Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Next Article in Journal
BAFormer: A Novel Boundary-Aware Compensation UNet-like Transformer for High-Resolution Cropland Extraction
Previous Article in Journal
Multispectral, Thermographic and Spectroradiometric Analyses Unravel Bio-Stimulatory Effects of Wood Distillate in Field-Grown Chickpea (Cicer arietinum L.)
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 16
Issue 14
10.3390/rs16142525
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Monitoring Methane Concentrations with High Spatial Resolution over China by Using Random Forest Model

by
Zhili Jin
,
Junchen He
and
Wei Wang
*
School of Geosciences and Info-Physics, Central South University, Changsha 410083, China
*
Author to whom correspondence should be addressed.
Remote Sens. 2024, 16(14), 2525; https://doi.org/10.3390/rs16142525
Submission received: 11 May 2024 / Revised: 5 July 2024 / Accepted: 7 July 2024 / Published: 10 July 2024
(This article belongs to the Topic Accessing and Analyzing Air Quality and Atmospheric Environment)
Browse Figures
Versions Notes

Abstract

:
Atmospheric methane is one of the major greenhouse gases with a drastic impact on climate change. This study developed a random forest model to obtain a daily 5 km resolution atmospheric methane concentration dataset with full spatial coverage (100%) from 2019 to 2021 in mainland China, thereby filling the gap in the methane product data from the Tropospheric Monitoring Instrument (TROPOMI). The coefficients of determination for a sample-based and spatial-based cross-validation are 0.97 and 0.93, respectively. The average deviation of the seamless methane product reconstructed by the random forest model is less than 1%, validated with the measured methane concentration data from the Total Carbon Column Observing Network sites. Methane concentrations in China show a distribution of high in the east and south and low in the west and north. The high-concentration areas include Central China, the Sichuan Basin, the Pearl River Delta, and the Yangtze River Delta. In terms of time scale, the methane concentration has evident seasonal variation, as it is low in spring (average 1852 ppb) and winter (average 1881 ppb) and high in summer (average 1885 ppb) and autumn (average 1886 ppb). This is mainly due to the significant increase in emissions from rice cultivation and wetlands during the summer and autumn. During the COVID-19 pandemic, the methane concentration decreases significantly and then starts to return to normal around 70 days after the Lunar New Year, indicating that the seamless methane product can potentially detect anomalous changes in methane concentration.

Graphical Abstract

1. Introduction

The scientific literature indicates that methane is the second most significant greenhouse gas, after carbon dioxide [1]. Atmospheric methane levels have been rising steadily since 2007. As of now, atmospheric methane levels have reached approximately 2.5 times pre-industrial levels [2]. The IPCC 5th Assessment Report [3] indicates that the global warming potential (GWP) of methane is 28 times higher than that of carbon dioxide over a 100-year time horizon. Furthermore, methane can compete with O3 for hydroxyl radicals (OH) in the atmosphere, resulting in the production of CO2 and H2O. This phenomenon affects the levels of ozone and water vapor in the atmosphere, consequently influencing the processes of the atmospheric cycle [4]. The continued growth of methane may increase the probability of extreme weather events, which in turn may affect people’s lives and socioeconomic development. Therefore, it is important to monitor and study the spatial and temporal distribution of atmospheric methane and its long-term trends.
The sources of atmospheric methane are diverse and complex and can be broadly classified into two categories: natural and anthropogenic. Natural sources include wetlands, oceans, vegetation, termites, and animal populations. Anthropogenic sources include rice cultivation, landfills, animal husbandry, biofuel burning, fossil fuel exploitation, and combustion [5,6]. In addition to the complexity of the sources, there is a degree of uncertainty in the proportion of each methane source due to various external natural or human factors, and research on methane source sinks is still in progress [7,8]. For example, the results of different scientists on global methane emissions from vegetation are inconsistent and even suggest new possible sources [9,10,11,12,13].
China has a large land area and complex methane sources. Therefore, it is difficult to obtain accurate methane concentration data with better spatial and temporal resolution based only on sparse and unevenly distributed ground-based monitoring stations. The inventory data of different methane concentrations are not only spatially poorly covered but also suffer from different degrees of underestimation and other problems [14,15]. In contrast, satellite-based remote sensing methods can monitor atmospheric methane on a large scale and over a long period. The main satellite sensors capable of detecting methane are SCIAMACHY carried by ENVISAT, TANSO-FTS carried by GOSAT, AIRS carried by Aqua and the Tropospheric Monitoring Instrument (TROPOMI) carried by Sentinel-5P [16,17,18,19,20]. Although the product data from these satellites can provide methane column concentrations on a global scale, they all lack spatial coverage. Several researchers have used SCIAMACHY, AIRS, and GOSAT data to invert the methane distribution in China and globally, and found that the trends from satellite observations are consistent with ground-based observations [10]. Rice cultivation and wetlands are significant sources of methane emissions in the China region, with a peak in the fall. However, the satellite data inversions of methane concentration data have low spatial and temporal resolution or are missing in the spatial coverage [21,22,23,24,25,26]. This is due to the satellite hardware and environmental factors, such as clouds and rainfall, which cause the satellite-acquired data to be missing or unusable because of poor quality. To address this issue, some researchers have employed deep learning methodologies and time series approaches to model methane statistics and predict its trends over time [27,28,29,30]. However, they have not yet incorporated the data spatially. Additionally, some researchers have modeled atmospheric methane top-down based on atmospheric chemistry methods, which obtain methane concentration data with high accuracy over a wide range, yet lack sufficient spatial resolution [31,32,33].
To gain a more comprehensive understanding of atmospheric methane and to analyze its spatial and temporal trends, it is essential to obtain seamless methane data with high spatial and temporal resolutions. Currently, there are complex sources of atmospheric methane, sparse data from ground-based monitoring stations, and missing spatial coverage of satellite products. Furthermore, atmospheric chemistry models have low spatial resolution, and emission inventory models have low temporal resolution. To address these limitations, a methane concentration simulation model was constructed by combining machine learning models with high spatial resolution time series data from TROPOMI. This was employed to fill the satellite methane data products with missing values, thereby obtaining daily seamless 5 km resolution atmospheric methane concentration data for the China region. Finally, the usability of this dataset was verified by analyzing the spatial distribution and temporal trends of the atmospheric methane concentrations in China.

2. Materials and Methods

2.1. Materials

Figure 1 shows the study area (China). The data used in our study included methane column product data from TROPOMI, methane observations from Total Carbon Column Observing Network (TCCON) sites, atmospheric reanalysis data from ERA5, and ground elevation data from SRTM (Shuttle Radar Topography Mission) provided by the USGS.
Previous studies have demonstrated that the methane products of TROPOMI exhibit higher spatial resolution and superior coverage compared to those of SCIAMACHY, GOSAT, and AIRS [34]. Additionally, the quantity of TROPOMI data is approximately one hundred times larger than that of GOSAT, yet it is accompanied by a correspondingly higher degree of bias [35]. Consequently, TROPOMI data are more suitable for methane retrieval in fine regions. In conclusion, the TROPOMI product was selected to obtain sufficient data for the methane simulations. TROPOMI is the satellite instrument on board the Copernicus Sentinel-5 Precursor satellite, which commenced operation in April 2018. The secondary product data selected included the secondary offline (OFEL) data product for the total column concentration of methane. The data primarily comprise the column-averaged dry air mixing ratio of methane (XCH4), with the bias-corrected data employed. Furthermore, the high-quality (quality assurance value > 0.5) XCH4 retrievals were selected by the recommendations. The data were available from 30 April 2018 with a spatial resolution of 7 km × 7 km, but increased to 5.5 km after 6 August 2019. Despite the global coverage and high spatial resolution of the data, the effect of cloud cover results in poor data coverage. We collected XCH4 data from 1 January 2019 to 31 December 2021, to obtain sufficient data. During the study period, the maximum value of the data was 2130 ppb (parts per billion) and the minimum value was 1649 ppb. We acquired 2 m temperature (T2M), surface net solar radiation clear sky (SSR), and top net solar radiation clear sky (TSR) data from the ERA5 hourly level atmospheric in-analysis product, with a spatial resolution of 0.25° × 0.25°. We acquired the ground data from the digital elevation model (DEM, 90 m × 90 m) products. The details of the data used are summarized in Table 1, and all data were resampled to a uniform spatial resolution of 0.05° × 0.05° (equal to approximately 5 km) and averaged daily.

