Spatial and Temporal Variations of Atmospheric CO 2 Concentration in China and Its Influencing Factors

: Over the past few decades, concentrations of carbon dioxide (CO 2 ), a key greenhouse gas, have risen at a global rate of approximately 2 ppm/a. China is the largest CO 2 emitter and is the principle contributor to the increase in global CO 2 levels. Based on a satellite-retrieved atmospheric carbon dioxide column average dry air mixing ratio (XCO 2 ) dataset, derived from the greenhouse gas observation satellite (GOSAT), this paper evaluates the spatial and temporal variations of XCO 2 characteristics in China during 2009–2016. Moreover, the factors influencing changes in XCO 2 were investigated. Results showed XCO 2 concentrations in China increased at an average rate of 2.28 ppm/a, with significant annual seasonal variations of 6.78 ppm. The rate of change of XCO 2 was greater in south China compared to other regions across China, with clear differences in seasonality. Seasonal variations in XCO 2 concentrations across China were generally controlled by vegetation dynamics, characterized by the Normalized Difference Vegetation Index (NDVI). However, driving factors exhibited spatial variations. In particular, a distinct belt (northeast–southwest) with a significant negative correlation ( r < − 0.75) between XCO 2 and NDVI was observed. Furthermore, in north China, human emissions were identified as the dominant influencing factor of total XCO 2 variations ( r > 0.65), with forest fires taking first place in southwest China ( r > 0.47). Our results in this study can provide us with a potential way to better understand the spatiotemporal changes of CO 2 concentration in China with NDVI, human activity and biomass burning, and could have an enlightening effect on slowing the growth of CO 2 concentration in China.


Introduction
Atmospheric carbon dioxide (CO2) is an important element of the global carbon cycle and a major greenhouse gas due to its extensive contributes to global warming. CO2 concentrations have increased by more than 40% since the Industrial Revolution [1]. In addition, a warmer climate with adverse consequences, such as rising sea levels and an increase in extreme weather conditions, mean that further increases in CO2 levels are expected [2]. More specifically, according to the National Oceanic and Atmospheric Administration (NOAA), the global CO2 concentration increased from 368.84 ppm in 2000 to 402.86 ppm in 2016, with an average growth rate of 2.06 ppm/a [3]. Since 2006, China has been the world's largest emitter of CO2 [4,5]. The annual growth rate of atmospheric XCO2 concentration reached 2.16 ppm/a in China, and the monthly average concentration of CO2 exceeded the global average by 1.63 ppm during 2006-2016. For example, in 2016, China released approximately 1.02 × 10 4 Mt CO2 into the atmosphere, contributing to 28% of total global carbon emissions [6]. Furthermore, CO2 emissions in China exceeded the total emissions of the United States, the European Union and Japan in 2016 [6]. Therefore, the CO2 emissions in China have a significant impact on global atmospheric CO2 concentrations.
Traditionally, the monitoring of atmospheric CO2 concentrations generally relies on measurements from ground stations distributed across the globe [7][8][9]. However, the development of satellite remote sensing technology allows for the real-time observation and mapping of the spatial and temporal distribution of global atmospheric CO2 concentrations [10,11]. For example, the widely used SCanning Imaging Absorption SpectroMeter for Atmospheric CHartographY (SCIAMACHY) datasets, developed by the European Space Agency (ESA), provide global spatial and temporal distributions of atmospheric trace gas concentrations, and are commonly used in global carbon source/sink simulations and atmospheric transport models [12][13][14]. Satellite observations effectively increase the breadth of observation data and save time costs. Therefore, satellite remote sensing inversion has become the dominant quantitative observation approach for atmospheric greenhouse gases [15][16][17][18].
