A Source-Level Estimation and Uncertainty Analysis of Methane Emission in China’s Oil and Natural Gas Sector

Abstract

A high-quality methane emission estimation in China’s oil and gas sector is the basis of an effective mitigation strategy. Currently, the published emission data and studies of China’s oil and gas sector only provide estimations of total emissions, which is not enough for good analysis of the trend and impact factors for the instruction of emission mitigation activities. The main problem is that published data for oil and gas infrastructure in China is incomplete, which makes it difficult to apply the conventional greenhouse gas inventory compiling method and the uncertainty estimation strategy. Therefore, this paper aims to develop a method to estimate infrastructure data using all available data, including partial data for the infrastructure, national production and consumption of oil and gas, and production and production capacity data of oil and gas enterprises, and then uses a Monte Carlo-based method to generate a source-based inventory and uncertainty analysis of methane emission for China’s oil and gas industry from 1995 to 2018. We found that methane emission increased from 208.3 kt in 1995 to 1428.8 kt in 2018. Methane emission in 2018 has an uncertainty of about ±3%. Compared to former studies, our research found that the production stage of natural gas is the main contributor, which is further driven by the growth of natural gas production. The mitigation potential introduced by technology development on methane emission remains large.

1. Introduction

Methane is the main constituent of natural gas and an abundant greenhouse gas, which contributes 15–20% of global anthropogenic greenhouse gas (GHG) [1]. Methane emissions from oil and natural gas accounts for about 20% of total anthropogenic methane emissions [2]. It can be released into the atmosphere by intended operations or by leaking components across the sector, including production, storage, transportation and distribution stages. The deduction of methane emissions in oil and natural gas sector is relatively easy, can bring about economic and health benefits to human society, and should also act as a necessity in the process to mitigate climate change [3,4,5,6,7,8]. Therefore, the deduction of oil and natural gas methane emissions is now highly prized by the international community. To support adequate methane-deduction actions, a deep understanding of methane emission sources and the total amount of emissions is required. Different countries and research groups have been working for a better estimation of methane emissions to achieve the goal.

The published methane emission data from relevant oil and natural gas research are not comprehensive enough for good analysis of the trends and impact factors of China’s oil and natural gas emissions. To estimate national methane emissions, the greenhouse gas inventory (GHGI) is a commonly used tool. It is a bottom-up approach based on a huge amount of data collected across the country, which estimates the total emissions of a country or a sector by adding up emission measurements or estimations from all activities or facilities within the researched scope [9]. GHGI provides more detailed information about emissions by estimating emission sources such as valves, liquid unloading or tanks, and their comparisons, which indicates the “key emitters” and guides strategic actions to mitigate these emissions. After becoming a member of UNFCCC, China published the Initial National Communication on Climate Change in 2004, which included the first national GHGI from 1994. Then, national GHGIs from 2005, 2010, 2012 and 2014 were published in national communications and biennial update reports [10,11,12,13,14]. However, the inventories gave only the total amount of oil and natural gas emissions and their uncertainties across only five years, without detailed information about emissions from different sources. The uncertainty estimation is rather rough. The national inventories gave the uncertainty analysis result for the category of “energy” for 2005, 2010, 2012 and 2014. Neither the uncertainty of oil and natural gas methane emissions, nor the uncertainty of sub-categories, was published. Some researchers have estimated methane emissions across a long-time scale [15,16,17,18,19,20,21], but we find that these studies present merely the total emissions of the oil and natural gas sector or emissions on different spatial scale. Some studies present emission estimation of carbon dioxide with high spatial and source resolution [22,23], but the method has not been found to be applied in methane emission estimation research. In addition, discrepancies exist in these estimations because of variance in activity data and emission factors [24]. Emissions from different emission sources and their uncertainties are also not given out. A year-by-year, source-based emission inventory might provide key information about methane emission for China’s oil and natural gas sector.

One of the key problems with compiling a more detailed methane inventory for China to estimate methane emissions in China’s oil and natural gas sector is that the systematic published oil and natural gas infrastructure data for academic use is inefficient for the traditional GHGI compilation method. Most GHGIs are compiled based on information about oil and natural gas infrastructures, operations such as liquid unloading and well completion, and measurement data from venting or leakages. The more data that can be found, the more detailed information an inventory may provide. For example, according to the national inventory of the United States, compilation is based upon infrastructure information collected from a compulsory information-reporting system, which is enforced by federal law. Other academic studies may also provide statistics about local emission sources [25,26]. However, official comprehensive statistics of oil and natural gas emission sources in China is rarely found, which means that the traditional GHGI compilation methods based on emission sources cannot be directly applied in this situation. We must estimate the number of emission sources as the first step in compiling the emission inventory. One possible method for this is to estimate emission sources using analysis correlation data, such as the total production of oil or natural gas. Methane emissions are physically generated by the production, transportation, distribution and use process, and the production process motivates the whole sector. We can regard the production of oil and natural gas as the final source of methane emissions, so it is reasonable to estimate the number methane emission sources from these data. On the other hand, these data can be easily collected from annual statistic yearbooks or commercial published reports, facilitating the practical compilation of a year-by-year inventory of methane emission sources.

