Enhanced seasonal CO2 exchange caused by amplified plant productivity in northern ecosystems

See allHide authors and affiliations

Science  21 Jan 2016:
DOI: 10.1126/science.aac4971


Atmospheric monitoring of high northern latitudes (> 40°N) has shown an enhanced seasonal cycle of carbon dioxide (CO2) since the 1960s but the underlying mechanisms are not yet fully understood. The much stronger increase in high latitudes compared to low ones suggests that northern ecosystems are experiencing large changes in vegetation and carbon cycle dynamics. Here we show that the latitudinal gradient of the increasing CO2 amplitude is mainly driven by positive trends in photosynthetic carbon uptake caused by recent climate change and mediated by changing vegetation cover in northern ecosystems. Our results emphasize the importance of climate-vegetation-carbon cycle feedbacks at high latitudes, and indicate that during the last decades photosynthetic carbon uptake has reacted much more strongly to warming than carbon release processes.

The seasonal cycle of atmospheric carbon dioxide (CO2) in the northern hemisphere is mainly controlled by carbon uptake and release processes of the land biosphere (1), specifically by the difference between photosynthetic carbon uptake (gross primary production, GPP) and ecosystem respiration (Reco) of the land biosphere (2). Airborne and surface data show that the amplitude of the seasonal cycle (hereinafter called “CO2 amplitude”) has increased since 1960, particularly north of 45°N, where increases as large as 50% have occurred. (14). The strong seasonality of GPP and Reco in northern land ecosystems causes a larger average CO2 amplitude in northern high latitudes than in low ones (2, 5). The larger trends in CO2 amplitude at higher latitudes consequently imply that they are caused by changing carbon cycle dynamics in northern ecosystems (4).

A variety of factors may contribute to the CO2 amplitude trend. Arctic and boreal regions have experienced strong warming in the last decades (6) and a “greening” trend has been detected from satellites indicating enhanced plant growth (7, 8) (Fig. 1, A and B). These satellite observations are confirmed by ground observations showing increases in shrub coverage in the tundra (9), tree growth along the tundra-boreal forest transition zone (10), and deciduous tree cover from recovery after severe boreal forest fires (11). Additionally, various estimates show positive trends in both annual amplitudes and annual totals of GPP (12, 13) and in net biome productivity (NBP) (14) in northern ecosystems (Fig. 1, C and D).The intensification of agriculture in mid-latitudes also likely contributes to the CO2 amplitude trends (15, 16). These multiple observational signals point to amplified plant productivity as a likely cause of the increase in CO2 amplitude (1, 3, 7, 17). Nevertheless, a quantitative explanation of the amplitude trends is still lacking. Current earth-system models consistently underestimate the CO2 amplitude trend (4) and its gradient with latitude, suggesting they are missing or under-representing key processes (18).

Fig. 1 Amplification of plant activity in the northern biosphere.

Annual time series and linear trends of (A) mean annual air temperature, (B) peak FAPAR (fraction of absorbed photosynthetic active radiation), (C) annual amplitude of GPP (gross primary production), (D) annual amplitude of NBP (net biome productivity), and (E) seasonal amplitude of atmospheric CO2 at Barrow (BRW) and Mauna Loa (MLO). Time series in (A-B) were spatially averaged and in (C, D) aggregated for boreal and arctic land regions north of 41°N (fig. S12), and the 1982–2011 mean has been subtracted. Dashed lines indicate the trends for the overlapping period of LPJmL simulations and observations. P-values were calculated with the Mann-Kendall trend test.

Here we examine the cause of the CO2 amplitude increase by combining observations from long-term monitoring sites of atmospheric CO2 concentration, satellite observation of vegetation greenness (19), and global observation-based datasets of GPP (12) and NBP (14), with results from the LPJmL dynamic global vegetation model (20, 21) coupled with the TM3 atmospheric transport model (22) (hereinafter called LPJmL+TM3 (23)), to explain the observed latitudinal gradient of CO2 amplitude trends. Unlike other biosphere models that were previously evaluated against CO2 amplitude trends (4), LPJmL considers several processes that potentially contribute to a better explanation of these trends, including agriculture, irrigation and land use change (21); vegetation dynamics; and processes that control northern vegetation dynamics such as permafrost (24), and fire (25) driven by observed burned area data (26). Moreover, LPJmL uses an improved phenology module (26), which has been optimized against satellite observations of FAPAR (fraction of absorbed photosynthetic active radiation), albedo, and an observation-based data set of GPP, resulting in a better representation of climate controls on vegetation dynamics as well as global carbon fluxes and stocks (26, 27). Note that atmospheric CO2 data were not used to constrain LPJmL.

