City agglomeration carbon sequestration capability driving factor key threshold determination method

Abstract

The invention discloses an urban agglomeration carbon sequestration capability driving factor key threshold determination method, and relates to the field of ecological environment, and the method comprises the steps: determining the causal effect between each driving factor and the carbon sequestration amount in a historical data set based on causal reasoning; determining reliable driving factors of the carbon sequestration amount according to a causal effect, and training a carbon sequestration amount regression model in combination with historical carbon sequestration amount data; calculating a Shapley value of each feature in the training sample; and drawing a dependency graph of each driving factor based on the Shapley value of each feature in all the training samples, determining a change point of scatter data in each dependency graph by adopting a change point detection method, and determining a key threshold value of the corresponding driving factor according to a driving factor slope value corresponding to the change point. According to the method, the internal causal mechanism of the driving factors and the carbon sequestration amount can be analyzed, the key threshold value is confirmed, and the urban agglomeration carbon cycle can be deeply and scientifically understood.

Description

Technical Field

[0001] The present invention relates to the field of ecological environment technology, and more specifically to a method for determining a key threshold value of a driving factor of an urban agglomeration's carbon sequestration capacity. Background Art

[0002] The carbon sequestration (CS) capacity of ecosystems plays a vital role in maintaining ecological balance and carbon cycle. With the rapid development of urbanization, urban agglomerations have become an important part of terrestrial ecosystems, playing an irreplaceable role in regulating atmospheric carbon dioxide concentration and carbon cycle. Therefore, studying the threshold of driving factors of urban agglomeration carbon sequestration capacity is of great significance for scientific understanding of carbon cycle, promoting sustainable development and mitigating climate change. It is also helpful for carbon sink assessment of urban agglomeration ecosystems under future climate change conditions.

[0003] However, most of the existing studies on the driving factors of carbon sequestration capacity use linear statistical regression and machine learning methods, which often only explore the correlation between one or several driving factors and CS, without fully considering the inherent causal relationship between the driving factors and CS, as well as the direct and indirect effects between the driving factors; in addition, the specific response threshold of urban agglomeration carbon sequestration to driving factors is still unclear. This makes it difficult to have a deeper and more scientific understanding of the carbon cycle of urban agglomerations, which is not conducive to building a sustainable urban ecosystem.

[0004] Therefore, how to explore the causal relationship between carbon sequestration capacity and driving factors, and determine the key threshold of driving factors, are issues that technical personnel in this field urgently need to solve. Summary of the invention

[0005] In view of this, the present invention provides a method for determining the key thresholds of driving factors of carbon sequestration capacity in urban agglomerations, which combines causal reasoning, machine learning, Shapley addition interpretation and change point detection to analyze the key thresholds of driving factors in urban agglomerations.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] The present invention discloses a method for determining a key threshold value of driving factors of carbon sequestration capacity of an urban agglomeration, and the specific steps are as follows:

[0008] Obtain historical carbon sequestration data of urban agglomerations and corresponding historical driving factor data to form a historical data set;

[0009] Based on causal reasoning, determining the causal effects between the driving factors in the historical data set;

[0010] Determine a reliable driving factor for carbon sequestration according to the causal effect, and obtain training samples based on historical data of the reliable driving factor for carbon sequestration and the historical carbon sequestration data, and train a carbon sequestration regression model;

[0011] The Shapley addition method is used to determine the Shapley value of each feature in each of the training samples, and the model output result is interpreted as a linear function of a binary variable;

[0012] Based on the Shapley values ​​of each feature in all the regression samples, a dependency diagram of each driving factor is drawn, and a change point detection method is used to determine the change point of the scattered data in each dependency diagram, and the key threshold of the corresponding driving factor is determined according to the slope value of the driving factor corresponding to the change point.

[0013] Preferably, the carbon sequestration data is obtained through net primary productivity, including carbon sequestration data of urban areas and non-urban areas; the driving factors include: annual afforestation area, population density, night lights, solar radiation, temperature, precipitation, and chlorophyll fluorescence.

