Influence of carbon-nitrogen cycle coupling on land model response to CO2 fertilization and climate variability



[1] Nutrient cycling affects carbon uptake by the terrestrial biosphere and imposes controls on carbon cycle response to variation in temperature and precipitation, but nutrient cycling is ignored in most global coupled models of the carbon cycle and climate system. We demonstrate here that the inclusion of nutrient cycle dynamics, specifically the close coupling between carbon and nitrogen cycles, in a terrestrial biogeochemistry component of a global coupled climate system model leads to fundamentally altered behavior for several of the most critical feedback mechanisms operating between the land biosphere and the global climate system. Carbon-nitrogen cycle coupling reduces the simulated global terrestrial carbon uptake response to increasing atmospheric CO2 concentration by 74%, relative to a carbon-only counterpart model. Global integrated responses of net land carbon exchange to variation in temperature and precipitation are significantly damped by carbon-nitrogen cycle coupling. The carbon cycle responses to temperature and precipitation variation are reduced in magnitude as atmospheric CO2 concentration rises for the coupled carbon-nitrogen model, but increase in magnitude for the carbon-only counterpart. Our results suggest that previous carbon-only treatments of climate-carbon cycle coupling likely overestimate the terrestrial biosphere's capacity to ameliorate atmospheric CO2 increases through direct fertilization. The next generation of coupled climate-biogeochemistry model projections for future atmospheric CO2 concentration and climate change should include explicit, prognostic treatment of terrestrial carbon-nitrogen cycle coupling.

1. Introduction

[2] The response of land ecosystems to increasing atmospheric CO2 concentration is an important control on the fraction of fossil fuel emissions of CO2 that remains in the atmosphere [Tans et al., 1990]. This control is modulated by the response of land ecosystems to climate variability and climate changes [Bousquet et al., 2000; Ciais et al., 2005; Goulden et al., 1996] induced in part by the atmospheric accumulation of CO2 [Friedlingstein et al., 2006]. Interactions between carbon and nitrogen cycles within land ecosystems play an important role in determining the long-term evolution of plant, litter, and soil organic matter pools of carbon and nitrogen [Chapin et al., 1986; Vitousek and Howarth, 1991], as well as the response of these pools to changes in atmospheric composition and climate [Field et al., 1992]. Modeling studies have noted that these interactions are likely to influence the future trajectories of atmospheric CO2 concentration and associated climate changes [Holland et al., 1997; McGuire et al., 1992, 2001; Moorhead et al., 1986; Rastetter et al., 1997, 1991], but to date no coupled climate-carbon cycle modeling studies have included an explicit treatment of the terrestrial nitrogen cycle [Friedlingstein et al., 2006].

[3] Hungate et al. [2003] showed that the responses of several current land ecosystem models to increasing CO2 concentration over the next century are stoichiometrically inconsistent with independent estimates of mineral nitrogen supply. Their analysis suggests that treating the nitrogen cycle explicitly in such models would lead to reduced land ecosystem sensitivity to changes in atmospheric CO2. Enhanced growth due to increasing CO2 increases plant demand for mineral nitrogen, while fresh litter inputs associated with increased growth can increase microbial demand for mineral nitrogen through enhanced immobilization potential. These effects could increase nitrogen limitation under conditions of increasing CO2, producing a decline over time in the strength of the negative feedback between land ecosystems and atmospheric CO2 [Luo et al., 2004; Miller, 1986; Reich et al., 2006].

[4] A recent intercomparison of coupled climate-carbon cycle models demonstrates a range of responsiveness of modeled land ecosystems to increasing temperature and increasing atmospheric CO2 [Friedlingstein et al., 2006]. For all models the land ecosystem response to temperature is a positive feedback, with warmer temperatures producing an increase in atmospheric CO2. Both net primary production (NPP) and heterotrophic respiration (HR) are enhanced by warming in these models, but the temperature effect on HR is stronger, resulting in a positive feedback. By excluding a prognostic nitrogen cycle and carbon-nitrogen (C-N) cycle coupling, this current generation of models misses a potentially important feedback connecting HR and NPP, whereby the supply of mineral nitrogen necessary to support new plant growth in nonagricultural systems comes mainly from nitrogen mineralization accompanying the decay of older soil organic matter [Bonan, 1990; Field et al., 1992; Pastor and Post, 1988; Schimel et al., 1990; Vitousek and Howarth, 1991]. NPP and HR are both sensitive to soil moisture status, and so we also expect that introducing realistic C-N feedbacks within the terrestrial ecosystem will alter the model responses to variation in precipitation [Betts, 2005; Prior et al., 1997; Santiago et al., 2005; Wahren et al., 2005].

[5] Here we test two major hypotheses: first, that introducing a prognostic nitrogen cycle and C-N coupling into a global land surface process model significantly reduces the strength of the terrestrial CO2 fertilization effect; and second, that C-N coupling results in a positive feedback between HR and NPP which reduces the strength of the net land carbon cycle response to changes in temperature. Both of these hypotheses are stated relative to a model which does not include a prognostic nitrogen cycle or C-N coupling. In terms of the signs and strengths of major feedbacks between the terrestrial biosphere and the global climate system, we expect that introducing C-N coupling reduces the strength of the negative carbon cycle feedback from CO2 fertilization of land ecosystems, leading to larger atmospheric accumulations of fossil fuel-derived CO2 than predicted by carbon-only models. We also expect a global reduction in the sensitivity of net land carbon exchange to temperature change, leading to a smaller net land carbon source to the atmosphere due to warming.

[6] Our study uses a recently developed model of coupled terrestrial carbon, nitrogen, water, and energy dynamics [Thornton and Zimmermann, 2007], implemented as a component of the Community Climate System Model (CCSM) [Collins et al., 2006]. The model includes the capability to run with and without a fully prognostic nitrogen cycle. We use that capability to evaluate the influence of terrestrial carbon-nitrogen cycle coupling on the response of the terrestrial biosphere to increasing atmospheric CO2 concentration, increasing mineral N deposition, and variability in temperature and precipitation. Simulations here are forced with multiyear output from the atmospheric model component of CCSM in an offline mode. Results documenting the influence of terrestrial C-N coupling in the fully coupled CCSM will be reported separately.

