Method for predicting carbon reserves and analyzing driving mechanism of carbon reserves

Abstract

The invention provides a method for predicting carbon reserves and analyzing a driving mechanism of the carbon reserves, and the method comprises the following steps: firstly collecting land utilization data of two periods, and constructing a land utilization change and carbon reserve correlation model based on an RF-FLUS-InVEST model in combination with carbon density data; furthermore, a driving mechanism of natural and social economic factors on carbon reserve space differentiation is quantified through an optimal parameter geographic detector, which comprises the following steps of: checking and screening key driving factors based on interaction, optimizing parameter configuration of the geographic detector by utilizing a differentiation factor (q value), and revealing a non-linear influence rule of a multi-factor synergistic effect on carbon reserve change. According to the method, the limitation of subjective parameter setting in traditional drive analysis is overcome, high-precision spatial analysis of a carbon reserve evolution mechanism is realized, and a scientific basis is provided for regional ecological protection and carbon neutralization policy formulation.

Description

Technical Field

[0001] The present invention relates to the technical field of ecosystem service assessment and carbon cycle analysis, and in particular to a method for carbon reserve prediction and its driving mechanism analysis. Background Art

[0002] As global climate change becomes increasingly severe, carbon storage, as a key indicator of ecosystem carbon sequestration capacity, has become a core component of regional ecological management and climate change response research. Carbon storage not only reflects the ability of ecosystems to absorb and store carbon dioxide but also directly impacts the balance of the global carbon cycle and the stability of the climate system. Therefore, in-depth research on the spatiotemporal distribution of carbon storage and its driving mechanisms is crucial for formulating sound ecological protection policies and addressing climate change.

[0003] In the field of carbon storage research, with the continuous development of remote sensing technology, geographic information systems, and ecosystem modeling, the FLUS-InVEST model, which integrates land-use change simulation with ecosystem service assessment, has been widely used in carbon storage research. The FLUS model can simulate the spatiotemporal patterns of land-use change based on historical data and future scenarios, providing a reliable tool for predicting future land-use change. The FLUS model has been improved using the random forest algorithm to predict the adaptability probability of land-use types based on driving factors. The feature importance of the random forest algorithm quantifies the contribution of driving factors, which can improve the accuracy and robustness of land-use change simulations. The InVEST model, on the other hand, accurately estimates carbon storage under different land-use types by quantifying the spatial distribution of ecosystem services. The combination of the two not only provides strong technical support for dynamic simulations of carbon storage but also opens up new research avenues for uncovering the mechanisms that influence land-use change on carbon storage. Furthermore, the Optimal Parameter Geographic Detector (OPGD), an emerging spatial statistical method, can effectively reveal the independent and interactive effects of natural and anthropogenic factors on the spatial variation of carbon storage, providing a new perspective for understanding the driving mechanisms of carbon storage change. Compared to traditional geodetectors, OPGD features optimized parameter settings, enabling it to more accurately capture complex relationships between factors, thereby providing more reliable and detailed spatially differentiated information. This optimization gives OPGD an advantage in processing complex geographic data, allowing it to better adapt to the characteristics and needs of different research areas, further enhancing the value and accuracy of geodetector methods in carbon storage research. Summary of the Invention

[0004] The purpose of the present invention is to solve the technical problems in the above background and to propose a method for carbon storage prediction and its driving mechanism analysis, comprising the following steps:

[0005] S1. Collect land use data and driving factor datasets in the target area;

[0006] S2, construct land use change based on FLUS model and simulate land use change in different scenarios;

[0007] S3, combining the land use change and the corrected carbon density data in step S2, and simulating the spatial distribution of carbon storage in different periods based on the INVEST model;

[0008] S4. Quantify the contribution of driving factors to the spatial differentiation of carbon storage using the optimal parameter geographic detector;

[0009] S5. Output carbon storage prediction results and driving mechanism analysis report to provide a basis for ecological protection decision-making.

[0010] In a preferred embodiment, step S4 includes the following steps:

[0011] S41. Screen key driving factors through interaction tests;

[0012] S42. Optimize the parameter configuration of the geographic detector based on differentiation factors and calculate the nonlinear impact weight of the synergistic effect of multiple factors on carbon storage changes.

