A method for spatial targeting of river methane emission reduction based on bivariate spatial clustering

Abstract

The application provides a river methane space targeted emission reduction method based on a bivariate space clustering, and belongs to the technical field of environmental protection.The method takes a watershed landscape feature and water quality as core regulation factors of methane emission in bivariate, and sets a planned space targeted emission reduction strategy based on an action mechanism between the core regulation factors and methane emission, so that the methane emission reduction effect can be improved and environmental protection can be promoted.

Description

Technical Field

[0001] This invention relates to the field of environmental protection technology, and in particular to a spatially targeted method for reducing methane emissions from rivers based on bivariate spatial clustering. Background Technology

[0002] Methane (CH4) is a potent greenhouse gas. River methane emissions refer to the process by which methane is generated and released into the atmosphere through microbial metabolism and other processes within river ecosystems. Methane emissions into the atmosphere have harmful effects, including exacerbating the greenhouse effect, damaging watershed aquatic environments, and disrupting the carbon cycle.

[0003] Currently, targeted adjustment measures are often used to reduce methane emissions in order to achieve the goal of environmental protection. However, existing methane emission reduction measures have the defect of unreasonable planning, resulting in poor methane emission reduction effect and failure to protect the environment well. Summary of the Invention

[0004] This invention proposes a spatially targeted methane emission reduction method for rivers based on bivariate spatial clustering. The method uses watershed landscape characteristics and water quality as bivariates to determine the core regulatory factors of methane emissions. Based on the interaction mechanism between the core regulatory factors and methane emissions, a reasonable spatially targeted emission reduction strategy is set up, which can improve the methane emission reduction effect and promote environmental protection.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a spatially targeted method for river methane emission reduction based on bivariate spatial clustering, comprising: First, the core regulatory factors for controlling methane emissions in the target watershed are identified; these core regulatory factors include primary landscape factors and primary water quality factors. Next, the spatial correlations among these core regulatory factors within the target watershed are identified. Based on the numerical distribution and spatial correlations of these core regulatory factors, the dominant type of methane emissions in the target watershed is determined; the dominant type indicates the proportion of methane emission control by the core regulatory factors. Finally, based on the dominant type and the interaction effects of the core regulatory factors on methane emissions, a spatially targeted emission reduction strategy for controlling methane emissions in the target watershed is generated.

[0006] In one implementation method, determining the core regulatory factors for controlling methane emissions in a target watershed includes: first, acquiring multi-source data for the target watershed, including methane emission flux, water quality parameters, and landscape characteristic data for each of the multiple sub-watersheds within the target watershed; then, constructing a machine learning regression model to learn the interaction between landscape characteristic data, water quality parameters, and methane emission flux in the multi-source data; and finally, using SHAP values ​​to interpret and analyze the machine learning regression model to identify the primary landscape factor with the greatest impact on methane emission flux among the landscape characteristic data and the primary water quality factor with the greatest impact on methane emission flux among the water quality parameters, and determining the primary landscape factor and primary water quality factor as the core regulatory factors.

[0007] In one implementation, identifying the spatial correlation between core regulatory factors within a target watershed includes: calculating the global spatial autocorrelation index among the core regulatory factors of the target watershed; the global spatial autocorrelation index satisfies: in, Represents the global spatial autocorrelation index; This represents the total number of sub-basins within the target watershed; This represents the spatial weight between the i-th and j-th sub-basins within the target watershed; This represents the primary landscape factor value of the i-th sub-basin. This represents the mean of the primary landscape factors in the target watershed; This represents the primary water quality factor value for the i-th sub-basin. This represents the mean of the primary water quality factor in the target watershed. Given that the bivariate global spatial autocorrelation index satisfies the first condition, the local spatial autocorrelation index of each sub-watershed is calculated; the local spatial autocorrelation index satisfies: in, This represents the local spatial autocorrelation index of the i-th sub-basin. It indicates the spatial correlation between core regulatory factors within the i-th sub-basin of the target watershed.

[0008] In one implementation, the dominant type of methane emissions in the target watershed includes the dominant type of methane emissions in each of the multiple sub-watersheds contained within the target watershed. Based on the numerical distribution and spatial correlation of the core regulatory factors in the target watershed, the dominant type of methane emissions in the target watershed is determined, including: for each of the multiple sub-watersheds, when the sub-watershed satisfies... When the dominant type of the sub-basin is determined to be the dual-high synergistic type; when the sub-basin meets the following conditions... When the dominant type of the sub-basin is determined to be the low background type, the sub-basin will be determined to be the low background type. When the dominant type of the sub-basin is determined to be the dominant landscape anomaly type; when the sub-basin meets the following conditions... At that time, the dominant type of the sub-basin was determined to be the dominant water quality anomaly type; among them, Represents the local spatial autocorrelation index of the i-th sub-basin; This represents the primary landscape factor value of the i-th sub-basin. This represents the mean of the primary landscape factors in the target watershed. and Indicates the numerical distribution of primary landscape factors; This represents the primary water quality factor value for the i-th sub-basin. This represents the mean value of the primary water quality factor in the target watershed. and Indicates the numerical distribution of primary water quality factors.