The world's first greenhouse gas observation satellite (GOSAT), launched in 2009 by the Japan Aerospace Exploration Agency (JAXA), the National Institute for Environmental Studies (NIES) and the Ministry of the Environment (MOE), is specifically designed to record atmospheric greenhouse gas column average dry air mixing ratios (XCO2 and XCH4) [19][20][21]. The GOSAT satellite uses solar radiation in the near-infrared/shortwave-infrared (NIR/SWIR) regions and variations in the thermal infrared (TIR) bands [22,23] to determine the target mixing ratios. In particular, SWIR observations are sensitive to the CO2 abundance from the surface to the top of the troposphere, while TIR observations are sensitive to CO2 in the upper and middle troposphere, particularly the CO2 concentration profile [24,25]. GOSAT provides global coverage and multi-year datasets that have been used in multiple applications, including atmospheric simulations, ecosystem flux retrievals and regional environmental quality assessments [26][27][28][29]. Furthermore, GOSAT provides a large number of atmospheric data products, including XCO2 concentration data, which allows for the determination of spatial and temporal trends in global atmospheric XCO2 concentrations, as well as atmospheric XCO2 concentration variations and potential carbon source-sink distributions [30][31][32]. Some scholars have done meaningful related research in this regard [25,29,33]. However, the study periods or spatial scales are limited [34,35]. In addition, most scholars focus their research on accuracy/uncertainty analysis [5,36,37], and some studies compared the difference of CO2 concentration in the same area from different satellites [31,37] or investigated the effect of a single carbon source/sink on the concentration of CO2; for example, the studies of CO2 emissions from anthropogenic activities [32,38], the impact of vegetable absorption [14,39] and the estimated CO2 emissions from forest fires [10,40]. However, these studies lack comprehensive consideration of multiple factors.
As a major greenhouse gas emitter, long-term serial observations of atmospheric CO2 concentrations in China have become very important. Spatial and temporal distribution differences based on long-term sequence changes can help us to understand the underlying characteristics of atmospheric XCO2 concentration changes in China. In this study, we interpolated GOSAT Level-2 CO2 data from 2009 to 2016 to obtain a monthly CO2 concentration dataset in China. Then, based on the results of its changing characteristics, the three main influencing factors of vegetation, anthropogenic activities and biomass burning are combined to explain the changing characteristics of atmospheric CO2 concentration in mainland China. The current study aims to investigate the spatial and temporal characteristics of XCO2 concentrations and the influencing factors of XCO2 variations in China during 2009-2016. Our results in this study can provide us with a potential way to better understand how the spatiotemporal changes of CO2 concentration in China with NDVI, human activity and biomass burning. It could also have an enlightening effect on slowing the growth of CO2 concentration in China.

CO2 Data
The GOSAT satellite was successfully launched by JAXA with a sun-synchronous orbit in Japan 2009. It has a spatial resolution of 10.5 km × 10.5 km, transits at 13:00 local time and has a replay period of 3 days [41,42]. There are two main sensors in the GOSAT, the Fourier Transform Spectrometer (FTS) and the Cloud Aerosol Imager (CAI), measuring atmospheric greenhouse gases, clouds and radiation, respectively. The FTS sensor can observe the average column concentration distribution of CO2 and CH4 by analyzing the atmospheric gas composition derived from spectral information of the atmosphere. GOSAT data products are widely used in the detection of fire emissions and human activity [43][44][45].
We selected monthly data from TANSO FTS SWIR based on the NIES Level-2 V02.72 algorithm during April 2009 to December 2016. However, due to the observation orbit of GOSAT, data was unavailable for each pan-winter (November to March), resulting in a large number of missing values for high latitudes across the globe. Therefore, we analyzed two conditions of the whole year, as well as the vegetation growing season during 2009-2016. Furthermore, we compared the corresponding satellite data with site observation datasets from atmospheric background stations (Mt. Waliguan NDVI quantifies vegetation by measuring the difference between the near-infrared and red regions of the electromagnetic spectrum, which can indicate the state of vegetation growth and dynamics over time. Vegetation is able to reduce atmospheric CO2 concentrations via photosynthesis. NDVI values close to 1 (0) denote vigorous (sparse) vegetation growth. Hence, NDVI indirectly indicates the photosynthetic capacity of vegetation, which has a strong impact on atmospheric CO2 concentrations [43,46]. Therefore, we used monthly Moderate-resolution Imaging Spectroradiometer (MODIS) NDVI Level-3 data from EOS (NASA) to represent the carbon sink.
