Carbon-nitrogen feedbacks in the UVic ESCM

Introduction Conclusions References

plant tissue, litter, soil and the mineral pools before being taken up again by the plant. Biological N 2 fixation and nitrogen deposition represent the external input and loss from the plant-soil system can occur via leaching. Simulated carbon and nitrogen pools and fluxes are in the range of other models and data. Gross primary production (GPP) for the 1990s in the CN-coupled version is 129.6 Pg C a −1 and net C uptake is 10 0.83 Pg C a −1 , whereas the C-only version results in a GPP of 133.1 Pg C a −1 and a net C uptake of 1.57 Pg C a −1 . At the end of a transient experiment for the years 1800-2000, where temperature is held constant but CO 2 fertilisation for vegetation is allowed to happen, the CN-coupled version shows an enhanced net C uptake of 1.05 Pg C a −1 , whereas in the experiment where CO 2 is held constant and temperature is transient 15 the land turns into a C source of 0.60 Pg C a −1 by the 1990s. The arithmetic sum of the temperature and CO 2 effects results in 0.45 Pg C a −1 , which is 0.38 Pg C a −1 lower than seen in the fully forced model, suggesting a strong non-linearity in the CN-coupled version. Anthropogenic N deposition has a positive effect on Net Ecosystem Production of 0.35 Pg C a −1 . Overall, the UVic CN-coupled version shows similar characteristics in 20 terms of C and N pools and fluxes to other CN-coupled Earth System Models.

Introduction
There is growing evidence that the availability of nitrogen (N) in terrestrial ecosystems has an important effect on the global carbon (C) cycle Gerber et al., 2010;Zaehle et al., 2010b;Bonan and Levis, 2010). The sensitivity of the terres-C-only models estimate β L to be 1.4±0.5 Pg C ppm −1 and γ L to be −79±45 Pg C K −1 (Denman et al., 2007). Models that include the interactions between the terrestrial C and N cycle show a decrease in β L , i.e. a suppressed CO 2 fertilisation effect (Thornton et al., 2009;Bonan and Levis, 2010) and γ L either becomes less negative or switches from being negative to being positive Bonan and Levis, 2010), i.e. a smaller release of C from the soil and vegetation pools or even an increase in these pools with increasing temperature. The overall effect of CN interactions on the terrestrial C balance is model dependent and ranges from less C storage to no change in C storage in the future when compared to C-only models (Friedlingstein and Prentice, 2010).
Due to the growing evidence that N potentially has an important impact on the terrestrial C cycle, it is necessary to develop a suite of models that represent the CN interactions. So far, the models range from low resolution models (4 • in latitude × 5 • in longitude, i.e.  to high resolution models (0.5 • × 0.5 • ), i.e. Yang et al. (2009) ;Jain et al. (2009). The only fully-coupled models in terms of climate-20 carbon feedbacks so far are that of  and Thornton et al. (2009).
With this study, we add another model to this list: we further develop the University of Victoria Earth System Climate Model (UVic ESCM) through the incorporation of the terrestrial CN feedback mechanisms. The UVic ESCM falls in the category of Earth System Models of Intermediate Complexity (EMIC) and is a fully coupled model the N model is adopted from Gerber et al. (2010) with modifications in order to fit the UVic ESCM structure; wherever we use Gerber et al.'s approach, we mention it in the respective section below.

Organic pools
Litterfall for C (C LF ) is determined for each plant functional type (PFT) by the size of 15 the carbon pools, C leaf , C root and C wood and by a pool specific turnover rate, η root and η wood (Table 2): where η leaf = η 0 leaf f (T )f (Θ); η 0 leaf is given in Table 2, f (T ) and f (Θ) are given in Eqs. (9) and (10). 20 Before plants drop their leaves, a fraction of the N is reabsorbed. This is taken account of by the factor r leaf in the calculation of litterfall for N, N LF : where CN leaf , CN root and CN wood are the C/N ratios of leaves, roots and wood (see Sect. 2.4.1). The C/N ratio of litterfall is different than that of the plant source because a portion of leaf nitrogen (r leaf ) is reabsorbed by the plant before abscission. Litterfall (C LF , N LF ) is added to the litter pools (C L , N L ), whereas humification (C HUM , N HUM ) and litter respiration (C RESPL ) and mineralisation (N MINL ) are subtracted 5 from the litter pools: Humification is the transfer of organic material from the litter to the soil pool (Eqs. 5 and 6), litter respiration is the decomposition of organic C in litter to form CO 2 (Eq. 7) 10 and litter mineralisation is the decomposition of organic N in litter to form ammonium (NH + 4 ) (Eq. 8): N HUM = f (T )f (Θ)k L N L (1 + ξ[N min,av ])τ (6) 15 where the temperature dependence f (T ) is a function of soil temperature (Cox, 2001, Eq. 17): 72 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | where q 10 = 2.0 and T s is the soil temperature in • C and f (Θ) is a function of soil moisture (Cox, 2001, Eq. 18) where S, S w and S 0 are the soil moisture, the wilting point soil moisture and the optimum soil moisture (Cox, 2001, Eqs. 19-21). Other terms used in Eqs. (5) and (8) are 5 a specific litter turnover rate k L (Table 3), the litter pool size (C L , N L ) and the concentration of available, mineral N, [N min,av ] (see Table 4 for relationships between various mineral N pools and concentrations). The parameter ξ (Table 3) describes the dependence of respiration and mineralisation on available mineral nitrogen concentration and is taken from Gerber et al. (2010). The fraction τ (Table 3) defines how much 10 of the litter is humified and transferred to the soil pool and how much is decomposed (C RESPL , N MINL ). Humified litter material is transferred to the soil pools, C s and N s , which are decreased by respiration (C RESPS ) in case of C (Eq. 11) and by mineralisation (N MINS ) in case of N (Eq. 12). The organic N soil pool, N s , is further increased by the immobilisa-15 tion of ammonium and nitrate (NH IMM 4 where soil respiration, C RESPS , 20 and soil mineralisation, N MINS , 73