2. Methods

2.1. Model Description

[7] The land biogeochemistry model used here is the result of merging the biophysical framework of the Community Land Model (CLM 3.0) [Bonan and Levis, 2006; Dickinson et al., 2006; Oleson et al., 2004] with the fully prognostic carbon and nitrogen dynamics of the terrestrial biogeochemistry model Biome-BGC (version 4.1.2) [Thornton et al., 2002; Thornton and Rosenbloom, 2005]. The resulting model, CLM-CN (Community Land Model with prognostic Carbon and Nitrogen) is fully prognostic with respect to all carbon and nitrogen state variables in the vegetation, litter, and soil organic matter, and retains all prognostic quantities for water and energy in the vegetation-snow-soil column from CLM 3.0. Detailed descriptions for all biogeochemical components of CLM-CN, and for those aspects of the biophysical framework modified to accommodate prognostic vegetation structure, are provided as an electronic supplement (see Text S1).

2.2. Simulation Protocol

[8] Our objective here is to examine the effects of introducing coupled C-N dynamics in the land component of a coupled climate system model. Our approach is to perform a series of offline simulations in advance of the much more computationally demanding fully coupled carbon-(nitrogen)-climate experiments. By “offline” we mean a simulation in which the land model component of the coupled system is forced by a prescribed data set of atmospheric fluxes and states. For these simulations we created such a data set by extracting 25 years (a) of hourly results from the atmospheric model component of CCSM (the Community Atmosphere Model, CAM [Collins et al., 2006]), from an experiment in which CAM and CLM were run in a partially coupled mode (prescribed sea surface temperatures and sea ice distributions). The coupling frequency between CLM and CAM in the coupled system is 1 h, so this sampling strategy does not represent any additional aggregation of atmospheric states or fluxes.

[9] Our goal here was to obtain a sample of CAM output that would be similar in mean state and variability to the climate simulated by the fully coupled model [cf. Doney et al., 2006], such that our analysis provides a preliminary indication of the dynamics of the fully coupled system. Preliminary evaluation of carbon cycle predictions from CLM-CN when forced with reanalysis surface weather fields showed reasonable results for predicted net primary production in most vegetation types, with underpredictions in the coldest regions (arctic tundra and larch forest) [Thornton and Zimmermann, 2007].

[10] The model includes carbon and nitrogen pools with long turnover times, and the long-term accumulation of mass in these slow pools depends in part on a balance between inputs and outputs of nitrogen that are very small relative to the rates of internal nitrogen cycling [Chapin et al., 1986; Vitousek and Howarth, 1991]. The model spin-up strategy described by Thornton and Rosenbloom [2005] (accelerated decomposition method) was used to bring the carbon and nitrogen states into dynamic equilibrium with respect to the 25-a sample of CAM output. Spin-up required about 750 model years, achieved by cycling the 25-a time series of atmospheric forcing. During spin-up, atmospheric CO2 was kept constant at 283.6 parts per million by volume (ppmv), its assumed value at 1850 A.D. Land cover was assumed constant during spin-up and throughout the experiments, using the values circa 1850 A.D. from Feddema et al. [2005].

[11] The nitrogen deposition fields used in this study were generated by the three-dimensional chemistry transport MOZART-2 (Model for Ozone and Related Tracers, version 2 [Horowitz et al., 2003]). In all simulations (preindustrial, present-day and future), MOZART uses meteorological data sets valid for the period of interest, on the basis of simulations by the Parallel Climate Model [Washington et al., 2000]. The MOZART-2 simulations were performed at the horizontal resolution of 2.8°. All the dynamical and chemical processes simulated by MOZART-2 are performed with a model time step of 20 min, while the nitrogen deposition fluxes are archived monthly. For additional information on the present-day and future simulations, the reader is referred to Lamarque et al. [2005]. The preindustrial simulation is similar to the present-day simulations, except that all emissions associated with anthropogenic activities (excluding biomass burning) are explicitly set to 0. Nitrogen deposition from the MOZART-2 preindustrial simulation was used for the CLM-CN spin-up simulation.

[12] To allow direct evaluation of the effects of C-N coupling, the spin-up was performed with the regular nitrogen cycle behavior in effect and also in carbon-only mode. Switching from the carbon-nitrogen to the carbon-only model configuration results in a more productive model mean state, characterized by increased vegetation productivity and larger steady state carbon accumulations in vegetation, litter, and soil organic matter. To help distinguish the effects of increased model mean state in the absence of C-N coupling from the direct effects of nitrogen limitation, carbon-only experiments were repeated with a 32% reduction in maximum carboxylation rate (Vcmax) (Table 1). The magnitude of this scaling factor was based on comparison of gross primary production from initial carbon-only simulation (about 177 PgC a−1, preindustrial) with the estimate from the third IPCC assessment (120 PgC a−1) [Houghton et al., 2001]. A similar scaling strategy was used by McGuire et al. [1992] in their evaluation of how C-N interactions influence modeled net primary production.

Table 1. Summary of Experimentsa
ExperimentCO2 ForcingNdep ForcingbVcmax Scalingc
  • a

    All experiments are 251 years (a) duration (1850–2100).

  • b

    Nitrogen deposition forcing (not relevant for experiments C and Cr).

  • c

    Multiplier applied to Vcmax (via fraction of leaf nitrogen in Rubisco enzyme) for all plant functional types.

CNca. 1850ca. 18501.0
CN+co2transientca. 18501.0
CN+ndepca. 1850transient1.0
Cca. 18501.0
Crca. 18500.68

[13] Multiple off-line experiments of duration 251 a (nominally 1850–2100 A.D.) were initiated from the spun-up states (Table 1). Control experiments were performed from each spin-up with constant CO2 and nitrogen deposition. Experiments with increasing atmospheric CO2 used the historical record through year 2000, and followed the SRES A2 concentration scenario for years 2000–2100 [Nakicenovic and Swart, 2000]. Experiments with increasing nitrogen deposition used a linear interpolation between MOZART-2 outputs for years 1890, 2000, 2050, and 2100.