ODIAC is a 1 × 1 km monthly global high-resolution gridded fossil fuel CO2 emissions data product (unit: g CO2/m 2 /d) [47]. It can effectively represent the amount of CO2 emitted by human activity to the atmosphere on a monthly basis and includes emissions from thermal power generation and cement manufacturing. Fossil fuel CO2 emission datasets from 2000 to 2017 was selected in this study.
GFED quantifies global trace gases and aerosol emissions from global biomass burning [48]. The CO2 emissions from biomass burning are determined via satellite-retrieved burned area, biogeochemical model-simulated fuel loads, moisture-regulated combustion factors, and land coverbased emission factors (unit: g CO2/m 2 /month) [49]. The latest version of GFED (v4.1s) includes small fire emissions that were missed in previous datasets, as well as daily resolution updates. In the current study, we derived the monthly fire CO2 emissions from different types of biomass burning with a 0.25° ×0.25° grid size from 2009 to 2016. These were subsequently linearly resampled to a 0.1°×0.1° grid size for consistency with the above datasets.

Standard Deviation Ellipse Analysis
Standard deviation ellipse (SDE) analysis is commonly used to analyze spatial distribution directional characteristics with time. The size of the ellipse reflects the concentration of the overall elements of the spatial pattern, while the declination (long semi-axis) reflects the dominant direction of the spatial pattern. Using years of standard deviation statistics enables us to understand average center, major/minor axes and azimuths trends of the discrete point set in time series [50,51]. Therefore, the standard deviation ellipse can express the main distribution direction of a set of points and the degree of dispersion in all directions. These two factors are usually used to describe the overall characteristics of a geospatial distribution [51]. In the current study, we applied SDE analysis results of the annual mean XCO2 concentrations in China from 2009 to 2016. By plotting the abstract images of the elliptical motion trajectories of China's annual average XCO2 concentration, the dynamic changes of these ellipses over time are obtained, which cannot be demonstrated using a pixel-level analysis. Therefore, we used an SDE analysis of the annual mean XCO2 concentrations in China from 2009 to 2016.

Pixel-Based Time Series Analysis
In order to investigate the interannual variability of the local spatial non-stationarity of atmospheric XCO2 concentrations over 2009-2016, we used the linear long-term trend (LLT) to quantitatively analyze long-sequence XCO2 concentration images during the growing season. By using LLT, we can comprehensively understand the spatial and temporal changes in atmospheric XCO2 concentrations. The time series of monthly atmospheric XCO2 concentration data was fitted by pixel-by-pixel linear regression. The LLT slope represents XCO2 concentration change trends (k), as it embodies the direction and magnitude of XCO2 changes across time. More specifically, positive (negative) slopes denote increases (decreases) in the XCO2 concentration in the specific pixel, while for slopes close to zero, the pixel is identified as "stable." The LLT slope was calculated as follows: where XCO2 is the grid unit XCO2 concentration, n is the time span and i is the time unit. Figure 1 presents the derived multi-year average atmospheric XCO2 concentrations in most part of China, with high latitudes masked due to absent values in winter. Spatial variations were observed in the eastern and western regions of China, with concentrations gradually decreasing from east to west. The maximum concentration (395.50 ppm) was observed in Shanghai, located in the coastal area of east China, with the lower atmospheric XCO2 concentrations (less than 393.00 ppm) observed over the Tibetan Plateau.

Validation of XCO2 Concentrations via in situ Data
In order to assess the accuracy of GOSAT data, we compared the satellite-derived XCO2 data with that measured from the WLG and LL stations ( Figure 2). The atmospheric environment and components of the WLG station were relatively stable, due to its distance from built-up areas. There was a good agreement between the two datasets, (r = 0.94, RMSE = 2.2 ppm). Similarly, to the WLG station, the LL station is located in the mountains, yet it is influenced by a tropical-subtropical monsoon climate and marine atmospheric environment. The measured CO2 variation range of the LL station is larger than that of the WLG station during 2009-2016 (r = 0.95, RMSE = 2.5 ppm).
