Monitoring of Atmospheric Carbon Dioxide over Pakistan Using Satellite Dataset

Abstract

Satellites are an effective source of atmospheric carbon dioxide (CO₂) monitoring; however, city-scale monitoring of atmospheric CO₂ through space-borne observations is still a challenging task due to the trivial change in atmospheric CO₂ concentration compared to its natural variability and background concentration. In this study, we attempted to evaluate the potential of space-based observations to monitor atmospheric CO₂ changes at the city scale through simple data-driven analyses. We used the column-averaged dry-air mole fraction of CO₂ (XCO₂) from the Carbon Observatory 2 (OCO-2) and the anthropogenic CO₂ emissions provided by the Open-Data Inventory for Anthropogenic Carbon dioxide (ODIAC) product to explain the scenario of CO₂ over 120 districts of Pakistan. To study the anthropogenic CO₂ through space-borne observations, XCO₂ anomalies (MXCO₂) were estimated from OCO-2 retrievals within the spatial boundary of each district, and then the overall spatial distribution pattern of the MXCO₂ was analyzed with several datasets including the ODIAC emissions, NO₂ tropospheric column, fire locations, cropland, nighttime lights and population density. All the datasets showed a similarity in the spatial distribution pattern. The satellite detected higher CO₂ concentrations over the cities located along the China–Pakistan Economic Corridor (CPEC) routes. The CPEC is a large-scale trading partnership between Pakistan and China and large-scale development has been carried out along the CPEC routes over the last decade. Furthermore, the cities were ranked based on mean ODIAC emissions and MXCO₂ estimates. The satellite-derived estimates showed a good consistency with the ODIAC emissions at higher values; however, deviations between the two datasets were observed at lower values. To further study the relationship of MXCO₂ and ODIAC emissions with each other and with some other datasets such as population density and NO₂ tropospheric column, statistical analyses were carried out among the datasets. Strong and significant correlations were observed among all the datasets.

1. Introduction

Atmospheric carbon dioxide (CO₂) is an important greenhouse gas (GHG) due to its significant contribution to climate change [1]. The concentration of atmospheric CO₂ is continuously increasing primarily due to anthropogenic activities [2], and if it keeps on increasing at the same rate, then around 1.5 °C of global warming may be reached by the middle of this century, which may cause more climate extremes [3]. Under the Paris agreement, it has been pledged to limit global warming to well below 2 °C relative to the pre-industrial levels [4]. To meet the goals of the Paris agreement, immediate efforts are required to reduce anthropogenic emissions including CO₂ emissions. Moreover, the regular monitoring of CO₂ emissions using reliable datasets is also important to investigate the spatiotemporal trends of CO₂ and evaluate the effectiveness of the reduction policies. However, larger uncertainties in the existing datasets and the unavailability of any reliable global system for monitoring CO₂ emissions from local sources create a hindrance in monitoring CO₂ at smaller scales with sufficient accuracy [5].

The atmospheric CO₂ concentration is continuously increasing, and accurate measurement of changes in the concentration of atmospheric CO₂ is a prerequisite to determining its influence on climate change. Several networks have been established in the world for the precise monitoring of atmospheric CO₂ including the Global Atmospheric Watch (GAW) sites [6], the Collaborative Carbon Column Observing Network (COCCON) [7,8], and the Total Carbon Column Observing Network (TCCON) [9]. However, the measurements obtained from these ground stations are not sufficient for the accurate monitoring of atmospheric CO₂ at regional and global scales due to certain limitations including their uneven distribution and limited spatial coverage [10]. Atmospheric CO₂ monitoring has been improved with the launch of satellites because satellites can cover a very large area and provide large quantities of sampling data. The information related to atmospheric CO₂ concentration has been retrieved using various sensors onboard different satellites. For instance, the concentration of atmospheric CO₂ has been retrieved using the thermal infrared (TIR) bands of the Infrared Atmospheric Sounding Interferometer (IASI) [11], the Atmospheric Infrared Sounder (AIRS) [12], and Tropospheric Emission Spectrometer (TES) [13]; however, the TIR bands have lower sensitivity to the lower troposphere where most of the CO₂ sources and sinks reside. The Scanning Imaging Absorption Spectrometer for Atmospheric Chartography (SCIAMACHY) employed on the European Space Agency’s (ESA) ENVISAT was the pathfinder instrument that first detected atmospheric CO₂ signals using shortwave infrared (SWIR) and near-infrared (NIR) bands, which were more sensitive to lower tropospheric CO₂ concentrations and provided a reliable column-averaged dry-air mole fraction of CO₂ (XCO₂) observations at global scales [14]. SCIAMACHY suspended its services in 2012; however, several satellites have been launched since then for the exclusive monitoring of greenhouse gases with great spatiotemporal resolutions. The Greenhouse Gases Observing Satellite (GOSAT) was the first satellite that was launched by Japan in 2009 for the monitoring of atmospheric CO₂ and methane (CH₄) [15]. In 2014, the National Aeronautics and Space Administration (NASA) launched a satellite, the Orbing Carbon Observatory 2 (OCO-2) that is dedicatedly monitoring the atmospheric CO₂ and providing high-resolution observations. The Chinese CO₂-monitoring satellite TanSat that launched in December 2016 is also capable of providing space-based CO₂ measurements and preliminary XCO₂ maps generated using the satellite measurements, which have been discussed in a recent study [16]. Moreover, most recently GOSAT-2 and OCO-3 have also been successfully launched for monitoring atmospheric CO₂ and the initial results have been discussed in recent studies [17,18].