[0009] In one implementation, a spatially targeted emission reduction strategy for methane emissions in a target watershed is generated based on the dominant type and the interaction effect of core regulatory factors on methane emissions. This includes: using a two-dimensional partial dependency graph to analyze the interaction effect of core regulatory factors on methane emission flux in a machine learning regression model, identifying key thresholds for the core regulatory factors; these key thresholds include the structural transformation threshold of primary landscape factors and the functional transformation threshold of primary water quality factors. Based on the key thresholds and dominant type, the regulation results for the target watershed under various regulation scenarios are predicted; the regulation results include methane emission reduction rate and socioeconomic costs. Based on the regulation results and dominant type, a spatially targeted emission reduction strategy for methane emissions in the target watershed is generated.

[0010] In one implementation, multiple control scenarios include a baseline scenario, a single-factor control scenario, a multi-factor combined control scenario, and a unified control scenario for the entire watershed. The baseline scenario does not control the target watershed. The single-factor control scenario only adjusts the landscape / controls the water quality of sub-watersheds with dominant landscape anomalies. The multi-factor combined control scenario adjusts the landscape of sub-watersheds with dominant landscape anomalies and controls the water quality of sub-watersheds with dominant water quality anomalies. The unified control scenario for the entire watershed only controls the sub-watersheds where the core control factor value is greater than the critical threshold.

[0011] In one implementation, based on key thresholds and dominant types, the regulation results of the target watershed under multiple regulation scenarios are predicted. This includes: acquiring regulation parameters for each of the multiple regulation scenarios, including water quality parameter regulation values ​​and landscape feature data regulation values. For each of the multiple regulation scenarios, the water quality parameters, landscape feature data, and regulation parameters of the target watershed are input into a machine learning regression model, and the machine learning regression model outputs the regulation results of the target watershed under the regulation scenario.

[0012] In one implementation, the spatially targeted emission reduction strategy for regulating methane emissions in the target watershed includes: the dominant type of methane emissions in each sub-watershed of the target watershed, the type of regulation measures, the key threshold, the expected methane emission reduction rate, and the socio-economic costs.

[0013] Compared with the prior art, the present invention has the following beneficial effects.

[0014] This invention provides a spatially targeted methane emission reduction method for rivers based on bivariate spatial clustering. First, the primary landscape factors and primary water quality factors for methane emissions in the core target watershed are determined. Then, based on their spatial correlation and numerical distribution, the dominant type of methane emissions in the target watershed is identified. Next, a spatially targeted emission reduction strategy is generated based on the dominant type and the interaction effects between the primary landscape factors, primary water quality factors, and methane emissions. This process follows the actual mechanism of methane emissions within the target watershed, ensuring that the formulated spatially targeted emission reduction strategy is reasonable, improves methane emission reduction effectiveness, and promotes environmental protection. Attached Figure Description

[0015] Figure 1 This is one of the schematic diagrams of a spatially targeted methane emission reduction method for rivers based on bivariate spatial clustering provided in the embodiments of this application; Figure 2 This is the second schematic diagram of a spatially targeted methane emission reduction method for rivers based on bivariate spatial clustering provided in this application embodiment; Figure 3 This is a logical schematic diagram of a spatially targeted methane emission reduction method for rivers in an application scenario provided by an embodiment of this application; Figure 4 This is the third schematic diagram of a spatially targeted methane emission reduction method for rivers based on bivariate spatial clustering provided in this application embodiment; Figure 5 This is the fourth of the river methane spatially targeted emission reduction methods based on bivariate spatial clustering provided in this application. Detailed Implementation

[0016] In the specification and claims of this invention, the terms "first" and "second," etc., are used to distinguish different objects, rather than to describe a specific order of objects.

[0017] In the embodiments of this application, "and / or" indicates a relationship between objects. For example, A and / or B can represent the following three situations: A exists alone, B exists alone, and A and B exist simultaneously.

[0018] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0019] In the description of this invention, unless otherwise stated, "a plurality of" means two or more. For example, a plurality of sub-basins means two or more sub-basins.

[0020] The method provided in this application relates to river methane emission reduction, and river methane emission reduction measures can be set through the interaction mechanism between landscape features, water quality and methane emissions.

[0021] Understandably, current river methane (CH4) emissions continue to increase due to the complex impacts of human activities, but existing watershed-scale emission reduction strategies mainly face the following five technical challenges.

[0022] 1. Spatial heterogeneity is overlooked: Existing technologies often adopt a "one-size-fits-all" approach to intervention across the entire watershed, which fails to adapt to the spatial heterogeneity of landscape characteristics and water quality within the watershed, leading to inefficient allocation of emission reduction resources. Spatial heterogeneity refers to the differences in landscape (such as land use and topography) at different locations within the watershed, resulting in varying impacts on water quality depending on spatial location.

