Energy carbon emission reduction path extraction method based on spatial-temporal heterogeneity

Abstract

The invention relates to an energy carbon emission reduction path extraction method based on spatio-temporal heterogeneity, and belongs to the field of cross application of energy, environmental science and resource utilization. The method comprises the following steps: estimating county-level energy carbon emission ECE based on remote sensing image inversion; detecting the change of county-level dominant factors by using a significant coefficient generated by space-time weighted regression STWR; and performing county-level path analysis based on the correlation between the change of the county-level dominant factors and the dynamic change of the county-level ECE. According to the method, the space-time heterogeneity relationship between the energy carbon emission and different driving factors can be detected, the optimal transfer paths at different times and in different areas are learned and extracted by matching the change quantity of the local carbon emission and the transfer (path) relationship between the change quantity of the local carbon emission and the dominant factors of the local carbon emission, and a basis is provided for a decision maker to formulate differential carbon emission reduction strategies.

Description

Technical Field

[0001] The present invention belongs to the cross - application field of energy, environmental science and resource utilization, and specifically relates to a method for extracting energy carbon emission reduction paths based on spatio - temporal heterogeneity. Background Technique

[0002] Reducing energy - related carbon emissions (ECE) and developing a green and low - carbon economy are of profound significance. The analysis of ECE reduction paths can effectively promote the development of a low - carbon economy.

[0003] Due to the spatial differences in the natural environment and social and economic activities, the ECE reduction paths in different counties may also vary

[14] . Estimating local ECE is a prerequisite for path analysis. Data for accounting ECE are published every year. However, the finest resolution of the published datasets is at the provincial level rather than the county level, so it is impossible to estimate county - level energy consumption.

[0004] Benefiting from the rapid development of night - time light (NTL) remote sensing and its related technologies, calculating local ECE based on NTL remote sensing inversion has gradually become the mainstream. Xiang et al. (2024) estimated the county - level ECE in a certain place using NPP - VIIRS NTL remote sensing images, with a relative error of less than 10% [1]. Zheng et al. (2024) used NTL data products to estimate the energy - related carbon emissions at multiple scales such as cities, counties, and towns in a certain place, with an average relative error of - 5.27% [2].

[0005] Although night-time light (NTL) image inversion has achieved high accuracy in estimating county-level energy carbon emissions (ECE), research on carbon emission reduction path analysis remains relatively lagging. Currently, there are mainly two types of carbon emission reduction analysis methods: decomposition methods and scenario methods. The former mainly includes index decomposition analysis (IDA) and structural decomposition analysis (SDA) methods. IDA has two well-known models, namely the STIRPAT (Stochastic Impacts by Regression on Population, Affluence and Technology) model and the LMDI (Logarithmic Mean Divisia Index) model. Although the IDA method helps to reveal the composition structure of carbon emissions, it has certain limitations in capturing complex dynamic relationships. For the STIRPAT model, multicollinearity is prone to occur among different factors [3]. For the LMDI model, it is difficult to support in-depth research on more detailed influencing factors [4]. The SDA method helps to analyze the impact of economic development and changes in the final demand structure on energy-related carbon emissions. However, this method requires specific data such as input-output tables and energy balance tables [5]. The scenario method mainly includes the Long-term Energy Alternatives Planning System (LEAP) and the System Dynamics (SD) model. LEAP can simulate the development paths of various industries in detail, but it may underestimate the limitations of technology application [6]. Although the SD model can demonstrate the dynamic process of energy consumption and carbon emissions, its adaptability to unforeseen future developments and changes is limited, and it cannot integrate various dynamic simulation scenarios, which further limits its prediction accuracy [7].

[0006] Although these carbon emission reduction analysis methods have achieved positive results, they all ignore exploring the dominant role of local driving factors and their transformation relationships from the perspective of spatio-temporal dynamics. The local path analysis of county-level ECE reduction should consider the main driving factors affecting the dynamic changes of ECE. Wang et al. (2024) [8] found that population (P) has largely contributed to the surging energy-related carbon emissions in African countries. Zhao et al. (2024) [9] also reported that impervious surfaces (I), which reflect urbanization, have significantly increased China's ECE. Su et al. (2018)

[10] claimed that the proportion of the secondary industry (manufacturing, M) has greatly promoted the increase in ECE. Su et al. (2020)

[11] found that the increase in the proportion of the tertiary industry (services, S) can reduce ECE. The above factors have direct or indirect impacts on ECE, but these impacts vary in different regions and periods and need further in-depth analysis.

[0007] Exploring the spatio-temporal heterogeneity relationship between local energy carbon emissions and their driving factors is the key to formulating the optimal carbon emission reduction path. Most previous studies analyzed energy carbon emission reduction and its influencing factors through global modeling methods such as STIRPAT and LMDI, ignoring the potential local heterogeneity. This limits the formulation of regional differentiated emission reduction targets and path analysis. Although national-level emission reduction targets are formulated, applying the same emission reduction path in different regions may be inappropriate and may even have a negative impact on regional economic development

[12] . To address this issue, Wang et al., (2025)

[13] used Geographically and Temporally Weighted Regression (GTWR) to explore the relationship between carbon emissions (CE) and influencing factors (such as urbanization (U) and per capita GDP (PGDP)), and found that the effects of U and PGDP on CE gradually changed from positive to negative, indicating that the heterogeneity relationship does change over time. However, the time interval decay weighting strategy adopted by GTWR cannot capture heterogeneity when the rate of numerical change is inconsistent at different locations, which limits its accuracy in analyzing local spatio-temporal heterogeneity and formulating carbon emission reduction paths and decision support reliability. The Spatio-Temporal Weighted Regression (STWR) model

[14] adopts a new decay weighting strategy based on the rate of change of numerical differences, which can better capture local spatio-temporal heterogeneity.

