Accuracy Verification of Satellite Products and Temporal and Spatial Distribution Analysis and Prediction of the CH₄ Concentration in China

Abstract

In this study, the spatiotemporal variations in CH₄ concentrations in China from 2003 to 2021 are investigated, and their trends are forecasted over the next decade. Based on the seventh edition standard product retrieved by the atmospheric infrared detector (AIRS) at an altitude of 500 hPa, we verified monthly CH₄ products using observational data provided by the World Data Center for Greenhouse Gases (WDCGG) from six ground stations in and around China. The correlation coefficients (R values) between the two data sets ranged from 0.68 to 0.92, signifying the ability of AIRS inversion data to represent temporal and spatial changes in surface CH₄ concentrations. Additionally, China was classified into three regions (steps) based on terrain, and the changes in CH₄ concentrations were assessed from three perspectives: spatial distribution, interannual variation, and seasonal variation. The results revealed that the CH₄ concentration decreased with elevation along a topographic gradient, with high-value areas located in the first and second steps, corresponding to the eastern Qinghai–Tibet Plateau, northern Xinjiang Uygur Autonomous Region, and Inner Mongolia Autonomous Region. Over 19 years, the average increase in CH₄ concentration has ranged from 65 to 175 ppb. In addition, the CH₄ concentrations were higher during summer and autumn and lower during spring and winter. Finally, a SARIMA model was used to predict the near-surface CH₄ concentration trend in China over the next ten years, which indicated a continued seasonal increase.

1. Introduction

As an important greenhouse gas, CH₄ contributes to approximately one-third of global warming [1,2]. Since the industrial revolution, the CH₄ concentration in the atmosphere has rapidly increased. In the sixth assessment report of the Intergovernmental Panel on Climate Change (IPCC), the CH₄ concentration was reported to have increased by nearly 262% over preindustrial levels. Approximately 0.5 °C of the 1.1 °C rise over previous levels was attributable to CH₄ emissions [3]. CH₄ sources are classified as either anthropogenic or natural sources, depending on whether they directly involve human activities. Natural sources of CH₄ largely include wetlands, termites, and oceans [4]. Among them, wetlands constitute the greatest single source of CH₄, accounting for 40–50% of total CH₄ emissions [5,6,7,8]. Anthropogenic sources primarily include energy consumption, enteric fermentation in ruminants, landfills, and biomass burning [9,10]. China set a carbon neutrality goal in 2020 that included a carbon emissions peak in 2030 and aimed for net zero carbon emissions by 2060. Therefore, understanding the spatial and temporal distributions of the atmospheric CH₄ concentration in China is very important for achieving carbon neutrality goals [11,12].

Atmospheric CH₄ monitoring mainly includes ground, space-based, and satellite remote sensing monitoring activities [13]. Ground monitoring has relatively high accuracy and covers a relatively long detection period. However, due to the relatively small number of ground stations and their uneven distribution, the ground monitoring scope is limited. The cost of using space-based monitoring, such as aircraft sampling, for research purposes, is too high [14,15]. Satellite remote sensing monitoring can provide large-area, all-day, all-weather data, with a relatively large measurement range and at relatively low cost. Atmospheric CH₄ monitoring satellites mainly include the Greenhouse Gas Observation Satellite (GOSAT) launched by the Japan Aeronautics and Space Administration [16], the Scanning Imaging Absorption Spectrometer (SCIAMACHY), the Environmental Satellite (ENVISAT) launched by the European Space Agency [17], and the Atmospheric Infrared Sounder (AIRS) onboard the Aqua satellite [18]. Compared to the SCIAMACHY and GOSAT, the AIRS sensor began providing CH₄ concentration information in 2002 and also has the ability to remove cloud effects and invert partly cloudy scenes [19,20,21]. The AIRS retrieval data have been used to analyze the distribution, evolution, and transmission of the CH₄ concentration in the atmosphere in China. For example, Wang [22] employed AIRS products to explore the temporal and spatial distribution characteristics of the near-surface CH₄ concentration in China from 2003 to 2013 from three perspectives, namely, region, seasonal variation, and interannual variation, and they concluded that the lowest CH₄ concentration occurred in Tibet. The high-value areas were located in northern Xinjiang and northern Heilongjiang. The concentration was high in summer and low in winter. The seasonal variation was obvious, and the interannual variation basically exhibited an increasing trend. Zhang [23] analyzed the temporal and spatial distribution characteristics of the CH₄ concentration globally and in East Asia from December 2002 to November 2016 using AIRS observation data, and found that the increase in the CH₄ concentration was obvious on the Qinghai–Tibet Plateau and in Northeast China, and that its distribution revealed obvious seasonal changes. In the Northern Hemisphere, the CH₄ concentration in summer was higher than that in winter in most areas. Wu [24] used AIRS data to detect the temporal and spatial changes in atmospheric CH₄ in China from 2002 to 2016 and found that the concentration values in northern Xinjiang, northeastern Inner Mongolia, and Heilongjiang Province were relatively high and indicated obvious seasonal changes. The CH₄ concentration was the highest in summer, followed by autumn, winter, and spring. In addition, the CH₄ concentration exhibited a significant increasing trend in all parts of China. In this study, the AIRS sensor was used for spatiotemporal analysis of the monthly average CH₄ concentration in China.