[0023] 2. The nonlinear threshold effect has not been quantified: Landscape components (such as the proportion of cultivated land) and water quality indicators (such as total nitrogen concentration) exhibit significant nonlinear characteristics and critical thresholds in regulating river CH4 emissions, but current technologies lack quantitative identification methods for these key thresholds. In reality, CH4 emissions from rivers in agricultural areas are primarily induced by acetic acid fermentation, while CH4 emissions from urban rivers are primarily induced by hydrogen trophic pathways.

[0024] 3. Unclear coupling mechanism of driving factors: Existing models fail to fully consider the heterogeneity within sub-basins and cannot clearly characterize the coupling relationship between water pollution load and landscape driving factors, making it difficult to determine priority control areas.

[0025] 4. Lack of predictive strategies for the future: Existing emission reduction methods are mostly based on historical data and do not effectively couple climate change and socio-economic development scenarios (such as the Shared Socioeconomic Pathways (SSPs) of CMIP6), making it impossible to assess the effectiveness of strategies under future high-pressure scenarios. CMIP6 (Coupled Model Intercomparison Project Phase 6) refers to the sixth Coupled Model Intercomparison Project, which combines Shared Socioeconomic Pathways (SSPs) and Representative Concentration Pathways (RCPs) to simulate future climate change. In the embodiments of this application, CMIP6 can be used to simulate the landscape characteristics and water quality of the target watershed over a future period.

[0026] 5. It is difficult to balance the benefits of emission reduction with socio-economic costs: There is a lack of spatial decision-making tools to coordinate and optimize environmental emission reduction targets with regional socio-economic impacts (population, GDP).

[0027] To address the shortcomings of existing methane emission reduction measures in the background art, which suffer from poor planning and thus fail to effectively protect the environment, this application provides a river methane spatially targeted emission reduction method based on bivariate spatial clustering. This method uses watershed landscape characteristics and water quality as bivariates to determine the core regulatory factors for methane emissions. Based on the interaction mechanism between these core regulatory factors and methane emissions, a rationally planned spatially targeted emission reduction strategy is established, which can improve methane emission reduction effectiveness and promote environmental protection.

[0028] like Figure 1 As shown in the embodiments of this application, a spatially targeted methane emission reduction method for rivers based on bivariate spatial clustering is provided, comprising S101-S104.

[0029] S101. Determine the core regulatory factors for methane emissions in the target watershed.

[0030] The aforementioned core regulatory factors may include primary landscape factors and primary water quality factors, etc.

[0031] Optionally, combined Figure 1 ,like Figure 2 As shown, S101 includes S1011-S1013.

[0032] S1011. Obtain multi-source data for the target watershed.

[0033] The aforementioned multi-source data may include methane emission flux (hereinafter referred to as river CH4 flux), natural hydrological data, landscape feature data, and water quality parameters for each sub-basin within the target watershed. The river CH4 flux is the measured value of the total emission flux, including diffusion flux, bubbling flux, and plant-mediated flux, and can be mapped to the corresponding sub-basin using the Spatial Join function of ArcGIS 10.8. The aforementioned natural hydrological data may include water temperature, flow rate, and water level fluctuation amplitude. The aforementioned landscape feature data may include land use structure data (such as the proportion of cultivated land, forest land, grassland, and urban land use). The aforementioned water quality parameters may include total nitrogen, total phosphorus, chemical oxygen demand, and ammonia nitrogen concentration. Furthermore, the aforementioned multi-source data may also include socioeconomic data (such as population, GDP, and wastewater treatment capacity) and future scenario data under the CMIP6 framework (covering future land use structure data, flow and water temperature data, population and GDP data), etc., which are not limited in this embodiment.

[0034] The following is a detailed process for acquiring multi-source data for a target watershed, including steps 1.1-1.3.

[0035] Step 1.1: For socioeconomic factors (population, GDP), perform spatial interpolation using the Spatial Analyst tool in ArcGIS 10.8 → Interpolation → Inverse Distance Weighting (IDW) to generate a continuous raster surface.

[0036] Step 1.2: For water quality indicators, spatial interpolation is performed using Geostatistical Wizard → Kriging, and then the geostatistical layer is converted into a standard raster using Geostatistical Analyst tool → GA Layer To Grid.

[0037] Step 1.3: After generating the two types of interpolation results, use the Spatial Analyst tool → Zonal Analysis → Zonal Statistics by Sub-basin and display the zonal statistics as a table. Finally, link the output statistical table to the attribute table of the sub-basin layer through the Spatial Join function to complete the aggregation of multi-source data to the sub-basin.

[0038] In one embodiment, reference Figure 3Taking the Yangtze River Basin as the research object, ArcGIS 10.8 was used to divide the Yangtze River Basin into 1932 independent sub-basin units based on elevation. The minimum catchment area threshold required to establish a permanent channel is 600 km². 2 Population and GDP were spatially interpolated using the inverse distance weighting method (power parameter p=2) and then aggregated to each unit; water quality indicators were spatially interpolated using the ordinary kriging method and then aggregated to each sub-basin unit; land use data were obtained by area statistics to obtain the proportion of various land uses within each unit.

