Regional Changes in Carbon Dioxide Fluxes of Land and Oceans Since 1980

See allHide authors and affiliations

Science  17 Nov 2000:
Vol. 290, Issue 5495, pp. 1342-1346
DOI: 10.1126/science.290.5495.1342


We have applied an inverse model to 20 years of atmospheric carbon dioxide measurements to infer yearly changes in the regional carbon balance of oceans and continents. The model indicates that global terrestrial carbon fluxes were approximately twice as variable as ocean fluxes between 1980 and 1998. Tropical land ecosystems contributed most of the interannual changes in Earth's carbon balance over the 1980s, whereas northern mid- and high-latitude land ecosystems dominated from 1990 to 1995. Strongly enhanced uptake of carbon was found over North America during the 1992–1993 period compared to 1989–1990.

Over the past two decades, on average, about half of the CO2 emissions caused by fossil fuel combustion have remained in the atmosphere, the rest having been absorbed by the ocean and by land ecosystems. Year-to-year variations in the rate of atmospheric CO2 accumulation are of the same magnitude as the decadal mean annual accumulation and result primarily from shifts in the natural carbon fluxes (1). Previous investigations into which reservoir (land or ocean) and which regions caused such year-to-year changes have produced conflicting answers. Carbon stable isotope studies all infer large shifts of both land and ocean fluxes of up to several gigatonnes (1015 g) of carbon per year (GtC year−1) (2, 3). In contrast, ocean carbon models and measurements of the CO2 partial pressure difference between the ocean surface and the atmosphere (Δp co 2) suggest relatively small changes in the air-sea fluxes (4–6). Global biogeochemical models of the land biosphere generally produce large interannual shifts in terrestrial fluxes, but they differ in where or how they attribute these shifts to underlying processes (photosynthesis or respiration).

Inverse models using atmospheric CO2 observations and atmospheric transport have been applied to infer the mean spatial distribution of CO2 fluxes (7–9), but rarely to estimate their interannual variability (10). Here, we constructed an inversion using 20 years of atmospheric CO2measurements, mostly from the NOAA Climate Monitoring and Diagnostics Laboratory air sampling network, to infer monthly changes in the carbon balance of large regions. The carbon balance of continents and oceans can be considered as the sum of two components, a long-term mean net flux (over 20 years) and a monthly varying flux anomaly. In the following, we present and discuss the monthly varying flux anomalies. The results of the inverse approach (“top-down”) are compared with predictions of two state-of-the-art global models (“bottom-up”) of the carbon fluxes over land ecosystems and oceans.

Over the past 20 years, the annual accumulation of CO2 in the atmosphere has varied between 1 and 6 GtC year−1 (1). Because fossil CO2emission changes do not vary much from year to year, the observed changes in accumulation rate reflect variations of ocean and land fluxes. At present, about 120 CO2 data records from around the globe are available (11). Most CO2 stations are in the marine boundary layer; they can be influenced directly by ocean fluxes and more indirectly by land fluxes. A few CO2stations, however, are close to or within the continents and can better capture year-to-year changes in terrestrial fluxes. Among the 120 sites available in 1998, we have selected 67 sites (12). At each site, we have analyzed the variance of the deseasonalized trend after subtraction of the trend at the South Pole (13). This analysis indicates that there is “excess” variance at low frequency at continental sites compared to oceanic sites (Fig. 1). This excess variance may reflect short-term spatial and temporal variability of land fluxes or of atmospheric transport. However, it also suggests that terrestrial carbon fluxes exhibit larger year-to-year variations than their oceanic counterparts. This inference can be evaluated by using an inverse model to calculate regional carbon balance variations from observed concentration variations.

