A county carbon storage influence analysis and optimization method based on landscape pattern

Abstract

This invention discloses a method for analyzing and optimizing the impact of landscape patterns on county-level carbon storage, relating to the field of energy conservation and carbon reduction technology. It addresses the problems of weak quantitative foundations, inaccurate analysis of impact mechanisms, and a lack of differentiated optimization strategies in existing technologies for analyzing the correlation between carbon storage and landscape patterns. This solution includes constructing standardized land use data and a regionally adapted carbon density parameter library to quantify the landscape pattern index and county-level carbon storage; determining the spatial clustering characteristics of both through spatial autocorrelation analysis and hotspot identification; analyzing nonlinear interaction effects and identifying optimal thresholds for key factors using the XGBoost-SHAP model; analyzing spatial heterogeneity using the GWR model; and revealing the impact mechanisms based on the analysis results to formulate differentiated landscape pattern and land space optimization strategies. This invention also proposes a method and analysis system for enhancing carbon sequestration capacity, achieving accurate analysis of the correlation between carbon storage and landscape patterns. The optimization strategies are highly targeted and practical, providing scientific guidance for enhancing county-level carbon sequestration.

Description

Technical Field

[0001] This invention relates to the field of energy conservation and carbon reduction technology, specifically a method for analyzing and optimizing the impact of landscape patterns on county-level carbon storage. Background Technology

[0002] Against the backdrop of advancing the "dual carbon" goals, counties, as the basic units for territorial spatial governance and ecological carbon sink management, have become a research hotspot in the field of energy conservation and carbon reduction, particularly in the accurate measurement and improvement of their carbon storage. Landscape patterns, as a direct representation of regional ecosystem structure, are closely related to the spatial distribution, aggregation characteristics, and heterogeneity of carbon storage in terms of their spatiotemporal variations. Therefore, conducting carbon storage impact analysis and optimization based on landscape patterns has become an important technical approach to enhancing county-level carbon sink capacity. Currently, existing technologies have combined landscape pattern indices with carbon storage, calculating carbon storage through land use data and analyzing the impact of landscape patterns. Ecological models, spatial analysis methods, and machine learning models are also being gradually introduced into related research, providing fundamental technical support for county-level carbon sink research.

[0003] Existing technologies for analyzing the correlation between carbon storage and landscape pattern at the county level suffer from a disconnect between carbon storage measurement and landscape pattern analysis. They often only measure carbon storage or analyze landscape pattern evolution in isolation, without constructing a standardized land use classification system and carbon density parameter library suitable for the study area. Furthermore, the quantification of landscape pattern often selects only a few indices and fails to systematically characterize landscape pattern features from multiple dimensions such as patch structure, clustering characteristics, and heterogeneity. As a result, the correlation analysis between landscape pattern and carbon storage lacks an accurate quantitative basis, and the regional adaptability of carbon storage measurement results is insufficient, making it difficult to truly reflect the intrinsic relationship between the two at the county level.

[0004] Meanwhile, existing technologies have significant limitations in analyzing the intrinsic mechanisms by which landscape patterns affect carbon storage. On the one hand, they mostly employ linear analysis methods, neglecting the nonlinear characteristics and threshold effects of the landscape pattern index on carbon storage, and failing to deeply analyze the interactions between key landscape pattern factors, thus making it impossible to accurately identify the optimal threshold range of landscape pattern factors for enhancing carbon sink functions. On the other hand, they lack analysis of the spatial heterogeneity of the impact of landscape patterns on carbon storage, often using global regression models for overall analysis, failing to clarify the differences in the intensity and direction of the impact of key factors on carbon storage in different counties. This results in landscape pattern optimization strategies lacking spatial differentiation, with insufficient pertinence and practicality, making it difficult to effectively guide county-level land space optimization and carbon sink capacity enhancement.

[0005] Therefore, a method for analyzing and optimizing the impact of landscape patterns on county-level carbon storage is proposed to address the above issues. Summary of the Invention

[0006] 1. The technical problem to be solved by the present invention

[0007] The purpose of this invention is to propose a method for analyzing and optimizing the impact of landscape patterns on county-level carbon storage, in order to solve the following problems existing in the prior art: (1) Existing technologies lack a precise quantitative basis for the correlation analysis between county landscape pattern and carbon storage. There is a disconnect between carbon storage measurement and landscape pattern analysis. A standardized land use classification system and carbon density parameter library adapted to the study area have not been constructed. Moreover, the index dimension selected for landscape pattern quantification is singular, which cannot systematically depict the landscape pattern characteristics from multiple dimensions such as patch structure, clustering characteristics, and heterogeneity. It is difficult to truly reflect the intrinsic relationship between the two at the county scale.

[0008] (2) Existing technologies have limitations in analyzing the intrinsic mechanism of landscape pattern's influence on carbon storage. On the one hand, they ignore the nonlinear characteristics, threshold effect, and interaction between key factors of the influence of landscape pattern index on carbon storage, and cannot accurately identify the optimal landscape pattern factor threshold for improving carbon sink function. On the other hand, they lack spatial heterogeneity analysis and only use global analysis methods. They do not clarify the differences in the intensity and direction of the influence of key factors on carbon storage in different counties, resulting in the lack of spatially differentiated design of the optimization strategy and insufficient pertinence and operability.