[0008] The existing technologies have the following defects:

[0009] 1) Most carbon emission reduction path analyses are based on global models, ignoring the local variations in the relationship between energy carbon emissions and their influencing factors in space and time, and it is difficult to accurately capture the degree and intensity of the impact of energy carbon emissions on the surrounding natural environment, social economy and other factors within the local spatio-temporal window.

[0010] 2) The change in energy carbon emissions has not been associated with the transfer of its dominant factors, and it is impossible to form carbon emission reduction paths with regional and temporal differences, and thus provide fine-grained, map-based visualization support.

[0011] 3) Decomposition analysis is one of the mainstream methods in current carbon emission reduction analysis research, including Index Decomposition Analysis (IDA) and Structural Decomposition Analysis (SDA). Although IDA helps to reveal the composition structure of carbon emissions, it has certain limitations in capturing complex dynamic relationships. For the STIRPAT model, multicollinearity is also prone to occur among different factors [3]. For the LMDI model, it is difficult to support in-depth research on more detailed influencing factors [4]. The SDA method helps to analyze the impact of economic development and changes in the final demand structure on energy-related carbon emissions. However, this method requires specific data such as input-output tables and energy balance tables [5].

[0012] 4) As another mainstream method, the scenario method mainly includes the Long-term Energy Alternatives Planning System (LEAP) and the System Dynamics (SD) model. LEAP can simulate the development paths of various industries in detail, but it may underestimate the limitations of technology applications [6]. Although the SD model can show the dynamic process of energy consumption and carbon emissions, its adaptability to unforeseen future developments and changes is limited, and it cannot integrate various dynamic simulation scenarios, which further limits its prediction accuracy [7]. Summary of the Invention

[0013] The object of the present invention is to provide a method for extracting energy carbon emission reduction paths based on spatio-temporal heterogeneity. This method can detect the spatio-temporal heterogeneity relationship between energy carbon emissions and different driving factors, and by matching the transfer (path) relationship between the change amount of local carbon emissions and its dominant factors, learn and extract the optimal transfer paths at different times and in different regions, providing a basis for decision-makers to formulate differentiated carbon emission reduction strategies.

[0014] To achieve the above object, the technical solution of the present invention is: a method for extracting energy carbon emission reduction paths based on spatio-temporal heterogeneity, including:

[0015] Estimating county-level energy carbon emissions ECE based on remote sensing image inversion;

[0016] Using the significant coefficients generated by spatio-temporal weighted regression STWR to detect the change of county-level dominant factors;

[0017] Based on the association between the change of county-level dominant factors and the dynamic change of county-level ECE, conduct county-level path analysis.

[0018] Further, estimating county-level energy carbon emissions ECE based on remote sensing image inversion includes:

[0019] (1) Provincial ECE calculation

[0020] Provincial ECE is calculated according to the IPCC estimation method, and the formula is as follows:

[0021]

[0022] Among them, N and Q n respectively represent the total number of energy consumption types and the nth energy consumption, C n and F n respectively represent the carbon emission coefficient and standard coal conversion coefficient of the nth energy consumption;

[0023] (2) Nighttime light NTL correction

[0024] For nighttime light NTL correction, the formula is as follows:

[0025] VANUI = NTL(1 - NDVI)

[0026] Among them, VANUI is the corrected night light urban index, and NDVI is the normalized difference vegetation index;

[0027] (3) Intercept-free linear model

[0028] The inversion of ECE is calculated by fitting a provincial intercept-free linear regression model, and its formula is defined as:

[0029]

[0030] Among them, M P represents the total number of pixels within a given province, VANUI m represents the VANUI of the m-th pixel, represents the provincial ECE calculated by the method in (1), and Ave() is the average function; ECE Inversion i.e., the ECE estimated by inversion.

[0031] Furthermore, the inversion and estimation of county-level energy carbon emissions ECE based on remote sensing images also include:

[0032] (4) Trend analysis

[0033] The slope is used to measure the change trend of pixel-level ECE, and the formula is as follows:

[0034]

[0035] Among them, x represents the variable to measure the trend, i.e., the estimated pixel-level ECE, x (t) is the value of x at time t, and T n is the total time.

[0036] Furthermore, the ECE estimated in (3) is verified by the estimated value using the IPCC estimation method.

[0037] Furthermore, based on the analysis result of the change trend of pixel-level ECE in (4), it can be determined whether the variable to measure the trend is in the active period or the stable period.

[0038] Furthermore, using the significant coefficients generated by the spatio-temporal weighted regression STWR to detect the change of county-level dominant factors is as follows:

[0039] (1) Fitting a spatially varying coefficient surface

[0040] The STWR model uses the time distance, i.e., the numerical difference rate between observations within a time interval, and its basic formula is:

[0041]

[0042] Among them, and respectively represent the dependent variable and independent variable of the i-th regression point (u i , v i ) at time period t. u i , v i respectively represent the horizontal and vertical axis coordinates of point i, and k represents the number of independent variables. and respectively represent the regression coefficient and error term of the i-th regression point at time period t. O Δt represents all the observation points within the time interval Δt. represents the independent variable matrix of all the observation points within the time interval Δt. W Δt (u i , v i ) represents the spatio-temporal weight matrix.

[0043] (2) Extraction of dominant factors

[0044] (2.1) Fit the spatial varying coefficients based on the STWR model, conduct a significance test with the set value of α, and screen out the counties that pass the significance test in all periods.

