Method and system for precise inversion of urban vegetation aboveground biomass considering vegetation type

Abstract

The application provides a kind of urban vegetation aboveground biomass accurate inversion method and system considering vegetation type, carry out research area vegetation aboveground biomass calculation, including corresponding biomass calculation after plot sampling according to different research areas;High-resolution optical data and spaceborne photon counting lidar data preprocessing, fine extraction of urban vegetation information, high-resolution remote sensing image data spectral feature extraction and optical-biomass inversion sub-model construction, vegetation height data extraction of spaceborne photon counting lidar data and spaceborne LiDAR-biomass inversion sub-model construction;Under the support of fine classification information of urban surface vegetation, integrate biomass inversion sub-model, obtain integrated inversion model by training urban vegetation biomass sample data, which fuses urban surface vegetation horizontal information and three-dimensional structure information, and carry out large-scale urban vegetation aboveground biomass accurate mapping.

Description

Technical Field

[0001] This invention relates to the field of urban vegetation aboveground biomass inversion, and more specifically, to a scheme for accurate inversion of urban large-scale vegetation aboveground biomass that takes into account vegetation type. Background Technology

[0002] Global climate change has profoundly impacted human production and lives, with issues such as global warming, urban flooding, and air pollution hindering human development. Based on its own national conditions, China has pledged to achieve carbon peaking by 2030 and carbon neutrality by 2060, aiming to improve energy efficiency and promote green energy development by reducing carbon emissions and increasing carbon sequestration. Vegetation, as a crucial green infrastructure within natural resources, plays a vital role in "collecting" carbon from the atmosphere through its carbon storage capacity. In recent years, with a deeper understanding of major vegetation ecosystems such as forests, grasslands, and wetlands, methods for measuring the carbon storage capacity of large-scale homogeneous vegetation distributions using biomass have been preliminarily mastered. With the rise of concepts like "green cities" and "low-carbon cities," the carbon storage capacity of scattered vegetation in cities is receiving increasing attention. Understanding its distribution, biomass calculations, and the status and trends of related ecological processes across multiple temporal and spatial scales is crucial for building low-carbon green cities and a key step in reducing errors in global carbon sink estimation.

[0003] The vegetation composition of forests and grasslands is relatively simple, and their biomass estimation can generally be carried out by: delineating large-scale quadrats, conducting sampling surveys to estimate vegetation parameters, calculating quadrat biomass based on vegetation growth equations, and thus obtaining the biomass of the entire forest or grassland area. This method is suitable for surfaces with uniform and relatively simple vegetation distribution, such as forests and grasslands, and performs well in these areas. However, due to human planning and environmental factors, urban vegetation distribution often does not conform to the first law of geography, frequently exhibiting fragmentation (primarily woodland areas, enriched by shrubs and grasslands, while also considering landscape visibility). It possesses complex and dynamic characteristics, and the ecosystem services it provides depend on vegetation type, composition, and local environmental conditions. The diverse vegetation composition and scattered distribution make it impossible to rely on simple prior knowledge to determine urban vegetation types and perform simple calculations during urban biomass inversion. Therefore, aboveground biomass inversion methods for forest and grassland scenarios are not suitable for aboveground biomass assessment within cities. Meanwhile, the size and distribution of vegetation patches are insufficient to meet the sampling needs of large-scale vegetation ecosystems such as forests or grasslands. Therefore, there is an urgent need for methods to extract information on the distribution of different types of vegetation in cities, so as to obtain supporting data for biomass calculation.

[0004] Biomass inversion is generally performed using remote sensing estimation methods. After destructive sampling within a certain area, allometric growth equations are derived through empirical summarization to calculate biomass and establish the correlation between plot data and remote sensing characteristic variables. Optical remote sensing data can provide unique canopy spectral information of surface vegetation to analyze the growth status of different vegetation types and perform biomass inversion. LiDAR data can provide three-dimensional structural information of vegetation, such as vegetation height, for remote sensing biomass estimation. While optical data provides continuous surface information, LiDAR data is often limited by data acquisition methods and measurement costs, limiting measurements to small areas. Therefore, the data area limitation of LiDAR becomes a problem when performing biomass inversion at the urban, national, or even global scale. Spaceborne LiDAR data can achieve large-scale acquisition of three-dimensional vegetation information, enabling urban vegetation biomass inversion and carbon sink estimation.

[0005] In addition, biomass can be inverted using sample plot data or spaceborne LiDAR data. These two types of data provide horizontal and three-dimensional structural information of vegetation, respectively. How to integrate the information expressed by the two in urban application scenarios to improve the accuracy of biomass inversion is also an urgent problem to be solved in urban biomass estimation. At present, there are no reports on large-scale urban vegetation aboveground biomass inversion after integrating the two types of information. Summary of the Invention

[0006] To address the above problems, this invention provides a novel method for inverting urban vegetation aboveground biomass, achieving a precise inversion method for large-scale urban vegetation aboveground biomass by integrating spaceborne photon-counting lidar data and high-resolution optical imagery. This invention achieves precise extraction of urban vegetation type information through high-resolution optical imagery, acquires large-scale urban vegetation height through spaceborne photon-counting lidar data, and achieves precise inversion of urban vegetation biomass by integrating spaceborne LiDAR data and optical data, thus filling the gap in urban vegetation aboveground biomass inversion.