[0009] 2. Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a method for analyzing and optimizing the impact of landscape patterns on county-level carbon storage, comprising the following steps: S1. Research Data Construction: Obtain basic land use data from multiple periods in the county of the study area and perform standardized classification processing; integrate relevant research results on regional carbon density; and establish a carbon density parameter library for land use types that is suitable for the study area. S2. Landscape and Carbon Storage Quantification: Landscape pattern indices that characterize landscape structure, clustering characteristics and heterogeneity are selected and calculated. An ecological model is used in combination with land use classification data and a carbon density parameter library to calculate the carbon storage of the county as a whole and for each land use type in the study area. S3. Spatial Feature Analysis: Spatial autocorrelation analysis is used to determine the spatial clustering characteristics of landscape pattern index and carbon storage, and hotspot identification is used to accurately locate the high and low value clustering areas of the two. S4. Analysis of Nonlinear Interaction Effects: Construct a machine learning-interpretable model to quantify the relative importance of the impact of each landscape pattern index on carbon storage, identify the threshold effect of the landscape pattern index on carbon storage, and analyze the nonlinear interaction between key landscape pattern factors on carbon storage. S5. Spatial Heterogeneity Analysis: A spatial weighted regression model was used to analyze the spatial differentiation characteristics of the impact of key landscape pattern indices on carbon storage, and to clarify the differences in the intensity and direction of the impact of each key factor on carbon storage in different counties of the study area. S6. Derivation of Influence Mechanism and Optimization Strategy: Combining the results of nonlinear interaction effect and spatial heterogeneity analysis, this study reveals the overall influence mechanism of county-level landscape pattern on carbon storage. Based on the optimal threshold range and spatial differentiation law of key factors, differentiated landscape pattern and land space optimization strategies for the study area are formulated.

[0010] Preferably, the land use basic data in S1 is multi-period remote sensing monitoring data, and the standardized classification processing is to reclassify the original land use data into six primary land categories: cultivated land, forest land, grassland, water area, construction land and unused land; the carbon density parameter library contains the carbon density of aboveground, underground, soil and dead organic matter for each land use type.

[0011] Preferably, the landscape pattern indices in S2 include the maximum patch index, edge density, patch number, patch density, aggregation index, dispersion index, Shannon diversity index, Shannon evenness index, and sprawl index, which are calculated using landscape pattern analysis software; the ecological model is the carbon storage and sequestration module of the InVEST model, and the carbon storage is obtained by weighted summation of the total carbon density of each land use type and its corresponding area, using the following formula:

[0012]

[0013] In the formula, Land use type Total carbon density, Land use type Aboveground carbon density, Land use type underground carbon density, Land use type Soil carbon density, Land use type Carbon density of dead organic matter; This represents the total carbon reserves; The number of land use types; Land use type The total area.

[0014] Preferably, the spatial autocorrelation analysis method in S3 is the global Moran's index (Moran's I) analysis method, and the hotspot identification method is the Getis-OrdGi analysis method. )Law.

[0015] Preferably, the machine learning-interpretability model in S4 is the XGBoost-SHAP model, which captures nonlinear trends and identifies threshold effects by combining SHAP partial dependency graphs with locally weighted scatter plot smoothing curves, and deconstructs the interaction effects of key landscape pattern factors through two-dimensional partial dependency graphs; the key landscape pattern factors are the number of patches, the spread index, and the maximum patch index, and the optimal threshold range for their impact on carbon sink function improvement is: the number of patches is between 200 and 500, the spread index is between 70% and 75%, and the maximum patch index is >70%.

[0016] Preferably, the spatially weighted regression model described in S5 is the geographically weighted regression (GWR) model, and its calculation formula is as follows:

[0017] In the formula, In position Carbon storage value at that location, For the regression coefficient term, , The first i The research unit k The regression coefficients and observed values ​​of the independent variables. The random error term is represented by the natural discontinuity method, which is used to classify the local regression coefficients of the model to characterize the spatial heterogeneity of the influence of key factors on carbon storage.

[0018] A method for enhancing county-level carbon sequestration capacity based on landscape pattern optimization is proposed. This method determines the threshold range and spatial differentiation characteristics of key landscape pattern factors for enhancing the carbon sequestration function of counties in the study area. Differentiated landscape pattern optimization schemes are formulated for different counties in the study area to regulate the key landscape pattern factors to the optimal threshold range, thereby enhancing the county-level carbon sequestration capacity.

[0019] A landscape pattern and carbon storage analysis system, the system comprising: Data processing module: used to acquire basic land use data, standardize and classify it, and construct a carbon density parameter library; Quantitative calculation module: used to calculate the landscape pattern index and the carbon storage of the county as a whole and for each land use type in the study area; Spatial Feature Analysis Module: Used to perform spatial autocorrelation analysis and hotspot identification, outputting the spatial clustering characteristics and distribution of clustering areas of landscape pattern index and carbon storage; Nonlinear Analysis Module: Used to build machine learning-interpretable models, outputting the relative importance of landscape pattern indices to carbon storage, threshold effects, and nonlinear interaction effects of key factors; Spatial heterogeneity analysis module: used to construct a spatial weighted regression model and output the spatial differentiation characteristics of the impact of key landscape pattern indices on carbon storage; Strategy derivation module: This module combines the results of nonlinear analysis and spatial heterogeneity analysis to output the impact mechanism of landscape pattern on carbon storage and the differentiated landscape pattern and land space optimization strategies for the study area.

[0020] Compared with existing technologies, the present invention provides a method for analyzing and optimizing the impact of landscape patterns on county-level carbon storage, which has the following advantages: (1) Constructing a standardized data system to solidify the accurate quantitative foundation for the correlation analysis between carbon storage and landscape pattern; This scheme innovatively builds an integrated data processing system adapted to the county-level study area, standardizes and reclassifies multi-period remote sensing land use data into six primary land categories, and integrates regional research results to establish a dedicated parameter library containing four types of carbon density: aboveground, underground, soil, and dead organic matter, thus overcoming the problem of the disconnect between existing carbon storage measurement and landscape pattern analysis. At the same time, nine core landscape pattern indices covering patch structure, clustering characteristics, and landscape heterogeneity are selected to achieve multi-dimensional systematic quantification of the county-level landscape pattern, so that the correlation analysis between landscape pattern and carbon storage has accurate data support that fits the regional characteristics of the study area, greatly improving the authenticity and reliability of the analysis results.