[0039] S1012. Construct a machine learning regression model to learn the interaction between landscape feature data, water quality parameters and methane emission flux in multi-source data.

[0040] In one embodiment, the input of the above-mentioned machine learning regression model may include landscape feature data (hereinafter referred to as land use structure data) and water quality parameters of each sub-basin in multiple sub-basins of the target watershed, and the output of the above-mentioned machine learning regression model may include the river CH4 flux of each sub-basin. In this way, the above-mentioned machine learning regression model can learn the interaction relationship between landscape feature data, water quality parameters and river CH4 flux during the construction process.

[0041] In another implementation, the input to the aforementioned machine learning regression model may include natural hydrological data, land use structure data, water quality parameters, and socioeconomic and anthropogenic stresses for each of the multiple sub-basins of the target watershed. The output of the aforementioned machine learning regression model may include the CH4 flux of the river in each of the aforementioned sub-basins. Thus, the aforementioned machine learning regression model can learn the interaction between natural hydrological data, land use structure data, water quality parameters, and socioeconomic and anthropogenic stresses and the CH4 flux of the river during the construction process.

[0042] The process of constructing the above machine learning regression model includes the following steps 2.1-2.2.

[0043] Step 2.1: Construct a sample dataset. The sample dataset may include multiple samples, each of which contains the input and output of a machine learning regression model.

[0044] Step 2.2: Construct a machine learning regression model using the above sample dataset.

[0045] The aforementioned machine learning algorithms can be decision trees, random forests, or gradient boosting (XGBoost, LightGBM, CatBoost), etc., and are not limited in the embodiments of this application.

[0046] Continuing with an example from S1011 above, a random forest model is constructed using river CH4 flux as the dependent variable and water temperature, flow rate, proportion of cultivated land, proportion of forest land, proportion of grassland, proportion of built-up land, TN concentration, TP concentration, COD concentration, NH3-N concentration, population density, and GDP as independent variables. The model is trained using 5-fold cross-validation and optimized via grid search. The model's coefficient of determination R0 is... 2 =0.58, RMSE=0.32.

[0047] S1013. The SHAP value is used to interpret and analyze the machine learning regression model to identify the primary landscape factor that has the greatest impact on methane emission flux in the landscape feature data and the primary water quality factor that has the greatest impact on methane emission flux in the water quality parameters, and the primary landscape factor and primary water quality factor are identified as core regulatory factors.

[0048] Specifically, the SHAP value of each driving factor in the input content of each sample in the sample dataset is calculated to quantify the marginal contribution of that driving factor to the predicted output content of that sample (positive values ​​indicate positive driving force, and negative values ​​indicate negative inhibition). Then, the absolute mean of the SHAP values ​​of all samples is taken to obtain the global importance score of each driving factor, and all driving factors are ranked according to their importance. Based on this ranking, several top-ranked factors are selected as core regulatory factors for subsequent nonlinear interaction effect analysis.

[0049] Taking one embodiment of S1011 above as an example, the SHAP value is used to interpret the model. Combined with the perspective of management operability, the proportion of cultivated land is selected as the primary landscape factor, the total nitrogen concentration (TN) as the primary water quality factor, and the proportion of forest land as the mitigation factor, which are the core regulatory factors for subsequent analysis.

[0050] S102. Identify the spatial relationships among core regulatory factors within the target watershed.

[0051] In this embodiment, the Moran index is used to describe the autocorrelation between core regulatory factors to characterize the aforementioned spatial correlation. The process first calculates the global spatial autocorrelation index between core regulatory factors, and then, after judging the global spatial autocorrelation index, calculates the local spatial autocorrelation index between core regulatory factors. The local spatial autocorrelation index is used to describe the spatial correlation between core regulatory factors within the sub-basins of the target watershed.

[0052] In one implementation, combined with Figure 2 ,like Figure 4 As shown, S102 includes S1021-S1022.

[0053] S1021. Calculate the global spatial autocorrelation index among the core regulatory factors of the target watershed.

[0054] Optionally, the above global spatial autocorrelation index satisfies the following formula (1).

[0055] Formula (1) in, This represents the global spatial autocorrelation index, also known as the bivariate global Moran index. This represents the total number of sub-basins within the target watershed; This represents the spatial weight between the i-th sub-basin and the j-th sub-basin within the target watershed. For the aforementioned spatial weight... The value of is determined when the i-th sub-basin is adjacent to the j-th sub-basin. =1, otherwise =0; This represents the primary landscape factor value of the i-th sub-basin. This represents the mean of the primary landscape factors in the target watershed; This represents the primary water quality factor value for the i-th sub-basin. This represents the mean value of the primary water quality factor in the target watershed.

