<b>Jacobson</b>, A. R., J. B. <b>Miller</b>, A. Ballantyne, S. Basu, L. Bruhwiler, A. Chatterjee, S. Denning, and L. Ott, 2018: Chapter 8: Observations of atmospheric carbon dioxide and methane. In Second State of the Carbon Cycle Report (SOCCR2): A Sustained Assessment Report [Cavallaro, N., G. Shrestha, R. Birdsey, M. A. Mayes, R. G. Najjar, S. C. Reed, P. Romero-Lankao, and Z. Zhu (eds.)]. U.S. Global Change Research Program, Washington, DC, USA, pp. 337-364, https://doi.org/10.7930/SOCCR2.2018.Ch8.
Observations of Atmospheric Carbon Dioxide and Methane
The global average atmospheric CO2 concentration in 2015 of about 401 ppm (see Figure 8.1) is roughly 20 ppm (5%) higher than in 2007. The anthropogenic excess of CO2—the concentration in the atmosphere above the preindustrial level of about 280 ppm—has grown by 20% in just the 8 years since 2007. The 2015 global average concentration of CH4 was about 1,833 parts per billion (ppb), which is 3% higher than in 2007 (a 5% increase in the anthropogenic excess).
8.3.1 Advances in Atmospheric Measurements and Platforms
The observation network for atmospheric CO2 and CH4 has grown dramatically since SOCCR1 (see Figure 8.2). Networks are now run by 1) governmental institutions such as the National Oceanic and Atmospheric Administration (NOAA), Environment and Climate Change Canada, U.S. Department of Energy, and California Air Resources Board; 2) research institutions including the National Center for Atmospheric Research (NCAR) and National Ecological Observatory Network (NEON); 3) universities such as Scripps Institution of Oceanography, The Pennsylvania State University, Oregon State University, and Red Universitaria de Observatorios Atmosfericos in Mexico; and 4) corporations (e.g., Earth Networks). Platforms and measurement techniques for observing greenhouse gas (GHG) distributions also have grown and become more diverse. In 2005, the North American CO2 and CH4 surface network mainly consisted of weekly surface flask–air sampling at a handful of sites and continuous observations at several observatories and three tall towers (see Figure 8.2). Sustained records are now available from many more towers, especially those of intermediate (~100 m) height. As the density of the North American GHG measurement network has grown, the emissions sensitivity of observations has moved from hemispheric scales (using background marine boundary layer observations), to regional scales (using tower and aircraft observations), and, more recently, to local scales from urban networks and oil and gas measurement campaigns. These new in situ measurements of CO2 and CH4 (see Figure 8.2) have been enabled by better availability of higher-precision, stable laser spectroscopic analyzers that require less-frequent calibration, although traceability to a common CO2 reference scale is critical for this collection of networks to be unified. Currently, about 90% of the CO2 network sites also report CH4 measurements.
Figure 8.2: Growth of the North American Carbon Dioxide (CO2) Monitoring Network from (a) 2005 to (b) 2015
New remote-sensing approaches have emerged such as the international Total Carbon Column Observing Network (TCCON), which now has six sites in North America among about 20 worldwide. TCCON measurements are made using high-resolution solar-tracking Fourier transform spectrometers (FTSs; Wunch et al., 2011), which are sensitive to the total CO2 content of the atmospheric column, can provide constraints on large-scale carbon fluxes (Chevallier et al., 2011; Keppel-Aleks et al., 2012), and also help identify biases in satellite-based remote sensors (e.g., Wunch et al., 2016). Since SOCCR1, first-generation CO2- and CH4-dedicated near-infrared space-based spectrometers have been deployed aboard the Greenhouse Gases Observing Satellite (GOSAT; Japan Aerospace Exploration Agency) and the Orbiting Carbon Observatory-2 (OCO-2; National Aeronautics and Space Administration [NASA]) satellites. Numerous carbon cycle data assimilation systems are attempting to assimilate these CH4 (GOSAT) and CO2 (GOSAT and OCO-2) column averages to derive surface fluxes. These efforts are challenged by small but spatially and temporally coherent biases in the data (Basu et al., 2013; Feng et al., 2016; Lindqvist et al., 2015). Estimating emissions anomalies (as opposed to absolute emissions), such as carbon flux variability driven by climate events, has proved to be more successful (Basu et al., 2014; Guerlet et al., 2013; Reuter et al., 2014; Turner et al., 2017). Assimilating column-average GHG data from both ground- and space-based instruments into carbon cycle models is still a rather new activity that requires modifications in traditional atmospheric inverse models. They need to be modified to handle a much larger data volume, extract information from full-column averages, and assimilate retrievals contaminated by coherent biases, which can masquerade as atmospheric gradients arising from surface exchange.