2.3. Analyses

[14] Following Friedlingstein et al. [2003], we express the sensitivity of the land carbon cycle to increasing atmospheric CO2L) as:

equation image

where ΔCL is the change in global total land carbon stock (PgC) over a given time period and ΔCA is the change in atmospheric CO2 concentration (ppmv) over the same period. Because our experiments are performed in offline as opposed to coupled mode and lack a radiatively driven climate change signal, we cannot calculate a temperature sensitivity parameter directly analogous to γL from Friedlingstein et al. [2003]. Instead, we estimate the sensitivities of the land carbon cycle to interannual variations in temperature (ST) and precipitation (SP) as the multiple least squares regression slopes for net ecosystem exchange of carbon (NEE) versus annual mean temperature and annual mean precipitation.

2.4. Model Archive

[15] Exact source code and routines for analysis of model output used in this study are archived at the Oak Ridge National Laboratory Distributed Active Archive Center, in the Biogeochemistry Model Archive (, following the model archiving guidelines provided by Thornton et al. [2005].

3. Results

3.1. Steady State Stocks and Fluxes

[16] Introduction of C-N coupling has a significant effect on the prognostic model carbon stocks at steady state (Figure 1 and Table 2). N limitation reduces total carbon stock by 43% compared to the carbon-only model in the control experiments. The reduction is greater for soil organic matter (55%) than for vegetation carbon (36%), with intermediate reductions for litter and coarse woody debris pools. N limitation also has a significant effect on the modeled carbon fluxes at steady state (Figure 1 and Table 3). Gross primary production (GPP) is reduced by 43% in control experiments, with a smaller reduction in autotrophic respiration (39%) and a larger reduction in heterotrophic respiration (47%), and a substantially larger reduction in C losses due to fire (54%). Under the carbon-only model many regions with moderate to high net primary production are maintained at relatively low total vegetation carbon by large mean annual fire fluxes. Under the C-N model nitrogen loss from fire in these same regions constrains productivity and reduces steady state fire fluxes (Table 3, fire amounts for experiment C relative to experiment CN).

Figure 1.

Example annual mean flux and state variables from final 25 a of control simulations for C-N (Experiment CN) and carbon-only (Experiment C) model configurations. (a) Net primary production (NPP). (b) Total vegetation carbon (Cveg). (c) Total soil organic matter carbon (CSOM).

Table 2. Summary of Carbon Stocks for the C-N and Carbon-Only Simulations, Showing Global Totals for Each Pool (PgC), With Percent of Total Global C Stock Shown in Parenthesesa
ExperimentWood CbVeg. CCWD CcLitter CSOM CdTotal C
  • a

    Values are given for the final 25 a of the control experiments, and averaged over the periods 1976–2000 and 2076–2100 for the transient experiments.

  • b

    Wood component of vegetation (veg.) pool.

  • c

    Coarse woody debris.

  • d

    Soil organic matter (SOM), not including litter or coarse woody debris (CWD).

CN613 (53)653 (57)147 (13)16 (1)334 (29)1150
C943 (47)1014 (50)247 (12)28 (1)736 (36)2026
Cr712 (49)771 (53)167 (11)19 (1)496 (34)1452
Years 1976–2000
CN+co2649 (54)690 (58)153 (13)16 (1)339 (28)1199
CN+ndep619 (53)660 (57)149 (13)16 (1)339 (29)1163
CN+co2ndep656 (54)698 (57)155 (13)17 (1)344 (28)1213
C+co21047 (48)1125 (51)269 (12)31 (1)776 (35)2201
Cr+co2803 (50)870 (54)184 (11)22 (1)537 (33)1612
Years 2076–2100
CN+co2801 (57)845 (60)176 (13)18 (1)357 (26)1397
CN+ndep642 (53)684 (57)154 (13)17 (1)352 (29)1206
CN+co2ndep847 (57)895 (60)186 (13)19 (1)379 (26)1480
C+co21527 (52)1625 (55)363 (12)39 (1)936 (32)2963
Cr+co21225 (54)1309 (57)263 (12)28 (1)683 (30)2284
Table 3. Summary of Carbon Fluxes for the C-N and Carbon-Only Simulations, Showing Means With Interannual Standard Deviations in Parenthesesa
  • a

    Values are given for the final 25 a of the control experiments, and over the periods 1976–2000 and 2076–2100 for the transient experiments. AR is autotrophic respiration. HR is heterotrophic respiration. NEE is net ecosystem exchange of carbon. NPP is net primary production. GPP is gross primary production. Units are PgC a–1.

  • b

    Positive upward (carbon release from land).

CN102.1 (1.0)41.6 (0.6)60.5 (0.6)40.6 (0.4)1.1 (0.1)0.01 (0.6)
C177.5 (2.4)79.0 (1.4)98.5 (1.3)76.6 (0.8)2.4 (0.2)−0.01 (1.5)
Cr146.2 (1.7)61.8 (1.0)84.4 (0.9)60.4 (0.5)1.4 (0.1)0.01 (1.2)
Years 1976–2000
CN+co2106.3 (1.2)43.7 (0.7)62.6 (0.7)41.6 (0.4)1.1 (0.1)−0.96 (0.6)
CN+ndep104.1 (1.0)42.5 (0.6)61.4 (0.6)41.2 (0.4)1.1 (0.1)−0.20 (0.6)
CN+co2ndep108.3 (1.3)44.7 (0.7)63.6 (0.7)42.3 (0.5)1.2 (0.1)−1.24 (0.6)
C+co2197.7 (4.4)89.4 (2.4)108.3 (2.2)83.1 (1.5)2.8 (0.2)−3.62 (1.7)
Cr+co2166.4 (3.9)71.9 (2.0)94.6 (2.0)66.9 (1.3)1.7 (0.1)−3.25 (1.4)
Years 2076–2100
CN+co2118.8 (1.4)49.9 (0.8)68.9 (0.8)45.8 (0.7)1.3 (0.1)−2.81 (0.6)
CN+ndep109.0 (1.2)44.9 (0.7)64.0 (0.7)43.1 (0.5)1.2 (0.1)−0.70 (0.6)
CN+co2ndep127.9 (2.2)54.5 (1.1)73.4 (1.2)49.0 (1.0)1.5 (0.1)−4.08 (0.6)
C+co2256.5 (5.9)120.3 (3.2)136.2 (2.9)105.7 (2.6)4.0 (0.3)−10.65 (2.0)
Cr+co2220.5 (4.8)99.4 (2.6)121.2 (2.4)87.5 (2.2)2.6 (0.2)−9.21 (1.6)