The global climate crisis poses several threats to Pakistan and the German Watch, an independent development and environmental non-governmental organization (NGO) that lists Pakistan among the ten countries that have been most affected by climate change during the last two decades. According to the Global Climate Risk Index (GCRI), Pakistan has suffered serious economic losses per unit GDP (Gross Domestic Product) due to 152 extreme weather events from 1999 to 2018 [19]. In a report published by the ActionAid, Bread for the World, and the Climate Action Network—South Asia (CANSA), it has been notified that even with the active emission reduction efforts, around 600,000 people will migrate by 2030 due to the impacts of climate change in Pakistan, and this number will be doubled if active actions are not carried out. The Government of Pakistan (GOP) is actively working at policy, management, and operational levels to deal with the effects of climate change. The Eco-system Restoration Initiative (ESRI) has been launched by the GOP to facilitate the transition towards an environmentally resilient Pakistan by mainstreaming adaptation and mitigation through ecological initiatives. The ESRI includes several programs including biodiversity conservation, mass-scale afforestation, and formulating environmental policies in line with the objectives of the country’s Nationally Determined Contribution (NDC) and attaining Land Degradation Neutrality (LDN). Monitoring the long-term trends of CO₂ over Pakistan under the background of climate change is important for predicting future CO₂ scenarios in the country.

OCO-2 was not a mapping mission, but it was designed as a sampling mission, so it samples a small portion of the globe every day [20]. OCO-2 was launched with a primary objective to monitor the atmospheric CO₂ at a regional scale; however, several studies have used the OCO-2 dataset for local-scale studies due to its precise measurements [4,21,22,23,24,25]. In this study, we used the XCO₂ retrievals from OCO-2 along with other datasets including the ODIAC CO₂ emissions, and the Ozone Monitoring Instrument (OMI) NO₂ tropospheric column to study the CO₂ scenario over 120 districts of Pakistan. Moreover, the districts were also ranked based on mean anthropogenic CO₂ concentrations and emissions. Additionally, spatial distribution patterns of various datasets were also compared. The description of the methods and datasets is given in Section 2; results are discussed in Section 4 and conclusions are provided in Section 5.

2. Materials and Methods

2.1. Study Area

Pakistan, with a population of around 220 million, is the fifth-most populated country in the world and is located in south Asia between 23°35′ to 37°05′ north and 60°50′ to 77°50′ east [26]. Geographically, it shares its borders with the largest CO₂-emitting nations such as China, India, and Iran. Administratively, Pakistan is divided into six units: Azad Kashmir, Gilgit Baltistan, Balochistan, Khyber Pakhtunkhwa, Punjab, and Sindh. Punjab is the most populous administrative unit followed by Sindh, Khyber Pakhtunkhwa, and Balochistan. In terms of area, Balochistan is the largest administrative unit of Pakistan followed by Punjab, Sindh, and Khyber Pakhtunkhwa. The administrative units are further subdivided into districts and this study is carried out over 120 districts in Pakistan. Although Pakistan has more than 120 districts in total, some smaller districts are merged into the neighboring districts to meet the defined area requirements (>1000 km²). The area requirements were defined so that the number of satellites were sufficient. The spatial boundaries of these districts located in various administrative units are shown in Figure 1a. In addition, the topography and landcover distribution of Pakistan are shown in Figure 1b,c, respectively.

2.2. Datasets

2.2.1. OCO-2 XCO₂ Dataset

OCO-2 was successfully launched by NASA from the Vandenberg Air Force Base in California on 2 July 2014 for the exclusive monitoring of atmospheric CO₂ at regional levels. After completing a series of check-out activities and orbit-raising maneuvers, it joined the front of the Afternoon Constellation (A-Train) on 3 August 2014 [20]. The referenced constellation consisted of six satellites orbiting around the earth at an altitude of 705 km. OCO-2 completes an orbit in 98.8 min and samples at a local time of about 1:30 pm, and it has a set of 233 orbit paths that repeat in 16-day cycles. OCO-2 has been routinely providing around 1 million soundings every day since 6 September 2014. The spatial resolution of OCO-2 at nadir is about 1.3  × 2.25 km.