[0007] This invention provides a method for accurate inversion of aboveground biomass of urban vegetation that takes into account vegetation type, comprising the following steps:

[0008] Step a, Biomass calculation of sample plots in the study area, including using different allometric growth models, weighing methods or other calculation methods according to the different main urban vegetation types in different study areas to calculate the biomass after sample plot sampling;

[0009] Step b, high-resolution optical data and spaceborne photon counting lidar data preprocessing, including geometric and radiometric correction of high-resolution optical images, and cleaning and data extraction of spaceborne photon counting lidar data.

[0010] Step c, refined extraction of urban vegetation information, including obtaining the dominant vegetation and its characteristics in the city through field sampling and various statistical data, defining the urban surface land use types separately, extracting the characteristics of different types of urban vegetation, and performing refined classification of urban vegetation types to provide basic data for accurate estimation of urban vegetation aboveground biomass.

[0011] Step d involves extracting spectral features from high-resolution remote sensing image data and constructing an optical-biomass inversion sub-model. This includes using spectral features as independent variables and biomass as dependent variables, constructing the relationship between features and biomass through sample points, and using the sample plot biomass data calculated in step a and the fine classification information of urban surface vegetation obtained in step c. Based on the obtained spectral features, a regression model with better estimation performance is selected to construct an optical-biomass quantum model, providing a sub-model for integrated inversion.

[0012] Step e, vegetation height data extraction from spaceborne photon counting lidar data and construction of spaceborne LiDAR-biomass inversion sub-model, includes: after filtering and denoising urban adaptive photon point cloud data, determining the window size for vegetation height data statistics based on the high-resolution image resolution, and extracting vegetation height features; using the sample plot biomass data calculated in step a and the urban surface vegetation fine classification information obtained in step c, selecting a regression model with better estimation performance based on the obtained vegetation height features to construct a spaceborne LiDAR-bioquantum model, providing a sub-model for integrated inversion;

[0013] Step f, large-scale urban surface vegetation aboveground biomass inversion, includes integrating the biomass inversion sub-models obtained in steps d and e with the support of the fine classification information of urban surface vegetation obtained in step c, and training the integrated inversion model that integrates the horizontal information and three-dimensional structural information of urban surface vegetation using the urban vegetation biomass quadrat data obtained in step a, and performing large-scale accurate mapping of urban vegetation aboveground biomass.

[0014] Furthermore, in step c, the urban surface category is first classified, and the urban surface category is defined as impermeable surface LU. i Water bodies LU w , bare ground LU s Vegetation LU v Superpixel segmentation is performed on the high-resolution multispectral data of the study area obtained in step b. For the defined urban surface category objects, spectral, textural, and other features are extracted to maximize the inter-class differences and minimize the intra-class differences between different land use categories. Land use types are then determined by combining different influencing factors, as expressed in the following expression:

[0015] LU = a × f(Feature)S )+b×f(Feature T )+(1-ab)×f(Feature O )

[0016] Where, f(Feature) S ) is a spectral feature discrimination model, f(Feature) T ) is a texture feature discrimination model, f(Feature) O Other feature discrimination models, where a, b, and c are the weights of the three discrimination models respectively;

[0017] And according to LU v The scope was further divided, and urban vegetation types within the study area were refined based on dominant tree species. Spectral, textural, and other information of different typical vegetation types were extracted. The vegetation type expression in the refined urban vegetation type extraction process is as follows:

[0018] Furthermore, the optical-biomass inversion sub-model constructed in step d is implemented by using spectral features. S Using spectral characteristics as the independent variable and biomass AGB as the dependent variable, the relationship between spectral characteristics and biomass is simulated. The optical-biological quantum model AGB is obtained by solving for the spectral characteristics and biomass of the sample points as the true values. S =f(Feature) S In addition, during the model selection process, different regression analysis methods are compared to select the better model.

[0019] Furthermore, the implementation of the spaceborne LiDAR-biomass inversion sub-model constructed in step e is as follows: using the spaceborne LiDAR vegetation height feature... L Using vegetation height as the independent variable and biomass as the dependent variable, the relationship between vegetation height and biomass in spaceborne LiDAR was simulated. Using sample points as the true values, the model was solved using the eigenvalues ​​of vegetation height and biomass in spaceborne LiDAR, resulting in the spaceborne LiDAR-biological quantum model AGB. L =f(Feature) L In addition, during the model selection process, different regression analysis methods are compared to select the better model.

[0020] Moreover, the implementation of the spaceborne LiDAR-biomass inversion sub-model constructed in step f is as follows: the model with better performance obtained in steps d and e is used as the independent variable, biomass is used as the dependent variable, and the vegetation category obtained in step c is used as the constraint condition. The relationship between several models and biomass is simulated through decision tree, and the sample points are used as the true values. The simulated biomass of several models and the real biomass are solved to obtain the spaceborne LiDAR-optical integrated model.

[0021] On the other hand, a method for accurately inverting aboveground biomass of urban vegetation that takes into account vegetation type, as described above, is used.

