Modelling changes in nitrogen cycling to sustain increases in forest productivity under elevated atmospheric CO 2 and contrasting site conditions

If increases in net primary productivity (NPP) caused by rising concentrations of atmospheric CO 2 (Ca) are to be sustained, key N processes such as soil mineralization, biological fixation, root uptake and nutrient conservation must also be increased. Simulating the response of these processes to elevated Ca is therefore vital for models used to project the effects of rising Ca on NPP. In this modelling study, hypotheses are proposed for changes in soil mineralization, biological fixation, root nutrient uptake and plant nutrient conservation with changes in Ca. Algorithms developed from these hypotheses were tested in the ecosystem modelecosysagainst changes in N and C cycling measured over several years under ambient vs. elevated Ca in Free Air CO2 Enrichment (FACE) experiments in the USA at the Duke Forest in North Carolina, the Oak Ridge National Laboratory forest in Tennessee, and the USDA research forest in Wisconsin. More rapid soil N mineralization was found to be vital for simulating sustained increases in NPP measured under elevated vs. ambient Ca at all three FACE sites. This simulation was accomplished by priming decomposition of N-rich humus from increases in microbial biomass generated by increased litterfall modelled under elevatedCa. Greater nonsymbiotic N 2 fixation from increased litterfall, root N uptake from increased root growth, and plant N conservation from increased translocation under elevated Ca were found to make smaller contributions to simulated increases in NPP. However greater nutrient conservation enabled larger increases in NPP with Ca to be modelled with coniferous vs. deciduous plant functional types. The effects of these processes on productivity now need to be examined over longer periods under transient rises in Ca and a greater range of site conditions.


Introduction
The extent to which forest net primary productivity (NPP) increases under elevated atmospheric CO 2 concentration (C a ) has been found to vary greatly with site conditions.These increases are greater in warmer environments (Myers et al., 1999) because elevated C a suppresses photorespiration (Long, 1991), in water-limited environments (Hättenschwiler et al., 1997) because elevated C a reduces transpiration and raises water use efficiency (Rogers et al., 1983), and in nutrient-rich environments which enable more rapid nutrient uptake under elevated C a (Oren et al., 2001).
In temperate forests, increases in NPP with C a have been found to be constrained by the availability of nitrogen (N).This constraint may be greater in deciduous than in coniferous forests because of their greater nutrient demands (Norby et al., 2010).Consequently studies of elevated C a effects on forest NPP and growth have given variable results.Norby et al. (2005) found an increase in NPP of 23 % after 1-3 yr under 560 vs. 360 µmol mol −1 in a meta-analysis of Free Air CO 2 Enrichment (FACE) experiments with young, temperate-zone forests.However longer-term exposure of trees to elevated C a has shown this increase to decline over time (Hättenschwiler et al., 1997;Idso, 1999;Medlyn et al., 1999;Norby et al., 2010) unless N uptake increases commensurately with NPP (Oren et al., 2001).

R. F. Grant: Modelling changes in nitrogen cycling to sustain increases in forest productivity
Several changes in N processes have been observed in experimental studies by which N uptake and accumulation may be hastened under elevated C a , thereby moderating long-term N limitations to NPP: 1. more rapid mineralization of soil N primed by increased litterfall (Crow et al., 2009;Drake et al., 2011;Phillips et al., 2011), 2. more rapid symbiotic or nonsymbiotic N 2 fixation driven by increased root C allocation or litterfall (Hofmockel and Schlesinger, 2007;Norby, 1987), 3. more rapid N uptake from denser and deeper root growth (Iversen, 2010;Luxmore et al., 1986), 4. greater nutrient conservation through increased translocation and recovery of N from senescing plant material (Drake et al., 2011).
Biogeochemical models used to forecast the response of forest productivity to rising C a must therefore simulate changes in these processes that result in additional plant N uptake and retention (Crow et al., 2009;Finzi et al., 2007).These processes are either absent or poorly resolved in current models used to study the effects of rising C a on forest growth (Iversen, 2010).Different representations of these processes among terrestrial ecosystem models with coupled C-N cycles cause variation in productivity gains simulated with rising C a .This variation can cause substantial differences in the magnitude and even in the direction of terrestrial feedbacks to C a among these models during climate change scenarios (Sokolov et al., 2008;Zaehle et al., 2010).There is therefore a need for greater accuracy and consistency in modelling changes in N cycling and hence in N constraints to productivity under rising C a .
All processes in hypotheses 1-4 above are represented in detail in the ecosystem model ecosys (Grant et al., 2007(Grant et al., , 2009a(Grant et al., , b, c, 2010(Grant et al., , 2011a)).The importance of these processes to simulating sustained increases in forest NPP and growth was assessed by comparing changes in forest N and C cycling modelled under elevated vs. ambient C a with those measured or calculated over several years from forest Free Air CO 2 Enrichment (FACE) experiments in the Duke Forest, Orange County NC, in the Oak Ridge National Laboratory (ORNL) forest at Oak Ridge TN, and the USDA Forest Service experimental farm in Rhinelander WI. Results from this testing were used to estimate the relative contribution from each of these four processes to sustaining increases in forest NPP with elevated C a at each of the three experimental sites.

Model Description
The key algorithms governing the simulation of C and N transformations in ecosys are described in the supplement to this article, in which equations and variables referenced below are described and listed in Supplement A through H. Algorithms representing biological processes in soil (Supplement A, G and H), physical processes driving soil-plantatmosphere water transfer (Supplement B), biological processes in plants (Supplement C and F), and chemical processes governing soil solute transformations (Supplement E) were solved at an hourly time step from hourly changes in atmospheric boundary conditions.Algorithms representing physical processes driving soil water, heat, gas and solute transfers (Supplement D) were solved at a 5 min time step assuming constant boundary conditions during each hour.All parameters in these algorithms remained unchanged from those in earlier studies of forests, crops and grasslands cited in the Introduction.The key model hypotheses for the effects of elevated C a on the four N processes given in the Introduction are described in further detail below with reference to equations in the Supplement.
3 Model hypotheses for changes in N cycling under elevated C a 3.1 More rapid mineralization of soil N primed by increased litterfall