[0056] It should be noted that the above This is used to determine whether the primary landscape factor and the primary water quality factor change synchronously. In order to achieve standardization, the above formula (1) uses the sum of the differences of the primary landscape factors of all sub-basins as the standard to measure the synchronicity of the two, and then to measure the global correlation between the two.

[0057] S1022. Under the condition that the bivariate global spatial autocorrelation index satisfies the first condition, calculate the local spatial autocorrelation index of each sub-basin in multiple sub-basins.

[0058] For example, the first condition mentioned above is or ,in This indicates a positive correlation between the primary landscape factor and the primary water quality factor in the target watershed. This indicates that the primary landscape factor and the primary water quality factor in the target watershed are negatively correlated. This indicates that the primary landscape factor and the primary water quality factor are uncorrelated in the target watershed. Therefore, under the first condition, the aforementioned local spatial autocorrelation index satisfies the following formula (2).

[0059] Formula (2) in, This represents the local spatial autocorrelation index of the i-th sub-basin. It indicates the spatial correlation between core regulatory factors within the i-th sub-basin of the target watershed.

[0060] It should be noted that the above The physical meaning is the local spatial correlation pattern (such as high-efficiency synergy or landscape-dominant anomaly) between the primary land use factor of the i-th sub-basin and the primary water quality factor of its neighboring sub-basins, thereby identifying local synergistic hotspots or spatial anomaly areas.

[0061] S103. Based on the numerical distribution and spatial correlation of core regulatory factors in the target watershed, determine the dominant type of methane emissions in the target watershed.

[0062] The dominant type indicates the proportion of control of methane emissions by the core regulatory factor. The dominant type of methane emissions in the target watershed can include the dominant type of methane emissions in each of the multiple sub-watersheds included in the target watershed.

[0063] In one application scenario, for each sub-basin within multiple sub-basins: When the sub-basin satisfies This indicates that the sub-basin and its neighboring sub-basins exhibit spatial synergistic clustering of high land use intensity and high water quality load. The sub-basin is an anomalous sub-basin with both high landscape pressure and high water quality load. Therefore, the dominant type of the sub-basin is determined to be the dual-high synergistic type.

[0064] When the sub-basin satisfies This indicates that the sub-basin and its neighboring sub-basins exhibit spatial synergistic clustering of low land use intensity and low water quality load. The sub-basin is an anomalous sub-basin with both landscape pressure and water quality load at low values. Therefore, the dominant type of the sub-basin is determined to be the dual-low background type.

[0065] When the sub-basin satisfies This indicates that the sub-basin is an abnormal sub-basin area with high landscape pressure but relatively low water quality load. Therefore, the dominant type of the sub-basin is determined to be the landscape-dominated abnormal type.

[0066] When the sub-basin satisfies This indicates that the sub-basin is an abnormal sub-basin area with low landscape pressure but high water quality load. Therefore, the dominant type of the sub-basin is determined to be the water quality-dominated abnormal type.

[0067] When the sub-basin satisfies If the value approaches 0 (e.g., p>0.05, meaning it fails the significance test at the 95% confidence level), it indicates that the landscape pressure and water quality load of this sub-basin are at a moderate level, and there is no obvious spatial correlation between this sub-basin and its neighboring sub-basins. Therefore, this sub-region is identified as a background normal area.

[0068] It should be noted that the above and Indicates the numerical distribution of primary landscape factors; the above and Indicates the numerical distribution of primary water quality factors.

[0069] Taking one embodiment of S1011 above as an example, the multiple sub-basins of the target watershed are divided into the following five dominant types, as shown in Table 1 below.

[0070] Table 1. Five Dominant Types of Sub-basins

[0071] S104. Based on the dominant type and the interaction effect of core regulatory factors on methane emissions, a spatially targeted emission reduction strategy for regulating methane emissions in the target watershed is generated.

[0072] In some embodiments, combined with Figure 4 ,like Figure 5 As shown, S104 includes S1041-S1043.

[0073] S1041. Two-dimensional partial dependency graph analysis was used to analyze the interaction effect of core regulatory factors on methane emission flux in the machine learning regression model, and the key thresholds of the core regulatory factors were identified.

[0074] The aforementioned key thresholds may include the structural transformation threshold of primary landscape factors and the functional transformation threshold of primary water quality factors. The aforementioned interaction effects may be synergistic, antagonistic, or have no significant interaction.

[0075] The following is a detailed identification process for the key thresholds of the aforementioned core regulatory factors, including steps 3.1-3.4.

[0076] Step 3.1: Based on the machine learning regression model obtained from S1012, extract the feature importance ranking of each driving factor (the average SHAP value can be used), select the top n core regulatory factors (such as total nitrogen, cultivated land ratio, forest land ratio, etc.), and pair all core factors to generate a total of C(n,2) factor pairs. X i , X j ), construct the factor pair analysis matrix.

[0077] Step 3.2: Calculate and output each factor pair on the machine learning regression model ( X i , X j A two-dimensional partial dependency graph of ).