[0022] Moreover, it includes the following modules,

[0023] The first module is used for biomass calculation of sample plots within the study area. It includes calculating the biomass of sample plots after sampling, based on the different main vegetation types of cities in different study areas, using allometric growth models, weighing methods, and other calculation methods.

[0024] The second module is used for high-resolution optical data and spaceborne photon counting lidar data preprocessing, including geometric and radiometric correction of high-resolution optical images, and cleaning and data extraction of spaceborne photon counting lidar data.

[0025] The third module is used to extract detailed information on urban vegetation in the study area, including the extraction of features of different types of urban vegetation and the fine classification of urban vegetation types, providing basic data for accurate estimation of aboveground biomass of urban vegetation.

[0026] The fourth module is used for spectral feature extraction of high-resolution remote sensing image data and construction of optical-biomass inversion sub-models. It includes the sample plot biomass data calculated by the first module and the fine classification information of urban surface vegetation obtained by the third module. Based on the obtained spectral features, the regression model with better estimation effect is selected to construct the optical-biomass quantum model, providing a sub-model for integrated inversion.

[0027] The fifth module is used for vegetation height data extraction from spaceborne photon counting lidar data and construction of a spaceborne LiDAR-biomass inversion sub-model. This includes filtering and denoising urban adaptive photon point cloud data, determining the window size for vegetation height data statistics based on the high-resolution image resolution, extracting vegetation height features, and using the sample plot biomass data calculated by the first module and the fine classification information of urban surface vegetation obtained by the third module. Based on the obtained vegetation height features, a regression model with better estimation performance is selected to construct a spaceborne LiDAR-bioquantum model, providing a sub-model for integrated inversion.

[0028] The sixth module is used for large-scale urban surface vegetation aboveground biomass inversion. It includes integrating the biomass inversion sub-models obtained from the fourth and fifth modules with the support of the fine classification information of urban surface vegetation obtained in the third module through the ensemble model constructed by decision tree, and training the ensemble inversion model that integrates the horizontal information and three-dimensional structural information of urban surface vegetation through the urban vegetation biomass quadrat data obtained in the first module, so as to carry out large-scale accurate mapping of urban vegetation aboveground biomass.

[0029] Alternatively, it may include a processor and a memory, the memory being used to store program instructions, and the processor being used to invoke the stored instructions in the memory to execute, as described above, a method for accurately inverting urban vegetation aboveground biomass that takes into account vegetation type.

[0030] Alternatively, it may include a readable storage medium storing a computer program that, when executed, implements a method for accurately inverting urban vegetation aboveground biomass that takes into account vegetation type, as described above.

[0031] This invention, in constructing a biomass inversion model jointly based on spaceborne photon-counting lidar data and optical data, proposes a biomass inversion method that combines vegetation 3D structural information and horizontal optical features after separately performing vegetation vertical structure parameter-biomass inversion and planar optical parameter-biomass inversion. Specifically, it constructs a sub-model based on the influence of vegetation 3D structural parameters and horizontal optical features acquired by spaceborne photon-counting lidar on biomass. Then, a decision-level fusion of the spaceborne LiDAR-biomass inversion results and the optical-biomass inversion results is performed using a decision tree-based stacking ensemble model to determine their influence on biomass and the joint mechanism. This invention combines vegetation 3D structural information acquired by spaceborne LiDAR with the optical features of high-resolution remote sensing imagery to construct a biomass remote sensing inversion model that combines vegetation height and other vegetation 3D structural parameters with optical features. This provides a new data source for urban biomass inversion, leveraging the advantages of combining spaceborne LiDAR data and optical data in urban vegetation biomass inversion to achieve biomass inversion in urban areas. Attached Figure Description

[0032] Figure 1 This is a flowchart of an embodiment of the present invention. Detailed Implementation

[0033] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0034] Unlike previous inversion methods, this invention introduces vegetation height data and high-resolution optical data acquired by ICESat-2, and performs regression modeling with urban biomass. Within the extracted urban vegetation area, appropriate vegetation height factors and optical features are selected to construct a spaceborne LiDAR-biomass inversion model and an optical-biomass inversion model. Through sensitivity analysis of the two models, the weights of the two in jointly performing biomass inversion are determined, and an urban aboveground biomass inversion model integrating spaceborne LiDAR and optical features is established.

[0035] See Figure 1 The present invention provides a method for accurate inversion of urban aboveground biomass over a large area, taking into account vegetation type. This method integrates spaceborne LiDAR and optical features for urban aboveground biomass inversion, and its specific implementation includes the following steps:

[0036] Step a: Calculate the aboveground biomass of the sample points and quadrats in the study area. This includes using quadrat biomass calculation methods (such as different allometric growth models, weighing methods, etc.) based on the main urban vegetation types in different study areas to calculate the biomass after sampling.

[0037] Taking trees as an example, within the study area, a dataset was constructed by collecting vegetation parameters such as height, diameter at breast height (DBH), tree species, and quantity based on sampling and data planning. The aboveground biomass of a single tree was calculated using Equation 1, and then the plot biomass was obtained by calculating the ratio of recorded vegetation biomass within the plot to the plot area.