[0014] Preferably, the causal reasoning specifically includes:

[0015] Taking carbon sequestration and driving factors as nodes of a directed acyclic graph, the edges of the directed acyclic graph are determined based on the assumed causal relationship, and a directed acyclic graph describing the assumed causal relationship between different driving factors and carbon sequestration is established;

[0016] The correctness of the initial assumptions of the directed acyclic graph is verified by using methods such as the front-door criterion and the back-door criterion to determine the final causal effect.

[0017] Preferably, the calculation of the causal effect includes: identifying the causal relationship between nodes (variables) of the directed acyclic graph according to the front-door criterion and the back-door criterion; calculating the causal effect value between two points according to the causal effect formula corresponding to the front-door criterion and the back-door criterion;

[0018] The backdoor criterion is: a pair of ordered variables (X, Y) is selected in a directed acyclic graph, and if the variable set Z satisfies that there is no descendant node of variable X in Z, and Z blocks all paths between variable X and variable Y that point to X, then the variable set Z satisfies the backdoor criterion about (X, Y);

[0019] The front door criterion is: a pair of ordered variables (X, Y) is selected in a directed acyclic graph. If the variable set Z satisfies that Z cuts off all directed paths from X to Y, and there is no back door path from X to Z, and all back door paths from Z to Y are blocked by X, then the variable set Z satisfies the front door criterion about (X, Y);

[0020] The backdoor causal effect formula is:

[0021] P(Y=y|do(X=x))=∑ z P(Y=y|X=x,Z=z)P(Z=z);

[0022] The front door causal effect formula is:

[0023] P(Y=y|do(X=x))=∑zP(Z=z|do(X=x))P(Y=y|do(z))=∑ z P(z|x)∑ x' P(y|x',z)P(x').

[0024] Preferably, in the training samples, the sample characteristics include: urban area and non-urban area characteristics, and driving factor characteristics; the sample label of the training samples is the carbon sequestration slope, and the driving factor characteristics are the slopes of each driving factor.

[0025] Preferably, the carbon fixation regression model is a LightGBM model, which discretizes continuous features based on a histogram algorithm, and adopts a Leaf-wise growth strategy based on leaf growth. The objective function of the LightGBM model is:

[0026]

[0027] In the formula, Obj t is the objective function; yi is the actual value of the regression target, is the t-th regression value of the model; n is the number of samples; Ω is the regularization term of the model.

[0028] Preferably, the LightGBM model uses a grid search algorithm to find the global optimal hyperparameters, and the specific steps are as follows:

[0029] Define hyperparameter space: set the value range of hyperparameters;

[0030] Create hyperparameter combinations: divide the hyperparameter space into different subspaces and generate hyperparameter combinations;

[0031] Training and evaluating the model: According to the settings of each hyperparameter combination, the model is trained and the model performance is evaluated through cross-validation method;

[0032] Select the best model: Based on the evaluation results, select the hyperparameter combination with the best performance.

[0033] Preferably, the binary variable linear function is:

[0034]

[0035] In the formula, x' is the input variable after being simplified by the mapping function; M is the number of input features.

[0036] Preferably, the dependency graph is a scatter plot, used to show the impact of a single driving factor on carbon fixation; each point in the dependency graph represents a sample point, the x-axis is the slope of the driving factor, and the y-axis is the Shapley value of the driving factor, indicating the degree of change of the driving factor characteristics on the output of the carbon fixation regression model.

[0037] Preferably, the change point detection method is a dynamic programming method, which determines the change points of the scattered data in the dependency graph by minimizing a cost function; the minimization cost function is:

[0038]

[0039] In the formula, τ represents the change point, K represents the number of change points, Represents the loss function of the sequence.

