The effects of nitrogen (N) constraints on future terrestrial carbon (C) dynamics are investigated using the O-CN land surface model. The model's responses to elevated [CO2] and soil warming agree well with observations made in ecosystem manipulation studies. N dynamics reduce terrestrial C storage due to CO2 fertilization over the period 1860–2100 by ∼50% (342 Pg C) mainly in mid-high latitude ecosystems, compared to a simulation not accounting for N dynamics. Conversely, N dynamics reduce projected losses of land C due to increasing temperature by 16% (49 Pg C); however, this effect is prevalent only in mid-high latitude ecosystems. Despite synergistic interactions, the balance of these opposing effects is a significant reduction in future net land C storage. Terrestrial N dynamics thereby consistently increase atmospheric [CO2] in the year 2100 with a median value of 48 (41–55) ppmv, corresponding to an additional radiative forcing of 0.29 (0.28–0.34) W m−2.
 While there is good evidence that terrestrial ecosystem productivity and C storage are responding positively to increasing atmospheric [CO2] [Norby et al., 2005], it is expected that N feedbacks will strongly constrain such behavior in the future [Hungate et al., 2003; Luo et al., 2004]. Conversely, increased soil organic matter decomposition resulting from higher soil temperatures could, while reducing soil C through increased respiration, increase net N mineralization and thereby enhance future plant productivity and C storage [Melillo et al., 2002]. Predicting the relative importance of these responses at the global scale is important as the outcome will strongly impact future terrestrial C storage; and hence climate. Two recent global modeling studies have suggested that N dynamics significantly alter the future behavior of the terrestrial C cycle, reversing the terrestrial carbon-climate interaction from strongly positive to weakly negative, with important consequences for future climate [Sokolov et al., 2008; Thornton et al., 2009].
 We first use observations from ecosystem manipulation studies to evaluate the sensitivity of C dynamics to N constraints as simulated by the O-CN land-surface model [Zaehle and Friend, 2010], a component of the Institut Pierre Simon Laplace's (IPSL) climate model [Marti et al., 2005]. We then use this model to assess the significance of terrestrial N dynamics on the interactions between the global C cycle and climate. O-CN has been developed from ORCHIDEE [Krinner et al., 2005], and combines a comprehensive treatment of C, water, and energy fluxes with current understanding of N effects on plant and soil processes, including a mechanistic simulation of N losses to leaching and trace gas emissions. Zaehle et al.  found that O-CN simulates historical C and N fluxes and stocks that are consistent with current understanding, in particular those patterns associated with seasonal and interannual variability in net land-atmosphere C exchanges. We present results from two sets of simulations, one explicitly accounting for N dynamics (“O-CN”), and a second using the same model but with N dynamics switched off (“O-C”) to isolate the marginal effects of N dynamics. Supplementary material is available describing details of the modeling procedure.
3.1. Model Sensitivity
 The realism of the sensitivity of O-CN to anticipated changes in atmospheric [CO2] was assessed by comparing site-specific simulations with observations from the Duke free air CO2 enrichment experiment [Oren et al., 2001]. The model agrees well with both observed productivity in the control plots and the observed enhancement of net primary production (NPP; Figure 1a) due to increased [CO2] [Norby et al., 2005]. The observed increase of N stress with time, exhibited by a strong sensitivity of the NPP CO2 response to N addition, falling foliar N content, and increasing belowground allocation [Palmroth et al., 2006] is also reproduced by the model (see Figure S1).
 Model responses to soil warming were tested by comparing predictions with the results of the Harvard Forest soil warming experiment [Melillo et al., 2002]. Simulated enhancement of soil respiration and cumulative mineralization of C (950 g C m−2) and N (40 g N m−2) over 10 years of soil warming agree well with the observations. Previous work has shown that O-CN's predictions of the fate of increases in mineral N supply, and the response of net C storage to chronic increases in N inputs, are in close agreement with monitoring and manipulative studies [Zaehle and Friend, 2010]. O-CN reproduces accurately the observed fraction of mineralized N transferred to woody biomass, and the observed stimulation of vegetation growth. The latter compensates for the enhanced soil organic matter decomposition [Magill et al., 2004] (Figures 1b and S2). This important response cannot be captured unless terrestrial N dynamics and their effects are considered.