[0078] Specifically, the distribution will be factorX i and X j The observed values ​​are evenly divided into N equally spaced values, forming an N*N grid matrix; and for each value combination point in the grid ( x i , x j ),make X i = x i , X j = x j The marginal mean of the sample dataset is taken for all other feature variables, and the CH4 flux prediction value is calculated by inputting it into the model point by point. Finally, the predicted value is generated. X i and X j A two-dimensional partial dependency surface is generated with the x and y axes and the CH4 flux prediction value as the response value. The two-dimensional partial dependency map is visualized and output as a contour plot.

[0079] Step 3.3: Based on the generated two-dimensional partial dependency plot, determine the interaction effect type of each pair of factors: When both factors are in the high value region, the predicted CH4 flux is significantly higher than the sum of the effects of the two factors acting alone, which is manifested as an arc-shaped distribution of contour lines convex to the origin, or a clear high-value "hot spot area" appears in the heat map, and the interaction effect is determined to be a synergistic effect; When one factor increases and the other factor also increases, the predicted CH4 flux is lower than the expected value of acting alone, which is manifested as an arc-shaped distribution of contour lines concave to the origin, or a clear low-value "cold spot area" appears in the heat map, and the interaction effect is determined to be an antagonistic effect; If the contour lines are approximately parallel to the coordinate axis, it indicates that the effects of the two factors on CH4 flux are independent of each other, and the interaction effect is determined to be no significant interaction.

[0080] Step 3.4: Identify the key thresholds for nonlinear interaction responses on the two-dimensional partial dependency graph.

[0081] The above process can be summarized as follows: Calculate the second-order partial derivative of the partial dependency surface on the two-dimensional partial dependency plot to detect abrupt gradient changes (the locations of the most drastic gradient changes), which serve as key inflection points in the interactive response; simultaneously identify the region with the densest contour lines, corresponding to the threshold interval where CH4 flux is most sensitive to factor changes; and extract one-dimensional partial dependency slices along the direction of the key factor, fitting them using piecewise linear regression or spline regression. The inflection point of the fitted curve is the response threshold along that factor direction. For each pair of core factors ( X i , X jOutput interaction effect type (synergistic effect / antagonistic effect / no significant interaction), key threshold coordinates ( X i , X j The magnitude and direction of CH4 flux change at the threshold, as well as two-dimensional partial dependence plot contour maps or heat maps.

[0082] Regarding the aforementioned key thresholds, when the primary landscape factor exceeds its structural transformation threshold, the CH4 flux shows an accelerated growth trend; when the primary water quality factor exceeds its functional transformation threshold, the increase in CH4 flux is significantly aggravated.

[0083] Understandably, the aforementioned structural transition threshold refers to the turning point where the CH4 emission mechanism or flux level undergoes a qualitative change after the driving factor reaches a certain critical value. For example, when the proportion of cultivated land in a sub-basin exceeds 50%, agricultural non-point source pollution inputs (fertilizer application, livestock and poultry breeding wastewater, etc.) increase significantly, the organic matter and nutrient load in the water body rises, providing sufficient substrate for the CH4 production process, and the CH4 flux shows a significant jump. This threshold can be used as the structural transition critical point for the driving effect of land use structure on river CH4 emissions. The aforementioned functional transition threshold refers to the inflection point where the direction or intensity of the driving factor's effect on CH4 flux changes significantly after reaching a certain critical value. For example, when the total nitrogen (TN) concentration in the water body exceeds 0.57 mg·L⁻¹. -1 At that time, the increase in nitrogen load began to significantly promote the CH4 production process in water bodies and sediments, and the correlation between CH4 flux and TN changed from weak to significant positive correlation. This threshold can be used as the critical point for the functional transformation of the driving effect of water quality parameters on river CH4 emissions.

[0084] Continuing with an example from S1011 above, a two-dimensional partial dependency plot (2D-PDP) was used to analyze the interaction effect of cultivated land proportion and TN concentration on CH4 flux. The following key thresholds were identified: Cultivated land proportion structural transformation threshold: approximately 50%. When the cultivated land proportion is below 50%, CH4 flux increases slowly and linearly with increasing cultivated land proportion; when it exceeds 50%, CH4 flux shows an accelerated growth trend, marking a structural inflection point in the watershed landscape's transformation from natural / semi-natural habitats to intensive agricultural landscapes.

[0085] S1042. Based on key thresholds and dominant types, predict the control results of the target watershed under various control scenarios.

[0086] Specifically, the aforementioned control results may include methane emission reduction rates and socioeconomic costs. The prediction process for these control results includes the following steps 4.1-4.2.

[0087] Step 4.1: Obtain the control parameters for each control scenario in multiple control scenarios.