[0021] (2) Integrating a multi-model analytical system to accurately reveal the nonlinear mechanism of landscape pattern on carbon storage; This scheme innovatively constructs the XGBoost-SHAP machine learning-interpretability model, which breaks through the limitations of existing technologies that mostly use linear analysis. It can not only quantify the relative importance of the impact of each landscape pattern index on carbon storage, but also accurately capture the threshold effect of landscape pattern index on carbon storage through SHAP partial dependence plot and two-dimensional partial dependence plot, and deeply analyze the nonlinear interaction between key factors. At the same time, it clarifies for the first time the optimal threshold range for the three core factors of patch number, sprawl index and maximum patch index to enhance carbon sink function, filling the technical gap that existing technologies cannot accurately identify the optimal parameters of carbon sink-oriented landscape patterns.

[0022] (3) Introducing spatial heterogeneity analysis to achieve refined spatial analysis of the impact of landscape patterns on carbon storage; This scheme innovatively adopts the geographically weighted regression (GWR) model to conduct spatial heterogeneity analysis, which differs from the existing technology that only uses a global regression model for overall analysis. It can accurately characterize the differences in the intensity and direction of the impact of key landscape pattern indices on carbon storage in different counties, and clearly present the spatial differentiation characteristics of the impact of each factor by classifying the local regression coefficients of the model through the natural discontinuity method. This design breaks through the problem of the fuzzy analysis of the spatial characteristics of the impact of landscape patterns on carbon storage in the existing technology, and makes the county-level impact mechanism analysis move from "global average" to "spatial refinement".

[0023] (4) Deriving differentiated optimization strategies to realize the practical implementation of carbon sink-oriented landscape pattern and land space optimization; This scheme innovatively combines the dual results of nonlinear interaction effect and spatial heterogeneity analysis. First, it systematically reveals the overall impact mechanism of county-level landscape pattern on carbon storage. Then, based on the optimal threshold range and spatial differentiation law of key factors, it customizes differentiated landscape pattern and land space optimization strategies for different counties in the study area. This design overcomes the shortcomings of existing technical optimization strategies, which lack spatial differentiation, pertinence and practicality. It directly transforms the technical analysis results into a feasible county-level carbon sink enhancement scheme, realizing a technical closed loop from "mechanism analysis" to "strategy implementation", and providing scientific and specific practical guidance for improving county-level carbon sink capacity. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the research area in an embodiment of the present invention; Figure 2 This is a schematic diagram of the spatial distribution pattern of the county-level landscape pattern index from 2000 to 2020 in Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the hotspot analysis of the county-level landscape pattern index from 2000 to 2020 in Embodiment 2 of the present invention; Figure 4 This is a schematic diagram illustrating the spatial distribution pattern and hotspot analysis of county-level carbon storage from 2000 to 2020 in Embodiment 2 of the present invention. Figure 5 This is a schematic diagram of SHAP digest parsing in an embodiment of the present invention; Figure 6 This is a schematic diagram of a two-dimensional partial dependence graph of key landscape pattern indicators in an embodiment of the present invention; Figure 7 This is a schematic diagram of the spatial distribution of the regression coefficients of key landscape pattern indices on carbon storage from 2000 to 2020 in Embodiment 2 of the present invention. Figure 8 This diagram illustrates a process flow for analyzing and optimizing the impact of landscape patterns on county-level carbon storage. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0026] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0027] Example 1, please refer to Figures 1 to 7As shown: To address the problems mentioned in the technical solutions, this application provides a method for analyzing and optimizing the impact of landscape patterns on county-level carbon storage, comprising the following steps: S1. Research Data Construction: Obtain basic land use data from multiple periods in the county of the study area and perform standardized classification processing; integrate relevant research results on regional carbon density; and establish a carbon density parameter library for land use types that is suitable for the study area. S2. Landscape and Carbon Storage Quantification: Landscape pattern indices that characterize landscape structure, clustering characteristics and heterogeneity are selected and calculated. An ecological model is used in combination with land use classification data and a carbon density parameter library to calculate the carbon storage of the county as a whole and for each land use type in the study area. S3. Spatial Feature Analysis: Spatial autocorrelation analysis is used to determine the spatial clustering characteristics of landscape pattern index and carbon storage, and hotspot identification is used to accurately locate the high and low value clustering areas of the two. S4. Analysis of Nonlinear Interaction Effects: Construct a machine learning-interpretable model to quantify the relative importance of the impact of each landscape pattern index on carbon storage, identify the threshold effect of the landscape pattern index on carbon storage, and analyze the nonlinear interaction between key landscape pattern factors on carbon storage. S5. Spatial Heterogeneity Analysis: A spatial weighted regression model was used to analyze the spatial differentiation characteristics of the impact of key landscape pattern indices on carbon storage, and to clarify the differences in the intensity and direction of the impact of each key factor on carbon storage in different counties of the study area. S6. Derivation of Influence Mechanism and Optimization Strategy: Combining the results of nonlinear interaction effect and spatial heterogeneity analysis, this study reveals the overall influence mechanism of county-level landscape pattern on carbon storage. Based on the optimal threshold range and spatial differentiation law of key factors, differentiated landscape pattern and land space optimization strategies for the study area are formulated.

[0028] The land use basic data in S1 consists of multi-period remote sensing monitoring data. The standardized classification process reclassifies the original land use data into six primary land categories: cultivated land, forest land, grassland, water area, construction land, and unused land. The carbon density parameter library contains the carbon density of aboveground, underground, soil, and dead organic matter for each land use type.

[0029] Landscape pattern indices in S2 include the largest patch index, edge density, patch number, patch density, aggregation index, dispersion index, Shannon diversity index, Shannon evenness index, and sprawl index, calculated using landscape pattern analysis software. The ecological model is the carbon storage and sequestration module of the InVEST model. Carbon storage is obtained by weighted summation of the total carbon density of each land use type and its corresponding area. The calculation formula is as follows:

[0030]

[0031] In the formula, Land use type Total carbon density, Land use type Aboveground carbon density, Land use type underground carbon density, Land use type Soil carbon density, Land use type Carbon density of dead organic matter; This represents the total carbon reserves; The number of land use types; Land use type The total area.