2.2. Methods

2.2.1. Model Selection and Construction

The objective of this study is to simulate a methane model based on a data-driven approach, given the complexity of the sources of methane in the atmosphere and the processes that affect it. As illustrated in Table 2, the study subjected several commonly employed models to empirical testing using sample data. Given that the relationship between methane and other data is not purely linear, linear regression is insufficient to meet the demand. Moreover, methods based on neural networks are relatively complex and have low accuracy. The application of SVM and Gaussian process regression algorithms for testing proved to be excessively time-consuming, yet yielded unsatisfactory results. The random forest method employed in this study has the potential to achieve high accuracy and to consume a relatively short time.
The random forest algorithm, a widely used method in integration learning, was selected for this purpose. The method is based on decision trees for parallel integration, which is a simple and efficient approach to calculating the importance of each variable in comparison to the complexity of deep learning. It has superior performance and generalization ability compared to a single decision tree due to its integration learning capabilities, as evidenced by studies [36,37,38].
As shown in Figure 2, the number of original samples is assumed to be N by the random forest method. Randomly select data from the sample using the bootstrapping method to construct sub-sample datasets with the same size as the original sample. Bootstrapping means randomly extracting data from a sample, and then reinserting it into the original dataset, repeating the extraction until the requisite quantity is reached. Subsequently, sub-sample datasets (which may include duplicate data) are employed to train a decision tree. This process is repeated k times, resulting in the generation of k decision trees. The final outputs of the model are the results predicted by the k decision trees, which are combined, and the optimal result is determined by voting. Therefore, the parameters to consider when constructing the model are mainly those corresponding to the constructed trees, including NumTrees, MinLeafSize, Fboot, and the need to set the mode of the random forest to regression. NumTrees is the number of decision trees in the bag. Increasing the number of trees usually improves the stability and prediction accuracy of the model because more trees can reduce the variance of the model, but more trees will increase the computation time and resource requirements. MinLeafSize is the minimum number of leaf node observations. Each leaf has at least one observation per tree leaf. A smaller number of leaf nodes makes the model more complex and may capture more noise in the data, leading to overfitting. A larger number of leaf nodes makes the model simpler and may lead to underfitting. Fboot is the proportion of the total sample that is sampled at each self-service sampling. The diversity and generalization of the model can be improved by setting the appropriate proportion of samples, as each tree has a different set of samples.
The probability of a sample not being selected when constructing the model is ( 1 − 1 / N ) N , which converges to 1 / e ≈ 0.37 when N is infinite. Consequently, each sample has an approximate 37% probability of not being selected. With these Out-of-Bag (OOB) data, it is possible to estimate the performance of the model and determine the parameters of the model [39]. By calculating the number of trees in the random forest and the curve of OOB error, an appropriate value for the number of trees in the random forest can be determined to avoid underfitting and overfitting. Based on OOB error, after considering the accuracy and performance of the model, the study sets the number of trees in the model to 100 and the minimum number of observations for leaf nodes to 5.
The study employed satellite methane column concentration data from TROPOMI as the response variable, alongside other data such as meteorology and topography, which were used as estimation factors. The data from 2019 to 2021 (with an interval of eight days, which is the maximum allowable under the equipment performance limitations) was used to train and build the model within the China region.

2.2.2. Validation Method

The study assesses the model’s accuracy based on sample data using the 10-fold cross-validation method (CV method). The original samples were initially randomly divided into ten parts, and then one part was selected as the test sample. The remaining nine samples were used as training samples to train the model and predict the results corresponding to the test samples. Furthermore, each sample was cycled as the test sample to obtain the predicted results corresponding to all samples, thereby enabling an assessment of the overall accuracy of the simulated methane column concentrations. This study divided the raw data samples by the samples, the time of the samples, and the spatial location of the samples, given that XCH4 exhibits a certain periodicity in time with significant seasonal variations [40]. In the case of the sample-based 10-fold cross-validation, the primary validation criterion is the fit of the model to the sample. Consequently, the data in the sample are randomly selected to obtain the divided sample. The missing data from TROPOMI should be filled in. Therefore, this study divides the sample into 10 groups for 10-fold cross-validation based on the spatial location of the data. This allows the predictive ability of the model to be assessed in space, that is to say, the accuracy of XCH4 measurements in areas where TROPOMI products are unavailable. However, if the sample data are divided randomly based on time, the results are consistent with the sample-based 10-fold cross-validation. The atmospheric system is dynamic, and the gas content within the atmosphere does not change abruptly under normal conditions. Consequently, the atmospheric methane concentration is relatively consistent over adjacent time and location. To address this, a novel approach was devised to initially divide the original data into 100 equal parts according to the latitude and longitude of China. The 100 data sets were then divided into 10 data sets at random to obtain the final division results. The 10-fold cross-validation based on sample time primarily assesses the model’s predictive capacity concerning time series, specifically its ability to identify seasonal fluctuations in XCH4. Similarly, the original data were initially sorted by time and then divided into 100 equal parts in chronological order. Finally, the 100 parts were randomly divided into 10 parts to obtain the final division results.
Following the modeling and estimation of the estimated methane concentration, the study validates its accuracy by comparing it with the monitoring data from the ground stations TCCON. The TCCON employs ground-based Fourier transform infrared (FTIR) spectrometers to measure the direct solar radiation in the near-infrared spectral region [41], from which the total column-averaged dry-air mole fractions of XCH4 have been retrieved [42]. The data employed in the study were derived from the TCCON Xianghe and Hefei stations. The Hefei station is situated at 31.9°N and 117.17°E, at an altitude of 30 m above sea level. It is located in a rural area northwest of Hefei, Anhui Province, China, in an area characterized by flat terrain and proximity to lakes. The station is a ground-based high-resolution Fourier transform spectrometer (FTS) that was completed in 2014 [43]. The longitude and latitude of the Xianghe station are 39.8°N and 116.96°E, respectively, and the altitude is 36 m above sea level. The station is situated in the Beijing–Tianjin–Hebei region, one of the most densely populated and economically active areas in China. It is also used for gas monitoring using the FTIR spectrometer.

3. Results

3.1. Model Performance

3.1.1. Seamless XCH4 Verification by the CV Methods

After rigorous quality screening and interpolation, the spatial coverage of the TROPOMI XCH4 data used in this study was limited. The average coverage of these data within China between 2019 and 2021 is approximately 2%. As shown in Figure 3a, the study presents the distribution of the TROPOMI data as of 1 October 2019. The figure shows that the XCH4 data from TROPOMI on this date is mainly distributed in Xinjiang, Gansu, Inner Mongolia, Northeast China, and some areas in South and Central China, and the data volume is low. Figure 3b shows an example of the spatial distribution of XCH4 data obtained by our model on a certain day (1 October 2019). The figure shows that the data gaps are well-filled by the model. The model can provide XCH4 data for any location in the country and has good agreement with related studies in terms of spatial distribution [26]. The figure also shows that the model can reconstruct the XCH4 information in the missing areas in the light and highly polluted (gray and red) areas.
In addition, we examined the fitting accuracy of the model by tabulating the results of the sample-based stochastic 10-fold cross-validation (Figure 3c), time-series-based 10-fold cross-validation (Figure 3d), and spatial location-based 10-fold cross-validation (Figure 3e). The figure shows that R2 = 0.97, RMSE = 9.5 (ppb), and MAE = 6.9 (ppb) for the 10-fold cross-validation sample. The R2 = 0.84, RMSE = 19.6 (ppb), and MAE = 14.9 (ppb) for the 10-fold cross-validation of time, and R2 = 0.93, RMSE = 13.4 (ppb), and MAE = 9.9 (ppb) for the 10-fold cross-validation of spatial location. The validation results indicate that the XCH4 data fitted by the model are highly consistent with the original XCH4 data and the model runs effectively (10CV-Sample R2 = 0.97). The R2 of the temporal cross-validation is slightly lower than that of the sample cross-validation. This finding is probably due to the involvement of time as a factor in the model construction, resulting in a heavier sensitivity and dependence of the model on time. In this study, we mainly consider the reconstruction of the original data with low coverage, using the model to improve its coverage. The results of the spatial 10-fold cross-validation (10CV-Space R2 = 0.93) show that the model has excellent prediction ability at different spatial locations and can be used to reconstruct the original data for XCH4 concentration data in the surface domain.