[0040] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method for determining the key thresholds of driving factors of carbon sequestration capacity of urban agglomerations. Based on causal reasoning, the causal effects between the driving factors are determined, thereby considering the inherent causal relationship between the driving factors and analyzing the impact between the driving factors and the carbon sequestration amount. The machine learning model and Shapley addition interpretation are further used to analyze the contribution of each influencing factor to the carbon sequestration amount, and the key thresholds of each driving factor are accurately determined through dependency graphs and change point detection methods, so as to achieve the detection of the key thresholds of each driving factor, and provide reference and reference for the scientific management of carbon cycle and ecosystem. The present invention can analyze the inherent causal relationship between each driving factor and carbon sequestration, accurately detect the key thresholds of each driving factor, so as to have a deeper and more scientific understanding of the carbon cycle of urban agglomerations, better mitigate climate change, and achieve sustainable development of urban ecosystems. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0042] Figure 1 It is a schematic diagram of the overall process of an embodiment of the present invention.

[0043] Figure 2 The figure is a schematic diagram of the overall process of the CMSC framework according to an embodiment of the present invention.

[0044] Figure 3It is a DAG directed acyclic graph of an embodiment of the present invention.

[0045] Figure 4 This is a ranking diagram of the importance of different driving factors to CS according to an embodiment of the present invention.

[0046] Figure 5 This is a driving factor dependency diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] The embodiment of the present invention discloses a method for determining a key threshold value of a driving factor of carbon sequestration capacity of an urban agglomeration, such as Figure 1 As shown, the specific steps are as follows:

[0049] Obtain historical carbon sequestration data of urban agglomerations and corresponding historical driving factor data to form a historical data set;

[0050] Based on causal reasoning, determine the causal effect between each driving factor and carbon sequestration in the historical data set;

[0051] Determine the reliable driving factors of carbon sequestration based on the causal effect, organize the training samples based on the historical data of the reliable driving factors of carbon sequestration and the historical carbon sequestration data, and train the carbon sequestration regression model;

[0052] The Shapley addition interpretation is used to determine the Shapley value of each feature in each training sample, and the output of the model is interpreted as a linear function of binary variables;

[0053] Based on the Shapley value of each feature in all training samples, the dependency diagram of each driving factor is drawn, and the change point detection method is used to determine the change point of the scattered data in each dependency diagram, and the critical threshold of the corresponding driving factor is determined according to the slope value of the driving factor corresponding to the change point.

[0054] In a specific embodiment, the carbon sequestration data is obtained through net primary productivity, including carbon sequestration data for urban areas and non-urban areas; driving factors include: annual afforestation area, population density, night lights, solar radiation, temperature, precipitation, and chlorophyll fluorescence.

[0055] In a specific embodiment, causal reasoning specifically includes:

[0056] Taking carbon sequestration and driving factors as nodes of a directed acyclic graph, the edges of the directed acyclic graph are determined based on the assumed causal relationship, and a directed acyclic graph describing the assumed causal relationship between different driving factors and carbon sequestration is established;

[0057] The front-door criterion, back-door criterion and other methods are used to verify the correctness of the initial assumptions of the directed acyclic graph and determine the final causal effect.

[0058] In a specific embodiment, the calculation of the causal effect includes: identifying the causal relationship between nodes (variables) of the directed acyclic graph according to the front-door criterion and the back-door criterion; calculating the causal effect value between two points according to the causal effect formula corresponding to the front-door criterion and the back-door criterion;

[0059] The backdoor criterion is: select a pair of ordered variables (X, Y) in a directed acyclic graph. If the variable set Z satisfies that there is no descendant node of variable X in Z, and Z blocks all paths between variables X and Y that point to X, then the variable set Z satisfies the backdoor criterion about (X, Y);

[0060] The front door criterion is: select a pair of ordered variables (X, Y) in a directed acyclic graph. If the variable set Z satisfies that Z cuts off all directed paths from X to Y, and there is no back door path from X to Z, and all back door paths from Z to Y are blocked by X, then the variable set Z satisfies the back door criterion about (X, Y);

[0061] The backdoor causal effect formula is:

[0062] P(Y=y|do(X=x))=∑zP(Y=y|X=x,Z=z)P(Z=z);

[0063] The front door causal effect formula is:

[0064] P(Y=y|do(X=x))=∑ z P(Z=z|do(X=x))P(Y=y|do(z))=∑ z P(z|x)∑ x' P(y|x',z)P(x').