Furthermore, based on the derived statistics of each site, the XCO2 concentrations were lower than those from the measured ground data by 0.88% (WLG) and 0.05% (LL), respectively (Table 1). In addition, the WLG station showed more accurate verification results, followed by the LL station. This is attributed to the different measurement scales. More specifically, the station-measured dataset is made up of surface point data, while the GOSAT XCO2 data compromises regional tropospheric column averages at a spatial resolution of 10.5 km × 10.5 km. This observation is similar to Guo et al. (2015), who used MODIS and GOSAT L2 data to model XCO2 concentrations during the 2012-2013 growing seasons and determined an average bias of 2.25 ppm [33]. This indicates that the monthly XCO2 data obtained by interpolation is relatively reliable and can be used for subsequent analysis.

Seasonal XCO2 Variations
The temporal characteristics of atmospheric tropospheric XCO2 concentrations in China exhibited strong seasonal variations and rising trends ( Figure 3). The XCO2 concentrations rose from 387.24 ppm in April 2009 to 404.09 ppm in April 2016, while interannual variations exhibited an increasing trend of 2.28 ppm/a ( Figure 3). The gray regions in Figure 3a represents the data unavailability in high latitudes during parts of the winter period. We de-trended the monthly average XCO2 concentrations in order to investigate monthly variations in XCO2 in China (Figure 3b). During the study period, XCO2 concentrations reached a maximum in spring (Apr-May) and minimum in summer (Jul-Aug), with the average changes of XCO2 amplitude reaching 6.78 ppm/a.  In particular, the spatial distribution of atmospheric XCO2 concentrations was observed to have drastically changed during summer, with concentrations decreasing rapidly with increasing latitude. China's low latitude coastal areas were associated with higher concentrations, while regions close to Mohe, Heilongjiang Province, exhibited low concentrations (only 387.00 ppm). In autumn, concentrations began to increase and the variance in spatial distribution was reduced. However, atmospheric XCO2 concentrations in the densely populated areas of the North China Plain increased rapidly. In winter, atmospheric XCO2 concentrations in eastern China continued to increase, widening the gap between the east and the west. The maximum concentration was observed close to the Henan and Shandong Provinces, while concentrations in the Ngari Prefecture and Qinghai-Tibet Plateaus were similar to those in spring (approximately 394.00 ppm). In terms of spatial distribution, atmospheric XCO2 concentrations dropped drastically from spring to summer, particularly in the three northeastern provinces, falling to −8.0 ppm (Figure 4). In addition, from summer to autumn, the concentrations in northeast and north China were observed to rise rapidly, with an increase of approximately 4.0 ppm. There was a slight increasing trend in concentrations in the western high latitude, while other parts of China exhibited relatively stable values. From autumn to winter, there was a visible growth in XCO2 concentrations in China, particularly in the intersection of Shandong and Henan Provinces in the central and eastern parts of China, as well as in Chengdu and Chongqing. The growth rate of atmospheric XCO2 slowed down from winter to spring, with most of China exhibiting a relatively stable trend. However, an increase of 1-2 ppm was observed in most parts of China. In addition, during this period, the concentration of atmosphere XCO2 reached a maximum, with a multi-year average of 396.44 ppm.

