Adaptive spatial fusion weight carbon sink estimation method, system, device and medium

Abstract

This invention discloses an adaptive spatial fusion weighted carbon sink estimation method, system, device, and medium. The method includes: acquiring a multi-source dataset of the study area and preprocessing it to obtain environmental factor data; dividing the study area into hierarchical regions using a comprehensive complexity index in an adaptive spatial grid map and adaptively setting the data point density for each hierarchical region; constructing a carbon sink estimation model and a carbon sink sample dataset for the study area. The carbon sink estimation model contains a spatial regression model, which is trained using the carbon sink sample dataset for carbon sink prediction, and the spatial regression model is trained using spatial regression of the response coefficients corresponding to environmental factors; acquiring net primary productivity data of the study area, and the carbon sink estimation model predicts the carbon sink estimation value for pixel m. This invention improves the accuracy and consistency of regional-scale carbon sink estimation results, providing reliable technical support for carbon sink assessment, ecological restoration effect quantification, and carbon management decision-making in mining areas and similar strongly disturbed areas.

Description

Technical Field

[0001] This invention relates to the fields of geographic information systems and ecological environment technology, and in particular to an adaptive spatial fusion weighted carbon sink estimation method, system, device and medium based on disturbance intensity driven, which is applicable to the spatial simulation, evaluation and optimization of carbon sinks in highly disturbed areas such as mining areas. Background Technology

[0002] With the intensification of global climate change, regional carbon sink assessment has become an important foundation for ecological environment monitoring and carbon cycle research. In particular, the carbon sink changes in areas with strong human disturbance, such as mining areas, are of great significance for assessing the effectiveness of ecological restoration and regional carbon neutrality and management. However, the significant differences in surface disturbance intensity and highly heterogeneous spatial structure in mining areas pose a great challenge to the stable and accurate estimation of regional carbon sinks. Existing regional carbon sink estimation methods mainly include model methods based on ecological process mechanisms and empirical model methods based on statistical regression or machine learning. Although spatial statistical models such as Geographically Weighted Regression (GWR) have been widely used for regional carbon sink spatial modeling by introducing spatial weight mechanisms to a certain extent to consider spatial non-stationarity, existing methods still have the following technical defects in practical applications in highly disturbed areas such as mining areas: (1) Difficulty in adapting to spatial structures with significant disturbance differences; Existing carbon sink estimation methods usually use fixed spatial resolution or single-scale parameters for modeling, and the spatial weight and bandwidth remain unchanged throughout the study area. It is difficult to dynamically adjust the modeling scale according to the differences in disturbance intensity in different areas, resulting in insufficient local fitting ability in strongly disturbed areas and over-smoothing in relatively stable areas. (2) Spatial weight construction does not fully consider the differences in disturbance states; existing spatial regression methods mostly use geographical distance as the basis for weight construction, ignoring the impact of differences in surface disturbance intensity on sample similarity and spatial correlation structure; in areas with strong disturbances such as mining areas, the differences in ecological processes between samples under different disturbance states are significant, and relying solely on distance to propagate sample information can easily lead to carbon sink estimation bias. (3) Lack of adaptive adjustment mechanism; in existing geographically weighted regression (GWR) and related spatial modeling methods, spatial bandwidth and weight parameters are usually determined through global optimization or manual experience, making it difficult to make local adaptive adjustments as spatial disturbance conditions change, thus limiting the model's stability and generalization ability.

[0003] In summary, existing regional carbon sink estimation techniques generally suffer from insufficient perception of disturbance differences and a lack of adaptive adjustment of modeling scale and spatial weights in highly disturbed and strongly non-stationary regions such as mining areas. In complex disturbance environments, they also suffer from insufficient ability to characterize spatial heterogeneity, large estimation errors in highly disturbed regions, and fixed model parameters that are difficult to adapt to dynamic spatial structure changes. It is difficult to simultaneously achieve a detailed representation of strongly disturbed regions and the stability of the overall estimation results. Summary of the Invention

[0004] The purpose of this invention is to provide an adaptive spatial fusion weighting carbon sink estimation method, system, device, and medium based on disturbance intensity-driven carbon sink estimation. The carbon sink estimation model introduces an adaptive spatial modeling mechanism driven by disturbance intensity, dynamically adjusting the spatial weights and response coefficients of the modeling scale according to the degree of surface disturbance and ecological environment differences. The carbon sink estimation model has a spatial regression model, and the core of the spatial regression model includes a dual-core fusion weighting function. The dual-core fusion weighting function integrates a dual-core spatial weighting system that combines geographical proximity and disturbance similarity, thereby improving the model's ability to express spatial non-stationary characteristics and its overall estimation stability.

[0005] The objective of this invention is achieved through the following technical solution: An adaptive spatial fusion weighted carbon sink estimation method, the method comprising: S1. Construct an adaptive spatial grid map of the study area, obtain multi-source datasets including satellite multispectral images, meteorological data, land use data, and elevation data of the study area, and preprocess them to obtain environmental factor data; S2. Using the comprehensive complexity index obtained from multi-source datasets, the comprehensive complexity index is used to divide the hierarchical regions in the adaptive spatial grid map and to adaptively set the data point density of each hierarchical region. S3. Construct a carbon sink estimation model and a carbon sink sample dataset for the study area. The carbon sink sample dataset contains environmental factor sample data, net primary productivity sample data and carbon sink labels corresponding to the data points. The carbon sink estimation model has a spatial regression model. The carbon sink estimation model uses the carbon sink sample dataset for carbon sink prediction training, and the spatial regression model performs spatial regression training on the response coefficients corresponding to the environmental factors. S4. Obtain net primary productivity data for the study area. The carbon sink estimation model uses the following expression to obtain the carbon sink estimate for cell m. : ,in For the net primary productivity of m pixels, Let m be the response coefficient of the environmental factor k. The standardized data corresponding to pixel m and environmental factor k This represents the total number of environmental factors.