3.2. Response to Increasing CO2 and Nitrogen Deposition

[17] Introduction of C-N coupling significantly reduces the carbon uptake response to increasing atmospheric CO2 concentration (Tables 2 and 4). Total carbon uptake due to increasing atmospheric CO2 concentration over the historical period (years 1850–2000) was 3.7 times higher for the carbon-only model than for the for the C-N model (Table 4), with mean uptake over the period 1981–2000 of 3.8 ± 0.4 and 1.1 ± 0.1 PgC a−1 for experiments C + co2 and CN + co2, respectively. These differences persist under the assumed future CO2 trajectory: total uptake is 3.8 times higher for carbon-only than for C-N model over the period 2000–2100 AD, with mean uptake over the period 2081–2100 of 10.8 ± 0.5 and 2.8 ± 0.2 PgC a−1 for experiments C + co2 and CN + co2, respectively (Table 4). Anthropogenic N deposition by itself produces a mean uptake of 0.24 ± 0.03 and 0.73 ± 0.09 PgC a−1 for the periods 1981–2000 and 2081–2100, respectively (experiment CN + ndep, Table 4). The interaction effect of increasing CO2 and increasing N deposition on total land carbon uptake is +2.5% for the period 1850–2000, but increases to +11.3% for the period 2000–2100 (experiment CN + co2ndep, Table 4), suggesting an increase in N limitation under rising CO2.

Table 4. Change in Land Carbon Stock (ΔCTOT, PgC) and Its Percent Distribution in Various Pools, Under Increasing Atmospheric CO2 and Increasing N Deposition, for Historical (1850–2000) and Future (2000–2100) Periods

[18] Introduction of C-N coupling shifts the partitioning of carbon accumulated due to CO2 fertilization away from soil organic matter and toward vegetation pools. Of the total carbon uptake under increasing CO2, the fraction entering soil organic matter is 50–58% lower and the fraction entering vegetation is 18–22% higher for experiment CN + co2 compared to experiment C + co2 (Table 4). Fractions entering coarse woody debris are similar for the two models, and only very small fractions (1–2%) enter litter for either model. For the carbon-only model, partitioning of CO2-fertilized carbon uptake is insensitive to changes in the model mean state induced by scaling Vcmax (experiments C + co2 versus Cr + co2).

[19] Nitrogen fertilization in the C-N model shifts partitioning of accumulated carbon away from vegetation pools and toward soil organic matter, compared to the CO2 fertilization response. Partitioning of carbon to soil organic matter under N fertilization increased by a factor of three while partitioning to vegetation decreased by 29% compared to that for CO2 fertilization (experiments CN + ndep versus CN + co2), with little difference between historical and future periods (Table 4). Because the total carbon uptake due to N fertilization is 23–25% of that due to CO2 fertilization, carbon partitioning in the combined CO2 + N fertilization experiment is weighted toward the CO2-fertilized pattern. Partitioning of carbon to coarse woody debris and litter are similar for CO2, N, and CO2 + N fertilization experiments.

[20] Next we consider the sensitivity of the land uptake to CO2 concentration (βL). Applying equation (1) in a moving window to 25-a segments of the time series for global mean atmospheric CO2 concentration and total ecosystem carbon, we find that βL is lower by approximately a factor of four over the period 2000–2100 for the C-N model compared to the carbon-only model (Figure 2a). Both models show a decline in βL which is accelerating over time. Potential mechanisms for this decline that could operate in both the carbon-only and C-N models include increasing rate of rise in atmospheric CO2 concentration in the SRES A2 scenario in conjunction with constrained time constants for ecosystem response to CO2 fertilization [Fung et al., 2005], and approach to saturation in the leaf-scale photosynthetic CO2 response [Wullschleger, 1993]. In the C-N model, this decline can also be partly attributed to increasing nitrogen limitation as both plant and heterotrophic nitrogen demands increase in response to CO2 fertilization. Figure 2b shows that the ratio of βL for experiments CN + co2:C + co2 gradually declines over this period, indicating the presence of additional mechanisms for declining βL unique to the C-N model. The same figure shows that with the addition of N fertilization, the ratio of βL for experiments CN + co2ndep:C + co2 increases over this period, supporting the interpretation of increasing N limitation in CN + co2 that is being alleviated toward the end of the future scenario by N fertilization in CN + co2ndep. The modification of model mean state by imposing a 32% decrease of Vcmax in experiment Cr + co2 results in only a 13% decline in βL at year 2100, compared to experiment C + co2 (Figure 2a), indicating that the large difference in βL between CN + co2 and C + co2 is due mainly to factors other than the reduced mean state in CN + co2.

Figure 2.

Comparison of βL between C-N and carbon-only model configurations. (a) Trends in βL for multiple experiments over the period 2000–2100. Symbols indicate values for eleven carbon-only models participating in C4MIP (diamonds) and the mean of the C4MIP models (square). C4MIP results from Friedlingstein et al. [2006]. (b) Trends in ratio of βL between C-N and carbon-only model configurations over the period 2000–2100.