[0032] In S3, the spatial autocorrelation analysis method is the global Moran's I analysis method, and the hotspot identification method is the Getis-OrdGi analysis method. )Law.

[0033] In S4, the machine learning-interpretability model is the XGBoost-SHAP model. It captures nonlinear trends and identifies threshold effects by combining SHAP partial dependency graphs with locally weighted scatter plots to smooth curves. It deconstructs the interaction effects of key landscape pattern factors through two-dimensional partial dependency graphs. The key landscape pattern factors are the number of patches, the spread index, and the maximum patch index. The optimal threshold range for their impact on carbon sink function improvement is: the number of patches is between 200 and 500, the spread index is between 70% and 75%, and the maximum patch index is >70%.

[0034] The spatial weighted regression model in S5 is the geographically weighted regression (GWR) model, and its calculation formula is as follows:

[0035] In the formula, In position Carbon storage value at that location, For the regression coefficient term, , The first i The research unit k The regression coefficients and observed values ​​of the independent variables. The random error term is represented by the natural discontinuity method, which is used to classify the local regression coefficients of the model to characterize the spatial heterogeneity of the influence of key factors on carbon storage.

[0036] Furthermore, a method for enhancing county-level carbon sequestration capacity based on landscape pattern optimization is proposed. The threshold range and spatial differentiation characteristics of key landscape pattern factors for enhancing county-level carbon sequestration function in the study area are determined. Differentiated landscape pattern optimization schemes are formulated for different counties in the study area to regulate key landscape pattern factors to the optimal threshold range, thereby enhancing county-level carbon sequestration capacity.

[0037] This solution also proposes a landscape pattern and carbon storage analysis system, which includes: Data processing module: used to acquire basic land use data, standardize and classify it, and construct a carbon density parameter library; Quantitative calculation module: used to calculate the landscape pattern index and the carbon storage of the county as a whole and for each land use type in the study area; Spatial Feature Analysis Module: Used to perform spatial autocorrelation analysis and hotspot identification, outputting the spatial clustering characteristics and distribution of clustering areas of landscape pattern index and carbon storage; Nonlinear Analysis Module: Used to build machine learning-interpretable models, outputting the relative importance of landscape pattern indices to carbon storage, threshold effects, and nonlinear interaction effects of key factors; Spatial heterogeneity analysis module: used to construct a spatial weighted regression model and output the spatial differentiation characteristics of the impact of key landscape pattern indices on carbon storage; Strategy derivation module: This module combines the results of nonlinear analysis and spatial heterogeneity analysis to output the impact mechanism of landscape pattern on carbon storage and the differentiated landscape pattern and land space optimization strategies for the study area.

[0038] Example 2: Based on Example 1, but with a difference, the present invention provides a method for analyzing and optimizing the impact of landscape patterns on county-level carbon storage, using specific examples and accompanying drawings. This example uses counties in Hebei Province as the study area to conduct research on the impact of landscape patterns on carbon storage, verifying the feasibility and practicality of the method. The specific details are as follows.

[0039] I. Data Acquisition and Processing (a) Land use data This embodiment selects 30m resolution land use remote sensing monitoring data from five periods (2000, 2005, 2010, 2015, and 2020), sourced from the Resource and Environmental Science and Data Center of the Chinese Academy of Sciences. This dataset, generated from Landsat remote sensing images through visual interpretation, is a core product covering multi-temporal land use / cover change across China. The original data includes 25 secondary land categories, such as paddy fields, dry land, and forest land. Based on the standardized classification requirements of this invention, the five periods of land use remote sensing monitoring data are reclassified into six primary land categories: cultivated land, forest land, grassland, water area, construction land, and unused land. The correspondence between these land categories is detailed in Table 1, the land category classification system table.

[0040] Table 1 shows the land classification system.

[0041] (II) Construction of carbon density parameter library To construct a land use type carbon density parameter database adapted to the regional characteristics of Hebei Province, this scheme integrates existing research results on carbon density in Hebei Province and other areas with similar climate zones and ecosystems. By extracting measured carbon density data from different land use types from existing data and calculating the average, the carbon density values ​​for four categories—aboveground, belowground, soil, and dead organic matter—are determined for the six primary land categories. Finally, a land use type carbon density parameter database for Hebei Province is established. Specific carbon density values ​​are detailed in Table 2, "Carbon Density Values ​​of the Four Major Carbon Pools in Hebei Province / t". hm -2 Among them, the overall carbon density level of forest land is significantly higher than that of other land types, there is no carbon density of dead organic matter in unused land, and the carbon density of water bodies is the lowest among all types. The regional adaptability of the parameter database lays the foundation for accurate carbon storage calculation in the future.

[0042] Table 2 shows the carbon density values ​​(t•hm) of the four major carbon pools in Hebei Province. -2

[0043] (III) Calculation of Landscape Pattern Index Based on the professional landscape pattern analysis software Fragstats 4.2, and in accordance with the method of this invention, nine landscape level indices were selected for calculation: Maximum Patch Index (LPI), Edge Density (ED), Number of Patches (NP), Patch Density (PD), Aggregation Index (AI), Dispersion Index (DIVISION), Shannon Diversity Index (SHDI), Shannon Evenness Index (SHEI), and Spread Index (CONTAG).

[0044] This index system can systematically depict the spatiotemporal evolution of county-level landscape patterns in Hebei Province from 2000 to 2020 from dimensions such as patch structure, spatial clustering and fragmentation, and landscape heterogeneity. The ecological meaning of each index is detailed in Table 3, Landscape Pattern Index Details Table, which provides a quantitative basis for subsequent correlation analysis between landscape patterns and carbon storage.

