CarbonTracker CT2022

CarbonTracker is a CO2 measurement and modeling system developed by NOAA to keep track of sources (emissions to the atmosphere) and sinks (removal from the atmosphere) of carbon dioxide around the world. CarbonTracker uses atmospheric CO2 observations from a host of collaborators and simulated atmospheric transport to estimate these surface fluxes of CO2. The current release of CarbonTracker, CT2022, provides global estimates of surface-atmosphere fluxes of CO2 from January 2000 through December 2020.

What is CarbonTracker?

CarbonTracker is a global model of atmospheric carbon dioxide with a focus on North America, designed to keep track of CO2 uptake and release at the Earth's surface over time. [read more]

Who needs CarbonTracker?

Policy makers, industry, scientists, and the public need CarbonTracker information to make informed decisions to limit greenhouse gas levels in the atmosphere. [read more]

What does CarbonTracker tell us?

North America is a source of CO2 to the atmosphere. The natural uptake of CO2 that occurs mostly east of the Rocky Mountains removes about a third of the CO2 released by the use of fossil fuels. [read more]

CarbonTracker CO2 weather for June-July, 2008. Warm colors show high atmospheric CO2 concentrations, and cool colors show low concentrations. As the summer growing season takes hold, photosynthesis by forests and crops draws concentrations of CO2 down, opposing the general increase from fossil fuel burning. The resulting high- and low-CO2 air masses are then moved around by weather systems to form the patterns shown here. [More on CO2 weather]

Global CO2 budget

Since the beginning of the atmospheric measurement record, annually-averaged CO2 concentrations have risen without interruption. This continuous increase is driven by the burning of fossil fuels. Over the CarbonTracker analysis period, CO2 emissions to the atmosphere from the burning of fossil fuels generally have risen year-on-year, starting at 6.8 Pg C yr-1 in 2000 and reaching a maximum of 10.1 Pg C yr-1 in 2019 (1 petagram of carbon is 1015 g C, or 1 billion metric tons C, or 3.67 billion metric tons CO2). The two significant exceptions to this steady increase in emissions are both attributable to major global economic disruptions: emissions fell by 0.1 Pg C yr-1 from 2008 to 2009 as a result of the global economic recession, and by about 0.6 Pg C yr-1 from 2019 to 2020 due to the COVID pandemic (Figure 1). These small decreases are not large enough to significantly slow the growth of atmospheric CO2 concentrations. It is estimated that fossil fuel emissions would have to be reduced by about 80% to halt the growth in atmospheric CO2.

Fossil fuel emissions are concentrated in areas with high population density and economic activity, and emissions inventory information used in CT2022 indicates that 82% of fossil fuel emissions come from the industrialized northern extratropics.

Figure 1. Annual global emissions. The bars in this figure represent CO2 emissions for each year in Pg C yr-1 over the globe. CarbonTracker models four types of surface-to-atmosphere exchange of CO2, each of which is shown in a different color: fossil fuel emissions (tan), terrestrial biosphere flux excluding fires (green), direct emissions from fires (red), and air-sea gas exchange (blue). Negative emissions indicate that the flux removes CO2 from the atmosphere. The net surface exchange, computed as the sum of these four components, is shown as a thick black line. [Explore these data further]

The other major source of CO2 is wildfires, which in CT2022 add an additional 2.0-2.6 Pg C yr-1 to the atmosphere. In contrast to fossil fuel emissions, wildfire CO2 comes principally from tropical and southern land. 87% of wildfire emissions in CT2022 are in those regions.

Offsetting these sources are natural sinks on land and in the ocean. Together, these sinks absorb about half the anthropogenic CO2 emitted into the atmosphere. Over the period 2001-2020, the CT2022 global sum of "natural" fluxes (fire emissions, the land biosphere sink, and the ocean sink) is 46% of fossil fuel emissions over the same time period. The atmospheric CO2 growth rate would be about twice the observed rate without these sinks. CarbonTracker is designed to identify these sinks in order to better understand the mechanisms behind them.