Decomposition
Organic transformations in ecosys occur concurrently in five organic matter-microbe complexes (coarse woody litter, fine nonwoody litter including root exudates, animal manure (if present), particulate organic matter (POM), and humus), each of which consists of five organic states (three decomposition substrates -solid organic C, N and P; sorbed organic C, N and P; and microbial residue C, N and P; their decomposition products -dissolved organic C, N and P (DOC, DON and DOP); and the decomposition agent, active microbial biomass (M)) in a surface residue layer and in each soil layer.The rates at which each of the three substrates decompose in each complex are first-order functions of M in diverse heterotrophic microbial functional types, including obligate aerobes (bacteria and fungi), facultative anaerobes (denitrifiers), obligate anaerobes (fermenters), heterotrophic (acetotrophic) and autotrophic (hydrogenotrophic) methanogens, and aerobic and anaerobic heterotrophic diazotrophs (nonsymbiotic

Mineralization
Growth of M by each microbial functional type in each organic matter-microbe complex [A25] is calculated from its uptake of DOC

Priming
Values of M used to drive decomposition [A1] are modelled by allowing M to transfer among the five organic mattermicrobe complexes according to differences in the concentration of M with respect to that of its substrate in each complex [A3].These transfers are indicated by changes in the relative sizes of the state variables for biomass vs. those of their substrates before and after redistribution in Fig. 1.The larger specific decomposition rates used in the fine litter and manure complexes [A4] cause concentrations of M vs. substrate in these complexes to be greater than those in the POM and humus complexes (Fig. 1).Consequently some M is transferred from the litter and manure to the POM and humus complexes, thereby priming decomposition of the substrates in these complexes [A1].Similarly, some M may also be transferred from the POM to the humus complex.This transfer implements the hypothesis of Fontaine et al. (2004) that fast-growing microbes specializing in utilization of fresh litter inputs produce extracellular enzymes that metabolize not only fresh litter but also existing SOC.The priming caused by this transfer in ecosys has been tested against changes in decomposition, CO 2 emission and N mineralization measured from litter-amended vs. non-amended soil in Grant et al. (1993).These transfers and consequent priming are greater with larger litter substrate and hence larger M generated by increased litterfall, as occurs under elevated C a .Because POM and humus have lower C : N ratios than does litter, this priming favours mineralization over immobilization.

Greater biological N 2 fixation primed by increased litterfall
Nonsymbiotic diazotrophic M populations in each complex conduct N 2 fixation [A27] and associated respiration [A28] to remedy N deficits calculated from their biomass and C : N ratios.These deficits arise when biomass N from uptake of DON [A22] and immobilization of NH + 4 and NO − 3 [A26b, c] is less than that needed to attain set C : N ratios with biomass C from uptake of DOC [A21].These deficits increase with larger litter substrate and hence larger diazotrophic M generated by increased litterfall, as occurs under elevated C a , particularly when higher litterfall C : N ratios increase DOC vs. DON uptake.Thus N 2 fixation rates increase with litterfall, as observed experimentally by Jurgensen et al. (1992) and modelled in Grant et al. (2007).The subsequent decay of diazotrophic M [A23] causes transfer of diazotrophic N [A24] to microbial residues and humus for decomposition and uptake by other M [A36], and to NH + 4 through mineralization [A26a], from which it can contribute to root and mycorrhizal uptake [C23].These same algorithms are used to model symbiotic diazotrophic M populations in roots of leguminous host plants (Supplement F).

C allocation and root growth
Growth of branch, root and mycorrhizal organs is driven by growth respiration R g [C17] and consequent assimilation of the nonstructural C product of CO 2 fixation (σ C ) [C20], coupled with assimilation of the nonstructural N and P products of root and mycorrhizal uptake (σ N and σ P ).Amounts of σ C , σ N and σ P in each organ are maintained by translocation along branch-root-mycorrhizal concentration gradients [C50-C53] generated by production of σ C from branch CO 2 fixation [C1-C12] and of σ N and σ P from root and mycorrhizal uptake [C23] vs. consumption of σ C , σ N and σ P from R g and branch, root and mycorrhizal growth [C20] (Grant, 1998).Non-remobilizable C associated with translocated C is lost as litterfall C [C18], so that litterfall is not prescribed from assumed turnover rates, but arises from fixation vs. respiration of C and uptake vs. assimilation of N and P as determined by soil and weather.Elevated C a could therefore hasten litterfall and increase translocation by increasing R m with respect to R c , or by reducing σ N : σ C with respect to structural N : C required for growth.Amounts of litterfall N and P are reduced with respect to that of litterfall C by additional translocation of remobilizable N and P to σ N and σ P .This additional translocation increases as ratios of σ N : σ C or σ P : σ C decline [C19], so that N : C and P : C ratios in litterfall decline from those in the living phytomass from which it originates.This decline in litterfall nutrient ratios therefore increases with declining nutrient status as frequently, although not always, observed in field studies (Barnes et al., 1998;Kimmins, 2004).Elevated C a could therefore increase this additional translocation by lowering ratios of σ N : σ C .The parameters used to calculate translocation in ecosys were selected to give the range in N and P translocation fractions from senescing material found in Kimmins (2004) across a range of sites with differing nutrient availability.

Root growth and N uptake
In branches, translocation and hence litterfall start at the lowest node at which leaves are present, and proceed upward until the requirement for R s , σ N or σ P by the branch is met.In roots or mycorrhizae, translocation and hence litterfall start with secondary axes and proceed to primary axes in each soil layer until the requirement for R s , σ N or σ P by roots or mycorrhizae in the soil layer is met.For deciduous plant functional types, litterfall is hastened by phenologically driven withdrawal of σ C , σ N and σ P into N storage pools under shortening photoperiods and declining temperatures during autumn.These storage pools then replenish σ C , σ N and σ P under lengthening photoperiods and rising temperatures to drive leafout during the following spring (e.g.Grant et al., 2009c).

Effects of changes in N cycling on CO 2 fixation modelled under elevated C a
Adaptation to elevated C a by each of the four N processes described above contributes to maintaining foliar σ N vs. σ C under elevated C a , the first three through increased N uptake, and the fourth through increased N conservation.In so doing, these changes help to maintain σ C and σ N assimilation and  Finzi et al. (2007), Norby et al. (2005) and in other references cited below.Modelling methodology at each site followed a common approach described below.

