Introduction 

Forest ecosystems worldwide are currently acting as carbon (C) sinks ([70]). Several factors may, however, influence the magnitude and direction of the net C balance, including recovery from historical land use (e.g., abandoned agricultural land reverting to forested land), increases in atmospheric CO₂ concentration and nitrogen (N) deposition, and climate change ([96], [108]). Nonetheless, while much research has been done to understand the controls on net ecosystem C balance ([111], [89], [85]), we know little about the controls on C sink partitioning between plant biomass and soil organic matter (SOM) pools. Soils may store C for long periods of time ([53]), accumulating on average three times the C in terrestrial vegetation ([76]). On the other hand, more N is required per unit of C stored in soil as compared to plant biomass ([118]). Hence, while an allocation to SOM may increase C sequestration in the long term, a preferential allocation to plant biomass is a more nutrient-efficient C sequestration process in the shorter term.

Studying ecosystem C sink partitioning is challenging due to the difficulties associated with quantifying the different ecosystem fluxes. Especially complex is the assessment of rapid and small changes in SOM which are linked to the balance between microbial respiration and plant inputs, including both litter and root-derived C ([98]). Thus, belowground C allocation and subsequent C dynamics are still far from being accurately quantified and understood ([74], [114]). Root C inputs have been shown to influence soil C sequestration, but both the magnitude and direction of this root effect are variable ([44], [72], [7], [82], [20]).

A robust definition of net ecosystem production (NEP) should be based on a full ecosystem mass balance ([81]), which accounts for both plant and soil sinks. When it is flux-based, NEP is defined as the difference between ecosystem-level gross photosynthetic gain of C (gross primary production, GPP) and ecosystem respiratory losses (R_eco). Alternatively, NEP (g C m^-2 y^-1) can be expressed as ([6] - eqn. 1):

In deciduous forest ecosystems, ΔC_biomass is the annual change in plant biomass (wood, branches, coarse roots), and ΔC_soil is the annual net change in soil organic C (SOC) stock. In this equation, litterfall and fine root turnover are considered as soil C input and therefore contributing to the ΔC_soil (see eqn. 2).

Net ecosystem productivity can be directly determined using eddy covariance techniques starting from net ecosystem exchange (NEE = ’€’NEP - [3], [2]). Plant biomass changes are usually estimated via a combination of repeated inventories and allometric relationships ([10]). On the other hand, direct SOC determination methods are generally unable to quantify ΔC_soil in the short term ([98]), and, at annual timescales, alternative methods are required to estimate soil C changes.

Considering that the dissolved organic C (DOC) is typically negligible, representing around 1% of forest NPP ([59]), ΔC_soil can also be written as (eqn. 2):

where Input_litter is the above-ground litterfall (i.e., leaves, branches, wood, etc.), Input_roots is the root-derived C input (i.e., exudates, root slashing and turnover), R_{C-rhizosphere} is the rhizosphere respiration of root-derived C, and Rh is the heterotrophic respiration. Litter input is conventionally measured by litter traps, while wood input is measured using repeated sampling ([34]), and rhizosphere and heterotrophic respiration can be estimated by a variety of methods (e.g., trenching, girdling, isotopes), as reviewed by Subke et al. ([104]) and Kuzyakov ([51]). The largest challenge is estimating gross root inputs. However, methods exist to estimate net annual root-derived C input (Net-C_root), which is the difference between Input_roots and R_{C-rhizosphere} (eqn. 3):

Different tracer methods have been used to date to estimate Net-C_root, such as pulse labeling, continuous labeling, and ¹³C natural abundance ([50]). The latter uses the difference in the stable C isotope composition of native SOM and new plant-derived organic matter to quantify Net-C_root.. When natural isotope abundances do not allow the use of this approach, distinct C isotope signatures in the soil organic C (SOC) pool and plant-derived organic matter can be obtained in manipulation experiments, by growing C₃ plants (δ¹³C of approximately -27‰) in soil with organic matter derived from C₄ plants (δ¹³C of approximately -12‰) or vice versa. This approach has been successfully applied in pot ([43], [113]) and field studies ([40], [12]) and was used in this investigation.