According to CT2022, the world oceans absorb 1.4 to 4.0 Pg C yr-1. This natural sink exists as a direct result of increasing atmospheric CO2 concentrations, because dissolved carbon concentrations in the ocean increase to reach equilibrium with the atmosphere. However, the large-scale circulation of the ocean and biological, physical, and chemical carbon cycling cause there to be a source of carbon to the atmosphere in the tropics. In CT2022, this natural source of between 0.4 and 1.1 Pg C yr-1 in the tropics is offset by large extratropical sinks of 2.4 - 4.5 Pg C yr-1.

The terrestrial biosphere is also a net sink of CO2, due mainly to two processes. These are CO2 fertilization, in which plants grow faster since they can more easily acquire carbon dioxide for photosynthesis, and the effects of human land-use practices, including fertilization, irrigation, fire suppression, and recovery from past land use. CarbonTracker attempts to locate these land sinks spatially and temporally, so hypotheses about their mechanisms can be tested. CT2022 finds widely-scattered terrestrial CO2 sinks, with significant absorption of carbon dioxide by northern temperate and boreal regions (1.3 - 2.5 Pg C yr-1, about 55% of the global land total of 2.1 - 4.5 Pg C yr-1 ).

In this text, fluxes reported are reported as ranges when possible, to provide some context for how well CarbonTracker constrains the long-term average flux. These ranges are computed as the minimum and maximum values from the sequence of annual CT2022 flux estimates, often excluding the year 2000 as we consider it a spin-up year in which estimats may not be representative of the actual long-term mean. A large range indicates a high degree of interannual variability in the flux estimate.

CO2 sources and sinks over North America

CT2022 results indicate that North American ecosystems have been a net sink of 0.7 (range 0.4 - 1.0) Pg C yr-1 over the period 2001-2020. This natural sink offsets about one-third of the emissions of about 1.7  Pg C yr-1 from the burning of fossil fuels in the U.S.A., Canada and Mexico combined.

Figure 2. Drought and land sinks over North America. Top Panel: United States Drought Monitor percent area of the continental U.S. undergoing different levels of drought intensity. Bottom Panel: Annual land sink estimates (including fire emissions) from CT2022 for temperate North America.

Whereas North American fossil emissions are generally steady over the CarbonTracker period, ranging between 1.4 and 1.8 Pg C yr-1, the amount of CO2 taken up by the North American biosphere varies significantly from year to year (Figure 2, bottom panel). In terrestrial biosphere models, inter-annual variability in land uptake can be related to anomalies in large-scale temperature and precipitation patterns. While the CarbonTracker period of analysis is relatively short compared to the dynamics of slowly-changing pools of biospheric carbon, episodes of extremes in net ecosystem exchange (NEE) have been associated with climatic anomalies (see e.g. Peters et al., 2007). CT2022 annual estimates of the land sink over temperate N. America are clearly related to continental-scale drought intensity (Figure 2). It is interesting to note that the inferred year-to-year variabilty (the "range") of land uptake is actually as big as the average sink itself.

Widespread droughts in the U.S. west and Canada during 2002, 2006, 2008, 2011, and 2012 resulted in relatively small annual uptake by terrestrial ecosystems in temperate North America (Figure 2). In these years, land ecosystems accounted for a sink of only about 0.3 Pg C yr-1. This is about six-tenths the sink of an "average" year, in which these same land ecosystems remove about 0.5 Pg C yr-1.

Spatial distribution of North American surface fluxes

CarbonTracker flux estimates include sub-continental patterns of sources and sinks coupled to the distribution of dominant ecosystem types across the continent (Figure 3). We have greater confidence in countrywide totals than in estimates of regional sources and sinks, but we expect that such finer-scale estimates will become more robust with future expansion of the CO2 observing nework. Our results indicate that the sinks are mainly located in the agricultural regions of the U.S. and Canadian midwest, and boreal forests in Canada.

Figure 3. Average ecosystem fluxes. The pattern of net ecosystem exchange (NEE) of CO2 of the land biosphere averaged over 2001-2015, as estimated by CarbonTracker CT2022. This NEE represents land-to-atmosphere carbon exchange from photosynthesis and respiration in terrestrial ecosystems, and a contribution from fires. It does not include fossil fuel emissions. Negative fluxes (blue colors) represent CO2 uptake by the land biosphere, whereas positive fluxes (red colors) indicate regions in which the land biosphere is a net source of CO2 to the atmosphere. Units are g C m-2 yr-1.