[0006] To better achieve the present invention, the present invention also includes the following methods: S5. Following method S4, calculate the estimated carbon sink values ​​of all pixels in the study area and summarize them to obtain the total estimated carbon sink value of the study area.

[0007] Preferably, in method S3, the spatial regression model obtains the response coefficients corresponding to environmental factors using the following method: S31. The spatial regression model constructs the following dual-core fusion weight function using each data point in the study area method S2: ,in The fusion weights between data points i and j For geographic proximity kernel function, For perturbation similarity kernel function; S32. In the carbon sink estimation model carbon sink prediction training, the spatial regression model is trained with model constraints using a dual-core fusion weight function, and the carbon sink estimation model outputs the response coefficients corresponding to the environmental factors for each data point. S33. The spatial regression model is extended by local regression interpolation to obtain the response coefficients corresponding to environmental factor k for all pixels.

[0008] Preferably, in method S31, the method for constructing the geographic proximity kernel function between data points i and j is as follows: S311. Construct a disturbance intensity function that includes topographic heterogeneity, vegetation heterogeneity, and surface heterogeneity. The expression for the disturbance intensity function is as follows: The perturbation intensity of data point i, For terrain heterogeneity, For vegetation heterogeneity, For surface heterogeneity, , , These are the weighting coefficients; S312. Normalize the disturbance intensity to generate a normalized disturbance index. ; S313. Preset global reference bandwidth based on all data points across the entire study area. Construct the adaptive local bandwidth for data point i according to the following expression. : ; S314, the distance between data point i and data point j Construct the following geographic proximity kernel function: .

[0009] Preferably, the perturbation similarity kernel function expression between data points i and j is as follows: , Let be the normalized perturbation exponent for data point i. Let j be the normalized perturbation exponent for data point j. The perturbation similarity sensitivity coefficient; perturbation similarity sensitivity coefficient The method for obtaining this information is as follows: the difference in perturbation between data points i and j. Construct a carbon sink semivariogram conditioned on perturbation differences. ; Semivariogram of carbon sink The initial value of γ is obtained by using an exponential semivariogram model. Then, the objective function containing the root mean square error and the Akaike information criterion is constructed by combining the carbon sink estimation model and performing joint iterative optimization to obtain the optimal γ. The optimal γ is used as the perturbation similarity sensitivity coefficient.

[0010] Preferably, the adaptive spatial grid map is divided into: The system comprises several hierarchical regions, each with a different grid resolution and a different data point density; the adaptive local bandwidth... Set the following bandwidth constraints: minimum bandwidth (Average distance between adjacent data points); Maximum bandwidth (Diagonal length of the study area); the bandwidth should not be less than the grid scale of the area to which the bandwidth belongs.

[0011] Preferably, in method S1, the satellite multispectral image source includes Sentinel-2 satellite multispectral image data with cloud cover below 10%, and the satellite multispectral image source undergoes preprocessing including radiometric correction and FLAASH atmospheric correction; the meteorological data includes monthly average temperature, monthly cumulative precipitation, and monthly average solar radiation; the elevation data is digital elevation model data; the adaptive spatial grid map internally constructs a WGS84 coordinate system, and the multi-source dataset is based on the WGS84 coordinate system for data mapping and spatial registration and alignment processing; the environmental factor items of the environmental factor data include vegetation index NDVI, vegetation index gradient, vegetation cover, bare land ratio, slope factor, and meteorological factors, and the meteorological factors include monthly average temperature, monthly cumulative precipitation, and monthly average solar radiation.

[0012] Preferably, in method S2, the comprehensive complexity index R is constructed as follows: , This is the normalized data corresponding to the slope factor in the environmental factor data; This is the normalized data corresponding to the vegetation index gradient in the environmental factor data; In environmental factor data The corresponding normalized data, i.e., the normalized data of the percentage of bare land surface area; , , These are the weighting coefficients.

[0013] Preferably, in the adaptive spatial grid map, the study area is divided into three levels of low, medium, and high levels using the comprehensive complexity index. Different-resolution grids are set for each level area, and each grid has a different data point density. The range of the comprehensive complexity index R corresponding to the low-complexity level area in the study area is R ≤ 0.33, and the preset grid data is N1. The range of the comprehensive complexity index R corresponding to the medium-complexity level area in the study area is 0.33 < R ≤ 0.66, and the preset grid data is N2. The range of the comprehensive complexity index R corresponding to the high-complexity level area in the study area is R > 0.66, and the preset grid data is N3, where N1 < N2 < N3. The grid scale of the study area decreases sequentially from low to high according to the level area, and the data point density of the grid is set from low to high sequentially according to the level area from low to high.

[0014] Preferably, in method S4, a regional ecosystem is constructed based on the study area, and the minimum carbon sink value allowed for the regional ecosystem of the study area is set and the maximum carbon sink value , and the following ecological constraint conditions are constructed for the carbon sink estimation value of pixel m : , and the ecological constraint adjustment of the carbon sink estimation value is performed for each pixel.

[0015] Preferably, the net primary productivity sample data and the net primary productivity of pixel m are both obtained through the CASA light use efficiency model using multi-source datasets to obtain the net primary productivity at time t according to the following expression : , where is the photosynthetically active radiation absorbed by vegetation, is the light use efficiency.