To prevent temperature rise and global warming, CH₄ emissions must be reduced, and a greater understanding of future CH₄ emissions may be one of the most critical steps in planning to reduce these emissions [25]. To date, researchers have adopted different methods and models to predict greenhouse gas emissions. Li et al. [26] used the autoregressive integral moving average (ARIMA) model to predict CO₂ emissions in China in 2030 and formulated practical suggestions for effectively reducing the emission intensity. Javanmard et al. [27], based on relevant data on the greenhouse gas emission rate in Iran from 1990 to 2018, used artificial neural network (ANN), autoregressive (AR), ARIMA, and other models to predict CO₂, N₂O, and CH₄ emissions; they found (through model evaluation indicators) that the prediction accuracy of the ARIMA model was higher than that of the other models. Jiang et al. [28] used ARIMA, random forest (RF), and long short-term memory (LSTM) models to predict the emissions of five gases (SO_x, NO₂, NH₃, H₂S, and CH₄) and found that the ARIMA model predicted gas emissions more accurately than the other two learning-based methods. Niu et al. [29] combined machine learning and algorithms to predict carbon emissions in China in 2030. Pakr et al. [30] used the autoregressive part of the ARIMA model combined with the support vector regression technique (ARIMA-SVR) to predict the air pollution concentration in Iran over the next 35 years.

As satellite observations for atmospheric composition research in China began later, large-scale, long-time series analyses and predictions of CH₄ concentration change characteristics in China have not been conducted. Therefore, the aim of this paper was to analyze the temporal and spatial changes in CH₄ concentrations in China from 2003 to 2021 based on AIRS-retrieved monthly CH₄ concentration data. First, to determine whether the AIRS-retrieved CH₄ concentration is correlated with the ground-observed CH₄ concentrations, the AIRS CH₄ products were compared with the ground-observed CH₄ concentration. Second, the spatial distribution of CH₄ and its age and seasonal variations from January 2003 to December 2021 were analyzed, and finally, the SARIMA model was used to predict the CH₄ concentration for the next 10 years. China has always been an area of high CH₄ emissions, so studying the distribution and change characteristics of CH₄ concentrations in China is critical to adapting to regional climate change and formulating a reasonable emission reduction plan [31].

2. Materials and Methods

2.1. AIRS V7 CH₄ Product

The satellite data used in this study were obtained from the AIRS sensor [23] onboard the Aqua satellite. The AIRS sensor was launched from the National Aeronautics and Space Administration (NASA) EOS/Aqua platform in May 2002. The sensor comprises 2378 channels and includes three band ranges, namely, longwave infrared (LWIR) (8.8–15.4 μm), medium-wave infrared (MWIR) (6.20–8.22 μm), and shortwave infrared (SWIR) (3.74–4.61 μm), covering the CH₄ detection band at 7.66 μm [23]. The AIRS sensor orbits Earth twice a day, and the error of the CH₄ products retrieved by the AIRS sensor is basically controlled between 1.4% and 0.1%. The root mean square error (RMSE) varies between 0.5% and 1.6% [24,32]. The AIRS data are stratified according to atmospheric pressure characteristics, and the highest sensitivity of the AIRS instrument is attained at altitudes of 200~500 hPa in the Northern Hemisphere [33,34]. Due to the uneven distribution of the land surface in China due to the different altitudes, there are no data for low-level areas such as the Qinghai–Tibet Plateau and the Yunnan–Guizhou Plateau. Because of its continuous coverage of CH₄ data across China and data on surface emissions, the seventh edition (V7) of the L3 monthly average CH₄ product retrieved by the AIRS sensor was used in this study, covering a total of 228 months from January 2003 to December 2021 (AIRS monthly average data at 500 hPa after systematic processing; URL: https://airs.jpl.nasa.gov/data/get_data, accessed on 12 April 2022), with a horizontal resolution of 1° × 1°.