The spectrometer installed on the OCO-2 satellite measures the near-infrared spectra of sunlight reflected off the Earth’s surface in three spectral regions centered at 0.765, 1.61, and 2.06 μm [27,28]. OCO-2 observations are processed using the Atmospheric CO₂ Observations from Space (ACOS) Full Physics (FP) retrieval algorithm to generate a level 2 (L2) product that provides XCO₂ retrievals. Technical details about the ACOS FP algorithm are given in the previous literature [29,30]. Several versions of XCO₂ L2 products have been released by the OCO-2 team and have been regularly validated against accurate measurements [2,31,32,33,34]. The validation results showed that the OCO-2 datasets were consistent and reliable for atmospheric CO₂ monitoring. A recent study [35] compared the latest OCO-2 XCO₂ product (v10) against accurate measurements and reported a sounding precision of ~0.8 ppm over land and ~0.5 ppm over water, and RMS biases of 0.5–0.7 ppm over both land and water. Our study incorporated the latest ACOS/XCO₂ Lite product version (v10r) that was available to download from the EARTHDATA website (https://earthdata.nasa.gov/, accessed on 15 February 2022).

2.2.2. ODIAC CO₂ Dataset

The Open-Data Inventory for Anthropogenic Carbon dioxide (ODIAC) is a high-resolution fossil fuel CO₂ (ffCO₂) emission dataset that was originally developed in 2009 at the National Institute of Environmental Studies (NIES), Japan [36]. ODIAC first time-incorporated the individual power plant emissions and nighttime light datasets to estimate and spatially distribute the CO₂ emissions. Several changes have been made since the first version of the product and details about the evolution are given in a recent study [37]. The current version of the ODIAC dataset (ODIAC2020b) available at the time of this study provided monthly CO₂ emissions in two spatial resolutions, i.e., 1 × 1 km and 1 × 1 degree [38]. We used the 1 × 1 km version of the product to study the CO₂ emission scenario over Pakistan. This global monthly CO₂ emission dataset spanning from 2000 to 2019 downscaled the national level emission statistics from the Carbon Dioxide Information Analysis Center (CDIAC) [39] and the global fuel statistical review data from the British Petroleum Company (after 2014) to 1 km resolution [40]. The downscaling was carried out based on 1 km global nighttime light data provided by the Defense Meteorological Satellite Program (DMSP) satellite and Carbon Monitoring for Action (CARMA) database. The ODIAC anthropogenic CO₂ dataset has been widely accepted by the carbon cycle community and several researchers have used the dataset to investigate anthropogenic carbon emissions for smaller to larger-scale studies [4,40,41]. The dataset can be downloaded from the NIES website (https://db.cger.nies.go.jp/dataset/ODIAC/DL_odiac2020b.html, accessed on 26 March 2022).

2.2.3. Other Datasets

The description of the other datasets used in this study is given in the following:

• The Ozone Monitoring Instrument (OMI) NO₂ tropospheric column dataset [42] from 2015 to 2020 was obtained from the EARTHDATA website (https://earthdata.nasa.gov/, accessed on 1 March 2022).
• The Visible Infrared Imaging Radiometer Suite (VIIRS) nighttime lights data [43] was downloaded from (https://eogdata.mines.edu/products/vnl/, accessed on 15 February 2022).
• Moderate Resolution Imaging Spectroradiometer (MODIS) Collection 6 global monthly Fire Location product (MCD14ML) from 2015 to 2020 was downloaded from (https://firms.modaps.eosdis.nasa.gov/download/, accessed on 27 February 2022).
• LandScan population density data [44] from 2015 to 2019 was downloaded from (https://landscan.ornl.gov/, accessed on 27 February 2022).
• Copernicus landcover data [45] from 2015 to 2019 was downloaded from the Copernicus Global Land Service website (https://land.copernicus.eu/global/products/lc, accessed on 11 February 2022).

2.3. Methodology

To understand the CO₂ scenario over Pakistan, the following methodology was adopted:

• The OCO-2 dataset comes with a quality flag to distinguish the cloud-contaminated and cloud-free XCO₂ retrievals (OCO-2 Data User Guide). It is generally advised to use cloud-free observations for local- and regional-level studies because the cloud-contaminated retrievals contain biases that might compromise the quality of the results. In this study, we incorporated the cloud-free OCO-2 retrievals with the daily standard deviation of the soundings less than 1 ppm. To determine the monthly, annual, and seasonal spatiotemporal trends of atmospheric CO₂, the OCO-2 XCO₂ retrievals were averaged on monthly, annual, and seasonal time intervals within the 0.5 × 0.5 degree spatial grid and the spatial boundaries of the administrative units (districts and provinces). Seasons were defined based on three months, i.e., DJF (December, January, February), MAM (March, April, May), JJA (June, July, August), and SON (September, October, November). To avoid uncertainties, the administrative boundaries with fewer than 300 satellite observations were not considered in the study. Moreover, the districts with an area <1000 km² were also not included in the study.
• Previous studies [41,46,47,48] have suggested that anthropogenic CO₂ could be detected using space-borne CO₂ observations. However, estimating the anthropogenic CO₂ concentration through these space-based observations is a challenging task. CO₂ is a greenhouse gas with a longer atmospheric life and a very large background concentration. Because of this, XCO₂ retrieved through satellite-based observations varies by only about 2% from pole to pole and over the seasonal cycle. The seasonal variability and the larger background concentration of atmospheric CO₂ must be removed to determine the anthropogenic CO₂ concentration. To do this, XCO₂ anomalies (MXCO₂) were calculated using an approach suggested by [46,47]:

  $$MXC{O}_2=XC{O}_{2 \left(ind\right)}-XC{O}_{2 \left(dbg\right)}$$

  (1)

  where $MXC{O}_2$ is the $XC{O}_{2 }$ anomaly, $XC{O}_{2 \left(ind\right)}$ is the individual OCO-2 $XC{O}_{2 }$ retrieval, and $XC{O}_{2 \left(dbg\right)}$ is the $XC{O}_{2 }$ daily background. The daily background concentration was estimated by calculating the daily median of $XC{O}_{2 }$ retrievals over the study area. The benefit of this method is that it provides the $XC{O}_{2 }$ anomaly for a single point. More details about the method are given in [46,47]. Once $MXC{O}_2$ was calculated against each OCO-2 retrieval, the mean of $MXC{O}_2$ was estimated within a spatial grid of 0.5 × 0.5 degrees and the spatial boundaries of the administrative areas for a defined period. $MXC{O}_2$ returns positive and negative values. Positive values show that CO₂ is being emitted into the atmosphere (sources) while negative values represent that CO₂ is being absorbed on the surface (sinks).
• The mean anthropogenic CO₂ emissions for each district was calculated by averaging the ODIAC CO₂ datasets from 2015 to 2019 and then summing the pixels/cells within the spatial boundaries of the districts. ODIAC results were combined with other datasets including the satellite-based XCO₂ anomalies, OMI NO₂ tropospheric column, nighttime lights data, and population density to study the spatial distribution of CO₂ over the study area.
• The ODIAC CO₂ and OCO-2-derived MXCO₂ datasets were compared in terms of correlation, spatial distribution, and ranking of districts based on the mean CO₂ emissions and mean MXCO₂ concentration values. Moreover, the relationship of the ODIAC and OCO-2 datasets with other datasets such as population density and NO₂ tropospheric column was also studied through cluster-based correlation analyses. To create the clusters, we segmented the datasets using the method described in [41,49], and then finally the correlation analyses were carried out.

3. Results

3.1. Spatial Distribution of OCO-2 XCO₂ Retrievals

Figure 2 shows the spatial distribution of monthly averaged XCO₂ retrievals over districts of Pakistan against the number of cloud-free OCO-2 soundings observed by each of the districts for a period of six years from 2015 to 2020. Overall, the satellite-based observations are good in quantity in January and February, and this quantity gradually keeps on decreasing from March and reaches a minimum in August, and then it starts increasing and reaches the maximum in December. This temporal change in the quantity of satellite-based observations is more significant in the districts located in Punjab and Sindh. Moreover, the results from Figure 2 show that the monthly averaged concentration of atmospheric CO₂ shows an increasing trend in January over most parts of the country and reaches the maximum concentration in May, then it starts decreasing and reaches the minimum concentration in September, and then again it starts increasing and follows this trend until the end of the year. The interannual variation trend of atmospheric CO₂ is also shown in Figure 3b. The highest concentration of atmospheric CO₂ during the pre-monsoon period can be attributed to several phenomena. For instance, a large amount of biomass burning takes place during this period and it significantly contributes to the increased levels of atmospheric CO₂ [50,51]. The higher temperature and radiation prevailing during summer stimulate the assimilation of atmospheric CO₂ in the daytime and respiration at night [52]. Moreover, the speed of the wind is low during the pre-monsoon period, causing a slight mixing in the boundary layer that is also a potential reason for increased atmospheric CO₂ [53]. The concentration of atmospheric CO₂ decreases during the monsoon period. A recent study [54] also reported a decreased atmospheric CO₂ concentration during the monsoon over Oman. During the rainy season, the soil moisture is increased, thereby enhancing the photosynthesis process, which eventually decreases the atmospheric CO₂ [55]. The low temperature due to the presence of clouds reduces the leaf and respiration rate, which also increases the carbon uptake. After the monsoon period, the atmospheric CO₂ again starts increasing, which might be linked to the consumption of fossil fuels during winter that produces an excessive amount of CO₂ and emits it into the atmosphere [34]. This increasing trend that starts in September reaches its peak in May. Moreover, microbial activity also increases after winter, which also complements the increases in the atmospheric CO₂ [53,56].