[0013] In a preferred solution, the driving factors in step S1 include natural factors and socioeconomic factors.

[0014] In a preferred embodiment, the land use data in step S1 includes a land use type map interpreted from remote sensing images, with a time span of no less than two periods;

[0015] Carbon density data are obtained through field sampling or literature databases, and are classified and assigned values according to vegetation type and soil type, and corrected using the carbon density correction formula.

[0016] In the preferred solution, in step S2, a random forest regression algorithm is used to predict the adaptability probability of land use types based on driving factors, and the obtained adaptability probability matrix is input into the spatial competition module of FLUS to replace the original ANN output; the random forest includes several decision trees.

[0017] In a preferred embodiment, step S3 includes the following steps:

[0018] The total carbon storage in the region is obtained by multiplying the area of each land use type by the carbon density of the four major carbon pools and adding them up. The calculation formula is:

[0019]

[0020] Where Ctot is the total ecosystem carbon storage; i is a certain land use type; n is the number of land use types; Ai is the area of land use type i; Ci-above, Ci-below, Ci-soil and Ci-dead are the aboveground biomass carbon density, belowground biomass carbon density, soil carbon density and dead organic carbon density of land use type i, respectively.

[0021] In a preferred embodiment, step S4 includes the following steps:

[0022] S401. Statistical model for quantifying spatial heterogeneity and its driving mechanism. The core indicator is the q value, which represents the explanatory power of the independent variable on the dependent variable. The q value calculation formula of the geographic detector is:

[0023]

[0024] Among them, L is the number of classification layers of the independent variable, N h and are the number of samples and the variance of the dependent variable in the hth layer, N and σ 2 is the total number of samples and the overall variance. The larger the q value is, the stronger the ability of the independent variable to explain the spatial differentiation of the dependent variable is.

[0025] S402, Optimal Parameter Geographic Detector improves the model's accuracy in interpreting spatial driving mechanisms through parameter optimization strategies, including optimizing the number of classes, optimizing discretization methods, and quantifying interactions:

[0026] In the optimization of the number of classifications, the optimal number of discretization layers of the independent variables is determined based on the grid search or information entropy maximization principle to maximize the q value of the representation explanatory power;

[0027] The optimal discretization method is to select the classification method of the optimal segmentation dependent variable variance through the natural breakpoint method, the equal interval method or the machine learning clustering algorithm to enhance the objectivity of the stratification; combined with the interaction detector, the synergistic effect of the two factors is quantified according to the formula:

[0028]

[0029] When satisfied When , it is determined that there is a two-factor synergistic enhancement effect.

[0030] In the preferred solution, the key driving factors in step 4 include GDP, average annual temperature, DEM, slope, and distance to railway.

[0031] In a preferred solution, the analysis report in step S5 includes a carbon storage spatial distribution map, a driving factor contribution radar map, and a multi-factor interaction heat map.

[0032] In the preferred solution, in step S2, the construction process of each decision tree includes:

[0033] S21, Gini impurity is used to measure the purity of the data set. The smaller the value, the purer the data set. The formula is:

[0034]

[0035] Among them, D is the dataset, K is the number of categories, and p k is the proportion of the k-th class samples in the data set D;

[0036] S22, information gain is used to select the best splitting feature. The greater the information gain, the greater the purity of the data set after splitting. The formula is:

[0037]

[0038] Among them, D is the data set, a is the split feature, V is the possible value of feature a, D v is the subset of feature a with value v, and Entropy(D) is the entropy of the dataset D;

[0039] S23. Entropy is used to measure the degree of disorder of a data set. The smaller the value, the purer the data set. The formula is:

[0040]

[0041] Among them, pk is the proportion of samples of the kth category in the dataset D.

[0042] The beneficial effects of this invention include significantly outperforming traditional single-model or static parameter methods in terms of prediction accuracy, driving mechanism analysis, dynamic adaptability, and decision support. Its findings can provide key technical support for regional carbon peak pathway planning and ecological protection and restoration projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Flowchart for the implementation of the present invention.

[0044] Figure 2 This is the principle diagram of the random forest algorithm.

[0045] Figure 3 Predict the land use in the Three Gorges Reservoir area for a specified year and under different scenarios.