[0045] Table 3. Details of Landscape Pattern Index

[0046] II. Analysis of the Spatial Characteristics of Carbon Storage and Landscape Pattern (I) Spatiotemporal characteristics and spatial clustering of landscape pattern index From 2000 to 2020, the landscape pattern index of counties in Hebei Province showed a significant spatiotemporal evolution pattern: the number of counties with high and high values ​​of edge density (ED), patch density (PD), and dispersion index (DIVISION) increased by approximately 16.98%, 100%, and 56.72%, respectively; the number of counties with high values ​​of maximum patch index (LPI) and aggregation index (AI) decreased by approximately 52.78% and 88.33%, respectively; and the number of counties with low values ​​of Shannon diversity index (SHDI) and number of patches (NP) decreased by approximately 51.02% and 16.67%, respectively. Overall, this reflects that the degree of landscape fragmentation in counties of Hebei Province continued to increase during the study period, and the landscape dominance of dominant patches gradually weakened.

[0047] The global Moran's I was calculated for the landscape pattern index for each year from 2000 to 2020. The results all passed the significance test with p < 0.01, and the z score was much greater than 2.58. For specific values, please refer to Table 4, Moran's I for the landscape pattern of Hebei Province from 2000 to 2020. This shows that the landscape pattern index of counties in Hebei Province has a significant positive spatial autocorrelation.

[0048] Through hot and cold spot analysis (Getis-OrdGi) Further precise identification of spatial clustering areas of landscape pattern indices revealed that: hotspots for ED, SHDI, SHEI, and DIVISION are concentrated in Zhangjiakou, Chengde, and counties along the Taihang Mountains; hotspots for NP are concentrated in Zhangjiakou and Chengde counties; hotspots for LPI, AI, and CONTAG are distributed in Shijiazhuang, Handan, Xingtai, Hengshui, and southeastern Baoding counties; and hotspots for PD have gradually evolved from scattered distribution in 2000 to concentrated distribution in parts of Chengde, Qinhuangdao, and Tangshan counties in 2020. The spatial clustering characteristics of landscape patterns are highly correlated with regional landform types and ecological background conditions.

[0049] The carbon storage and sequestration module using the InVEST model was used to calculate the carbon storage capacity of Hebei Province. This module maps aboveground carbon density, belowground carbon density, soil carbon density, and dead organic matter carbon density of different land types to different land use type rasters based on land use type data and a carbon density database. [20,21] First, the carbon density per unit area for each land type is calculated separately. Then, combining the areas of each land type, a weighted sum is used to obtain the total carbon storage for the region. The formula for calculating the total carbon storage is as follows:

[0050] (1) In the formula, Land use type Total carbon density, Land use type Aboveground carbon density, Land use type underground carbon density, Land use type Soil carbon density, Land use type Carbon density of dead organic matter; This represents the total carbon reserves; The number of land use types; Land use type The total area.

[0051] Furthermore, to study the spatial characteristics of county-level landscape pattern indices and carbon storage in Hebei Province, the global Morans' I index was first used to determine its overall spatial autocorrelation and clustering pattern. The formula for calculating the global Morans' I index is as follows: (2) In the formula, and They are respectively counties i and j A certain landscape pattern index or carbon storage; and These are the mean and variance of a certain landscape pattern index or carbon storage, respectively. The number of counties; Let be the spatial weight matrix, if the county i and j Adjacent, then = 1. Conversely = 0.

[0052] Based on this, hot and cold spot analysis (Getis-OrdGi) is used. It accurately identifies significant high-value clusters (hotspots) and low-value clusters (cold spots) within the study area.

[23] The calculation formula is as follows: (3) In the formula, For the county i The aggregation index, For the county j A certain landscape pattern index or carbon storage value, Spatial weights, The number of counties, and These are the mean and standard deviation of a certain landscape pattern index or carbon storage, respectively.

[0053] To quantify the relative importance and nonlinear relationship of different landscape pattern indices on carbon storage, this study uses the XGBoost machine learning algorithm and combines it with the SHAP (SHapley Additive ex Planation) value interpretation framework for analysis.

[0054] XGBoost is a highly efficient gradient boosting ensemble learning method that iteratively constructs multiple decision trees and integrates their predictions. It effectively captures complex nonlinear relationships and interactions between variables and has significant advantages in identifying key driving factors. Compared to traditional models such as GBDT and Random Forest, XGBoost typically exhibits higher prediction accuracy on small to medium-sized datasets. Therefore, this model has been widely applied to research on driving mechanisms in fields such as ecological environment management and urban sustainable development.

[0055] To further enhance the interpretability of the model and delve into the mechanisms of action of various landscape pattern indices, the SHAP method is introduced to interpret the model output. Based on the Shapley value in game theory, SHAP assigns a contribution value to each feature to quantify its impact on individual prediction results. It not only supports global importance ranking but also reveals the specific direction and intensity of each variable's influence in local predictions. Building upon this, the SHAP partial dependency graph is further utilized, combined with a locally weighted scatter plot smoothing (LOWESS) curve, to capture the nonlinear trend between landscape pattern indices and carbon storage. Furthermore, the threshold effect is identified by locating the intersection of the LOWESS curve and the SHAP=0 curve.

[0056] To understand the spatial differences in the intensity of landscape pattern's influence on carbon storage, the GWR model was further used to study the spatial heterogeneity of various landscape pattern indices on carbon storage, clarifying the differences and causes of the impact of different factors on carbon storage in different counties. The calculation formula is as follows: (4) In the formula, In position Carbon storage value at that location, For the regression coefficient term, , The first i The research unit k The regression coefficients and observed values ​​of the independent variables. This is the random error term.

[0057] Table 4 shows the Moran index of landscape pattern in Hebei Province from 2000 to 2020;

[0058] (II) Spatiotemporal characteristics and spatial clustering of carbon reserves Based on the InVEST model's carbon storage and sequestration module, combined with land use classification data and a carbon density parameter library, the carbon storage of the entire county area and various land use types in Hebei Province from 2000 to 2020 was calculated. Specific values ​​are detailed in Table 5: Carbon Storage of Hebei Province from 2000 to 2020 (10 6 The results show that the total carbon storage in counties of Hebei Province has been decreasing year by year, from 3605.69 × 10⁻⁶ in 2000 to 2020. 6 t decreased to 3578.94×10 6 t, a cumulative reduction of approximately 26.75 × 10 6 t.