[0038]

[0039] Where D is the diameter at breast height (DBH) of the vegetation (cm), H is the tree height (m), and W is the diameter at breast height (DBH). S For trunk biomass, W B For branch biomass, W L Let represent leaf biomass, a1, a2, a3, b1, b2, and b3 be the corresponding coefficients, and W be the total aboveground biomass.

[0040] Step b involves preprocessing high-resolution optical data and spaceborne photon-counting lidar data, including geometric and radiometric correction of the high-resolution optical images and cleaning and data extraction of the spaceborne photon-counting lidar data.

[0041] In this embodiment, high-resolution optical images and ICESat-2ATL03 data of the study area are preprocessed.

[0042] Furthermore, geometric correction, radiometric calibration, and orthorectification are performed on the multispectral data of the high-resolution remote sensing image. Atmospheric correction is then performed on the multispectral image using FLAASH based on parameters such as the high-resolution image sensor type, the geographical location of the imaging center, imaging time, and altitude information to obtain surface reflectance data. Simultaneously, panchromatic data is preprocessed and fused with the multispectral data to obtain high-resolution multispectral data, providing data support for the fine extraction of urban vegetation in step c.

[0043] Furthermore, for spaceborne photon counting lidar data (the example uses ICESat-2 ATL03 data), photon point data for the corresponding strips within the study area are first acquired. After cleaning for null values ​​and outliers, the photon point elevations and geographic coordinates of each photon point within the study area are obtained. For the acquired ATL03 raw point cloud data, its maximum and minimum elevation values ​​are calculated and denoted as E. max and E min Based on the natural discontinuity method, the statistical interval is set to count the frequency of photons at elevation points; and the point cloud density is calculated as a threshold for coarse point cloud denoising.

[0044] Furthermore, after coarse denoising of the point cloud, fine denoising is performed based on the distance between photon points. First, the distance between photon points is calculated based on the latitude, longitude, and corresponding elevation data obtained from the satellite data. Since some photon points are very close together, to avoid rounding errors between closely spaced points, this example calculates the latitude and longitude distance d between two photon points using Equation 2. L :

[0045] d L =R*cos{1 / [sin y1sin y2 + cos y1cos y2cos(x2-x1)]} (Formula 2)

[0046] Where R is the Earth's radius; y1 and y2 are the latitudes of the two points; and x1 and x2 are the longitudes of the two points.

[0047] Then, the Euclidean distance between any two photon points is calculated using elevation data. (Δh is the elevation difference between the two points), and calculate their maximum distance d. max Minimum distance d minThe difference between the maximum and minimum distances, Δd, is used to determine the segment interval. The frequency of the distance D between photon points in each interval is counted, and the interval with the highest frequency is set as the neighborhood EPS value. Then, based on the calculated minimum neighborhood value EPS, denoising experiments on ICESAT-2ATL03 data are conducted using denoising algorithms such as DBSCAN. The optimal MinPts is determined through multiple experiments. Photon points clustered within the point cloud clusters are saved, while photon points determined to be outside the point cloud clusters are considered noise and removed, resulting in usable height data.

[0048] Step c involves the detailed extraction of urban vegetation information in the study area. This includes obtaining the dominant vegetation and its characteristics in the city through field sampling and various statistical data, defining the urban surface land use types separately, extracting the characteristics of different types of urban vegetation, and performing detailed classification of urban vegetation types to provide basic data for accurate estimation of urban vegetation aboveground biomass.

[0049] In this embodiment, the dominant vegetation types within the study area are obtained in advance through surveys and literature review. Based on the field survey data obtained in step a, the urban surface categories are first classified, and the urban surface category is defined as impermeable surface LU. i Water bodies LU w , bare ground LU s Vegetation LU v The high-resolution multispectral data of the study area obtained in step b is used for superpixel segmentation. For the defined urban land surface categories, spectral, textural, and other features are extracted to maximize inter-class differences and minimize intra-class differences. Land use types are then determined by combining different influencing factors. The linear algorithm expression is as follows:

[0050] LU = a × f(Feature) S )+b×f(Feature T )+(1-ab)×f(Feature O (Equation 3)

[0051] Where, f(Feature) S ) is a spectral feature discrimination model, f(Feature) T ) is a texture feature discrimination model, f(Feature) O Other feature discrimination models, where a, b, and c are the weights of the three discrimination models respectively.

[0052] And according to LU vThe study area was further subdivided, and urban vegetation types within the study area were refined based on dominant tree species. Spectral, textural, and other information of different typical vegetation types was extracted. The vegetation type expression in the refined urban vegetation type extraction process is as follows:

[0053] in, For spectral feature discrimination model, For texture feature discrimination model, Other feature discrimination models, a v b v and c v The weights assigned to each of the three discrimination models are given.