3.3. Response to Interannual Variation in Temperature and Precipitation

[21] Introduction of C-N coupling reduces the sensitivity of NEE to variation in temperature and precipitation (ST and SP, respectively). The sensitivity to temperature (ST) is positive (warm anomalies associated with land carbon release) and the sensitivity to precipitation (SP) is negative (wet anomalies associated with land carbon uptake) for both carbon-only and C-N models. The magnitudes of these sensitivities, however, are much smaller for the C-N model: ST and SP for the coupled carbon-nitrogen model (experiment CN) are 23% and 42%, respectively, of their values for the carbon-only model (experiment C, Figure 3). Precipitation explained ∼16% of the interannual variance in global NEE for both model configurations, while temperature explained 37% for the carbon-only model and 14% with C-N coupling. Sensitivities to temperature and precipitation (ST and SP) were reduced in magnitude by 17% and 32%, respectively, for the carbon-only model with reduced mean state compared to the default carbon-only model (experiments Cr versus C, Figure 3).

Figure 3.

Sensitivity of global net ecosystem exchange of carbon (NEE) to interannual variability in global mean air temperature (Tair, ST) and precipitation (Prcp, SP) over land, estimated over the final 25 a of the control experiments. Error bars show 1 standard deviation around mean response.

[22] There is considerable spatial variation in these responses for both carbon-only and coupled carbon-nitrogen models, with regions of positive and negative sensitivity to temperature (ST, Figure 4a) and precipitation (SP, Figure 4b). Spatial patterns for experiments CN and C are qualitatively similar, but the frequencies of large positive and negative values for the sensitivities (ST and SP) are reduced in CN. There is a general pattern of negative sensitivity to temperature (ST) and positive sensitivity to precipitation (SP) at higher latitudes (>45°), shifting to positive ST and negative SP in the mid latitudes and tropics. The spatial patterns for sensitivity to precipitation (SP) are similar to those found by Doney et al. [2006] for covariance between net land ecosystem carbon exchange and soil moisture in a long, stable control simulation from a fully coupled climate-carbon model.

Figure 4.

Sensitivity of NEE to (a) interannual variation in temperature (ST), and (b) interannual variation in precipitation (SP), calculated independently at each grid cell, using the final 25 a of the control simulations (experiments CN and C).

[23] To address the mechanisms responsible for these spatial patterns, we disaggregated the NEE versus temperature and precipitation responses into independent responses for NPP, HR, and carbon loss due to fire. On the basis of comparisons of sign and magnitude we can say that NPP variability dominates the NEE response to temperature in most regions, reinforced in the tropics and subtropics by variability in fire (Figures 5a, 5b and 5d). HR variability dominates the temperature response in Scandinavia, central Asia, southeastern India, and parts of coastal northwestern North America (Figure 5c). The NPP response is driven largely by variations in soil water availability, quantified here using the CLM biophysical variable Btran, which is the scaling factor (range 0–1) on stomatal conductance related to plant-available soil water. Warming in warm climates leads to soil drying (lower Btran, Figure 5e) due to high evaporative demand, while warming in cold climates is wetting the soil by melting soil ice (Figures 5e and 5f).

Figure 5.

Dissection of NEE response to interannual variation in air temperature (Tair) from experiment CN, showing the regression slopes for various model outputs versus Tair, from multiple regressions against Tair and Prcp. Color scales are arranged so that shades of green (red) indicate factors favoring carbon uptake (release) under warmer Tair. (a) NEE versus Tair (from Figure 4a, replicated here for ease of comparison). (b) NPP versus Tair, scaled by factor −1.0 to use a common color scale with other carbon fluxes. (c) Heterotrophic respiration (HR) versus Tair. (d) Fire flux versus Tair. (e) Btran versus Tair, scaled by factor 100.0 (see text for definition of Btran). (f) Soil ice versus Tair (total soil ice in first five soil layers, mm water equivalent). Figures 5a–5d share a color scale. Figures 5e and 5f show color scale and relevant units above map.

[24] NPP variability dominates the NEE response to precipitation in the tropics and mid latitudes, reinforced by variability in fire, with a clear shift to dominance by HR variability at high latitudes and in cold climates (Figure 6a–6d). The NPP response is driven directly by soil moisture (results not shown). The HR response in cold climates is partly explained by increased snow depth in anomalously wet years (Figure 6e), which insulates the soil in winter and keeps soil temperature relatively high (Figure 6f), leading to increased soil respiration (Figure 6c).

Figure 6.

Dissection of NEE response to interannual variation in precipitation (Prcp) from experiment CN, showing the regression slopes for various model outputs versus Prcp, from multiple regressions against Tair and Prcp. Color scales are arranged so that shades of green (red) indicate factors favoring carbon uptake (release) under higher Prcp. (a) NEE versus Prcp (from Figure 4b, replicated here for ease of comparison). (b) NPP versus Prcp, scaled by factor −1.0 to use a common color scale with other carbon fluxes. (c) HR versus Prcp. (d) Fire flux versus Prcp. (e) Snow depth (mm) versus Prcp. (f) Soil temperature at ∼20 cm depth (T soil) versus Tair. Figures 6a–6d share a color scale. Figures 6e and 6f show color scale and relevant units above map.

3.4. Changes in Sensitivities to Temperature and Precipitation Under Future Scenarios of CO2 and Nitrogen Deposition

[25] Introduction of C-N coupling reverses the sign of changes in sensitivity to both temperature and precipitation (ST and SP) over time under scenario of increasing CO2 concentration. For experiment C + co2, global values for ST and SP around year 2100 increase in magnitude by 34% and 38%, respectively, compared to their 1850 values (Figure 7). That is, the value for sensitivity to temperature becomes more positive, and the value for sensitivity to precipitation becomes more negative. When carbon-nitrogen cycle coupling is included (experiment CN + co2), ST and SP decreased in magnitude by 16% and 18%, respectively, over the same time period. When increased anthropogenic nitrogen deposition is included (experiment CN + co2ndep), the change in ST from 1850 to 2100 is smaller (7% decrease) and the change in SP is larger (33% decrease) than for CO2 increasing alone. We have not identified a mechanism explaining these differences in the transient responses to temperature and precipitation variation under increasing anthropogenic nitrogen deposition.

Figure 7.

Changes in NEE sensitivity to interannual variation in air temperature (ST) and precipitation (SP) at year 2100 in simulations with increasing atmospheric CO2 concentration, relative to ST and SP from the relevant control simulations (C, Cr, and CN). Values at 2100 are estimated from global total NEE and global mean temperature and precipitation over land for years 2076–2100.