3.1.2. XCH4 Verification by TCCON Site-Based Measurements

The model predictions were also validated based on data from the TCCON station (Figure 4). According to a related study [44], the average deviation of the TROPOMI and station XCH4 data at the Xianghe station is 0.60%, which is approximately 20 ppb. We interpolated the XCH4 results of the model, fit from 2019 to 2021, to obtain XCH4 data at the Hefei and Xianghe stations, as shown in Figure 4b and Figure 4c, respectively. At the Hefei station, the data sampling frequency of the TCCON station is lower. The mean XCH4 value is 1892 (ppb), and our estimated mean XCH4 value is 1888 (ppb), a difference of 4 (ppb), which is approximately equal to 0.2%. The data are sampled more frequently at the Xianghe station, where the mean value of XCH4 is 1890 (ppb), and the mean value of XCH4 estimated by the model is 1889 (ppb), with a deviation of 1 (ppb), which is approximately equal to 0.05%. The figure demonstrates that the model-fitted XCH4 data exhibit reduced accuracy in the reconstruction of extreme values. This may be attributed to the resolution factor. The data measured on-station represents data within a radius of the site, while the XCH4 data used for comparison is obtained through data interpolation within 100 km of the station. This range is selected based on relevant studies [44], which may cause bias. In general, the fitting results of the model are reliable and can satisfy the high coverage reconstruction of XCH4.

3.2. Spatiotemporal Characteristics of Surface XCH4

3.2.1. XCH4 Concentrations in Different Seasons in China

Figure 5 shows the changes in XCH4 in four seasons from 2019–2021, with March-May as spring, June–August as summer, September–November as autumn, and December and January–February of the following year as winter. The methane concentration shows a trend of increasing and decreasing throughout the year, but increases in total; the lowest methane concentration is observed in spring and the highest in autumn.
The average methane concentration distribution for the four seasons from 2019 to 2021 was plotted to explore the spatial and temporal variation in the methane concentration in the Chinese region, as shown in Figure 6. The images in each column in Figure 6 show that for a single season, the methane concentration increases steadily with increasing time. The images in each row in Figure 6 show that for a single year, the methane concentration shows a trend of increasing and then decreasing, with the lowest in spring and the highest in autumn. Spatially, the images show a distribution of high in the southeast and low in the northwest. Relevant studies have shown that in the China region, methane emissions from wetlands and rice cultivation dominate during the summer and autumn seasons due to extensive rice cultivation [45,46]. This leads to the emergence of a large number of methane emission sources, which causes the methane concentration in the China region to show the above spatio-temporal distribution. In the following, the study further analyses the possible specific reasons for this situation by examining certain high-value distribution areas.

3.2.2. XCH4 Concentrations in Typical Regions of China

The model was employed to generate a daily, full-coverage (100%), high-resolution (5 km), and high-quality XCH4 dataset for the China region from 2019 to 2021. This was achieved by synthesizing monthly and annual data by counting the daily data. The model-simulated data exhibited an effective fit with the ground station observations, with an overall deviation of approximately 0.2%. The results demonstrate that our dataset can be employed to examine the spatial and temporal variability of XCH4 over the terrestrial region of China.
Figure 7 presents the distribution of the annual average XCH4 concentrations in China in 2021, with a focus on the main high-concentration regions. In China, the distribution of XCH4 concentration is generally high in the east and low in the west. The high concentration areas, as illustrated in the subplot of the figure, are primarily situated in the coastal region, including the Pearl River Delta and Yangtze River Delta. In contrast, inland regions exhibit a more uniform distribution across the Sichuan Basin, Hebei, and Central China (the larger lakes in this region include Dongting Lake, Poyang Lake, and Hong Lake). These regions exhibit a common feature, namely a more developed water system and a flatter terrain.
When considering the sources of atmospheric methane, we find that rice cultivation and wetlands are very important emission sources, and related studies show that wetlands are the largest single source of XCH4, accounting for approximately 20–40% of the total XCH4 emissions—approximately 70% of which is obtained from the southern and tropical regions; rice cultivation occupies 6% to 20% of the total XCH4 emissions, most of which is obtained from higher temperature regions [47]. In the high-value regions in Figure 7, the water systems are more developed. Thus, methane from humidity may be an important source of atmospheric methane in these regions, resulting in a higher overall XCH4 concentration in eastern China than in the West. Moreover, the methane concentrations in the Yangtze River Delta and Pearl River Delta regions are significantly higher than in the surrounding areas. At the same time, rice cultivation, which is more widely performed in these regions, requires large amounts of water, and it produces more methane in southern China than in northern China. Thus, methane concentrations are significantly higher in the Sichuan basin and central China than in the surrounding areas. Temperature is also an important factor affecting methane emission from rice fields and wetlands [48], contributing to the phenomenon of higher methane concentration in the south and lower in the north.
Atmospheric methane is also heavily influenced by anthropogenic factors. According to EDGAR [49], fossil fuel exploitation and combustion are important sources of emissions in China, resulting in regional areas of high XCH4 concentrations in parts of Xinjiang due to fossil fuel exploitation and other circumstances. In the case of Hebei, although emission sources, such as rice cultivation, are limited, the high population density in the region has a high demand for fossil fuels, such as natural gas. Thus, the methane concentration is affected, resulting in a higher local methane concentration than the surrounding areas.
The most important removal process of atmospheric methane is its oxidation reaction with OH in the troposphere, especially in the tropics where light is abundant and temperature is high [50], considering the sink of atmospheric methane. Accordingly, the atmospheric methane scavenging process is stronger in regions close to the equator, leading to the possibility of lower methane concentrations in the southern than in the northern regions, under other similar conditions. The reaction with free chloride ions in the marine boundary layer was found to be another sink for methane [51], which may affect some of the atmospheric methane concentrations. As shown in Figure 7, some southern coastal regions of the Yangtze River Delta have slightly lower methane concentrations than those in the north, probably due to a combination of source sinks of atmospheric methane.