[0054] Furthermore, in the process of extracting sample features, the spectral features selected in this example include Normalized Difference Vegetation Index (NDVI), Difference Vegetation Index (DVI), Ratio Vegetation Index (RVI), Green Chlorophyll Index (CIgreen), Modified Red Edge Normalized Difference Vegetation Index (MNDVI), Green Normalized Difference Vegetation Index (GNDVI), Enhanced Vegetation Index (EVI), Soil-Adjusted Vegetation Index (SAVI), and Modified Soil-Adjusted Vegetation Index (MSAVI), and their calculation formulas are shown in Equation 4-12:

[0055]

[0056] DVI=NIR-R (Formula 5)

[0057]

[0058]

[0059]

[0060]

[0061]

[0062]

[0063]

[0064] In the formula, NIR, R, G, and B are the reflectivities of the near-infrared band, red band, green band, and blue band, respectively; L is the canopy background adjustment coefficient.

[0065] In this example, the texture feature selected is the band average value for each object. Brightness b, standard deviation σ L The length / width ratio γ, shape index s, and density d of the image object are given by formulas 13-18.

[0066]

[0067]

[0068]

[0069]

[0070]

[0071]

[0072] Where n is the number of pixels that make up an image object. The value of n for each pixel in this layer L The number of layers containing spectral information. The layer contains the average value of spectral information, where a is the bounding box length of the object, b is the bounding box width, f is the fill degree of the bounding box, A is the area of ​​the image object, e is the boundary length of the image, X is the x-coordinate of all pixels constituting the image object, Y is the y-coordinate of all pixels constituting the image object, and Var(X) and Var(Y) are the variances.

[0073] Step d involves extracting spectral features from high-resolution remote sensing image data and constructing an optical-biomass inversion sub-model. This includes using spectral features as independent variables and biomass as dependent variables, constructing the relationship between features and biomass through sample points, and using the sample plot biomass data calculated in step a and the fine classification information of urban surface vegetation obtained in step c. Based on the obtained spectral features, a regression model with better estimation performance is selected to construct an optical-biomass quantum model, providing a sub-model for integrated inversion.

[0074] Furthermore, the method for constructing the optical-biomass inversion sub-model in step d is as follows: using spectral features... S Using spectral characteristics as the independent variable and biomass AGB as the dependent variable, the relationship between spectral characteristics and biomass is simulated. The sample points are used as the true values, and the optical-biological quantum model AGB is obtained by solving for the spectral characteristics and biomass. S =f(Feature) S Meanwhile, during the model selection process, different regression analysis methods were compared, and the better models were selected and defined as AGB. S1 =f(Feature) S ), AGB S2 =f(Feature) S ...

[0075] Step e involves extracting vegetation height data from spaceborne photon-counting lidar data and constructing a spaceborne LiDAR-biomass inversion sub-model. This includes filtering and denoising urban adaptive photon point cloud data, determining the window size for vegetation height data statistics based on the high-resolution image resolution, extracting vegetation height features, and using the sample plot biomass data calculated in step a and the fine classification information of urban surface vegetation obtained in step c. Based on the obtained vegetation height features, a regression model with better estimation performance is selected to construct a spaceborne LiDAR-bioquantum model, providing a sub-model for integrated inversion.

[0076] In this embodiment, vegetation height data extraction from ICESat-2 ATL03 data and construction of a spaceborne LiDAR-biomass inversion sub-model are performed. Urban adaptive ICESat-2 data filtering and denoising are carried out. The window size for vegetation height data statistics is determined according to the resolution of high-resolution images, and vegetation height features are extracted. The biomass data of the sample plots calculated by the first module and the fine classification information of urban surface vegetation obtained by the third module are used to select a regression model with better estimation effect based on the obtained vegetation height features to construct a spaceborne LiDAR-bioquantum model, providing a sub-model for integrated inversion.

[0077] Furthermore, based on the resolution of the optical image, this example constructs a 4m*4m window to statistically analyze the height of the spaceborne LiDAR data within the window. Based on the obtained vegetation height, this example selects the 98th percentile height value h, the central canopy height h_centroid, the difference between the 98th percentile canopy height value and the median canopy height h_dif, the average canopy height h_mean, the median canopy height h_median, and the minimum canopy height h_min from each segment of the extracted data for calculation.

[0078] Furthermore, the method for constructing the spaceborne LiDAR-biomass inversion sub-model in step e is as follows: based on the vegetation distribution area obtained in step c, extract the photon point data of the overlapping part with the quadrat, and use the vegetation height percentage data and its statistical data obtained in step d. L Using the biomass observations obtained in step a as the independent variable and the biomass observations as the dependent variable, regression models such as stepwise linear regression, Bayesian ridge regression, ordinary linear regression, elastic network regression, support vector machine regression, random forest regression, gradient boosting regression, and deep learning were used to model the relationship between vegetation height and biomass in spaceborne LiDAR. Using sample points as the true values, regression analysis was performed using the characteristic values ​​of vegetation height and biomass in spaceborne LiDAR to obtain the spaceborne LiDAR-biological quantum model AGB. L =f(Feature) LThe model was evaluated using K-fold cross-validation, based on the regression coefficients R0 and the coefficient of determination R0. 2 The root mean square error (RMSE) was used to determine the accuracy of the regression model. Appropriate inversion parameters and the model with the highest overall accuracy were selected, with the model exhibiting the highest overall accuracy being chosen as the candidate for the spaceborne LiDAR-biomass inversion model. These were defined as AGB. L1 =f(Feature) L ), AGB L2 =f(Feature) L ...and thus obtain vegetation biomass strip data corresponding to the vegetation height strips obtained in step d;

[0079] Step f, large-scale urban surface vegetation aboveground biomass inversion, includes integrating the biomass inversion sub-models obtained in steps d and e with the support of the fine classification information of urban surface vegetation obtained in step c, and training the integrated inversion model that integrates the horizontal information and three-dimensional structural information of urban surface vegetation using the urban vegetation biomass quadrat data obtained in step a, and performing large-scale accurate mapping of urban vegetation aboveground biomass.