Another problem is providing an estimation of uncertainty. Uncertainty naturally exists in the process of measurement and the statistical analysis of methane emissions. It is the combination of natural randomness and errors, reflecting the characteristics of the emission estimation as well. Uncertainty is usually presented as a function of the 95% confidence interval. The existence of super-emitters proven by recent research indicates that the emissions of both a single emitter and a particular kind of emitter were probabilistic. The distributions of certain kinds of emission sources are not Gaussian [26,27,28,29,30]. This means that, although the integration of variance still works mathematically, this integration process cannot be applied to the uncertainty integration, because variance cannot be directly converted into uncertainty when emissions are not Gaussian. To tackle this problem, the Monte Carlo (MC) method is now widely applied in research [31,32,33,34,35,36]. The MC method can give out both estimators and empirical distributions of an estimated variable by carrying out random trials with the help of a computer. The MC method will be helpful for estimating emission uncertainties at different stages, and better considering the properties of actual emission sources.

In this research, an emission-source-based Monte Carlo model was established for estimating methane emissions from China’s oil and natural gas sector. Methane emissions from China’s oil and natural gas sector from 1995 to 2018 were estimated. We also estimated the yearly emissions of different emission sources in the oil and natural gas sector, and discussed the factors that may affect the emissions. The uncertainties of the total emission and source emission of each year were also estimated based on the model. There are three main contributions of this research. First, we established a tool to estimate methane emissions without the direct input of emission source data, which means innovation on emission estimation methods on the condition that the basic infrastructure data are inefficient. Second, methane emissions at different stages and operations within China’s oil and natural gas sector were estimated over 20 years. This improves the estimation resolution, making trend analysis possible. Third, more detailed uncertainty estimations were given out to better characterize the methane emissions from China’s oil and natural gas system. This article is organized as follows. The first part is the method, introducing the model structure and data sources. The second part is the results. The third part is the discussion of the results, and comments on this research. The last part summarizes this research and states recommendations for future research.

2. Methods

2.1. Research Scope

Methane emissions from China’s oil and natural gas sector were considered in this research. According to IPCC guidelines, methane emissions from oil and natural gas are classified as category 1.5, including venting and fugitive emission from the oil and natural gas sector. Based on the IPCC guidelines and considering China’s data availability, we considered the following emission sources: newly drilled well completion, liquid unloading, leakage from well heads in daily operations, gathering stations, process stations, compressor stations, gas storage stations, gas receiving stations, pipeline leakage (mainly valves), city gate stations (terminals), city regular stations, city pipeline maintenance and sources in the oil system. Emissions caused by city users (i.e., combustion) are also counted in this research. For the oil sector, the total emission rate is relatively low. The following emission sources are considered in the model: oil well completion, emissions from well heads, storage tanks of wells, crude oil storage tanks, association stations of oil transmission. Emission sources considered in this research includes mainly infrastructure sources, and some emissions from operations are also considered. Emissions of natural gas are further divided into four stages. We consider the emissions from well completion as the exploration emission. The production stage includes liquid unloading, well heads, processing, gather stations and compressors. The transport stage includes gas storage, receiving stations, and pipeline leakage. City gate stations, regular stations, pipeline maintenance and user emissions are marked as downstream emissions in our study. Here, a facility-level emission estimation is given out. We did not make a component- or equipment-level estimation due to the lack of deep investment in China’s actual production sites.

2.2. Model Structure

The structure of the model can be described as per Figure 1. There are three steps in this model to estimate the total methane emissions from China’s oil and natural gas sector.

Step 1: Emission Source Estimation

In the first step in this model, the number of the emission sources was estimated to offer a detailed estimation of total methane emissions and make comparisons between different sources possible. Methane is emitted by infrastructures and production operation activities across the oil and natural gas sector, and the number of facilities and activities show relevance to other statistics such as the total production of natural gas and oil, gas imports, number of receiving and storage stations, etc. These data are known as activity level in our research, and it is reasonable to estimate the number of emission sources from the activity level. All the emission sources considered in this research were introduced in Section 2.1.

The number of oil and natural gas wells, natural gas gathering stations, process stations, compress stations, storage stations, city terminals, transportation tanks, regular stations, oil tanks and operations including drilling, liquid unloading, and pipeline leakage are estimated according to the annual total production or consumption of oil and natural gas from 1995 to 2018. These data were retrieved from China’s statistics yearbook [31,37,38]. For those years where activity levels were not published in the yearbooks, vacant values were calculated using linear interpolation. The published statistics of long-distance natural gas transmission networks and LNG terminals were collected to estimate emission sources in the transportation and distribution process. The number of the natural gas receiving stations is directly taken from published reports [39,40,41,42,43].

We estimated the source numbers by multiplying a coefficient with the correlative data. The number of upstream sources is estimated from annual production, as shown in

Table 1. Emission sources and the corresponding activity level. The latter is used to estimate the number of the emission sources.

Emission Sources	Corresponding Activity Level
Well completion, liquid unloading, well head emission, gather station, processing, compressor station	Natural gas production
Oil-related sources	Oil production
Gate stations, regular stations, emission caused by city users	Gas consumption in cities
City pipeline maintenance	Gas pipeline in cities
Gas storage stations	Gas storage stations
Gas receiving stations	Gas receiving stations
Pipeline leakage	Length of transmission gas pipeline

. We estimated the coefficients from the environment assessment report from Changqing oil field [44,45,46]. Sources at the transportation stage, such as valves, were estimated using the total pipeline mileage. Downstream sources, including city gate stations and sources coming after in the supply chain, were estimated using the total amount of city combustion. Coefficients were confirmed using the public report of the Beijing Gas Co., Ltd. We further assumed that the technical level was stable between 1995 and 2018, which means that we can use the same coefficient across the studied years.