3.2.3. XCH4 Changes during the COVID-19 Pandemic

During the COVID-19 pandemic, local human activity was significantly curtailed. To ascertain the impact of this isolation period on local atmospheric methane concentrations, we conducted an investigation. However, the changes in methane concentrations over a short period are small and are susceptible to the influence of meteorological conditions. Based on the previous validation of the data simulated by the model, the error between the methane concentration data from TROPOMI and the TCCON ground site is approximately 0.6%. In contrast, the base is approximately 1900 (ppb). In this case, deviations may be observed between the daily methane concentrations and the actual results. Thus, we only analyze the volatility of methane concentrations to attenuate the effect of errors. However, the increasing trend of methane concentration dramatically affects the fluctuation curve of the time series of methane concentration. In contrast, the growth rate of the methane concentration is relatively stable in a short period (in years) [52]. Therefore, we assume that the natural growth rate of methane concentration is a constant value, and we perform a linear fit of the three-year Wuhan area methane. The methane concentration data were linearly fitted to obtain the increase curve and then subtracted from the corresponding values on the growth curve to obtain the methane concentration change data other than the natural growth. However, this approach would result in negative values in the concentration change data. We also performed an overall offset of the results to obtain the fluctuation curve of the methane concentration from 2019 to 2021, and the results are shown in Figure 8. The final coefficient of the fitted straight line is 0.5, and the constant is approximately 1848 (ppb).
Figure 8 shows the time-series variation in the daily three-day moving average of the methane concentration fluctuations in Wuhan for four periods around the Chinese Lunar New Year from 2019 to 2021; they are classified into different epidemic phases according to related studies [53]. In the first period, before the outbreak in 2019, the fluctuations of the methane concentrations were consistent with those in 2020 and 2021. During the lockdown period, due to the COVID-19 pandemic, the methane concentration fluctuation profile was relatively stable in the first two weeks. However, after two weeks, a decreasing trend began to appear, whereas the fluctuation profile of methane concentration remained stable in 2020 and 2021. From 24 to 72 days after the New Year, the impact of the lockdown starts to appear clearly, and the methane concentration fluctuation curve continuously shows a decreasing trend in 2019. The lockdown caused the local methane concentration fluctuation to decrease significantly, whereas the fluctuation curve remained consistent in the remaining years. Moreover, the isolation period ends approximately 72 days into the new year, and the methane concentration in 2019 rapidly increases to levels similar to those in 2020 and 2021 and remains high for some time. Ninety-six days after the new year, the fluctuations of methane concentrations in 2019 and from 2020 to 2021 tend to be consistent. The figure shows that the fluctuation of methane concentration decreases significantly during the new crown pneumonia isolation period but increases rapidly and returns to normal levels afterwards. Some related studies on the effect of the COVID-19 lockdown on the atmosphere are consistent with our results; the COVID-19 lockdown causes a significant downward trend in the atmospheric methane concentration. However, this effect rapidly diminishes after the isolation period, and the atmospheric methane concentration recovers [54,55]. The phenomenon may be attributed to the notable reduction in the impact of human activities on methane during the period of the COVID-19 lockdown. The production activities before and during the lockdown are compared, to analyze to some extent the degree of influence of different human activities on atmospheric methane. The finding indicates that the dataset can detect the abnormal variation in atmospheric methane concentration.

3.2.4. XCH4 Concentrations in Typical Days in China

Figure 9a presents the changes in XCH4 levels during the Chinese New Year period and the periods preceding and following the New Year in 2019. The two weeks preceding and following the Chinese New Year (5 February 2019) are regarded as the Chinese New Year period. The two-week periods preceding and following this period are designated as the pre- and post-Chinese New Year periods. The XCH4 data during this period demonstrate statistically significant changes in the methane concentrations throughout China and the regions of interest before and after the Chinese New Year. This is illustrated in Figure 9a. In winter, the formation of snow creates low oxygen and other environments, which in turn promote methane production by anaerobic bacteria. However, low temperatures inhibit various chemical processes in the source sink of methane, and methane emissions from the vegetation are substantially reduced in winter [56]. Consequently, the change in methane throughout the Chinese region is difficult to determine. The results show that the XCH4 concentration decreases throughout China during the Chinese New Year period. This condition is probably due to the gradual closure of most factories and the reduction in human production and consumption activities during the Chinese New Year period, thereby reducing anthropogenic methane emissions. However, the four regions with high methane concentrations show a continuous increase during this period, consistent with the previous section. This may be because human industrial activity decreased during this period, but other energy activities did not, and natural environmental factors, such as the water systems and temperatures during the winter months, also contributed to methane production [57,58]. As a result, the impact of the Spring Festival shutdown on methane production is reduced, allowing the methane concentrations in these regions to continue to exhibit an increasing trend. The results indicate that the methane concentrations in Central China, Yangtze River Delta, and Pearl River Delta regions are significantly higher than those in the Sichuan Basin, which may be more influenced by their surrounding water systems. This may be due to nutrient loading caused by human activities, which promotes the generation of methane [59].
Similarly, Figure 9b shows a phenomenon consistent with the above in the variation in XCH4 in the time scale around the National Day, which is considered the time point (1 October 2019). The week commencing on National Day is designated as the National Day period. The preceding and subsequent weeks are designated as the pre and post-National Day periods, respectively. Furthermore, the XCH4 data during this period are employed to ascertain the fluctuations in methane concentrations near National Day in China and the regions of interest. The methane concentrations in China and the regions of interest were obtained before and following National Day. The figure illustrates that the methane concentration in China declined during this period, yet remains stable. In contrast, the methane concentration in Central China, the Yangtze River Delta, and the Pearl River Delta is considerably higher than the average and continues to demonstrate an increasing trend. In the Sichuan Basin region, the methane concentrations peaked during the National Day period and declined before and after the National Day. Given that the Sichuan Basin is predominantly terrestrial and more susceptible to human influence than the other three regions, it should be noted that fluctuations in XCH4 concentrations in the short term may be due to changes in meteorological conditions [53].

4. Discussion

4.1. Comparison with Other XCH4 Products

In China, studies have mainly focused on the analysis of methane concentrations using existing satellite products or ground monitoring data, but have not filled the gaps in the methane concentration data [34]. However, the existing methods for methane concentration estimation do not obtain methane concentration data with good spatial and temporal resolution at the same time [60,61]. In comparison, our results provide better spatial coverage than satellite and ground-based monitoring products [15,18], better temporal continuity than emission inventory acquisition results [62], and the detection of anomalous changes for small regions [34]. Compared with the methane concentration results simulated by atmospheric chemistry models at the global spatial scale, the research results have a better spatial resolution [32].
Compared with methane products from ground stations or satellite monitoring, the methane concentration data products obtained by the proposed method have wide coverage, complete coverage in space with high resolution (0.05°), and strong continuity. On the contrary, methane data products with good spatial and temporal resolution are difficult to obtain from the ground station and satellite monitoring. Ground station monitoring has high accuracy and strong continuity, but the number of stations is small, and the distribution is sparse [14]. Satellite monitoring has a wide coverage, but the existing products from methane monitoring satellites have various deficiencies in the spatial coverage due to the influence of weather on the available data [15,16,17,18,19].
Compared with the products of methane concentration measured by bottom-up emission inventory methods and top-down atmospheric chemistry modeling methods, the proposed method directly obtains methane concentration data. Moreover, the data can have high timeliness and do not require a priori knowledge, and the methodology is simpler. But at the same time, it also makes it difficult to explore the science behind this method. The bottom-up method, based on emission inventories, can calculate methane emissions from different sources and can analyze the sources and destinations of atmospheric methane more directly. However, this method obtains indirect methane concentrations and relies on high-precision activity data and emission factors for estimation, due to the complexity of methane sources. Therefore, its products have lower temporal resolution, and the emission factors obtained are difficult to apply to all situations [62]. The top-down methods based on atmospheric chemistry models are commonly used to simulate methane concentrations globally, and can obtain methane concentration products with full global coverage, but their principles are complex [32].

4.2. Disadvantage of the XCH4 Model

The advantages of this study are clear, but the disadvantages are also evident due to the limitations of the proposed method. The random forest model was selected for this study. Although the parameters were adjusted by constraining them through the OOB errors in the previous section, the model was not specifically used for methane concentration prediction, leading to some shortcomings in the model results and difficulty in confirming whether the factors considered in this study could characterize the methane concentration more completely. In addition, Figure 4 shows that the model does not perform effectively in simulating predictions at extreme and very small values. Moreover, the method is data-driven and can be greatly influenced by the raw data. Figure 7 shows that the model predictions are less than perfect in terms of detail, due to the coarse resolution of the raw data.
Furthermore, the absence of genuine methane concentration observations has compelled the study to rely solely on TCCON field data to validate the estimated outcomes. Nevertheless, the data from these stations are unable to represent all regions. For instance, regions characterized by high albedo or elevated surfaces may also be contributors to the absence of satellite data, in addition to cloud cover. However, it is challenging to ascertain the accuracy of these regions through research.