We estimated CO2 amplitude trends in observed time series at 19 monitoring sites with at least 20 years of data (table S1). We found much stronger positive CO2 amplitude trends at high latitude sites (e.g., 0.08 ppm yr−1 ≈ 0.53% yr−1 at Point Barrow, BRW, in 1971–2011) than at low latitude sites (e.g., 0.005 ppm yr−1 ≈ 0.076% yr−1 at Mauna Loa, MLO, in 1970–2011, Fig. 1E). These estimated trends were similar to those of previous studies (2, 4) with small differences due to station selection, time series analysis methods, and time series length. We found weaker trends in the CO2 amplitude, especially at MLO, because this trend originates mostly from low CO2 amplitude values in the 1960s and weakens from 1970 onwards. To account for the effect of time series length, we estimated the uncertainties in CO2 amplitude trends by computing trends for different combinations of start and end years (fig. S1). The estimated uncertainties (Embedded Image% yr−1 at BRW, Embedded Image% yr−1 at MLO, Embedded Image%-iles of trend slope ensemble) demonstrate that only high latitude sites have persistent long-term increases in CO2 amplitude.

In comparison to surface-level site observations, LPJmL+TM3 had on average stronger CO2 amplitude trends (Fig. 2A and table S2). However, LPJmL+TM3 reproduced the observed changes in CO2 amplitude at higher atmospheric levels (fig. S2). LPJmL+TM3 simulations were well correlated with site observations regarding spatial patterns of mean CO2 amplitude values (r = 0.84) and trends (r = 0.51, p ≤ 0.05) (fig. S3). LPJmL+TM3 had a modest performance in representing the year-to-year variability of the CO2 amplitude at some sites (23) which may indicate the importance of regional extreme events on the land carbon balances (28). LPJmL+TM3 reproduced the observed pattern of strong positive CO2 amplitude trends north of 45°N, the large variability of trends in mid-latitudes, and the small or non-significant trends south of 20°N. We found that simulations of CO2 amplitudes were sensitive to the choice of the meteorological forcing dataset for the TM3 transport model (fig. S4A). Therefore we propagated the uncertainty both from time series length and meteorological forcing to the overall uncertainty of simulated CO2 amplitude trends for a more robust model evaluation. The inter-annual variability of ocean CO2 uptake had no distinct contribution to CO2 amplitude trends in comparison to a climatology of ocean uptake (fig. S4B). LPJmL yields positive trends in annual maximum FAPAR in northern ecosystems which are in good agreement with satellite observations (27), and increases in annual amplitudes and totals of GPP and NBP that agree with independent observation-based GPP and NBP estimates (Fig. 1, A to E, and fig. S5). Although the model does not fully account for trends in agricultural fertilizer usage, the simulated trends in GPP of agricultural regions are comparable to an independent estimate (fig. S7). This comparison of LPJmL simulations and independent datasets demonstrate an amplification of plant productivity in northern ecosystems.

Fig. 2 Latitudinal gradients of trends in the seasonal CO2 amplitude and its drivers.

(A) Simulated and observed CO2 amplitude trends with 95% confidence intervals (dashed lines) and the uncertainty distribution of site-level trend slopes (solid lines are inter-quartile range, dots are median values). Site-level uncertainty distributions are not shown in B and C for clarity. (B) Global effects of CO2 fertilization and climate change on the latitudinal gradient. Removing the effect of CO2 fertilization on photosynthesis reduces CO2 amplitude trends globally but has no effect on the latitudinal gradient. The latitudinal gradient disappears with a constant climate. (C) Separation of the indirect effect of changing vegetation cover and the direct of climate change on photosynthesis in high-latitude regions on the latitudinal gradient. The latitudinal gradient disappears both without changes in vegetation cover (i.e., no vegetation cover change but still climate change) and without climate change (i.e., no climate change but still forcing changes in vegetated area).