Another remote-sensing approach for CO2 uses light detection and ranging (LIDAR), which has been deployed at surface sites to measure the mean CO2 along horizontal paths (Gibert et al., 2008, 2011) and aboard aircraft to measure partial-column integrals (Dobler et al., 2013). Space-based LIDAR total column CO2 and CH4 measurements are under development (Ehret et al., 2008), and a CH4 system will be deployed on the MERLIN satellite sensor. LIDAR instruments have narrow beams and thus can often obtain data in partly cloudy regions that confound passive sensors. Because they are active, LIDAR instruments can obtain data in the absence of sunlight (at high latitudes or at night). Despite this appealing feature, LIDAR instruments are not yet broadly distributed for atmospheric research.
Vertical In Situ
Calibrated CO2 and CH4 total column values can be measured using in situ approaches. The AirCore is a thin steel tube that samples an air profile, typically during a balloon flight (Karion et al., 2010). Profiles (and thus column integrals) of CO2 and CH4 (Karion et al., 2010) extend to altitudes that allow sampling of nearly 99% of the atmospheric column of air. In addition to defining the vertical structure of CO2 and CH4 in both the troposphere and stratosphere, these data provide calibrated total columns that can be directly compared to remotely sensed soundings from space (e.g., OCO-2 and GOSAT) and the ground (TCCON). Time series of AirCore measurements are being established at Sodankylä, Finland; Orleans, France; Lamont, Oklahoma; and Boulder, Colorado. While not sampling the total column, in situ measurements taken aboard light aircraft flying between the surface and 6 to 8 km above sea level also are ongoing. These regular (biweekly to monthly) measurements capture the seasonal and interannual distribution of CO2, CH4, and other GHGs throughout North America (Sweeney et al., 2015; see Figure 8.2). Although the number of air samples collected has not significantly increased since 2007, the number of gases measured has increased from eight to more than 50, including gases like carbonyl sulfide (COS) and the 14C:C ratio of CO2 (Δ14CO2) that are tracers for biogenic and fossil fuel emissions.
Carbon monoxide (CO) retrievals from the Measurements Of Pollution In The Troposphere (MOPITT) and Infrared Atmospheric Sounding Interferometer (IASI) satellite instruments have been used to constrain biomass burning GHG emissions and help separate intact ecosystem carbon uptake from biomass burning emissions (e.g., van der Laan-Luijkx et al., 2015). Although CO retrievals from these platforms can be biased by 10% or more (De Wachter et al., 2012; Deeter et al., 2016; George et al., 2009), robust signals can still be gleaned since the variation in CO from large biomass burning events can be up to 500% of the background. While not a GHG measurement, solar-induced fluorescence (SIF), a direct by-product of photosynthesis, can be measured from space and is emerging as an important marker of terrestrial gross primary production (Frankenberg et al., 2011; Joiner et al., 2011) and complement to remotely sensed CO2. Direct estimation of gross primary production from SIF retrievals remains an area of active research.
Concentrations and isotopic ratios of carbon cycle process tracers such as COS, CO, Δ14CO2, halogenated species, 13CO2, 13CH4, propane, and ethane are now being regularly analyzed in North American air and as part of the NOAA tower and aircraft networks and targeted regional and local measurement campaigns. These include programs such as the Mid-Continent Intensive (MCI; NACP) campaign, Carbon in Arctic Reservoirs Vulnerability Experiment (CARVE; NASA), Atmospheric Carbon and Transfer-America (ACT-America) program (NASA), Indianapolis Flux Experiment (INFLUX), and Los Angeles megacities effort (see Section 8.3.2). These process tracers allow for constraints on carbon cycle processes such as photosynthetic CO2 fixation, fossil fuel emissions, and transport model fidelity.