Step 2: Estimate the Emission Rate and Uncertainty of a Single Source

In this step, we estimated the emission rate and its uncertainty of a single emission source by applying the Monte Carlo method. We assumed that the fugitive emission rate of sources followed a log-normal distribution so as to simplify our model, as the log-normal distribution is the most commonly seen heavy-tailed distribution in the research [47,48,49]. Emissions from venting and other organized emission activities follows the normal distribution, as we regard the uncertainty of venting emission estimation resulting from the observation process. Parameters of these distributions were determined by the mean value and variance (mathematically called the first moment and the second central moment). Parameters can be calculated by the following formula. For normal distribution,

$$p\left(x\right)=\frac{1}{\sqrt{2\pi }{\sigma }_{nor}}\exp \left(-\frac{{\left(x-{\mu }_{nor}\right)}^2}{2{\sigma }_{nor}{}^2}\right)$$

(1)

$${\mu }_{nor}=E$$

(2)

$${\sigma }_{nor}=\sqrt{D}$$

(3)

and for log-normal distributions,

$$p\left(x\right)=\frac{1}{x\sqrt{2\pi }{\sigma }_{lognor}}\exp \left(-\frac{{\left(\ln x-{\mu }_{lognor}\right)}^2}{2{\sigma }_{lognor}{}^2}\right)$$

(4)

$${\mu }_{lognor}=\frac{1}{2\ln \frac{E^4}{D+E^2}}$$

(5)

$${\sigma }_{lognor}=\sqrt{\ln \left(\frac{D}{E^2}+1\right)}$$

(6)

After setting the distribution and the parameters, a random test was conducted. Using MATLAB, a random value following a given distribution was assigned to every single emitter of a certain kind as the emissions were observed. The number of emission sources of this kind was estimated in Step 1. After that, we added up the emissions of all the emitters and obtained an estimation of the total emissions of this kind of source. This process was repeated 10,000 times to create a sample set. The mean value and variance of the samples gave an estimation of the source emission characteristics. In addition, an uncertainty bond (95% confidence interval, CI) can be obtained using the 2.5% and 97.5% percentile of the sample.

We considered technology development in our model. The development of technology is presented as the change in the average emission factor. We assumed the average emission factor linearly declined from 1995 to 2018, which means using

$$\begin{array}{l}Average emission facto{r}_k\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}=\frac{2018-k}{23}Average emission facto{r}_{1995}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}+\frac{k-1995}{23}Average emission facto{r}_{2018} ,k=1995,1996,\dots ,2018\end{array}$$

(7)

to simplify the model. The ratio of emission factors of 1995 and 2018 is defined as

$$r=\frac{Average emission facto{r}_{1995}}{Average emission facto{r}_{2018}}$$

(8)

where $r>1$ because of technical advancement. The average emission rate of 2018 was fixed by literature review, which will be introduced in Section 2.3. Therefore, the average emission factor can be calculated by

$$\begin{array}{l}Average emission facto{r}_k=\frac{r\left(2018-k\right)+k-1995}{23} Average emission facto{r}_{2018}\\ \left(k=1995,1996,\dots ,2018\right)\end{array}$$

(9)

As a key parameter of the model, $r$ was confirmed by optimization. As the emission data of years 2005, 2010, 2012 and 2014 were published in the national inventory, we can minimize the following Euclidean metric $d$ to find the optimized $r$:

$$d^2={\displaystyle \sum }{\left(Emissio{n}_{t,model}-Emissio{n}_{t,inventory}\right)}^2, t=2005,2010,2012,2014$$

(10)

where $Emissio{n}_{t,inventory}$ is the emission in the published national inventory of year $t$, and $Emissio{n}_{t,model}$ is the average emission calculated by the model of year $t$. It is easy to prove that $Emissio{n}_{t,model}$ is a function of $t$ and emission factor ratio $r$. In this way, we can find the optimized $r$ that fits the model best to known statistics.

We mainly used reference emission factors from “Guidelines for greenhouse gas emission accounting and report in China oil and natural gas industry (trial)” as the mean values of distributions. For those sources whose emission factors were not published, we adapted emission factors of similar from the Guidelines. Since the variances for Chinese cases were not published in the literature, they were derived by multiplying a factor to the mean value. This factor, the ratio of variance and mean value, is set as a key parameter of this model. We determined the value of these parameters according to Balcombe, et al. [50] This factor will influence the distribution and uncertainty estimation of the final result, which will be discussed in the coming section.

Step 3: Estimate the Total Emission Rate and Uncertainty of the Oil and Natural Gas Sector

The Monte Carlo method is also applied when estimating the total emission rate and uncertainty. We added up the sampling results of all the sources to obtain one sample of the total emissions. This process was repeated 10,000 times. Then, we obtained a sample of the total emission rate from a size of 10,000. The same method as Step 3 was applied to calculate the mean value, variance and 95% CI.

2.3. Decomposition of the Impact of Activity Level and Technical Change

It is natural to realize that the total production and import of oil and natural gas will have a positive effect on methane emissions. The more oil and natural gas that is imported or produced at a particular time, the more active facilities or operations are required, and the more methane might be emitted by these sources. On the other hand, technology developments will result in a lower emission rate, which may offset the effect of total production and import. The interaction of these two factors may lead to different methane mitigation strategies, so it is useful to evaluate the conflicting impacts of these two factors.