[0088] The aforementioned various control scenarios can include baseline scenarios, single-factor control scenarios, multi-factor combined control scenarios, and basin-wide unified control scenarios. Correspondingly, the baseline scenario does not control the target basin; the single-factor control scenario only adjusts the landscape / controls the water quality of sub-basins with dominant landscape anomalies; the multi-factor combined control scenario adjusts the landscape of sub-basins with dominant landscape anomalies and controls the water quality of sub-basins with dominant water quality anomalies; and the basin-wide unified control scenario only regulates sub-basins where the core control factor value exceeds the critical threshold. The control parameters can include water quality parameter control values ​​and landscape characteristic data control values.

[0089] Taking one embodiment of S1011 above as an example, with the predicted CH4 flux in 2100 under the SSP5-8.5 scenario as a baseline (baseline scenario), the following four regulation scenarios are set as shown in Table 2 below.

[0090] Table 2. Four Regulation Scenarios and Regulation Measures

[0091] Step 4.2: For each of the multiple regulation scenarios, input the water quality parameters, landscape feature data, and regulation parameters of the target watershed into the machine learning regression model. The machine learning regression model outputs the regulation results of the target watershed under the regulation scenario.

[0092] In one application scenario, the control parameters corresponding to each control scenario in step 4.1 are input into the machine learning model constructed in S1012 to simulate and output the predicted values ​​of river CH4 emission flux in the target watershed under each control scenario. Based on the above predicted values, the expected methane emission reduction rate (hereinafter referred to as the expected CH4 emission reduction rate) of each control scenario relative to the baseline scenario is calculated as a core indicator for measuring environmental benefits. As shown in the following formula (3).

[0093] Formula (3) in, This represents the predicted CH4 flux under the baseline scenario. This represents the predicted CH4 flux under other control scenarios (such as single-factor control scenario, multi-factor combined control scenario, or unified control scenario across the entire watershed).

[0094] Furthermore, after calculating the expected CH4 emission reduction rate... Based on this, the proportion of the population and GDP directly affected by the control measures under each control scenario are estimated as agency cost indicators. Then, the environmental emission reduction benefits (emission reduction rate) and socio-economic costs (proportion of affected population, proportion of affected GDP) are calculated for each control scenario.

[0095] S1043. Based on the regulation results and dominant type, generate a spatially targeted emission reduction strategy for methane emissions in the target watershed.

[0096] In one application scenario, a multi-dimensional comprehensive comparative analysis is conducted on the control results of the target watershed under the above-mentioned control scenarios to obtain the optimal targeting strategy. This strategy represents the best balance between emission reduction efficiency and socio-economic costs. The selection principle for the optimal targeting strategy can be: under the premise of achieving acceptable emission reduction benefits, select the scheme with the lowest socio-economic costs; or under the constraint of controllable socio-economic costs, select the scheme with the greatest emission reduction benefits.

[0097] The aforementioned spatially targeted emission reduction strategy for methane emissions in the target watershed can include the dominant type of methane emissions in each sub-watershed of the target watershed, the type of control measures (such as landscape adjustment or water quality control), key thresholds, expected methane emission reduction rates, and socio-economic costs, in order to form an operable spatial management plan for CH4 emission reduction in the watershed rivers.

[0098] Continuing with an example from S1011 above, based on the evaluation results of S1043, a spatially targeted emission reduction strategy for the Yangtze River Basin is output. Strategy 3 (returning farmland to forest in high-arable land-low-nitrogen areas and controlling TN in low-arable land-high-nitrogen areas) covers approximately 66.20% of the sub-basin, affecting approximately 62.41% of the population and 67.20% of GDP, achieving an 8.03% emission reduction, representing the optimal balance between overall emission reduction effectiveness and impact scope.

[0099] In summary, the river methane spatial targeted emission reduction method based on bivariate spatial clustering provided in this application first determines the primary landscape factor and primary water quality factor of methane emissions in the core control target watershed, and then determines the dominant type of methane emissions in the target watershed based on the spatial correlation and numerical distribution of the two factors. Finally, a spatial targeted emission reduction strategy is generated based on the dominant type and the interaction effect between the primary landscape factor, primary water quality factor, and methane emissions. This process follows the actual mechanism of methane emissions within the target watershed, making the formulated spatial targeted emission reduction strategy reasonable, improving methane emission reduction effectiveness, and promoting environmental protection.

[0100] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A spatially targeted methane emission reduction method for rivers based on bivariate spatial clustering, characterized in that, include: Identify the core regulatory factors for controlling methane emissions in target river basins; The core regulatory factors include primary landscape factors and primary water quality factors. Identify the spatial correlations among core regulatory factors within the target watershed; Based on the numerical distribution of the core regulatory factors in the target watershed and the spatial correlation, the dominant type of methane emissions in the target watershed is determined; the dominant type indicates the regulatory weight of the core regulatory factors on methane emissions. Based on the dominant type and the interaction effect of the core regulatory factors on methane emissions, a spatially targeted emission reduction strategy for regulating methane emissions in the target watershed is generated.