4.3. Potential Applications of the XCH4 Seamless Products

Related studies have shown that the current source sink of methane is constantly changing, and the changes in methane concentration are difficult to relate directly to the corresponding source [63]. Moreover, different methane products are subject to significant uncertainties. The top-down inversion approach has an uncertainty of 30% in the results, and the emissions calculated by the bottom-up approach are overestimated compared with the actual emissions [8]. By contrast, our results are based on a data-driven approach that reduces the impact of uncertainty in methane sources. The overall deviation of methane results simulated by our method is less than 1% compared with the data from the TCCON stations. The 5 km atmospheric XCH4 concentration dataset significantly improves the existing XCH4 concentration data in China. It is potentially valuable for studying greenhouse gas’s spatial and temporal distribution and dynamics. According to this study, the data cover the China region seamlessly (Figure 7). Thus, this dataset can be used to conduct a study on the spatial distribution of XCH4. In addition, the data are in good agreement with the ground station data in terms of time series (Figure 4), and this dataset can be used to analyze and study the time series variation in XCH4. Finally, this dataset can detect the changes in methane concentration during the COVID-19 and holiday periods (Figure 8 and Figure 9). Thus, it can be used to detect the changes in methane concentration due to particular situations on the one hand, and on the other hand to detect whether special situations occur, and the corresponding locations and times, based on the changes in methane concentration. Therefore, the dataset of this study can be used to explore the geographic location of high methane emission sources accordingly, or to measure methane emissions by statistical methods. Monthly or annual methane data can also be calculated using statistical methods.

5. Conclusions

The traditional bottom-up or top-down methods cannot simultaneously obtain methane concentration data with high spatial and temporal resolution. In this study, a random forest model is used to simulate methane concentration data and achieve the reconstruction of TROPOMI methane concentration data with high coverage. It is also used to obtain a daily 5 km resolution atmospheric methane concentration dataset with full spatial coverage (100%) from 2019 to 2021 in mainland China. The proposed model considers temporal continuity and spatial coverage compared to the traditional methods. Three 10-CV approaches, namely, sample-, spatial-, and temporal-based CVs, are used to assess and compare the models’ performance. Methane concentration simulation models were found to have ultra-high accuracy (Sample-10 CV: R2 = 0.97, RMSE = 9.5 ppb, and MAE = 6.9 ppb; Space-10 CV R2 = 0.93, RMSE = 13.4 ppb, and MAE = 9.9 ppb). The final atmospheric methane concentration dataset in mainland China was then obtained using the model predictions and compared with the monitoring data from TCCON sites (Xianghe and Hefei) for validation. The overall deviation of methane results simulated by our method is less than 1% compared with the data from TCCON stations. Therefore, the proposed approach can be used to estimate the methane emissions and explore the time series variation in XCH4. The coverage of XCH4 data predicted by the model is 100%, providing information on the distribution of XCH4 in different regions. Through experimental analysis, the predicted XCH4 is sensitive to changes in methane concentration due to special conditions. Thus, it can detect the source of methane emissions or the occurrence of special conditions to a certain extent.
The comprehensive performance of the methane concentration data obtained from the research is, to some extent, superior to the widely used traditional bottom-up or top-down method models. The newly generated daily full-coverage 5 km methane concentration products have potential value for applications in carbon emission source monitoring, carbon emission measurement, and the spatial and temporal variation analysis of greenhouse gases.
In the future, the predictive power of the proposed model will be examined at different time scales (e.g., monthly, seasonal, and annual scales) and space scales, and compared with other related methane products. Subsequently, further exploration will be conducted to improve the spatiotemporal resolution of methane data, and the application potential of data in methane point source emissions and detection will be studied.

Author Contributions

Conceptualization, J.H.; Methodology, Z.J. and W.W.; Validation, J.H.; Investigation, Z.J. and J.H.; Writing—original draft, Z.J.; Writing—review & editing, W.W.; Visualization, J.H.; Supervision, W.W. All authors have read and agreed to the published version of the manuscript.

Funding

This study was supported by the Fundamental Research Funds for the Central Universities of Central South University, the National Natural Science Foundation of China (42371392), Basic Science-Center Project of National Natural Science Foundation of China (72088101), Natural Science Foundation of Hunan Province, China (2023JJ30660), Department of Natural Resources of Hunan Province (2022) No.5, and the Key Program of the National Natural Science Foundation of China (41930108). We are also grateful to the Data Center of US NASA (https://search.earthdata.nasa.gov/search, accessed on 18 October 2022), Climate Data Store (https://cds.climate.copernicus.eu/, accessed on 7 November 2022), USGS (http://srtm.csi.cgiar.org/srtmdata/, accessed on 12 April 2022), and TCCON (https://tccondata.org/, accessed on 1 November 2022) for providing the datasets used here.

Data Availability Statement

The DEM data are available from the USGS (http://srtm.csi.cgiar.org/srtmdata/, accessed on 12 April 2022); The ERA5 data are available from the Climate Data Store (https://cds.climate.copernicus.eu/, accessed on 7 November 2021); The site data are available from the TCCON (https://tccondata.org/, accessed on 31 October 2022); The Methane mixing ratio data are available from the Data Center of US NASA (https://search.earthdata.nasa.gov/search, accessed on 15 September 2022). The results of this study can be obtained from https://doi.org/10.6084/m9.figshare.25954993, accessed on 10 May 2024.