2.2. Ground Measurements of CH₄ Concentration

The ground-based observations used in this study were obtained from the World Data Center for Greenhouse Gases (WDCGG) as part of the Global Atmosphere Watch (GAW) program. The GAW program is mainly responsible for long-term observations of greenhouse gases, ozone layer properties, solar radiation, precipitation chemistry, aerosols, reactive gases, persistent organic pollutants, heavy metals, isotopes, and meteorological elements [24,35]. The WDCGG (URL: https://gaw.kishou.go.jp, accessed on 18 April 2022) provides monitoring data from several atmospheric CH₄ monitoring ground stations. In this paper, we selected 6 ground stations in and near China. The locations of the selected ground stations are shown in

Table 1. Ground site information sheet.

Ground Station Location	Coordinates (°)	Time	Altitude (m)
Waliguan (WLG)	(36.28°N, 100.89°E)	2003.01~2020.12	3810
Shangdianzi (SDZ)	(40.65°N, 117.12°E)	2009.09~2015.09	293
Lulin (LLN)	(23.47°N, 120.87°E)	2006.08~2020.12	2862
Minamitorishima (MNM)	(24.29°N, 153.98°E)	2003.01~2021.11	9
Tae-ahn Peninsula (TAP)	(36.73°N, 126.13°E)	2003.01~2020.12	20
Ryori (RYO)	(39.03°N, 141.82°E)	2003.01~2021.11	280

and Figure 1. Here, we used monthly CH₄ concentration data for the corresponding period to compare the ground monitoring data with the AIRS retrieval data. According to the division of climate statistics, March to May is considered spring, June to August is considered summer, September to November is considered autumn and December to February of the following year is considered winter.

2.3. Methodology

As a commonly used time series forecasting analysis method, the ARIMA model has been favored by researchers in recent years. The seasonal autoregressive integrated moving average (SARIMA) model is an extension and improvement of the ARIMA model that considers the seasonality or periodicity of the sequence. Since the CH₄ concentration series used in this paper exhibits obvious seasonality and periodicity, the SARIMA model was adopted to predict the future change trend of the CH₄ concentration in China. The SARIMA model can be expressed as SARIMA(p,d,q)(P,D,Q)s, where p and P denote the order of the autoregressive and seasonal autoregressive parts, respectively, q and Q denote the order of the moving average and seasonal moving average parts, respectively, and s denotes the period [36]. Since the monthly average CH₄ concentration is considered in this study, s is set to 12. Then, the general form of the SARIMA model can be expressed as:

$$\mathcal{\ell }(B)Ø(B^S){\nabla }^d{\nabla }_S^{D}y(t)=\theta (B)\vartheta (B^S)\epsilon (t)$$

(1)

with:

$$Ø(B^S)=1-{Ø}_1-\dots {Ø}_P(B^S)$$

(2)

$$\vartheta (B^S)=1+{\vartheta }_1{B}^S+\dots {\vartheta }_q{B}^S$$

(3)

$${\nabla }_S^D={(1-B^S)}^D$$

(4)

$B$, $\nabla$, $y(t)$, $\mathcal{\ell }(B)$, $Ø(B)$, and $\epsilon (t)$ constitute the ARIMA model with the seasonal index, $Ø(B^S)$ denotes the seasonal regression polynomial, ${Ø}_1$...${Ø}_P$ is a seasonal regression parameter, $\vartheta (B^S)$ is a seasonal moving average polynomial, ${\vartheta }_1$, ${\vartheta }_2$, ${\vartheta }_3$...${\vartheta }_q$ is a seasonal moving average parameter, and ${\nabla }_S^D$ denotes the seasonal difference value in the $D$ series [37].

To use the SARIMA model for CH₄ concentration change trend prediction in China, we first established a time series of the near-surface CH₄ concentration in China. The main steps of SARIMA modeling are as follows:

• First, it is determined whether the CH₄ concentration sequence remains stable. If the sequence is stable, we can proceed to the next step. Conversely, we can obtain a stable sequence through differential processing. The augmented Dickey–Fuller unit root test method is a strict statistical test method that relies on whether a unit root exists in the time series. If the sequence is nonstationary, a unit root occurs. In the case of a stationary sequence, there is no unit root;
• In determining whether the CH₄ concentration sequence is a white noise sequence, the white noise test method is also known as the pure randomness test method. When the data are purely random, the sequence provides no research significance, so it is necessary to perform a white noise test of the sequence to ensure that the data can be used for analysis. If a white noise sequence exists, then it is unsuitable for SARIMA model analysis. Conversely, we can proceed to the next step;
• The order of the model is determined by both the autocorrelation function (ACF) and partial autocorrelation function (PACF), and the minimum information criterion (the Akaike information criterion, hereinafter referred to as AIC) is used to determine the model order (p,d,q) and (P,D,Q)_s;
• Modeling and residual analysis are performed;
• The trained SARIMA model is used to generate forecasts based on the time series data;
• The results are finally evaluated in terms of forecasting accuracy via the mean absolute error (MAE), mean square error (MSE), and RMSE.