The Duke Forest FACE experiment
Productivity of a loblolly pine (Pinus taeda) stand planted in 1983 was measured from 1997 to 2005 under ambient (∼ 371 µmol mol −1 ) vs. elevated (∼ 571 µmol mol −1 ) C a (Drake et al., 2011;McCarthy et al., 2010).The history of this site was simulated by planting a temperate coniferous functional type (e.g.Grant et al., 2007Grant et al., , 2009a) ) in 1958 on a soil with properties given in Oh and Richter (2005) and growing it until 1983 under historical C a and repeating sequences of half-hourly weather data recorded from 1997 to 2005 at the Duke Forest.Ammonium and nitrate concentrations in precipitation were set to give wet N deposition rates reported in Drake et al. (2011), and atmospheric ammonia concentrations used to calculate ammonia deposition [D15] were set to 5 nmol mol −1 .This spinup allowed ecosys to achieve stable changes in C stocks during successive weather sequences.The modelled stand was then clearcut (Grant et al., 2007) and a second stand was planted in 1983 with the same plant density as that at which the field site was planted in the same year.This stand was grown until 2006 under 371 µmol mol −1 C a and repeating sequences of half-hourly weather data recorded from 1997 to 2005 at the Duke Forest.This run was then repeated from 1997 to 2005 under 571 µmol mol −1 C a .To simulate the response to N fertilizer measured at the site, both runs were repeated with annual fertilizer applications of 11.2 g N m −2 as urea from 1998 to 2005 as described in McCarthy et al. (2010).Model output for NPP, phytomass and key N transfers were compared with those measured over the same period under the same changes in C a and N.

The ORNL FACE experiment
Productivity and N relations of a sweetgum (Liquidambar styraciflua L.) stand planted in 1988 was measured from 1998 to 2008 under ambient (∼ 370 µmol mol −1 ) vs. elevated (∼ 550 µmol mol −1 ) C a (Norby et al., 2010).The history of this site was simulated by planting a temperate deciduous functional type (e.g.Grant et al., 2009a) in 1954 on a soil with properties given in Johnson et al. (2004), and growing it until 1988 under historical C a and repeating sequences of half-hourly weather data recorded from 1990 to 1998 at the ORNL Walker Branch site, and from 1999 to 2008 at the ORNL FACE site (Riggs et al., 2010).Ammonium and nitrate concentrations in precipitation were set to give wet N deposition rates reported in Norby et al. (2010), and atmospheric ammonia concentrations were set to 5 nmol mol −1 .This spinup allowed ecosys to achieve stable changes in C stocks during successive weather sequences.The modelled stand was then clearcut and a second stand was planted in 1988 with the same plant density as that at which the field site was planted in the same year.This stand was grown until 2008 under 370 µmol mol −1 C a and repeating sequences of half-hourly weather data recorded from 1990 to 1998 at the ORNL Walker Branch site and from 1999 to 2008 at the ORNL FACE site.This run was then repeated from 1998 to 2008 under 550 µmol mol −1 C a .To simulate the response to N fertilizer measured at the site (Norby et al., 2010), the run under 370 µmol mol −1 C a was repeated with annual fertilizer applications of 20 g N m −2 as urea from 2004 to 2008.Model results for NPP, wood increment, leaf N concentrations and root mass were compared with those measured over the same period under the same changes in C a and N.

The Rhinelander FACE experiment
Productivity and N relations of an aspen (Populus tremuloides Michx.)stand planted in 1997 was measured from 1998 to 2006 under ambient (∼ 374 µmol mol −1 ) vs. elevated (∼ 541 µmol mol −1 ) C a (King et al., 2005;Kubiske et al., 2006;Talhelm et al., 2012).The history of this site was simulated by planting an annual cereal functional type (e.g.Grant et al., 2011b) N and P in [C19c, d] to a common value that was independent of changes in nonstructural N:C ratios caused by changes in C a .Results for NPP with each disabled process were then compared with those from the model runs with all processes enabled.To examine whether sensitivity of NPP modelled under elevated vs. ambient C a differed with plant functional type, the ORNL FACE runs were re-executed with the temperate deciduous functional type replaced by the temperate coniferous functional type used in the Duke FACE runs.Similarly the Duke FACE runs were re-executed with the temperate coniferous functional type replaced by the temperate deciduous functional type used in the ORNL FACE runs.NPP modelled under elevated vs. ambient C a at both sites were then compared for the deciduous vs. coniferous types.2010), with error bars representing standard errors from replicated measurements (n = 4).Root exudation (Grant, 1993) was excluded from modelled NPP to simulate biometric measurements.