[0016] An adaptive spatial weight carbon sink estimation system includes a carbon sink estimation model, an adaptive spatial calculation planning module, an adaptive spatial grid map, a carbon sink sample dataset, and a data acquisition and calculation module. The carbon sink sample dataset internally stores environmental factor sample data, net primary productivity sample data, and carbon sink labels corresponding to data points. The data acquisition and calculation module is used to obtain multi-source datasets including satellite multispectral images, meteorological data, land use data, and elevation data in the study area and preprocess them to obtain environmental factor data. The adaptive spatial calculation planning module uses the comprehensive complexity index obtained from the multi-source datasets to perform level area division and adaptively set the data point density of each level area in the adaptive spatial grid map. The carbon sink estimation model has a spatial regression model internally. The carbon sink estimation model uses the carbon sink sample dataset for carbon sink prediction training, and the spatial regression model performs spatial regression training of the corresponding response coefficients of environmental factors. The carbon sink estimation model obtains the carbon sink estimation value of pixel m according to the following expression based on the net primary productivity data of the study area obtained : ,in For the net primary productivity of m pixels, Let m be the response coefficient of the environmental factor k. The standardized data corresponding to pixel m and environmental factor k This represents the total number of environmental factors.

[0017] An electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform the steps of the adaptive spatial fusion weighted carbon sink estimation method of the present invention.

[0018] A storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the adaptive spatial fusion weighted carbon sink estimation method of the present invention.

[0019] Compared with the prior art, the present invention has the following advantages and beneficial effects: (1) The carbon sink estimation model of this invention introduces an adaptive spatial modeling mechanism driven by disturbance intensity. It dynamically adjusts the spatial weights and response coefficients of the modeling scale according to the degree of surface disturbance and the differences in ecological environment. The carbon sink estimation model has a spatial regression model. The core of the spatial regression model includes a dual-core fusion weight function. The dual-core fusion weight function integrates the dual-core spatial weight system of geographical proximity and disturbance similarity, which improves the model's ability to express spatial non-stationary characteristics and the overall estimation stability.

[0020] (2) The present invention can effectively reduce the uncertainty of regional carbon sink estimation under complex disturbance conditions, improve the accuracy and consistency of regional carbon sink estimation results, and provide reliable technical support for carbon sink assessment, ecological restoration effect quantification and carbon management decision-making in mining areas and similar strongly disturbed areas.

[0021] (3) This invention improves the ability to express spatial heterogeneity under complex disturbance environments. By introducing the surface disturbance intensity index and adaptively adjusting the spatial modeling scale and regression weight accordingly, the modeling parameters can be adjusted according to the disturbance differences in different regions. This effectively improves the problem of insufficient expression of traditional fixed parameter models in spatially heterogeneous regions and enhances the adaptability of carbon sink estimation results to complex surface structures.

[0022] (4) This invention reduces the carbon sink estimation bias in highly disturbed areas and enhances the stability of carbon sink estimation results. By considering both geographical proximity and disturbance similarity constraints during the spatial weight construction process, it reduces the unreasonable information propagation between samples with different disturbance states, which helps to reduce the estimation bias in areas with strong disturbances and improve the stability of spatial modeling results under complex disturbance conditions.

[0023] (5) The present invention enhances the adaptive and generalization capabilities of the model parameters; through the perturbation-driven bandwidth adjustment mechanism and the parameter correction process based on residual characteristics, the model parameters can be adjusted according to changes in spatial structure and error distribution characteristics, overcoming the problems of fixed modeling parameters and insufficient adaptability in existing methods, and improving the applicability of carbon sink estimation methods in different regions and under different perturbation conditions. Attached Figure Description

[0024] Figure 1 This is a flowchart of the adaptive spatial fusion weighted carbon sink estimation method of the present invention; Figure 2 This is a schematic diagram illustrating the principle of adaptive bandwidth adjustment based on disturbance intensity in the embodiment. Figure 3 This is a schematic diagram illustrating the principle of the dual-core spatial weighting function in the embodiment; Figure 4 This is a schematic diagram illustrating the spatial distribution of carbon sink estimates in the example study area. Detailed Implementation

[0025] The present invention will be further described in detail below with reference to embodiments: Example like Figure 1 As shown, an adaptive spatial fusion weighted carbon sink estimation method includes the following steps: S1. Construct an adaptive spatial grid map of the study area (this invention takes a mining area as an example, preferably an open-pit mining area), obtain a multi-source dataset of the study area including satellite multispectral images, meteorological data, land use data, and elevation data, and preprocess it to obtain environmental factor data.

[0026] In some embodiments, the satellite multispectral image source includes Sentinel-2 satellite multispectral image data with cloud cover below 10%. The satellite multispectral image data includes blue, green, red, and near-infrared bands, with a spatial resolution of 10m to meet the requirements for vegetation index and vegetation cover inversion. The satellite multispectral image source (containing 12 images of the study area) undergoes preprocessing including radiometric correction and FLAASH atmospheric correction. Meteorological data includes monthly average temperature, monthly cumulative precipitation, and monthly average solar radiation. In this embodiment, these are obtained by combining ground meteorological station observation data with spatial interpolation results, and a continuous meteorological raster is generated through inverse distance weighted interpolation. Elevation data is digital elevation model data, preferably using airborne lidar mapping or high-resolution DEM data to ensure the ability to characterize slopes and micro-topographic disturbances in the mining area. The adaptive spatial grid map internally utilizes the WGS84 coordinate system. Multi-source datasets undergo data mapping and spatial registration / alignment based on this system. All data from the source datasets are mapped using the WGS84 coordinate system. In this embodiment, remote sensing imagery is used as a reference, and meteorological, land use, and DEM data are resampled to the resolution of the remote sensing imagery. The resampling method uniformly employs bilinear interpolation (nearest neighbor is used for categorical data). After inspection, the pixel offset error between different data sources is controlled within 0.5 pixels. Environmental factor items in the environmental factor data include the vegetation index NDVI, vegetation index gradient, vegetation cover, bare land ratio, slope factor, and meteorological factors. Meteorological factors include monthly average temperature, monthly cumulative precipitation, and monthly average solar radiation. The vegetation index NDVI, vegetation index gradient, and vegetation cover FVC are calculated using satellite multispectral imagery from the multi-source dataset, while the slope factor is calculated using elevation data from the multi-source dataset. Land use types in the land use data include: forest land, grassland, cultivated land, bare land, mining-disturbed areas, and construction land. In the implementation case using a specific open-pit coal mine as the study area, bare land and mining-disturbed areas account for approximately 27.4% of the study area. Mining-disturbed areas and bare land are considered high-disturbance areas and used for subsequent normalized disturbance index construction. Data preprocessing of the multi-source dataset includes image radiometric correction, atmospheric correction, cropping, raster alignment, outlier removal, and missing value imputation. Outlier removal includes removing outlier pixels with NDVI < -0.1 or NDVI > 0.95, and removing extreme pixels with a slope greater than 65°. This results in the construction of a multi-source dataset that can be used for subsequent model processing. This embodiment uses a specific open-pit coal mine as the study area, with a mining area of ​​approximately 138 km². The multi-source dataset includes remote sensing image data, meteorological observation data, land use / cover data, and digital elevation data.