3. Results

3.1. Verification of the Accuracy of AIRS Measurement Data

To verify the accuracy of the monthly mean CH₄ concentration in the China region retrieved by the AIRS, this paper uses the data from the six ground stations provided by the WDCGG and the AIRS retrieval data. First, the AIRS retrieval data are synthesized according to the longitude–latitude grid (1° × 1°) to obtain the monthly mean value products, and then the CH₄ concentration data retrieved by AIRS within 100 km of the ground observation stations are compared with the data of the ground observation stations, which spatiotemporally aligns the ground station and satellite data [22].

Figure 2 shows a comparison of the monthly average CH₄ data retrieved by AIRS and the CH₄ data measured by the six ground stations. Among these data comparisons, the highest correlation coefficient was between the MNM station and the AIRS data, which reached a value of 0.92, followed by that of the WLG station and AIRS data (0.87), while the lowest correlation coefficient was obtained between the SDZ station and the AIRS data (0.68). The greatest difference was between the SDZ station measurements and the satellite data, which may have been because the SDZ station is located in the Miyun area of Beijing and is heavily affected by human activities. The RMSEs of the measurements from the six stations relative to the AIRS data varied between 10 and 25, and the RMSE for the MNM station was the lowest. MNM is a volcanic island in the Pacific Ocean in Japan. No one lives on the island, and it is basically undisturbed by human factors. Thus, measurements at this site correlated more closely with the concentration data from the AIRS retrieval data. Figure 2 shows that the RMSE values for the MNM and LLS stations were lower than those for the other four stations. The other four stations are located in relatively densely populated areas, and the CH₄ emissions in densely populated areas are relatively high. The near-surface CH₄ concentration was high but did not cause a high atmospheric concentration because the near-surface CH₄ was transferred to adjacent regions. According to the above data, there was a strong correlation between the monthly mean CH₄ data obtained by AIRS and the measurement data at the six ground stations. Wu et al. [24] compared the CH₄ data retrieved by AIRS to the CH₄ observation data of the ground stations and found that these two data types exhibited satisfactory consistency. Zhang et al. [38] compared different satellite data to ground station observations and determined that the AIRS data attained the highest correlation with ground station observations. Zhang et al. [39] also obtained CH₄ observations on the roof of the National Satellite Meteorological Center in Beijing from January to June 2009 using ground-based Fourier transform infrared (FTIR) hyperspectral data. Via comparison of the two product layers, they found that the error between the two ranged from 0.2% to 1.1%. In summary, the CH₄ data retrieved by the AIRS were basically consistent with the CH₄ observation data at the six ground observation stations, passing the accuracy verification tests. Therefore, AIRS retrieval of CH₄ data can be used to study further the temporal and spatial distribution of CH₄ in China.

3.2. Spatial Variations of CH₄ from AIRS over China

Figure 3 shows the distribution of the average CH₄ concentration in China from January 2003 to December 2021. Overall, the CH₄ concentration showed a distribution with high values in the north and low values in the south. Figure 3 clearly shows that there are some very obvious high-value areas, such as the northwestern part of Xinjiang, the northeastern part of Inner Mongolia, the northwestern part of Heilongjiang Province, and the Sichuan Province. China has a high terrain in the west and a low terrain in the east, and its terrain is roughly stepped, forming three levels of topographic steps. It is common to use the “three steps” concept to summarize the basic terrain characteristics of China [40]. The first step is characterized by high terrain, which is influenced by topography and results in a cold climate and low population density, leading to slow economic development. On the other hand, the third step has flat terrain and is situated near the ocean, resulting in a warm and humid climate, high population density, dense transportation routes, and favorable conditions for economic development, making it the most developed region in China. The second step is in a transitional state. Figure 3 shows that a high concentration of CH₄ is distributed in each step, and there is a significant difference in CH₄ concentrations among the different steps. It is evident that population density, GDP per unit of land area, and land type have a significant impact on the spatial distribution of the CH₄ concentration. Therefore, this study divides the distribution of CH₄ in China into “three steps”.