 Despite these encouraging model-data comparisons we acknowledge that the set of available ecosystem manipulation experiments is limited with respect to spatial and ecosystem coverage as well as duration. We therefore apply our model to the global scale with important caveats, while nevertheless noting that if these warming and CO2 experiments are representative of larger scale responses, then so is our model. A global sensitivity experiment assessing the short-term response of O-CN to increased atmospheric [CO2] and soil warming suggests that N dynamics play a key role in boreal, and a strong role in temperate ecosystems, whereas tropical ecosystems show a rather low sensitivity to N dynamics (see supplementary material). These latitudinal differences in the role of N dynamics for C exchange are entirely consistent with ecological understanding of the spatial distribution of nutrient controls on primary production [Vitousek and Howarth, 1991].
 Predictions by both O-CN and O-C are equally compatible with current estimates of global terrestrial C stocks and fluxes (Table S1), with the terrestrial biosphere predicted to have accumulated 83 Pg C (O-CN) or 108 Pg C (O-C) between 1860 and 2000. This difference in C sequestration would have led to a difference in atmospheric [CO2] of 6 ppmv, based on a three-box global C cycle model incorporating the responses of the IPSL coupled carbon-climate model [Friedlingstein et al., 2006]. These results imply land sink fractions of total anthropogenic C emissions of 0.26 (O-C) and 0.20 (O-CN) for the period 1960–2000, both compatible with current understanding of the dynamics of the contemporary global C cycle (0.20–0.27) [Raupach et al., 2008].
 N dynamics, expressed as O-CN minus O-C, cause projected additional land C accumulation between 1860 and 2100 to fall by 37%, from 388 Pg C to 244 Pg C (Figure 2). The simulations using CO2 forcing alone show that N availability strongly limits terrestrial C accumulation in response to rising CO2 concentrations (so-called “CO2 fertilization”) in northern boreal and temperate ecosystems, but has little effect within the tropics (Figure S4). Global terrestrial C accumulation by 2100 is 50% (342 Pg C) lower than when not explicitly accounting for N dynamics (Figure 2), resulting in a reduction of the land CO2 sensitivity parameter βl from 1.27 to 0.63 Pg C ppmv−1 (Figure 3a). The simulations with only climate change forcing show that the stimulation of plant growth due to warming induced soil net N mineralization reduces temperate and boreal C losses in response to climate change by 16% (49 Pg C) (Figures 2 and S4). However, this reduction is not sufficient to compensate for the heat-stress and drought related C losses from tropical ecosystems. These opposing effects of temperature increases in different regions due to N dynamics result in a small global reduction in the magnitude of the land temperature sensitivity (γl) from −61 to −51 Pg C K−1 (Figure 3b). Increasing N deposition, although important in regions of high deposition rates, has only minor impacts at the global scale (Figure 2), leading to a comparatively small increase in net land C uptake (27 Pg C), mostly in the northern extra-tropics, in agreement with previous findings [Townsend et al., 1996; Nadelhoffer et al., 1999].
 The simulated effect of N dynamics on net land C storage in temperate and boreal ecosystems is consistent with overwhelming evidence that N availability strongly controls terrestrial C cycling in temperate and boreal ecosystems [Vitousek and Howarth, 1991], limiting their response to elevated [CO2] [Hyvonen et al., 2007]. N dynamics increase simulated C storage in boreal ecosystems due to warming induced net N mineralization as suggested by Jarvis and Linder ; however, N availability also limits net C gains in Arctic regions expected from increased growing season length and air temperature, as observed by Shaver et al. . O-CN results qualitatively agree with those of Sokolov et al.  and Thornton et al.  in that the strength of CO2 fertilization is reduced and that warming-induced N mineralization increases vegetation growth, but the three studies differ in the relative importance of the two effects as the strength of the N constraint varies geographically.