[0046] Figure 4 A map showing changes in land use types between specified years.

[0047] Figure 5 The predicted carbon storage distribution map of the Three Gorges Reservoir area for a specified year and under different scenarios.

[0048] Figure 6is the contribution of a single driving factor to carbon storage distribution.

[0049] Figure 7 It shows the contribution of the interaction of multiple driving factors to the distribution of carbon stocks. DETAILED DESCRIPTION

[0050] The present invention provides a high-precision, dynamic carbon storage prediction method that combines spatial analysis with a driving factor quantification model, with the aim of revealing the impact mechanism of natural and human factors on carbon storage.

[0051] The above-mentioned object of the present invention is achieved through the following technical steps:

[0052] Step 1: Collect land use data, carbon density data, and driving factor datasets for the target area, including natural factors and socioeconomic factors;

[0053] The land use data in step 1 include land use type maps interpreted from remote sensing images, with a time span of no less than two periods; the carbon density data are obtained through field sampling or literature databases, and are classified and assigned values according to vegetation type and soil type.

[0054] Step 2: Construct land use change based on the FLUS model and simulate land use change in different scenarios;

[0055] Step 2.1: Use the random forest algorithm to perform feature importance analysis on the driving factors to determine the relative importance of each driving factor to land use change, so as to more effectively explore the relationship between driving factors and land use change.

[0056] Random forests are composed of multiple decision trees. The construction process of each decision tree is as follows:

[0057] Step 2.2: Gini impurity is used to measure the purity of a dataset. The smaller the value, the purer the dataset. The formula is as follows:

[0058]

[0059] Among them, D is the dataset, K is the number of categories, and p k is the proportion of samples of the kth class in the dataset D.

[0060] Step 2.3: Information gain is used to select the best splitting feature. The greater the information gain, the greater the purity of the dataset after splitting. The formula is as follows:

[0061]

[0062] Among them, D is the data set, a is the split feature, V is the possible value of feature a, D vis the subset of feature a whose value is v, and Entropy(D) is the entropy of the dataset D.

[0063] Step 2.4: Entropy is used to measure the degree of disorder in a dataset. A smaller value indicates a purer dataset. The formula is as follows:

[0064]

[0065] Among them, pk is the proportion of samples of the kth category in the dataset D.

[0066] Step 2.5: The specific calculation process of carbon density correction method is as follows:

[0067] C SP =3.3968×MAP+3996.1

[0068] C BP =6.798×e 0.0054×MAP

[0069] C BT =28×MAT+398

[0070] Among them, CSP and CBP represent the soil carbon density and biomass carbon density calculated by the average annual rainfall (t·hm-2), respectively; CBT represents the biomass carbon density calculated by the average annual temperature (t·hm-2); MAP and MAT represent the average annual rainfall (mm / a) and the average annual temperature (℃ / a), respectively.

[0071]

[0072] K B =K BP ×K BT

[0073] KBP and KBT are the correction coefficients for the precipitation and temperature factors of biomass carbon density, respectively. C'BP and C”BP represent the biomass carbon densities of the study area and the country, respectively, calculated using average annual rainfall; C'SP and C”SP represent the soil carbon densities of the study area and the country, respectively, calculated using average annual rainfall data; C'BT and C”BT represent the biomass carbon densities of the study area and the country, respectively, calculated using average annual temperature data; KS and KB represent the soil carbon density correction coefficients and biomass carbon density correction coefficients, respectively.

[0074] Step 3: Combined with the land use changes in step 2, the spatial distribution of carbon storage in different periods is simulated based on the INVEST model;

[0075] Step 3.1: Multiply the area of each land use type by the carbon density of the four major carbon pools and add them up to obtain the total carbon storage in the region. The calculation formula is:

[0076]

[0077] Where: Ctot is the total ecosystem carbon storage (t); i is a certain land use type; n is the number of land use types; Ai is the area of land use type i (hm2); Ci-above, Ci-below, Ci-soil and Ci-dead are the aboveground biomass carbon density (t·hm-2), belowground biomass carbon density (t·hm-2), soil carbon density (t·hm-2) and dead organic carbon density (t·hm-2) of land use type i, respectively.