Among the three regions, the average CH₄ concentration value is highest in the second step at 1848 ppb. The second step encompasses several obvious high-value areas, such as northwestern Xinjiang, which is sparsely populated and contains a relatively suitable ecological environment. However, the area is harsh. No rice is planted in this area, and there are no sources of CH₄ emissions, such as wetlands. However, there are two major oil and gas production bases in the Tarim Basin and Junggar Basin [41], with raw coal processing, natural gas mining, and leakage processes [22]. The reason for the high concentration of CH₄ in the northeastern part of Inner Mongolia is the presence of permafrost in this region. As the greenhouse effect intensifies, the temperature continues to rise, and the permafrost begins to melt, releasing a large amount of CH₄. The first step mainly affects the Qinghai–Tibet Plateau and the western part of Sichuan. The reason for the high CH₄ concentration on the Qinghai–Tibet Plateau may be that this plateau contains the largest area of alpine grassland, the livestock industry is relatively well developed, and there are many alpine glaciers. As temperatures continue to rise, many glaciers are beginning to melt, releasing gases that were previously trapped in the permafrost. Summer and autumn are the most intense periods for biological and permafrost-related CH₄ emissions [42]. The reason for the high CH₄ concentration in western Sichuan is the emission of CH₄ from paddy fields in the Sichuan Basin in summer, leading to a significant increase in CH₄ within a large range around natural sources. Paddy field emissions constitute one of the main sources of CH₄. The high-value areas of the third step mainly occur in northwestern Heilongjiang and at the junction of Inner Mongolia and Heilongjiang. Permafrost persistently exists in these areas, and permafrost can release a large amount of CH₄ upon melting. In winter, when the temperature is typically low, a large amount of coal and natural gas is burned, which is the main cause of the CH₄ emissions, and there is a large wetland forest area in the region, which is a major source of CH₄ [43]. Ding and Wang reported that 67% of CH₄ emissions originated from swamps in Northeast China [24,44]. This is also a reason for the high CH₄ concentration observed in this area.

There are also several obvious low-value areas in China, such as the southern part of Xinjiang and the southwestern part of Tibet. These areas are relatively sparsely populated, occur at higher altitudes and are less affected by anthropogenic emissions. Moreover, there are no oil fields or wetlands in these areas. Areas with low CH₄ values are located in the western region where human habitation, industry, and agriculture are relatively underdeveloped, which may be the reason for the low CH₄ concentration values in this area.

3.3. Interannual Variation in the CH₄ Concentration in China

Figure 4 shows the spatial distribution of the CH₄ concentration in China in 2003, 2007, 2011, 2015, 2019, and 2021. Overall, the CH₄ concentration in China increased over time. In 2003, the annual average CH₄ concentration was 1794 ppb; by 2021, the annual average concentration reached 1906 ppb, and the annual average growth rate was 0.34%. This was due to the rapid industrial development in China during this period, and the CH₄ that was produced in energy consumption, such as that resulting from the combustion of coal, oil, and natural gas, has far exceeded previous levels. In addition, large areas of forest and agricultural land have been converted into urban and industrial areas, destroying vegetation and making the conditions for CH₄ conversion into organic matter less suitable. Coupled with the gradual shrinking of surface water areas and greatly reduced precipitation, the conditions for CH₄ absorption and dissolution become less suitable, and the dynamic balance between CH₄ formation and transformation is disrupted, resulting in an annual increase in atmospheric CH₄. Between 2003 and 2021, the regions of maximum and minimum CH₄ concentrations also changed. Figure 4 shows that between 2003 and 2007, the CH₄ concentration in the Qinghai–Tibet Plateau was much lower than that in other regions of China, and in 2011, the area with the highest CH₄ concentration changed, and the Qinghai–Tibet Plateau, no longer had the lowest concentration. By 2021, the Qinghai–Tibet Plateau exhibited the highest CH₄ concentration in China.

Figure 5a shows the temporal and spatial distributions of the CH₄ concentration difference between 2003 and 2021. The CH₄ concentration difference in the first step was the most obvious, reaching more than 125 ppb, and even approximately 170 ppb in peak areas within the first step. Figure 5b shows the change in the annual average CH₄ concentration in different regions of China from 2003 to 2021. For example, the CH₄ concentration in the first stage was 1786 ppb in 2003 and reached 1926 ppb in 2021, and the annual average growth rate was 0.42%, which was much higher than the growth rate in China. Before 2010, the CH₄ concentration in the first step was the lowest among the three steps, but after 2011, the CH₄ concentration in the first step became the highest among the four regions. This occurred because the first step mainly comprises the Qinghai–Tibet Plateau, which is gradually warming and wetting. The rising temperature causes the frozen active layer to deepen, thereby releasing a large amount of CH₄. Under the influence of permafrost melting, the area of lakes on the Qinghai–Tibet Plateau is expanding, and the CH₄ emissions are enhanced. In addition, the migration of China’s industrial centers from east to west and the adjustment of the industrial structure may lead to an increase in fuel-burning activities and fossil fuel emissions [45,46,47]. As demand increases, so do the CH₄ emissions from livestock and aquaculture. These factors lead to a rapid increase in the rate of change of CH₄ in the first step. Wu [24] found that the annual average temperature on the Qinghai–Tibet Plateau significantly increased and was much higher than that in other regions, while the temperature in the other regions basically remained stable. Temperature increase is also a factor that indicates great regional changes. The atmospheric CH₄ concentration increased on the Qinghai–Tibet Plateau. High temperatures can lead to temperature rise, and the increase in temperature can promote CH₄ emissions [48,49,50]. The change trends remained relatively stable in the other two regions, with growth rates of 0.31% and 0.29%, which were basically consistent with the change trends in China, indicating that the anthropogenic and natural sources in these regions were relatively stable.