We quantified the contribution of land NBP from different regions to CO2 amplitude trends (table S5 and fig. S6B). NBP from boreal regions contributed 51% to the average CO2 amplitude in the 1970s at northern high latitude sites (> 45°N). In the 2000s, the contribution of boreal regions increased to 54% corresponding to an annual increase of Embedded Image% yr−1. The contribution of arctic NBP was 17% in the 1970s and increased with Embedded Image% yr−1. NBP from global agricultural regions contributed with 11% in the 1970s and increased with Embedded Image% yr−1. Temperate and tropical regions made only minor contributions to the trends in CO2 amplitude. However, only trends in the contribution of arctic and boreal NBP were significant (p = 0.03 and p = 0.01, respectively) at northern sites. At low latitude sites (0 – 45°N), NBP from boreal regions still contributed dominantly to the increase in CO2 amplitude with Embedded Image% yr−1 (p = 0.12), followed by NBP from agricultural regions with Embedded Image% yr−1 (p = 0.37) and NBP from arctic regions with Embedded Image% yr−1 (p = 0.12). Therefore boreal regions contributed approximately with 57%, arctic regions with 25% and agricultural regions with 17% to the overall CO2 amplitude trend at northern latitude sites (41%, 14% and 20% at low latitude sites, respectively). The agricultural contribution in LPJmL is within the ranges found in previous studies (15, 16). Similar to (4), we find that fossil fuel emissions and ocean CO2 exchange have little impact on the trends in CO2 amplitude (table S5). Consistent with Graven et al. (4), LPJmL attributes a dominant role to boreal and arctic ecosystems in driving the CO2 amplitude increase.

Both GPP and Reco can potentially contribute to the increasing CO2 amplitude. We found stronger trends in annual total GPP (Embedded ImagePgC yr−2, LPJmL in 1970–2011) than in annual total Reco (Embedded ImagePgC yr−2, LPJmL in 1970–2011) in northern ecosystems (fig. S8) and stronger trends in the GPP than Reco annual amplitudes across all northern latitudes (fig. S9). The contribution of GPP to the CO2 amplitude increased by Embedded Image% yr−1 whereas the increasing Reco contributed to a decrease of the seasonal amplitude of Embedded Image% yr−1. Consequently, given the opposite signs of GPP and Reco fluxes, the effect of GPP on the CO2 amplitude is attenuated but not compensated by Reco (table S5). This is consistent with earlier results (29), that showed that the spatial variability of NEE and NBP amplitudes is strongly related to GPP. The stronger increase in GPP compared to Reco leads to a positive trend in northern ecosystem NBP of Embedded ImagePgC yr−2 (LPJmL) which is confirmed by independent estimates from the Jena CO2 Inversion scheme (Embedded ImagePgC yr−2, version s81_v3.6 (14)) (fig. S10). Trends in annual total GPP in northern ecosystems and CO2 amplitude trends at northern latitude sites show a strong linear relation across different LPJmL+TM3 model experiments (r2 = 0.96, Fig. 3 and fig. S11). For example, a 0.08 ppm yr−1 increase in CO2 amplitude at Barrow requires a 0.07 PgC yr−2 increase in boreal and arctic GPP. Thus, the increase in the seasonal CO2 amplitude can be explained by a photosynthesis-driven increase in net carbon uptake of northern ecosystems.

Fig. 3 Trend in the CO2 amplitude at Barrow against trends in northern ecosystem gross primary production across different factorial model experiments with LPJmL.

Dots and error bars represent median values and 95% confidende intervals of the estimated trends. Gray-colored horizontal lines are the median and 95% confidence interval for the estimated trend in the observed CO2 amplitude time series.