[0078] Step 4: Use the optimal parameter geographic detector to quantify the contribution of driving factors to the spatial variation of carbon storage, including:

[0079] Screen key driving factors through interaction tests;

[0080] The parameter configuration of the geographic detector is optimized based on the differentiation factor (q value), and the nonlinear impact weight of the synergistic effect of multiple factors on carbon storage changes is calculated.

[0081] Step 4.1: Quantify the statistical model of spatial heterogeneity and its driving mechanism. The core indicator is the q value, which represents the explanatory power of the independent variable on the dependent variable. The q value calculation formula of the traditional geographic detector is:

[0082]

[0083] Among them, L is the number of classification layers of the independent variable, N h and are the number of samples and the variance of the dependent variable in the hth layer, N and σ 2 is the total number of samples and the overall variance. The larger the q value, the stronger the ability of the independent variable to explain the spatial differentiation of the dependent variable.

[0084] Step 4.2: The Optimal Parameter Geographic Detector (OPGD) improves the model's accuracy in explaining spatial driving mechanisms through parameter optimization strategies. Its core components include optimizing the number of clusters, optimizing discretization methods, and quantifying interactions. In optimizing the number of clusters, grid search or information entropy maximization principles are used to determine the optimal number of discretization layers for the independent variables to maximize the q-value representing the explanatory power. Discretization method optimization uses natural breakpoints, equal spacing, or machine learning clustering algorithms to select the classification method that can optimally divide the variance of the dependent variable, thereby enhancing the objectivity of the stratification. Furthermore, the interaction detector is combined to quantify the synergistic effect of the two factors, according to the formula:

[0085]

[0086] If satisfied It is determined that there is a two-factor synergistic enhancement effect.

[0087] Step 5: Output carbon storage prediction results and driving mechanism analysis report to provide a basis for ecological protection decision-making.

[0088] Compared with the prior art, the present invention has the following beneficial effects:

[0089] It significantly outperforms traditional single-model or static parameter methods in terms of prediction accuracy, driving mechanism analysis, dynamic adaptability, and decision support. Its findings can provide key technical support for regional carbon peak pathway planning and ecological protection and restoration projects.

[0090] The present invention will be further described below with reference to the accompanying drawings. The following examples are only used to more clearly explain the technical solutions of the present invention.

[0091] like Figure 1 As shown, the method first requires preparing relevant land use drivers for the study area over the two required time periods. For example, natural drivers include data on average annual temperature, elevation, and slope; transportation drivers include data on distance to water systems and national highways; and social drivers include data on GDP. High-resolution land use data for the study area for the specified years is also required, along with carbon density data for the four major carbon pools corresponding to each land use type.

[0092] Step 1: simulate the change of land use types in the study area;

[0093] Step 1.1 Figure 2 As shown, the random forest algorithm (RF) is used to predict the adaptability probability of land use types based on driving factors;

[0094] Step 1.2: Based on the land cover change patterns and planning policies of the research area during a specific period, a land use cost transfer matrix is established under different scenarios.

[0095] Step 1.3: Set the domain weights of each land use type in the study area according to relevant literature;

[0096] Step 1.4 Figure 3 The land use data under different scenarios were obtained using ArcGIS and drawn using origin. Figure 4 Land use conversion map shown;

[0097] Step 2 Figure 5 As shown in the figure, the carbon storage distribution of the region can be obtained by multiplying the area of each land use type by the carbon density of the four major carbon pools and summing them up.

[0098] Step 3 Figure 6 、 Figure 7The figure shows the use of the optimal parameter geographic detector to study the individual and interactive effects of each driving factor on carbon storage changes in the study area.

[0099] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for carbon storage prediction and analysis of its driving mechanism, characterized by: The following steps are involved: S1. Collect land use data and driving factor datasets in the target area; S2, construct land use change based on FLUS model and simulate land use change in different scenarios; S3, combining the land use change and the corrected carbon density data in step S2, and simulating the spatial distribution of carbon storage in different periods based on the INVEST model; S4. Quantify the contribution of driving factors to the spatial differentiation of carbon storage using the optimal parameter geographic detector; S5. Output carbon storage prediction results and driving mechanism analysis report to provide a basis for ecological protection decision-making.