Mineral pools
The UVic-CN model has two separate N mineral pools, ammonium (NH + 4 ) and nitrate (NO − 3 ); for simplicity, the pools are labelled NH 4 and NO 3 hereafter. The rates of change of these two pools are given by:  (17) and (18)) and a passive uptake (second part on RHS in Eqs. (17) and (18)). The active plant uptake represents the part of the uptake that is driven by exchange of ions between the roots and the soil, i.e. for each NH + 4 molecule taken up, a proton is exuded, whereas the 5 passive uptake transports the N contained in the soil water via the transpirational water stream.
Active plant uptake depends on the PFT-dependent maximum uptake rate ν max per 10 unit root mass (kg N (kg root C) −1 a −1 ), C root (kg C m −2 ), soil depth h S (m), the halfsaturation constant k p,1/2 (kg N m −3 ) (see Tables 3 and 2 for values), the available ammonium NH 4(av) (kg N m −2 ) and the total concentration of available mineral N, [N min(av) ] (kg N m −3 ) (Table 4 lists the relationships between different mineral N pools).
The second part on RHS in Eqs. (17) and (18)  3 ) via microorganisms occurs when the soil quality decreases, i.e. soil C/N ratio (CN soil ) increases; in the UVic-CN immobilisation happens when soil C/N is greater than 13 -a value in the range used by Gerber et al. (2010) and Zaehle and Friend (2010); consequently, the soil C/N ratios in the UVic CN model are kept more or less constant. Relating immobilisation rates to 5 the C/N ratio is controversial as biomass and metabolic state of microorganisms seem to be better predictors of immobilisation rates (Bengtsson et al., 2003) than soil C/N ratios. However, microbial biomass and their metabolic state are not variables that are ready to be included into a global Earth System Model. Further, we do not know if soil C/N ratios will not change under CO 2 fertilisation and higher N demands. The concept 10 of the equations for immobilisation is taken and adjusted from Gerber et al. (2010): (20) where NH IMM 4 and NO IMM 3 are in (kg N m −2 a −1 ), the temperature f (T ) and moisture function f (Θ) are given in Eqs. (9) and (10), k L is the specific litter turnover rate (a −1 , 15  (Table 3), C L is the litter C pool (kg C m −2 ) and CN soil is the C/N ratio of the soil (kg C (kg N) −1 ).

Nitrogen input
External nitrogen inputs consist of biological nitrogen fixation (BNF) and atmospheric 5 deposition of NH ). During the spin-up period, we use the relationship between BNF and evapotranspiration (ET) based on Cleveland et al. (1999) that has been used in CN models of Zaehle and Friend (2010) and Yang et al. (2009): where the original units are modified to kg N m −2 a −1 for BNF and mm a −1 for ET. However, using this approach for transient simulations  in the UVic ESCM leads to a significant reduction in NPP at the end of the 20th century due to a reduction of ET with increasing CO 2 concentrations and we therefore opt for the apparently more robust relationship used by the Community Climate Model CLM4 (Thornton et al., 2009) 15 and relate total annual BNF to NPP. After the UVic-CN model has come to equilibrium for the year 1800 using the relationship between evapotranspiration (ET) and BNF following Eq. (22) we obtain a relationship, , between BNF and net primary production (NPP): 20 where NPP is in kg C m −2 a −1 and is 2.73 g N (kg C) −1 and gives 180 Tg N a −1 for an NPP of 66 Pg N a −1 . Nitrogen deposition takes place in dry and wet form of NH x (NH DEP 4 ) and NO y (NO DEP 3 ) and occurs close to the source of pollution. Nitrogen deposition onto terrestrial ecosystems has increased by a factor of 3.6 since the pre-industrial period and Introduction is projected to double again between 1990 and 2050 (Galloway et al., 2004). The main centres of deposition in the early 1990's are the Eastern United States, Central Europe, India, Southeast Asia and Southeastern Brazil, which are likely to intensify and spread in the future (Galloway et al., 2004). Here, we use the global annual, natural and anthropogenic, deposition rates of NH x and NO y by Dentener (2006) for the time 5 slices of 1860, 1993 and 2050, which are regridded from the original 5 • × 3.75 • map to UVic ESCM's resolution of 3.6 • × 1.8 • and interpolated between time slices in order to obtain annual deposition rates for the years 1860-1999; deposition rates of the year 1860 are used for the period 1800-1859. 10 Mineral N in the UVic ESCM can be lost from the soil via leaching:

Nitrogen loss
and is related to the runoff Q D (m a −1 ) and the concentration of available NH + 4 and NO − 3 (kg N m −3 ). The available N depends on the sorption factor b NH 4 and b NO 3 (Table 3) 15 and makes NH + 4 less available for leaching than NO − 3 due to the cation binding capacity of soils. Gaseous losses of N are not considered in the current version.