The Duke Forest FACE experiment
Net primary productivity measured and modelled without fertilizer at the Duke Forest rose by 23-30 % from 1997 through 2001 after C a was increased from 371 to 571 µmol mol −1 (Fig. 2a).Both rises were greater than 30 % in 2002, when drought reduced modelled NPP less under elevated C a than under ambient.This smaller reduction was modelled from slower soil drying by transpiration which had been reduced from that under ambient C a [B1c] since the start of the experiment in 1997.Reduced transpiration was modelled from increased canopy stomatal resistance (r c ) caused by the different responses of carboxylation and diffusion to elevated C a [B2a] (Grant et al., 1999(Grant et al., , 2004)).However measured rises in NPP remained near 30 % during 2003 and 2004, while modelled rises declined to near 20 %. Almost 80 % of the measured and modelled rises in NPP were allocated to leaves (Fig. 3a) and wood (Fig. 3b), and only about 20 % to roots (Fig. 3c).2010), with error bars representing standard errors from replicated measurements (n = 4).Root exudation (Grant, 1993) was excluded from modelled NPP to simulate biometric measurements.
Annual fertilizer applications caused modelled and measured NPP and phytomasses to rise above unfertilized values over time (Fig. 2a, b), indicating N limitation to forest productivity at the Duke site.However rises in NPP modelled under 571 vs. 371 µmol mol −1 C a with fertilizer were similar to those without fertilizer (20-30 %), while measured rises were smaller (10-20 %) (Fig. 2a).Consequently gains in phytomass modelled under elevated C a with fertilizer were larger than those measured (Fig. 2b).
These increases in NPP were driven by ones in gross primary productivity (GPP in [C1]) modelled under 571 vs. 371 µmol mol −1 C a (Table 1) caused by increases in mesophyll aqueous CO 2 concentrations [C6a], calculated from intercellular gaseous CO 2 concentrations assumed to rise proportionally with C a .These increases in GPP were partially offset by increases in autotrophic respiration (R a in [C13]) driven by increased biomass R m [C16] and R g [C17] (Fig. 2b) when calculating increases in NPP.These increases drove a cumulative gain in NPP of 2032 g C m −2 (2317 g C m −2 including root exudation) from 1997 to 2004 that was similar to one of 2216 g C m −2 derived from biometric measurements by McCarthy et al. (2010) (Table 1).
This gain was partially offset by a cumulative increase in R h of 685 g C m −2 that was consistent with one of 800 g C m −2 estimated by Drake et al. (2011).This gain was further offset by small increases in C losses from CH 4 emission, leaching and runoff, when calculating net biome productivity (NBP) (= Flux Gains -Flux Losses in Table 1).
Almost 70 % of the cumulative gain in NPP modelled from 1997 to 2004 appeared in plant foliage, wood, root and nonstructural C stocks, giving a total gain in plant biomass of 1593 g C m −2 (Fig. 2b), which was similar to one of 1735 g C m −2 derived from biometric measurements by Mc-Carthy et al. (2010).The remaining 724 g C m −2 of this gain in NPP was returned to the soil as litterfall, consistent with an increase of 79 g C m −2 yr −1 measured in 1998 by Finzi et al. (2001).Of this increase in litterfall, 316 g C m −2 was above ground, corresponding to increased inputs of 48 ± 14 g C m −2 yr −1 measured in the forest floor by Lichter et al. (2008), which included some root litterfall.The remaining 408 g C m −2 of the increase in modelled litterfall including exudation was below ground, which was greater than an increase of 15 g C m −2 yr −1 in root litterfall averaged from measurements over this period by Pritchard et al. (2008a).
This increase in modelled litterfall generated a rise in midseason M of 21 % under elevated C a [A25], similar to one of 15 % in microbial C estimated by Drake et al. (2011), which drove the modelled increase in R h [A11] (Table 1).Differences between increases in litterfall and R h caused changes in soil C stocks, including a gain in litter, but a loss in POC and humus, the decomposition of which was primed by the increase in M from litterfall [A3] (Fig. 1).These gains in soil C were slightly smaller than those estimated by Drake et al. (2011) (Table 1).
Additional N required to sustain the increase in GPP modelled under elevated vs. ambient C a was partially provided by a small cumulative increase in nonsymbiotic N 2 fixation [A27] driven by that in R h (Table 1).This increase arose from modelled rates of N 2 fixation averaging 0.180 vs. 0.175 g N m −2 yr −1 under elevated vs. ambient C a , close to potential N 2 fixation rates of ∼ 0.02 g N m −2 mol −1 measured in the forest floor plus mineral soil by Hofmockel and Schlesinger (2007).However the modelled increase in N 2 fixation was mostly offset by a decrease in atmospheric NH 3 deposition to the canopy [D15] and by an increase in N leaching, both caused by reduced stomatal conductance (g c ) and hence transpiration modelled under elevated C a [B2a] as described earlier.Wet N deposition was the same under both ambient and elevated C a and so was not included in Table 1.
The cumulative difference in N inputs modelled under elevated vs. ambient C a did not account for the gains in plant structural N stocks needed to sustain the increases in plant structural C stocks (Table 1).Most of these gains were consequently withdrawn from plant nonstructural N stocks in seasonal reserves used in ecosys to buffer differences between plant N requirements and root N uptake under variable growth conditions.Consequently these stocks were more Table 1.Differences in cumulative fluxes and stocks of C and N estimated from biometric measurements (E) or modelled (M) at the Duke Forest FACE experiment after 8 yr (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004) under 571 vs. 371 µmol mol −1 CO 2 .
C heavily drawn upon to meet increased N requirements under elevated C a .However most of this withdrawal occurred within the first two years of the model experiment, indicating that its magnitude may have been an artifact of the sudden onset of elevated C a in 1997.The remainder of these gains was drawn from soil N stocks through increased mineralization of POM and humus N [A26a], driven by the more rapid decomposition of POM and humus C described earlier (Table 1).The consequent decline in POM and humus N stocks (Table 1) was partially offset by a gain in litter N stocks from immobilization of some of the additional NH + 4 and NO − 3 [A26b, c] mineralized from POM and humus.This loss in POM and humus N stocks and the gain in litter N stocks were consistent with the loss in mineral-associated C and gain in organic horizon C estimated by Drake et al. (2011) (Table 1).
The gain of plant N and loss of soil N modelled under elevated vs. ambient C a (Table 1) was achieved by increased root and mycorrhizal uptake [C23] of the N mineralized from POM and humus and not immobilized in litter, and by increased translocation of N from senescing plant material [C19] (Fig. 4).This increase in translocation caused litterfall N modelled under elevated C a to remain similar to that modelled under ambient C a , in spite of the increase in litterfall C (Table 1), and so enabled the increase in N uptake modelled under elevated C a to be almost fully retained in the forest canopy.These modelled increases in uptake and translocation were corroborated by experimental findings (Drake et al., 2011) (Fig. 4).
The modelled increases in N uptake and retention (Table 1; Fig. 4) maintained σ N vs. σ C and hence increases in GPP [C6, C11, C12] and NPP under elevated C a as the experiment progressed (Fig. 2).However modelled gains in plant N through increased root N uptake and translocation under elevated C a (Fig. 4) were not commensurate with those in plant C through increased NPP (Fig. 2a).Consequently ecosys simulated a rise in the NPP : N uptake ratio from 162 to 173 g C g N −1 under 371 vs. 571 µmol mol −1 C a .The increase in this ratio was consistent with one from 158 to 170 g C g N −1 (assuming 0.5 g C g DM −1 ) measured at the Duke FACE site by Finzi et al. (2007).Consequently foliar N concentrations modelled  in 1998 declined from 24.0 to 22.6 mg N g C −1 , while those measured in the same year by Finzi et al. (2001) declined from 21.0 to 18.8 mg N g C −1 , although this measured decline was considered to be nonsignificant.These rises in NPP : N uptake ratios and consequent declines in foliar N concentrations suggest a gradually increasing N limitation that could eventually reduce future increases in NPP.