[26] In the carbon-only model configuration with reduced photosynthesis and increasing CO2 (experiment Cr + co2), changes in sensitivity to temperature and precipitation over time are in the same direction as for the default carbon-only model configuration (experiment C + co2), with ST and SP at year 2100 increasing in magnitude by 36% and 51%, respectively, compared to their 1850 values (Figure 7).

3.5. Measures of N Limitation

[27] Short-term N limitation can be assessed in terms of an instantaneous imbalance between supply and demand, which, when integrated over time, results in a long-term limitation expressed as a diminished ecosystem mean state and diminished nitrogen demand [Chapin et al., 1986; Vitousek and Howarth, 1991]. We are able to estimate both instantaneous and long-term N limitation from our experiments. In the C-N experiments gross primary production prior to down-regulation by N limitation (referred to here as “potential GPP”) is calculated at each time step, allowing us to estimate instantaneous N limitation as the ratio of down-regulated GPP to potential GPP. In the carbon-only experiments nitrogen limitation is eliminated by adding the nitrogen required to meet demand at each time step, allowing us to estimate long-term N limitation as the ratio of C-N (down-regulated) GPP to carbon-only GPP. These measures are expressed as scalars ranging from 0 to 1, with values closer to 0.0 indicating stronger N limitation.

[28] Instantaneous N limitation results in a 31% reduction in GPP, on the basis of down-regulated versus potential global total GPP in experiment CN. Long-term N limitation results in a 42% reduction in GPP, on the basis of comparison of global total GPP between experiments CN and C. Regional patterns differ substantially between instantaneous and long-term N limitation (Figure 8). In the boreal zone, for example, instantaneous limitation is weak and long-term limitation is strong, while in cold-dry and warm-wet regions both instantaneous and long-term limitations tend to be weak. Changes in instantaneous N limitation in the transient experiments show progressively stronger N limitation under increasing atmospheric CO2 (Figure 9, experiment CN + co2). Increasing mineral nitrogen deposition has the expected effect of reducing overall N limitation (experiment CN + ndep), and the net effect of increasing CO2 and increasing mineral nitrogen deposition is a reduction over time in global mean N limitation (experiment CN + co2ndep). These transient changes are small, on the order of ±5%, indicating that the global mean instantaneous N limitation is relatively stable on multihundred year timescales.

Figure 8.

Spatial distribution of nitrogen limitation in the C-N control simulation, expressed as (a) instantaneous and (b) long-term limitation scalar (see section 3.5). Shades of red (green) show stronger (weaker) N limitation. Areas with gross primary production (GPP) <1 gC m−2 a−1 are shown in gray.

Figure 9.

Changes in instantaneous N limitation scalar over the course of transient experiments, expressed as a percent change from N limitation in the C-N control, estimated from global total actual versus potential GPP.

[29] The supplemental nitrogen input required in experiment C to completely eliminate N limitation for every land grid cell at every time step is 441 TgN a−1, 4 times higher than the estimated nitrogen inputs from atmospheric deposition (6 TgN a−1) and biological nitrogen fixation (104 TgN a−1) in experiment CN. Total accumulation of nitrogen necessary to support the carbon accumulation in experiment C + co2 is 19.2 Pg nitrogen over the period 2000–2100. This is about 3 times higher than the high end of nitrogen accumulation estimated by Hungate et al. [2003] for this period on the basis of supply limitations. The total accumulation of nitrogen in experiment CN + co2ndep over this same period is 4.7 Pg nitrogen, which falls between the high and low estimates from Hungate et al. [2003], and represents an independent, mechanistic estimate of supply-limited nitrogen accumulation in land ecosystems over the coming century.

4. Discussion

4.1. Comparison to Observations and Previous Model Results

[30] Our results showing that accumulation of carbon in soil organic matter is higher and accumulation in vegetation carbon is lower for N fertilization compared with CO2 fertilization are consistent with findings from recent observational studies. In labeled N tracer studies for forests in Europe and North America, Nadelhoffer et al. [2004, 1999] have shown that most tracer N is recovered in soil organic matter (∼80%), with little recovery in tree biomass (3%–17%, depending on tissue). In a meta-analysis of the influence of CO2 and nitrogen fertilization across multiple ecosystem types, van Groenigen et al. [2006] found that elevated CO2 leads to accumulation of soil C and N only when N is added at rates well above background atmospheric inputs. Several studies have shown that CO2 fertilization tends to shift N away from soil organic matter and into plant biomass, which, given the higher C:N for biomass versus soil organic matter, is associated with a preferential increase in vegetation carbon [Gill et al., 2006; Hungate et al., 2006; Luo et al., 2006].

[31] The issue of fractionation in carbon uptake between vegetation and soil pools is important since these pools can have very different residence times and are affected differently by disturbances such as fire and forest harvest. Dufresne et al. [2002] note that there is a factor of 2 difference in the ratio of carbon stored in vegetation to that stored in soil organic matter because of increasing CO2 between the land components of two coupled climate-carbon cycle models, with the IPSL model storing carbon preferentially in vegetation (ratio ∼2:1), while the Hadley model stores preferentially in soil organic matter (ratio ∼1:2). Our results show carbon storage due to increasing CO2 strongly weighted toward vegetation, with a ratio of ∼8:1 for CO2 fertilization, dropping to ∼5:1 when N fertilization is included.

[32] Our estimate of present-day carbon sink due to anthropogenic mineral nitrogen deposition (0.24 PgC a−1 for the period 1981–2000) is in reasonable agreement with several independent estimates. Field et al. [1992] used simple stoichiometric logic to provide a range of recent N-fertilized uptake of 0.3–2.5 PgC a−1, and suggested that the correct value is likely closer to the low end of that range. Nadelhoffer et al. [1999] used a simple budget based on a synthesis of studies in temperate forests to estimate a current global forest-only uptake of 0.25 PgC a−1 due to anthropogenic N fertilization. In a modeling study, Holland et al. [1997] give a significantly higher range for the estimated N-fertilized carbon sink, ∼0.7–2.0 PgC a−1. Most of the models included in the nitrogen deposition intercomparison of Lamarque et al. [2005] did not have a representation of ammonia atmospheric chemistry, and so the model-generated N deposition data set used in our study does not include ammonia deposition. Since this component represents more than 50% of the present-day total N deposition [Dentener et al., 2006], it is likely that our current estimate is biased low.

