation result.

[0097] Specifically, the carbon sink estimation result is the total carbon storage data of the mangrove ecosystem, which can be the total carbon storage value per unit area or region.

[0098] In verifying accuracy, the actual carbon storage is calculated by combining the allometric growth equation with the measured biomass and carbon content coefficients on the ground. The root mean square error, mean absolute error, and coefficient of determination are used to compare the actual carbon storage with the carbon storage predicted by the model. The model fitting effect is evaluated based on the index results. If the coefficient of determination is close to the preset threshold, the verification is completed. Otherwise, the model parameters are adjusted and retrained until the model meets the accuracy requirements.

[0099] Specifically, the allometric growth equation is an equation reflecting the relationship between biomass and morphological parameters, and can be the allometric growth equation for biomass of mangrove trees; actual carbon storage is the real carbon storage calculated from measured data; root mean square error is an indicator reflecting the degree of deviation between predicted and actual values; mean absolute error is an indicator reflecting the average deviation between predicted and actual values; coefficient of determination is an indicator reflecting the goodness of fit of the model, and can be 0.95; preset threshold is the standard for judging the model's qualification based on the coefficient of determination, and can be 0.85; model parameters are the adjustable variables in the model, and can be regression coefficients, learning rates, etc.

[0100] Ground-measured data from multiple independent mangrove plots were collected, including morphological parameters such as tree diameter at breast height (DBH), tree height, and ground diameter. Biomass for each plot was calculated using the allometric growth equation. Combined with measured carbon content coefficients, the actual carbon storage for each plot was calculated. A comprehensive inversion model was used to predict the carbon storage for these plots, yielding the predicted carbon storage. The actual and predicted carbon storage were paired, and three accuracy indices—root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (CCD)—were calculated. A preset threshold of 0.85 was set for the CCD. If the calculated CCD was close to or greater than 0.85, the model showed good fit, and accuracy verification was completed. If the CCD was less than 0.85, the sources of error were analyzed, and parameters such as regression coefficients and input variable weights in the model were adjusted. The model was retrained and validated again until the CCD reached the preset threshold, indicating that the model met the accuracy requirements.

[0101] For example, regarding the remote sensing-predicted aboveground biomass TC of mangroves pred And the actual aboveground biomass TC of mangroves calculated using the allometric growth equation.obs The three commonly used formulas for accuracy verification are as follows: ; ;

[0102] In the formula, TC represents the mean total plant carbon storage in H independent mangrove ecosystems, where H is the number of validation samples (i.e., H different mangrove ecosystems). pred TC is the model's predicted value. obs This represents the actual value calculated using the allometric growth equation. The fit is judged based on the values ​​of three indicators, including R0. 2 The closer the value is to 1, the closer the model's predicted value is to the true value, which means the better the fit.

[0103] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0104] The above embodiments are merely illustrative of the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made based on the essence of the content of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A composite carbon sink estimation method based on multimodal mangrove canopy data, characterized in that, Including the following steps: Acquire lidar point cloud data, hyperspectral remote sensing images, and ground-measured data for mangrove areas; The lidar point cloud data, hyperspectral remote sensing images, and ground-measured data are preprocessed to obtain standardized data. Structural parameters and biochemical parameters were retrieved based on the standardized data. A comprehensive inversion model is constructed by integrating the structural and biochemical parameters. The carbon storage of mangroves was calculated using the comprehensive inversion model described above. The accuracy of the comprehensive inversion model is verified by ground-based measured data, and the final carbon sink estimation result is output.

2. The method according to claim 1, characterized in that, When acquiring data, airborne or drone-based lidar was used to scan the mangrove area to obtain three-dimensional point cloud data of the canopy; three-dimensional point cloud data of the mangrove understory vegetation was collected using handheld lidar; hyperspectral remote sensing images of a specific spectral range were acquired; sample plots were set up in the mangrove understory vegetation area, and the plant height and spatial distribution data of the understory vegetation were extracted using handheld lidar. Tree diameter at breast height was obtained using measurement tools, and organ samples of different tree species were collected. Carbon content coefficients were obtained through laboratory analysis, resulting in ground-based measured data containing vegetation growth information and carbon content.