Figure 1

Year-to-year variability of the CO2 trend at the monitoring sites used in the inverse procedure over the period 1980–1998. The variability is estimated by the standard deviation of the weekly CO2 deseasonalized trend after subtraction of the South Pole trend (13). Sites further toward the red end of the scale indicate stronger trend variations, which are presumably caused by larger variability of the sources influencing the station. The 11 continental regions (gray areas) and the eight ocean regions (separated by lines) used in the inverse procedure are also shown. For the continents, from north to south: Arctic regions (far North America and far north Eurasia), boreal and temperate North America, boreal and temperate Europe, boreal and temperate North Asia, tropical Africa, south America, tropical Asia, and all lands south of 20°S. For the oceans: North Atlantic, >50°N; North Pacific, >50°N; North Atlantic, 20° to 50°N; North Pacific, 20° to 50°N; Equatorial Pacific, 20°S to 20°N; sum of Equatorial Indian and Atlantic, 20°S to 20°N; subtropical, sub-Antarctic oceans, 20° to 50°S; and Southern Ocean, >50°S. A total of 67 sites are used in this study. The Mace-Head (MHD, Ireland) continuous record is split in two time series according to continental and oceanic origin of air at the station (two dots).

We have developed such a model, extending the work of (9), to retrieve the net CO2 fluxes every month from 1980 to 1998 (13). The inverse model optimizes CO2 ocean and land fluxes for 11 continental regions and eight ocean regions (Fig. 1) by minimizing the differences between the CO2 concentrations simulated by a three-dimensional atmospheric transport model and those observed at measurement sites. Fossil CO2 fluxes are prescribed from energy use statistics (14). The control inversion is described in (13). The atmospheric CO2 data used for the inversion are calculated from 67 selected monitoring sites (12) over the period 1980–1998. Raw flasks and in situ records are smoothed in the time domain to remove synoptic variability (11) and are used in the form of monthly means. An increasing number of stations is available over time, from 20 sites in 1980 to 67 sites in 1997, with 35 new sites appearing between 1987 and 1991. Data uncertainties are estimated each month at each station from the (synoptic) scatter and measurement uncertainties of the original flask data (13). In addition to the control inversion (13), we have carried out a sensitivity study consisting of seven additional inversions in which key parameters are varied individually (15), providing a range of uncertainty on the inferred fluxes. The sensitivity study was performed to better account for uncertainties that are not explicitly represented in the inverse procedure, which only returns a residual uncertainty.

At the regional scale, we find that the long-term mean net fluxes are significantly different among the eight inversions, within a typical range of 0.4 GtC year−1 for ocean regions and of 0.8 GtC year−1 for land regions. Independent inverse modeling studies of the mean net fluxes tend to confirm this result (7–9). On the other hand, the inferred flux anomalies are substantially more similar among our sensitivity tests, which suggests that anomalies are retrieved more robustly than long-term mean fluxes (16). The latter are inferred from mean spatial concentration differences among stations, which are rather small within a given latitude band. For instance, the apportionment of sources and sinks between North America and Eurasia relies on spatial mean differences on the order of 0.5 parts per million (ppm) at mid-northern latitudes (7). In contrast, flux anomalies for these regions are inferred from temporal changes of concentration differences between stations, which are larger than mean longitudinal differences (see below). Accordingly, we focus on monthly flux anomalies.

During the 1980–1998 period, land fluxes are found to be twice as variable as ocean fluxes (Fig. 2), in agreement with the qualitative analysis in Fig. 1. The anomalies have a peak-to-peak amplitude of 5.0 GtC for the total global terrestrial flux and 2.5 GtC for the global air-sea flux. Anomalous uptake over land is inferred during 1981–1983, 1991–1993, and 1997, and anomalous uptake over oceans persists from 1989 to 1996. Unlike previous studies that relied on atmospheric carbon isotope records (2, 3), we infer no significant anticorrelation in the flux anomalies between ocean and land. In this study, the ocean flux anomalies are always smaller than the land anomalies. The inversion with the minimum variability in the air-sea flux anomalies (half as large as the control run) corresponds to larger observational uncertainties, allowing a looser fit to the ocean stations. The inversion with the maximum variability in the air-sea flux anomalies (1.3 times as large as the control run) occurs when no constraint is applied to the global mean ocean uptake (13).