[33] In a recent synthesis of observed influence of increasing CO2 on NPP in temperate forests, Norby et al. [2005] found a consistent increase in NPP of ∼23% for CO2 about 200 ppmv higher than ambient (∼550 ppmv) across a broad productivity range. We find an increase in global NPP at present compared to preindustrial steady state of 7% (∼100 ppmv increase in CO2), and a 22% increase in NPP at year 2100 relative to present (∼460 ppmv increase in CO2). Our results are not directly comparable to the experimental findings, since the experiments impose a step change in CO2 while our results are from transient simulations, for which a smaller response is expected.

[34] Declining βL under increasing CO2 observed in our transient experiments for both carbon-only and C-N models (Figure 2a) is similar in pattern and mechanism to results from Fung et al. [2005] for a fully coupled climate-carbon cycle model with a carbon-only land biogeochemistry component. Our estimates of βL at 2100 from carbon-only experiments C + co2 and Cr + co2 (1.4 and 1.2 PgC ppmv−1, respectively) are similar to the results from Fung et al. [2005] (1.1 PgC ppmv−1), and very close to the mean βL from the eleven models participating in the recent C4MIP intercomparison (1.35 PgC ppmv−1, Figure 2a) [Friedlingstein et al., 2006]. This suggests that reductions in βL similar to those observed between our carbon-only and C-N model might also be obtained from the addition of explicit C-N coupling in other models.

[35] The global stock of vegetation C in experiment CN + co2ndep (ca. 2000) is in good agreement with recent estimates, but the total soil organic matter carbon stock is significantly lower than current estimates [Houghton et al., 2001]. As described previously [Thornton, 1998; Thornton and Rosenbloom, 2005], the converging cascade model of litter and soil organic matter dynamics used here can be parameterized with either three or four soil organic matter pools, with equivalent fidelity to data from radio-labeled substrate decomposition experiments. Additional experiments showed that the vegetation states and carbon uptake transient responses to increasing atmospheric CO2 and nitrogen deposition are affected very little by the choice of three or four soil organic matter pools (differences generally <2%, results not shown). The total soil organic matter stock when using the four-pool model is approximately twice as large, in better general agreement with inventory-based estimates [Houghton et al., 2001], and so we recommend that future simulations use the four-pool model configuration.

4.2. Influence of Model Mean State

[36] Our analysis of the influence of mean state is intended to address the pragmatic question of whether it would be adequate to substitute a simpler model formulation to capture the main C-N induced climate-carbon cycle responses. If changes in fundamental carbon-climate responses induced by a simple alteration of the carbon cycle mean state in the carbon-only model are similar to the changes obtained by the introduction of full prognostic C-N coupling, then an argument could be made for avoiding the added complexity of the C-N coupling mechanisms, instead parameterizing the C-N coupling effects as a down-regulation of mean rates of photosynthesis. Conversely, an argument can be made for explicitly including the C-N coupling if the responses to altered mean state and added C-N mechanisms differ in ways that are likely to be important for the purposes of a particular model application. Here we are specifically concerned with model applications in the context of a fully coupled carbon cycle-climate simulation for projection of future climate on timescales of several hundred years.

[37] The relative importance of model mean state on the finding of greatly reduced land carbon uptake sensitivity to atmospheric CO2 concentration (βL) in the C-N compared to carbon-only model (Figure 2a) can be assessed by comparing some indicators of mean state, such as vegetation carbon and GPP, with differences in βL between models C, Cr and CN. Since the mean states for Cr and CN are not the same, we normalize responses by comparing the changes in mean state relative to C. Reductions in vegetation carbon and GPP, compared to C, are 1.5 and 2.3 times larger, respectively, for CN than for Cr. We therefore expect that if the effect on βL of introducing C-N coupling is mainly due to the effect of N limitation on the model mean state, then the ratio of differences in βL between CN and Cr should fall close to the range 1.5–2.3. In fact the ratio of differences in βL (at year 2000) is 6.6, several times larger than the range for mean state differences. This suggests that most of the difference in βL between models CN and C is due not to the reduced mean state in CN, but to more fundamental C-N coupling mechanisms. Results are similar when considering model behavior at year 2100.

[38] The same analysis applied to steady state estimates of land carbon cycle sensitivity to interannual variation in temperature and precipitation (ST and SP, Figure 3) suggests that C-N coupling mechanisms play an important role in the C-N model response to temperature, but that the differences between carbon-only and C-N model in response to precipitation could be due mainly to differences in the model mean state. For the case of changes in ST and SP over time under the influence of rising atmospheric CO2 concentration, the influence of C-N coupling (CN + co2 versus C + co2 in Figure 7) has the opposite sign compared to the influence of reduced mean state in the carbon-only model (Cr versus C in Figure 7). This is a strong indication that the difference in temporal dynamics of ST and SP between the C-N and carbon-only models is not primarily due to different mean states. We conclude that for the βL and ST responses, as well as for the temporal dynamics of the SP response, the behavior of the C-N model cannot be adequately captured through simple down-regulation of photosynthesis in the carbon-only model.

4.3. Model Limitations and Future Directions

[39] The CLM-CN configuration described here has several significant shortcomings. First, these simulations have not considered the influence of changing land cover and land use on C-N cycle dynamics. Disturbance history has been shown to have a strong influence on carbon and nitrogen cycle dynamics in previous observational studies [Bautista-Cruz and del Castillo, 2005; Gholz et al., 1985; Law et al., 2003; Prober et al., 2005; Torn et al., 2005] and modeling studies [Bond-Lamberty et al., 2005; Bugmann and Solomon, 2000; Schimel et al., 1997; Thonicke et al., 2001; Thornton et al., 2002; Turner et al., 2007]. Without an explicit treatment of disturbance history it is difficult to assess the behavior of the model against historical observations of atmospheric CO2 concentration, since an important part of that signal is due to carbon sources and sinks from vegetation disturbance, land management practices, and regrowth. This capability is currently being added to the model, and future studies will report on the influence of these dynamics on the climate-carbon response mechanisms discussed here.