In our model, we assume that the number of emission sources is linearly correlated with the activity levels. After calculating the number of emission sources by multiplying by a coefficient, the emission of the sources is calculated following a certain distribution whose expectation is given and incorporates the change in technology level. Adding up the emissions from single sources to calculate the average total emission of this kind means a multiplication operation of emission expectation and the number of emission sources. As a result, a linear relationship between emissions and the activity level is established in this process, and the average emission of this kind of emission source at year $t$, $Emissio{n}_{t,k}$, can be expressed as the following equation:

$$Emissio{n}_{t,k}=T{L}_{t,k}\times A{L}_{t,k}$$

where $A{L}_{t,k}$ stands for the activity level, and $T{L}_{t,k}$ stands for the technology influence for emission source $k$ in year $t$. It must be noted that $T{L}_{t,k}$ combines the influence of different factors, including the number of the single sources, average emission rate, etc. They all reflect the influence of emission mitigation technology, but are not further divided in this research. To show the impact of these two factors on emissions, we break down the increment of methane emission as follows:

$$\begin{array}{ll}Emissio{n}_{t,k} -& Emissio{n}_{t-1,k}=T{L}_{t,k}\times A{L}_{t,k}-T{L}_{t-1,k}\times A{L}_{t-1,k}\\ & \hspace{1em}=\left(T{L}_{t-1,k}+\Delta T{L}_{t-1,k-1}\right)\times \left(\frac{A{L}_{t,k}}{A{L}_{t-1,k}}\right)A{L}_{t-1,k}-T{L}_{t-1,k}\times A{L}_{t-1,k}\\ & \hspace{1em}=T{L}_{t-1,k}\left(r-1\right)A{L}_{t-1,k}+\Delta T{L}_{t-1,k-1}·r·A{L}_{t-1,k}\\ & \hspace{1em}=Emissio{n}_{t-1,k}\left(r-1\right)+\Delta T{L}_{t-1,k-1}A{L}_{t,k}\end{array}$$

(11)

where $r=\frac{A{L}_{t,k}}{A{L}_{t-1,k}}$ stands for the ratio of the activity level of the following two years, and $\Delta T{L}_{t-1,k-1}=T{L}_{t,k}-T{L}_{t-1,k}$ stands for the increment of technology influence factors. Here, we define

$$IA{L}_{t,k}=Emissio{n}_{t-1,k}\left(r-1\right)$$

(12)

$$IT{L}_{t,k}=\Delta T{L}_{t-1,k-1}A{L}_{t,k}=Emissio{n}_{t,k}-Emissio{n}_{t-1,k}-IA{L}_{t,k}$$

(13)

In these equations,$IA{L}_{t,k}$ and $IT{L}_{t,k}$ stand for the impact of the activity level and technology, respectively, for emission source $k$ in year $t$. It must be noticed that the emission increases of different sources are driven by different activity levels. As a result, the impact of a certain type of emission increase is deposited based on the correlated activity level data. Depositions of emission sources are conducted, and the total impact of activity level change and technology development are calculated by adding up the impact of different sources.

3. Results

3.1. Total Emissions from the Oil and Natural Gas Sector in China

Total emissions from the oil and natural gas sector in China from 1995 to 2018 are estimated in this research. By optimization, the ratio of emission factors between 1995 and 2018 is 1.30914. Based on the estimation of $r$, the total emission in 1995 is 208.4 kt, and this number rises to 1430.3 kt in 2018. The uncertainty bands (95% CI) of methane emission depend on the estimation of the ratio of the variance and the mean value of mean emission factor. When setting $rati{o}_{lognor}=3$ and $rati{o}_{nor}=0.2$, the uncertainty band in 1995 is 204.0–214.0 kt and 1402.6–1471.9 kt in 2018. Detailed results can be seen in Figure 2 and the emission inventory in Table 2.

A brief comparison between the result and other research is shown in Figure 3. We selected two other studies: estimations from EDGAR5.0 and Schwietzke, 2014 [17,51]. The estimation of this research is the lowest estimation. There might be several reasons for this divergence. First, the researched emission sources might be different between these studies. Second, due to the method difference, average emission rates (emission factors) may vary and cause great divergence in results. Field tests and studies of native average emission rates might be helpful to study the reason for this divergence.

3.2. Emissions from Different Sources

The emissions from different sources are shown in Figure 4. We categorize the NG sector into exploration, production, transportation and downstream. The comparison of the NG sector and oil sector are also shown. Upstream and downstream of the natural gas sector are the greatest emission sources across the studied years. They also contribute most of emission growth, whose emissions are counted about 7 times and 16 times in 2018 compared to 1995, respectively. Emissions from the oil system changed slightly around 46–56 kt.

From our estimation, the methane emissions from the oil and natural gas sector have increased intensely since 1995. This increase is mainly caused by emission increases in the natural gas sector. According to our estimation, methane emissions from oil changes from 50.29 kt in 1995 to 48.52 kt in 2018; the percentage of oil-related emissions in total methane emissions drop from 24.1% to 3.4%. By contrast, methane emissions from natural gas sources increase sharply to 1381.8 kt, and take a share of 96.6% in 2018. Natural gas-related sources have become the main contributing sources in the oil and natural gas sector.

To better analyze source emission differences, emissions from the top eight emitters (which contribute over 95% of the total emission) are shown in Figure 5. Sources in the natural gas supply chain make up seven out of top eight emitters. Among them, liquid unloading, gather stations and regular stations are the three major emitters in this system. Emissions from regular stations have increased sharply over the studied years, and become the largest emitter in 2018. Emissions from liquid unloading and gather stations also obviously increased in the studied years. The oil system was the largest emitter before 2000, but emissions from the oil system has stayed rather stable, and its contribution to the total emission has gradually decreased.

