We explore the value of multispectral CO retrievals from NASA/Terra Measurement of Pollution In The Troposphere (MOPITT v5), along with Atmospheric CO2 Observations from Space (ACOSv2.9) CO2 retrievals from the Japan Aerospace Exploration Agency Greenhouse Gases Observing Satellite (GOSAT), for characterizing emissions from anthropogenic combustion. We use these satellite retrievals to analyze observed CO2/CO enhancement ratios (ΔCO2/ΔCO) over megacities. Since CO is coemitted with CO2 in anthropogenic combustion, the observed ΔCO2/ΔCO characterizes the general trend in combustion activity. Our analyses show patterns in ΔCO2/ΔCO that correspond well with the developed/developing status of megacities, and ΔCO2/ΔCO that agree well with available literature and emission inventories to approximately 20%. Comparisons with ΔCO2/ΔCO derived from Total Carbon Column Observing Network measurements show similar agreement, where some of the differences in observed ΔCO2/ΔCO are due to representativeness and limited GOSAT data. Our results imply potential constraints in anthropogenic combustion from GOSAT/MOPITT, particularly in augmenting our carbon monitoring systems.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Tracking greenhouse gas emissions from anthropogenic combustion is imperative due to the role of these gases in long-term climate change, and in supporting activities focused on monitoring, reporting, and verifying anthropogenic emissions. However, current methods to quantify these emissions lack the sufficient accuracy necessary to support these activities. Attributing the sources of these emissions is challenging due to the dearth of accurate CO2 measurements with sufficient spatiotemporal coverage, the difficulty in teasing out the small anthropogenic signal from the large natural sources and sinks dominating the carbon cycle, and the uncertainties in modeling atmospheric transport [Pacala et al., 2010].
 This study aims to demonstrate the use of observations from combustion-related atmospheric constituents (traditionally associated with air pollution) to augment the anthropogenic CO2 signal in urban environments. CO is the most suitable constituent to study since: (1) it is coemitted with CO2 in the combustion of carbonaceous fuels, (2) it is relatively well-observed from in situ, aircraft, and satellites, and (3) it has a relatively short lifetime and well-known sink from its reaction with the OH radical. Along with CO2, we use CO to investigate combustion properties over large urban areas or megacities, where significant CO2 and CO emissions (and anthropogenic CO2 signal) are expected especially over regions of rapid growth and development [e.g., Duren and Miller, 2012; Schneising et al., 2013].
 We note that the relationship of CO2 and CO has been previously studied as a means to disentangle anthropogenic CO2 from its large natural (biospheric) signal [e.g., Suntharalingam et al., 2004; Wunch et al., 2009; Wang et al., 2010; Turnbull et al., 2011]. These studies used ground-based and/or aircraft-based collocated CO2 and CO measurements at sites located in specific regions, such as East Asia and California, to derive sensitivities of CO2 to CO (ΔCO2/ΔCO) and to subsequently infer emissions for the region. This work extends these studies by generalizing to satellite observations, which can provide the needed spatiotemporal coverage in critically important yet poorly observed urban regions. This directly follows the discussion in Duren and Miller  on using the Measurement of Pollution In The Troposphere (MOPITT) CO column retrievals to supplement the column-averaged dry air mole fractions of CO2 currently derived from the Japanese Space Agency's Greenhouse Gases Observing Satellite (GOSAT or “Ibuki”) [Morino et al., 2011] and in the near future from NASA's Orbiting Carbon Observatory (OCO-2), in order to improve our capability to monitor and verify carbon emissions from megacities. A unique aspect of this study is the use of new multispectral CO retrievals from MOPITT [Worden et al., 2010; Deeter et al., 2012, 2013] that provide improved characterization of lower-tropospheric CO comparable to the vertical sensitivity of GOSAT.
2.1 Data Description
 Our analysis focuses on CO2 and CO observations for the period of June 2009 to May 2010. This period coincides approximately with the first year set of CO2 data from GOSAT.
 We used the GOSAT data products (http://oco.jpl.nasa.gov) derived using the Atmospheric CO2 Observations from Space (ACOS) L2 algorithm (build 2.9) as initially described by Crisp et al. . The algorithm used the radiance spectra from GOSAT at O2 A-band (0.76 µm) and at two CO2 bands (near 1.61 and 2.06 µm) to retrieve and other atmospheric and surface state properties. We used all postscreening filters as recommended by the ACOS team and described in Crisp et al. . GOSAT collects observations near 13:00 local time with a revisit time of 3 days. It uses a nominal raster scan that spans ~150 km between ~10 km footprints (at nadir). GOSAT data over megacities come from the GOSAT project by S. Maksyutov (T. Oda, personal communication, 2013). A similar set of retrievals has been used by Kort et al.  to study CO2 distribution in Los Angeles and Mumbai and by Keppel-Aleks et al.  to infer regional fossil-fuel CO2 emissions. Samples of data used in this study are shown in Figure 1a corresponding to its seasonal mean for spring 2010.