Figure 3 shows the long-term time series (Figure 3a), monthly averaged (Figure 3b), and seasonal-averaged concentrations (Figure 3c) of atmospheric CO₂ over various administrative units of Pakistan including Punjab, Sindh, KPK, and Balochistan for a period of six years from 2015 to 2020. Gilgit Baltistan and Kashmir were not included in the analysis due to the insufficient number of OCO-2 retrievals. A similar varying trend of atmospheric CO₂ concentration was observed in all the provinces of Pakistan; however, some differences were found in the magnitudes. The atmospheric CO₂ concentrations during each month were higher than those in the same month of the previous year (Figure 3a). It reflected a continuous increase in CO₂ concentrations. The seasonal-averaged XCO₂ concentration of various provinces is shown in Figure 3c. Atmospheric CO₂ concentration was higher during DJF and MAM, and lower in JJA, and SON. There was a slight difference between the magnitudes of seasonal-averaged XCO₂ concentrations among the various provinces.

3.2. Spatial Distribution of MXCO₂ and Anthropogenic CO₂ Emissions

Figure 4b,c show the spatial distribution of the total number of cloud-free OCO-2 retrievals over a spatial grid of 0.5 × 0.5 degrees and within the district boundaries of Pakistan from 2015 to 2020. Overall, the number of observations was sufficient for most of the areas; however, the northern areas including Gilgit Baltistan and Kashmir received the lowest number of OCO-2 retrievals. Any district with observations of fewer than 300 retrievals was not considered in the study.

Figure 4d,e show the annually averaged XCO₂ concentration over a spatial grid of 0.5 × 0.5 degrees and within the district boundaries of Pakistan from 2015 to 2020. While the MXCO₂ estimated using the OCO-2 retrievals is also displayed over a 0.5 × 0.5 degree spatial grid and the spatial boundaries of the districts (Figure 4f,g). MXCO₂ represents anthropogenic CO₂ concentrations. Surprisingly, higher MXCO₂ concentrations were observed over the districts located along the China–Pakistan Economic Corridor (CPEC) routes shown in Figure 4a. Moreover, most of the districts in Punjab and Sindh are surrounded by cropland (Figure 1b). The spatial pattern of increased MXCO₂ over Punjab and Sindh was somehow similar to that of the cropland (Figure 1b and Figure 4f,g). This higher concentration of MXCO₂ over the cropland might be due to the pre- and post-harvest burning. A fire location product derived using the MODIS data from 2015 to 2020 is also shown in Figure 4h, which also exhibited a similar spatial distribution pattern. Moreover, NO₂ is an indicator of atmospheric pollution, and it is co-emitted with CO₂ when fossil fuels are combusted at high temperatures. It has a short lifetime on the order of hours, so NO₂ columns often greatly exceed the background and noise levels of modern satellite sensors near sources, which makes it a suitable tracer of recently emitted CO₂ [4]. Figure 4i shows the spatial distribution of mean NO₂ tropospheric columns from 2015 to 2020. The spatial distribution pattern of increased NO₂ is similar to that of MXCO₂. Figure 4k shows the mean spatial distribution of anthropogenic CO₂ emissions from 2015 to 2019 over Pakistan, estimated using the ODIAC dataset. The emission pattern showed that CO₂ emissions were higher in Punjab followed by Sindh, KPK, and Balochistan. The anthropogenic emission pattern was similar to OCO-2-derived MXCO₂ (Figure 4f,g,k) over Punjab and Sindh where the CO₂ emissions were higher. However, this pattern was slightly different for other provinces where the emissions were lower. The mean nighttime lights dataset provided by the VIIRS from 2015 to 2019 is also displayed in Figure 4l. The results from both the ODIAC and nighttime light datasets showed similarities in spatial patterns as well as magnitudes. Moreover, the spatial distribution of the satellite-based MXCO₂ (Figure 4f,g) was also compared with that of the population density (Figure 4j) and a similarity between the spatial distributions of the two datasets was observed. In summary, the MXCO₂ derived using OCO-2 retrievals showed a spatial distribution pattern similar to cropland, fire locations, NO₂ tropospheric column, population density, ODIAC emissions, and nighttime lights. To further determine the relationship of MXCO₂ with these variables, statistical analyses were carried out between MXCO₂ and other datasets including the population density and the NO₂ tropospheric column, and the results are described in Section 3.3.