The difference of the contributions of oil and gas can be summarized in two ways. First, methane is only a by-product of the oil system, and the methane emission potential of the oil system is naturally smaller than the natural gas system. Difference in the emission potential can be proven by the emission factors used in our model. Compressor-based facilities, such as compressor stations and regular stations, have rather large average emission rates. These factors make the natural gas system more likely to produce more methane emissions. Second, the total production and import of oil represents a much smaller change than the total production and import of natural gas. Take the production data as an example. The production of oil in China changed from 150.05 million tons to a maximum of 209.74 million tons in 2015, and then dropped to 189.32 million tons in 2018. The production of natural gas, by contrast, rapidly increased from 17.95 billion cubic meters to 158.16 billion cubic meters. The difference in activity level will naturally cause the different of methane emissions.

From the above analysis, we may also conclude that the rapidly increasing natural gas production and import, which indicates high total natural gas demand in reality, is the final motivator of methane emissions from China’s oil and natural gas sector. This deduction can be proven by comparing the trend of methane emissions and the activity level data. We calculate the correlation coefficient of the total methane emissions and different activity level data series of 1995–2018. The result shows that natural gas production has the highest correlation coefficient of 0.9995, indicating a close relationship between them. As a simple conclusion, we found that the activity level, especially total natural gas production, has motivated methane emission growth in the past 20 years, and resulted in the difference between oil and natural gas’ contribution to total emissions.

3.3. Emission Inventory, 1995–2018

To demonstrate the results of this research, an emission inventory from China’s oil and natural gas sector is shown in Table 2. We take 2018 as an example to show the estimation of the uncertainty bond of each type of source, as shown in Table 3. The total emission has an asymmetric distribution, as shown in Figure 6.

It can be noticed that the uncertainty of a single source is rather large, but the uncertainty of total emissions is smaller. For example, the upper uncertainty range of a single wellhead is about four times that of the average emission rate (9.4898 versus 2.5000), but the uncertainty of total emission is 3%. This difference result can be explained by the central limitation theorem, which means that the probabilistic distribution of total emission will converge to the normal distribution. The more emission sources that are considered, the better the total emission will converge to the normal distribution. This makes a good explanation of why the distribution of the total emissions is much closer to a Gaussian distribution, rather than a heavy-tail distribution. In addition, it also indicates that the distribution characteristic of single sources, despite being heavy tailed according to much recent research, may not have as great an influence on the distribution of total emissions as we had imagined. Heavy-tailed distributions discovered by field tests are responsible for the skewed estimation of average emission rate, since lots of heavy emitters might be ignored or poorly considered in the average emission rate estimation. They minorly influence the total emission estimation, since the number of emission sources results in the convergence to Gaussian distribution.

3.4. The Impact of Activity Level and Technical Change

Following the above procedures shown in Section 2.3, the impact of these two factors can be calculated and shown in Figure 7. The emission growth caused by the activity level greatly overruns the mitigation effect of technology development, and we can see that the total emission increment shown by the yellow line has nearly the same growing trend as the activity impact. We further picture the change in increment of natural gas production, one of the most important activity level data, and found that the increment of natural gas production has nearly the same trend as the total methane emission increment. This further indicates natural gas’ impact on the change in total emissions.

Table 2. Methane emission inventory from China’s oil and natural gas sector, 1995–2018. Unit of methane emission is kt.