[0080] Furthermore, the ensemble model selected in this invention is based on a regression decision tree. This ensemble model has two features: candidate results from the spaceborne LiDAR-biomass inversion model and the optical-biomass inversion result. If the number of samples in the test sets of the spaceborne LiDAR-biomass inversion model and the optical-biomass inversion model is n, and the vegetation categories are m, then the feature space dimension of the ensemble model is n*m. Using the actual biomass measurements as the true value, the outputs of the four base models (candidates from the spaceborne LiDAR-biomass inversion model and the optical-biomass inversion model) are used as inputs to train the sub-models of the decision tree; finally, the output of the decision tree is used as the true inversion result to train the decision tree ensemble model. By using Gaofen-2 and ICESat-2 data of the study area, urban vegetation aboveground biomass inversion can be achieved.

[0081] The method for constructing the spaceborne LiDAR-biomass inversion sub-model is as follows: using the AGB obtained in steps d and e... S1 AGB S2 AGB L1 =f(Feature) L ) and AGB L2 =f(Feature) LUsing models with good performance as independent variables and biomass as the dependent variable, and with the vegetation category obtained in step c as the constraint, the relationship between several models and biomass is simulated using a decision tree. Sample points are used as the true values, and the simulated biomass of several models is compared with the actual biomass to solve the problem. The regression coefficient Ri and the coefficient of determination Ri are then used to determine the model. 2 The root mean square error (RMSE) is used to determine the model accuracy, resulting in the spaceborne LiDAR-optical integrated model.

[0082]

[0083] Where AGB represents the final biomass result obtained from the ensemble model; a m b m The weights of each parameter in the ensemble model are determined by the vegetation category; Among several candidate models for vegetation type determination, including the optical-biomass model and the spaceborne LiDAR-biomass model, the LU model is the most accurate for inverting this vegetation type. V The vegetation type is a limiting condition for each formula.

[0084] Furthermore, during the training of the integrated model, the inversion values ​​of the spaceborne LiDAR-biomass model and the optical-biomass model need to be normalized, as shown in Equation 23:

[0085]

[0086] In the formula, x i For the i-th biomass scenario, the inversion values ​​of the spaceborne LiDAR-biomass model and the optical-biomass model are given; y i This is the corresponding normalized result; x min and x max These represent the minimum and maximum values ​​of the inversion results of the spaceborne LiDAR-biomass model and the optical-biomass model under all scenarios.

[0087] In specific implementation, the method proposed in the technical solution of this invention can be automatically executed by those skilled in the art using computer software technology. System devices for implementing the method, such as computer-readable storage media storing the corresponding computer program of the technical solution of this invention and computer equipment including the computer program running the corresponding computer program, should also be within the protection scope of this invention.

[0088] In some possible embodiments, a precise inversion system for urban vegetation aboveground biomass that takes into account vegetation type is provided, including the following modules:

[0089] The first module is used for biomass calculation of sample plots within the study area, including calculating the corresponding biomass after sampling based on the main urban vegetation types within the study area.

[0090] The second module is used for high-resolution optical data and spaceborne photon counting lidar data preprocessing, including geometric and radiometric correction of high-resolution optical images, and cleaning and data extraction of spaceborne photon counting lidar data.

[0091] The third module is used to extract detailed information on urban vegetation in the study area, including the extraction of features of different types of urban vegetation and the fine classification of urban vegetation types, providing basic data for accurate estimation of aboveground biomass of urban vegetation.

[0092] The fourth module is used for spectral feature extraction of high-resolution remote sensing image data and construction of optical-biomass inversion sub-models. It includes the sample plot biomass data calculated by the first module and the fine classification information of urban surface vegetation obtained by the third module. Based on the obtained spectral features, the regression model with better estimation effect is selected to construct the optical-biomass quantum model, providing a sub-model for integrated inversion.

[0093] The fifth module is used for vegetation height data extraction from spaceborne photon counting lidar data and construction of a spaceborne LiDAR-biomass inversion sub-model. This includes filtering and denoising urban adaptive photon point cloud data, determining the window size for vegetation height data statistics based on the high-resolution image resolution, extracting vegetation height features, and using the sample plot biomass data calculated by the first module and the fine classification information of urban surface vegetation obtained by the third module. Based on the obtained vegetation height features, a regression model with better estimation performance is selected to construct a spaceborne LiDAR-bioquantum model, providing a sub-model for integrated inversion.