[0045] (2.2) Compare the absolute values of the fitting coefficients of each factor for the screened counties, and determine the factor with the largest absolute value of the fitting coefficient as the ECE dominant factor.

[0046] Furthermore, α = 0.05.

[0047] Furthermore, based on the association between the change of the county-level dominant factor and the dynamic change of county-level ECE, conduct a county-level path analysis as follows:

[0048] (1) Define the path and the occurrence of the path: Consider the change of the dominant factor as the path, including the path type and positive and negative impacts. Different counties may experience the same path, i.e., the change of the dominant factor, at different times, and one occurrence is regarded as one occurrence of the corresponding path.

[0049] (2) Path filtering rule: Remove the paths with the number of occurrences less than or equal to 5% of the total number of path occurrences.

[0050] (3) Associate with the change of the corresponding county-level ECE; Each occurrence of the path is linked to the change of ECE, ΔECE, of the corresponding county.

[0051] (4) Path sorting and extraction: For each path, average all the associated ECE changes, and then, through a sorting method, extract the best paths for different counties.

[0052] The present invention also provides an energy carbon emission reduction path extraction system based on spatio-temporal heterogeneity, including a memory, a processor, and computer program instructions stored on the memory and executable by the processor. When the processor runs the computer program instructions, the method steps described in any of the above can be implemented.

[0053] The present invention also provides a computer-readable storage medium, on which computer program instructions executable by a processor are stored. When the processor runs the computer program instructions, the method steps described in any of the above can be implemented.

[0054] Compared with the prior art, the present invention has the following beneficial effects: The method of the present invention can detect the spatio-temporal heterogeneity relationship between energy carbon emissions and different driving factors. By matching the transfer (path) relationship between the change in local carbon emissions and its dominant factors, the optimal transfer paths at different times and in different regions are learned and extracted, providing a basis for decision-makers to formulate differentiated carbon emission reduction strategies. The advantages of the present invention are as follows:

[0055] 1. Considering the spatio-temporal heterogeneity of energy carbon emission driving factors is a key area often overlooked in previous research. Many previous analyses relied on global models, which did not take into account the differences in the relationship between carbon emissions and their influencing factors at different locations and time periods. The present invention innovatively starts from the perspective of spatio-temporal heterogeneity, establishes a spatio-temporal non-stationary relationship between local (county-level) energy carbon emissions and their related factors, and forms regional (county-level) and time-effective countermeasures for energy carbon emission reduction on the basis of statistical tests and diagnoses.

[0056] 2. The present invention innovatively links the local (county-level) energy carbon emissions with the transfer of identified dominant factors to carry out the learning and exploration of carbon emission reduction paths. The proposed method can detail the specific situation of the occurrence of carbon emission paths, including the specific associated carbon emission changes, the time and location of occurrence, providing strong support for in-depth explanation and analysis. The detection of dominant factors is achieved through the local significance test of spatio-temporal heterogeneity coefficients, ensuring the accuracy and reliability of the analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is the path analysis framework for reducing energy consumption of the present invention.

[0058] Figure 2 is the trend analysis according to the ECE slope: active period (2014 - 2018) and stable period (2018 - 2021).

[0059] Figure 3 is the population from 2015 to 2021 and its corresponding coefficient surface: (a) P, (b) spatial coefficient surface of P.

[0060] Figure 4 Impervious surface (I) and its corresponding coefficient surface from 2015 to 2021: (a) I, (b) Spatial coefficient surface of I.

[0061] Figure 5 Proportion of the secondary industry (manufacturing, M) and its corresponding coefficient surface from 2015 to 2021: (a) M, (b) Spatial coefficient surface of M.

[0062] Figure 6 Proportion of the tertiary industry (service industry, S) and its corresponding coefficient surface from 2015 to 2021: (a) S, (b) Spatial coefficient surface of S.

[0063] Figure 7 Spatial distribution and transfer of dominant factors. (a) Spatial distribution; (b) Transfer.

[0064] Figure 8 Path analysis: (b) and (e) are the relevant extraction paths of the average change of ECE in the active and stable periods respectively. (a) and (c) are their corresponding spatial distributions. (b) and (d) are the optimal paths for county-level carbon emission reduction. (“+” and “-” respectively indicate that the factor has a positive and negative impact on carbon emissions).

[0065] Figure 9 Details of the path occurrence. Specific implementation manners

[0066] The technical solution of the present invention will be specifically described below in conjunction with the accompanying drawings.

[0067] The present invention provides a method for extracting energy carbon emission reduction paths based on spatio-temporal heterogeneity, including:

[0068] Estimating county-level energy carbon emissions ECE by remote sensing image inversion;

[0069] Detecting the change of county-level dominant factors by using the significant coefficients generated by spatio-temporal weighted regression STWR;

[0070] Based on the association between the change of county-level dominant factors and the dynamic change of county-level ECE, county-level path analysis is carried out.

[0071] The following is the specific implementation process of the present invention.

[0072] A method for extracting energy carbon emission reduction paths based on spatio-temporal heterogeneity of the present invention designs and proposes a path analysis framework for reducing energy-related carbon emissions (ECE) ( Figure 1), the framework includes the following steps: 1) Estimate the county-level ECE by remote sensing image inversion. 2) Detect the shift of county-level dominant factors using the significant coefficients generated by Spatiotemporal Weighted Regression (STWR). 3) Conduct county-level path analysis based on the association between historical shifts and the dynamic changes of ECE.