8.3.2 Atmosphere-Based Fluxes from Local to Continental Scales
Short-Term and Regional to Local Emissions
Since SOCCR1 (CCSP 2007), studies of the carbon cycle have expanded to include regional campaigns designed to understand and quantify ecosystem and anthropogenic sources and sinks in particular regions and seasons. The NACP MCI campaign intensively sampled the atmosphere above the Midwest agricultural region during 2007 and 2008 and compared sources and sinks derived from atmospheric CO2 data to those based on bottom-up inventories. The results showed a high degree of convergence between surface fluxes inferred from three atmospheric inversions and bottom-up inventories (Ogle et al., 2015; Schuh et al., 2013). CARVE studied boreal and Arctic ecosystem carbon cycling in Alaska using aircraft and tower CO2 and CH4 measurements between 2012 and 2015 (e.g., Chang et al., 2014). One significant finding was that an ensemble of process-based wetland emission models (Melton et al., 2013) systematically underestimated atmospherically constrained CH4 emissions from tundra ecosystems on Alaska’s North Slope (Miller et al., 2016). Recently launched regional studies also should provide new insights into North American carbon cycling. The ACT-America (2015 to 2019) program is designed to explore the structure of GHG distributions within synoptic weather systems and reduce atmospheric transport error in inverse flux estimates using a variety of aircraft observations. The new NASA CARbon Atmospheric Flux Experiment (CARAFE) airborne payload, which is designed for validation of regional carbon flux estimates, was recently deployed to collect airborne eddy covariance measurements for CO2 and CH4 (Wolfe et al., 2015). Other studies such as NASA’s Deriving Information on Surface Conditions from Column and Vertically Resolved Observations Relevant to Air Quality (DISCOVER-AQ) and Arctic Research of the Composition of the Troposphere from Aircraft and Satellites (ARCTAS), as well as the Texas Air Quality Study (TexAQS), have focused primarily on reactive gas compounds and air quality research but also have measured and interpreted CO2 and CH4 data (e.g., Brioude et al., 2012; Townsend-Small et al., 2016; Vay et al., 2011). At much larger scales, the HIAPER (High-Performance Instrumented Airborne Platform for Environmental Research) Pole-to-Pole Observations (HIPPO; 2009 to 2011) and the Atmospheric Tomography Mission (ATom; 2016 to 2018) projects have measured atmospheric trace gas species, including CO2 and CH4, along north-south transects in the Pacific and Atlantic oceans. These measurements are not significantly sensitive to North American emissions, but they are expected to help constrain large-scale carbon fluxes and atmospheric transport and, by extension, improve understanding of the North American carbon balance.
Many studies at more local scales have been designed to provide constraints on urban CH4 and CO2 emissions. A large global trend in urban migration is making cities loci of both emissions and their mitigation, thus driving interest in atmospheric measurement approaches to inform decision making (e.g., Duren and Miller 2012). There have been projects outside of North America (e.g., Bréon et al., 2015; Levin et al., 2011); some North American urban carbon balance studies include those in Indianapolis (INFLUX; Davis et al., 2017), Los Angeles (Feng et al., 2016; Wong et al., 2015; Wunch et al., 2009), Salt Lake City (McKain et al., 2012), and Boston (McKain et al., 2015). In general, these studies have deployed small networks of GHG sensors in and around cities and used the observed gradients, in conjunction with high-resolution atmospheric transport models and bottom-up inventories, to determine urban CH4 and net CO2 emissions (fossil and biogenic). Comparisons between atmospherically derived and bottom-up CO2 emissions show varying degrees of agreement, even in the same city. In Indianapolis, a CO2 flux calculation using tower observations and a high-resolution (1-km) atmospheric inversion system (Lauvaux et al., 2016) yielded emissions about 20% larger than either the Hestia Project (Gurney et al., 2012; Arizona State University) or Open-source Data Inventory for Anthropogenic CO2 (ODIAC; Oda and Maksyutov 2011) inventory products, while aircraft mass-balance fluxes (Heimburger et al., 2017) were about 20% lower than the inventories. Indianapolis airborne mass balance CH4 emissions were about 30% higher than a custom-made urban inventory, and the tower-based inversion suggested CH4 emissions twice as large as the aircraft mass balance estimate. In Salt Lake City, another atmospheric inversion approach using high-resolution (1.3-km) meteorology also showed a high level of correspondence with the Vulcan Project. The California Research at the Nexus of Air Quality and Climate Change (CalNex) mission, which sampled CO2 above Los Angeles, derived emissions 20% to 30% higher than ODIAC and Vulcan (Brioude et al., 2013; Gurney et al., 2012). In the Los Angeles megacities experiment and INFLUX, additional biogenic and anthropogenic process tracers like CO, Δ14CO2, and numerous hydro- and halocarbons also have been measured (Newman et al., 2016; Turnbull et al., 2015). These data could enable partitioning the net CO2 signals into anthropogenic and biogenic components.