[0094] The sixth module is used for large-scale urban surface vegetation aboveground biomass inversion. It includes integrating the biomass inversion sub-models obtained from the fourth and fifth modules with the support of the fine classification information of urban surface vegetation obtained in the third module through the ensemble model constructed by decision tree, and training the ensemble inversion model that integrates the horizontal information and three-dimensional structural information of urban surface vegetation through the urban vegetation biomass quadrat data obtained in the first module, so as to carry out large-scale accurate mapping of urban vegetation aboveground biomass.

[0095] In some possible embodiments, a system for accurately inverting urban vegetation aboveground biomass that takes into account vegetation type is provided, including a processor and a memory. The memory is used to store program instructions, and the processor is used to call the stored instructions in the memory to execute the method for accurately inverting urban vegetation aboveground biomass that takes into account vegetation type as described above.

[0096] In some possible embodiments, a system for accurately inverting urban vegetation aboveground biomass that takes into account vegetation type is provided, including a readable storage medium on which a computer program is stored. When the computer program is executed, it implements the method for accurately inverting urban vegetation aboveground biomass that takes into account vegetation type as described above.

[0097] In summary, the present invention has the following characteristics:

[0098] Compared to existing methods, estimating aboveground biomass in urban vegetation presents a unique challenge. Unlike forests and grasslands where the first law of geography can be applied, urban vegetation distribution is constrained by planning requirements. This invention first obtains the distribution of various urban vegetation types through vegetation classification and fine-grained vegetation category classification. Then, it introduces ICESat-2 data to provide a wide range of three-dimensional vegetation structure parameters. Combining this with spectral parameters from high-resolution optical data better describes the vertical and horizontal structure of urban vegetation. Through an integrated model, the two are jointly inverted based on vegetation type conditions, giving the model mathematical and physical meaning. Simultaneously, the fusion of optical and spaceborne LiDAR photon point information provides richer information for biomass calculation, improving the accuracy of urban vegetation biomass inversion under complex underlying surface conditions. After obtaining biomass results from spaceborne LiDAR and optical features respectively, this invention uses a decision tree-based Stacking integrated model to dynamically fuse the biomass outputs from the two models according to the vegetation type of the studied sample points to obtain the final inversion result. This invention can fully leverage the advantages of spaceborne LiDAR data and high-resolution imagery in acquiring vegetation height over a wide area and obtaining full-coverage surface optical parameters. The combination of vegetation height information and spectral characteristics can effectively leverage the advantages of both, fill the gap in urban vegetation aboveground biomass estimation methods, and improve the accuracy of urban vegetation aboveground biomass model inversion.

[0099] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A method for accurate inversion of aboveground biomass of urban vegetation that takes into account vegetation type, characterized in that, Includes the following steps: Step a, Biomass calculation of sample plots in the study area, including calculating the corresponding biomass after sampling of sample plots based on the main urban vegetation types in the study area; Step b, high-resolution optical data and spaceborne photon counting lidar data preprocessing, including geometric and radiometric correction of high-resolution optical images, and cleaning and data extraction of spaceborne photon counting lidar data. Step c, refined extraction of urban vegetation information, including obtaining the dominant vegetation and its characteristics in the city through field sampling and various statistical data, defining the urban surface land use types separately, extracting the characteristics of different types of urban vegetation, and performing refined classification of urban vegetation types to provide basic data for accurate estimation of urban vegetation aboveground biomass. Step d involves extracting spectral features from high-resolution remote sensing image data and constructing an optical-biomass inversion sub-model. This includes using spectral features as independent variables and biomass as dependent variables, constructing the relationship between features and biomass through sample points, and using the sample plot biomass data calculated in step a and the fine classification information of urban surface vegetation obtained in step c. Based on the obtained spectral features, a regression model with better estimation performance is selected to construct an optical-biomass quantum model, providing a sub-model for integrated inversion. Step e, vegetation height data extraction from spaceborne photon counting lidar data and construction of spaceborne LiDAR-biomass inversion sub-model, includes: after filtering and denoising urban adaptive photon point cloud data, determining the window size for vegetation height data statistics based on the high-resolution image resolution, and extracting vegetation height features; using the sample plot biomass data calculated in step a and the urban surface vegetation fine classification information obtained in step c, selecting a regression model with better estimation performance based on the obtained vegetation height features to construct a spaceborne LiDAR-bioquantum model, providing a sub-model for integrated inversion; Step f, large-scale urban surface vegetation aboveground biomass inversion, includes integrating the biomass inversion sub-models obtained in steps d and e with the support of the fine classification information of urban surface vegetation obtained in step c, and training the integrated inversion model that integrates the horizontal information and three-dimensional structural information of urban surface vegetation using the urban vegetation biomass quadrat data obtained in step a, and performing large-scale accurate mapping of urban vegetation aboveground biomass.