Allocation of N to plant organs
Nitrogen is allocated to leaves, roots and wood: the allocation of N to wood follows 20 a fixed C/N ratio of 330 kg C (kg N) −1 for broad-leaved and needle-leaved trees and for shrubs (Sitch et al., 2003). While the C/N ratio of wood is fixed, the C/N ratios of leaves and roots vary between a minimum and maximum value ( Table 2). The change in total vegetation N (N V ) is estimated by: where NH UP 4 and NO UP 3 are the N that the plant takes up in form of NH + 4 and NO − 3 (Eqs. 17 and 18) and N LF is the N lost via litterfall (Eq. 2). Vegetation N (N V ) is spread 5 over the three plant N pools by first allocating N to wood following the fixed C/N ratio, then allocating a minimum amount of N to roots to meet the maximum C/N ratio and then adding the remaining N to the leaf N pool. If there is more N available than needed to fill up the N leaf pool and CN Leaf < CN Leaf,min then we set CN Leaf = CN Leaf,min and any excess N is added to the roots. In that way, the N requirements for leaves are 10 met before the ones for roots and only if there is sufficient N available, root N levels increase. If CN Root < CN Root,min then we set CN Root = CN Root,min and any excess N is added back to the NO − 3 pool and subtracted from the uptake. If both, CN Leaf and CN Root are at their minimum level, the plant N status is optimal for plant growth. The reason for choosing this setup is to allow flexible root and leaf C/N ratios in order to 15 avoid immediate N deficiency stress when enhancing C acquisition rates. It has been shown that root C/N ratios (Pendall et al., 2004;Gai-ping et al., 2006) as well as leaf C/N ratios (Liu et al., 2005) can increase in Free Air CO 2 Experiments experiments (FACE), though the interdependence between changes in root and leaf C/N ratios still needs investigation. 20 Under N limitation in the model, i.e., there is not enough N available to meet the requirements CN Leaf > CN Leaf,max , then, first, leaching is reduced by up to 100 % and if more N is needed, then immobilisation is reduced by up to 50 % and added to the plant uptake. In both cases, the NO given the current C stocks in the plant biomass, the minimal requirement for N to fulfil the C/N ratios are always met.

N effect on photosynthesis
One of the determining factors in the rate of photosynthesis is the activity of the enzyme Rubsico, which correlates well with leaf N concentration (e.g., Evans, 1983). This 5 relationship is reflected in the UVic ESCM by linking the maximum rate of carboxylation by Rubisco, V c,max (mol CO 2 m −2 s −1 ) to leaf N, or in this case to the inverse of the leaf C/N ratio.
We substitute the top canopy leaf N concentration (n l ) in the original equation (Cox 10 et al., 1999, Eq. 21), which is fixed for each PFT, by the inverse of the calculated average canopy leaf C/N ratio (CN leaf ). The constant of proportionality λ is 0.004 for C 3 and 0.008 for C 4 PFTs (Cox et al., 1999). Equation (27) means that photosynthetic activity and therefore plant productivity is reduced when CN Leaf increases, but in the model it never goes towards zero because of N limitation as CN Leaf has a maximum 15 value (Table 2). We opt for using the average canopy leaf C/N ratio rather than top leaf C/N ratio as done in Cox et al. (1999) as there is evidence that it is not the C/N ratio of leaves that varies within a canopy, but the leaf mass area per unit area and with it the N mass per unit area (Hollinger, 1996). Hence, as long as N concentration is expressed in kg N (kg C) −1 , i.e. as inverse of the C/N ratio, as done in the UVic model, 20 we can assume that there is no need for varying C/N ratios of leaves within the canopy (Thornton and Zimmermann, 2007).

Model simulations
The model is integrated either with CN feedbacks switched on (UVic CN-coupled mode) or with both, the vegetation and soil CN feedbacks, switched off (UVic C-only mode). Introduction To switch off the soil CN feedbacks, the term (1 + ξ[N min,av ]) is omitted from Eqs. (5) to (8) and to turn off the vegetation CN feedback, the leaf N concentrations (n l ) given as inverse (1/n l ) in Table 2 are used in Eq. (27) instead of the calculated leaf C/N ratios (CN leaf ). Values for n l in the UVic C-only mode are set so that a comparable GPP between the C-only and the CN-coupled mode is achieved.