[0027] S2. Using the comprehensive complexity index obtained from the multi-source dataset, the adaptive spatial grid map is used to divide the region into levels and adaptively set the data point density for each level. The adaptive spatial grid map of this invention adaptively divides the region into levels according to the study area and adaptively sets the data point density according to the level (sample data points in the carbon sink sample dataset and subsequent carbon sink estimation model training for carbon sink prediction are all subject to adaptive spatial processing driven by perturbation intensity based on data point density). In some embodiments, the adaptive spatial grid map of this invention is divided into... Each region is divided into several levels, with a different grid resolution and a different data point density within each grid. Adaptive local bandwidth is also implemented. Set the following bandwidth constraints: minimum bandwidth (Average spacing between adjacent data points). Maximum bandwidth. (Diagonal length of the study area). The bandwidth should not be less than the grid scale of the area to which the bandwidth belongs.

[0028] In some embodiments, the comprehensive complexity exponent R is constructed as follows: , This refers to the normalized data corresponding to the slope factor in the environmental factor data, namely the normalized data of the elevation change rate (slope calculated based on the digital elevation raster). This is the normalized data corresponding to the vegetation index gradient in the environmental factor data, that is, the normalized data of the vegetation index change gradient of adjacent pixels or data points. Specifically, the NDVI gradient (i.e. the vegetation index gradient) is calculated by the neighborhood sliding window method (5×5 sliding window) to calculate the NDVI difference of adjacent pixels and then normalized. In environmental factor data The corresponding normalized data is the normalized data of the percentage of bare land surface area. , , The weighting coefficients are a+b+c=1. In the example of an open-pit coal mine as the study area, the weighting coefficients are set as follows: a=0.3, b=0.4, c=0.3. This embodiment can specifically adopt the following method: the bare land ratio is calculated by using each 30m×30m grid cell as the unit for determining the percentage of bare land surface area. This embodiment uses three indicators—slope, vegetation index gradient, and bare land ratio—to comprehensively reflect the surface spatial structure characteristics of the mining area, representing the topographic complexity, vegetation spatial variation intensity, and mining disturbance degree, respectively, thus comprehensively reflecting the surface spatial structure characteristics of the mining area.

[0029] In the case of taking a certain open-pit coal mine area as the research area in this embodiment, in the adaptive spatial grid map, the research area is divided into three levels of regions with low, medium, and high levels by using the comprehensive complexity index. Different-resolution grids are set in each level of region, and each grid has a different data point density. The range of the comprehensive complexity index R corresponding to the low-complexity level region in the research area is R ≤ 0.33, and the preset grid data is N1 (example: the number of grids is 1842). The range of the comprehensive complexity index R corresponding to the medium-complexity level region in the research area is 0.33 < R ≤ 0.66, and the preset grid data is N2 (example: the number of grids is 3965). The range of the comprehensive complexity index R corresponding to the high-complexity level region in the research area is R > 0.66, and the preset grid data is N3 (example: the number of grids is 6412), and N1 < N2 < N3. The grid scale of the research area decreases in turn from low to high according to the level regions. For example, the research area is divided into three levels of regions with low, medium, and high levels, that is, corresponding to the low-complexity level region, the medium-complexity level region, and the high-complexity level region, with low complexity (R ≤ 0.33, the number of grids 1842), medium complexity (0.33 < R ≤ 0.66, the number of grids 3965), and high complexity (R > 0.66, the number of grids 6412). Different-resolution spatial grid units are generated for the three regions respectively to adapt to local differences: low-complexity region (coarse grid, side length about 50m); medium-complexity region (medium grid, side length about 30m); high-complexity region (fine grid, side length about 10m). The high-complexity region is mainly distributed in the pit edge, waste dump, and exposed slope area; the data point density of the grids is set from low to high in turn according to the level regions from low to high. For example, in the high-complexity region (fine grid 10m): sample points are densely arranged, and 2 - 3 sample points are arranged in the grid; in the medium-complexity region (medium grid 30m): sample points are arranged with medium density, and 1 sample point is arranged in each grid; in the low-complexity region (coarse grid 50m): sample points are sparsely arranged, and 1 sample point is arranged in every 2 - 3 grids. Thus, the present invention constructs a three-layer adaptive spatial region of level regions, grids, and data points in the adaptive spatial grid map, improving the model's ability to express the spatial heterogeneity of the mining area.