Acknowledgments

We express our sincere gratitude to those who provided support and advice for this article.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Skeie, R.B.; Hodnebrog, Ø.; Myhre, G. Trends in atmospheric methane concentrations since 1990 were driven and modified by anthropogenic emissions. Commun. Earth Environ. 2023, 4, 317. [Google Scholar] [CrossRef]
  2. Basu, S.; Lan, X.; Dlugokencky, E.; Michel, S.; Schwietzke, S.; Miller, J.B.; Bruhwiler, L.; Oh, Y.; Tans, P.P.; Apadula, F.; et al. Estimating emissions of methane consistent with atmospheric measurements of methane and δ13C of methane. Atmos. Chem. Phys. 2022, 22, 15351–15377. [Google Scholar] [CrossRef]
  3. Summary for Policymakers. In Climate Change 2013—The Physical Science Basis: Working Group I Contribution to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change; Intergovernmental Panel on Climate Change (Ed.) Cambridge University Press: Cambridge, UK, 2014; pp. 1–30. [Google Scholar]
  4. Rotmans, J.; Swart, R.J.; Vrieze, O.J. The role of the CH4-CO-OH cycle in the greenhouse problem. Sci. Total Environ. 1990, 94, 233–252. [Google Scholar] [CrossRef]
  5. Peng, S.S.; Piao, S.L.; Bousquet, P.; Ciais, P.; Li, B.G.; Lin, X.; Tao, S.; Wang, Z.P.; Zhang, Y.A.; Zhou, F. Inventory of anthropogenic methane emissions in mainland China from 1980 to 2010. Atmos. Chem. Phys. 2016, 16, 14545–14562. [Google Scholar] [CrossRef]
  6. Fung, I.; John, J.; Lerner, J.; Matthews, E.; Prather, M.; Steele, L.P.; Fraser, P.J. Three-dimensional model synthesis of the global methane cycle. J. Geophys. Res. Atmos. 1991, 96, 13033–13065. [Google Scholar] [CrossRef]
  7. Fu, B.; Jiang, Y.; Chen, G.; Lu, M.; Lai, Y.; Suo, X.; Li, B. Unraveling the dynamics of atmospheric methane: The impact of anthropogenic and natural emissions. Environ. Res. Lett. 2024, 19, 064001. [Google Scholar] [CrossRef]
  8. Saunois, M.; Stavert, A.R.; Poulter, B.; Bousquet, P.; Canadell, J.G.; Jackson, R.B.; Raymond, P.A.; Dlugokencky, E.J.; Houweling, S.; Patra, P.K.; et al. The Global Methane Budget 2000–2017. Earth Syst. Sci. Data 2020, 12, 1561–1623. [Google Scholar] [CrossRef]
  9. Parsons, A.J.; Newton, P.C.D.; Clark, H.; Kelliher, F.M. Scaling methane emissions from vegetation. Trends Ecol. Evol. 2006, 21, 423–424. [Google Scholar] [CrossRef]
  10. Kirschbaum, M.; Bruhn, D.; Etheridge, D.; Evans, J.; Farquhar, G.; Gifford, R.; Paul, K.; Winters, A. Comment on the quantitative significance of aerobic methane release by plants. Funct. Plant Biol. 2006, 33, 521–530. [Google Scholar] [CrossRef]
  11. Keppler, F.; Hamilton, J.T.G.; Braß, M.; Röckmann, T. Methane emissions from terrestrial plants under aerobic conditions. Nature 2006, 439, 187–191. [Google Scholar] [CrossRef]
  12. Smith, P.; Reay, D.; Smith, J. Agricultural methane emissions and the potential formitigation. Philos. Trans. R. Soc. A Math. Phys. Eng. Sci. 2021, 379, 20200451. [Google Scholar] [CrossRef] [PubMed]
  13. Bizic, M. Phytoplankton photosynthesis: An unexplored source of biogenic methane emission from oxic environments. J. Plankton Res. 2021, 43, 822–830. [Google Scholar] [CrossRef]
  14. Hmiel, B.; Petrenko, V.V.; Dyonisius, M.N.; Buizert, C.; Smith, A.M.; Place, P.F.; Harth, C.; Beaudette, R.; Hua, Q.; Yang, B.; et al. Preindustrial 14CH4 indicates greater anthropogenic fossil CH4 emissions. Nature 2020, 578, 409–412. [Google Scholar] [CrossRef] [PubMed]
  15. Frankenberg, C.; Meirink, J.F.; van Weele, M.; Platt, U.; Wagner, T. Assessing Methane Emissions from Global Space-Borne Observations. Science 2005, 308, 1010–1014. [Google Scholar] [CrossRef] [PubMed]
  16. Aumann, H.H.; Chahine, M.T.; Gautier, C.; Goldberg, M.D.; Susskind, J. AIRS/AMSU/HSB on the Aqua mission: Design, science objective, data products, and processing systems. IEEE Trans. Geosci. Remote Sens. 2003, 41, 253–264. [Google Scholar] [CrossRef]
  17. Alexe, M.; Bergamaschi, P.; Segers, A.; Detmers, R.; Butz, A.; Hasekamp, O.; Guerlet, S.; Parker, R.; Boesch, H.; Frankenberg, C. Inverse modelling of CH4 emissions for 2010–2011 using different satellite retrieval products from GOSAT and SCIAMACHY. Atmos. Chem. Phys. 2015, 15, 113–133. [Google Scholar] [CrossRef]
  18. Sadavarte, P.; Pandey, S.; Maasakkers, J.D.; Lorente, A.; Borsdorff, T.; van der Gon, H.D.; Houweling, S.; Aben, I. Methane Emissions from Superemitting Coal Mines in Australia Quantified Using TROPOMI Satellite Observations. Environ. Sci. Technol. 2021, 55, 16573–16580. [Google Scholar] [CrossRef] [PubMed]
  19. de Gouw, J.A.; Veefkind, J.P.; Roosenbrand, E.; Dix, B.; Lin, J.C.; Landgraf, J.; Levelt, P.F. Daily Satellite Observations of Methane from Oil and Gas Production Regions in the United States. Sci. Rep. 2020, 10, 1379. [Google Scholar] [CrossRef] [PubMed]
  20. Pei, Z.; Han, G.; Mao, H.; Chen, C.; Shi, T.; Yang, K.; Ma, X.; Gong, W. Improving quantification of methane point source emissions from imaging spectroscopy. Remote Sens. Environ. 2023, 295, 113652. [Google Scholar] [CrossRef]
  21. Schneising, O.; Buchwitz, M.; Burrows, J.P.; Bovensmann, H.; Bergamaschi, P.; Peters, W. Three years of greenhouse gas column-averaged dry air mole fractions retrieved from satellite—Part 2: Methane. Atmos. Chem. Phys. 2009, 9, 443–465. [Google Scholar] [CrossRef]
  22. Kuze, A.; Kikuchi, N.; Kataoka, F.; Suto, H.; Shiomi, K.; Kondo, Y. Detection of Methane Emission from a Local Source Using GOSAT Target Observations. Remote Sens. 2020, 12, 267. [Google Scholar] [CrossRef]
  23. Chang, Y.; Deng, X.; Liu, H.; Ding, J.; Ww, H. Temporal and Spatial Distribution Character of CH4 Near Surface. Environ. Sci. Technol. 2017, 40, 161–166. [Google Scholar]
  24. Zhang, X.; Bai, W.; Zhang, P.; Wang, W. Spatiotemporal variations in mid-upper tropospheric methane over China from satellite observations. Chin. Sci. Bull. 2011, 56, 3321–3327. [Google Scholar] [CrossRef]
  25. Bergamaschi, P.; Frankenberg, C.; Meirink, J.F.; Krol, M.; Villani, M.G.; Houweling, S.; Dentener, F.; Dlugokencky, E.J.; Miller, J.B.; Gatti, L.V.; et al. Inverse modeling of global and regional CH4 emissions using SCIAMACHY satellite retrievals. J. Geophys. Res. Atmos. 2009, 114, D22301. [Google Scholar] [CrossRef]
  26. Zhang, X.; Jiang, H.; Wang, Y.; Han, Y.; Buchwitz, M.; Schneising, O.; Burrows, J.P. Spatial variations of atmospheric methane concentrations in China. Int. J. Remote Sens. 2011, 32, 833–847. [Google Scholar] [CrossRef]
  27. Meng, X.; Chang, H.; Wang, X. Methane Concentration Prediction Method Based on Deep Learning and Classical Time Series Analysis. Energies 2022, 15, 2262. [Google Scholar] [CrossRef]
  28. Rehman, S.U.; Husain, I.; Hashmi, M.Z.; Elashkar, E.E.; Khader, J.A.; Ageli, M. Forecasting and modeling of atmospheric methane concentration. Arab. J. Geosci. 2021, 14, 1667. [Google Scholar] [CrossRef]
  29. Liu, B.; Ma, X.; Guo, J.; Wen, R.; Li, H.; Jin, S.; Ma, Y.; Guo, X.; Gong, W. Extending the wind profile beyond the surface layer by combining physical and machine learning approaches. Atmos. Chem. Phys. 2024, 24, 4047–4063. [Google Scholar] [CrossRef]
  30. Yang, J.; Gan, R.; Luo, B.; Wang, A.; Shi, S.; Du, L. An Improved Method for Individual Tree Segmentation in Complex Urban Scenes Based on Using Multispectral LiDAR by Deep Learning. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2024, 17, 6561–6576. [Google Scholar] [CrossRef]