5
Both model versions are spun-up until the soil C pool changed less than 0.5 % per century. The models are then integrated transiently from 1800 to 1999 in either the CNcoupled mode or the C-only mode. We use the usual set of forcing for the UVic ESCM (orbital parameters, solar constant, volcanic activity, sulphate concentrations, land ice cover, atmospheric CO 2 concentrations, non-CO 2 greenhouse gas concentrations and land use change), the only new forcing for the UVic-CN version is the nitrogen deposition derived as described in Sect. 2.2. Nitrogen deposition affects the C cycle in the model only when the CN feedbacks are switched on.
Five experiments are conducted, three with the CN-coupled version (E1-E3), two with the C-only version (E4-E5). The experiments are listed in Table 5 and are sim- 15 ilar to other studies that used radiatively coupled/uncoupled runs (e.g., Zaehle et al., 2010b). The runs are transient runs for the time period 1800 to 1999.
A fully forced simulation is conducted for the CN-coupled version (FF1) and the Conly version (FF2), in which all of the relevant forcings are used. The experiments also included radiatively coupled simulations, where the climate experiences the radiative 20 effect of increasing atmospheric CO 2 concentrations, but the vegetation experiences no CO 2 fertilisation effect due to atmospheric CO 2 concentrations being held constant at the 1800 level (E1 and E4) and radiatively uncoupled simulations, where the climate sees a constant CO 2 concentration at 1800 levels, but the vegetation experiences the transient CO 2 concentrations (E2 and E5). The third experiment for the CN-coupled version held N deposition constant at 1800 levels, whereas in E1 and E2 they are transient. Sensitivities of the terrestrial C pool to CO 2 concentration (β L ) and air temperature (γ L ) are calculated following Bonan and Levis (2010, Eqs. 2a and 3a): and are changes in land C in the different experiments (Table 5), ∆C A is the change of atmospheric CO 2 concentration and ∆T L the change in 2 m land surface temperature between the period 1800-1804 and the period 1995-1999. Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the vegetation is the wood CN ratio as wood contributes between 70-94 % of tropical plant biomass (Vitousek et al., 1988;Malhi et al., 1999 (Fig. 2b) and therefore lower than the N values shown in the IGBP-DIS database (Fig. 2c). Since 15 soil N content in the model is tied to soil C content via a fixed CN ratio, lower C stocks in the UVic-CN model (Fig. 7) lead to lower N stocks compared to the IGBP-DIS data (frequently over 30 kg C in the boreal zone) (Global Soil Data Task Group, 2000). The lack of permafrost and peatlands in the UVic-CN model is the likely reason for the underestimation of boreal C stocks (Wania et al., 2009) Post et al. (1985); Batjes (1996) (95-140 Pg N).

Results and discussion
In general, tropical forests show the highest simulated vegetation C/N ratios (Fig. 3a), with some extra-tropical exceptions such as in Chile, Mexico and South Africa where 25 both tree PFTs occur. C/N ratios in temperate forests in North America are between 200 and 250 and decrease northwards to 150-200 kg C (kg N) −1 , a value range also seen for the Eurasian boreal zone. Litter C/N ratios follow the vegetation C/N ratio closely (Fig. 3b) and correspond well to observed values (White et al., 2000), who 83 Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | reported litter C/N ratios of 55 for deciduous broadleaved forests, 93 for evergreen needleleaved forests, 45 for grass and 75 for shrubs. Both the modelled litter and vegetation C/N ratios are lower in areas where the percentage of leaf and root biomass is higher. In boreal areas simulated leaves and roots constitute about 10 % of total biomass, whereas in some tropical regions leaves and roots constitute only 3 % of the 5 total biomass compared to observed 4.5 % in northern conifer ecosystems and 1.9 % in tropical closed forests (Vitousek et al., 1988). The difference in the percentage of leaves and roots explains the higher C/N ratios in vegetation and litter in the tropics.
In the absence of reliable observation based estimates of NO − 3 in desert areas, the evaluation of simulated N pools from any model remains difficult. Ammonium and ni-10 trate pools (Fig. 4) show some similarity to the results shown in Xu-Ri and Prentice (2008, Figs. 12 and 13). Both Xu-Ri and Prentice (2008) (2000). Without the knowledge of how much NO − 3 is stored in desert areas, it will remain difficult to evaluate the simulated mineral N pools in any model.

Nitrogen fluxes
Global annual rates of plant N uptake (873.2 Tg N a −1 ) are lower than estimates from 25 other models (1002 to 1126 Tg N a −1 ) (Xu-Ri and Prentice, 2008;Yang et al., 2009;Zaehle et al., 2010b). As discussed above, the vegetation CN ratios in the UVic-CN 84 model are higher compared to other models, which reduces the demand for plant N uptake and explains the lower uptake rates. Generally, uptake rates in the UVic-CN model range from 3-9 g N m −2 a −1 in temperate and boreal regions to 3-15 g N m −2 a −1 in tropical regions (Fig. 5a). Higher values of 15-23 g N m −2 a −1 can be found in tropical grasslands (in this case in sub-Saharan Africa, India, Southern Brazil and Northern 5 Australia). Nitrogen uptake rates in the ORCHIDEE-CN model are estimated to be 4, 8 and 13 g N m −2 a −1 in boreal, temperate broadleaved and tropical regions respectively with maximal uptake rates of 30 g N m −2 a −1 found in grasslands (Zaehle et al., 2010b).
The spatial distribution of leaching is similar to that of runoff with highest values in the tropics and negligible in drier and colder regions (Fig. 5b). Global annual N losses

