Enhanced Seasonal Exchange of CO2 by Northern Ecosystems Since 1960

See allHide authors and affiliations

Science  06 Sep 2013:
Vol. 341, Issue 6150, pp. 1085-1089
DOI: 10.1126/science.1239207

Downs and Ups

Every spring, the concentration of CO2 in the atmosphere of the Northern Hemisphere decreases as terrestrial vegetation grows, and every fall, CO2 rises as vegetation dies and rots. Climate change has destabilized the seasonal cycle of atmospheric CO2 such that Graven et al. (p. 1085, published online 8 August; see the Perspective by Fung) have found that the amplitude of the seasonal cycle has exceeded 50% at some latitudes. The only way to explain this increase is if extratropical land ecosystems are growing and shrinking more than they did half a century ago, as a result of changes in the structure and composition of northern ecosystems.


Seasonal variations of atmospheric carbon dioxide (CO2) in the Northern Hemisphere have increased since the 1950s, but sparse observations have prevented a clear assessment of the patterns of long-term change and the underlying mechanisms. We compare recent aircraft-based observations of CO2 above the North Pacific and Arctic Oceans to earlier data from 1958 to 1961 and find that the seasonal amplitude at altitudes of 3 to 6 km increased by 50% for 45° to 90°N but by less than 25% for 10° to 45°N. An increase of 30 to 60% in the seasonal exchange of CO2 by northern extratropical land ecosystems, focused on boreal forests, is implicated, substantially more than simulated by current land ecosystem models. The observations appear to signal large ecological changes in northern forests and a major shift in the global carbon cycle.

Observations of atmospheric CO2 concentration from the longest-running Northern Hemisphere records at Mauna Loa, Hawaii (MLO, 20°N) and Barrow, Alaska (BRW, 71°N) show increasing trends in seasonal amplitude (Fig. 1) (14). Shorter records from other ground-based sites in the Northern Hemisphere also show increasing trends in amplitude, with the largest changes at high latitudes (5), but the long-term change and especially its spatial pattern have not been well characterized by existing data sets because only MLO and BRW have records extending back to the 1950s and 1960s (69). The two sites show similar trends in the timing of CO2 drawdown, advancing by roughly 1 week, but differing trends in amplitude: The amplitude at MLO increased by 15 ± 5% over the past 50 years, while the amplitude at BRW increased by at least twice as much (Fig. 1). The best estimate for BRW is a 35% increase in amplitude, with a range of 30 to 49%, including uncertainty from changes in measurement site location, sampling inlet height, and local vegetation effects (10, 11).

Fig. 1 Long-term trends in seasonal CO2 cycles at Barrow and Mauna Loa.

Observed peak-to-trough seasonal amplitude (A) and phase (B), given by the day of year of downward zero crossing, of CO2 concentration at Barrow (71°N, blue) and Mauna Loa (20°N, black) measured by the Scripps CO2 Program (7, 8) and the NOAA Global Monitoring Division (9). Growth rate of amplitude is given in percentage change per year, with 1 SD uncertainty of ±0.05 to 0.07% year–1. Seasonal CO2 cycles observed at Barrow (C) and Mauna Loa (D) for the 1961 to 1963 or 1958 to 1963 and 2009 to 2011 time periods. The first 6 months of the year are repeated.

Here, we present seasonal CO2 cycles observed by the recent High-Performance Instrumented Airborne Platform for Environmental Research (HIAPER) Pole-to-Pole Observations (HIPPO) large-scale aircraft campaign (12) and regular aircraft profiles over several fixed sites by the National Oceanic and Atmospheric Administration (NOAA) Carbon Cycle Group Aircraft Program from 2009 to 2011 and compare them to observations made 50 years earlier during the International Geophysical Year (IGY) (13). The IGY observations were obtained by 160 weather reconnaissance flights above the North Pacific and Arctic Oceans from 1958 to 1961 along 500 and 700 millibar (mb) pressure surfaces (~6 and 3 km altitude). To assess changes in seasonal CO2 amplitude over time, we examine HIPPO and NOAA data over the same regions and pressure surfaces as the IGY data (14). The comparison reveals a strikingly large (~50%) amplitude increase north of 45°N, indicating that major ecological changes must have occurred in northern land ecosystems over the past 50 years.