3.4. Seasonal Variation in the CH₄ Concentration in China

Figure 6 shows the four-season distribution of the CH₄ concentration in China from 2003 to 2021. Figure 6a–d denote spring, summer, autumn, and winter, respectively. Figure 6 shows that in the western regions of China, the CH₄ concentration in summer is significantly higher than that in autumn. In these areas, the population is relatively sparse, and the influence of anthropogenic factors is limited, but in the eastern regions of China, the CH₄ concentration in autumn is high. In summer, these areas experience more human activities and industrial coal burning and contain a large area of rice cultivation. It is worth noting that in the first step, the CH₄ concentration rapidly increases in summer and then decreases with decreasing temperature. The Qinghai–Tibet Plateau is sparsely populated, and the CH₄ emissions mainly stem from biological sources. Therefore, in summer, with high temperatures and high humidity levels, the necessary anaerobic environmental and humidity conditions for CH₄ generation are provided, while on the other hand, a large amount of organic matter is present in the soil. CH₄ is decomposed and released into the atmosphere, resulting in a rapid increase in the atmospheric CH₄ content, while the contribution of biological sources is greatly reduced during the cold winter season. In addition, a hotspot with the highest concentration throughout the year occurs in this area. Through a query, it was found that this hotspot is located in the city of Chengdu, Tibet Province. This may be related to the geographical location of this area. Influenced by factors such as latitude and geographical location, Qamdo exhibits characteristics of obvious vertical distribution and notable regional differences. The north-south vertical arrangement of mountains and rivers in this area is conducive to the north-south transportation of warm and humid air masses. The great difference in canyon height affects atmospheric movement, while Wu [24] and other researchers found that the satellite-monitored CH₄ distribution is not only closely related to emission sources on the ground but also related to transmission due to atmospheric motion.

Figure 7a,d show that the average seasonal concentration of CH₄ in China is highest in autumn, followed by summer, winter, and spring (in descending order), with the highest concentration (1853 ppb) observed in autumn and the lowest (1828 ppb) in spring. The higher concentrations in summer and autumn can be attributed to the positive correlation between temperature and CH₄ emissions, as reported in several studies [51,52,53]. These seasons exhibit the highest temperatures and microbial activity, leading to extensive waste decomposition, while biological sources and permafrost release the most CH₄ during this period, making paddy fields, rivers, and lakes the primary biological sources of CH₄. As demonstrated in Figure 6b,c, the average concentration of CH₄ in summer is higher than that in autumn, primarily influenced by natural sources in the first step. In contrast, the population on the Qinghai–Tibet Plateau is sparse, resulting in a minimal impact on CH₄ concentration due to human activities. The summer season in this region has high temperatures, humidity, and frequent microbial activity, leading to vigorous permafrost release, which is primarily responsible for the increased CH₄ concentration. The seasonal variation in CH₄ concentration is not significantly influenced by industrial activities. In the second step, the rice fields in the Sichuan Basin substantially contribute to the elevated CH₄ concentration in summer. The overall characteristics of high CH₄ concentrations in summer and autumn and low concentrations in winter and spring are the results of various factors. The spatial distribution of seasonal variation in CH₄ concentration varies greatly due to regional differences in economic development and CH₄ source distribution.

3.5. Influence of Human Factors on Atmospheric CH₄ Concentrations in China

The sources of CH₄ emissions are mainly classified as either anthropogenic or natural sources, among which the former play a leading role. We verified the correlation between the annual mean CH₄ concentration in China and the annual average GDP and total energy consumption (URL http://www.stats.gov.cn/, accessed on 29 April 2023) from 2003 to 2021, and the correlation coefficients (R values) were 0.99 and 0.97, respectively. The results indicated that economic development and energy consumption were important reasons for the increase in CH₄ concentration. In addition, we obtained the total annual average CH₄ emissions from 2003 to 2021 and the four sectors (solid fuel, rice cultivation, wastewater treatment and discharge, and oil and natural gas), of which solid fuels accounted for approximately 30%, rice cultivation accounted for approximately 24%, wastewater treatment and discharge accounted for approximately 14%, and oil and gas accounted for approximately 4% of the total emissions. AIRS data were used to retrieve the annual average CH₄ concentration for a correlation comparison with the abovementioned sectors, and the results are shown in

Table 2. AIRS retrieval of annual average CH₄ concentration and correlation between different sectors.

Emission Source	R
Total emissions	0.92
Solid fuels	0.84
Rice cultivations	0.72
Wastewater treatment and discharge	0.99
Oil and natural gas	0.96

. The wastewater treatment and discharge sector had the highest correlation with the AIRS-retrieved annual average concentration of CH₄, rice planting and AIRS-retrieved CH₄. On the other hand, the correlation with the annual average concentration was the lowest, which indicated that the CH₄ concentration retrieved by AIRS data was closely related to the surface CH₄ emissions.

3.6. SARIMA Model Prediction of the CH₄ Concentration Trend

In this paper, the monthly CH₄ concentration values for China and its three regions that were retrieved by the AIRS sensor were adopted as the original time series. Choosing all of China as an example, the average monthly CH₄ concentration in the entire country was used as the original sequence. Before the SARIMA model is established, the sequence should be stabilized, and the unit root test (the ADF test) should be performed to determine whether the sequence is stationary. The ADF test results are listed in

Table 3. Results of the ADF unit root test on the CH₄ concentration sequence.

	Original Sequence	Differential Sequence
T statistics	1.11	−5.88
p-value	0.99	2.99 × 10−7
1% threshold	−3.46	−3.46
5% threshold	−2.88	−2.88
10% threshold	−2.57	−2.57

, indicating that the original p-value of the sequence is 0.99 and that the Τ statistics were all greater than the critical values of 1%, 5%, and 10%. Therefore, it could be directly determined that the sequence was nonstationary. Since the original data comprised a nonstationary sequence, the interference of trend and seasonal items was eliminated. Therefore, the seasonal difference was first obtained for the original sequence, the seasonal factor was removed, and the first-order difference was obtained. The ADF unit root test was applied to the sequence after difference elimination, as indicated in Table 3. The result revealed a p-value of 2.99 × 10⁻⁷, and the T statistics were all under the critical values of 1%, 5%, and 10%. Hence, it could be determined that the difference series was a stationary series.

The stationarity test in the previous step verified that the first-order difference sequence was stationary. Next, we determined whether the original sequence was a white noise sequence. Here, we used the Ljung–Box (LB) method to test for white noise. First, we assumed that H₁ could not be rejected at p > 0.05 (if the sequence were purely random, it was a white noise sequence) and that H₂ could not be rejected at p < 0.05 (if the sequence were subject to correlation, it was not a white noise sequence). According to the LB test result, p was much smaller than 0.05, so assuming that H₂ could not be rejected, it could be determined that the sequence after the first-order 12-step difference was not a white noise sequence.

For the SARIMA model establishment, the order of the model must first be determined, and the values of (p,d,q) and (P,D,Q)_s must be determined. In the previous step, the number of first-order differences was determined to be 1, and the d value was determined to be 1. Moreover, D was set to 1, and s was set to 12. The approximate range of the remaining parameters was determined through the use of ACF and principal component analysis (PCA) diagrams, and the optimal model was obtained and screened in combination with the AIC. The optimal model was the SARIMA(1,1,1)(0,1,1)₁₂ model. The residual sequence was then subjected to the LB test, and the results indicated that the p-value of the LB statistic, with a lag ranging from 1 to 12, was greater than 0.05 (the significance level) and passed the significance test. The results demonstrated that the SARIMA(1,1,1)(0,1,1)₁₂ model was a white noise sequence and that the model was significantly effective.

After the above analysis, the SARIMA(1,1,1)(0,1,1)₁₂ model finally passed the significance test. With the use of this model to predict the CH₄ concentration in China, we first employed the data from January 2003 to December 2010 to predict the CH₄ concentration from January 2011 to December 2021. As shown in the orange part of Figure 8a, the SARIMA(1,1,1)(0,1,1)₁₂ model performed well, with little difference between the fitted and actual values. By comparing the original values from January to December 2020 to the predicted values, as summarized in

Table 4. Comparison of the original values and predicted values of the CH₄ concentration from January to December 2020.