Effect of CN feedbacks on carbon pools and fluxes
Before integrating the UVic model versions transiently, experiments are used to readjust the leaf nitrogen values (n l ) used in the C-only version in place of 1/CN leaf in Eq. (27) in order to achieve a comparable annual gross primary productivity (GPP) in 25 both model versions for the pre-industrial simulations ( beginning of the transient simulation, by the 1990s the GPP of the C-only version is 133.1 Pg C a −1 and therefore higher than in the CN version (129.6 Pg C a −1 ) ( where 0.012 is a factor to convert units of mol CO 2 m −2 s −1 to kg C m −2 s −1 . R m and R d are the maintenance and dark respiration, respectively, S the soil moisture, and 15 N root , N stem , N leaf are the N contents in root, stem and leaf in kg N kg C. In UVic C-only, as in the original TRIFFID code, N contents in root and sapwood are calculated in relation to the leaf N content Cox et al. (1999, Eqs. 31-33). In UVic-CN however, N contents are based on explicit N fluxes into and out of the plant tissues and N levels in the plant depend on the availability of N in the environment. This leads to 20 lower sapwood N contents in the CN-coupled version, which leads to lower autotrophic respiration rates, which in turn results in higher NPP values. The effect of this can also be seen in the zonal averaged NPP values in Fig. 6 (Arora and Matthews, 2009). Nevertheless, the high NPP/GPP ratio of 0.58 in the CN-coupled version points towards the necessity of re-visiting the autotrophic respiration calculation in TRIFFID. The increase in NPP in the UVic C-only version from 1800 to 1999 is 19 %, whereas 5 it is only 12 % for the CN version. It is still uncertain, how much of an CO 2 fertilisation effect we can expect. Early results from the FACE experiments suggest an increase in productivity by 23 ± 2 % for approximately 550 ppmv CO 2 (Norby et al., 2005), which is also reproduced in a modelling study (Hickler et al., 2008). More recent results from one of the Free Air CO2 Experiments (FACE) show that the initial increase of 10 NPP of deciduous sweetgum trees due to enhanced CO 2 wore off after an initial 4-5 yr period and dropped from an enhancement effect of 24 % in 2001-2003 to 9 % in 2009, which is hypothesised to be caused by N-limitation (Norby et al., 2010;Garten Jr. et al., 2011), supporting the kind of N limitation seen in models. However, the decrease in NPP in deciduous, and therefore more N-demanding sweetgum, is not Over the 1980-1990 period, the zonally averaged GPP values from both model versions are quite close to each other (Fig. 6). The main difference between the two model versions arises between the latitudes 30 • S and 60 • S. The CN-coupled model 20 simulates lower average GPP values for this part of the Southern Hemisphere than the C-only version, which fit the observed, data derived median GPP values from Beer et al. (2010) better. Both UVic model version simulate a lower, but wider peak around the tropics than the data by Beer et al. (2010) (Fig. 6). The global simulated GPP of the UVic-CN version is in good agreement with the most recent, observation-based, 25 estimate of 123±8 (Beer et al., 2010).
Vegetation carbon stocks are driven by wood density and are highest in tropical forests followed by temperate and boreal forests in the UVic-CN version (Fig. 7  top  Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 4-12 kg C m −2 for temperate and boreal forests, which is close to observations that show mean values of 12.1 kg C m −2 for tropical and 5.7-6.4 for temperate and boreal forests (Malhi et al., 1999). Soil carbon stocks are highest in cold regions where decomposition rates are low (Fig. 7 bottom left). The difference between the UVic-CN and the UVic-C version are shown on the right hand side of Fig. 7. The largest vegetation 5 carbon gains in UVic-CN compared to UVic-C are in the range of 3-5 kg C m −2 found in the circumpolar region, tropical regions gain less C. Soil carbon gains are highest in cold areas including the circumpolar region and the Tibetan plateau area, which decreases the discrepancy between modelled soil C stocks in the boreal region in UVic-CN and observations, e.g., Malhi et al. (1999) report an average of 34.3 kg C m −2 10 for boreal soils and the IGBP-DIS data shows high abundance of gridcells with a soil C content of 30 kg C or more (Global Soil Data Task Group, 2000). Soil C in extra-boreal regions in the UVic-CN version is generally higher than in the UVic-C, which brings the model results closer to observations, which are in the range of 9.6 and 12.3 kg C m −2 for the temperate and tropical regions, respectively (Fig. 7c,d). 15 Total global vegetation stocks are higher in the UVic-CN version than in the UVic-C version due to higher NPP (Table 6), for which the reasons are discussed above. This is in contrast to Zaehle et al. (2010b) and Bonan and Levis (2010), who found a decrease in vegetation productivity when including CN interactions in their models and therefore lower vegetation C stocks. The soil C stocks are higher in the UVic-CN 20 version than in the C-only version, which is in agreement with Bonan and Levis (2010) but in disagreement with Zaehle et al. (2010b). In our case, soil C stocks increased when including CN interactions because the consideration of mineral N concentration in Eq. (5) leads to a faster humification process than when not including CN interactions. Higher humification rates mean more input into the soil C pool while at the same 25 time soil decomposition is not influenced by the mineral N concentration (Eq. 13) and does therefore not increase. The faster humification process and with it the faster litter decomposition (Eq. 7) lead to a smaller litter C pool in the UVic-CN version (