Interannual Variations in XCO2
Due to the temporal coverage of the GOSAT observation data, we focused on atmospheric XCO2 concentrations during the vegetation growing season (GS, April-October each year). The area was divided into seven sub-regions based on seasonal variation characteristics (  (Table 2). Overall, the multi-year average concentration of XCO2 in growing season reached 392.28 ppm in China.   During the growing season, large differences were detected between the XCO2 variations (∆XCO2) in sub-regions per year (Table 2). In China, the ∆XCO2 values exhibited a gradual downward trend (y=−0.04x+2.61) during the growing seasons of 2009-2016 ( Figure 6). However, at high latitudes, the average ∆XCO2 rates in NWC and NEC was rapidly reduced (−0.1 ppm/a). The second is North China, where the average ∆XCO2 decreased at a rate of −0.05 ppm/a. Furthermore, ∆XCO2 values in SWC were observed to fluctuate by 2-3 ppm/a, with relatively stable growth rates. However, in CC and EC, atmospheric XCO2 levels exhibited a small upward trend (+0.04 ppm/a). The fastest annual growth trend was observed in SC, with the average ∆XCO2 rate reaching + 0.19 ppm/a and concentrations exhibiting an accelerated upward trend. ppm, 4.54 ppm, 4.63 ppm and 4.22 ppm, respectively. In particular, the increase of XCO2 concentration in the Jilin Province was 142.9% of the national average. Furthermore, the average increase in NC also reached 3.7 ppm in 2010. In March 2010, the most severe continuous low temperature hazard since 1977 took place in northeastern China [52]. As a result, vegetation growth was slow in the spring of March-April, and thus NDVI values were only 59% of the multi-year average. This, in turn, affected atmospheric XCO2 concentrations. Essentially, the large-scale and long-term persistence of low temperatures had a serious negative impact on the growth of vegetation in spring 2010, leading to restrictions in the ability of vegetation to absorb CO2 from the atmosphere. Therefore, in 2010, the median center of China's atmospheric XCO2 concentrations shifted in the northeast.
In September 2013, China issued an action plan for the prevention and control of atmospheric pollution, which effectively controlled the production of high-energy consumption, high polluting and high emission enterprises in the Beijing-Tianjin-Hebei (BTH), Yangtze River Delta (YRD) and Pearl River Delta (PRD) regions. This was also an effective mitigation of atmospheric CO2 emissions. According to the ODIAC CO2 emissions dataset, fossil fuel combustion from human activities prior to 2013 exhibited a significant period of rapid growth (+0.13 g CO2/m 2 /d), while the growth rate decreased significantly (+0.04 g CO2/m 2 /d) following 2013. Furthermore, the most severe El Niño event since 2000 occurred in 2014-2016, which impacted precipitation and average temperatures in southeastern China. In particular, in 2015, precipitation in north China was reduced, with other parts of China experiencing severe drought. During the 2015/16 winter, average precipitation levels in south China (Guangdong, Guangxi and Hainan Provinces) reached historic values, exceeding those in perennial years by more than 1.6 times [53]. At the same time, the El Niño phenomenon also had a large impact on the variability of the atmosphere [54][55][56]. Thus, a series of related issues resulting from policy regulation and the El Niño phenomenon moved the median center of atmospheric XCO2 concentrations to the east of China.

Major Influencing Factors of XCO2 in China
NDVI was generally observed as an important carbon sink affecting the long-term spatial distribution of atmospheric XCO2 in China (r = −0.83, p < 0.05). This is consistent with results derived from SCIAMACHY data in China during 2003-2011 [14]. There was an obvious negative correlation belt from northeast to southwest China (r < −0.75), particularly in northeast China (Figure 7). This can control the absorption of carbon dioxide in the atmosphere through the photosynthesis of green vegetation [57,58]. Our results indicate that variations in atmospheric XCO2 concentration were mainly affected by changes in vegetation cover. However, the effect of NDVI on XCO2 concentrations varied spatially. In addition, a weak correlation (r < 0.2) was observed in the low latitudes of SWC and some areas of EC. Due to the low latitudes of Yunnan Province in the subtropical region, vegetation cover is present all year round. Consequently, the annual average NDVI was observed as 0.69; 176.9% of the national average. In eastern China, the vegetation growth was severely disturbed by anthropogenic activities. The majority of the agricultural waste formed by winter crops is incinerated in the open, resulting in June NDVI values of only 57% of those in May and July. NWC compromises a large area of bare land with a monthly average NDVI of less than 0.25. Thus, in this region, the influence of vegetation on atmospheric XCO2 concentrations is limited.