Year	New Gas Well	Liquid Unloading	Wellhead	Gather Station	Process Station	Compress Station	Storage	Receive	Pipeline Leakage	Gate	Regular	Pipeline Downstream	Combustion	Oil System	Total Emission	Uncertainty (95% CI)
1995	0.00	49.77	20.19	30.61	9.85	18.05	0.09	0.00	0.08	4.72	24.17	0.06	0.54	50.29	208.43	[203.97, 214.00]
1996	4.70	54.35	22.05	33.43	10.75	19.71	0.09	0.00	0.16	4.43	22.63	0.12	0.53	50.60	223.56	[218.77, 229.58]
1997	4.66	58.82	23.85	36.19	11.63	21.22	0.09	0.00	0.19	4.56	23.32	0.14	0.56	50.88	236.12	[231.00, 242.65]
1998	4.61	63.17	25.62	38.88	12.48	22.81	0.09	0.00	0.25	4.87	24.91	0.19	0.61	51.16	249.62	[244.13, 256.78]
1999	4.56	67.44	27.35	41.44	13.36	24.36	0.60	0.00	0.30	5.18	26.45	0.23	0.66	51.42	263.33	[257.46, 271.00]
2000	4.51	71.58	29.03	44.01	14.13	25.87	0.59	0.00	0.35	5.47	27.96	0.28	0.70	51.66	276.14	[270.00, 284.20]
2001	10.64	82.32	33.39	50.62	16.31	29.79	0.59	0.00	0.41	6.85	35.01	0.32	0.88	52.26	319.38	[312.48, 328.63]
2002	10.53	92.80	37.64	57.07	18.45	33.63	0.58	0.00	0.46	8.19	41.85	0.37	1.05	52.84	355.46	[347.64, 366.15]
2003	10.41	103.03	41.79	63.37	20.44	37.25	0.58	0.00	0.53	9.50	48.58	0.41	1.22	53.40	390.52	[382.02, 402.10]
2004	10.30	113.01	45.84	69.52	22.35	40.91	0.57	0.00	0.61	10.79	55.15	0.46	1.39	53.90	424.80	[415.52, 437.41]
2005	10.18	122.74	49.78	75.51	24.29	44.35	0.56	0.00	0.77	13.26	67.74	0.59	1.69	54.41	465.86	[456.02, 479.55]
2006	20.70	143.73	58.30	88.39	28.46	52.01	0.56	0.11	0.99	15.25	77.89	0.78	1.98	55.06	544.21	[532.63, 560.34]
2007	20.46	164.22	66.61	101.02	32.39	59.38	0.78	0.11	1.23	19.00	97.10	0.99	2.46	55.68	621.42	[608.14, 639.94]
2008	20.22	184.17	74.70	113.29	36.36	66.67	0.78	0.32	1.42	22.39	114.39	1.18	2.90	56.26	695.06	[680.24, 715.69]
2009	19.99	203.63	82.58	125.22	40.15	73.66	0.77	0.43	1.69	24.35	124.44	1.40	3.21	56.82	758.34	[741.81, 781.08]
2010	19.75	222.54	90.26	136.88	43.98	80.46	0.76	0.42	1.99	28.96	147.98	1.64	3.92	57.35	836.89	[819.16, 861.78]
2011	16.80	238.02	96.53	146.38	47.04	86.05	0.82	0.62	2.28	39.82	203.52	1.91	5.32	56.62	941.74	[922.38, 969.01]
2012	9.44	245.32	99.50	150.88	48.67	88.70	0.96	0.82	2.58	46.07	235.48	2.19	6.18	56.61	993.41	[973.33, 1021.91]
2013	28.64	273.25	110.82	168.07	53.89	98.85	1.53	1.12	2.89	50.83	259.78	2.48	6.96	56.34	1115.44	[1093.29, 1147.64]
2014	19.19	290.57	117.85	178.70	57.42	105.13	1.80	1.40	3.17	54.49	278.52	2.78	7.57	56.38	1174.96	[1151.44, 1209.39]
2015	9.07	296.68	120.32	182.48	58.70	107.34	1.78	1.39	3.55	58.06	296.74	3.18	8.10	57.21	1204.60	[1180.06, 1239.05]
2016	4.54	297.75	120.77	183.13	58.73	107.72	1.76	1.46	3.86	64.53	329.79	3.52	9.06	52.43	1239.04	[1214.71, 1274.50]
2017	22.17	317.84	128.90	195.46	62.64	114.97	1.73	1.64	4.28	68.69	351.00	3.98	9.88	49.69	1332.86	[1306.96, 1371.25]
2018	19.83	335.08	135.90	206.06	66.35	121.16	1.71	2.09	4.70	77.42	395.72	4.46	11.33	48.52	1430.33	[1402.57, 1471.90]

Table 2. Methane emission inventory from China’s oil and natural gas sector, 1995–2018. Unit of methane emission is kt.

Year	New Gas Well	Liquid Unloading	Wellhead	Gather Station	Process Station	Compress Station	Storage	Receive	Pipeline Leakage	Gate	Regular	Pipeline Downstream	Combustion	Oil System	Total Emission	Uncertainty (95% CI)
1995	0.00	49.77	20.19	30.61	9.85	18.05	0.09	0.00	0.08	4.72	24.17	0.06	0.54	50.29	208.43	[203.97, 214.00]
1996	4.70	54.35	22.05	33.43	10.75	19.71	0.09	0.00	0.16	4.43	22.63	0.12	0.53	50.60	223.56	[218.77, 229.58]
1997	4.66	58.82	23.85	36.19	11.63	21.22	0.09	0.00	0.19	4.56	23.32	0.14	0.56	50.88	236.12	[231.00, 242.65]
1998	4.61	63.17	25.62	38.88	12.48	22.81	0.09	0.00	0.25	4.87	24.91	0.19	0.61	51.16	249.62	[244.13, 256.78]
1999	4.56	67.44	27.35	41.44	13.36	24.36	0.60	0.00	0.30	5.18	26.45	0.23	0.66	51.42	263.33	[257.46, 271.00]
2000	4.51	71.58	29.03	44.01	14.13	25.87	0.59	0.00	0.35	5.47	27.96	0.28	0.70	51.66	276.14	[270.00, 284.20]
2001	10.64	82.32	33.39	50.62	16.31	29.79	0.59	0.00	0.41	6.85	35.01	0.32	0.88	52.26	319.38	[312.48, 328.63]
2002	10.53	92.80	37.64	57.07	18.45	33.63	0.58	0.00	0.46	8.19	41.85	0.37	1.05	52.84	355.46	[347.64, 366.15]
2003	10.41	103.03	41.79	63.37	20.44	37.25	0.58	0.00	0.53	9.50	48.58	0.41	1.22	53.40	390.52	[382.02, 402.10]
2004	10.30	113.01	45.84	69.52	22.35	40.91	0.57	0.00	0.61	10.79	55.15	0.46	1.39	53.90	424.80	[415.52, 437.41]
2005	10.18	122.74	49.78	75.51	24.29	44.35	0.56	0.00	0.77	13.26	67.74	0.59	1.69	54.41	465.86	[456.02, 479.55]
2006	20.70	143.73	58.30	88.39	28.46	52.01	0.56	0.11	0.99	15.25	77.89	0.78	1.98	55.06	544.21	[532.63, 560.34]
2007	20.46	164.22	66.61	101.02	32.39	59.38	0.78	0.11	1.23	19.00	97.10	0.99	2.46	55.68	621.42	[608.14, 639.94]
2008	20.22	184.17	74.70	113.29	36.36	66.67	0.78	0.32	1.42	22.39	114.39	1.18	2.90	56.26	695.06	[680.24, 715.69]
2009	19.99	203.63	82.58	125.22	40.15	73.66	0.77	0.43	1.69	24.35	124.44	1.40	3.21	56.82	758.34	[741.81, 781.08]
2010	19.75	222.54	90.26	136.88	43.98	80.46	0.76	0.42	1.99	28.96	147.98	1.64	3.92	57.35	836.89	[819.16, 861.78]
2011	16.80	238.02	96.53	146.38	47.04	86.05	0.82	0.62	2.28	39.82	203.52	1.91	5.32	56.62	941.74	[922.38, 969.01]
2012	9.44	245.32	99.50	150.88	48.67	88.70	0.96	0.82	2.58	46.07	235.48	2.19	6.18	56.61	993.41	[973.33, 1021.91]
2013	28.64	273.25	110.82	168.07	53.89	98.85	1.53	1.12	2.89	50.83	259.78	2.48	6.96	56.34	1115.44	[1093.29, 1147.64]
2014	19.19	290.57	117.85	178.70	57.42	105.13	1.80	1.40	3.17	54.49	278.52	2.78	7.57	56.38	1174.96	[1151.44, 1209.39]
2015	9.07	296.68	120.32	182.48	58.70	107.34	1.78	1.39	3.55	58.06	296.74	3.18	8.10	57.21	1204.60	[1180.06, 1239.05]
2016	4.54	297.75	120.77	183.13	58.73	107.72	1.76	1.46	3.86	64.53	329.79	3.52	9.06	52.43	1239.04	[1214.71, 1274.50]
2017	22.17	317.84	128.90	195.46	62.64	114.97	1.73	1.64	4.28	68.69	351.00	3.98	9.88	49.69	1332.86	[1306.96, 1371.25]
2018	19.83	335.08	135.90	206.06	66.35	121.16	1.71	2.09	4.70	77.42	395.72	4.46	11.33	48.52	1430.33	[1402.57, 1471.90]