The ORNL FACE experiment
Net primary productivity modelled at ORNL was generally larger than that derived from measurements (Fig. 5a), although wood growth increments were similar (Fig. 5b).NPP modelled under 550 µmol mol −1 C a rose sharply above that under 370 µmol mol −1 C a during the first two years of the experiment (Fig. 5a), but declined relatively more thereafter so that gains in NPP modelled under elevated C a became smaller as the experiment progressed.Almost 80 % of the modelled rise in NPP was allocated to leaves (Fig. 6a) and wood (Fig. 6b), and only ca.20 % to roots (Fig. 6c) as was modelled at Duke (Fig. 3).Gains in leaf NPP modelled under elevated C a in ecosys were larger, while those in root NPP were smaller, than gains derived from biometric measurements by Norby et al. (2010), who attributed most of the total gain in NPP at ORNL to root production, particularly in 2000-2002 (Fig. 6c).The greater modelled vs. measured NPP (Fig. 5a) was mostly attributed to greater modelled vs. measured root NPP (Fig. 6c).Norby et al. (2010) attributed these declining gains in NPP to more rapid declines in tree N status, apparent in lower foliar N concentrations measured under elevated C a   2010), with error bars representing standard errors from replicated measurements (n = 2).Root exudation (Grant, 1993) was excluded from modelled NPP to simulate biometric measurements.
(Fig. 7).Lower foliar N concentrations were also modelled [C6, C12] through changes in σ N vs. σ C [C11], although values were consistently 15 mg N g C −1 larger than those measured (Fig. 7).Strong N limitations to NPP were also indicated by the sharp rise in wood increment measured and modelled under ambient C a after fertilizer application started in 2004 (Fig. 5b).In ecosys, declining N status caused greater allocation of σ C , σ N and σ P to roots vs. leaves and wood [C50-C53], so that root NPP was maintained (Fig. 6c) while leaf and wood NPP declined (Fig. 6a, b) as the experiment progressed.
The modelled trend towards smaller gains in NPP as the experiment progressed was reversed in 2002, 2006and 2007 (Fig. 5a) (Fig. 5a).During these years soil drying following lowerthan-average precipitation in 2001 and in 2005-2007 caused smaller declines in NPP to be modelled under elevated vs. ambient C a (Fig. 5a), as at the Duke Forest in 2002 (Fig. 2a).Smaller declines under elevated C a were also apparent in NPP derived from measurements in 2002, but not in 2006 and 2007.The greater decline in NPP measured under elevated vs. ambient C a in 2007 was attributed by Warren et al. (2011)  elev.gain under elevated C a during the 2007 drought.In ecosys, however, soil water conserved before droughts by reductions in transpiration under elevated vs. ambient C a , as was observed by Warren et al. (2011), allowed smaller declines in transpiration, g c and hence in net C uptake to be modelled under elevated C a as droughts progressed.

Year (c) root
Increases in GPP [C1] modelled under 550 vs. 370 µmol mol −1 C a drove a cumulative gain in NPP from 1998 to 2008 of 1858 g C m −2 (1971 g C m −2 including root exudation), slightly greater than one of 1542 g C m −2 derived from biometric measurements by Norby et al. (2010) (Table 2).This gain was partially offset by an increase of 680 g C m −2 in R h , and by a small increase in C losses from leaching and runoff, when calculating NBP (Table 2).
About two-thirds of the gain in NPP modelled from 1998 to 2008 appeared in plant wood, root and nonstructural C stocks (Table 2).Gains in foliar stocks were not modelled because values were calculated at the end of each year after leafoff.The remaining 640 g C m −2 of this gain in NPP was returned to the soil as litterfall, of which 302 g C m −2 was above ground, consistent with an annual increase of 40 g C m −2 yr −1 in surface litterfall measured in 2000 by Johnson et al. (2004).The remaining 338 g C m −2 of the increase in modelled litterfall including exudation was below ground, which was less than the 675 ± 500 g C m −2 increase in total root litterfall estimated to 2006 by Iversen et al. (2008).
This increase in modelled litterfall generated a rise in midseason M of 9 % [A25], similar to one of 13 % in microbial C measured in July 2000 by Johnson et al. (2004), which drove the modelled increase in R h [A11] (Table 2).This increase slightly exceeded that in litterfall, causing small declines in litter, POM and humus C stocks primed by the increase in M from litterfall [A3] (Fig. 1).These declines contrasted with a substantial increase in soil C measured by Iversen et al. (2012) under elevated vs. ambient C a in 2009, but this increase was also apparent in pretreatment measurements of soil C by Johnson et al. (2004) and so may have been caused by pre-existing soil variability.
Additional N required to sustain the increase in GPP modelled under elevated vs. ambient C a was partially provided by a small cumulative increase in nonsymbiotic N 2 fixation [A27], and in atmospheric NH 3 deposition due to greater N immobilization at the soil surface [D15] (Table 2).Additional N was provided by reductions in N losses from emissions and leaching caused by greater N immobilization [A26] in the soil profile at this N-limited site (Table 2).The reduction in N loss from leaching in ecosys was consistent with one of 0.2 g N m −2 yr −1 measured experimentally in 1999 by Johnson et al. (2004).
The cumulative net gains in N inputs partly accounted for the substantial increase in plant structural N stocks needed to sustain increases in plant structural C stocks (Table 2).Some of this increase in plant structural N stocks was withdrawn from plant nonstructural N stocks.This withdrawal was less  2010), 3 Iversen et al. (2012).
than that at Duke (Table 1) because greater withdrawals for leafout in spring and greater losses from senescence in autumn caused plant nonstructural N stocks to be lower in deciduous trees.The remainder of the increase in plant structural N stocks was drawn from soil N stocks through increased mineralization of POM and humus N [A26a], driven by the more rapid decomposition of POM and humus C described earlier.
As for the Duke Forest, the gain of plant N and loss of soil N modelled under elevated vs. ambient C a at ORNL (Table 2) was achieved by increased root and mycorrhizal uptake [C23], some of which might be attributed to increased root and mycorrhizal growth as well as to priming of humus decomposition.Detailed measurements of root productivity by Iversen et al. (2008) enabled testing of changes in root growth and function modelled under ambient vs. elevated C a (Fig. 8a).Modelled root and mycorrhizal mass densities increased under 550 vs. 370 µmol mol −1 C a (Fig. 7), but less than did measured values during 2001 (Fig. 8a), the year of greatest root productivity reported in Iversen et al. (2008) (Fig. 6c).
In ecosys, greater root and mycorrhizal growth [C20b] under elevated C a was driven by greater shoot-root and rootmycorrhizal transfers of σ C [C50, C52] from greater concentration gradients generated by greater CO 2 fixation [C1] (Grant, 1998) (Fig. 8a).Greater growth drove greater root and mycorrhizal elongation [C21b, c] (Fig. 8b) used to calculate path lengths and surface areas for mineral N uptake [C23a-d].Greater σ C concentrations drove an increase in C exudation [C19e-i] of 53 vs. 42 g C m −2 yr −1 (Grant, 1993) (Fig. 8c), while more rapid root and mycorrhizal respiration [C15] and lower σ N vs. σ C drove an increase in root and mycorrhizal litterfall [C18] of 234 vs. 208 g C m −2 yr −1 .This increase in litterfall was smaller, but not significantly different, than one of 255 ± 148 vs. 161 ± 87 measured in 2001 by Iversen et al. (2008).These increases in exudation and litterfall provided more substrate for litter M growth [A1] and hence priming of POM and humus decomposition [A3] (Fig. 1) under elevated C a .However greater root and mycorrhizal elongation generated only slightly more rapid mineral N uptake [C23a-d], in which increases in mycorrhizal uptake offset reductions in root uptake below 0.5 m (Fig. 8d).Gains in NPP modelled and measured at ORNL were smaller than those at the Duke Forest after the third year under elevated C a (Fig. 5a vs. Fig.2a) in spite of similar soil attributes and climate, suggesting that deciduous forests may respond less to elevated C a than do coniferous.When the deciduous functional type used the ORNL simulations was replaced with the coniferous one used at Duke, similar values of NPP were modelled for both types under ambient and elevated C a from 1998 to 2004 (Fig. 9a).However NPP modelled for the coniferous functional type did not decline thereafter (Fig. 9a), but followed a time course at ORNL similar to that at Duke (Fig. 2a).
The smaller gains in NPP modelled for the deciduous functional type under elevated C a from 2005 through 2008 were attributed to larger declines in leaf N concentrations than those for the coniferous functional type (Fig. 9b).These greater declines were caused by greater N loss from above ground litterfall relative to N gain from root uptake, and hence less retention of N in the canopy and in reserves by the deciduous functional type than by the coniferous one.This lesser retention was caused by phenology-driven withdrawal in deciduous functional types of reserve σ C , σ N and σ P to drive leafout in spring, and of branch σ C , σ N and σ P and consequent greater litterfall during leafoff in autumn.Less retention of N was apparent in smaller ratios of NPP : N uptake modelled under ambient vs. elevated C a at ORNL (90 vs. 108 g C g N −1 ) than at Duke (162 vs. 173 g C g N −1 ).These smaller ratios were consistent with ones of 120 vs. 112 g C g N −1 at ORNL and 158 vs. 170 g C g N −1 at Duke (assuming 0.5 g C g DM −1 ) measured by Finzi et al. (2007).Consequently ratios of NPP modelled under elevated vs. ambient C a remained smaller for the deciduous vs. coniferous functional type at ORNL (Fig. 9c), explaining the smaller response of NPP to elevated C a modelled at ORNL vs. Duke (Fig. 2a vs. Fig.5a).A similar reduction in NPP ratios under elevated vs. ambient C a was modelled by replacing the coniferous functional type with a deciduous one in the simulations at Duke (results not shown).