2.1.2 XCO Retrievals
 As mentioned, we used retrievals from MOPITT derived from the combination of thermal-infrared (TIR, 4.7 µm) and near-infrared (NIR, 2.3 µm) spectral bands. These multispectral (TIR/NIR) retrievals have been demonstrated to exhibit enhanced sensitivity in the lower troposphere with maximum degrees of freedom (DFS ~ 2.2) higher than TIR-only (~1.8) and NIR-only (~1.0) [Deeter et al., 2012, 2013]. In this analysis, we used MOPITT v5 L2 TIR/NIR and TIR-only total column retrievals (converted to XCO using MOPITT-retrieved surface pressure) along with the recommended quality control screening from the MOPITT team (http://www.acd.ucar.edu/mopitt). MOPITT collects observations near 10:30 am local time with global coverage every 3–4 days. It uses a cross-track scan with ~22 km footprints (at nadir). Recently, Jiang et al.  used a similar set of MOPITT retrievals to show the value of this product in diagnosing model errors in convective transport. While MOPITT retrievals are available for the years 2000 to present, our analysis is limited to the GOSAT first year period. In Figure 1b, the XCO retrievals clearly show strong urban signatures compared to even at seasonal timescale. This is not surprising as XCO can be enhanced by a factor of 2–3 over an urban environment whereas can be only enhanced by 1%.
2.2 Regression Analyses
 We derived ΔCO2/ΔCO for each urban region based on pairs of and XCO retrievals collected over the region for the entire study period. First, we define the center location and spatial extent of each region using population density data from NASA Socioeconomic Data and Applications Center (http://sedac.ciesin.columbia.edu), which has a spatial resolution of 2.5 min, in conjunction with the Google Earth application. We then refine the spatial boundaries of this region by selecting the population density pixels that sum to the population reported by United Nations World Population Prospectus prioritizing the inclusion of the most densely populated pixels. We show a sample of our urban designation for Los Angeles, California, in Figure 1c.
 We then collect pairs of and XCO retrievals falling within the pixels of the region assuming that the retrievals have a spatial footprint corresponding to their respective horizontal resolutions. Due to GOSAT sampling limitations, data pairs were selected by choosing the closest MOPITT retrieval, in both space and time, to the GOSAT retrieval. We recognize that the pair is ~2.5 h apart and may not sample the same air column. This may have an impact given the diurnal variability of these gases and local meteorology. We note, however, that this can be mostly addressed through data assimilation, which is beyond the scope of this analysis. Here we assume that each pair is a sample of a distribution that characterizes the annual-mean combustion properties of the urban airshed, which is larger and longer in scale than the column spatiotemporal footprints. This is a necessary assumption given the limited number of GOSAT/ACOS retrievals for this period and differences in sampling characteristics between GOSAT and MOPITT. To minimize the biospheric influence on the sample distribution, we filter out pairs of retrievals whose enhancements about the seasonal mean exhibit statistically significant (p value < 0.1) negative correlation with the assumption that those data are largely associated with biospheric CO2 and biogenic CO. We use seasonal rather than monthly enhancements to utilize more GOSAT data in our analysis. We understand that this filter may not fully disentangle the biospheric CO2 signal. It removes most GOSAT data during the summer and focuses the analysis to fall to early spring when biogenic CO over urban airsheds is expected to be small.
 We used the linear regression method outlined in York et al.  to determine ΔCO2/ΔCO for the collected and XCO pairs for each urban region. The slope of the line corresponds to ΔCO2/ΔCO. This is the same method used by Wunch et al.  in their study of greenhouse gas emissions in California using Total Carbon Column Observing Network (TCCON). This is shown in Figure 1d for the case of Los Angeles.
 Finally, we carried out a suite of regression analysis for this study (see supporting information) to test the sensitivity of the calculated slopes to assumptions on how well the data represents the urban region. This includes sets of analyses for pairs of and XCO collected using: (1) our urban designation but assuming larger representative footprints of the retrievals and (2) larger areas of isotropic influence around the center of the urban region. We select our final ΔCO2/ΔCO estimates by finding the “local” minima within statistically significant estimates in (1) or (2). This accounts for the fact that, in many regions, factories and power plants are located outside of regions where the population is densest. As a sanity check, we verify that the TIR/NIR estimate is lower than its corresponding TIR-only estimate to ensure that TIR/NIR retrievals have enhanced sensitivity in the lowermost troposphere; otherwise, the estimate is representative of a free-tropospheric ΔCO2/ΔCO or confounded by retrieval biases. In the following section, we report our final estimates (typically a minimum across our analyses), together with the associated errors derived from the regression scheme and the resulting interquartile range across our sets of analysis.