[0059] Table 5 shows the carbon storage (10⁻⁶ tons) in Hebei Province from 2000 to 2020. 6 t)

[0060] Looking at the changes in carbon storage across different land use types, forest land, water areas, and construction land all showed an overall upward trend, with construction land experiencing the largest increase, approximately 85.78 × 10⁻⁶. 6 Forest carbon storage increased by approximately 27.31 × 10⁻⁶ tons. 6 The carbon storage in the water area increased by approximately 2.26 × 10⁻⁶ t. 6 The carbon storage of arable land, grassland, and unused land showed an overall downward trend, with the most significant decrease in arable land carbon storage, from 1355.91 × 10⁻⁶. 6 t decreased to 1258.13×10 6 t, a cumulative decrease of approximately 97.78 × 10 6 The characteristics of changes in land-use carbon storage are highly correlated with the land use structure changes during the study period, which showed the largest area of ​​construction land transferred in and the largest area of ​​cultivated land transferred out.

[0061] The spatial distribution pattern of carbon reserves in Hebei Province's counties remains generally stable, exhibiting a significant characteristic of being higher in the west and lower in the east, and higher in the north and lower in the south. Using the natural discontinuity method, carbon reserves are divided into five levels: low, relatively low, medium, relatively high, and high. Counties with high carbon reserves are stably distributed in areas such as Weichang Manchu and Mongolian Autonomous County, Fengning Manchu Autonomous County, and Longhua County in Chengde City, northern Hebei Province. In 2005, Chicheng County in Zhangjiakou City was newly added as a county with high carbon reserves. Counties with slightly higher carbon reserves are mainly distributed in Guyuan County and Zhangbei County in Zhangjiakou City, the southeastern counties of Chengde City, and Zhuolu County, Wei County, and Laiyuan County in the Taihang Mountains of western Hebei Province.

[0062] The global Moran's index calculation results show that the Moran's I index of carbon storage in Hebei Province counties from 2000 to 2020 all passed the significance test with p < 0.01, and the z score was much greater than 2.58. For specific values, please refer to Table 6 Moran's I index of carbon storage in Hebei Province counties from 2000 to 2020, indicating that carbon storage has a significant positive spatial autocorrelation.

[0063] The analysis of hot and cold spots shows that the carbon storage hot spots are stably and concentrated in the Zhangjiakou-Chengde ecological conservation area in northern Hebei Province, while the cold spots are mainly distributed in the central and southern plains of Hebei Province, especially in the counties and districts surrounding the central cities of Shijiazhuang, Baoding, Cangzhou, Handan and Xingtai. The spatial clustering characteristics of carbon storage are highly consistent with the regional ecological background, land use structure and intensity of human activities.

[0064] Table 6 shows the Moran Index of carbon storage in counties of Hebei Province from 2000 to 2020.

[0065] III. Analysis of the Influence Mechanism of Landscape Pattern on Carbon Storage (I) Identification of Key Influencing Factors – XGBoost-SHAP Model Analysis Using the XGBoost-SHAP machine learning-interpretable model proposed in this invention, the relative importance of each landscape pattern index on carbon storage is quantified, and key influencing factors and their directions of influence are accurately identified. See details for the results. Figure 5 .

[0066] The results show that the dominant landscape pattern factors and their influencing directions of carbon storage varied in different years from 2000 to 2020. However, the number of patches (NP), the contagion index (CONTAG), and the maximum patch index (LPI) were key influencing factors throughout the study period and are the core landscape pattern indicators driving changes in carbon storage in counties of Hebei Province. In 2000, carbon reserves were mainly affected by NP, CONTAG, and LPI, with CONTAG having a positive impact and LPI having a negative impact. In 2005, the dominant factors shifted to NP, ED, and CONTAG. The positive promoting effect of CONTAG weakened, ED had a negative impact, and the negative effect of LPI strengthened. In 2010, LPI, NP, and CONTAG once again became key factors, with CONTAG maintaining a positive contribution and LPI's impact on carbon storage trending towards a positive direction. In 2015, the dominant factors were LPI, CONTAG, and DIVISION. Both LPI and CONTAG had a positive impact, and the positive effect of LPI was further enhanced. In 2020, CONTAG, ED, and NP ranked first, and the distribution of SHAP values ​​corresponding to high and low values ​​of each factor tended to be mixed, indicating that their impact on carbon storage is regulated by nonlinear response and multiple interactions, and the explanatory power of a single indicator has decreased.

[0067] (II) Analysis of Nonlinear Interaction and Threshold Effect Based on SHAP partial dependence graphs and two-dimensional partial dependence graphs, the nonlinear interaction effects of three key factors—NP, CONTAG, and LPI—on carbon storage were analyzed, and the optimal threshold range for enhancing carbon sink function was accurately identified. For details, see [link to results]. Figure 6 There is a significant nonlinear interaction among the three, and the carbon sequestration function reaches its optimum under a specific combination of thresholds: NP and CONTAG interaction: When NP is between 200 and 500 and CONTAG is above 70%, the SHAP value remains positive overall. Under the background of high landscape connectivity, a moderate number of patches can form structurally complete and functionally stable carbon sink units. When NP is greater than 800 and CONTAG is between 55% and 65%, the increase in the number of patches has a limited effect on improving carbon storage.

[0068] LPI and NP interaction: When LPI>70% and NP<800, SHAP value is at a high positive value, and the landscape structure with "moderate number of patches and dominant patches" has higher carbon sequestration efficiency; when LPI is below 40% and NP is above 800, SHAP value drops significantly, and the surge in the number of patches accompanied by the decrease in the dominance of core patches will exacerbate landscape fragmentation and weaken carbon sequestration capacity.