The Rhinelander FACE experiment
The forest stands at Rhinelander were exposed to elevated vs. ambient C a within a year of planting, so that elevated C a raised GPP [C1] and hence NPP during early growth by increasing both CO 2 fixation rates per unit leaf area [C6a] and leaf area growth [C21a].Consequently gains in NPP modelled under elevated C a were relatively large during 1998 and 1999, when elevated C a increased leaf area growth at low values of LAI limiting to CO 2 fixation (Fig. 10a).These gains declined from 2000 to 2004, when LAI attained values less limiting to CO 2 fixation.Only ∼ 60 % of these gains in NPP were allocated to foliar and wood production in ecosys (Fig. 11a, b), with the remainder allocated to root production (Fig. 11c).However measured gains were allocated more to foliar and wood production and less to root production than were modelled gains, so that root and hence total NPP in ecosys rose above that measured as the experiment progressed.
Declining gains in NPP modelled from 2001 to 2004 (Fig. 10a) were not attributed to increasing N limitation, because litterfall N concentrations measured and modelled under ambient and elevated C a did not decline as the experiment progressed, in contrast to ORNL, where foliar N concentrations declined (Fig. 7).Some N limitation to gains in NPP was apparent in lower litterfall N concentrations measured and modelled under elevated C a (Fig. 12), due in ecosys to lower foliar N concentrations and greater N translocation as modelled at the other sites (Fig. 4 and Fig. 7).Kubiske et al. (2006) attributed declining gains in tree growth measured from 2001 to 2004 under elevated vs. ambient C a (Fig. 10b) to declining July radiation recorded at Rhinelander.This decline was accompanied by one in temperature, causing modelled and measured rises in NPP and wood increment to slow under ambient C a , and more so under elevated C a , during 2001-2004 (Fig. 10a, b) (Fig. 10a, b).Conversely, radiation and T a recorded at Rhinelander were higher in 2005 than in 2001-2004, causing modelled rises in NPP and wood increment to increase under ambient C a , and more so under elevated C a .
Increases in GPP [C1] modelled under 541 vs. 374 µmol mol −1 C a [C1] drove a cumulative gain in NPP from 1998 to 2006 of 860 g C m −2 (Table 3), which on an annual basis was smaller than one of 807 g C m −2 from 1998 to 2003 derived from biometric measurements by King et al. (2005) (Fig. 9a).However the modelled gain rose to 929 g C m −2 if exudation was included.This gain was partially offset by an increase in R h of 387 g C m −2 and by small increases in C losses from emission, leaching and runoff (Table 3).
About 55 % of modelled gains in NPP appeared in plant wood, root and nonstructural C stocks (Table 3).Gains in foliar stocks were not modelled because values were calculated at the end of each year after leafoff.The remaining 415 g C m −2 of this gain in NPP was returned to the soil as litterfall, of which 149 g C m −2 was above ground, consistent with an increase in annual surface litterfall of 32 g C m −2 yr −1 measured in 2003 by Liu et al. (2005) and an average of 30 g C m −2 yr −1 measured from 2002 to 2008 by Talhelm et al. (2012).The remaining 266 g C m −2 of the   2011) in 2006, measured wood increments from Kubiske et al. (2006), with error bars representing standard errors from replicated measurements (n = 3).Root exudation (Grant, 1993) was excluded from modelled NPP to simulate biometric measurements.increase in modelled litterfall including exudation was below ground.
This increase in modelled litterfall drove the modelled increase in R h (Table 3), which almost entirely offset that in litterfall, causing a small rise in litter C stocks and a small decline in POM and humus C stocks primed by an increase in M from litterfall [A3] (Fig. 1).These changes in soil C stocks in ecosys were consistent with a small increase in coarse POM stocks and concurrent small decrease in mineral-associated organic matter observed by Hofmockel et al. (2011b) under elevated vs. ambient C a at Rhinelander.
Additional N required to sustain the increase in GPP modelled under elevated vs. ambient C a was partially provided by a small cumulative increase in nonsymbiotic N 2 fixation [A27], partially offset by a small decrease in atmospheric NH 3 deposition (Table 3).Additional N was provided by reductions in N losses from emissions and leaching caused by greater N immobilization in the soil profile [A26].
The cumulative net gain in soil N inputs partly accounted for the increase in plant structural N stocks needed to sustain the increase in plant structural C stocks (  humus C described earlier.The contribution of this increased mineralization to plant N stocks was partially offset by increased immobilization of N in litter [A26b, c] and consequent gain in litter N stocks (Table 3).