[0030] S3. Construct a carbon sink estimation model and a carbon sink sample data set for the research area. The carbon sink sample data set internally stores the environmental factor sample data, net primary productivity sample data, and carbon sink labels corresponding to the data points. In some embodiments, the net primary productivity sample data is obtained by the CASA model of light use efficiency using multi-source data sets according to the following expression for the net primary productivity at time t : , where is the photosynthetic active radiation absorbed by vegetation, is the light use efficiency; , is the total solar radiation, Photosynthetically active radiation, The absorption ratio of the vegetation canopy (estimated by remote sensing vegetation index). , To maximize light energy utilization, , These are the temperature stress coefficients, This represents the water stress coefficient. The carbon sink estimation model incorporates a spatial regression model. The carbon sink estimation model is trained using a carbon sink sample dataset for carbon sink prediction, while the spatial regression model is trained using spatial regression of the response coefficients corresponding to environmental factors.

[0031] In some embodiments, the spatial regression model obtains the response coefficients corresponding to environmental factors as follows: S31, such as Figure 3 As shown, the spatial regression model constructs the following dual-core fusion weight function between data points i and j based on each data point in the study area method S2: ,in The fusion weights between data points i and j Let i be the kernel function for the geographical proximity between data points i and j. Let be the perturbation similarity kernel function between data points i and j. To avoid insufficient sample information due to excessive weight decay in regions of high perturbation difference, this invention preferably sets a lower limit constraint on the fusion weights. When the value is below a preset threshold, the minimum weight value is used for substitution or a neighborhood smoothing compensation mechanism is introduced to ensure the numerical stability and continuity of results in the local regression calculation process. Preferably, the method for constructing the geographic proximity kernel function between data points i and j is as follows: S311. Construct a disturbance intensity function that includes topographic heterogeneity, vegetation heterogeneity, and surface heterogeneity. The expression for the disturbance intensity function is as follows: The perturbation intensity of data point i, To represent the terrain heterogeneity, the standard deviation of the slope within a local window is calculated based on the slope value of the study space (grid and its neighboring grids) where the sample point is located, which characterizes the local undulation of the terrain. To determine vegetation heterogeneity, the coefficient of variation of the NDVI gradient magnitude within a local window is calculated based on the NDVI gradient value of the study space (grid and its neighboring grids) where the sample point is located, which characterizes the degree of drastic local changes in vegetation cover. To determine surface heterogeneity, based on the proportion of bare land in the study space (grid and its neighboring grids) where the sample point is located, the information entropy of the proportion of bare land within a local window is calculated to characterize the local mixing degree of surface cover types. , , These are the weighting coefficients. In the example case of an open-pit coal mine area as the study area, this embodiment sets them as follows: This reflects the main contribution of vegetation heterogeneity to the intensity of disturbance.

[0032] S312. Normalize the disturbance intensity to generate a normalized disturbance index. The expression is as follows: , The value ranges from 0 to 1, with larger values ​​indicating stronger local disturbances. In this embodiment... The value range is 0.03-0.91.

[0033] S313. Preset global reference bandwidth based on all data points across the entire study area. ,like Figure 2 As shown, the adaptive local bandwidth of data point i is constructed according to the following expression. : α represents the disturbance response sensitivity coefficient (used to adjust the magnitude of bandwidth variation with disturbance); when the normalized disturbance exponent... Large, adaptive local bandwidth Automatic reduction improves modeling accuracy in highly perturbation regions; when the normalized perturbation index... Hours, adaptive local bandwidth Maintaining a larger scale improves computational efficiency in stable regions. Bandwidth controls the spatial influence range of the local regression model; specifically, bandwidth determines the spatial distance range of surrounding samples participating in the weight calculation when calculating the regression parameters of a sample point i at a certain location. Sample points closer to sample point i have a larger weight; sample points farther away have a gradually decreasing weight; when the distance exceeds the bandwidth range, its influence approaches zero. Therefore, the bandwidth parameter actually determines the spatial neighborhood scale considered by the local regression model.

[0034] Preferably, adaptive local bandwidth Set the following bandwidth constraints: minimum bandwidth (Average spacing between adjacent data points). Maximum bandwidth. (Diagonal length of the study area). The bandwidth should not be less than the grid scale of the region to which the bandwidth belongs, such as a highly perturbable region (or a highly complex region) h. i ≥10m, medium disturbance region (or medium complexity region) h i ≥30m, low-disturbance zone (or low-complexity zone) h i ≥30m. When calculated When the constraint interval is exceeded, pruning is performed to ensure model stability; bandwidth is used to construct the spatial weight matrix; when the calculated h i When the above limits are exceeded, the system executes pruning rules to ensure that the bandwidth remains within the effective range.

[0035] S314, the distance between data point i and data point j Construct the following geographic proximity kernel function: ,in Let be the distance between data point i and data point j.

[0036] S32. In the carbon sink estimation model's carbon sink prediction training, the spatial regression model undergoes model constraint training with a dual-core fusion weight function, and the carbon sink estimation model outputs the response coefficients corresponding to the environmental factors for each data point.

[0037] S33. The spatial regression model is extended by local regression interpolation to obtain the response coefficients corresponding to environmental factor k for all pixels.

[0038] In some embodiments, the perturbation similarity kernel function expression between data points i and j is as follows: , Let be the normalized perturbation exponent for data point i. Let j be the normalized perturbation exponent for data point j. The difference between the normalized perturbation exponents of data point i and data point j is given by the kernel function when the perturbation states of the two samples are significantly different, which effectively suppresses the spread of unreasonable information between heterogeneous perturbation regions. This is the perturbation similarity sensitivity coefficient. The method for obtaining this information is as follows: the difference in perturbation between data points i and j. Construct a carbon sink semivariogram conditioned on perturbation differences. The semivariogram of carbon sinks An initial value of γ is obtained through an exponential semivariogram model. Then, an objective function incorporating root mean square error and the Akaike information criterion is constructed using a carbon sink estimation model, and the optimal γ is obtained through joint iterative optimization. This optimal γ is used as the perturbation similarity sensitivity coefficient. In the case study of an open-pit coal mine area, the determined perturbation similarity sensitivity coefficient is γ = 2.3. When the perturbation index of two data points differs by 0.3, the perturbation similarity weight decreases to approximately 50% (exp(-2.3×0.3)≈0.5); when the perturbation index of two data points differs by 0.6, the weight decreases to approximately 25%, effectively reducing the mutual influence between samples with high perturbation differences. The model parameters of the carbon sink estimation model of this invention include the perturbation similarity sensitivity coefficient. Global reference bandwidth The disturbance response sensitivity coefficient α, the response coefficient, and the weighting coefficient. , , To ensure optimal performance of the carbon sink estimation model, a two-stage grid search strategy is employed to collaboratively optimize the aforementioned parameters. This invention uses a phased search strategy to jointly optimize the core parameters. The optimization process includes two stages: coarse-scale parameter scanning and fine-tuning of fine-scale parameters. The modified Akaike Information Criterion (AICc) and cross-validation error are used as evaluation indicators. In the case study of an open-pit coal mine area, the preferred method is as follows: Phase 1: Coarse-grained global search Parameter range: , and Determined based on the average spacing of sample points in the study area; , Search step size: The step size is 10% of the interval length, the α step size is 0.5, and the γ step size is 0.3; the root mean square error (RMSE-CV) of five-fold cross-validation is used to determine the approximate region where the optimal parameters are located.

[0039] Phase Two: Fine-grained Local Search A fine-grained search is performed near the optimal parameter region obtained in the first stage to further improve parameter accuracy. Search step size settings: The step size is 2% of the interval length, with α and γ steps both being 0.1. The modified Akaike Information Criterion (AICc) is used as the evaluation metric. The optimal parameter combination with the smallest AICc is selected by traversing all candidate parameter combinations. , , ).

[0040] In this embodiment, the optimal parameter combination determined through the above optimization process is: , , Under this parameter combination, the model's AICc is reduced by 18.7% compared to the traditional GWR model with fixed bandwidth, and the cross-validation R² increases from 0.52 to 0.68, indicating that parameter optimization effectively improves the model's fitting accuracy and generalization ability.

[0041] After obtaining the optimal parameter combination, the spatial regression model is extended by local regression interpolation to obtain the response coefficients corresponding to environmental factor k for all pixels (i.e., the response coefficients of pixel m and environmental factor k). ),in These represent different environmental factors.

[0042] This embodiment can generate an environmental response coefficient grid. Unified processing with CASA-NPP (baseline NPP data calculated by the CASA model) data; specifically including: unifying the coefficient raster and NPP raster to the same coordinate reference system (WGS84); setting a unified spatial resolution (e.g., 30m); ensuring alignment of different data sources through raster resampling and spatial registration algorithms; controlling the pixel spatial offset between different data sources to within one pixel range through registration accuracy evaluation, with an actual average offset error of approximately 0.37 pixels; and processing the pixel-level environmental response parameters output by the spatial regression model. Local rationality screening is conducted, including outlier removal, parameter stability assessment, and ecological driving constraint screening, to exclude parameter combinations that are clearly inconsistent with ecological mechanisms. After screening, the number of pixels that can effectively participate in subsequent calculations accounts for more than 85% of the initial total number of pixels.

[0043] S4. Obtain net primary productivity data for the study area. In some embodiments, the net primary productivity of pixel m is obtained using a multi-source dataset through the light energy utilization model CASA at time t according to the following expression. : ,in Photosynthetically active radiation absorbed by vegetation The carbon sink estimation model uses the following expression to obtain the estimated carbon sink value for pixel m, representing the light energy utilization rate. : ,in For the net primary productivity of m pixels, Let m be the response coefficient of the environmental factor k. This refers to the standardized data corresponding to pixel m and environmental factor k (i.e., the standardized data of environmental factor data). This represents the total number of environmental factors.

[0044] In some embodiments, a regional ecosystem is constructed based on the study area, and a minimum allowable carbon sink value for the regional ecosystem of the study area is set. and maximum carbon sink value Carbon sequestration estimate for pixel m The following ecological constraints are established: Ecological constraints are adjusted for the carbon sink estimates of each pixel.

[0045] In some embodiments, if the estimated carbon sink values ​​of all pixels in a certain local area of ​​the study area differ significantly from the actual carbon sink values, local correction can be performed as follows: This yields the estimated carbon sink value for each pixel m in the study area. A continuous raster dataset is formed. Through raster stitching and regional statistical processing, a spatial distribution map of carbon sinks for the local area can be generated (i.e., the regional carbon sink estimation result for the local area). The carbon sink estimation values ​​of all pixels in the local area are statistically analyzed to calculate the average carbon sink level of the local area, as shown in the following expression: , This is an estimated value for the carbon sink of pixels in this local area. The number of pixels in this local area is then calculated; next, the results for this local area are compared with the reference carbon sink level provided by regional ecological statistics or existing monitoring studies. Compare the results and calculate the consistency calibration coefficient for the local area: In this embodiment, the reference carbon sink level of a local area is used as an example. The initial estimated mean of the model is Therefore, the regional consistency calibration coefficient is obtained: To ensure the stability of the calibration process, this embodiment preferably sets a reasonable range: When the calibration coefficient falls within this range, the model estimation results are considered to have good consistency at the regional scale, and there is no need to trigger the anomaly review mechanism; finally, the pixel carbon sink is calibrated using the following formula: After calibration, the average regional carbon sequestration value is: It is consistent with the reference statistical level.

[0046] S5. Following method S4, calculate the estimated carbon sink values ​​for all pixels in the study area, and then statistically summarize them to obtain the total estimated carbon sink value for the study area; such as Figure 4 As shown, this invention calculates the carbon sink estimate for all pixels in the study area, and can obtain... Figure 4 The spatial distribution map of the carbon sink estimates for the study area is shown.