[0073] 1. Estimate the county-level ECE by remote sensing image inversion

[0074] 1.1 IPCC estimation method

[0075] The provincial-level ECE can be calculated according to IPCC

[15] , and the formula is as follows:

[0076]

[0077] Among them, N and Q n respectively represent the total number of energy consumption types and the nth energy consumption, C n and F n respectively represent the carbon emission coefficient and standard coal conversion coefficient of the nth energy consumption;

[0078] 1.2 Nighttime light NTL correction

[0079] The saturation of nighttime light (NTL) may lead to an underestimation of ECE in highly developed areas, resulting in inaccurate ECE estimation. The Human Settlement Index (HSI), the Modified Nighttime Light Urban Index (VANUI), and the Enhanced Vegetation Index Modified Nighttime Light Index (EANTLI) are three methods commonly used to reduce the saturation effect. HSI uses the normalized NTL and the Normalized Difference Vegetation Index (NDVI) to eliminate the saturation effect, but when NDVI approaches zero, the result may increase exponentially, leading to overcorrection. In areas with high vegetation coverage, the Enhanced Vegetation Index (EVI) is considered more sensitive than NDVI, but EVI is also more easily affected by terrain undulations. Considering that there are many hills in the study area, the VANUI method is adopted, and the formula is as follows:

[0080] VANUI = NTL(1 - NDVI) (2)

[0081] Among them, VANUI is the Modified Nighttime Light Urban Index, and NDVI is the Normalized Difference Vegetation Index;

[0082] 1.3 Intercept-free linear model

[0083] The inversion of ECE is calculated by fitting a provincial-level intercept-free linear regression model, and its formula is defined as:

[0084]

[0085] Among them, M PRepresents the total number of pixels within a given province, VANUI m Represents the VANUI of the m-th pixel, Represents the provincial ECE calculated by the method in (1), where Ave() is the average function; ECE Inversion That is, the ECE estimated by inversion.

[0086] 1.4 Precision Verification

[0087] 1.4.1 Verification of Estimation Results

[0088] The present invention uses the estimation values of the Intergovernmental Panel on Climate Change (IPCC) of the United Nations to verify the inversion results, and the results are shown in Table 1. At the provincial level, the average relative errors (RE) of places G, F, and Z are 12.29%, 7.91%, and 13.03% respectively. In addition, the public data of a certain city in 2020 and 2021 are collected and verified by the method of the Intergovernmental Panel on Climate Change (IPCC) of the United Nations. The results show that the relative errors of the inversion results of the present invention are 8.8% and 1.5% respectively (the data of other years are incomplete and thus cannot be compared).

[0089] Table 1. Relative Error (RE) (%)

[0090]

[0091] 1.4.2 Collinearity Diagnosis

[0092] To avoid the problem of multicollinearity, multicollinearity diagnosis is performed on factors such as population (P), impervious surface (I), proportion of secondary industry (M), and proportion of tertiary industry (S). As shown in Table 2, all variance inflation factors (VIF) are less than 5, and the collinearity tolerance values are all less than 1 and greater than 0.1, indicating that there is no obvious collinearity problem.

[0093] Table 2 Results of Multicollinearity Diagnosis

[0094]

[0095] 1.4.3 Model Comparative Analysis

[0096] The present invention compares the results of the ordinary least squares (OLS), geographically weighted regression (GWR), and spatio-temporal weighted regression (STWR) models between 2014 and 2021. As shown in Table 3, all the results of the spatio-temporal weighted regression (STWR) model perform best in terms of the coefficient of determination (R 2 ), sum of squared errors (SSE), and corrected Akaike information criterion (AICc).

[0097] Table 3 Model Performance Comparison of OLS, GWR, and STWR

[0098]

[0099] 1.5 Trend Analysis

[0100] The present invention uses the slope to measure the change trend of ECE, and its formula is as follows:

[0101]

[0102] where x represents the variable for which the trend is to be measured (x refers to the estimated value of pixel-level ECE retrieved based on remote sensing pixels. The estimated value at the county or provincial level is obtained by summarizing the pixel-level ECE estimated values within the administrative boundary), x (t) is the value of x at time t, and T n is the total time.

[0103] Based on the analysis results of the change trend of county-level ECE, it is possible to determine whether the variable for which the trend is to be measured is in the active period or the stable period.

[0104] (2) Detect the change of county-level dominant factors by using the significant coefficients generated by spatio-temporal weighted regression (STWR).

[0105] 2.1 Fitting the spatially varying coefficient surface

[0106] The STWR model adopts a time distance (i.e., the numerical difference rate between observations within a time interval) rather than a decay weighting strategy of time intervals, and can better explore local spatio-temporal heterogeneity. Its basic formula is:

[0107]

[0108] where and respectively represent the dependent variable and independent variable of the i-th regression point (u i , v i ) at time t. u i , v i respectively represent the horizontal and vertical axis coordinates of the i-th point. k represents the number of independent variables (here it is 4, namely P, I, M, S), and respectively represent the regression coefficient and error term of the i-th regression point at time t. O Δt represents all the observation points within the time interval Δt, represents the independent variable matrix of all the observation points within the time interval Δt, and W Δt (u i , v i ) represents the spatio-temporal weight matrix;

[0109] Similar to the Geographically Weighted Regression (GWR) model, the Spatio-Temporal Weighted Regression (STWR) model has the following assumptions: (1) There is a local linear relationship between the dependent variable and the independent variables. (2) The expected value of the random error term is zero and it follows a normal distribution. (3) The variance is homogeneous and independent of time and space. (4) Different independent variables are independent of each other and independent of the error term. (5) There is non-stationarity in space and time: the relationship between variables changes with geographical location and time and is not stable globally. (6) Local stationarity: within a certain local spatial range and adjacent time, the relationship between variables is relatively stable, so that parameter estimation can be carried out through local weighted regression. (7) The influence of adjacent observation points on the current point is greater. The model assigns different weights to observation points at different distances through a weight matrix to reflect the spatio-temporal decay effect.

[0110] 2.2 Extraction of dominant factors

[0111] 2.2.1 Fit the spatially varying coefficients based on the STWR model and conduct a significance test with α = 0.05 to screen out the counties that pass the significance test in all periods.

[0112] 2.2.2 Compare the absolute values of the fitting coefficients of each factor for the screened-out counties and determine the factor with the largest absolute value as the dominant factor of ECE.

[0113] (3) Path analysis

[0114] 3.1 Define paths and path occurrence situations. In the present invention, the transformation of the dominant factor is regarded as a path, including the path type and positive and negative impacts (represented as + / −). Different counties may experience the same path (transformation of the dominant factor) at different times, and one occurrence is regarded as one path occurrence situation of the corresponding path.

[0115] 3.2 Path filtering rules. To reduce noise and enhance the robustness of the learned paths, some rules need to be formulated to remove paths with fewer occurrences. In the present invention, we remove paths with the number of occurrences less than or equal to 5% of the total number of path occurrences (the minimum number of occurrences is 4).

[0116] 3.3 Associate with the change of local (county-level) energy carbon emissions (ECE). Each path occurrence situation can be linked to the change of ECE (ΔECE) in a specific county.

[0117] 3.4 Path sorting and extraction. For each path, the average of all associated ECE changes can be calculated. Then, through sorting or other calculation methods, the best paths for different counties can be extracted.

[0118] The research area is the GFZ region, which is the main engine of economic growth. The total GDP of GFZ is 2.1928 trillion yuan, accounting for about 21.6% of the total GDP in 2020. However, the total energy consumption in this region has also increased from 4.8 billion tons in 2010 to 7.21 billion tons in 2020, posing severe challenges to energy consumption and carbon emission control. Many cities within GFZ play a pioneering role in the development of a low-carbon economy. Exploring the emission reduction path of ECE in the GFZ region is representative and can be extended to other late-developing regions.

[0119] All source data (see Table 4) were converted to the WGS_1984_UTM coordinate system and preprocessed according to the following steps: (1) Resample the NTL data to 1000m, eliminate negative values and remove outliers. (2) Correct the NTL saturation using the Normalized Difference Vegetation Index (NDVI) data according to Equation (2). (3) Crop the corrected NTL data according to the county boundaries of the GFZ region.

[0120] Table 4. Data Sources

[0121]

[0122] The present invention first calculates the energy consumption of each province in different years, considering seven types of energy consumption, namely coal, coke, gasoline, kerosene, diesel, fuel oil, and natural gas. The carbon emission coefficients and standard coal conversion coefficients corresponding to each type of energy consumption refer to IPCC and https: / / www.nea.gov.cn / .

[0123] To test the reliability of the ECE inversion results of the present invention, we used the IPCC estimation results to test the ECE inversion results, and the results are shown in Table 1. The average relative errors of the ECE estimation values in the three provinces of G, F, and Z are 12.29%, 7.91%, and 13.03% respectively.

[0124] The pixel-level slope is calculated using formula (4), and two different trends are found. Considering that these two trends may reflect two different stages or patterns of economic development. Therefore, by distinguishing these two different stages or patterns, we should be able to conduct a more in-depth analysis of the carbon emission reduction paths under different trends, thus providing more effective carbon emission reduction policy analysis results for decision-makers. We divide it into two periods: the active period (2014 - 2018) and the stable period (2018 - 2021) (the ECE data of 2018 is used for both periods, mainly to compare and examine the differences in slopes before and after 2018). In the previous period, the average value and standard deviation (SD) of the ECE slope were 0.014 and 0.054 respectively. In the latter period, the average value and standard deviation of its slope were -0.0145 and 0.037 respectively. To illustrate the differences, we divide the slope into five groups (Table 5): "< -1.5SD>" (i.e., less than the average value - 1.5SD), "-1.5SD to -0.5SD" (i.e., between the average value - 1.5SD and the average value - 0.5SD), "-0.5SD to 0.5SD" (i.e., between the average value - 0.5SD and the average value + 0.5SD), "0.5SD to 1.5SD" (i.e., between the average value + 0.5SD and the average value + 1.5SD), and "≥1.5SD" (i.e., greater than the average value + 1.5SD). As Figure 2 shown, there are obvious differences between the standard deviation and the mean of the ECE slope in many counties during the active period, while they are relatively close to the mean during the stable period.

[0125] Table 5. Slope estimation values of ECE.

[0126]

[0127] Figure 3 The spatial distribution of P and its corresponding coefficients is given, indicating that in most counties and districts, P has a positive impact on ECE, especially around some big cities, but the scope and degree of the impact show a shrinking trend. The coefficients of the counties and districts near the boundaries of F and Z are negative, indicating that when the population increases to a certain extent, it may reduce the ECE in some areas with relatively low population density.

[0128] Figure 4 The spatial distribution of I and its corresponding coefficients is shown. For the counties and districts with a relatively low level of urbanization, the positive impact of I continues to increase; for the counties and districts with a relatively high level of urbanization, it shows an obvious negative impact. This can be explained that in the early stage of urbanization, the increase in the area of impervious surfaces has a significant positive impact on the increase in ECE, while in the later stage of urbanization, the impact of impervious surfaces decreases.

[0129] Figure 5It shows the spatial distribution of M and its corresponding coefficients. Counties with strong manufacturing industries are significantly negatively affected, which may be related to the relatively low carbon emission intensity per unit of output value.

[0130] Figure 6 It shows the spatial distribution of S and its corresponding coefficients, indicating that most counties have a negative impact on ECE, and the scope and degree of the impact show an expanding trend. Some counties in the northern part of Area Z show some positive impacts, which may indicate that the large-scale growth of S may still promote ECE to a certain extent.

[0131] To obtain more reliable path analysis results, the present invention conducted a significance test with α = 0.05 on each coefficient surface generated by STWR, and plotted the dominant factors of counties passing the test in different periods on a graph. Based on the transformation of the dominant factors and the changes in ECE, path learning and analysis were carried out. As Figure 7 shown, most counties are dominated by P (about 50.58% on average), followed by impervious surfaces (I), which is consistent with the results of previous studies. P and S are more sensitive to changes in ECE than I and M. The number of counties dominated by P decreased from 180 in 2015 to 102 in 2018, and then increased slightly to 109 in 2021. During the active period (2015 - 2018), most counties dominated by P changed to being dominated by I, M, and S, especially in the counties of Area F and Area Z. However, during the stable period (2018 - 2021), many counties dominated by S reverted to being dominated by S. The numbers of counties dominated by I and M remained relatively stable. These transformations may be related to the reduction in service activities and the slowdown of urbanization.

[0132] Figure 8 It shows the county-level path results learned from the framework of the present invention, and these paths are associated with the average changes in ECE during the active period and the stable period. During the active period, the path "S+ → P+" is most significantly correlated with ECE, contributing approximately 1.119×10 6 tons of carbon emissions. It is recommended to adopt the path "M+ → S-", because it can reduce ECE by approximately 7.747×10 5 tons. During the stable period, it is recommended to adopt the path "S+ → I+", because it can reduce the amount of ECE by approximately 1.122×10 6 . The path "M- → P+" is the worst, because it stimulates the growth of ECE, increasing by approximately 1.978×10 6 tons. The path "I+ → P+" is the second worst, increasing the amount of ECE by approximately 4.107×10 5 tons, indicating that the slowdown of urbanization does not necessarily reduce ECE.

[0133] For most counties currently dominated by P+ factors, the "P+→M-" path is a wise choice, as it reduces approximately 1.765×10 5 tons and 1.28×10 4 tons of ECE during the active and stable periods, respectively. This path indicates that improving the quality of manufacturing and transferring the population from industries with high per capita carbon emissions to high-tech manufacturing is conducive to reducing ECE.

[0134] The characteristics of the method of the present invention are as follows:

[0135] 1. Support for the analysis of local carbon emission reduction dynamic paths in (counties). Most carbon emission reduction path analyses are based on global models, ignoring the local changes of relationships in space and time. The proposed framework incorporates spatio-temporal heterogeneity using the STWR model and establishes a channel for path learning, that is, learning by combining the transformation of dominant factors with the change of carbon emissions. Therefore, it provides fine-grained, map-based visualization support for forming carbon emission reduction paths with regional and temporal differences.

[0136] 2. The paths are linked to carbon emissions and can indicate the increase or decrease of carbon emissions in local areas (counties). Local dominant factors are detectable, so the local optimal paths and expected carbon emission changes under different development periods (modes) can be predicted and mapped. Suppose the current dominant factor in a certain county is "M+". If the current carbon emissions are in the active period, the path "M+→S-" is better, as it reduced approximately 7.747×10 5 tons of carbon emissions in the past; if in the stable period, the path "M+→P+" is better, as it has the least increase in carbon emissions, approximately 1.112×10 5 tons. The practical guiding significance of these two paths is also very strong. For rapidly growing regions, the former path can be taken to reduce carbon emissions and achieve high-quality growth; while for regions with slower growth, the latter path may be more suitable for controlling the growth of carbon emissions. Figure 8 (c) and (f) respectively show the optimal paths for county-level reduction of ECE during the active and stable periods.

[0137] 3. The occurrence of each path can be traced to their corresponding carbon emission change amounts, including the time and location of their occurrence. This characteristic can provide a reference for regions with similar development models and is conducive to analyzing regional carbon emission anomalies. Many paths detected in the empirical analysis of the present invention have good interpretability. Figure 9Shows three paths: "S+→I+", "M+→S-", and "S+→P+", along with all the corresponding counties and the associated ECE amounts. Counties with the same path exhibit clustering characteristics spatially. "S+→I+" occurs in counties and cities related to the effectiveness of actively promoting the construction of smart cities. The reduction of this path is significant and relatively uniform. "M+→S-" occurs in counties and cities where the average proportion of the secondary and tertiary industries (M and S) changed from 50.4% and 43.6% in 2015 to 47.95% and 46.83% in 2018. The service industry (S) with low energy consumption and high added value gradually increased, reducing the ECE. "S+→P+" occurs in counties and cities with the loss of high-tech population and the improvement of the service quality of S (tertiary industry). That is, the proportion of the population engaged in high-tech (low-carbon emission) industries in these counties and cities may be insufficient, while the proportion of the population engaged in low-end industries with high carbon emissions is relatively large, resulting in a sharp increase in ECE. It is worth noting that although the overall change in carbon emissions fluctuates around the average value, the deviation of individual counties and cities is relatively large. If more data is used, the number of occurrences of each path will increase, making it more stable and reliable.

[0138] 4. It also supports comparative analysis of the selected counties and counties with similar conditions. As shown in Table 6, City A and County B are very close in terms of spatial location, energy carbon emissions (ECE), dominant factors, population size, industrial structure, etc. Their energy carbon emissions are both around 1.5 million tons, with a change range of less than 0.5 million tons. Their population sizes are both around 470,000. The proportion of their secondary industry (manufacturing, M) is about 30%, lower than the proportion of the tertiary industry (service industry, S). Their dominant factors have changed from an increase in population factors (P+) to an increase in impervious surface factors (I+). However, the situation of County C is very different from that of City A and County B. Its energy carbon emissions exceed 3 million tons, with a change range of more than 1 million tons. Its population size has doubled to 1.4 million. The proportion of its tertiary industry (S) is higher than that of the secondary industry (M), and its dominant factor has not changed.

[0139] Table 6. Comparison of the occurrence of paths and related information

[0140]

[0141] 5. Based on statistical tests and diagnoses of spatio-temporal heterogeneity relationships, energy carbon emission countermeasure methods are formed. Through STWR, the spatio-temporal non-stationary relationship between local energy carbon emissions (at the county level) and their related factors is established, and on the basis of statistical tests and diagnoses, regional (county-level) and time-effective energy carbon emission reduction countermeasures are formed.

[0142] The embodiments of the present invention also provide an energy carbon emission reduction path extraction system based on spatio-temporal heterogeneity, including a memory, a processor, and computer program instructions stored on the memory and executable by the processor. When the processor runs the computer program instructions, the method steps described in any of the above can be implemented.

[0143] The embodiments of the present invention also provide a computer-readable storage medium, on which computer program instructions executable by the processor are stored. When the processor runs the computer program instructions, the method steps described in any of the above can be implemented.

[0144] References:

[0145] [1] Xiang, C.; Mei, Y.; Liang, A. Analysis of Spatiotemporal Changes in Energy Consumption Carbon Emissions at District and County Levels Based on Nighttime Light Data—A Case Study of Jiangsu Province in China. Remote Sensing 2024, 16, 3514, doi: https: / / doi.org / 10.3390 / rs16183514.

[0146] [2] Zheng, Y.; Fan, M.; Cai, Y.; Fu, M.; Yang, K.; Wei, C. Spatio-temporal pattern evolution of carbon emissions at the city-county-town scale in Fujian Province based on DMSP / OLS and NPP / VIIRS nighttime light data. Journal of Cleaner Production 2024, 442.

[0147] [3] Li, W.; Gao, S. Prospective on energy related carbon emissions peak integrating optimized intelligent algorithm with dry process technique application for China's cement industry. Energy 2018, 165, 33 - 54.

[0148] [4] Wang, P.; Wu, W.; Zhu, B.; Wei, Y. Examining the impact factors of energy - related CO2 emissions using the STIRPAT model in Guangdong Province, China. Applied Energy 2013, 106, 65 - 71.

[0149] [5] Chong, C. H.; Tan, W. X.; Ting, Z. J.; Liu, P.; Ma, L.; Li, Z.; Ni, W. The driving factors of energy - related CO2 emission growth in Malaysia: The LMDI decomposition method based on energy allocation analysis. Renewable and Sustainable Energy Reviews 2019, 115.

[0150] [6] Huang, R.; Zhang, S.; Wang, P. Key areas and pathways for carbon emissions reduction in Beijing for the “Dual Carbon” targets. Energy Policy 2022, 164, 112873.

[0151] [7]Lin, C.;Li, X. Carbon peak prediction and emission reduction pathways exploration for provincial residential buildings: Evidence from Fujian Province. Sustainable Cities and Society 2024, 102.

[0152] [8]Wang, J.;Shan, Y.;Cui, C.;Zhao, C.;Meng, J.;Wang, S. Investigating the fast energy-related carbon emissions growth in African countries and its drivers. Applied Energy 2024, 357, 122494.

[0153] [9]Zhao, J.;Chen, W.;Liu, Z.;Liu, W.;Li, K.;Zhang, B.;Zhang, Y.;Yu, L.;Sakai, T. Urban expansion, economic development, and carbon emissions: Trends, patterns, and decoupling in mainland China’s provincial capitals (1985–2020). Ecological Indicators 2024, 169, 112777, doi: https: / / doi.org / 10.1016 / j.ecolind.2024.112777.

[0154]

[10] Su, W.;Liu, Y.;Wang, S.;Zhao, Y.;Su, Y.;Li, S. Regional inequality, spatial spillover effects, and the factors influencing city-level energy-related carbon emissions in China. Journal of Geographical Sciences 2018, 28, 495-513.

[0155]

[11] Yang, L.;Wang, J.;Shi, J. Can China meet its 2020 economic growth and carbon emissions reduction targets?Journal of Cleaner Production 2017, 142, 993 - 1001, doi: https: / / doi.org / 10.1016 / j.jclepro.2016.08.018.

[0156]

[12] Wang, Y.;Yang, H.;Sun, R. Effectiveness of China's provincial industrial carbon emission reduction and optimization of carbon emission reduction paths in “lagging regions”: Efficiency - cost analysis. Journal of Environmental Management 2020, 275.

[0157]

[13] Wang, N.;Qu, Z.;Li, J.;Zhang, Y.;Wang, H.;Xi, H.;Gu, Z. Spatial - temporal patterns and influencing factors of carbon emissions in different regions of China. Environmental Research 2025.

[0158]

[14] Que, X.;Ma, X.;Ma, C.;Chen, Q. A Spatiotemporal Weighted Regression Model(STWRv1.0) for Analyzing Local Non - stationarity in Space and Time. Geoscientific Model Development Discussions 2020, 2020, 1 - 33, doi: https: / / doi.org / 10.5194 / gmd - 13 - 6149 - 2020.

[0159]

[15] Eggleston, H.; Buendia, L.; Miwa, K.; Ngara, T.; Tanabe, K. 2006 IPCC guidelines for national greenhouse gas inventories. 2006.

[0160] The above are the preferred embodiments of the present invention. All changes made according to the technical solution of the present invention, when the functions and effects produced do not exceed the scope of the technical solution of the present invention, shall fall within the protection scope of the present invention.

Claims

1. A method for extracting energy carbon emission reduction paths based on spatio-temporal heterogeneity, characterized in that, Including: Estimating county-level energy carbon emissions (ECE) by remote sensing image inversion; Detecting the transformation of county-level dominant factors using the significant coefficients generated by spatio-temporal weighted regression (STWR); Conducting county-level path analysis based on the association between the transformation of county-level dominant factors and the dynamic changes of county-level ECE.

2. The method for extracting an energy carbon emission reduction path based on spatio-temporal heterogeneity according to claim 1, wherein Estimating county-level energy carbon emissions (ECE) by remote sensing image inversion, including: (1) Provincial ECE calculation Provincial ECE is calculated according to the IPCC estimation method, and the formula is as follows: Among them, N and Q n respectively represent the total number of energy consumption types and the nth type of energy consumption, C n and F n respectively represent the carbon emission coefficient and the standard coal conversion coefficient of the nth type of energy consumption; (2) Nighttime light (NTL) correction The formula for nighttime light (NTL) correction is as follows: VANUI = NTL(1 - NDVI) Where VANUI is the corrected nighttime light urban index and NDVI is the normalized difference vegetation index; (3) Intercept-free linear model The inversion of ECE is calculated by fitting a provincial intercept-free linear regression model, and its formula is defined as: Among them, M P represents the total number of pixels within a given province, and VANUI m represents the VANUI of the m-th pixel, and ECE I P PCC represents the provincial ECE calculated by the method in (1), and Ave() is the average function; ECE Inversion That is, the ECE obtained by inversion estimation.

3. A method for extracting an energy carbon emission reduction path based on spatio-temporal heterogeneity according to claim 2, characterized in that, Estimating county-level energy carbon emissions (ECE) by remote sensing image inversion also includes: (4) Trend analysis The slope is used to measure the change trend of pixel-level ECE, and the formula is as follows: where x represents the variable for which the trend is to be measured, namely the estimated pixel-level ECE, and x (t) is the value of x at time t, and T n is the total time.

4. A method for extracting an energy carbon emission reduction path based on spatio-temporal heterogeneity according to claim 2, characterized in that, For the ECE estimated in (3), the estimated value of the IPCC estimation method is used to verify the estimation result.

5. A method for extracting an energy carbon emission reduction path based on spatio-temporal heterogeneity according to claim 2, characterized in that, Based on the analysis result of the change trend of pixel-level ECE in (4), it can be determined whether the variable to be measured for the trend is in the active period or the stable period.

6. The method for extracting an energy carbon emission reduction path based on spatio-temporal heterogeneity according to claim 1, wherein Detecting the transformation of county-level dominant factors using the significant coefficients generated by spatio-temporal weighted regression (STWR), specifically as follows: (1) Fitting the spatially varying coefficient surface The STWR model uses the time distance, that is, the numerical difference rate between the observed values within a time interval, and its basic formula is: Among them, and represent the dependent variable and independent variable of the \(i\)-th regression point \((u i , v i )\) at time \(t\), respectively. \(u i , v i represent the horizontal and vertical axis coordinates of point \(i\) respectively, \(k\) represents the number of independent variables, and represent the regression coefficient and error term of the \(i\)-th regression point at time \(t\) respectively. \(O Δt represents all the observation points within the time interval \(\Delta t\), represents the independent variable matrix of all the observation points within the time interval \(\Delta t\), and \(W Δt (u i , v i ) represents the spatio-temporal weight matrix; (2) Extracting the dominant factors (2.1) Based on the STWR model to fit the spatially varying coefficient, a significance test is carried out with the set value of α, and the counties that pass the significance test in all periods are screened out; (2.2) Comparing the absolute values of the fitting coefficients of each factor of the screened-out counties, and the factor with the largest absolute value of the fitting coefficient is determined as the dominant factor of ECE.

7. A method for extracting an energy carbon emission reduction path based on spatio-temporal heterogeneity according to claim 6, characterized in that α＝0.05。 8. A method for extracting an energy carbon emission reduction path based on spatio-temporal heterogeneity according to claim 1, characterized in that, Conducting county-level path analysis based on the association between the transformation of county-level dominant factors and the dynamic changes of county-level ECE, specifically as follows: (1) Defining the path and the path occurrence situation: Regarding the transformation of the dominant factor as the path, including the path type and the positive and negative impacts. Different counties may experience the same path, that is, the transformation of the dominant factor, at different times, and one occurrence is regarded as one occurrence situation of the corresponding path; (2) Path filtering rule: Removing the paths with the number of path occurrences less than or equal to 5% of the total number of path occurrences; (3) Associating with the changes of the corresponding county-level ECE; Each path occurrence situation is linked to the change of ECE (ΔECE) of the corresponding county; (4) Path sorting and extraction: For each path, the average of all associated ECE changes is calculated, and then, through a sorting-including method, the best paths of different counties are extracted.

9. An energy carbon emission reduction path extraction system based on spatio-temporal heterogeneity, characterized in that Including a memory, a processor, and computer program instructions stored on the memory and executable by the processor. When the processor runs the computer program instructions, it can implement the method steps as described in any one of claims 1-8.

10. A computer-readable storage medium storing computer program instructions executable by a processor, which, when executed by the processor, can implement the method steps described in any one of claims 1-8.

Images