Historical changes of C fluxes and pools
In Fig. 8 we compare how carbon fluxes and pools in the CN-coupled and the C-only version have evolved over the 19th and 20th century. GPP values of both versions increase over the last two centuries, remaining comparable up to the 1880s, but diverging from then on with the C-only version increasing faster than the CN-coupled 5 version (Fig. 8a). The point of divergence coincides with a change in radiative forcing caused by volcanic eruption of Krakatoa in 1883. Around 1883, both model versions show an increase in GPP followed by a decrease, though the decrease for the C-only version is much smaller than the decrease for the CN-coupled version. This difference in GPP fluctuations following volcanic eruptions can also be seen between 1800 and 10 1840. After each of the volcanic events, GPP first increases and then drops again. In case of the C-only version, GPP rates drop back to the value observed before the volcanic event, but the GPP in the CN-coupled version shows a much stronger decrease after a preceding spike. The reason why the UVic ESCM simulates an increase in GPP directly after volcanic 15 eruptions is twofold: On the one hand, air temperature drops after volcanic eruptions due to an increase in aerosols (e.g., Harris and Highwood, 2011), which cause higher carbon assimilation rates in TRIFFID/MOSES (Cox et al., 1999, Eq. 15). On the other hand, soil moisture increases due to a decrease in evaporation that exceeds the decrease in precipitation. Following (Cox et al., 1999, Eq. 18), an increase in soil mois-20 ture leads to higher carbon assimilation rates. The difference between the CN-coupled and the C-only version arises from the accumulation of C biomass through increased GPP; the CN-coupled version lags behind in acquiring enough N to maintain stable C/N concentration ratios within the plant tissue and the increase in C/N ratios affects photosynthesis negatively (Eq. 27). Global average C/N ratios in leaves increase dur-25 ing each volcanic event and return to pre-event values afterwards. When comparing NPP (Fig. 8b) to heterotrophic respiration, HR (Fig. 8c) we can see complementary patterns, i.e. when NPP shows a positive anomaly after a volcanic eruption, HR shows a negative one due to the opposite effect of temperature on those two variables. Lower 89 temperature increases GPP and hence NPP in the UVic model, but it decreases soil and litter respiration rates. The CN-coupled and C-only versions show very similar trends up to 1960, when they start diverging for both NPP and HR due to the higher GPP values. The total land C pool shown in Fig. 8f is determined by the soil and litter C pools 5 (Fig. 8e) which are much larger than the vegetation C pool (Fig. 8d).

15
The soil and litter C pool anomaly for the CN-coupled model is higher during the the first 150 yr of the simulation, but even though it increases after 1960, it does not increase as fast as the C-only version and the anomaly by the year 1999 is therefore higher for the C-only version (72 Pg C) than for the CN-coupled version (60.5 Pg C) (Fig. 8e). This pattern also dominates the total C shown in Fig. 8f, which shows that be-20 tween 1800 and 1960, the terrestrial biosphere gained up to 20 Pg C andbetween 1960 and1999 it gained another 26-47 Pg C depending on the model version. A difference of 21 Pg C in total C accumulation by the year 1999 compares well to the ORCHIDEE model, in which the C-only version gained 25 Pg C more than the CN-coupled version in the period 1860-2000 (Zaehle et al., 2010a).

25
The climate sensitivities for the period 1800-1999 β L and γ L are listed in Table 7. A previous version of the UVic model had a β L value of 1.2 Pg C ppmv −1 and a γ L value of −98 Pg C K −1 (Friedlingstein et al., 2006). The current version of the UVic model simulates a β L value of 1.2 for the C-only and 0.8 Pg C ppmv −1 for the CN-90 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | coupled version. The γ L sensitivity changes from −98 in the previous UVic version to −107 Pg C K −1 in the C-only version described here due to the inclusion of the litter pool. The CN-coupled version shows a reduced sensitivity to temperature of −91 Pg C K −1 .
The reduction of the β L sensitivity when including CN interactions is also found in all 5 other models listed in Table 7, though the UVic model shows the smallest reduction in β L . The C sensitivity to temperature of the UVic model is the highest one compared to other models that have included CN interactions but is lower than the Hadley Centre model in the C4MIP study (Friedlingstein et al., 2006). The change of γ L when including CN interactions ranges from a decrease in the negative value (UVic ESCM, 10 ORCHIDEE) to a switch from a negative to a positive value (CLM4, IGSM) ( Table 7). The strongest effect on γ L has been found in the IGSM model which is integrated up to the year 2350, which could have an effect on the resulting γ L values. However, Zaehle et al. (2010a) find that γ L values stayed relatively stable at least over the 21st century. 15 In order to evaluate the sensitivity of the land C uptake to the introduction of N into the UVic model, we compare the spatial distribution and the zonal averages of Net Ecosystem Production (NEP), i.e. the CO 2 flux from the atmosphere to the land, of the UVic CN-coupled model to the UVic C-only model under different forcing regimes for the 1990s (Fig. 9). The C-only version of the UVic ESCM simulates a strong C sink 20 in tropical regions and a less strong C sink for the extra-tropical regions for the 1990s under the "Fully Forced" experiment ( Fig. 9c). Almost all of the Amazon, tropical Africa and parts of Southeast Asia take up C at a rate of over 20 g C m −2 a −1 . A large proportion of these tropical C sinks disappears in the UVic CN-coupled version (Fig. 9b), but the boreal C sinks remain.