  31. Xu, J.; Liu, Q.; Wang, K.; Wang, Q.; Wang, L.; Liu, Y.; Li, M. Spatiotemporal variation in near-surface CH4 concentrations in China over the last two decades. Environ. Sci. Pollut. Res. Int. 2021, 28, 47239–47250. [Google Scholar] [CrossRef]
  32. Dalsøren, S.B.; Myhre, C.L.; Myhre, G.; Gomez-Pelaez, A.J.; Søvde, O.A.; Isaksen, I.; Weiss, R.F.; Harth, C.M. Atmospheric methane evolution the last 40 years. Atmos. Chem. Phys. 2016, 16, 3099–3126. [Google Scholar] [CrossRef]
  33. Zhang, D.; Liao, H.; Wang, Y. Simulated spatial distribution and seasonal variation of atmospheric methane over China: Contributions from key sources. Adv. Atmos. Sci. 2014, 31, 283–292. [Google Scholar] [CrossRef]
  34. Zhang, J.; Han, G.; Mao, H.; Pei, Z.; Ma, X.; Jia, W.; Gong, W. The Spatial and Temporal Distribution Patterns of XCH4 in China: New Observations from TROPOMI. Atmosphere 2022, 13, 177. [Google Scholar] [CrossRef]
  35. Qu, Z.; Jacob, D.J.; Shen, L.; Lu, X.; Zhang, Y.; Scarpelli, T.R.; Nesser, H.; Sulprizio, M.P.; Maasakkers, J.D.; Bloom, A.A.; et al. Global distribution of methane emissions: A comparative inverse analysis of observations from the TROPOMI and GOSAT satellite instruments. Atmos. Chem. Phys. 2021, 21, 14159–14175. [Google Scholar] [CrossRef]
  36. Breiman, L. Random forest. Mach. Learn. 2001, 45, 5–32. [Google Scholar] [CrossRef]
  37. Prasad, A.M.; Iverson, L.R.; Liaw, A. Newer Classification and Regression Tree Techniques: Bagging and Random Forests for Ecological Prediction. Ecosystems 2006, 9, 181–199. [Google Scholar] [CrossRef]
  38. He, J.; Wang, W.; Fu, M.; Wang, Y. Insights into Global Visibility Patterns: Spatiotemporal Distributions Revealed by Satellite Remote Sensing. J. Clean. Prod. 2024; in press. [Google Scholar] [CrossRef]
  39. Breiman, L. Out-of-Bag Estimation. 1996. Available online: https://api.semanticscholar.org/CorpusID:17166335 (accessed on 14 October 2023).
  40. Wu, X.; Zhang, X.; Chuai, X.; Huang, X.; Wang, Z. Long-Term Trends of Atmospheric CH4 Concentration across China from 2002 to 2016. Remote Sens. 2019, 11, 538. [Google Scholar] [CrossRef]
  41. He, J.; Wang, W.; Wang, N. Seamless Reconstruction and Spatiotemporal Analysis of Satellite-based XCO2 Incorporating Temporal Characteristics: A Case Study in China during 2015–2020. Adv. Space Res. 2024; in press. [Google Scholar]
  42. A method for evaluating bias in global measurements of CO2 total columns from space. Atmos. Chem. Phys. 2011, 11, 12317–12337. [CrossRef]
  43. Wang, W.; Tian, Y.; Liu, C.; Sun, Y.; Liu, W.; Xie, P.; Liu, J.; Xu, J.; Morino, I.; Velazco, V.A.; et al. Investigating the performance of a greenhouse gas observatory in Hefei, China. Atmos. Meas. Tech. 2017, 10, 2627–2643. [Google Scholar] [CrossRef]
  44. Yang, Y.; Zhou, M.; Langerock, B.; Sha, M.K.; Wang, P. New ground-based Fourier-transform near-infrared solar absorption measurements of XCO2, XCH4 and XCO at Xianghe, China. Earth Syst. Sci. Data 2020, 12, 1679–1696. [Google Scholar] [CrossRef]
  45. Zhang, G.; Xiao, X.; Dong, J.; Xin, F.; Zhang, Y.; Qin, Y.; Doughty, R.B.; Moore, B. Fingerprint of rice paddies in spatial–temporal dynamics of atmospheric methane concentration in monsoon Asia. Nat. Commun. 2020, 11, 554. [Google Scholar] [CrossRef]
  46. Ni, Q.; Zhou, M.; Wang, J.; Wang, T.; Wang, G.; Wang, P. Intercomparison of CH4 Products in China from GOSAT, TROPOMI, IASI, and AIRS Satellites. Remote Sens. 2023, 15, 4499. [Google Scholar] [CrossRef]
  47. Bloom, A.A.; Palmer, P.I.; Fraser, A.; Reay, D.S.; Frankenberg, C. Large-scale controls of methanogenesis inferred from methane and gravity spaceborne data. Science 2010, 327, 322–325. [Google Scholar] [CrossRef] [PubMed]
  48. Sass, R.L.; Fisher, F.M.; Turner, F.T.; Jund, M.F. Methane emission from rice fields as influenced by solar radiation, temperature, and straw incorporation. Glob. Biogeochem. Cycles 1991, 5, 335–350. [Google Scholar] [CrossRef]
  49. Crippa, M.; Guizzardi, D.; Banja, M.; Solazzo, E.; Muntean, M.; Schaaf, E.; Pagani, F.; Monforti, F.; Olivier, J.; Quadrelli, R.; et al. CO2 Emissions of All World Countries. JRC/IEA/PBL 2022 Report; Publications Office of the European Union: Luxembourg, 2022. [Google Scholar] [CrossRef]
  50. Thompson, A.M.; Cicerone, R.J. Atmospheric CH4, CO and OH from 1860 to 1985. Nature 1986, 321, 148–150. [Google Scholar] [CrossRef]
  51. Allan, W. Interannual variation of 13C in tropospheric methane: Implications for a possible atomic chlorine sink in the marine boundary layer. J. Geophys. Res. 2005, 110, D11306. [Google Scholar] [CrossRef]
  52. Nisbet, E.G.; Manning, M.R.; Dlugokencky, E.J.; Fisher, R.E.; Lowry, D.; Michel, S.E.; Myhre, C.L.; Platt, S.M.; Allen, G.; Bousquet, P.; et al. Very Strong Atmospheric Methane Growth in the 4 Years 2014–2017: Implications for the Paris Agreement. Glob. Biogeochem. Cycles 2019, 33, 318–342. [Google Scholar] [CrossRef]
  53. Wei, J.; Liu, S.; Li, Z.; Liu, C.; Qin, K.; Liu, X.; Pinker, R.T.; Dickerson, R.R.; Lin, J.; Boersma, K.F.; et al. Ground-Level NO2 Surveillance from Space Across China for High Resolution Using Interpretable Spatiotemporally Weighted Artificial Intelligence. Environ. Sci. Technol. 2022, 56, 9988–9998. [Google Scholar] [CrossRef]
  54. Gulyaev, E.; Antonov, K.; Markelov, Y.; Poddubny, V.; Shchelkanov, A.; Iurkov, I. Short-term effect of COVID-19 lockdowns on atmospheric CO2, CH4 and PM2.5 concentrations in urban environment. Int. J. Environ. Sci. Technol. 2022, 20, 4737–4748. [Google Scholar] [CrossRef] [PubMed]
  55. Liang, M.; Zhang, Y.; Ma, Q.; Yu, D.; Chen, X.; Cohen, J.B. Dramatic decline of observed atmospheric CO2 and CH4 during the COVID-19 lockdown over the Yangtze River Delta of China. J. Environ. Sci. 2023, 124, 712–722. [Google Scholar] [CrossRef]
  56. Bussmann, I.; Fedorova, I.; Juhls, B.; Overduin, P.P.; Winkel, M. Methane dynamics in three different Siberian water bodies under winter and summer conditions. Biogeosciences 2021, 18, 2047–2061. [Google Scholar] [CrossRef]
  57. Gar’kusha, D.N.; Fedorov, Y.A. Methane in the water and bottom sediments of the mouth area of the Severnaya Dvina River during the winter time. Oceanology 2014, 54, 160–169. [Google Scholar] [CrossRef]
  58. Savić, S.; Selakov, A.; Milošević, D. Cold and warm air temperature spells during the winter and summer seasons and their impact on energy consumption in urban areas. Nat. Hazards 2014, 73, 373–387. [Google Scholar] [CrossRef]
  59. Li, X.; Qi, M.; Gao, D.; Liu, M.; Hou, L. Switches of methane production pathways and emissions with human activity intensity in subtropical estuaries. J. Hydrol. 2022, 612, 128061. [Google Scholar] [CrossRef]
  60. Wang, K.; Zhang, J.; Cai, B.; Liang, S. Estimation of Chinese city-level anthropogenic methane emissions in 2015. Resour. Conserv. Recycl. 2021, 175, 105861. [Google Scholar] [CrossRef]
  61. Zhang, Y.; Fang, S.; Chen, J.; Lin, Y.; Chen, Y.; Liang, R.; Jiang, K.; Parker, R.J.; Boesch, H.; Steinbacher, M.; et al. Observed changes in China’s methane emissions linked to policy drivers. Proc. Natl. Acad. Sci. USA 2022, 119, e2202742119. [Google Scholar] [CrossRef] [PubMed]
  62. Gong, S.; Shi, Y. Evaluation of comprehensive monthly-gridded methane emissions from natural and anthropogenic sources in China. Sci. Total Environ. 2021, 784, 147116. [Google Scholar] [CrossRef]