Figure 2

(left). Inferred anomalous changes in the global land (A) and ocean (B) carbon fluxes. Black lines are the average of the eight inversions. Shaded areas represent the range of values obtained from the eight sensitivity inversions. For each inversion, we calculated monthly flux anomalies by subtracting the long-term mean flux over the period 1980–1998, and then used a 12-month running mean to deseasonalize the anomalous flux.

No individual ocean basin is inferred to contribute predominantly to the global ocean flux anomaly, although both the Southern Ocean (south of 50°S) and the equatorial zone exhibit relatively larger flux anomalies. The Equatorial Pacific (20°S to 20°N) is known to exhibit year-to-year variations in the air-sea CO2 fluxes, as shown by repeated surveys (17–20) of measured Δp co 2. The year-to-year variations found in our inversions are in good agreement with the ocean flux anomalies derived from Δp co 2measurements (17–20). Both approaches estimate an anomalous CO2 sink of 0.1 to 0.5 GtC year−1 during the strong El Niño events of 1982–1983, 1986–1987, and 1997–1998, and an anomalous source of the same amplitude during the La Niña event of 1988–1989. The agreement in this region may be partially due to the relatively dense observational network in the atmosphere, with 20% of the flask data (Fig. 1), and in the ocean with the dense Δp co 2 coverage (17–20).

Our inversions also compare favorably with the ocean carbon model of (21) for the amplitude of the year-to-year variations both in the Equatorial Pacific (Fig. 3A) and in the Southern Ocean [not shown, see (22)]. However, some phasing differences remain between the two approaches. Outside the Equatorial Pacific and Southern Ocean, the inversions and the ocean carbon model disagree: The inversions give much larger variations than the ocean model (±0.4 versus ±0.1 GtC year−1). The few basin scale observations do not justify a preference of one estimate over the other. The ocean model could underestimate variability because it does not include continental margins, and because it underestimates the variability in ocean dynamics at high latitudes (21). On the other hand, the inversions could overestimate variability because the low density of the atmospheric stations makes it difficult to draw a precise line between land and ocean, leaving the possibility that some of the land variability has been attributed to the oceans.

Figure 3

(right). Carbon balance anomalies of tropical regions. Black lines are the average of the eight inversions. Shaded areas represent the range of values obtained from the eight sensitivity inversions. Thick arrows indicate strong El Niño events documented by minima in the Southern Oscillation Index. Thin arrows indicate weak El Niño conditions. (A) Equatorial Pacific region, 20°S to 20°N. The (deseasonalized) flux anomalies derived by the inversion are compared to those predicted by one global three-dimensional ocean carbon model (blue line) that computes changes in ocean circulation and in marine biology (21), as well as to those estimated on the basis of Δp co 2 measurements compiled in (17–20) (blue squares). Uncertainties in the flux estimates (rightmost blue squares) are composed of the seasonal variations of p co 2 in water, wind speed variability, and wind speed dependence of gas exchange (18). Oceanic flux anomalies from the model of (21) are averaged over the same regions as the inversion. Δp co 2-based flux anomalies are extrapolated to different surfaces of the equatorial Pacific Ocean as described in (17–20). (B) Tropical land regions, 20°S to 20°N. The deseasonalized flux anomalies derived by the inversion are compared to those predicted by a global biogeochemical model of the terrestrial biosphere [green line, see (23)] averaged over the same regions as the inversion. Note that land use–induced and disturbance-induced carbon flux anomalies are not included in the biogeochemical model.