3 Results and Discussion
3.1 Comparison With TCCON Measurements
 We began our analysis by testing our selection algorithm with collocated and XCO from all TCCON ground-based spectrometers (http://tccon.ipac.caltech.edu). The network provides retrievals of accurate and precise , XCO, and other constituents [Wunch et al., 2011]. We collected data from ACOS and MOPITT around the location of the station extending to a 1° radius and apply the same algorithm described in section 2.2 for TCCON and ACOS/MOPITT. In Table 1, we show comparisons of ΔCO2/ΔCO (in units of mol/mol) derived from these data sets. For this period, we have two stations (Lamont and Park Falls) with available measurements and results that satisfy the same filtering criteria described in section 2.2. This comparison is intended to test our regression algorithm, which is independent of urban designation. We find agreements (~11% to ~28%) in ΔCO2/ΔCO between TCCON and ACOS/MOPITT for both stations. Differences may be attributed to representativeness errors and retrieval sensitivities between ACOS/MOPITT and TCCON, particularly for Park Falls where urban influences are sampled around the 1° radius, reflecting higher CO in MOPITT TIR/NIR. It is also possible that nonanthropogenic sources confounded our analysis [Turnbull et al., 2011]. Lamont may have lower TCCON ΔCO2/ΔCO than Park Falls due to the influence of nearby coal-fired power plants. We note that, while these results (albeit limited) are promising, a more robust footprint analysis and more data from GOSAT will provide more accurate and precise comparison with TCCON.
Table 1. Comparison of ΔCO2/ΔCO (mol/mol) Derived From TCCON CO2/CO Column Measurements and ACOS v2.9 and MOPITT v5 XCO TIR/NIR and TIR-Only Retrievals Around a 1° Bin Centered at the Site Locationa
ACOS v2.9/MOPITT v5 ΔCO2/ΔCO
aPercentage differences with TCCON are shown in parenthesis.
Lamont, OK, USA (36.61°N, −97.49°E)
80 ± 5
90 ± 9 (~12%)
100 ± 11 (~25%)
Park Falls, WI, USA (45.95°N, −90.24°E)
112 ± 2
81 ± 11 (~28%)
124 ± 14 (~11%)
3.2 Sensitivity Over Megacities
 We performed our regression analyses, with the aforementioned selection criteria and seasonal filtering, over megacities with population larger than 5 million persons. The results are shown in Figure 2. Several megacities did not pass the selection criteria due to lack of GOSAT data over small urban footprints. We sorted the values of ΔCO2/ΔCO to show a trend in combustion properties among megacities, with higher values corresponding to higher overall combustion efficiency. An exception to this would be higher CO due to smoldering wildfires or other nonanthropogenic sources, lower CO due to dilution with “clean” air, and higher CO2 from human respiration [Turnbull et al., 2011]. In general, we find most megacities in “developed” nations show higher ΔCO2/ΔCO (>70, e.g., Germany and UK) than those in “developing” nations (<60, e.g., India and China). This result demonstrates constraints from satellite observations in distinguishing broad patterns of combustion properties that can be used to estimate related emissions given prior knowledge of emission of one of these gases [e.g., Suntharalingam et al., 2004; Wunch et al., 2009].
 Our results also agree well with literature. In particular, we find our ΔCO2/ΔCO in Tokyo (96 ± 13), Osaka/Kobe (89 ± 9), and Beijing/Tianjin (23 ± 4) to be reasonably consistent (albeit high) with values reported by Suntharalingam et al.  over Japan (60–80) and for the northeast China outflow (10–20) derived from Transport and Chemical Evolution over the Pacific aircraft campaign in 2001. Our Beijing/Tianjin ratio also agrees well with Turnbull et al.  of 21–23 ± 1 for China from flask samples taken at Shangdianzi, China, and Tae-Ahn Peninsula, Korea (TAP); as well as with Wang et al.  of 24–29 ± 6–10 derived from surface measurements in Miyun, China, near Beijing for 2005–2007, with dramatic increase in 2008 (46 ± 4.6) which they attributed to pollution measures during the 2008 Beijing Olympics. Such reduction in pollution was also captured by MOPITT [Worden et al., 2012]. It is very interesting to see close agreements even with inventory-based estimates over China (e.g., 23 for year 2006 in Zhang et al. ). In addition, Turnbull et al.  reports a value of 77 ± 12 for Korea based on measurements in TAP, while we find a higher value (119 ± 8) for Seoul. Over California, Wunch et al.  reported a value of 91 ± 17 based on TCCON measurements at Pasadena in 2008 representing the South Coast air basin (SoCAB). Our ΔCO2/ΔCO estimate is slightly lower (81 ± 2) for a similar urban region that includes Los Angeles. However, recent top-down estimates of CO2 and CO emissions, based on California Nexus (CalNex) aircraft campaign in 2010 [Brioude et al., 2013], show even a lower ratio of 75 in SoCAB.