Printer-friendly Version
Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | between 40 • N and 60 • N; here, a drop from about 0.2 Pg C a −1 in the C-only model to < 0.1 Pg C a −1 in the CN model is observed. A reduction in NEP in the tropics and the mid-northern latitudes are also observed in the MIT CN-TEM model when compared to the MIT C-TEM model . When comparing our results to the ones from the ORCHIDEE model, we find two main differences though: first, the zon-5 ally averaged NEP in both ORCHIDEE versions for the 1990s is larger in mid-latitudes (> 0.4 Pg C a −1 ) than in low latitudes (< 0.4 Pg C a −1 ), and second, the zonally averaged NEP south of 50 • N in the ORCHIDEE model is higher in the CN version than in the C-only version (Zaehle et al., 2010b), which contrasts the results of the UVic ESCM. This difference indicates that the N effect on the C cycle is stronger in the UVic 10 ESCM than in the ORCHIDEE model. In the "Fully Forced minus Vegetation" experiment, where the vegetation experiences constant atmospheric CO 2 concentrations at 1800 levels, whereas climate and N deposition are transient, almost all of the land area turns into a C source (Fig. 9d-f). In the C-only version the Amazon is a stronger C source than in the CN-coupled version but the opposite is true for Southeast Asia. When comparing the "Fully Forced" to the "Fully Forced minus Vegetation" experiments a larger decrease of tropical NEP is observed in the C-only version than in the CN version, bringing the NEP values of the C-only and CN version closer to each other between 50 • S and 20 • N (Fig. 9d). Tropical NEP in the C-only model decreases by 0.39-0.43 Pg C a −1 per 10 • -latitude band, whereas the 20 CN version decreases only by 0.20-0.23Pg C a −1 . The stronger reduction in NEP in the C-only compared to the CN version has also been found in the ORCHIDEE model (Zaehle et al., 2010b). In our model, this difference is due to a reduction in NPP that averages 16.4 % between 30 • S and 30 • N in the C-only version but only 12.9 % in the CN-coupled version.

25
The "Fully Forced minus Climate" experiment, basically a CO 2 fertilisation experiment, results in an increase of NEP compared to the "Fully Forced" experiment in the C-only version between 30 • S and 60 • N, with the strongest increase around the equator of 0.12 Pg C a −1 per 10 • -latitude band (Fig. 9g-i). In contrast, the CN-coupled version 92 Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | does not show increases of NEP around the equator, but rather in mid-latitudes, i.e. 10 • S-30 • S and 40 • N-50 • N.
The last experiment, "Fully Forced minus N deposition" is similar to the Fully Forced experiment, only that N deposition is excluded from the model. As the zonal average shows, the simulation without N deposition has a reduced C sink strength between 5 10 • S and 60 • N, in the areas where N deposition happens (Dentener, 2006). Zaehle et al. (2010b) found that the latitudes between 35 • N and 65 • N are most affected by N deposition in the ORCHIDEE-CN model. The sensitivity of the UVic model to N deposition in the tropics together with the lack of increase in NEP in the "Fully Forced minus Climate" experiment is likely to be related to changes in the Amazon basin. 10 Throughout the figures, the Amazon basin is different from other tropical regions with lower ammonium and nitrate concentration (Fig. 4), lower plant uptake (Fig. 5a), partially in GPP and NPP (Fig. 6). One difference between the Amazon and the rest of the tropics that we have found is a much higher simulated soil moisture. Higher soil moisture leads to higher runoff and despite lower ammonium and nitrate concen- 15 trations, leaching rates of mineral N are about the same in the Amazon as in other tropical regions (Fig. 5b), which means that in our model relatively more mineral N is lost via leaching in the Amazon than in other regions. Lower soil ammonium and nitrate concentrations cause lower plant uptake rates, leading to higher leaf C/N ratios in the Amazon, which lowers photosynthesis. 20 Global numbers for NEP are given in Table 8. The difference in NEP between the Conly (FF2) and the CN-coupled (FF2) is 0.74 Pg C a −1 in the "Fully Forced" simulations.
This drop in NEP is simulated despite the increase in NPP in the CN-coupled version discussed above; the lower NEP is caused by higher soil and litter respiration rates reducing the C sink strength in the CN-coupled version. A drop of 0.7 Pg C a −1 from Introduction The model experiments "Fully Forced minus Vegetation" result the land to become a strong C source in both model versions. The land in the C-only version represents a slightly stronger C source (−0.63 Pg C a −1 ) than the one in the CN-coupled version (−0.60 Pg C a −1 ). The difference between the CN-coupled version and the C-only version in the UVic model is smaller than that found by Zaehle et al. (2010b, approximately 5 −0.7 Pg C a −1 for the ORCHIDEE-C and −0.3 for ORCHIDEE-CN). When the model is integrated in the "Fully Forced minus Climate" mode, we observe a larger increase in global NEP in the C-only model (from 1.57 to 2.17 Pg C a −1 ) than in the CN-coupled version (from 0.83 to 1.05 Pg C a −1 ) compared to the "Fully Forced" simulations ( Table 8 (Table 8). A similar non-linearity has been found in the ORCHIDEE-CN by Zaehle et al. (2010b), who compared their "Fully Forced" version to the "Fully Forced minus Vegetation" + "Fully Forced minus Climate" (all three versions 20 are without N deposition) and got a difference of 0.4 Pg C a −1 , i.e. the NEP of the "Fully Forced" is 0.4 Pg C a −1 higher than the arithmetic sum of the other two simulations. In our case, that difference is 0.38 Pg C a −1 .