Local studies also have been undertaken in and around oil and gas extraction fields. Between 2005 and 2016, U.S. natural gas extraction increased by over 38% (U.S. Energy Information Administration, www.eia.gov/dnav/ng/hist/n9010us2m.htm). The fraction of CH4 that leaks during extraction and distribution is highly uncertain and is driving research on both bottom-up and top-down methods. Alvarez et al. (2012) estimated that if this CH4 leak rate is greater than about 3%, the climate impact of natural gas combustion could equal or exceed that of coal on a per-unit energy basis. Some recent studies of CH4 emissions from oil and gas production (e.g., Brandt et al., 2014) have found higher emissions compared to estimates from past U.S. Environmental Protection Agency (EPA) inventories. Field studies also have shown considerable variation among regions. For example, Karion et al. (2013) found that emissions from the Uintah Basin in Utah were about 9% of production, while Peischl et al. (2015) found leak rates well under 3% of production for the Haynesville, Fayetteville, and Marcellus shale regions. Based on a variety of studies at scales ranging from individual pieces of equipment to regional scales, Brandt et al. (2014) concluded that leakage rates are unlikely to be large enough to make the climate impact of natural gas as large as that of coal.
The answer to the question of why field studies suggest higher emissions than official inventories is likely related to the existence of a small number of “super emitters” that are difficult to capture in inventory-based approaches, but whose atmospheric signatures are often seen in measurements (Brandt et al., 2014; Schwietzke et al., 2017; Kort et al., 2014). For example, Zavala-Araiza et al. (2015) found that half of CH4 emissions from the Barnett Shale region were due to just 2% of oil and gas facilities, and the study achieved closure within error bounds between atmospheric methods and an inventory product derived from local emissions measurements. Although small in area and duration, these measurement campaigns have provided policy-relevant information using atmospheric CH4 concentration data.
Interannual and Continental Emissions
Inverse models such as CarbonTracker have been continuously improved and upgraded to exploit the improved density of atmospheric CO2 and CH4 observations (Bruhwiler et al., 2014). Global inversions with regularly updated flux estimates include CarbonTracker (Peters et al., 2007; carbontracker.noaa.gov), the European Union’s Copernicus Atmospheric Monitoring Service (CAMS; atmosphere.copernicus.eu; formerly MACC), Max Planck Institute Jena CarboScope project (Rödenbeck et al., 2003; www.bgc-jena.mpg.de/CarboScope), and CarbonTracker-Europe from Wageningen University (Peters et al., 2010; www.carbontracker.eu). These products constitute the ensemble of inverse models used in this chapter to estimate North American CO2 fluxes.