[40] While the current model includes a detailed treatment of many components of the terrestrial nitrogen cycle, it lacks detail on the processes of nitrification, denitrification, and volatilization, all of which are important components of the long-term nitrogen balance. The vertical distribution of nitrogen cycling processes in the soil column is not treated explicitly, in part because of large uncertainties associated with these processes, compounded by uncertainties in the treatment of vertical soil water dynamics. This prevents an explicit treatment of speciation of mineral nitrogen between ammonium and nitrate. More sophisticated models for these processes are available [e.g., Li et al., 2005; Neff and Asner, 2001], but considerable work is required to integrate the existing knowledge within the biophysical and biogeochemical framework of CLM-CN. Development in this direction is underway now.

[41] While the current model includes the fundamental controls on fire from fuels and climate, there has not yet been any systematic evaluation of the model behavior against observations. The transport of fire emissions is also not treated prognostically in our fully coupled system, although there is evidence that long-range transport from large biomass burning sources could play an important role in C-N dynamics [Fabian et al., 2005].

[42] The current treatment of biological nitrogen fixation is entirely empirical, and has an important influence on the long-term establishment of the C-N model mean state. We found that model response to increasing CO2 is not particularly sensitive to the parameterization of this process (results not shown), but a more mechanistic treatment of the process would help to address uncertainties in the detailed spatial patterns of N limitation. The results presented here will also be sensitive to the details of the modeled N deposition fields [Lamarque et al., 2005]. We are exploring the effects of replacing the MOZART-2 inputs with results from other models, from the work of Lamarque et al. [2005] and/or Dentener et al. [2006]. In the future we plan to have a prognostic nitrogen deposition capability operational within the fully coupled CCSM framework.

[43] Finally, the current model ignores potential limitations from other nutrients. Phosphorus limitation, in particular, is known to play an important role in tropical forests growing on highly weathered soils [Vitousek and Howarth, 1991] as well as in some temperate systems [Gosz et al., 1973], and there are also potentially important interactions between the phosphorus cycle and biological nitrogen fixation [Vitousek et al., 2002]. We note that, even for systems where phosphorus limitation is more important than nitrogen limitation, the inclusion of nitrogen dynamics will produce a result that is closer to reality than a carbon-only model. However, because of the critical role of the tropical forests in establishing the past, present, and future trajectories of the global carbon cycle and climate-carbon cycle interactions, we must very soon confront the challenge of developing a parsimonious treatment of carbon-nitrogen-phosphorus coupling for use in global coupled climate system modeling.

5. Conclusions

[44] We tested the hypothesis that inclusion of explicit prognostic coupling between the carbon and nitrogen cycles in the land biogeochemistry component of a coupled climate system model has important consequences for climate-carbon cycle interactions. We found that, in comparison to a carbon-only model configuration, the most critical mechanisms controlling the sign and magnitude of feedbacks between the global climate system and the terrestrial biosphere are significantly altered by the introduction of an explicit prognostic treatment of the nitrogen cycle. Specifically:

[45] 1. Sensitivity of land carbon uptake to increasing atmospheric CO2 concentration is smaller by a factor of 3.6 for C-N versus carbon-only model configurations, with a shift for the C-N model toward proportionally more carbon uptake in vegetation and less in soil organic matter. Total carbon uptake due to increasing atmospheric CO2 over the period 2000–2100 is smaller by a factor of 3.8 for the C-N versus carbon-only model.

[46] 2. Land carbon cycle responses to interannual variation in both temperature and precipitation have significantly smaller magnitudes for the C-N model, compared to its carbon-only counterpart: globally integrated responses are smaller by factors of 4.3 and 2.4 for temperature and precipitation, respectively.

[47] 3. Under the influence of rising atmospheric CO2 concentration, land carbon cycle sensitivities to interannual variation in temperature and precipitation are shown to decrease in magnitude over time for the C-N model, while increasing in magnitude over time for the carbon-only counterpart.

[48] 4. The influence of model mean state does not appear to explain the large decrease in sensitivity to CO2, the smaller sensitivity to temperature variation, or the transient changes in temperature and precipitation sensitivity under increasing CO2 that result from introduction of C-N coupling.

[49] We conclude that introduction of terrestrial C-N coupling is likely to have a fundamental impact on the climate-carbon cycle feedbacks in a fully coupled climate-biogeochemistry simulation. We further conclude that a simple reparameterization of a carbon-only model to produce an altered mean state resembling that obtained under explicit C-N coupling is not likely to result in climate-carbon cycle dynamics similar to those obtained under the explicit C-N coupled system. C-N coupling is certain to reduce the direct CO2 fertilization effect in a coupled simulation, producing a tendency toward higher atmospheric CO2 concentrations for identical fossil fuel emissions scenarios. The complex spatial patterns of land carbon cycle response to temperature and precipitation variation suggest that the strength and sign of the globally integrated carbon-climate feedbacks will depend on the convolution of these patterns with the spatial patterns of climate change resulting from a particular model or ensemble member. Investigations of climate-carbon cycle feedback mechanisms using CLM-CN as a component of a fully coupled climate-carbon system model are now underway.


[50] We thank S. Doney, D. Schimel, G. Bonan, and E. Holland for helpful discussions, and we thank two anonymous reviewers for their thoughtful and constructive comments on the manuscript. This work was supported in part by NASA Earth Science Enterprise, Terrestrial Ecology program, grant W19,953 to P. E. Thornton. Additional support was provided by the National Center for Atmospheric Research (NCAR) through the NCAR Community Climate System Modeling program, and through the NCAR Biogeosciences program. NCAR is sponsored by the National Science Foundation.