Conclusions
The or leaching), which cannot be eliminated until we have gathered better data. The UVic ESCM agrees with some models and disagrees with others but shows in general a similar behaviour to other CN-coupled models; where disagreement happened, we found an explanation for the different behaviour between our model and others. One of the main attributes of the UVic CN-coupled ESCM is that the inclusion of N leads to an in-5 crease of the NPP:GPP ratio which is caused by a reduction in autotrophic respiration due to its relationship with plant N content. Even though the current formulation of the autotrophic respiration served the C-only version well, the high NPP:GPP ratio suggests that it is unrealistic for a CN-coupled version and should be replaced in a future version.

10
The high soil moisture contents in the Amazonian basin are the reason why this region stands out compared to other tropical regions. Higher soil moisture contents lead to faster microbial processes, causing low mineral N concentrations. The Amazonian basin shows lower productivity values and a nearly neutral NEP and only minor changes in our forcing experiments. The overall cause for the high soil moisture values 15 in the Amazon is a bias towards high precipitation in that region in the UVic ESCM.

Variable
Units Description  Table 2. List of PFT-dependent parameters used in UVic ESCM-CN: Leaf base turnover rate (η 0 leaf ), root turnover rate (η root ) and wood turnover rate (η wood ) are all taken from the UVic-ESCM, and retranslocation of N before leaf abscission (r leaf ) and maximum N uptake rate (ν max ) are taken from Gerber et al. (2010). Minimum and maximum C/N ratios for leaves (CN Leaf,min , CN Leaf,max ) and roots (CN Root,min ,CN Root,max ) for each PFT are chosen as follows: CN Leaf,min are the inverted maximum leaf N concentrations used in the previous UVic ESCM (Meissner et al., 2003) with the exception of the value for C3G which is raised from 18 to 25. CN Leaf,max are allocated in order to allow a wide range of possible CN ratios, CN Root,min and CN Root,max are set to be higher than CN Leaf,min and CN Leaf,max (White et al., 2000). Average leaf nitrogen concentration, n l , is used in the C-only version to calculate Rubisco activity V c,max . BT = broadleaved trees, NT = needle-leaved trees, C3G = C 3 grasses, C4G = C 4 grasses, SH = shrubs.  Available mineral N pool [N min(av) ] [NO 3(av) ] + [NH 4(av) ] kg N m −3 Available mineral N concentration Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Table 5. Description of the UVic ESCM experiments and the forcings used. "FF" are the fully forced simulations and "E" are the experimental simulations in which the forcings are modified. In the forcing column, "FF" indicates a fully forced model, using transient CO 2 concentration for the vegetation and climate and N deposition, "FF minus Vegetation" means that the CO 2 concentration for the vegetation is held constant, "FF minus Climate" means that the CO 2 concentration for the climate is held constant and "FF minus Ndep" means that the N deposition is held constant. "CN" indicates the use of the UVic CN-coupled version and "C-only" indicates the use of the UVic C-only version, "CO 2 for the vegetation/climate" gives the year or period that is used and Ndep gives the year or period of natural and anthropogenic N deposition.   (2010) Table 7. Climate sensitivities β L in (Pg C ppmv −1 ) and γ L in (Pg C K −1 ) of the land C pool in the UVic C-only and CN-coupled version compared to other models.  . Leaf, stem and root N content depend on the size of the C pools and fixed C/N ratios. The UVic inherent leaf, root and stem turnover rates are used to calculate litterfall with the only modification that N in leaves is partially reabsorbed before abscission. The litterfall goes first into the litter pool, which is partially decomposed and enters the NH 4 pool and part of it is humified and enters the soil N pool. The soil N pool is mineralised and adds to the NH 4 pool. Ammonium is turned into NO −