[0065] In the above-mentioned front-door and back-door causal effect formulas, Z represents the set of driving factors and carbon sequestration; X represents the driving factors; and Y represents the carbon sequestration.

[0066] In a specific embodiment, in the training samples, the sample characteristics include: urban area and non-urban area characteristics, and driving factor characteristics; the sample label of the training sample is the carbon sequestration slope, and the driving factor characteristics are the slopes of each driving factor.

[0067] In a specific embodiment, the carbon fixation regression model is a LightGBM model. The LightGBM model discretizes continuous features based on a histogram algorithm. The LightGBM model adopts a Leaf-wise growth strategy based on leaf growth. The objective function of the LightGBM model is:

[0068]

[0069] Where, Obj t is the objective function; yi is the actual value of the regression target, is the t-th regression value of the model; n is the number of samples; Ω is the model regularization term; f i Represents the i-th decision tree.

[0070] In a specific embodiment, the hyperparameter range of the LightGBM machine learning algorithm to be optimized is shown in Table 1.

[0071] Table 1 LightGBM model hyperparameter range

[0072]

[0073] For the test set, the LightGBM model was used for regression, and the regression results were compared with the actual CS slopes of the corresponding cities and counties and non-cities and counties. The reliability of the regression model was verified by the mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE). The three evaluation indicators of MSE, RMSE, and MAE are calculated by the following formula:

[0074]

[0075] Where Y j is the true value of the test set sample data; is the regression value obtained by the model, n is the total number of samples in the test set, and the smaller the MSE, RMSE, and MAE values ​​are, the better the model performance is.

[0076] In a specific embodiment, the LightGBM model uses a grid search algorithm to find the global optimal hyperparameters. The specific steps are as follows:

[0077] Define hyperparameter space: set the value range of hyperparameters;

[0078] Create hyperparameter combinations: divide the hyperparameter space into different subspaces and generate hyperparameter combinations;

[0079] Training and evaluating the model: According to the settings of each hyperparameter combination, the model is trained and the model performance is evaluated through cross-validation method;

[0080] Select the best model: Based on the evaluation results, select the hyperparameter combination with the best performance.

[0081] In a specific embodiment, the binary variable linear function is:

[0082]

[0083] In the formula, Shapley value R is a set of real numbers; x' is the input variable after simplification by the mapping function; M is the number of input features.

[0084] In a specific embodiment, the dependency graph is a scatter plot used to show the impact of a single driving factor on carbon sequestration; each point in the dependency graph represents a sample point, the x-axis is the slope of the driving factor, and the y-axis is the Shapley value of the driving factor, indicating the degree of change in the output of the carbon sequestration regression model caused by the characteristics of the driving factor.

[0085] In a specific embodiment, the change point detection method is a dynamic programming method, which determines the change points of scattered data in the dependency graph by minimizing the cost function; the minimized cost function is:

[0086]

[0087] Wherein, τ represents the change point; K represents the number of change points. In this embodiment, K is set to 1, that is, one change point is detected; represents the loss function of the sequence. In this embodiment, the loss function is the sum of squared errors.

[0088] In a specific embodiment, in order to determine the threshold of the driving factor affecting carbon sequestration, the following steps are taken: (1) Data collection: Obtain the driving factor trend (slope) and the dependency graph with the corresponding Shapley value. (2) Data sorting: Arrange the slope of each driving factor in ascending order, ensuring that the Shapley value is aligned with the order of the driving variable slope. (3) Threshold detection: Perform a change point detection algorithm to detect the key change points of the Shapley value. The corresponding change point is determined as the threshold of the specific driving factor.

[0089] In a specific embodiment, the county division of a certain urban agglomeration from 2001 to 2020 is taken as an example for explanation.

[0090] Specifically, the county-level carbon sequestration (CS) from 2001 to 2020 was obtained through net primary productivity (NPP); NPP, as a product of CS in the ecosystem, represents the efficiency of vegetation in fixing and converting photosynthetic products.