Figure 8.3: Inverse Model Estimates of Annual Emissions of (a) Methane (CH4) and (b) Nonfossil Fuel Carbon Dioxide (CO2) from 2000 to 2014
Mean annual CO2 fluxes over North America from this ensemble are shown in Figure 8.3 and listed in Table 8.1. These inverse model flux estimates show some level of agreement about mean fluxes and patterns of interannual variability. However, they also manifest notable differences. These differences remain one of the most important indicators of the overall uncertainty in inverse model fluxes. The uncertainty in fluxes derived from inverse models has proven to be a difficult quantity to estimate directly, since those models depend on results from upstream analyses with complicated, unknown uncertainties. For instance, some of the overall difference in inverse model fluxes can be attributed to differing atmospheric transport among the models, which assume that the winds and diffusive mixing of the transport model are unbiased and subject only to random error. Another element of overall uncertainty comes from the structure of the flux estimation scheme in each inverse model. This structure includes the choice of prior emissions from the burning of fossil fuels, terrestrial biosphere, and the ocean used in the model. The interpretation of results from inverse models is further complicated by the fact that these models retrieve spatiotemporal patterns of CO2 and CH4 fluxes that do not necessarily correspond with patterns expected from differing theories about ecosystem carbon exchange; therefore, they do not map directly onto improvements in process knowledge. Despite these limitations, inverse model results are important because their net carbon flux estimates are by construction consistent with atmospheric data constraints. Ensembles of inverse models using different transport, structure, data inputs, and priors are particularly useful since they mitigate some of these limitations.
Table 8.1. Estimates of Annual, North American, Land Biosphere Carbon Dioxide (CO2) Fluxes (Including Fire) Derived from Atmospheric CO2 Measurements Using Inverse Models and the U.S. Environmental Protection Agency (EPA) Inventory over the Period 2004 to 2013
|CT2015||CAMSa||CTE2015||CarboScopeb||Inverse Models||EPA||Fossil Fuel Emissions|
|Boreal North America||160 ± 77||–356 ± 61||–302 ± 50||–407 ± 64||–306 ± 43||30 ± 1|
|Temperate North America||–352 ± 111||–602 ± 95||–252 ± 126||–365 ± 109||–393 ± 67||–202 ± 5c||1744 ± 37|
|North America||–511 ± 106||–959 ± 117||–555 ± 147||–773 ± 107||–699 ± 82||1774 ± 24|
Emissions in teragrams of carbon (Tg C) per year are listed for the Atmospheric Tracer Transport Model Intercomparison Project’s (TransCom) temperate and boreal North American regions (Gurney et al., 2002). The “inverse models” column averages across the four inverse models (CarbonTracker [CT], Copernicus Atmospheric Monitoring Service [CAMS], CarbonTracker-Europe [CTE], and CarboScope) and represents the best estimate from this ensemble. Fossil fuel emissions are derived from Boden et al. (2017). Values reported are the 2004 to 2013 mean plus or minus a measure of interannual and across-model variability (twice the standard error of the mean of annual emissions). Negative emissions represent a sink.
a Version v15r4, atmosphere.copernicus.eu.
b Version v3.8.
c U.S. EPA (2017) estimates correspond to “managed lands” in the United States, which largely corresponds to the TransCom temperate North American region.
Previous comparisons of inverse models such as Baker et al. (2006) and Peylin et al. (2013) indicated that, while each inversion manifests a different long-term mean flux estimate, the patterns of interannual variability tend to have better agreement. There is some indication of interannual variation coherence in the present collection of models, but with some significant disagreement, mainly from the Jena CarboScope model. Averaging across the inversions, the land biosphere sink in North America, including fire emissions, averaged over 2004 to 2013 is 699 ± 82 Tg C per year (mean ± two standard errors of the mean of the interannual and intermodel variability). This sink offsets about 39% of the fossil fuel emissions of 1,774 ± 24 Tg C per year for the same geographic area, although 98% of these anthropogenic emissions come from just the temperate North American region. Disagreement remains among these inversions about the average size of the North American sink, but they all estimate significant interannual variability in that sink. Over the temperate North American region, these inverse models estimate interannual variability (one sigma) of between 163 and 277 Tg C per year, equivalent to 45% to 83% of each model’s mean flux.