Figure 5a shows the multiyear mean XCO₂ concentration in Pakistani districts. The results showed that the atmospheric CO₂ concentration was continuously increasing in most of the districts. Figure 5b shows the mean MXCO₂ concentrations in districts for five years from January 2015 to December 2019. The districts were ranked based on the mean MXCO₂ concentration. The results showed that the districts with the highest concentrations of atmospheric CO₂ were either among the most populous cities, industrial in nature or experiencing large-scale construction or economic activities. For instance, Karachi, the capital of Sindh, is the largest city in Pakistan with an estimated GDP of $164 billion (PPP) as of 2019 [57]. Being ranked as a beta-global city, Karachi is Pakistan’s premier industrial and economic hub with the two largest seaports, the Port of Karachi and Port Bin Qasim. Lahore is the second-largest city of Pakistan and the capital of Punjab with an estimated GDP of $65.14 billion (PPP) as of 2017 [58] and a projected average growth rate of 5.6% [22]. The ODIAC inventory v2019 suggested that Lahore whole-city CO_2ff emissions increased by about 646 kt C/year during October 2014–May 2019, translating into a total change of 27% over 2015–2019 (i.e., a mean annual 5.9% increase), which is consistent with Pakistan’s national emission estimates of 5.05% during 2001–2018 [59]. Sialkot, Narowal, Nowshera, Multan, and Faisalabad are the cities where large-scale industrial activities are carried out. Islamabad is the capital of Pakistan. Kohistan and Mohmand were also listed among the districts with the highest MXCO₂ concentrations. These cities contain large reservoir-based dams [60]. The reservoir-based dams emit a significant amount of CO₂ into the atmosphere, and the dams located in tropical regions have even larger emissions compared to boreal and temperate regions [61]. Most of the districts showing the lowest MXCO₂ concentrations were located in the Balochistan and KPK provinces. The districts located in KPK were mainly mountainous areas covered with forests (Figure 1b,c) and in the case of Balochistan, the districts with the lowest MXCO₂ concentrations had the lowest population density, which indicated less human activity. Some districts, such as Umerkot, Mithi, and Mastung, belonging to the Sindh and Balochistan provinces, also had the lowest atmospheric CO₂, and these districts were mainly covered with deserts. Figure 5c shows the annually averaged MXCO₂ concentrations in districts of Pakistan from 2015 to 2019. The districts where MXCO₂ concentrations were increasing were located in Punjab and Sindh; however, most of the districts with decreased annual concentrations were in the KPK and Balochistan provinces. In 2014, the government of KPK launched the Billion Tree Tsunami (BTT) project as a response to the challenge of global warming [62]. The project restored 350,000 hectares of forests and degraded land [63]. The BTT project was completed in August 2017, ahead of schedule. The decreased concentration of MXCO₂ in KPK areas might be attributed to the large-scale afforestation; however, more evidence is needed to support the statement. Overall, as per the OCO-2 MXCO₂ results, Punjab showed the highest MXCO₂ concentration followed by Sindh, KPK, and Balochistan.

Mean CO₂ emissions for districts were also calculated and the districts were then ranked based on the amount of mean CO₂ emissions (Figure 6a). The larger districts of Punjab and Sindh such as Karachi, Lahore, Multan, Muzaffargarh, Faisalabad, Rawalpindi, Hyderabad, and Dadu showed higher CO₂ emissions. These districts were also listed among the ones with higher MXCO₂ concentrations estimated using OCO-2 XCO₂ retrievals (Figure 5b). These areas are among the most populated districts of Pakistan and experience large-scale economic activities. Based on the ODIAC CO₂ emission dataset, most of the districts with the lowest amounts of CO₂ emissions were located in Balochistan. The spatial distribution of the ODIAC dataset largely relies on nighttime lights, which result due to human activity. Balochistan is the largest province of Pakistan in terms of area with the lowest population density (Figure 4j). The low ODIAC CO₂ emissions in districts located in Balochistan might be due to the lower population density, which results in producing fewer of the nighttime lights that primarily control the spatial distribution of the ODIAC emissions. Figure 6b shows the annually averaged CO₂ emissions for each district of Pakistan, and the results show that CO₂ emissions were continuously increasing in most of the districts located in the Punjab and Sindh provinces.

3.3. Correlation Analysis

Figure 7 shows the correlation analysis of the OCO-2 MXCO₂ and ODIAC CO₂ emissions between each other and with some other datasets such as population density and the OMI NO₂ tropospheric column. Population growth is one of the major causes of increased CO₂ emissions in developed as well as developing countries, whereas the impact of population growth on CO₂ emissions has not received enough attention. NO₂ is an indicator of atmospheric pollution, and it is co-emitted with CO₂ when fossil fuels are combusted at high temperatures.

Previous studies showed that compared to the single sounding of XCO₂, clusters of satellite-based XCO₂ retrievals showed a better and more significant correlation with other variables, which might be due to the fact that a single sounding is an instantaneous snapshot of the realistic atmosphere [41,48]. Therefore, we segmented the datasets and created clusters following the method described in [49] and then finally carried out correlation analyses between the datasets.