[0091] The annual afforestation area (AAF) reflects the contribution of human activities to the vegetation coverage area and represents the cumulative afforestation area each year. AAF is one of the important indicators for achieving ecological governance and improving the carbon sequestration capacity of ecosystems. AAF is extracted from the "China Forestry and Grassland Statistical Yearbook".

[0092] The annual NPP-VIIRS Night Lights (NTL) product is developed through a cross-sensor calibration scheme and the spatial resolution of the NTL product is 1 km.

[0093] Population density (POP) data uses WorldPop's population density product.

[0094] The Solar Induced Chlorophyll Fluorescence (CSIF) dataset is sourced from the National Tibetan Plateau / Third Pole Environmental Data Center. Specifically, CSIF provides global SIF data with a spatial resolution of 0.05° and a temporal resolution of 4 days, integrating the annual average CSIF data of counties in the urban agglomeration.

[0095] The county-level administrative division data across urban agglomerations come from the global administrative boundaries, and the vector boundary coordinate system of the 214 counties is WGS1984.

[0096] This paper proposes a new framework to detect the threshold of driving factors affecting CS changes. The proposed framework integrates causal reasoning, machine learning, Shapley addition interpretation and change point detection, and is named CMSC framework. Figure 2 As shown in Figure 2, CMSC is used to gain in-depth insights into the driving mechanisms of CS and determine the thresholds of driving factors that affect CS. After causal reasoning, historical data is imported into the LightGBM model to train the CS regression model.

[0097] In the embodiment, DoWhy is used to quantitatively study the causal relationship between the driving factors and CS. The present invention describes the assumed causal relationship between different variables and CS based on DAG (directed acyclic graph), where the nodes represent variables ( Figure 3 Indicated by blue ovals and green boxes), edges ( Figure 3 The black arrows in the figure represent the direct or indirect causal relationship between the variable and CS. All variables in the DAG represent the slope of the corresponding variable, and β represents the unobserved confounder.

[0098] like Figure 3 As shown in the figure, the direct or indirect impact of different driving factors on CS is explained by the causal graph model. The causal effects of urban and non-urban areas quantitatively estimated based on the graph model are 1.56×10 -5 (p<0.001) and 1.01×10 -5(p<0.05). The causal effects of the constructed DAGs for urban and non-urban areas were significant, indicating that there is a causal interaction between the effects of different driving factors on CS changes. In addition, the verification results of the causal effect are shown in Table 2. The test results of the random confounding factor and data subset methods are very close to the estimated value of the causal effect, while the results of the placebo intervention method are approximately equal to 0, confirming that there is a causal effect between the driving factors and CS. The results of causal inference show the potential causal mechanism of urban agglomeration urban and non-urban areas driving vegetation CS.

[0099] Table 2 Robustness and accuracy verification of causal effect estimation results

[0100]

[0101] The optimal parameter results of the LightGBM model are shown in Table 3, and the accuracy evaluation results are shown in Table 4, which shows that LightGBM has a stable CS regression ability in urban agglomerations, and its MSE, RMSE and MAE values ​​are all smaller than MLR. Compared with the traditional linear statistical regression (MLR) method, the MSE, RMSE and MAE of the LightGBM model are reduced by 33.44% (51.8%), 18.18% (30.25%) and 15.91% (43.3%) in urban areas (non-urban areas), respectively. The regression results show that the county-level CS slope obtained by regression is consistent with the actual CS slope of the test data set, which verifies the practicality and reliability of the LightGBM model. In summary, the LightGBM based on causal reasoning can fully reflect the nonlinear relationship between driving factors and CS, and accurately regress the CS of urban areas and non-urban areas in urban agglomerations.

[0102] Table 3 Parameter adjustment results of LightGBM regression model based on causal reasoning

[0103]

[0104] Table 4 Accuracy verification of LightGBM regression model based on causal reasoning

[0105]

[0106] For NTL, POP, T a The importance of the direct and indirect effects of TP and SSRD on the spatiotemporal pattern of CS is analyzed as follows: Figure 4 As shown in Figure 2, SSRD has a relatively weak contribution to the CS changes in urban and non-urban areas, and NTL is the most important factor affecting the CS trend changes in urban areas (feature importance: 0.0041) and non-urban areas (feature importance: 0.0034). This confirms the great impact of human activities on CS. aThe contribution to CS ranks third (feature importance: 0.0012) and second (feature importance: 0.0025) in urban areas and non-urban areas, respectively.

[0107] Based on the SHAP value, the degree and direction (positive or negative) of the impact of different driving factors on CS changes in each district and county were obtained, such as Figure 5 As shown in Figure 2. In the sample of all districts and counties, the TP and SSRD slopes in the urban agglomeration are positively correlated with the CS trend. With the increase of SSRD slope, the SHAP value of CS increases significantly. The SHAP value of TP also has a similar trend, which indicates that higher radiation and moisture conditions lead to the gain of CS. For urban and non-urban areas, NTL and T a The slope is negatively correlated with the SHAP value, indicating that an increase in the intensity of human activities and an increase in temperature are more likely to lead to a decrease in CS. On the other hand, population growth (decline) has a significant positive (negative) contribution to CS, and the carbon sequestration capacity of non-municipal areas decreases with the increase in the slope of POP, indicating that a decrease in POP leads to an increase in CS.

[0108] according to Figure 5 In the urban area, the NTL threshold, SSRD threshold, Ta threshold, and TP threshold affecting CS are 0.04 nW·cm -2 ·sr -1 ·yr -1 , -6.1×10 4 J.m -2 ·yr -1 、0.013K·yr -1 , 3.64×10 -5 m·yr -1 ; In non-urban areas, the NTL threshold, SSRD threshold, and T a The threshold, TP threshold, and POP threshold are 0.4 nW·cm -2 ·sr -1 ·yr -1 , -5.46×10 4 J.m -2 ·yr -1 、0.017K·yr -1 , 2.51×10 -5 m·yr -1 、-0.04people·km -2 ·yr -1 .

[0109] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the embodiments can be referred to each other. The above description of the disclosed embodiments enables professionals and technicians in this field to implement or use the present invention. Various modifications to these embodiments will be obvious to professionals and technicians in this field, and the general principles defined herein can be implemented in other embodiments without departing from the scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown in this article, but will comply with the widest range consistent with the principles and novel features disclosed herein.

Claims

1. A method for determining the key threshold of driving factors of urban agglomeration carbon sequestration capacity, characterized in that: The specific steps are as follows: Obtain historical carbon sequestration data of urban agglomerations and corresponding historical driving factor data to form a historical data set; Based on causal reasoning, determine the causal effect between each driving factor in the historical data set and carbon sequestration; Determine a reliable driving factor for carbon sequestration according to the causal effect, and obtain training samples based on historical data of the reliable driving factor for carbon sequestration and the historical carbon sequestration data, and train a carbon sequestration regression model; The Shapley addition method is used to determine the Shapley value of each feature in each of the training samples, and the output of the model is interpreted as a linear function of a binary variable; Based on the Shapley value of each feature in all the training samples, a dependency graph of each driving factor is drawn, and a change point detection method is used to determine the change point of the scattered data in each dependency graph, and the key threshold of the corresponding driving factor is determined according to the slope value of the driving factor corresponding to the change point.

2. The method for determining the key threshold of driving factors of urban agglomeration carbon sequestration capacity according to claim 1 is characterized in that: The carbon sequestration data is obtained through net primary productivity, including carbon sequestration data for urban areas and non-urban areas; the driving factors include: annual afforestation area, population density, night lights, solar radiation, temperature, precipitation, and chlorophyll fluorescence.

3. The method for determining the key threshold of driving factors of carbon sequestration capacity of urban agglomerations according to claim 1 is characterized in that: The causal reasoning specifically includes: Taking carbon sequestration and driving factors as nodes of a directed acyclic graph, the edges of the directed acyclic graph are determined based on the assumed causal relationship, and a directed acyclic graph describing the assumed causal relationship between different driving factors and carbon sequestration is established; The correctness of the initial assumptions of the directed acyclic graph is verified by using methods such as the front-door criterion and the back-door criterion to determine the final causal effect.