Emissions from human activity are the main contributors to CO2 emissions [47,58]. As demonstrated in Figure 7 (ODIAC), most regions in China exhibited a highly positive correlation with the changes of XCO2, particularly in the BTH, YRD, PRD and Sichuan-Chongqing economically developed areas. This shows that the intensity of human activity has an important impact on the spatial and temporal variation of XCO2 concentrations. However, the correlation between the XCO2 and Shandong and Henan Provinces, which have high population densities, were weak, whilst other regions in these two provinces exhibited negative correlations (Figure 7). This may be a result of the industrial structure and economic level of these regions. The CO2 emissions from fossil fuel combustion in these two provinces (81.39 g CO2/m 2 /day) was only 19.45% of those in Shanghai (418.38 g CO2/m 2 /day). Biomass burning emissions are a key source of atmospheric CO2 [45,59]. Based on the biomass burning emission inventory (GFEDv4.1s) and derived CO2 emissions in China (Figure 7), the annual average CO2 emissions from biomass burning in China was determined as 76.9 Tg, with emissions distributed in Heilongjiang Province (14.4 Tg/a) and the three provinces along the southeastern coast (13.8 Tg/a). Yunnan Province in SWC (7.78 Tg/a) exhibited strong correlations with forest fires (55%) and the burning of savanna, grassland and shrublands (36.7%). The correlations of Heilongjiang Province in northeast China were slightly weak, yet the largest annual average emissions in China were associated with this region, and among them, agricultural straw waste burning was the main source of emissions (76.3% of total biomass burning emissions). In the eastern coastal areas, biomass combustion emissions were more evident, and the main combustion type was agricultural waste combustion (51.5%). However, the correlation between biomass burning emission of this region with variations in atmospheric XCO2 concentration was low (r = 0.13). In particular, in the eastern part of China, several areas exhibit negative correlations (Figure 7, GFED2009-2016). In addition, it has been observed that this area is also an area of high human activity (Figure 7, ODIAC). The same situation also occurred in the Pearl River Delta region, where human activity is strong in low and middle latitudes. Biomass burning emissions were greatly affected by human activity, and mainly occurred during spring. This spatial distribution trend determined here is similar to Qiu et al. [48].
The partial correlation and regression coefficients of the standardized multiple regression equations reveal the influence of the NDVI, ODIAC and GFED on atmospheric XCO2 concentrations across different regions (Figure 8). NDVI (vegetation photosynthesis) had the greatest influence on the spatial pattern of atmospheric XCO2 concentrations. Although photosynthesis plays a key role in various regions of China, its inhibition of XCO2 concentrations is limited. Fossil fuel combustion, considered as the leading cause of the rapid growth in atmospheric XCO2 concentrations in China, was the main carbon source of CO2, particularly in medium and large cities (Figure 7). The regression coefficients almost reached 0.9 in sub-regions, with that of EC almost reaching 1.0. Biomass burning also contributed to XCO2 concentrations in China, with contributions exhibiting significant regional heterogeneity, particularly in Yunnan Province in the southwest and Heilongjiang Province in the northeast (Figure 8b).

Contribution of XCO2 in China
In summary, we used NDVI to represent carbon sinks, and human activities and fire emissions to represent carbon sources. We then quantified the effects of the fires, human activity and NDVI on the temporal changes in XCO2 concentration per pixel during 2009-2016 ( Figure 9). We found that NDVI dominated the whole study area, denoted by temporal XCO2 variations, particularly in southwest and northwest China. A strong correlation was observed between NDVI and XCO2, indicating that NDVI was the major influencing factor among the three factors in most parts of southwestern and northwestern China. However, the contribution rate of the three influencing factors in the other five regions exhibit significant seasonal variations, particularly in the east of Huline (red line in Figure 5). More specifically, the distribution of human activity and fire emissions were evaluated over the entire growing season. The areas where human activity (e.g., fossil fuel combustion) affected atmospheric XCO2 concentrations were generally located in Shandong, Jiangsu and Henan Provinces, followed by Xi'an, Chengdu and Kunming. In the southeastern coastal areas and main inland urban areas, although fire emissions were high and human activity was intense, they were not focal influencing factors in these areas. Fire emissions predominantly occurred in Yunnan Province, and were also detected close to Henan Province and Chengdu.