Figure 7a shows the scatterplot and correlation between ODIAC CO₂ emissions and OCO-2 XCO₂ anomalies (MXCO₂). The two datasets showed correlation coefficients (r_p represents Pearson’s correlation and r_s represents Spearman’s correlation) of above 0.5, and this correlation was significantly improved (over 0.91) when the analysis was carried out between higher values of the two datasets (Figure 7b). Figure 7c shows the scatterplot and correlation between the OCO-2 anomalies and population density derived using the Landscan data. XCO₂ anomalies showed a positive correlation with the population density, and this correlation was also improved when the analysis was carried out between higher values (Figure 7d). Figure 7e shows the scatterplot between the ODIAC emissions and the population density. ODIAC showed a relatively better correlation with the population density compared to OCO-2 XCO₂ anomalies, and unlike OCO-2, the correlation between the ODIAC and population density was improved when the analysis was carried out after removing higher values (Figure 7f). However, both the OCO-2 and ODIAC datasets showed a positive correlation with the population density. Figure 7g shows the scatterplot and correlation between the XCO₂ anomalies and NO₂ tropospheric columns retrieved from OMI observations. Both of the datasets showed a positive correlation, and the correlation was improved when the analysis was carried out with higher values (Figure 7h). Hakkarainen et al. correlated OCO-2-based XCO₂ anomalies with OMI NO₂ tropospheric column over various regions and reported a good correlation between the two datasets [46]. Figure 7i,j show the scatterplot and correlation between the ODIAC CO₂ emissions and NO₂ tropospheric column, and both datasets showed a good correlation. The ODIAC dataset showed a better correlation with the NO₂ tropospheric column compared to the OCO-2 XCO₂ anomalies. In all the correlation analyses carried out between the OCO-2 XCO₂ anomalies and other variables, the common results showed a relatively weaker correlation at lower values and the results significantly improved when the analyses were carried out with higher values.

4. Discussion

The temporal change in the quantity of satellite-based observations was more significant in the districts located in Punjab and Sindh. Pakistan is a monsoon-influenced country and generally, in Pakistan, March to May is treated as pre-monsoon, June to August as monsoon and September to November as post-monsoon season [2]. The rainy season in the country starts during the pre-monsoon period and meets the occurrence of the summer monsoon, from June to August, which is a major source of precipitation. The decreased number of OCO-2 soundings during the pre-monsoon and monsoon periods might be due to the presence of clouds, because the space-based observations are vulnerable to clouds and aerosols [33]. GOSAT and OCO-2 are processed using the ACOS FP algorithm that has evolved over time to improve the quality as well as the quantity of the retrievals. A recent study [64] showed that out of 37 million soundings collected by GOSAT through June 2020, only 5.4 % of the soundings (2 × 10⁶) were finally assigned a “good” XCO₂ quality flag when processed using the ACOS FP algorithm, and other soundings were discarded due to the presence of clouds and other artefacts. Moreover, northern parts of the country including Gilgit Baltistan and Kashmir consistently had the lowest number of OCO-2 retrievals throughout the year. These areas are covered with the highest mountain ranges that are mostly covered with snow. The satellite observations over such complex areas showed larger uncertainties, did not meet the filtering criteria of the retrieval algorithms and were finally discarded.

The satellite-detected XCO₂ anomaly (MXCO₂) exhibited a spatial distribution similar to the CPEC route, NO₂ tropospheric column, population density, fire location, cropland, and nighttime lights. The CPEC, a large-scale trading partnership between Pakistan and China, is a collection of infrastructure projects including a comprehensive road network connecting the Gwadar port to China [65,66,67]. Rapid development has been carried out on the CPEC project during the last decade. The notable works include the construction of the road network shown in Figure 4a, special economic zones along the CPEC routes, and several energy projects. The rapid development along the CPEC road network might be a potential reason for the increased levels of atmospheric CO₂ concentration in the neighboring districts. NO₂ is emitted when fossil fuels are burned at high temperatures, thus it is also an indicator of pollution. Hakkarainen et al. compared the spatial distribution of XCO₂ anomalies derived using OCO-2 retrievals with that of NO₂ tropospheric columns over various regions of the world and reported that both of the datasets showed similarities in terms of spatial distribution patterns [46,47]. In the case of the MXCO₂–ODIAC comparison, similarity in the spatial distribution pattern of the two datasets was found over the areas which had higher anthropogenic emissions; however, some differences were also observed over the areas which had lower amounts of anthropogenic emissions. Several regional-scale studies have been carried out comparing the ODIAC CO₂ emissions with the satellite-based XCO₂ anomalies and reported that the ODIAC dataset showed a good agreement with satellite-based XCO₂ concentration at higher values, whereas notable deviations occurred at lower values. Yang et al. studied the correlation between the ODIAC emissions and XCO₂ anomalies derived using GOSAT retrievals over China and reported that the correlation between the two datasets was improved at higher values [48]. Similar results were reported by [41,46] when OCO-2-based XCO₂ anomalies were compared with the ODIAC CO₂ emissions. The differences between these two datasets are due to the different approaches to measurement. Bottom-up-approach-based inventories, i.e., ODIAC and CDIAC, estimate the CO₂ emissions based on standardized protocols, combining activity data such as fuel production and consumption as well as traffic-monitoring data with pre-calculated emission factors for specific sources across different activity sectors. The datasets are distributed by either spatial proxies, i.e., nighttime lights, population density, or through combinations of line sources such as on-road emissions and point sources such as power plants [22]. ODIAC incorporates the nighttime lights and power plant point sources to produce a high-resolution CO₂ emission dataset [37,38].