Table 3. Average methane emission and their uncertainty band of emission sources in 2018. Unit of methane emission is kt.

	New Gas Well	Liquid Unloading	Wellhead	Gather Station	Process Station	Compress Station	Storage	Receive	Pipeline Leakage	Gate	Regular	Pipeline Downstream	Combustion	Oil System	Total Emission
Average	19.83	335.08	135.90	206.06	66.35	121.16	1.71	2.09	4.70	77.42	395.72	4.46	11.33	48.52	1430.33
Lower bond	19.35	333.16	134.66	204.75	38.81	120.05	1.58	1.95	4.60	76.47	393.87	4.46	11.33	46.48	1402.57
Upper bond	20.33	337.05	137.15	207.32	107.98	122.27	1.84	2.24	4.80	78.38	397.60	4.46	11.33	50.87	1471.90

Table 3. Average methane emission and their uncertainty band of emission sources in 2018. Unit of methane emission is kt.

	New Gas Well	Liquid Unloading	Wellhead	Gather Station	Process Station	Compress Station	Storage	Receive	Pipeline Leakage	Gate	Regular	Pipeline Downstream	Combustion	Oil System	Total Emission
Average	19.83	335.08	135.90	206.06	66.35	121.16	1.71	2.09	4.70	77.42	395.72	4.46	11.33	48.52	1430.33
Lower bond	19.35	333.16	134.66	204.75	38.81	120.05	1.58	1.95	4.60	76.47	393.87	4.46	11.33	46.48	1402.57
Upper bond	20.33	337.05	137.15	207.32	107.98	122.27	1.84	2.24	4.80	78.38	397.60	4.46	11.33	50.87	1471.90

On the other hand, the offset effect of technology development is small but keeps increasing. We estimate that the coefficient $r$ is 1.30914, which means that the average emission rate of 2018 has only a 24% decrease compared to 1995. There exists great technology development potential. In addition, from Equation (13), we may be also reminded that technological development may create a large mitigation effect when the total emission is large. This indicates that it might be more effective to promote emission mitigation technologies now for highly active energy production activities. Since natural gas may play a more important role in China’s energy supply during the energy transition period, mitigation technology should be highly focused on cutting China’s methane emissions.

3.5. The Influence of Different Parameters

There are three parameters in this model: technology development factor r, ratio of variance and mean value of normal distribution, $r_{\mathrm{nor}}$, and log-normal distribution, $r_{\mathrm{lognor}}$. Their influence on the estimation result will be discussed in this section.

Ratios of variance and mean value mainly influence the distribution of total emission. Here, combinations of the two ratios were selected, and total emissions were calculated. We take the results from 2018 as an example, and the results are presented in

Table 4. Uncertainty of total emission under different combinations of $r_{\mathrm{lognor}}$ and $r_{\mathrm{nor}}$ in 2018. Unit of methane emission is kt.