Time	Original	Forecast	Relative Error
2020/1	1886.78	1888.67	0.10%
2020/2	1873.80	1874.83	0.05%
2020/3	1878.21	1871.49	−0.36%
2020/4	1876.64	1879.53	0.15%
2020/5	1878.95	1881.83	0.15%
2020/6	1883.46	1887.11	0.19%
2020/7	1895.68	1897.39	0.09%
2020/8	1907.40	1915.07	0.40%
2020/9	1920.42	1912.67	−0.40%
2020/10	1909.40	1904.76	−0.24%
2020/11	1907.24	1897.04	−0.53%
2020/12	1905.06	1902.96	−0.11%

, we found that the relative error between the actual and predicted values was small (controlled to within 1%).

Table 5. Errors in the CH₄ data set under the SARIMA model.

Region	SARIMA(p,d,q)(P,D,Q)s	RMSE	MAE	MSE	R2
China	SARIMA(1,1,1)(0,1,1)12	4.11	3.34	16.94	0.98
First	SARIMA(1,1,1)(4,1,0)12	6.73	5.42	45.41	0.97
Second	SARIMA(1,1,1)(2,1,0)12	4.41	3.54	19.49	0.97
Third	SARIMA(1,1,1)(0,1,1)12	4.70	3.73	22.15	0.96

reveals that the MSE, REMS, and MAE values of the SARIMA model were all low and that the prediction accuracy was high. The results indicated that the model could achieve a satisfactory prediction effect. Therefore, this model could be accurately and reliably used to predict the changing trend of the CH₄ concentration in China in the future. Thus, we could use the SARIMA model to predict the CH₄ concentration from 2022 to 2030, as shown by the yellow line in Figure 8a. The CH₄ concentration is expected to increase over the next 10 years gradually, and corresponding countermeasures must be implemented in China.

The above operations were also performed for the three regions, and the results revealed the SARIMA(1,1,1)(4,1,0)₁₂ model of the first step, the SARIMA(1,1,1)(2,1,0)₁₂ model of the second step, and the SARIMA(1,1,1)(0,1,1)₁₂ model of the third step. The predicted results are shown in Figure 8b–d. The results indicated that the CH₄ concentration in China will continue to rise before 2030. Therefore, to achieve the carbon neutrality goal in 2060, notable efforts are still needed in China to save energy and reduce emissions.

4. Conclusions

In this paper, we used the monthly average CH₄ concentration retrieved from the AIRS data from 2003 to 2021 to analyze the temporal and spatial variation characteristics of CH₄. In addition, the distribution, interannual variation, and seasonal variation were analyzed and summarized. Finally, the SARIMA model was used to predict the future changes in the CH₄ concentration, and the following conclusions could be obtained:

• The CH₄ concentration data retrieved by the AIRS sensor and the ground observations exhibited a suitable correlation. The correlation coefficient relative to the WLG ground station was 0.87, and the correlation coefficient relative to the MNM ground station was 0.92. Therefore, the statistical p-value was less than 0.01, and the significant relationship could be used to analyze the change characteristics of the near-ground CH₄ concentration in China;
• The near-ground CH₄ concentration is generally high in northern China and low in southern China. The high-value areas are mainly located in northwestern Xinjiang, northeastern Inner Mongolia, and northwestern Heilongjiang Province. The low-value areas are located in southern Xinjiang and southwestern Tibet;
• This study investigated the temporal dynamics of CH₄ concentrations in China between 2003 and 2021. The results revealed an overall increasing trend in CH₄ concentration over this period. However, further analysis of the data also showed clear seasonal variations in the CH₄ concentration, with the highest concentrations occurring in summer and autumn and the lowest concentrations in spring and winter. These findings provide important insights into the spatiotemporal distribution of CH₄ in China and highlight the need for continued monitoring and mitigation efforts to address the potential environmental and health impacts of CH₄ emissions in this region. In future research, the underlying drivers of the observed temporal trends and seasonal variations in CH₄ concentration and their implications for climate change and air quality in China should be explored;
• Given the growing concern over the potential impact of CH₄ concentrations on climate change and their associated environmental risks, the development of accurate prediction models is essential for the effective monitoring and mitigation of CH₄ emissions. In this study, we explored the application of the SARIMA model for predicting changes in the CH₄ concentration over the next decade. The results of our analysis indicated that the SARIMA model provides a robust prediction of the future trend of the CH₄ concentration in China with a high degree of accuracy. Specifically, our predictions suggested that the near-surface CH₄ concentration in China will continue to exhibit an increasing seasonal trend in the coming years, with a very small relative error between the predicted and actual values. These findings highlight the potential utility of the SARIMA model as a tool for monitoring and forecasting CH₄ concentrations and provide important insights for policymakers and stakeholders seeking to address the environmental and health impacts of CH₄ emissions in China. However, further research is needed to validate and refine the model and to explore its potential application in other regions and contexts.