During the 1980s, tropical land regions are found to contribute more anomalous changes to the global carbon balance than mid- and high-latitude ecosystems, whereas the converse is true for the period 1990–1995. During the period 1996–1998, tropical and Northern Hemisphere land ecosystems contribute equally to the total flux anomaly (13). Observed changes in the CO2 growth rate partly illustrate that result. For the periods 1980–1985, 1989–1991, and 1995, growth rate anomalies start in the tropics and propagate toward high latitudes within a few months. During the period 1992–1993, a negative growth rate anomaly initiates at mid-northern latitudes and then propagates toward the Equator (13).

In the 1980s, tropical land regions predominantly influence the carbon flux anomalies in all eight inversions performed (Fig. 3A), except during 1984–1985. An anomalous source is inferred during El Niño years in 1987–1988 and 1998, whereas an anomalous sink occurs during 1982, 1985–1986, and 1989–1993. A strong El Niño occurred in 1982–1983, for which the inversions produce only a relatively small positive flux anomaly over tropical land regions (Fig. 3B). This may reflect the low station density in the tropics in the early 1980s. The interannual variations estimated from the inversion are in reasonable agreement with those predicted by the global biogeochemical model of (23) (r = 0.55, Fig. 3B), except for the period 1996–1998. Note that the inverse estimates combine changes in land use, biomass burning of natural origin and induced by humans, and biogeochemical carbon sources and sinks, whereas the model of (23) only accounts for climate-driven biogeochemical effects. Among tropical continents, the Amazon is a major contributor to anomalous carbon balance variations (24). However, the partition of the land fluxes within the tropical band can only be tentative, given the absence of sampling stations on the tropical continents and the small magnitude of atmospheric signals (ppm per GtC) caused by strong convective vertical “dilution” of the surface fluxes.

Northern Hemisphere land areas predominantly influenced the carbon flux anomalies during the early 1990s. A strong drop in growth rate occurred during 1992–1993 at mid-northern latitudes (1). We invert this signal into an enhanced terrestrial uptake over the Northern Hemisphere continents. Terrestrial carbon storage increased there by 1.4 GtC year−1 from the biannual mean in 1989–1990 to that of 1992–1993 (Fig. 4A), in accordance with previous analyses of atmospheric carbon isotopes (25) and oxygen data (26). Our regional inversions locate the 1992–1993 enhanced terrestrial sink predominantly over North America (∼2.0 GtC). This striking result is also directly visible in the CO2 observations. Figure 4B shows the annual mean difference in CO2 concentration between Atlantic and Pacific stations, which relates to the North American carbon balance (27). The Atlantic stations were 0.5 ppm higher than those in the Pacific during 1989–1990 but were 0.5 ppm lower during 1992–1993 (Fig. 4B). The enhanced North American uptake during 1992–1993 is inferred by all but one of the seven sensitivity runs (28). Between 1989 and 1993, North American and Eurasian carbon fluxes are anticorrelated (r = –0.65, see Fig. 4A), but the enhanced uptake over North America remains on average three times the size of the reduced uptake over Eurasia. Moreover, the error correlation between those two regions estimated by the inversion is not significant (r = –0.35), which indicates that the present atmospheric network may be able to correctly separate anomalous changes in Eurasia versus North America in the early 1990s.