${\mathit{r}}_{\mathbf{l}\mathbf{o}\mathbf{g}\mathbf{n}\mathbf{o}\mathbf{r}}=3$	Uncertainty Band	Average Emission	Band Length	${\mathit{r}}_{\mathbf{n}\mathbf{o}\mathbf{r}}=0.2$	Uncertainty Band	Average Emission	Band Length
$r_{\mathrm{nor}}=0.1$	[1402.8, 1470.0]	1430.3	67.2	$r_{\mathrm{lognor}}=1$	[1412.6, 1452.4]	1430.3	39.8
$r_{\mathrm{nor}}=0.2$	[1402.6, 1471.9]	1430.3	69.3	$r_{\mathrm{lognor}}=3$	[1402.6, 1471.9]	1430.3	68.9
$r_{\mathrm{nor}}=0.5$	[1403.2, 1471.5]	1431.2	68.3	$r_{\mathrm{lognor}}=5$	[1396.4, 1485.3]	1430.4	88.9
$r_{\mathrm{nor}}=0.7$	[1407.8, 1477.1]	1436.1	69.3

. We can see that $r_{\mathrm{lognor}}$ has a more obvious influence on the total distribution than $r_{\mathrm{nor}}$. This can be explained by two aspects. First, the change in $r_{\mathrm{lognor}}$ can lead to a highly skewed distribution, which may introduce obvious asymmetry to total distribution. Second, emissions caused by emitters following normal distribution is less than that from log-normal emitters. This can be concluded from the difference of average emission rate. A smaller contribution of total emission results in a smaller influence on the total fluctuation. The conclusion—$r_{\mathrm{lognor}}$ has a more obvious influence—indicates that we should pay more attention to the emission characteristics of these emission sources (mainly including fugitive emission sources) to obtain a better estimation of total emission uncertainty.

Another influence of $r_{\mathrm{nor}}$ is that the increase of $r_{\mathrm{nor}}$ will lead to an increase of average total emissions, which seems to violate the probability law. This phenomenon results from the model-sampling strategy. Unlike the log-normal distribution, Gaussian distribution does not ensure a positive sample result. The larger its variance, the more likelihood that it will generate a nonpositive sample. These nonpositive samples are set to be zero, since actual emission rates are nonnegative. As a result, these values are enlarged, and cause a rise of total emissions. This indicates that when the actual measurements of normal distribution variance are rather large, a new distribution model might be selected to better reflect the actual emission fluctuation.

Technology development factor r can be calculated using optimization as described above, but the analysis of the factor impact may further reveal the influence of methane mitigation technology to the total emission. We fixed $r_{\mathrm{nor}}=0.2$ and $r_{\mathrm{lognor}}=3$, and changed the value of r as 1, 1.309, 2 and 4. Total emissions under these circumstances are shown in Figure 8. A large r stands for rapid technology development, and decrease of average emission rate. We noticed that when the technology development was rapid enough, an emission peak appeared around 2015 and the total emission continuously decreased till now. The peak value increased because we set the emission of 2018 as a constant. The increment decomposition also shows that when r is big enough, the mitigation effect of technology development can totally offset the increase caused by activity level. These all indicate that technology development has a great methane mitigation potential, which should be paid attention.

4. Discussions and Conclusions

In this research, we present an inventory of methane emissions from China’s oil and natural gas sector, 1995–2018. To finish this task, we develop a method that matches the data availability from China. We evaluate the average emissions and their uncertainty by applying probabilistic methods to the studied years. This inventory also estimates emissions from different sources, thus making an improvement to former studies. The main findings of this research include:

• A quick rise of methane emissions is observed. Methane emissions increase from 208.4 kt in 1995 to 1430.3 kt in 2018. Methane emissions in 2018 have an uncertainty of about $\pm$3%. In addition, methane emissions are continuously rising with high probability based on the trends shown in this research.
• Emissions from the natural gas sector are the key motivator of total methane emissions. Natural gas-related emission sources contribute most of the methane emissions. Liquid unloading, gather stations and regular stations are the key emitters with obvious emission increase across the studied years. Methane emissions are driven by natural gas production and import increase, which represents economic development and energy structure adjustment.
• The mitigation potential introduced by technology development on methane emissions is still large. According to this research, the mitigation effect of comprehensive technology development is small, but as the emissions grow higher, technology development will create more methane mitigation opportunities. As natural gas is playing an increasingly important part in China’s carbon neutrality strategy, technology development might become a vital emission mitigation tool.

Based on the above findings, we make the following recommendations. First, to reduce methane emissions, sources such as liquid unloading and compressors should be paid attention. Research on the emission characteristics of these key emitters should also be focused on forthcoming research, because the emissions from key emitters may have a considerable impact on the estimation of total emission, as well as deepening the understanding of these emission sources compared to global research. Second, for China, emission mitigation technologies should play an important part in emission mitigation policies. The great mitigation potential of technologies makes the study and application of technologies vital. Third, since methane emissions were mainly motivated by natural gas production and consumption, the natural gas sector should be focused on emission mitigation policies from many perspectives. Policies may highlight the application of emission mitigation technologies. When setting the emission mitigation goals, the prediction of future natural gas demand and supply should be taken into consideration to make the emission goal realistic and achievable. Fourth, a comprehensive statistics database of emission-related data will greatly increase the inventory quality. For policymakers, establishing a data-collection system might be of the same importance as making emission mitigation policies, because a better estimation of methane emissions, as well as other greenhouse gases, can provide a concrete basis to effectively mitigation policies.

There are several limitations of this research. First, the model is rather simple, sketching the main structure of the supply chain. It might be possible to make a more detailed component-level modeling, to better reflect physical reality. We also make some bold simplifications in our research, which can be further specified, for example, by introducing different emission distributions based on field tests in China. In the near future, coming work is expected to compile the inventory to component level, considering different uncertainty sources such as measurement errors and natural fluctuation. Parameters used in our research can also be updated if data based on native research are available, in order to obtain a better national inventory.