[0069] CONTAG and LPI Interaction: When CONTAG is between 65% and 75% and LPI > 60%, the SHAP value is generally high and positive. A landscape pattern in which moderate spatial connectivity and dominant ecological patches coexist is conducive to building a functionally stable carbon sink system. When CONTAG is less than 60% and LPI is less than 40%, simply increasing the degree of patch connectivity will weaken the carbon sink function.

[0070] Based on the combined results of the multifactor interaction analysis, NP ranged from 200 to 500, CONTAG from 70% to 75%, and LPI was greater than 70%. The optimal threshold range for key landscape pattern factors for enhancing carbon sequestration function in counties of Hebei Province.

[0071] (III) Spatial Heterogeneity Analysis – Geographically Weighted Regression (GWR) Model Analysis Using the geographically weighted regression (GWR) model proposed in this invention, the spatial differentiation characteristics of the impact of three key factors—NP, CONTAG, and LPI—on carbon storage were analyzed. The natural discontinuity method was used to classify the local regression coefficients of the model into seven levels. For detailed results, see [link to results]. Figure 7 .

[0072] The results show that the intensity and direction of the impact of the three key factors on carbon storage exhibit significant spatial heterogeneity, which is highly consistent with the regional geographical background, development type, and ecological function positioning. The number of patches (NP) has a spatial impact on carbon storage that is higher in the north and lower in the south. In ecological conservation counties such as Zhangjiakou, Chengde, Qinhuangdao, and northern Tangshan, the NP regression coefficients are mostly positive, indicating that a moderate increase in the number of patches helps improve regional carbon sequestration capacity. In urban-agricultural counties such as Shijiazhuang, Baoding, Cangzhou, Hengshui, Handan, Xingtai, and southern Tangshan, the NP regression coefficients are generally negative. The increase in the number of patches stems from the expansion of construction land or the fragmentation of arable land, which easily leads to ecological degradation and inhibits carbon storage accumulation. From 2000 to 2020, the proportion of counties with a positive impact from NP increased slightly from 22.6% to 23.8%.

[0073] The CONTAG (Contagion and Affected Areas) index shows a spatial pattern of stronger impact on carbon storage in the south and weaker impact in the north. In southern urban and agricultural counties such as Shijiazhuang, Baoding, Cangzhou, Hengshui, Xingtai, Handan, and Langfang, the regression coefficients are mostly positive, indicating that improved landscape connectivity helps form contiguous ecological corridors and stable vegetation cover, promoting carbon storage accumulation. In northern ecological conservation areas such as Zhangjiakou, Chengde, Qinhuangdao, and Tangshan, the regression coefficients are mostly negative, suggesting that large-scale, homogeneous improvements in landscape connectivity reduce the structural complexity of the original ecosystems and inhibit carbon storage potential. From 2000 to 2020, the proportion of counties with a positive CONTAG impact increased from 56.0% to 58.9%, and the scope of the positive impact slightly expanded.

[0074] The Maximum Patch Index (LPI) also exhibits a spatial pattern of stronger impact on carbon storage in the south and weaker impact in the north. In southern urban and agricultural counties, the LPI regression coefficient is mostly positive, indicating that the presence of large, dominant ecological patches is an important carrier of regional carbon sequestration. In northern ecological conservation counties, the LPI regression coefficient is mostly negative, suggesting that the formation of single large patches reduces the heterogeneity of the natural landscape and weakens the carbon sequestration function. From 2000 to 2020, the proportion of counties with a positive impact of LPI decreased from 61.9% to 58.3%, indicating a slight contraction in the scope of the positive impact.

[0075] IV. Implementation Conclusions and Practical Implications; This embodiment takes counties in Hebei Province as the research object. Through the full-process application of the method of this invention, it systematically reveals the spatiotemporal correlation characteristics and influencing mechanisms of landscape patterns and carbon storage at the county scale. The main conclusions are as follows: From 2000 to 2020, the landscape pattern of counties in Hebei Province showed the characteristics of increased fragmentation and weakened dominance of dominant patches. Moreover, the landscape pattern indices showed significant positive spatial autocorrelation, and the clustering characteristics were highly consistent with the regional geomorphology and ecological background. The carbon storage in counties of Hebei Province shows a decreasing trend year by year. The spatial distribution is high in the west and low in the east, and high in the north and low in the south. The carbon storage hotspots are concentrated in the northern ecological conservation area, while the coldspots are distributed in the central and southern urban and agricultural areas. The changes in land use carbon storage are highly correlated with the transformation of land use structure. The number of patches (NP), the spread index (CONTAG), and the maximum patch index (LPI) are the core landscape pattern factors driving changes in carbon storage in counties of Hebei Province. The three factors have significant nonlinear interaction effects and threshold effects on carbon storage. The optimal thresholds for improving carbon sink function are NP 200~500, CONTAG 70%~75%, and LPI >70%. The impact of the three key factors on carbon storage exhibits significant spatial heterogeneity: the impact of NP is higher in the north and lower in the south, while the impact of CONTAG and LPI is stronger in the south and weaker in the north. These differences stem from variations in regional ecological function positioning, intensity of human activities, and land use types.

[0076] This embodiment verifies the feasibility, practicality, and accuracy of the landscape pattern-based county-level carbon storage impact analysis and optimization method proposed in this invention in practical research. This method can effectively reveal the impact mechanism of county-level landscape pattern on carbon storage and provide scientific and specific strategic guidance for the optimization of county-level landscape pattern and land space. It can provide methodological reference and practical guidance for other counties across the country to carry out carbon sink-oriented landscape pattern optimization.

[0077] Please refer to the above work process. Figures 1 to 8 .