2. The method as described in claim 1, characterized in that, The core regulatory factors for determining the target watershed methane emissions include: Acquire multi-source data for the target watershed, including methane emission flux, water quality parameters, and landscape feature data for each of the multiple sub-watersheds contained in the target watershed; Construct a machine learning regression model to learn the interaction between landscape feature data, water quality parameters and methane emission flux in the multi-source data; The SHAP value is used to interpret and analyze the machine learning regression model to identify the primary landscape factor that has the greatest impact on methane emission flux in the landscape feature data and the primary water quality factor that has the greatest impact on methane emission flux in the water quality parameters, and the primary landscape factor and the primary water quality factor are identified as the core regulatory factors.

3. The method as described in claim 1, characterized in that, The identification of spatial correlations among core regulatory factors within the target watershed includes: Calculate the global spatial autocorrelation index among the core regulatory factors of the target watershed; the global spatial autocorrelation index satisfies: in, This represents the global spatial autocorrelation index; This represents the total number of sub-basins within the target watershed; This represents the spatial weight between the i-th sub-basin and the j-th sub-basin in the target watershed; This represents the primary landscape factor value of the i-th sub-basin. This represents the mean value of the primary landscape factor in the target watershed; This represents the primary water quality factor value for the i-th sub-basin. This represents the mean value of the primary water quality factor in the target watershed; If the bivariate global spatial autocorrelation index satisfies the first condition, the local spatial autocorrelation index of each sub-watershed in the plurality of sub-watersheds is calculated; the local spatial autocorrelation index satisfies: in, This represents the local spatial autocorrelation index of the i-th sub-basin. This indicates the spatial correlation between the core regulatory factors within the i-th sub-basin of the target watershed.

4. The method as described in claim 3, characterized in that, The first condition is: or .

5. The method as described in claim 1 or 3, characterized in that, The dominant type of methane emissions in the target watershed includes the dominant type of methane emissions in each of the multiple sub-watersheds included in the target watershed; The determination of the dominant type of methane emissions in the target watershed based on the numerical distribution of core regulatory factors and the spatial correlation in the target watershed includes: for each of the multiple sub-watersheds, when the sub-watershed satisfies When the dominant type of the sub-basin is determined to be the dual-high synergistic type; when the sub-basin satisfies When the dominant type of the sub-basin is determined to be the double low background type; when the sub-basin satisfies When the sub-basin meets the following conditions, the dominant type is determined to be the landscape dominant anomaly type; when the sub-basin satisfies the following conditions, the dominant type is determined to be the landscape dominant anomaly type. At that time, the dominant type of the sub-basin is determined to be the water quality dominant anomaly type; wherein, Represents the local spatial autocorrelation index of the i-th sub-basin; This represents the primary landscape factor value of the i-th sub-basin. This represents the mean of the primary landscape factors in the target watershed. and Indicates the numerical distribution of primary landscape factors; This represents the primary water quality factor value for the i-th sub-basin. This represents the mean value of the primary water quality factor in the target watershed. and Indicates the numerical distribution of primary water quality factors.

6. The method as described in claim 2, characterized in that, The step of generating a spatially targeted emission reduction strategy for methane emissions in a target watershed based on the dominant type and the interaction effect of the core regulatory factors on methane emissions includes: Two-dimensional partial dependency graph analysis was used to analyze the interaction effect of the core regulatory factors on methane emission flux in the machine learning regression model, and the key thresholds of the core regulatory factors were identified. The key thresholds include the structural transformation threshold of the primary landscape factor and the functional transformation threshold of the primary water quality factor. Based on the key thresholds and the dominant type, predict the regulation results of the target watershed under various regulation scenarios; the regulation results include methane emission reduction rate and socio-economic costs. Based on the regulation results and the dominant type, a spatially targeted emission reduction strategy for methane emissions in the target watershed is generated.

7. The method as described in claim 6, characterized in that, The various control scenarios include a baseline scenario, a single-factor control scenario, a multi-factor combined control scenario, and a basin-wide unified control scenario. The baseline scenario does not control the target basin. The single-factor control scenario only adjusts the landscape / controls the water quality of sub-basins with dominant landscape anomalies. The multi-factor combined control scenario adjusts the landscape of sub-basins with dominant landscape anomalies and controls the water quality of sub-basins with dominant water quality anomalies. The basin-wide unified control scenario only controls sub-basins whose core control factor values ​​are greater than the key threshold.

8. The method as described in claim 6, characterized in that, The method of predicting the regulation results of the target watershed under multiple regulation scenarios based on the key threshold and the dominant type includes: Obtain the control parameters for each of the multiple control scenarios, including water quality parameter control values ​​and landscape feature data control values; For each of the various control scenarios, the water quality parameters, landscape feature data of the target watershed, and the control parameters are input into the machine learning regression model, and the machine learning regression model outputs the control results of the target watershed under the control scenario.

9. The method as described in claim 6, characterized in that, The spatially targeted emission reduction strategy for regulating methane emissions in the target watershed includes: the dominant type of methane emissions, the type of regulation measures, the key threshold, the expected methane emission reduction rate, and the socio-economic costs for each sub-watershed in the target watershed.

Images