Contributions of model hypotheses to modelled response of NPP to C a
The contributions of priming, N 2 fixation, root growth and translocation to increases in NPP modelled under elevated C a at each site were then examined by modelling changes in these increases when each process was disabled as described above.When the algorithm for priming POM and humus decomposition by litterfall [A3] (Fig. 1) was disabled, ecosys was unable to simulate rises in NPP under elevated C a consistent with those modelled with priming enabled at all three sites (Fig. 13).At Duke and Rhinelander, no rises were modelled without priming after 2002 and 2000 respectively (Fig. 13a, c), while at ORNL rises without priming remained about one-half of those with the full model (Fig. 13b).When nonsymbiotic N 2 fixation [A27] was dis-abled, NPP modelled under both ambient and elevated C a was little affected at all three sites (Fig. 13a, b, c).When the algorithm for the effect of σ N : σ C ratios on translocation [C19] was disabled, ecosys simulated rises in NPP under elevated C a at Duke and Rhinelander that were similar to those with the algorithm enabled (Fig. 13a, c).However rises simulated at ORNL declined from those with the algorithm enabled after 2003 (Fig. 13b) as N limitations to NPP became more severe (Fig. 5b, Fig. 7).When root growth under elevated C a was constrained to that under ambient C a , increases in NPP were little affected, except during early growth at Rhinelander (Fig. 13c).The algorithm for priming was apparently essential for modelling NPP response to elevated C a , while those for N 2 fixation, root growth and translocation were less important.Because C : N ratios in humus were lower than those in litter, priming also hastened net N mineralization [A26a] and uptake [C23] (Tables 1, 2 and 3 ;Figs. 4,9c).This model representation of priming was consistent with the experimental findings of increased root exudation, microbial biomass and exoenzyme activity (Phillips et al., 2011) from greater litterfall (Lichter, 2008) that hastened soil respiration and hence N mineralization and uptake from older SON fractions under elevated vs. ambient C a (Drake et al.,  2011; Hofmockel et al., 2011a, b).However gains in root N uptake modelled with priming under elevated C a were partially offset by increased N immobilization [A26b, c] apparent in greater litter N stocks (Tables 1, 2 and 3) generated by greater litterfall with higher C : N ratios (Fig. 12).These model results were consistent with experimental findings that increased litterfall under elevated C a induced concurrent increases in immobilization of plant-derived SOM and in mineralization of mineral-associated SOM (Hofmockel et al., 2011b;Holmes et al., 2006).When priming was disabled, ecosys was unable to simulate sustained increases in NPP under elevated C a at all three sites (Fig. 13a, b, c), indicating that simulation of priming was vital to modelling the response of NPP to C a in an ecosystem model with fully coupled C-N cycles.Models in which humus decomposition and mineralization are independent of litterfall are therefore unlikely to be able to accurately simulate long-term changes in forest productivity under elevated vs. ambient C a .

Greater biological N 2 fixation primed by increased litterfall
Greater litterfall under elevated C a in ecosys increased growth of diazotrophic M [A25] and hence increased nonsymbiotic N 2 fixation [A27] (Grant et al., 2007).However these increases were small with respect to those in plant N stocks (Tables 1, 2 and 3), contributed little to increases in NPP (Fig. 13), and so were unlikely to be the main source for increased N uptake.Hofmockel and Schlesinger (2007) did not detect significant increases in nonsymbiotic N 2 fixation from incubated samples of forest floor and mineral soil  taken from plots under elevated vs. ambient C a in the Duke FACE experiment.However they did measure substantial increases from these samples with additions of labile C representing litterfall, as has been modelled by ecosys in other studies (e.g.Fig. 1 in Grant et al., 2007).Hofmockel and Schlesinger (2007) concluded that increases in nonsymbiotic N 2 fixation would not enable the more rapid N uptake required to sustain increases in NPP under elevated C a .This conclusion was consistent with the relatively minor contribution of these increases to those in plant N stocks in ecosys, although this contribution may become more important in longer-term responses to elevated C a under more N-limited conditions than those at the FACE sites modelled here.

Greater N uptake from increased root growth
Root growth has consistently been found to increase under elevated C a (reviewed in Iversen, 2010), as modelled here (Fig. 8a, b, c).Mycorrhizal growth has been found to increase relatively more than root growth under elevated C a , particularly below 0.15 m (Pritchard et al., 2008b), as also modelled here (Fig. 8a, b, c).These increases in growth, and hence elongation (Fig. 8b), reduced path lengths for N diffusion [C23a, c] and increased surface areas for active N uptake [C23b, d] and hence increased N uptake rates (Figs. 4 and 8d).Increases in N uptake can be inferred from gains in biomass N modelled under elevated C a at each of the three sites (Tables 1, 2, 3).However uptake was also determined by aqueous concentrations of NH  1, 2, 3) which were products of root N uptake prior to the elevated C a treatments.Consequently increases in N uptake and assimilation modelled under elevated C a (Fig. 4; Tables 1, 2 and 3) could not be clearly attributed to increases in root and mycorrhizal growth (Fig. 13) since the effects on N uptake of this growth could not be separated from those of soil N transformations and plant N storage.