[0047] An adaptive spatial weighted carbon sink estimation system includes a carbon sink estimation model, an adaptive spatial computation planning module, an adaptive spatial grid map, a carbon sink sample dataset, and a data acquisition and computation module. The carbon sink sample dataset internally stores environmental factor sample data, net primary productivity sample data, and carbon sink labels corresponding to data points. The data acquisition and computation module acquires multi-source datasets of the study area, including satellite multispectral imagery, meteorological data, land use data, and elevation data, and preprocesses them to obtain environmental factor data. The adaptive spatial computation planning module uses the comprehensive complexity index obtained from the multi-source dataset to classify graded regions in the adaptive spatial grid map and adaptively sets the data point density for each graded region. The carbon sink estimation model internally includes a spatial regression model. The carbon sink estimation model is trained using the carbon sink sample dataset for carbon sink prediction, and the spatial regression model is trained using spatial regression of the response coefficients corresponding to environmental factors. Based on the acquired net primary productivity data of the study area, the carbon sink estimation model obtains the carbon sink estimate for pixel m according to the following expression. : ,in For the net primary productivity of m pixels, Let m be the response coefficient of the environmental factor k. The standardized data corresponding to pixel m and environmental factor k This represents the total number of environmental factors.

[0048] To assess the spatial rationality and stability of the carbon sink estimation results, this embodiment further includes a result verification and uncertainty assessment module. First, a spatial consistency check is performed on the carbon sink spatial distribution results. Specifically, the carbon sink estimation raster is spatially overlaid with vegetation index data and land cover data to verify whether the spatial distribution of carbon sinks conforms to ecological laws. For example, high vegetation cover areas correspond to higher carbon sink values, while bare land and mining-disturbed areas correspond to lower carbon sink levels. Subsequently, the accuracy of the model prediction results is evaluated using error statistics indicators. The following indicators are preferably used: root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R²), and relative prediction deviation (RPD). In this embodiment, the calculation results are as follows:

[0049] An RPD greater than 1.8 indicates a "good to excellent" level for spatial mapping of carbon sinks in the mining area. Further, this embodiment optionally incorporates an uncertainty analysis module to assess the impact of model parameter perturbations on the carbon sink estimation results. Preferably, the Monte Carlo stochastic simulation method is used to randomly perturb the following variables: environmental response coefficient. And NPP data. The preferred number of simulations is set as follows: In this embodiment, the uncertainty assessment results are as follows:

[0050] High uncertainty areas are mainly distributed in: open-pit mining activity boundaries; newly reclaimed areas; and areas with significant interannual NDVI fluctuations.

[0051] An electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which, when executed, cause the at least one processor to perform the steps of the adaptive spatial fusion weighted carbon sink estimation method of the present invention.

[0052] A storage medium storing a computer program, which, when executed by a processor, implements the steps of the adaptive spatial fusion weighted carbon sink estimation method of the present invention.

[0053] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive spatial fusion weighted carbon sink estimation method, characterized in that: The methods include: S1. Construct an adaptive spatial grid map of the study area, obtain multi-source datasets including satellite multispectral images, meteorological data, land use data, and elevation data of the study area, and preprocess them to obtain environmental factor data; S2. Using the comprehensive complexity index obtained from multi-source datasets, the comprehensive complexity index is used to divide the hierarchical regions in the adaptive spatial grid map and to adaptively set the data point density of each hierarchical region. S3. Construct a carbon sink estimation model and a carbon sink sample dataset for the study area. The carbon sink sample dataset contains environmental factor sample data, net primary productivity sample data and carbon sink labels corresponding to the data points. The carbon sink estimation model has a spatial regression model. The carbon sink estimation model uses the carbon sink sample dataset for carbon sink prediction training, and the spatial regression model performs spatial regression training on the response coefficients corresponding to the environmental factors. S4. Obtain net primary productivity data for the study area. The carbon sink estimation model uses the following expression to obtain the carbon sink estimate for cell m. : ,in For the net primary productivity of m pixels, Let m be the response coefficient of the environmental factor k. The standardized data corresponding to pixel m and environmental factor k This represents the total number of environmental factors.

2. The adaptive spatial fusion weighted carbon sink estimation method according to claim 1, characterized in that: It also includes the following methods: S5. Following method S4, calculate the estimated carbon sink values ​​of all pixels in the study area and summarize them to obtain the total estimated carbon sink value of the study area.

3. The adaptive spatial fusion weighted carbon sink estimation method according to claim 1, characterized in that: In method S3, the spatial regression model obtains the response coefficients corresponding to environmental factors as follows: S31. The spatial regression model constructs the following dual-core fusion weight function using each data point in the study area method S2: ,in The fusion weights between data points i and j For geographic proximity kernel function, For perturbation similarity kernel function; S32. In the carbon sink estimation model carbon sink prediction training, the spatial regression model is trained with model constraints using a dual-core fusion weight function, and the carbon sink estimation model outputs the response coefficients corresponding to the environmental factors for each data point. S33. The spatial regression model is extended by local regression interpolation to obtain the response coefficients corresponding to environmental factor k for all pixels.

4. The adaptive spatial fusion weighted carbon sink estimation method according to claim 3, characterized in that: In method S31, the geographic proximity kernel function between data points i and j is constructed as follows: S311. Construct a disturbance intensity function that includes topographic heterogeneity, vegetation heterogeneity, and surface heterogeneity. The expression for the disturbance intensity function is as follows: The perturbation intensity of data point i, For terrain heterogeneity, For vegetation heterogeneity, For surface heterogeneity, , , These are the weighting coefficients; S312. Normalize the disturbance intensity to generate a normalized disturbance index. ; S313. Preset global reference bandwidth based on all data points across the entire study area. Construct the adaptive local bandwidth for data point i according to the following expression. : ; S314, the distance between data point i and data point j Construct the following geographic proximity kernel function: .