The contribution of each factor varied across seasons. In spring, where vegetation respiration is stronger than photosynthesis, human activity and fire emissions contributed more to the atmospheric XCO2 concentration, particularly in north and northeast China. In addition, fire emissions were principally detected in southeastern China, where a large continuous fire occurred in the north of eastern China, and a large number of sporadic fire spots appeared in southern China. In summer, photosynthesis began to reach its maximum, and the impact of NDVI thus peaked. As a result, the atmospheric CO2 concentration in China rapidly decreased to the lowest value, with an average of 389.53 ppm in this study period. These three main influencing factors did not show a clear dominant role in Yunnan Province and central East China. However, human activity still played an important in major urban areas, particularly in the south-central region of China. In autumn, the impact of emissions from human activity was more evident, while the proportion of fire emissions in southeastern China also begun to increase rapidly. Our results demonstrate that in east China, there was a large area affected by the three major driving factors. The shape of this area is similar to the equivalent area in spring and is also one of the most densely populated areas in China. The contribution rate of each influencing factor experienced seasonal changes throughout 2009-2016. Throughout the study period, NDVI was a key factor in controlling atmospheric CO2 concentration levels. The XCO2 concentration in most areas of Henan Province, Chengdu and Kunming were affected by human activity; biomass burning emissions have a significant impact on CO2 concentration levels in Yunnan Province. In terms of seasonal changes, the impact of human activity in spring at low latitudes rapidly increased, and the range of effects of biomass combustion on XCO2 concentration increased. Furthermore, the contribution of biomass burning increased in southern China. In the winter, the impact of NDVI was further weakened. The emissions from biomass burning were concentrated in low latitudes below 25°N, particularly in Guangdong Province and Fujian Province. In addition to the impacts of NDVI, atmospheric XCO2 concentrations were significantly affected by human activity. Fire emissions were observed to contribute to atmospheric XCO2 concentrations, yet there were significant seasonal differences in the spatial distribution of its impacts.

Conclusions
In the current study, the seasonal cycle and inter-annual variations of XCO2 over China based on a long-term (2009-2016) GOSAT dataset were analyzed. The main results can be summarized as follows.
(1) Atmospheric XCO2 concentrations exhibited clear seasonal changes, with the highest concentrations observed in spring and the lowest in summer; (2) during 2009-2016, the average growth rate of XCO2 concentration over China reached 2.28 ppm/a, and the average amplitude of seasonal variation was 6.78 ppm/a. In the growing season, the average XCO2 concentration was determined as 392.28 ppm; (3) the regions with high XCO2 concentrations were mostly distributed in southeast coastal areas, particularly the Yangtze River Delta region. In addition, low concentrations of XCO2 were mainly observed in northern China; and (4) the contribution of each of the three driving factors exhibited significant spatial and temporal distribution heterogeneity. Atmospheric XCO2 concentrations over China were mainly controlled by NDVI across time, with a significant negative correlation belt (r < −0.75) from northeast to southwest China. In addition, human activity was the main source of CO2 emissions, and the biomass burning emissions also show an important impact on changes in atmospheric XCO2 concentrations in northeast China in autumn and southwest China in spring.
The results of this study can provide us with a potential way to better understand the spatiotemporal changes of CO2 concentration in China with NDVI, human activity and biomass burning, and could also have an enlightening effect on slowing the growth of CO2 concentration in China. However, our study determined the relative contribution rates of the main influencing factors of CO2, without quantifying the individual contributions of each factor. Therefore, a further study will focus on quantitatively attributing the effects of these factors, including the effects of atmospheric transport at various scales as well.