 We find that the ratios derived (and associated comparison with literature) are sensitive to the spatiotemporal extent of the urban influence considered for the analysis. This is exemplified by the range of estimates shown in Figure 2. Tianjin and Beijing represent issues in defining spatial boundaries since both cities are geographically close, leading us to effectively combine the two since we cannot separate their airsheds without considering local meteorology. For Moscow, we find limited retrievals (especially toward higher latitudes) resulting to issues in representativeness errors and finding statistical significance. This limitation in GOSAT observing strategy over megacities was also highlighted by Kort et al. . We note that the errors derived from the regression represent the precision (not accuracy) of our estimates. We find that these are typically lower than the range of estimates across our analyses, which is indicative of unaccounted representativeness errors in our regression. These errors are also small compared to literature since our analysis focuses on annual-mean estimates, effectively smearing out short-term variability. Additionally, our Los Angeles estimate corresponds to a specific case where we found no local minima across the range of ΔCO2/ΔCO, leading us to choose instead the estimate that had the largest number of data pairs.
3.2 Sensitivity Over Countries
 We show in Figure 3 our ΔCO2/ΔCO estimates for the countries corresponding to the megacities (defined here using geopolitical boundaries), along with inventory-based emission ratios from Emissions Database for Global Atmospheric Research (EDGAR) version 4.2 (http://edgar.jrc.ec.europa.eu). We used the country totals for year 2010 CO2 and CO emissions and converted them to emission ratios, eCO2/eCO (mol/mol). Overall, our estimates fit surprisingly well with the emission ratios (i.e., slope = 0.84 and R2 = 0.61). This agreement (with median difference of ~25%) is consistent with our limited comparisons with TCCON. Again, we see a clear trend of higher ΔCO2/ΔCO in more “developed” nations (USA to UK), with notable exceptions like Australia and Brazil, where high values of CO are observed due to biomass burning events in these regions. Russia has a large spread due to limited retrievals. We also note that our estimate for UK differs significantly with EDGARv4.2, which we attribute to a possible underestimate in EDGARv4.2 CO emissions.
 A quick comparison of our estimates over megacities (Figure 2) and countries (Figure 3) reveal that, in some cases, the megacity and country estimates are quite similar. While this might arise from sampling bias, the consistency could also shed light on how large urban areas within the country may represent the overall anthropogenic signal.
4 Summary and Future Implications
 This work demonstrates the use of satellite observations of atmospheric constituents associated with anthropogenic combustion in augmenting the anthropogenic CO2 signal in urban environments. We generalize the concept of using ΔCO2/ΔCO to retrievals of from ACOS/GOSAT and XCO from MOPITT v5. Our approach is to select a set of retrievals for specific urban regions, which we assume to be samples of a distribution characterizing the annual-mean combustion properties of the urban airshed, and perform linear regression analysis to determine ΔCO2/ΔCO. We find that this approach (albeit simple) provides ΔCO2/ΔCO estimates that are consistent (within ~20%) with TCCON and other concentration-based ratios from literature, as well as emission-based ratios (e.g., EDGAR). Discrepancies in our estimates are mostly due to limited GOSAT data at small footprints and unaccounted for representativeness errors. We vconclude that this work has important implications in providing potential constraints on emissions from anthropogenic combustion [e.g., Keppel-Aleks et al., 2013], especially in conjunction with tracer-transport inversions [e.g., Wang et al., 2009]. Additionally, it lends support to study how current and future carbon monitoring systems may be designed, used, and augmented. For instance, this study reiterates the relevance of multispecies measurements from different instruments that are fairly collocated and exhibit similar sampling characteristics, or more desirably, from the same instrument. We will extend this work in the future to include other combustion-related constituents (e.g., NO2) [Berezin et al., 2013].
 This work was supported by NASA grants NNX11AL31G and NNX13AK24G (Silva and Arellano) and NNX10AT42G (Worden). We thank ACOS/GOSAT, MOPITT, and TCCON teams for data, NASA/CMS team for insights, and Google Earth for the overlay map. GOSAT data were produced by ACOS/OCO-2 project at JPL/CalTech and obtained from ACOS/OCO-2 data archive maintained at NASA GESDISC. We thank Kerensa Gimre, Aishwarya Raman, and two reviewers for helping improve the manuscript.
 The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.