[0078] It should be noted that the term "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0079] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing and optimizing the impact of landscape patterns on county-level carbon storage, characterized in that, Includes the following steps: S1. Research Data Construction: Obtain basic land use data from multiple periods in the county of the study area and perform standardized classification processing; integrate relevant research results on regional carbon density; and establish a carbon density parameter library for land use types that is suitable for the study area. S2. Landscape and Carbon Storage Quantification: Landscape pattern indices that characterize landscape structure, clustering characteristics and heterogeneity are selected and calculated. An ecological model is used in combination with land use classification data and a carbon density parameter library to calculate the carbon storage of the county as a whole and for each land use type in the study area. S3. Spatial Feature Analysis: Spatial autocorrelation analysis is used to determine the spatial clustering characteristics of landscape pattern index and carbon storage, and hotspot identification is used to accurately locate the high and low value clustering areas of the two. S4. Analysis of Nonlinear Interaction Effects: Construct a machine learning-interpretable model to quantify the relative importance of the impact of each landscape pattern index on carbon storage, identify the threshold effect of the landscape pattern index on carbon storage, and analyze the nonlinear interaction between key landscape pattern factors on carbon storage. S5. Spatial Heterogeneity Analysis: A spatial weighted regression model was used to analyze the spatial differentiation characteristics of the impact of key landscape pattern indices on carbon storage, and to clarify the differences in the intensity and direction of the impact of each key factor on carbon storage in different counties of the study area. S6. Derivation of Influence Mechanism and Optimization Strategy: Combining the results of nonlinear interaction effect and spatial heterogeneity analysis, this study reveals the overall influence mechanism of county-level landscape pattern on carbon storage. Based on the optimal threshold range and spatial differentiation law of key factors, differentiated landscape pattern and land space optimization strategies for the study area are formulated.

2. The method for analyzing and optimizing the impact of landscape pattern on county-level carbon storage according to claim 1, characterized in that, The land use basic data in S1 consists of multi-period remote sensing monitoring data. The standardized classification process reclassifies the original land use data into six primary land categories: cultivated land, forest land, grassland, water area, construction land, and unused land. The carbon density parameter library contains the carbon density of aboveground, underground, soil, and dead organic matter for each land use type.

3. The method for analyzing and optimizing the impact of landscape pattern on county-level carbon storage according to claim 1, characterized in that, The landscape pattern indices in S2 include the maximum patch index, edge density, patch number, patch density, aggregation index, dispersion index, Shannon diversity index, Shannon evenness index, and sprawl index, which are calculated using landscape pattern analysis software. The ecological model is the carbon storage and sequestration module of the InVEST model. Carbon storage is obtained by weighted summation of the total carbon density of each land use type and its corresponding area, using the following formula: In the formula, Land use type Total carbon density, Land use type Aboveground carbon density, Land use type underground carbon density, Land use type Soil carbon density, Land use type Carbon density of dead organic matter; This represents the total carbon reserves; The number of land use types; Land use type The total area.

4. The method for analyzing and optimizing the impact of landscape pattern on county-level carbon storage according to claim 1, characterized in that, The spatial autocorrelation analysis method in S3 is the global Moran's I analysis method, and the hot spot identification method is the Getis-OrdGi cold hot spot analysis method.

5. The method for analyzing and optimizing the impact of landscape pattern on county-level carbon storage according to claim 1, characterized in that, The machine learning-interpretability model in S4 is the XGBoost-SHAP model. It captures nonlinear trends and identifies threshold effects by combining SHAP partial dependency graphs with locally weighted scatter plot smoothing curves. It deconstructs the interaction effects of key landscape pattern factors through two-dimensional partial dependency graphs. The key landscape pattern factors are the number of patches, the spread index, and the maximum patch index. The optimal threshold range for their impact on carbon sink function improvement is: the number of patches is between 200 and 500, the spread index is between 70% and 75%, and the maximum patch index is >70%.

6. The method for analyzing and optimizing the impact of landscape pattern on county-level carbon storage according to claim 1, characterized in that, The spatial weighted regression model described in S5 is the geographically weighted regression (GWR) model, and its calculation formula is as follows: In the formula, In position Carbon storage value at that location, For the regression coefficient term, , The first i The research unit k The regression coefficients and observed values ​​of the independent variables. The random error term is represented by the natural discontinuity method, which is used to classify the local regression coefficients of the model to characterize the spatial heterogeneity of the influence of key factors on carbon storage.

7. A method for enhancing county-level carbon sequestration capacity based on landscape pattern optimization, characterized in that, By applying the landscape pattern-based county carbon storage impact analysis and optimization method described in any one of claims 1-6, the threshold range and spatial differentiation characteristics of key landscape pattern factors for enhancing the carbon sink function of counties in the study area are determined. Differentiated landscape pattern optimization schemes are formulated for different counties in the study area, and the key landscape pattern factors are adjusted to the optimal threshold range to achieve the enhancement of county carbon sink capacity.

8. A landscape pattern and carbon storage analysis system, characterized in that, The system is used to implement the method for analyzing the impact of county-scale landscape patterns on carbon storage as described in any one of claims 1-7, the system comprising: Data processing module: used to acquire basic land use data, standardize and classify it, and construct a carbon density parameter library; Quantitative calculation module: used to calculate the landscape pattern index and the carbon storage of the county as a whole and for each land use type in the study area; Spatial Feature Analysis Module: Used to perform spatial autocorrelation analysis and hotspot identification, outputting the spatial clustering characteristics and distribution of clustering areas of landscape pattern index and carbon storage; Nonlinear Analysis Module: Used to build machine learning-interpretable models, outputting the relative importance of landscape pattern indices to carbon storage, threshold effects, and nonlinear interaction effects of key factors; Spatial heterogeneity analysis module: used to construct a spatial weighted regression model and output the spatial differentiation characteristics of the impact of key landscape pattern indices on carbon storage; Strategy derivation module: This module combines the results of nonlinear analysis and spatial heterogeneity analysis to output the impact mechanism of landscape pattern on carbon storage and the differentiated landscape pattern and land space optimization strategies for the study area.

Images