Furthermore, MXCO₂ exhibited good correlations with various variables such as the ODIAC CO₂ emissions, population density, and NO₂ tropospheric columns at higher values; however, these correlations deteriorated at lower values. The larger deviation between smaller values of the datasets might be due to several reasons: (i) The ODIAC and population datasets may contain uncertainties due to data gaps, inaccurate calculation factors, inaccurate data, and the unavailability of statistics, especially for developing countries; (ii) uncertainties may be present in the XCO₂ anomalies that are likely to be produced by the CO₂ uptake of the biosphere, which remains in the XCO₂ anomalies.

5. Summary and Conclusions

In this work, we used OCO-2 and the ODIAC CO₂ datasets to study the scenario of CO₂ over 120 districts of Pakistan. The spatial coverage of OCO-2, spatial distribution, and monthly and seasonal variations of atmospheric CO₂ were studied in detail. OCO-2 provided a reasonable spatial coverage over Pakistan except for the northern areas, which are mostly covered with the highest mountain ranges. Moreover, the quantity of OCO-2 retrievals started decreasing during the pre-monsoon season and reached the minimum during the monsoon season (June to August). This might be due to the presence of clouds that strongly affected the space-borne observations. The concentration of atmospheric CO₂ was continuously increasing in Pakistan with significant seasonal fluctuations. The maximum and minimum concentrations of atmospheric CO₂ were observed in May and September, respectively. Moreover, the atmospheric CO₂ concentration was higher during DJF and MAM, and lower in JJA and SON. A slight difference in the concentration of atmospheric CO₂ was also observed between various administrative units of Pakistan.

To study the anthropogenic CO₂ using a satellite dataset, OCO-2 retrievals were deseasonalized and detrended by calculating the XCO₂ anomalies (MXCO₂). The spatial distribution pattern of MXCO₂ showed higher concentrations over the cities located along the CPEC routes. The spatial distribution pattern of the satellite-based MXCO₂ was compared with those of various datasets including the ODIAC CO₂ emissions, OMI NO₂ tropospheric column, population density, nighttime lights, fire locations, and cropland, and all the datasets showed a similarity in the spatial distribution pattern. Mean MXCO₂ concentrations were calculated within the spatial boundaries of districts, and the districts were then ranked based on the mean MXCO₂ concentrations. The cities showing the highest MXCO₂ concentrations were either metropolitan cities or experiencing large-scale economic activities. Karachi, Lahore, Faisalabad, Multan, Sialkot and Narowal were among the cities with the highest MXCO₂ concentrations. The cities with the lowest MXCO₂ concentrations were mostly located in the KPK and Balochistan provinces. Moreover, Punjab showed the highest anthropogenic CO₂ concentration among the administrative units, followed by Sindh, KPK, and Balochistan.

Anthropogenic CO₂ emissions over Pakistan were studied using the ODIAC dataset. The spatial distribution pattern of the ODIAC emissions was also similar to those of the satellite-derived MXCO₂, OMI NO₂ tropospheric column, and population density. The ODIAC spatial distribution pattern showed some deviations from that of the satellite-derived MXCO₂ over the areas where population density was significantly low. This might be due to the fact that the ODIAC dataset largely relied on nighttime lights for the spatial distribution, and nighttime light is difficult to detect from a satellite over the areas with lower human activity. To determine the anthropogenic CO₂ emissions in each district, the mean CO₂ emissions were estimated for each of the districts, and the districts were then ranked based on the mean CO₂ emissions. The districts with the highest MXCO₂ concentrations such as Karachi, Lahore, Muzaffargarh, Multan, Faisalabad, and Rawalpindi were also among the cities with the highest CO₂ emissions. The results showed that CO₂ emissions were increasing in most of the cities.

The spatial distribution patterns of the satellite-derived MXCO₂ and ODIAC CO₂ emissions showed a similar trend with the population density and NO₂ tropospheric column. Therefore, to study the relationship between various variables, cluster-based correlation analyses were carried out. The results showed that OCO-2 MXCO₂ showed a positive and significant correlation with the ODIAC CO₂ emissions, and these correlations improved when the analysis was carried out with higher values. Both the OCO-2 and ODIAC datasets showed positive and significant correlations with the NO₂ tropospheric column and population density. The common results of OCO-2 MXCO₂ with all the variables showed that the correlation was improved when the smaller values were removed. This might be due to the uncertainties in the MXCO₂ caused by CO₂ uptake of the biosphere or atmospheric transport. However, the matter needs further investigation.

Several satellites are now orbiting around the earth and continuously monitoring atmospheric CO₂. In the future, we intend to study the potential of joint utilization of multiple CO₂-monitoring satellite datasets along with atmospheric transport modeling to monitor the atmospheric CO₂ at local scales.