Greater nutrient conservation through translocation of N from senescing plant material
Greater translocation was modelled under elevated C a [C19] (Fig. 4), but contributed to increases in NPP only at ORNL after several years (Fig. 13b) as N limitation increased.Both modelled and measured leaf N concentrations consistently declined under elevated C a (Fig. 7).Such declines frequently induce greater translocation of N from senescing plant material (Barnes et al., 1998), conserving nonstructural N stocks but further reducing N concentrations in litter (Fig. 12) from those in leaves.Greater translocation modelled under elevated C a caused rates of leaf litterfall N to remain similar to those under ambient C a in spite of greater rates of litterfall C, as observed experimentally by Johnson et al. (2004).However experimental evidence for greater translocation under elevated C a has been inconsistent, being found in some studies (Drake et al., 2011) but not in others (Johnson et al., 2004).Greater N conservation through translocation from senescing plant material is therefore likely to be a less important process in maintaining increases in NPP under elevated C a than is soil priming, except possibly under severe N limitation.
However model results indicated that nutrient conservation is an important adaptation to elevated C a , because differences in nutrient conservation among plant functional types affected responses of NPP to C a .Smaller increases in NPP under elevated C a modelled for deciduous vs. coniferous functional types (Fig. 9c) were attributed to more rapid leaf turnover, which increased canopy N losses in spite of translocation.These losses were apparent in smaller ratios of NPP : N uptake modelled and measured at ORNL vs. Duke. Franklin et al. (2009) attributed these smaller increases to more rapid fine root turnover at ORNL vs. Duke because a greater fraction of the increase in NPP measured under elevated C a at ORNL was allocated to roots (Fig. 6c) (Norby et al., 2010).However in ecosys, increases in NPP were allocated to roots similarly at ORNL and Duke (Fig. 6c vs. Fig.3c) because parameters in the algorithms for shoot-rootmycorrhizal transfers of σ C , σ N and σ P were assumed not to change with functional type [C50-C53].Both attributions indicate that NPP of functional types with more rapid turnover of plant material will experience greater N limitation and hence will increase less over time under elevated C a .

Conclusions
The response of forest NPP to elevated C a is limited by N in most temperate and boreal zones.However this limitation appears to be moderated by several N processes in soils and plants that adapt to increased C flows through forest ecosystems as C a rises.The extent of this mitigation will vary with the rates of these processes over time and so needs to be assessed over longer periods and broader conditions than those of the three experiments used here for testing model hypotheses.Of key concern is the gradual depletion of humus N from priming, which may reduce its contribution to moderating N limitations over longer periods of time.Also the adaptation of these N processes to elevated C a , for example the withdrawal of nonstructural N to sustain gains in structural N, may differ under the sudden rise in C a imposed in these experiments than under the gradually rising C a currently in progress.All these processes need therefore to be investigated over longer time periods under transient rises in C a .These investigations may be supported by longer-term modelling of forest productivity under diverse site conditions and transient changes in C a and climate.The development and testing of algorithms for these processes may help to resolve current uncertainty in modelling terrestrial feedbacks to C a during climate change.
2 fixers) [A1, A2].Decomposition rates are also Monod functions of substrate C concentrations in soil [A4], calculated from the fraction of substrate mass colonized by M [A5].These decomposition rates drive transfers of C, N and P from the substrates to DOC, DON and DOP [A7].
to greater declines in transpiration, g c and hence C

Fig. 9 .
Fig. 9. (a) Annual net primary productivity (NPP), (b) leaf N concentrations and (c) ratios of NPP modelled from 1998 to 2008 under ambient (∼ 370 µmol mol −1 ) vs. elevated (∼ 550 µmol mol −1 ) C a with deciduous vs. coniferous plant functional types at the ORNL FACE experiment.Values of NPP and leaf N concentrations modelled with the deciduous plant functional type are the same as those in Fig. 5a and Fig. 7.

Fig. 13 .
Fig. 13.Ratios of annual net primary productivity (NPP) modelled under ambient vs. elevated C a at the (a) Duke, (b) ORNL and (c) Rhinelander FACE experiments with the full model or with model algorithms for priming by litterfall [A3], nonsymbiotic N 2 fixation [A27], increased root growth driven by plant C status [C50], and translocation driven by plant N status [C19] disabled(-).Agreement between enabled and disabled algorithms for NPP modelled under ambient C a at each site was achieved by offsetting changes in model parameters affected by disabling as described in the Methods.

Sensitivity of modelled NPP to changes in N cycling under ambient vs. elevated C a
each year from 1956 to 1971 on a soil with properties given inDickson et al. (2000)under historical C a and repeating sequences of half-hourly weather data recorded from 1999 to 2006 at the Rhinelander FACE site.Ammonium and nitrate concentrations in precipitation were set to give wet N deposition rates reported in NADP maps for Wisconsin, and atmospheric ammonia concentrations were set to 5 nmol mol −1 .A boreal deciduous func-This stand was grown until 2006 under 374 µmol mol −1 C a and repeating sequences of half-hourly weather data recorded from 1999 to 2006 at the Rhinelander FACE site.This run was then repeated from 1998 to 2006 under 541 µmol mol −1 C a .Model results for NPP, wood increment and litter N concentration were compared with those measured over the same period under the same changes in C a .To examine sensitivity of NPP modelled under elevated vs. ambient C a to the N processes described above, the model runs at the three FACE sites were repeated for the duration of the elevated C a treatments with selected processes disabled.The extent to which priming of soil N mineralization (hy- Grant et al., 2009a)nt et al., 2009a)was then planted in 1972 and grown under 1999-2006 hourly weather data until 1996.This spinup allowed ecosys to achieve stable changes in C stocks during successive weather sequences.The stand was then clearcut and a second boreal deciduous stand was planted in 1997 with the same plant density as that at which www.biogeosciences.net/10/7703/2013/Biogeosciences, 10, 7703-7721, 2013 the field site was planted in the same year.pothesis 1) raised NPP under elevated vs. ambient C a was examined by setting the transfer rate constant in [A3a] to zero and offsetting the consequent slowing of POM and humus decomposition by increasing the specific rate constants for POM and humus decomposition in [A4a].In this way POM and humus decomposition was made independent of changes in litterfall caused by changes in C a .The extent to which more rapid nonsymbiotic N 2 fixation (hypothesis 2) raised NPP under elevated vs. ambient C a was examined by disabling diazotrophic fixation in [A27].The extent to which increased root growth (hypothesis 3) raised NPP under elevated vs. ambient C a was examined by reducing the rate constant for nonstructural C exchange among roots and branches [C50] under elevated C a so that root mass remained similar to that under ambient C a .The extent to which greater translocation and recovery of N (hypothesis 4) raised NPP under elevated vs. ambient C a was examined by setting the translocation fractions for nonstructural 1 1 Includes root exudation, 2 Drake et al. (2011), 3 McCarthy et al. (2010), 4 Hofmockel et al. (2011a), 5 Finzi et al. (2007).

Table 3 )
. The remainder of this increase was drawn from soil N stocks through increased mineralization of POM and humus N [A26a], driven by the more rapid decomposition of POM and