Figure 4

Carbon balance anomalies of Northern Hemisphere land regions after 1988. (A) Terrestrial carbon flux anomalies. Black line is the average of the eight inversions for northern extratropical lands. Shaded area represents the range of values obtained from the eight sensitivity inversions. Flux anomalies are separated between North America (purple line) and Eurasia (orange line). North America corresponds to the aggregated results of the inversion over three regions (temperate, boreal, and arctic in Fig. 1), and Eurasia corresponds to aggregated results over four regions (temperate and boreal Europe, temperate and boreal Siberia in Fig. 1). Note that we restrict this plot to the period 1987–1998 because the partition of land fluxes between Eurasia and North America is constrained by too few sites in the early 1980s (14 sites in 1980 and 20 sites in 1985 for the whole Northern Hemisphere). (B) Observed CO2 concentrations at Eastern Pacific stations (left) and Atlantic stations (right) between 20° and 70°N for two different biannual mean periods (1989–1990 and 1992–1993) and one annual period (1996.5–1997.5). Changes in the difference of concentrations between Pacific and Atlantic sites partly reflect shifts in the carbon balance of North America. For each of the periods 1989–1990, 1992–1993, and 1996–1997, we calculated anomalous concentrations at each site by subtracting the mean concentration calculated using Pacific sites only. This represents anomalies at Atlantic sites compared to Pacific sites [see (7), figure 3]. In the Atlantic, each period is summarized by the mean anomaly shown as a horizontal line. In the Pacific, all mean anomalies are zero (the segment “All”). Data are shown for the following locations: BME (32.37°N, 64.65°W), BMW (32.27°N, 64.88°W), CBA (55.20°N, 162.72°W), CMO (45.48°N, 123.97°W), CSJ (51.93°N, 131.02°W), ICE (65.00°N, 24.00°W), IZO (28.30°N, 16.48°W), KEY (25.67°N, 80.20°W), MHT (53.33°N, 9.70°W), STM (66.00°N, 2.00°E), P05 (25.00°N, 154.00°W), P04 (30.00°N, 148.00°W), P03 (35.00°N, 143.00°W), P02 (40.00°N, 138.00°W), and P01 (45.00°N, 131.00°W). For the Pacific Ocean, we exclude sites located too far from the west coast of America (limit of 160°W). For the Atlantic Ocean, we keep all stations between 20° and 70°N. For both oceans, we include the coastal sites (KEY, CSJ, and CMO) and we do not use sites located in the interior of the American continent (UTA, NWR, and ITN).

The carbon balance of North America has received much attention. We show here that this region experienced a large change in carbon fluxes of 2 GtC in the early 1990s, a change similar in magnitude to the mean annual uptake inferred by Fan et al.(7) over the period 1988–1992. Although the mean carbon balance of North America appears to be poorly constrained in our eight inversions (29), the large flux anomaly that we infer appears fairly robust. This anomalous increase in carbon storage over North America was of short duration; it was followed by a carbon loss of the same magnitude during 1994–1995 (Fig. 4A). A second large negative growth rate anomaly occurred between 1995 and 1997 (see Fig. 4A) that we attribute to an anomalous land uptake of 0.9 GtC in the Northern Hemisphere, both over North America (increase of 0.6 GtC) and over Eurasia (increase of 0.3 GtC).

Both 1992 and 1993 were colder years than average over the entire Northern Hemisphere, but North America experienced air temperatures 0.5°C colder than Eurasia in early 1993 (30). We find that North American flux anomalies strongly correlate with temperature anomalies over the early 1990s (r = 0.8). A decrease of ecosystem respiration larger than a decrease of photosynthesis could have enhanced carbon uptake over North America during 1992–1993. The global biogeochemical model (23), which captured tropical flux anomalies well, fails to reproduce the inferred changes over Northern Hemisphere land ecosystems.

It is important to determine which regions and processes are responsible for interannual changes in the carbon balance of oceans and continents for two reasons. First, the impact of climate variability on the carbon sources and sinks can provide a unique assessment of process models used to predict CO2 levels in response to emission and climate change scenarios. Second, to detect where CO2is being absorbed over land and oceans over the long term, it is necessary to separate interannual variations from long-term mean fluxes. Only when more atmospheric CO2 measurements become available will it be possible to better understand carbon flux anomalies, especially on land. One of the challenging issues then will be to link this top-down information to local measurements related to the interannual changes in the carbon balance of land ecosystems, such as climate anomalies, land use, and flux and tree-ring data.

  • * To whom correspondence should be addressed. E-mail: pbousquet{at}

  • Present address: Max-Planck-Institut für Biogeochemie, D-07701 Jena, Germany.


View Abstract

Stay Connected to Science

Navigate This Article