We calculate an amplitude gain factor for the 1958 to 1961 and 2009 to 2011 observations in 10° latitude bands at 500 and 700 mb after detrending the data by subtracting the observed deseasonalized trend at MLO and after specifying the shape of the seasonal cycle in each pressure and latitude band using fits to the NOAA aircraft data, which have higher temporal resolution than the HIPPO and IGY data (14). This method does not allow for the detection of changes in phase, which in any case are not clearly resolvable in the data (14), but shifts of up to 7 days as observed at MLO and BRW do not significantly affect the calculation of amplitude (1).

An increase in amplitude at least as large as that detected at Barrow is found over a wide range of pressures and at all latitudes north of 45°N (Fig. 2). The change in seasonal CO2 amplitude north of 45°N is about +5 parts per million (ppm) at all pressures, representing a 57 ± 7% increase at 500 mb since 1958 to 1961 (Figs. 2 and 3) (~0.9 ± 0.1% year−1 over 50 years). The seasonal CO2 amplitude increase is smaller between 35° and 45°N at 500 mb, 26 ± 18%, whereas there is no distinguishable amplitude change south of 35°N (Fig. 2). The same pattern appears at 700 mb, although the IGY data are limited to 10° to 30°N and 65° to 90°N at 700 mb. Low-latitude amplitude changes in the aircraft data are consistent, within their uncertainties, with the surface measurements from Mauna Loa at ~670 mb.

Fig. 2 Seasonal CO2 cycles observed by aircraft in 1958 to 1961 and 2009 to 2011.

Seasonal cycles of CO2 at 500 mb in the upper row (A) and 700 mb in the lower row (B) for each latitude band with aircraft data in both periods. Curves show the seasonal cycles fit to the data, with uncertainties indicated by shaded areas. Amplitudes of the seasonal cycle of CO2 for all locations with data available in both periods (C). The IGY aircraft data cover the period 1958 to 1961; the early BRW and MLO data cover 1961 to 1963 and 1958 to 1963. Uncertainties (55) are shown by the shaded areas for 500 mb data and by error bars for 700 mb and BRW in 2009 to 2011. Uncertainties for BRW in 1961 to 1963 and MLO in both periods are smaller than the symbol sizes.

Fig. 3 Regional attribution of the seasonal CO2 amplitude and its change over 50 years.

Contributions of different terrestrial regions to the seasonal cycle of CO2 at ground-based stations and at 500 mb for 20°N (A) and for 70°N (B) for 2009 to 2011, averaged over two atmospheric transport models, ACTM (20) and TM3 (21). Error bars reflect absolute differences between the two models. (C) Locations of the Mauna Loa and Barrow stations and the 155°W meridian, as well as the terrestrial regions considered. Arctic (red); boreal (orange); temperate (yellow); subtropics, tropics, and Southern Hemisphere land (shades of blue). (D to F) Change in CO2 amplitude, in percentages, in the observations and in simulations where the amplitude in NEP was optimally scaled in different regions, plotted by latitude along 155°W for different pressures: (D) 500 mb aircraft data, (E) 700 mb aircraft data and Mauna Loa (670 mb), and (F) Barrow at sea level.

Representativeness of the aircraft data is shown by the similarity between HIPPO and NOAA data for 2009 to 2011 (15), and between the IGY and independent data from 1963 to 1968 over the North Atlantic Ocean (16). The large changes in amplitude between 1958 to 1963 and 2009 to 2011 north of 45°N (35 to 66%) substantially exceed interannual variation at Barrow (±4%) (Fig. 1) and in high-latitude NOAA aircraft data for recent years (±15%) (15), indicating that the observations represent a long-term change rather than a short-term fluctuation.

Fossil fuel, wildfire, and ocean fluxes of CO2 contribute only a few percent to the seasonal CO2 cycle in the Northern Hemisphere (5, 17, 18), limiting their impact on trends in CO2 amplitude to less than a few percent (19). Because midtropospheric amplitude changes north of 45°N are at least as large as those at Barrow, the observations cannot be explained solely by trends in the intensity of vertical mixing (19), although changes in transport may cause secondary effects. Model simulations also indicate that potential biases caused by uneven sampling within each latitude band are much smaller than the uncertainties in measured CO2 amplitude change (19).