Word of caution about high-resolution biological flux maps Figure 3 shows estimated fluxes by ecoregion. While we also provide flux maps and data with a finer 1° x 1° spatial resolution, for example on our flux maps pages, these ecoregions define the actual scales at which CarbonTracker operates. With the present observing network, the detailed one-degree fluxes should not be interpreted as quantitatively meaningful for each block. Any within-ecoregion patterns come directly from results of the terrestrial biosphere model. Part of this high-resolution patterning comes from variations of temperature, precipitation, light, plant species, and soil type over each ecoregion. To spread the influence of measurements from the sparse observation network, CarbonTracker makes adjustments uniformly over an entire ecoregion. These adjustments scale the net ecosystem flux of CO2 predicted by the terrestrial biosphere model by the same factor across each ecoregion. Thus we caution that the 1° x 1° spatial detail in CarbonTracker land fluxes is based on the simulations of the terrestrial biosphere model and the assumption of large-scale ecosystem coherence. This has not been verified by observations.

The CarbonTracker observing system

CarbonTracker surface flux estimates are optimally consistent with atmospheric CO2 observations from the GLOBALVIEWplus 7.0 and 8.0 ObsPack, comprising 559 time series datasets from around the world using a variety of measurement techniques and platforms (Table 1, Figure 4). These observation are contributed by collaborators from 54 different laboratories. Most of the CO2 observational data can be accessed by downloading the GLOBALVIEWplusv8.0 ObsPack, or if modeled observations are also required, the CT2022 ObsPack. More information on CO2 measurements used in CT2022 can be found in the observations documentation.

ObservationNumber ofNumber ofNot ForAssimilation Observations
TypeDatasetsObservations AssimilationTotalAcceptedRejectedWithheld
surface-insitu209 9 772 615 8 183 7001 588 9151 459 32451 37578 216
surface-flask123 80 415 14 32366 09262 1688903 034
surface-pfp29 43 150 21 56821 58220 494219869
tower-insitu105 7 794 559 6 480 7811 313 7781 213 50836 45263 818
aircraft-pfp39 78 648 15 53563 11357 7192 2773 117
aircraft-flask6 16 032 12 4703 5623 4786717
aircraft-insitu35 8 131 898 8 031 639100 25993 4674416 351
shipboard-flask3 3 163 5032 6602 5148561
shipboard-insitu10 1 707 612 467 1911 240 4211 130 04348 34662 032
Total559 27 628 092 23 227 7104 400 3824 042 715140 152217 515
Table 1. CarbonTracker CT2022 observations by observation type, which comprises measurement platform and instrument (in ObsPack parlance this is a "project"). More information on CO2 measurements used in CT2022 can be found in the observations documentation.
Figure 4. CarbonTracker Observational Network Click on any site marker for more information. Double-click on a site marker to center the map on that site.

Calculated time-dependent CO2 fields throughout the global atmosphere

A "byproduct" of the data assimilation system, once sources and sinks have been estimated, is that the mole fraction of CO2 is calculated everywhere in the model domain and over the entire 2000-2016 time period, based on the optimized source and sink estimates (Figure 1). As a check on model transport properties and CarbonTracker inversion performance, calculated CO2 mole fractions are regularly compared with measurements from aircraft campaign datasets taken by NOAA/ESRL and collaborators. These independent samples are not used to estimate fluxes in CarbonTracker, but rather set aside for cross-validation.

Since CarbonTracker simulates CO2 throughout the entire atmospheric column, the model atmosphere can be sampled like satellite (GOSAT and OCO-2) and ground-based remote sensing instrument (TCCON) retrievals of CO2. Examples of our agreement with the latter can be found on our TCCON page.

Flux uncertainties

Figure 5. Carbon dioxide weather Shown is the daily average of the pressure-weighted average mole fraction of carbon dioxide in the free troposphere as modeled by CarbonTracker for March 20, 2009. Units are micromoles of CO2 per mole of dry air (μmol mol-1), and the values are given by the color scale depicted under the graphic. The "free troposphere" in this case is levels 6 through 10 of the TM5 model. This corresponds to about 1.2km above the ground to about 5.5km above ground, or in pressure terms, about 850 hPa to about 500 hPa. Gradients in CO2 concentration in this layer are due to exchange between the atmosphere and the earth surface, including fossil fuel emissions, air-sea exchange, and the photosynthesis, respiration, and wildfire emissions of the terrestrial biosphere. These gradients are subsequently transported by weather systems, even as they are gradually erased by atmospheric mixing.