4. The method for determining the key threshold of driving factors of carbon sequestration capacity of urban agglomerations according to claim 3 is characterized in that: The calculation of the causal effect includes: identifying the causal relationship between the nodes of the directed acyclic graph according to the front door criterion and the back door criterion; calculating the causal effect value between two points according to the causal effect formula corresponding to the front door criterion and the back door criterion; The backdoor criterion is: a pair of ordered variables (X, Y) is selected in a directed acyclic graph, and if the variable set Z satisfies that there is no descendant node of variable X in Z, and Z blocks all paths between variable X and variable Y that point to X, then the variable set Z satisfies the backdoor criterion about (X, Y); The front door criterion is: a pair of ordered variables (X, Y) is selected in a directed acyclic graph. If the variable set Z satisfies that Z cuts off all directed paths from X to Y, and there is no back door path from X to Z, and all back door paths from Z to Y are blocked by X, then the variable set Z satisfies the front door criterion about (X, Y); The backdoor causal effect formula is: P(Y=u|do(X=x))=∑ z P(Y=y|X=x,Z=z)P(Z=z); The front door causal effect formula is: P(Y=y|do(X=x))=∑ z P(Z=z|do(X=x))P(Y=y|do(z))=∑ z P(z|x)∑ x' P(y|x',z)P(x'); Where Z represents the combination of driving factors and carbon sequestration; X represents the driving factors; and Y represents the carbon sequestration.

5. The method for determining the key threshold of driving factors of urban agglomeration carbon sequestration capacity according to claim 1 is characterized in that: In the training samples, the sample characteristics include: urban district and non-urban district characteristics, and driving factor characteristics; the sample label of the training samples is the carbon fixation slope, and the driving factor characteristics are the slopes of each driving factor.

6. The method for determining the key threshold of driving factors of urban agglomeration carbon sequestration capacity according to claim 1 is characterized in that: The carbon fixation regression model is a LightGBM model. The LightGBM model discretizes continuous features based on a histogram algorithm. The LightGBM model adopts a Leaf-wise growth strategy based on leaf growth. The objective function of the LightGBM model is: Where, Obj t is the objective function; yi is the actual value of the regression target, is the t-th regression value of the model; n is the number of samples; Ω is the regularization term of the model.

7. The method for determining the key threshold of driving factors of urban agglomeration carbon sequestration capacity according to claim 6 is characterized in that: The LightGBM model uses a grid search algorithm to find the global optimal hyperparameters. The specific steps are as follows: Define hyperparameter space: set the value range of hyperparameters; Create hyperparameter combinations: divide the hyperparameter space into different subspaces and generate hyperparameter combinations; Training and evaluating the model: According to the settings of each hyperparameter combination, the model is trained and the model performance is evaluated through cross-validation method; Select the best model: Based on the evaluation results, select the hyperparameter combination with the best performance.

8. The method for determining the key threshold of driving factors of urban agglomeration carbon sequestration capacity according to claim 1 is characterized in that: The binary variable linear function is: In the formula, x' is the input variable after being simplified by the mapping function; M is the number of input features.

9. The method for determining the key threshold of driving factors of urban agglomeration carbon sequestration capacity according to claim 1 is characterized in that: The dependency graph is a scatter plot used to show the impact of a single driving factor on carbon fixation; each point in the dependency graph represents a sample point, the x-axis is the slope of the driving factor, and the y-axis is the Shapley value of the driving factor, indicating the degree of change in the output of the carbon fixation regression model caused by the characteristics of the driving factor.

10. The method for determining the key threshold of driving factors of urban agglomeration carbon sequestration capacity according to claim 1 is characterized in that: The change point detection method is a dynamic programming method, which determines the change points of the scattered data in the dependency graph by minimizing the cost function; the minimization cost function is: In the formula, τ represents the change point, K represents the number of change points, Represents the loss function of the sequence.

Images