The large increase in CO2 amplitude north of 45°N must therefore be attributed almost entirely to the terrestrial biosphere. We now focus on assessing the magnitude and driving factors for trends in the seasonal exchange of CO2 from net ecosystem production (NEP) by calculating optimal estimates of NEP change that best match the atmospheric CO2 amplitude change and then exploring potential mechanisms with historical simulations of terrestrial ecosystem models.

We begin with spatially resolved monthly estimates of NEP averaged over 2005 to 2010 that have been optimized using terrestrial ecosystem model simulations and current observations from the global CO2 network in two atmospheric transport models, Atmospheric Chemistry Transport Model (ACTM) (20) and TM3 (21). We simulate separate tracers in ACTM and TM3 representing large regions (Fig. 3C) to assess their contribution to the seasonal CO2 cycle at different tropospheric locations (Fig. 3, A and B), then we adjust monthly NEP for the large regions by a consistent fraction—that is, by changing the amplitude but not the phase in order to best match the observed CO2 amplitude changes, roughly centered on 155°W (Fig. 3, D to F) (22).

Exploiting our observation that CO2 amplitude changes are much larger north of 45°N, we optimize changes in NEP in boreal and Arctic regions (Case 1 in Fig. 3, D to F), the regions with stronger influence on CO2 cycles at high latitudes compared with low latitudes (5) (Fig. 3, A and B). Due to isentropic transport, the seasonal cycle at 70°N and 500 mb is less sensitive to the Arctic than to regions farther south (Fig. 3B), showing that changes cannot only be occurring in the Arctic. The optimal uniform change in NEP amplitude in boreal and Arctic regions, fitting to CO2 amplitude changes north of 45°N only, is an 85% [73 to 107% confidence interval (CI)] (23) increase. Extending NEP changes to include temperate regions as well as Arctic and boreal regions results in an optimal increase of 53% (46 to 65% CI), assuming uniform change over these three regions (Case 2 in Fig. 3, D to F), or optimal zone-specific increases of 23% (12 to 42% CI) in the Arctic, 67% (57 to 87% CI) in the boreal, and 67% (54 to 89% CI) in the temperate region (Case 3 in Fig. 3, D to F).

Allowing for changes in the temperate region in Cases 2 and 3 improves the fit to high-latitude data compared to Case 1 while significantly decreasing the change attributed to Arctic and boreal regions. The simulations tend to produce fractional changes in amplitude that are more uniform with altitude than the observations, which show fractional changes in amplitude increase with height at high latitudes, and they tend to overestimate the CO2 amplitude change at lower latitudes, which were not used in optimization. This may partly result from transport errors, caused by inaccuracies in modeled transport or unresolved transport changes since 1960, or from representation errors, caused by low resolution in the atmospheric transport models or in the terrestrial regions considered. Alternatively, the high-latitude effects on low-latitude CO2 amplitudes may have been counteracted by decreasing NEP amplitudes in subtropical and tropical land regions. Fitting to all data and allowing for changes in subtropical and tropical NEP results in a change of –15% (–31 to 9% CI) in the tropics and subtropics (Case 4 in Fig. 3, D to F) without significantly affecting the optimal zone-specific NEP increases in the Arctic and boreal regions, though the NEP increase in the temperate region is reduced from Case 3.

The observations therefore require a widespread increase in the seasonal exchange of carbon in northern regions. Our analysis suggests that an additional 1.3 × 1015 to 2.0 × 1015 g carbon (Pg C) is exchanged with the atmosphere seasonally by Arctic, boreal, and temperate ecosystems combined in 2009 to 2011 compared with 1958 to 1963, an increase of 32 to 59%. This seasonal exchange represents the accumulation of carbon on land during the growing season (typically May to August) and the release of carbon during the dormant season, after subtracting annual mean exchange (17). Our analysis does not distinguish between enhanced growing season uptake and enhanced dormant season release, but changes in these two opposing fluxes may not necessarily be in balance. We do find, counter to previous studies invoking dormant or early growing season trends alone (1, 24), that strong increases in CO2 amplitude are unlikely to occur without enhanced uptake in the main growing season. This is because the intense fluxes in the main growing season (June and July) make a much larger contribution to the seasonal CO2 amplitude than either the dormant season (September to April) or the shoulder seasons (May and August), as shown by calculations that examine contributions of the different seasons to the CO2 cycle (25).