Several factors can contribute to the increased photosynthetic carbon uptake, and thus the latitudinal gradient of the increasing CO2 amplitude. Rising atmospheric CO2 and climate change directly affect physiological processes that can enhance photosynthesis (30). To test the relative effect of CO2 fertilization and climate change on CO2 amplitude trends, we performed two model experiments with LPJmL where we kept temperature and precipitation at 1965–1975 levels for the period 1970–2011 (i.e., no climate change); and held CO2 constant at 325.7 ppm after 1970 to quantify the effect of CO2 fertilization (Fig. 2B). We found that both climate change and CO2 fertilization affect CO2 amplitude trends, but with regional differences: Climate change was the dominant factor on CO2 amplitude trends north of 40°N whereas CO2 fertilization was the dominant factor south of 40°N. Without the effect of CO2 fertilization, CO2 amplitude trends were generally lower (ca. –0.04 ppm yr−1 across all latitudes) but the latitudinal gradient of stronger CO2 amplitude trends in northern compared to southern latitudes was not affected. However, the strong CO2 amplitude trends in northern latitudes disappeared under constant climate and reverted the latitudinal gradient toward stronger trends south of 40°N (Fig. 2B). Therefore, the stronger CO2 amplitude trends at northern latitudes are mainly dominated by climate change-induced increases in boreal and arctic GPP (Fig. 3). Increasing GPP results in increasing plant growth which again enhances GPP. LPJmL simulates especially an increasing coverage of trees across the boreal zone at the expense of tundra (23). To quantify the role of this indirect vegetation cover feedback on GPP and CO2 amplitude trends, we performed two more model experiments: in the first, we again fixed climate in the period 1970–2011 according to the climate conditions in 1965–1975 but prescribed changes in vegetated area as simulated in the standard experiment (i.e., no climate change but still changes in vegetation cover). In the second, we used observed climate but fixed vegetated area after 1970 (i.e., constant vegetation cover but still climate change). Both experiments were performed only for northern ecosystems while the rest of the world was simulated following the normal simulation protocol. The latitudinal gradient of stronger CO2 amplitude trends in northern latitudes disappeared both without the direct effect of climate change and without the indirect effect of changing vegetation cover in northern ecosystems (Fig. 2C). Thus, the interaction between the direct climate effects on photosynthesis and the indirect effect of changing vegetation cover drive the trend in the CO2 amplitude. The climate effect is likely mostly exerted via temperature given earlier results from eddy covariance sites, indicating that variability in ecosystem GPP north of 42°N is driven by temperature (31). Additional other processes such as increasing plant available water from enhanced seasonal thawing of permafrost soils and changes in plant phenology contribute to plant productivity in northern ecosystems. Indeed we found weaker GPP and CO2 amplitude trends in a LPJmL simulation without considering permafrost dynamics (24) (LPJmL-noPF in Fig. 3), and overestimated observed GPP and CO2 amplitude trends with a too simplistic phenology model that only accounts for temperature effects but ignores radiation and hydrological effects on the seasonal leaf development (26) (LPJmL-oldPhen in Fig. 3). These examples demonstrate a strong but complex control of climate on plant productivity in northern ecosystems which ultimately results in the major contribution of enhanced plant growth to the strong CO2 amplitude trends in northern latitudes.

Overall, our results suggest that a major driver of the large increase in CO2 amplitude at high northern latitude involves the interaction of recent climate change with vegetation dynamics. Climate change affects processes such as plant physiology, phenology, water availability, and vegetation dynamics, ultimately leading to increased plant productivity and vegetation cover in northern ecosystems in the last decades. Our results further highlighted the gradual replacement of herbaceous vegetation with forests as a major specific factor. Lastly, we identified a dominance of changes in photosynthesis over respiration in driving the changes. Sensitivities of these processes to climate need to be carefully assessed in current ecosystem and earth system models against observational data to accurately reproduce observed changes in CO2 amplitude. However the stimulation of photosynthesis and vegetation growth by climate change cannot be unlimited, because of limitations by nutrients, radiation, and possibly increased mortality (32). Thus at some point in the future the positive trends in plant productivity and thus the CO2 amplitude increase might stall. Continued long-term observation of atmospheric CO2 together with ground and satellite observations of vegetation productivity and dynamics will be the key to detect, model, and better predict such change in high-latitude carbon cycle dynamics.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S13

Tables S1 to S7

References (3369)

References and Notes

  1. See supplementary materials on Science Online.
  2. Acknowledgments: We thank the following institutions and scientists that acquired and provided CO2 site data: E. Dlugokencky (NOAA), Scripps Institution of Oceanography, D. Worthy (Environment Canada), F. Meinhardt (Umweltbundesamt), R. Langenfelds and P. Krummel (CSIRO), and H. Koide (JMA). We further thank the following groups, institutions or scientists for sharing their datasets: AFS, CRU, EC/JRC, ECMWF, GIMMS, NCEP, NRCan, L. Giglio (UMD), M. Jung (MPI-BGC), S.E. Mikaloff-Fletcher (NIWA). We thank S. Schaphoff (PIK) for comments on LPJmL. We thank I.C. Prentice and R. Thomas for comments on the manuscript. We acknowledge financial support by the European Union (FP7) through the projects GEOCARBON (283080), CARBONES (242316), and EMBRACE (283201); from US DOE (DE-SC0012167); from US NSF (1304270); and from NOVA (UID/AMB/04085/2013). Table S7 provides an overview of used data and how it can be obtained. Results from factorial model experiments with LPJmL+TM3 are available at
View Abstract

Navigate This Article