It is important to note that at this time the uncertainty estimates for CarbonTracker sources and sinks are themselves quite uncertain. They have been derived from the mathematics of the ensemble data assimilation system, which requires several educated guesses for initial uncertainty estimates. The paper describing CarbonTracker (Peters et al. (2007), Proc. Nat. Acad. Sci. vol. 104, p. 18925-18930) presents different uncertainty estimates based on the sensitivity of the results to 14 alternative yet plausible ways to construct the CarbonTracker system. For example, the 14 realizations produce a range of the net annual average terrestrial emissions in North America of -0.40 to -1.01 Pg C yr-1 (negative emissions indicate a sink). The procedure is described in the Supporting Information Appendix to that paper, which is freely downloadable from the PNAS web site.

Furthermore, the estimates do not take into account several additional factors noted below. The calculation is set up for sources and sinks to slowly revert, in the absence of observational data, to first guesses of net ecosystem exchange, which are close to zero on an annual basis. This set-up may result in a bias. Also due to the sparseness of measurements, we have had to assume coherence of ecosystem processes over large distances, giving existing observations perhaps an undue amount of weight. The process model for terrestrial photosynthesis and respiration was very basic, and will likely be greatly improved in future releases of CarbonTracker. Easily the largest single annual average source of CO2 is emissions from fossil fuel burning, which are currently not estimated by CarbonTracker. We use estimates from emissions inventories (economic accounting) and subtract the CO2 mole fraction signatures of those fluxes from observations. As a result, the biosphere and ocean fluxes estimated by CarbonTracker inherit error from the assumed fossil fuel emissions. While these emissions inventories may have a small relative error on global scales (perhaps 5 or 10%), any such bias translates into a larger relative error in the annual average ecosystem sources and sinks, since those fluxes have smaller magnitudes. We expect to add a process model of fossil fuel combustion in future releases of CarbonTracker. Finally, additional measurement sites are expected to lead to the greatest improvements, especially to more robust and specific source/sink results at smaller spatial scales.

Consistency of modeled and observed atmospheric CO2 growth rates

Global atmospheric CO2 growth rates inferred directly from observed carbon dioxide at marine surface sites are consistent with those modeled by CarbonTracker, both in their average values and in their year-to-year variations (Figure 6). These global growth rates hovered at around 4 Pg C yr-1, or around 1.9 ppm yr-1 (Figure 6) in the decade of the 2000s, but have increased in recent years. The observed global growth rates in 2015 and 2016 are particularly high, at 6.3 Pg C yr-1 and 6.0 Pg C yr-1 (just under 3.0 ppm yr-1). This anomaly is believed to be related to the large 2015-16 El Niño (see below).

Figure 6. Atmospheric CO2 growth rates. Observed atmospheric CO2 growth rates (source: NOAA ESRL page on global trends in CO2) are plotted against the atmospheric CO2 growth rate inferred from CT2022 global fluxes. Note that error bars on the observed growth rates are relatively small and may not be visible on this plot.

The 2015-2016 El Niño

The large El Niño of 2015 and 2016 was responsible for two years of sustained record global growth rates in atmospheric CO2 (Figure 6). With its extended assimilation window and expanded network of measurement data, CT2019 simulates these growth rate anomalies quite well. In terms of surface fluxes, CT2019 finds that land ecosystems in the tropics were responsible for a CO2 emissions anomaly of about 1.2 Pg C during this event (see Figure 7).

Figure 7. Tropical land flux anomalies simulated by CarbonTracker. Our previous two releases are shown in green (CT2017) and red (CT2019B). CT2022, with its longer time span and modifications to represent the El Niño, is shown in blue. Depicted are monthly anomalies of the sum of imposed fire and optimized non-fire tropical land flux in Pg C yr-1. Thick lines have been low-pass filtered using a Hanning window of length 13 months.

2007 CarbonTracker PNAS publication

Full bibliography of studies using CarbonTracker.

CarbonTracker is a NOAA contribution to the North American Carbon Program

Other CarbonTrackers