 Available evidence suggests that the productivity of tropical ecosystems is limited more by the availability of phosphorus (P) than of N [Davidson et al., 2004; Cleveland and Townsend, 2006]; the lower importance of N dynamics for tropical C exchange predicted by O-CN appears therefore reasonable. However, this finding is counter to the results of Sokolov et al. , who show a strong attenuation of the tropical C cycle response due to N dynamics, and Thorton et al. , who found a large increase in tropical C storage in response to warming due to reduced N limitation. Our simulated increase in productivity and widening of foliar C:N ratios in response to elevated [CO2] have been observed in tropical rain forests [Lloyd and Farquhar, 2008], suggesting that elevated [CO2] may affect plant growth, although the evidence is inconclusive. Observational evidence suggests a negative effect of temperature increases on vegetation growth in tropical forests [Clark, 2004], exacerbated by increasing drought stress related to positive phases of ENSO [Phillips et al., 2009]. This evidence supports the trend towards reduced tropical biomass under moderate warming predicted by both O-C and O-CN.
 What do our results imply for future atmospheric [CO2] and climate? N dynamics induce a major constraint on the response of land C uptake to atmospheric CO2, coupled with a small increase in land carbon uptake due to increased temperature. This N constraint on CO2 fertilization would lead to an additional 71–99 ppmv of CO2 remaining in the atmosphere by 2100 (Figure 3c), depending on the assumed ocean sensitivity to atmospheric [CO2] [Friedlingstein et al., 2006]. The combined effect of N dynamics on both land CO2 (βl) and temperature sensitivities (γl), but without allowing for any synergistic interactions between them, results in an additional increase in atmospheric CO2 of 36–111 ppmv (relative to 28–85 ppmv for O-C), depending on the assumed ocean carbon and atmosphere sensitivity [Friedlingstein et al., 2006]. This increase is due to the effect of reduced CO2 “fertilization” outweighing the reduction in carbon loss from climate change. Taking account of the complete set of forcings (i.e., atmospheric [CO2], climate, and N deposition) and their synergistic interactions (Figure 2b) [see also Zaehle et al., 2010], the results of O-CN imply that atmospheric [CO2] will be in the range 834–1014 (median: 916) ppmv by the year 2100, whereas the C cycle-only model (O-C) implies an atmospheric concentration of 793–960 (median: 868) ppmv. Terrestrial N dynamics are therefore predicted to increase atmospheric [CO2] in year 2100 by 41–54 (median: +48) ppmv, with the range of values reflecting the different transient linear climate and ocean-atmosphere C flux sensitivities across the C4MIP coupled carbon-cycle climate models [Friedlingstein et al., 2006]. This additional amount of CO2 in the atmosphere due to N feedbacks would cause an additional radiative forcing of 0.28–0.34 (median: 0.29) Wm−2, corresponding to 0.18–0.37 (median: 0.31) K of additional mean global surface warming. While acknowledging that other important factors in the global C cycle, such as land use or P dynamics, would likely impact on the response of the terrestrial biosphere to N dynamics, the study presented here represents a key step towards realistically accounting for, and so better understanding, the effects of N dynamics on C cycle-climate feedbacks.
 These results demonstrate that modeling studies not accounting for terrestrial N dynamics [Friedlingstein et al., 2006] have likely substantially overestimated future C storage in terrestrial ecosystems in response to elevated atmospheric CO2, and thus underestimated C accumulation in the atmosphere. Contrary to previous studies [Sokolov et al., 2008; Thornton et al., 2009], we find that N dynamics have little effect on tropical C dynamics and that the stimulation of plant growth in N-limited ecosystems due to future warming of soils is only prevalent in mid- to high-latitude ecosystems. This effect is too small to preclude the existence of a positive feedback between the global C cycle and the climate system by the end of this century. Our model simulations reported here suggest that N dynamics consistently accelerate C accumulation in the atmosphere and thereby imply greater rates of climate change than estimated previously.
 We thank I. C. Prentice for useful discussions. This work was supported by the Marie Curie RTN GREENCYCLES (MRTN-CT-2004-512464) and ERG JULIA (PERG02-GA-2007-224775). Computing time was provided by the Centre d'Energie Atomique, Saclay, France.