  63. Turner, A.J.; Frankenberg, C.; Kort, E.A. Interpreting contemporary trends in atmospheric methane. Proc. Natl. Acad. Sci. USA 2019, 116, 2805–2813. [Google Scholar] [CrossRef]
Remotesensing 16 02525 g001
Figure 1. Research area in this study. The site-based measurements (TCCON) from the Hefei (119.17°E 31.9°N) and Xianghe stations (119.17°E 31.9°N) are used as the validation data. Background color indicates surface elevation.
Figure 1. Research area in this study. The site-based measurements (TCCON) from the Hefei (119.17°E 31.9°N) and Xianghe stations (119.17°E 31.9°N) are used as the validation data. Background color indicates surface elevation.
Remotesensing 16 02525 g001
Remotesensing 16 02525 g002
Figure 2. Random Forest Flowchart. After inputting sample data, bootstrapping is used to obtain different sub-sets of data, and decision trees are constructed separately. When estimating, the values of each decision tree are counted and averaged to obtain the final result.
Figure 2. Random Forest Flowchart. After inputting sample data, bootstrapping is used to obtain different sub-sets of data, and decision trees are constructed separately. When estimating, the values of each decision tree are counted and averaged to obtain the final result.
Remotesensing 16 02525 g002
Remotesensing 16 02525 g003
Figure 3. Changes in the coverage of predicted results and cross-validation results of the model. (a,b) indicate the XCH4 (ppb) estimated by TROPOMI and the model on 1 October 2019. (c–e) show the result of the model for sample and space-time 10-fold cross-validation.
Figure 3. Changes in the coverage of predicted results and cross-validation results of the model. (a,b) indicate the XCH4 (ppb) estimated by TROPOMI and the model on 1 October 2019. (c–e) show the result of the model for sample and space-time 10-fold cross-validation.
Remotesensing 16 02525 g003
Remotesensing 16 02525 g004
Figure 4. Comparison of model-predicted XCH4 data with XCH4 data at the TCCON sites. (a) indicates the location of the TCCCON sites and the spatial distribution of XCH4 simulated by the model, (b,c) indicate the comparison of TCCON observations and model-estimated values at the Hefei station and Xianghe site. The horizontal axis represents the number of days starting from 1 January 2019.
Figure 4. Comparison of model-predicted XCH4 data with XCH4 data at the TCCON sites. (a) indicates the location of the TCCCON sites and the spatial distribution of XCH4 simulated by the model, (b,c) indicate the comparison of TCCON observations and model-estimated values at the Hefei station and Xianghe site. The horizontal axis represents the number of days starting from 1 January 2019.
Remotesensing 16 02525 g004
Remotesensing 16 02525 g005
Figure 5. Comparison of the average XCH4 concentrations (ppb) during different seasons in China and four typical regions. CC, SCB, PRD, and YRD stand for Central China, the Sichuan Basin, the Pearl River Delta, and the Yangtze River Delta, respectively.
Figure 5. Comparison of the average XCH4 concentrations (ppb) during different seasons in China and four typical regions. CC, SCB, PRD, and YRD stand for Central China, the Sichuan Basin, the Pearl River Delta, and the Yangtze River Delta, respectively.
Remotesensing 16 02525 g005
Remotesensing 16 02525 g006
Figure 6. Seasonal mean changes in the XCH4 concentrations (ppb) in China from 2019 to 2021. (a–d) indicates 2019, (e–h) indicates 2020 and (i–l) indicates 2021.
Figure 6. Seasonal mean changes in the XCH4 concentrations (ppb) in China from 2019 to 2021. (a–d) indicates 2019, (e–h) indicates 2020 and (i–l) indicates 2021.
Remotesensing 16 02525 g006
Remotesensing 16 02525 g007
Figure 7. Spatial distributions of annual mean national and regional (zoomed-in subplots) XCH4 concentrations (ppb) in China. Regions shown in panel are Central China (CC), the Sichuan Basin (SCB), the Pearl River Delta (PRD), and the Yangtze River Delta (YRD).
Figure 7. Spatial distributions of annual mean national and regional (zoomed-in subplots) XCH4 concentrations (ppb) in China. Regions shown in panel are Central China (CC), the Sichuan Basin (SCB), the Pearl River Delta (PRD), and the Yangtze River Delta (YRD).
Remotesensing 16 02525 g007
Remotesensing 16 02525 g008
Figure 8. Time series of changes in the three-day moving average of the daily XCH4 concentrations in Wuhan, China before and after the Lunar New Year from 2019 to 2021. Four time periods, “Before the outbreak”, “Lockdown”, “Impact of the lockdown”, and “Return to normal level”, are divided according to the lockdown process in the Wuhan area in 2019 due to the COVID-19 pandemic. The corresponding dates for the Chinese Lunar New Year from 2019 to 2021 are 5 February 2019, 25 January 2020, and 12 February 2021.
Figure 8. Time series of changes in the three-day moving average of the daily XCH4 concentrations in Wuhan, China before and after the Lunar New Year from 2019 to 2021. Four time periods, “Before the outbreak”, “Lockdown”, “Impact of the lockdown”, and “Return to normal level”, are divided according to the lockdown process in the Wuhan area in 2019 due to the COVID-19 pandemic. The corresponding dates for the Chinese Lunar New Year from 2019 to 2021 are 5 February 2019, 25 January 2020, and 12 February 2021.
Remotesensing 16 02525 g008
Remotesensing 16 02525 g009
Figure 9. Comparison of the average XCH4 concentrations (ppb) before, during, and after (a) Spring Festival and (b) National Day holidays in China and four typical regions. CC, SCB, PRD, and YRD stand for Central China, the Sichuan Basin, the Pearl River Delta, and the Yangtze River Delta, respectively.
Figure 9. Comparison of the average XCH4 concentrations (ppb) before, during, and after (a) Spring Festival and (b) National Day holidays in China and four typical regions. CC, SCB, PRD, and YRD stand for Central China, the Sichuan Basin, the Pearl River Delta, and the Yangtze River Delta, respectively.
Remotesensing 16 02525 g009
Table 1. Data information.
Table 1. Data information.
Used DataVariableUnitSpatial ResolutionTime ResolutionSource
Methane mixing ratioXCH4ppb7 km × 7 km
(5.5 km × 5.5 km)		dayESA
DEMDEMM90 m——USGS
2 m temperatureT2MM0.25° × 0.25°hourERA5
Total column ozoneTCO3K0.25° × 0.25°hourERA5
Total column waterTCWKg/m20.25° × 0.25°hourERA5
Surface net solar radiationSSRKg/m20.25° × 0.25°hourERA5
Top net solar radiationTSRJ/m20.25° × 0.25°hourERA5
Methane mixing ratioSXCH4ppb——hourTCCON
Table 2. Different model effects.
Table 2. Different model effects.
Model TypeR2RMSE (ppb)Time (s)
Linear regression0.5225.32.62
SVM0.7747.55008
Gaussian process regression0.8812.733043
FFNN0.6322.1446.26
Decision Tree0.9111.1648.08
This study0.979.5575.95
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Jin, Z.; He, J.; Wang, W. Monitoring Methane Concentrations with High Spatial Resolution over China by Using Random Forest Model. Remote Sens. 2024, 16, 2525. https://doi.org/10.3390/rs16142525

AMA Style

Jin Z, He J, Wang W. Monitoring Methane Concentrations with High Spatial Resolution over China by Using Random Forest Model. Remote Sensing. 2024; 16(14):2525. https://doi.org/10.3390/rs16142525

Chicago/Turabian Style

Jin, Zhili, Junchen He, and Wei Wang. 2024. "Monitoring Methane Concentrations with High Spatial Resolution over China by Using Random Forest Model" Remote Sensing 16, no. 14: 2525. https://doi.org/10.3390/rs16142525

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop