Driving Factors of CO 2 Concentration in Mainland China Based on GWR

. Considering that the complexity and dynamicity of CO 2 emissions, the spatiotemporal distribution pattern of atmospheric CO 2 and its drivers remain unclear. In this study, we used the Geographically Weighted Regression (GWR) method to analyze the comprehensive distribution of CO2 concentration in mainland China from 2015 to 2019. We considered the relationship between nine factors, including natural and human activities, and CO2 concentration. To clarify the correlation between CO2 concentration and drivers, we utilized Pearson’s correlation coefficient. Then, the GWR analysis revealed the spatial heterogeneity across provinces, which reflects the extent to which impact factors influence CO 2 concentrations. Finally, we analysed CO 2 concentration spatiotemporal variation characteristics and predicted future trends of CO 2 concentration in 31 provinces in China. According to our research, GDP has a major impact on CO 2 growth, while natural factors have a minor influence on CO 2 concentration. Our study found significant regional differences in the effects of combined variables on CO 2 concentrations, with monthly rotational patterns temporally and clustering of high growth rates spatially and CO 2 concentration in mainland China will continue to steadily increase. The findings of this research are crucial for China’s future energy low-carbon transition and policy-making.


Introduction
Carbon dioxide (CO 2 ) is a greenhouse gas in regulating the earth's temperature by absorbing and releasing heat over time. The industrial revolution has led to a significant increase in atmospheric CO 2 concentration, particularly by 19 ppm from 2011 to 2019, making China the largest carbon emitter globally. Consequently, changes in CO 2 concentration in China will have a significant impact on global climate change and international carbon emissions negotiations. To comply with international demand for carbon reduction, China's State Council has issued guidelines to reach the peak of carbon emissions by 2030 and strive for carbon neutrality by 2060. Given the complexity and dynamism of CO 2 concentration, many scholars are studying the trends and influencing factors of atmospheric CO 2 concentration in China. Several studies, including those by Deng et al [1], Zhu et al [2], and Wang et al [3], have analyzed the drivers of human activities in carbon concentration by combining CO2 concentration with industrial structure, energy intensity, population density, and economic development. However, these studies lack in-depth discussion of natural factors. To address this gap, other studies, such as those by Martin Ju et al [4] and Piao and Liu et al [5], have constructed models and performed analysis to study the effects of temperature, water, and plant phenology changes on CO2 concentration. However, current researches focus primarily on anthropogenic factors, and neglect the role of natural factors. This study aims to employ Geographically Weighted Regression (GWR) to investigate the intrinsic relationship between China's CO 2 concentration and both human and natural factors. By identifying the key factors influencing CO 2 concentration and forecasting its future trend, the findings of this study will provide crucial information for China's low-carbon policy-making.

Data sources
The data sources of this study mainly include CO 2 concentration data and related influencing factors from 31 provinces in mainland China. The CO 2 concentration data from 2015-2019 is provided by Big Data Cloud Platform for Global Change (http://globid.cn). The total population, GDP and energy consumption data of 31 provinces come from the China Statistical Yearbook. Surface net solar radiation, soil water content, surface temperature, relative humidity, and precipitation come from the ERA5-Land(Copernicus Climate Change Service, 2019). Enhanced vegetation index(EVI) is calculated by the data of a Moderate-resolution Imaging Spectroradiometer(MODIS). All data needs to be preprocessed into grid data with the same range and resolution by ArcGIS and Matlab.

Research methods
To study the provincial differences and influencing factors of CO 2 concentration change in mainland China. Firstly, based on the collected data, we extracted the annual and monthly average data of corresponding products with provincial administrative regions as the unit. Then, the correlation analysis of nine driving factors and CO 2 concentration in Chinese provinces obtained by preprocessing was conducted. Based on multi-source data fusion, the geographical weighted regression analysis of comprehensive driving factors was carried out to realize the specific analysis of the difference of influence weights in different provinces. Finally, the spatial and temporal variation of CO 2 concentration was studied, and the future annual mean CO 2 concentration variation trend was predicted based on the re-scaling range analysis method(R/S analysis).

Correlation analysis of driving factors of CO2 concentration
To deeply analyze the spatial and temporal distribution of CO 2 , understanding the correlation between CO2 concentration and driving factors is essential. Pearson correlation coefficient, which provides clear test results with high efficiency, can be used to describe the degree of linear correlation between two random variables [6]. In the study, Pearson＇s correlation coefficient was utilized to calculate the correlation between annual correlation coefficients of nine driving factors and CO 2 concentration, which are presented in Fig 1. EVI, solar radiation, temperature, relative humidity, population, GDP and energy consumption showed a positive correlation with CO 2 concentration, among which GDP had the highest correlation coefficient and its promoting effect is the largest, followed by energy consumption, solar radiation, temperature and EVI. Precipitation and soil water content were negatively correlated with CO 2 concentration. According to the statistics in the last decade (2011-2020), 90% of the total CO 2 emissions come from fossil CO 2 emissions and 10% from land use change [7]. Hence, fossil fuel consumption is the most crucial anthropogenic driving force affecting CO 2 concentration. The overall inhibitory effect of the factors (precipitation and soil water content) on CO 2 concentration was not significant, which might be due to the comprehensive effect of multiple natural factors on CO 2 concentration. For example, water content affects vegetation growth and thus inhibits the increase of CO 2 concentration.

Comprehensive evaluation of driving factors of CO2 concentration
The geographically weighted regression(GWR) model is a local spatial regression approach that changes the parameter estimation results with different spatial and quantifies the heterogeneity or non-stationary characteristics of spatial data relations [8]. We can use the GWR model to explore the spatial heterogeneity of the combined drivers of CO 2 concentration in 31 provinces.
To better evaluate the driving mechanism of CO 2 concentration change, the driving factors were divided into five categories. Population, GDP and energy consumption were integrated by the normalization process and expressed as economic indicators. Surface water, relative humidity and precipitation were fused and expressed as humidity index in the same way. The mean values of economy, EVI, surface radiate, temperature and humidity in each province from 2015 to 2019 were analyzed by geographically weighted regression with the mean values of CO 2 concentration in each province.
Spatial autocorrelation analysis was conducted on the standardized residuals obtained by regression, and the results were shown in Tab 1. The Residual Squares of the GWR result (Tab 1) is 0.2770 and R2 Adjusted is 0.6330, indicating that the model can explain 63% of the spatial distribution of CO 2 concentration, and the GWR model has a good fitting effect. Meanwhile, At the same time, the spatial autocorrelation analysis of the GWR standardized residual showed that the p-value (Tab 1) was greater than 0.1, which indicated that the standardized residual was in a random distribution, and the GWR model of mathematical analysis could be applied to the comprehensive evaluation analysis. Regression coefficients of EVI(a), net surface reflection (b), humidity (c), temperature (d), and economy (e) were plotted in Fig2, which further analyses the relationship between the driving factors and the spatial distribution of CO 2 concentration.  Fig2(c) reveals that the effect of humidity on CO 2 concentration is negatively correlated in northern mainland China and positively correlated in southern mainland China, with the most significant correlation in southwest China. Therefore, humidity has a positive effect on CO 2 concentrations in most regions, increasing regional CO 2 concentrations. Regarding temperature (Fig2(d)), previous research has shown that water affects the globally integrated annual net ecosystem exchange in localized regional settings (Martin Jung et al.). In the dry and cold highland regions of the west, rising and dry temperatures limit vegetation growth, reduce carbon emissions and lower regional CO 2 concentrations, while in the hilly, humid and hot regions, increased atmospheric CO 2 concentrations accelerate warming and warmer temperatures cause more CO 2 to be released to the atmosphere from organic matter in the soil, increasing regional CO 2 concentrations. Fig2(e) demonstrates that the effect of human economic activities on CO 2 concentration is negatively correlated in the western part of China and positively correlated in the middle and eastern parts, with the human economic factor not being the dominant one in the sparsely populated western region. These results reveal that the different drivers of spatial heterogeneity exhibit spatial instability. The largest growth rate of 0.991% per annum was recorded in TAR and the smallest was in Shandong with 0.608% per annum. The annual growth rate of the overall spatial distribution is characterized by low in North and Central China and high in the West and Inner Mongolia Autonomous Region, according to the histogram of the annual growth rate, except for four provinces, namely Gansu, Qinghai, Sichuan and Heilongjiang, the annual growth rate of the remaining provinces has gradually decreased and stabilized since 2015. To further analyze the trends in CO 2 concentration, stability analysis of the spatial variation of CO 2 concentration growth in mainland China based on R/S analysis. The R/S analysis is a statistical technique proposed by the British hydrologist Harold Edwin Hurst for analyzing time series trends and predicting future trends [9]. The results (Fig4) show that the mean Hurst value for mainland China is 0.7462, with a very low proportion of inverse persistent sequences, mostly in northwestern China and Yunnan and Guizhou provinces, where persistent sequences are dominant. The results of the quantitative R/S stability show that parts of northern, central and southern China have a relatively strong trend persistence, implying that the region will continue to grow at a stable rate. In contrast, CO 2 concentrations in Northeast China and East China are increasing more slowly and with a stronger trend of persistence. Tibet is a sparsely populated region with little human activity, but in recent years its economy and industrial structure have improved compared to previous years, resulting in a rapid and sustained increase in CO 2 concentrations. In summary, the overall trend of CO 2 concentration in mainland China is strong and persistent. Focusing on improving carbon emission policies in provinces with high CO 2 concentration growth rates and significant persistence will facilitate the achievement of carbon reduction targets.

Conclusions
This study examines the spatial heterogeneity of CO 2 concentration trends and their drivers in mainland China from 2015-2019. The main findings are as follows: (1) A combined assessment of driving mechanisms reveals significant regional differences in the effects of combined variables on CO 2 concentrations, with environmental and locational factors leading to varying facilitative inhibitory effects of driving variables.
(2) Temporally, CO 2 concentration in mainland China exhibits a monthly rotation, while spatially, high growth rates cluster in adjacent regions. We conclude that CO 2 concentration in mainland China will continue to steadily increase for some time in the future.
This study provides a scientific basis for low-carbon policies in different regions by analyzing the spatiotemporal characteristics and driving factors of CO2 concentration in mainland China. However, due to data limitations, the spatiotemporal variations in driving factors at different time scales were not explored. Future research should conduct a comprehensive investigation of the spatiotemporal heterogeneity of CO2 concentration driving factors from both temporal and spatial perspectives to facilitate the development of diverse carbon emission policies in various regions and the formulation of effective carbon emission management policies.