3. The method according to claim 1, characterized in that, During preprocessing, the lidar point cloud data is filtered and topographically corrected to construct a canopy height model; the hyperspectral remote sensing image is subjected to strip stitching, atmospheric correction, geometric fine correction, and band dimensionality reduction; effective bands of the hyperspectral image are selected based on spectral index correlation, and the spectral reflectance curves of plants within the sample plots are recorded; the lidar point cloud data, hyperspectral remote sensing image, and ground-measured data are coordinate unified and spatially registered to obtain standardized data.

4. The method according to claim 1, characterized in that, When retrieving structural parameters, individual trees are segmented from the standardized canopy 3D point cloud data to extract key vegetation structure parameters; statistical analysis and spatial data interpolation are performed on the standardized understory 3D point cloud data to establish a spatial data model of understory vegetation, extract understory structural feature parameters, and obtain structural parameters including the top layer and understory.

5. The method according to claim 1, characterized in that, When retrieving biochemical parameters, spectral indices related to vegetation biochemistry are calculated based on the standardized hyperspectral remote sensing images; mangrove species are classified using machine learning methods, and leaf chemical parameters are retrieved by combining the spectral indices to obtain biochemical parameters that include species type and leaf chemical attributes.

6. The method according to claim 1, characterized in that, When constructing the comprehensive inversion model, the key dimensions of vegetation growth data in the structural parameters and the spectral index in the biochemical parameters are input into the fusion modeling algorithm. Corresponding regression coefficients are set for each species to establish a top-level plant carbon storage inversion model. The structural characteristic parameters and diameter parameters of the understory are substituted into the improved allometric growth equation, and corresponding regression coefficients are set for each understory species to establish an understory plant carbon storage inversion model. The top-level and understory plant carbon storage inversion models are integrated to obtain a comprehensive inversion model covering the entire canopy.

7. The method according to claim 6, characterized in that, The fusion modeling algorithm includes random forest regression, XGBoost regression, or a nonlinear regression model based on neural networks. When establishing the top-level plant carbon storage inversion model, tree height and diameter at breast height are used as structural input dimensions, and vegetation biochemically related spectral indices are used as biochemical input dimensions. The model is constructed by fitting regression coefficients to calculate the carbon storage per unit area of ​​the top-level plants.

8. The method according to claim 6, characterized in that, When establishing a carbon storage inversion model for understory vegetation, height quantiles, point cloud coverage, non-empty voxels per unit area, and leaf height diversity index are extracted from the structural characteristic parameters of the understory. Combined with the ground diameter parameters of the understory vegetation, the improved log-linear allometric growth equation is substituted into the model, and the carbon storage per unit area of ​​understory vegetation is calculated by fitting regression coefficients.

9. The method according to claim 1, characterized in that, When calculating the carbon storage of mangroves, the carbon storage per unit area of ​​top-layer plants and understory plants is calculated using the comprehensive inversion model. The total carbon storage of top-layer plants and the total carbon storage of understory plants are calculated by combining the distribution area of ​​each species in the mangrove area. The aboveground carbon storage of mangroves is obtained by superimposing the two data, and the total carbon storage of mangroves is calculated by combining the fixed ratio between underground biomass and aboveground carbon storage.

10. The method according to claim 1, characterized in that, When verifying accuracy, the actual carbon storage is calculated by combining the allometric growth equation with the measured biomass and carbon content coefficients on the ground; the root mean square error, mean absolute error and coefficient of determination are used to compare the actual carbon storage with the carbon storage predicted by the model; the model fitting effect is evaluated based on the index results. If the coefficient of determination is close to the preset threshold, the verification is completed. Otherwise, the model parameters are adjusted and retrained until the model meets the accuracy requirements.

Images