Net-C_root, combined with aboveground inputs to the soil (litter and dead wood), also provides interesting information about soil C dynamics. For soils at steady-state (ΔC_soil=0), the sum of Net-C_root and aboveground inputs is the amount of C that replaces SOC decomposition, thus becoming a measure for SOC turnover. For soils which are net C sinks (ΔC_soil>0), this sum exceeds SOC mineralization and a fraction of it enlarges the SOC pool, thus leading to soil C sequestration. In this context, for soils which are net C sinks, the ratio between ΔC_soil and Net-C_root + aboveground inputs indicates the fate of C input: the higher the ratio, the larger the contribution of fresh C to soil C sequestration. The opposite is true for soils that are net C sources (ΔC_soil<0).

Root C input rates vary considerably depending on tree species, mychorrhizal associations and environmental factors ([54]), with values of up to 40% of net assimilated C being reported ([112]). According to the microbial efficiency-mineral stabilization (MEMS) framework ([13]), the fraction of Net-C_root inputs sequestered in the soil depends on the efficiency of decomposers to convert C into bio-products as compared to tha amount of C lost as CO₂ ([99]) and on soil matrix interactions ([100], [48]). Soil organic matter mineralization is driven by both substrate stoichiometry and microbial demand for resources ([67], [38]): when N is limiting, microbes use labile substrate to mineralize recalcitrant SOM ([69], [14]). Root exudates can thus prime SOM decomposition ([56], [5], [26]). Clearly, root-derived soil C inputs can either stimulate soil C sequestration or, conversely, induce priming with consequent losses of stabilized SOM but likely enhancements in N availability, which in turn can stimulate plant growth. The key factors determining the direction (and magnitude) of this effect are, however, not yet clear. Understanding the fate of root-derived C, and its effects on N dynamics and ecosystem C sequestration, is relevant from an ecological perspective and is also an urgent challenge to address, particularly in the context of global changes such as atmospheric CO₂ increase and N deposition.

The aims of the present study were: (1) to obtain an estimate of Net-C_root in six different forest ecosystems; (2) to partition NEP into aboveground tree biomass production and soil C sinks; and (3) to investigate the controls of this partitioning. Specifically, we tested the hypothesis that soil C:N stoichiometry controls ecosystem C uptake (GPP) and sink partitioning (ANPP vs. soil C) across forest ecosystems. To verify if our hypothesis could be generalized to other forests, we tested it on several world forest sites for which ANPP, GPP and soil C:N data were available in the literature.

  Materials and methods 

Study sites

Six forests were considered in the present study. Three sites were in central Italy, two sites in northern Italy, and one in Croatia. All sites were equipped with an eddy covariance tower for mass, momentum and energy ecosystem exchange measurements and can be classified as high fertility sites, according to key soil properties ([114] - see also Appendix 1). Site characteristics and flux data are reported in Tab. 1, while a brief description for each site is given below.

Roccarespampani (42° 24′ N, 11° 55′ E - [11], [107]) is a Turkey oak (Quercus cerris L.) coppice forest at about 235 m a.s.l. in central Italy. Mean annual temperature is 14 °C and mean annual rainfall is 755 mm. Soil is sandy clay Luvisol (which is typically nutrient rich), derived from sedimentary material of volcanic origin and marine deposits, and is moderately acid (pH=5.7), with a total depth > 100 cm ([87]). Cation exchange capacity (CEC) is high, ranging between 19 and 42 meq 100g^-1 in the different soil layers ([107]). The forest has been managed as a “coppice with standards” over the last 200 years, with a rotation cycle varying between 15 and 20 years. Two stands were selected: a 6-year-old coppice (RO1) and a 15-year-old coppice (RO2).