Enhancement in seasonal NEP is likely to be greatest in boreal regions, where all four cases require that the NEP amplitude changed by at least 53%. Less definitive is the requirement for decreased NEP amplitude in subtropical and tropical regions, although this may be consistent with the conversion of highly productive forests to crop or pasture lands and increased drought (2630).

To explore potential mechanisms for changing NEP, we examine historical simulations of a subset of terrestrial ecosystem models currently participating in the fifth phase of the Coupled Model Intercomparison Project (CMIP5) (31). These models include simulations of CO2 fertilization, climate change, and land use change, as well as nonlinear interactions between these processes; some additionally include dynamic vegetation change, coupled nitrogen cycling, or simple crop modeling (32). We use the ACTM and TM3 atmospheric transport models to estimate atmospheric CO2 amplitudes resulting from the CMIP5 models’ simulated monthly NEP (Fig. 4).

Fig. 4 Comparison of observations with CMIP5 model simulations.

Change in amplitude of the seasonal cycle of CO2 between 1958 to 1961 and 2009 to 2011 versus amplitude of the seasonal cycle for 2009 to 2011 at 500 mb, averaged over 45° to 90°N, in observations and in simulations of the CMIP5 land models (56) using the two atmospheric transport models, ACTM and TM3. Diamonds indicate models with dynamic vegetation. Error bars in the simulated amplitude and change in amplitude show differences between ACTM and TM3. The simulations include changes in fossil fuel emissions and atmospheric transport between 1958 to 1961 and 2009 to 2011. Oceanic fluxes were given by the two models’ atmospheric CO2 inversions for the late 2000s and held constant for the 1958 to 1961 and 2009 to 2011 periods.

None of the CMIP5 models can account for the increase in CO2 amplitude north of 45°N at 500 mb between 1958 to 1961 and 2009 to 2011 (Fig. 4). Moreover, the recent CO2 amplitude varies strongly between models, with a range of 6 to 19 ppm compared with the observed value of 13.8 ± 0.7 ppm. Previous studies have similarly shown that most terrestrial ecosystem models underestimate the long-term trend in CO2 amplitude at Mauna Loa (33) and include large errors in the seasonal amplitude of NEP at eddy flux sites (34).

Simulated CO2 amplitude changes from the CMIP5 models are nearly proportional to their NEP amplitude changes in northern ecosystems (32), as found in our optimization (Fig. 3), implying that the models underestimate the 50-year changes in NEP. This underestimate is not surprising, considering that the models’ parameterizations of the direct effects of rising CO2 and temperature are informed by laboratory and field studies that show low response to manipulations of CO2 and temperature. Observed enhancement factors (3538) suggest that the increase in net primary productivity (NPP) from the observed 50-year increases in CO2 (23%) (8) and northern temperature (1.0°C north of 30°N) (39) should be no larger than 20 to 30%. Correspondingly, simulated increases in seasonal NPP amplitude are in the range 6 to 22% in the CMIP5 models, accompanied by higher amplitudes in heterotrophic respiration in most cases (32). Independent of the models, an explanation for the NEP changes based solely on a direct physiological response to warming is inconsistent with two features of the Barrow record (Fig. 1): (i) the weak interannual variability but strong long-term trend, and (ii) the increase in amplitude from the 1960s to the mid-1970s, a period when northern land temperatures did not rise but cooled slightly (39, 40).