5. The adaptive spatial fusion weighted carbon sink estimation method according to claim 3 or 4, characterized in that: The expression for the perturbation similarity kernel function between data points i and j is as follows: , Let be the normalized perturbation exponent for data point i. Let j be the normalized perturbation exponent for data point j. The perturbation similarity sensitivity coefficient; Perturbation similarity sensitivity coefficient The method for obtaining this information is as follows: the difference in perturbation between data points i and j. Construct a carbon sink semivariogram conditioned on perturbation differences. ; Semivariogram of carbon sink The initial value of γ is obtained by using an exponential semivariogram model. Then, the objective function containing the root mean square error and the Akaike information criterion is constructed by combining the carbon sink estimation model and performing joint iterative optimization to obtain the optimal γ. The optimal γ is used as the perturbation similarity sensitivity coefficient.

6. The adaptive spatial fusion weighted carbon sink estimation method according to claim 3, characterized in that: The adaptive spatial grid map is divided into: The system comprises several hierarchical regions, each with a different grid resolution and a different data point density; the adaptive local bandwidth... Set the following bandwidth constraints: minimum bandwidth (Average distance between adjacent data points); Maximum bandwidth (Diagonal length of the study area); the bandwidth should not be less than the grid scale of the area to which the bandwidth belongs.

7. The adaptive spatial fusion weighted carbon sink estimation method according to claim 1, characterized in that: In method S1, the satellite multispectral image sources include Sentinel-2 satellite multispectral image data with cloud cover below 10%, and the satellite multispectral image sources have undergone preprocessing including radiometric correction and FLAASH atmospheric correction; the meteorological data includes monthly average temperature, monthly cumulative precipitation, and monthly average solar radiation; the elevation data is digital elevation model data; the adaptive spatial grid map is internally constructed with a WGS84 coordinate system, and the multi-source dataset is based on the WGS84 coordinate system for data mapping and spatial registration and alignment processing; The environmental factor items in the environmental factor data include vegetation index NDVI, vegetation index gradient, vegetation coverage, bare land ratio, slope factor, and meteorological factors, including monthly average temperature, monthly cumulative precipitation, and monthly average solar radiation.

8. The adaptive spatial fusion weighted carbon sink estimation method according to claim 1 or 7, characterized in that: In method S2, the comprehensive complexity index R is constructed as follows: , This is the normalized data corresponding to the slope factor in the environmental factor data; This is the normalized data corresponding to the vegetation index gradient in the environmental factor data; In environmental factor data The corresponding normalized data, i.e., the normalized data of the percentage of bare land surface area; , , These are the weighting coefficients.

9. The adaptive spatial fusion weighted carbon sink estimation method according to claim 8, characterized in that: In the adaptive spatial grid map, the study area is divided into three levels of regions, namely low, medium, and high levels, using the comprehensive complexity index. Different-resolution grids are set in each level region, and each grid has a different data point density. The range of the comprehensive complexity index R corresponding to the low-complexity level region in the study area is R ≤ 0.33, and the preset grid data is N1. The range of the comprehensive complexity index R corresponding to the medium-complexity level region in the study area is 0.33 < R ≤ 0.66, and the preset grid data is N2. The range of the comprehensive complexity index R corresponding to the high-complexity level region in the study area is R > 0.66, and the preset grid data is N3, where N1 < N2 < N3. The grid scale of the study area decreases in order from the low-level region to the high-level region, and the data point density of the grid is set from low to high in order from the low-level region to the high-level region.

10. The adaptive spatial fusion weighted carbon sink estimation method according to claim 1, characterized in that: In method S4, a regional ecosystem is constructed based on the study area, and a minimum allowable carbon sink value for the regional ecosystem of the study area is set. and maximum carbon sink value Carbon sequestration estimate for pixel m The following ecological constraints are established: Ecological constraints are adjusted for the carbon sink estimates of each pixel.

11. The adaptive spatial fusion weighted carbon sink estimation method according to claim 1, characterized in that: The net primary productivity sample data and the net primary productivity of pixel m were obtained at time t using the CASA light energy utilization model and a multi-source dataset according to the following expression. : ,in Photosynthetically active radiation absorbed by vegetation This refers to the utilization rate of light energy.

12. An adaptive spatial weighted carbon sink estimation system, characterized in that: The system includes a carbon sink estimation model, an adaptive spatial computation planning module, an adaptive spatial grid map, a carbon sink sample dataset, and a data acquisition and computation module. The carbon sink sample dataset stores environmental factor sample data, net primary productivity sample data, and carbon sink labels corresponding to data points. The data acquisition and computation module acquires multi-source datasets of the study area, including satellite multispectral imagery, meteorological data, land use data, and elevation data, and preprocesses them to obtain environmental factor data. The adaptive spatial computation planning module uses the comprehensive complexity index obtained from the multi-source dataset to perform hierarchical region division and adaptively set the data point density for each hierarchical region in the adaptive spatial grid map. The carbon sink estimation model includes a spatial regression model, which is trained using the carbon sink sample dataset for carbon sink prediction, and the spatial regression model is trained for spatial regression of the response coefficients corresponding to environmental factors. Based on the acquired net primary productivity data of the study area, the carbon sink estimation model obtains the carbon sink estimate for pixel m according to the following expression. : ,in For the net primary productivity of m pixels, Let m be the response coefficient of the environmental factor k. The standardized data corresponding to pixel m and environmental factor k This represents the total number of environmental factors.

13. An electronic device, characterized in that: Comprising: At least one processor; And a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to cause the at least one processor to execute the steps of the method according to any one of claims 1-11.

14. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-11.

Images