The level of interannual variability from inverse models stands in stark contrast to the annual Inventory of U.S. Greenhouse Gas Emissions and Sinks, prepared by the U.S. EPA. EPA’s U.S. GHG inventory estimates land use, land-use change, and forestry (LULUCF) sector emissions on managed lands. Managed lands represent about 95% of total U.S. land cover and more than 99% of the conterminous United States, which corresponds well to the net biosphere fluxes estimated by inversion models for temperate North America. EPA’s LULUCF CO2 sink estimate has a 2004 to 2013 mean of 202 ± 5 Tg C per year (U.S. EPA 2017; mean plus or minus two standard errors of the mean). The small interannual variability in the EPA inventory of just 5 Tg C per year stands in contrast to all the inverse models. This low apparent variability may arise from the historical 5- to 14-year frequency at which U.S. Forest Service Forest Inventory and Analysis (FIA) plots have been resampled. Comparing the interannual variability of inventories and inversions is inherently difficult due to the mismatch in their temporal sensitivities.
Various estimates of North American surface CO2 emissions were collected as part of the recent NACP regional interim synthesis (Hayes et al., 2012; Huntzinger et al., 2012) and REgional Carbon Cycle Assessment and Processes (RECCAP) effort (Canadell et al., 2011; King et al., 2015). The RECCAP North America study included a suite of inverse models collected by Peylin et al. (2013) with a 2000 to 2009 mean CO2 sink of 890 ± 400 Tg C per year (mean and one sigma standard deviation), implying a larger sink than either inventory (270 Tg C per year) or terrestrial biosphere model (359 ± 111 Tg C per year) estimates (King et al., 2015). The current suite of inverse models collected for this report (see Table 8.1) suggests North American biosphere emissions of 699 ± 82 Tg C per year averaged over 2000 to 2014. The models collected for this chapter also supplied results from their earlier versions to the RECCAP ensemble of Peylin et al. (2013). That report showed a wide range of North American flux estimates, but the subset of models used in this chapter all manifested sinks smaller than 500 Tg C per year for North America over the reporting period 2001 to 2004, whereas the other models all estimated greater sinks between about 500 and 1,500 Tg C per year.
The North American sink estimated from the suite of inverse models collected for this report agrees well with previous bottom-up estimates. SOCCR1 (Pacala et al., 2007) reported a sink of 666 ± 250 Tg C per year for 2003. This estimate was derived from bottom-up inventories and models and did not include information from atmospheric inverse models. Hayes et al. (2012) attempted to reconcile net biosphere emissions estimates from inventories, terrestrial biosphere models, and atmospheric inverse models averaged over 2000 to 2006 for North America. That study found a sink of 511 Tg C per year simulated by terrestrial biosphere models and an inventory-based sink estimate of 327 Tg C per year (with an estimate of additional noninventoried fluxes that brings the total sink estimate to 564 Tg C per year). The collection of inverse models used in that study manifested significantly larger sinks (981 Tg C per year) than the current collection. See Ch. 2: The North American Carbon Budget for an assessment of the overall agreement of these various estimates of North American surface CO2 exchange with the atmosphere.
The use of regional models of CO2 and CH4 has become more common since SOCCR1. These models have focused, for example, on continental-scale processes (Butler et al., 2010; Gourdji et al., 2012; Schuh et al., 2010) or at the scale of the mid-continent (Lauvaux et al., 2012b; Schuh et al., 2013). Regional model CO2 flux estimates for North America so far have been published for periods of up to 1 year, with multiyear analyses only available from global inversion approaches. One prominent result from regional inverse CO2 studies is the sensitivity of the annual net CO2 flux to defining the inflow of atmospheric CO2 into the study region (Gourdji et al., 2012; Schuh et al., 2010). Lauvaux et al. (2012b) demonstrated that this sensitivity could be minimized with observations at the inflow boundaries. This finding highlights the importance of global-scale measurement networks and carbon reanalysis systems for understanding North American carbon fluxes. More recently, CH4 has received more attention with regional inversions for the continent (Kort et al., 2008; Miller et al., 2013), California (Jeong et al., 2013), and Alaska (Chang et al., 2014; Miller et al., 2016). Additional uncertainties in inverse modeling approaches arise from sparse data coverage. When the observational network is not strongly sensitive to particular land regions, inverse modeling systems must make assumptions about spatial and temporal patterns of emissions. As with the issue of boundary inflow, mitigating this sensitivity necessitates building a denser, intercalibrated measurement network.
See Full Chapter & References