We are led to conclude that ecological changes in boreal and temperate forests are driving additional increases in the summertime uptake of carbon. This inference from atmospheric data is qualitatively consistent with expanding evidence for substantial changes occurring in these ecosystems. Forest inventories show increased stand area and biomass (26, 28). Other ground-based studies show that evergreen shrubs and trees are migrating northward in response to warming (4143), and that fire, logging, and other disturbances (44, 45) are shifting the age composition toward younger, early-successional forests that experience shorter, more intense periods of seasonal carbon uptake (46, 47). Satellite observations generally show trends toward increased greenness in northern ecosystems (4), although many areas of the boreal forest show browning trends in recent decades (48, 49). The atmospheric evidence helps to quantify the aggregate effect of these, and other, types of ecological changes over the past 50 years. The changes are evidently underrepresented in the CMIP5 models, including those that allow for dynamic vegetation (Fig. 4) and coupled nitrogen cycling (Community Climate System Model 4 and Norwegian Earth System Model 1-M) (50).

The increase in seasonal CO2 amplitude may be closely related to the net sink of CO2 in northern extratropical land ecosystems that is currently ~2 Pg C year−1 (28, 5153). The relationship depends on how the sink has changed over time, which is not well constrained observationally but is likely to have increased since 1960 (54). Assuming that the sink was ~1 Pg C year−1 in 1960, the observed 1.3 to 2.0 Pg C year−1 increase in seasonal exchange over 50 years implies that growing season uptake increased by 40 to 60%, whereas dormant season release increased by only 20 to 50%, a result similar to the model scenarios described by Randerson et al. (5). A northern sink smaller than ~1 Pg C year−1 in 1960 would imply an even larger imbalance between growing season and dormant season trends.

There is no evidence that northern ecosystem trends are slowing down, because the largest CO2 amplitudes at Barrow have been observed in the past few years (Fig. 1). The inability of the CMIP5 models to account for the observed increase in the amplitude of atmospheric CO2 indicates that they underestimate the widespread ecological changes that occurred over the past 50 years and are likely to underpredict future changes.

Supplementary Materials

Materials and Methods

Figs. S1 to S9

Tables S1 to S7

References (5781)

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. See section SM1 in the supplementary materials.
  3. See section SM2 in the supplementary materials.
  4. See section SM3 in the supplementary materials.
  5. See section SM5 in the supplementary materials.
  6. See section SM4 in the supplementary materials.
  7. See sections SM6 and SM8 in the supplementary materials.
  8. We specify the uncertainty in NEP amplitude change with 95% confidence intervals, assessed by varying parameters in the optimization and including jackknife estimates of the data.
  9. See section SM7 in the supplementary materials.
  10. See section SM9 in the supplementary materials.
  11. We specify the uncertainty in CO2 amplitude with 95% confidence intervals estimated using a jackknife procedure (fig. S3).
  12. Only two models’ simulations extended to 2011 (HadGEM2-ES and NorESM-1). Other models’ output was available only through 2005, so NEP for 2009 to 2011 is given by the mean over 2001 to 2005. In these models, the change in NEP amplitude up to 2001 to 2005 may be 10% smaller than the change up to 2009 to 2011.
  13. Acknowledgments: The Scripps CO2 Program is supported by DOE grant DE-SC0005090. HIPPO was supported by NSF grants ATM-0628575, ATM-0628519, ATM-0628388, ATM-0628452, and ATM-1036399, and by the National Center for Atmospheric Research (NCAR). Recovery and updating of early aircraft, MLO, and BRW data was supported by NSF grant ATM-1036399. L.R.W. was supported by NASA award NNX11AF36G. Early observations at BRW were funded by U.S. Navy/Office of Naval Research contract N00014-67-A-0103-0007. Online access to all observational data is summarized in section SM10 of the supplementary materials. NCAR is supported by the NSF. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of NOAA, NSF, DOE or NASA. We thank the HIPPO science team and the crew and support staff at the NCAR Research Aviation Facility. We acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups for producing and making available their model output. Support of the CMIP data sets is provided by the Office of Science, U.S. Department of Energy. C. Roedenbeck provided assistance with the TM3 model. P.K.P. is partially supported by the Ministry of Education, Culture, Sports, Science and Technology Green Network of Excellence program. G.W.S. acknowledges support from the NSF Graduate Research Fellowship Program and the Environmental Protection Agency’s Science to Achieve Results program.
View Abstract

Navigate This Article