2. The method for accurate inversion of urban vegetation aboveground biomass considering vegetation type according to claim 1, characterized in that: In step c, the urban surface categories are first classified, and the urban surface category is defined as impermeable surface. water bodies bare land ,vegetation ..., superpixel segmentation is performed on the high-resolution multispectral data of the study area obtained in step b. For the defined urban surface category objects, spectral features, texture features, and other features are extracted. The texture features include the object's band mean, brightness, and standard deviation. The other features include the object's length / width ratio, shape index, and density. This maximizes the inter-class differences and minimizes the intra-class differences between different land use objects. Land use type is then determined by combining different influencing factors, expressed as follows: in, For spectral feature discrimination model, For texture feature discrimination model, Other feature discrimination models, where a, b, and c are the weights of the three discrimination models respectively; And according to The scope was further divided, and urban vegetation types within the study area were refined based on dominant tree species. Spectral, textural, and other information of different typical vegetation types were extracted. The vegetation type expression in the refined urban vegetation type extraction process is as follows: 。 3. A method for accurate inversion of urban vegetation aboveground biomass considering vegetation type, as described in claim 1 or 2, characterized in that: The optical-biomass inversion sub-model constructed in step d is implemented using spectral features. As an independent variable, biomass Using spectral characteristics as the dependent variable, the relationship between spectral features and biomass is simulated, and with sample points as the true values, the solution is obtained by solving for the spectral feature values ​​and biomass, thus deriving the optical-biological quantum model. Meanwhile, in the process of model selection, different regression analysis methods are compared and a better model is selected.

4. The method for accurate inversion of urban vegetation aboveground biomass considering vegetation type according to claim 3, characterized in that: The implementation of the spaceborne LiDAR-biomass inversion sub-model constructed in step e is as follows: using spaceborne LiDAR vegetation height characteristics... Using vegetation height as the independent variable and biomass as the dependent variable, the relationship between vegetation height and biomass in spaceborne LiDAR was simulated. Using sample points as the true values, the model was solved using the eigenvalues ​​of vegetation height and biomass in spaceborne LiDAR, resulting in a spaceborne LiDAR-biological quantum model. Meanwhile, in the process of model selection, different regression analysis methods are compared and a better model is selected.

5. The method for accurate inversion of urban vegetation aboveground biomass considering vegetation type according to claim 4, characterized in that: The implementation of the spaceborne LiDAR-biomass inversion sub-model constructed in step f is as follows: the better performing model obtained in steps d and e is used as the independent variable, biomass is used as the dependent variable, and the vegetation category obtained in step c is used as the constraint condition. The relationship between several models and biomass is simulated through decision tree, and the sample points are used as the true values. The simulated biomass of several models and the real biomass are solved to obtain the spaceborne LiDAR-optical integrated model.

6. A precise inversion system for urban vegetation aboveground biomass that takes into account vegetation type, characterized in that: This method is used to implement an accurate inversion method for urban vegetation aboveground biomass that takes into account vegetation type, as described in any one of claims 1-5.

7. The accurate inversion system for urban vegetation aboveground biomass considering vegetation type as described in claim 6, characterized in that: Includes the following modules, The first module is used for biomass calculation of sample plots within the study area, including the calculation of corresponding biomass after sampling of sample plots based on the main urban vegetation types within the study area. The second module is used for high-resolution optical data and spaceborne photon counting lidar data preprocessing, including geometric and radiometric correction processing of high-resolution optical images, and cleaning and data extraction of spaceborne photon counting lidar data. The third module is used to extract detailed information on urban vegetation in the study area, including the extraction of features of different types of urban vegetation and the function of fine classification of urban vegetation types, providing basic data for accurate estimation of aboveground biomass of urban vegetation. The fourth module is used for spectral feature extraction of high-resolution remote sensing image data and construction of optical-biomass inversion sub-models. It includes the sample plot biomass data calculated by the first module and the fine classification information of urban surface vegetation obtained by the third module. Based on the obtained spectral features, the regression model with better estimation effect is selected to construct the optical-biomass quantum model, providing a sub-model for integrated inversion. The fifth module is used for vegetation height data extraction from spaceborne photon counting lidar data and construction of a spaceborne LiDAR-biomass inversion sub-model. This includes filtering and denoising urban adaptive photon point cloud data, determining the window size for vegetation height data statistics based on the high-resolution image resolution, extracting vegetation height features, and using the sample plot biomass data calculated by the first module and the fine classification information of urban surface vegetation obtained by the third module. Based on the obtained vegetation height features, a regression model with better estimation performance is selected to construct a spaceborne LiDAR-bioquantum model, providing a sub-model for integrated inversion. The sixth module is used for large-scale urban surface vegetation aboveground biomass inversion. It includes integrating the biomass inversion sub-models obtained from the fourth and fifth modules with the support of the fine classification information of urban surface vegetation obtained in the third module through the ensemble model constructed by decision tree, and training the ensemble inversion model that integrates the horizontal information and three-dimensional structural information of urban surface vegetation through the urban vegetation biomass quadrat data obtained in the first module, so as to carry out large-scale accurate mapping of urban vegetation aboveground biomass.

8. The accurate inversion system for urban vegetation aboveground biomass considering vegetation type according to claim 6, characterized in that: It includes a processor and a memory, the memory being used to store program instructions, and the processor being used to call the stored instructions in the memory to execute the method for accurate inversion of urban vegetation aboveground biomass taking into account vegetation type as described in any one of claims 1-5.

9. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which, when executed, implements a method for accurate inversion of urban vegetation aboveground biomass that takes into account vegetation type, as described in any one of claims 1-5.

Images