2. The method for carbon storage prediction and its driving mechanism analysis according to claim 1 is characterized by: Step S4 includes the following steps: S41. Screen key driving factors through interaction tests; S42. Optimize the parameter configuration of the geographic detector based on differentiation factors and calculate the nonlinear impact weight of the synergistic effect of multiple factors on carbon storage changes.

3. The method for carbon storage prediction and its driving mechanism analysis according to claim 1 is characterized by: The driving factors in step S1 include natural factors and socioeconomic factors.

4. The method for carbon storage prediction and its driving mechanism analysis according to claim 1 is characterized by: The land use data in step S1 includes a land use type map interpreted from remote sensing images, with a time span of no less than two periods; Carbon density data are obtained through field sampling or literature databases, and are classified and assigned values according to vegetation type and soil type, and corrected using the carbon density correction formula.

5. The method for carbon storage prediction and its driving mechanism analysis according to claim 1 is characterized by: In step S2, the random forest regression algorithm is used to predict the adaptability probability of land use types based on driving factors. The obtained adaptability probability matrix is input into the spatial competition module of FLUS to replace the original ANN output; the random forest includes several decision trees.

6. The method for carbon storage prediction and its driving mechanism analysis according to claim 1 is characterized by: Step S3 includes the following steps: The total carbon storage in the region is obtained by multiplying the area of each land use type by the carbon density of the four major carbon pools and adding them up. The calculation formula is: Where Ctot is the total ecosystem carbon storage; i is a certain land use type; n is the number of land use types; Ai is the area of land use type i; Ci-above, Ci-below, Ci-soil and Ci-dead are the aboveground biomass carbon density, belowground biomass carbon density, soil carbon density and dead organic carbon density of land use type i, respectively.

7. The method for carbon storage prediction and its driving mechanism analysis according to claim 1 is characterized by: Step S4 includes the following steps: S401. Statistical model for quantifying spatial heterogeneity and its driving mechanism. The core indicator is the q value, which represents the explanatory power of the independent variable on the dependent variable. The q value calculation formula of the geographic detector is: Among them, L is the number of classification layers of the independent variable, N h and are the number of samples and the variance of the dependent variable in the hth layer, N and σ 2 is the total number of samples and the overall variance. The larger the q value is, the stronger the ability of the independent variable to explain the spatial differentiation of the dependent variable is. S402, Optimal Parameter Geographic Detector improves the model's accuracy in interpreting spatial driving mechanisms through parameter optimization strategies, including optimizing the number of classes, optimizing discretization methods, and quantifying interactions: In the optimization of the number of classifications, the optimal number of discretization layers of the independent variables is determined based on the grid search or information entropy maximization principle to maximize the q value of the representation explanatory power; The optimal discretization method is to select the classification method of the optimal segmentation dependent variable variance through the natural breakpoint method, the equal interval method or the machine learning clustering algorithm to enhance the objectivity of the stratification; combined with the interaction detector, the synergistic effect of the two factors is quantified according to the formula: When satisfied When , it is determined that there is a two-factor synergistic enhancement effect.

8. The method for carbon storage prediction and its driving mechanism analysis according to claim 1 is characterized by: The key driving factors in step 4 include GDP, mean annual temperature, DEM, slope, and distance to railway.

9. The method for carbon storage prediction and its driving mechanism analysis according to claim 1 is characterized by: The analysis report in step S5 includes a carbon storage spatial distribution map, a driving factor contribution radar map, and a multi-factor interaction heat map.

10. The method for carbon storage prediction and its driving mechanism analysis according to claim 5, characterized in that: In step S2, the construction process of each decision tree includes: S21, Gini impurity is used to measure the purity of the data set. The smaller the value, the purer the data set. The formula is: Where D is the dataset, K is the number of categories, and p k is the proportion of the k-th class samples in the data set D; S22, information gain is used to select the best splitting feature. The greater the information gain, the greater the purity of the data set after splitting. The formula is: Among them, D is the data set, a is the split feature, V is the possible value of feature a, D v is the subset of feature a with value v, and Entropy(D) is the entropy of the dataset D; S23. Entropy is used to measure the degree of disorder of a data set. The smaller the value, the purer the data set. The formula is: Among them, pk is the proportion of samples of the kth category in the dataset D.

Images