Lecceto (LE - 43° 18′ N, 11° 16′ E) is a Holm oak (Quercus ilex L.) coppice with a rotation period of 18-20 years at about 300 m a.s.l. in central Italy. Holm oak represents 81% of the total tree canopy; others species include Arbutus unedo L., Juniperus communis L., Quercus pubescens L., Phillyrea latifolia L., Fraxinus ornus L. Mean annual temperature is 13.5 °C and annual average rainfall is 780 mm.

Jastrebarsko (JA - 45° 37′ N, 15° 41′ E; [64], [65]) is a 35-year-old forest in Croatia dominated by pedunculate oak (Quercus robur L.) with 19% of black alder (Alnus glutinosa Haernt.), 14% hornbeam (Carpinus betulus L.) and 9% of narrow-leafed ash (Fraxinus angustifolia L.). Mean annual temperature is 10.4 °C with mean monthly temperatures of -0.2 °C and 20.7 °C in January and July, respectively. Average annual precipitation is 900 mm year^-1, of which around 500 mm falls during the active vegetation period (April-September). Soil is a Luvic Stagnosol with a depth > 100 cm and an acidic pH (4.9) in the upper mineral layer (0-20 cm) that linearly increases to neutral pH at depths > 100 cm. At the beginning of the growing season, the soil drains and water content soon drops below water holding capacity (46% v/v) allowing enough oxygen supply for root growth and substantially increasing nutrient availability in these soils, where nutrient availability can be constrained by high water levels.

La Mandria (LM - 45° 09′ N, 7° 34′ E) is an 80-year-old pedunculate oak-hornbeam forest (Quercus robur L. and Carpinus betulus L.) in northern Italy. Mean annual temperature at the site is 11.6 °C and annual precipitation is 1030 mm. Soil is Typic Fragiudalf with adequate moisture content throughout the year, neutral pH and good CEC (ranging from 17 to 11 meq 100 g^-1 at soil surface and Bh horizons, respectively).

Collelongo (CO - 41° 52′ N, 13° 38′ E; [110], [95]) is an 110-years-old pure beech (Fagus sylvatica L.) forest in northern Italy that has been part of the network of Long Term Ecological Research sites (LTER Italy) since 2006. Mean annual temperature at the site is 7.1 °C and mean annual rainfall is 1188 mm. The soil is a Humic Alisol with volcanic ash also present. Both CEC and N content are high in the different soil layers, ranging from 14.8 to 23.3 meq 100g^-1 and from 4 to 7.3 mg N g^-1, respectively ([73]). Wet N deposition rates in the period 2002-2009 averaged 10.8 kg N ha^-1 yr^-1 ([25])

Net root-derived C input to soil

Net-C_root was quantified using the in-growth core isotope technique, following Cotrufo et al. ([12]). A soil depleted in ¹³C (δ¹³C = -17.22‰) was collected from the USDA-ARS Central Plains Experimental Range located in NE Colorado, USA (40° 49′ N, 104° 46′ W). The soil is classified as a Zigweid soil series (Fine-loamy, mixed, superactive, mesic Ustic Haplocambid), with a pH of 7.4, N content of 1.37 g kg^-1, and P content of 0.5 g kg^-1 ([12]). At this site, plant cover is approximately 75% C₄ grasses, and for brevity we call henceforth this soil as “C₄ soil”. Soil was air-dried prior to being sealed and boxed for shipment to Italy. Upon arrival, the C₄ soil was ground and sieved to 2 mm and well mixed to make a homogeneous soil pool, before using it for in-growth cores and chemical (C% and δ¹³C) analyses as described below.

At each forest site, six cores, made of a 2 mm mesh net (thus allowing the penetration of fine roots) with a diameter of 4 cm and a height of 30 cm, were placed randomly within the eddy covariance tower footprint in October 2006 (2008 for Jastrebarsko) and filled with the C₄ soil to a bulk density similar to the average bulk density for the site. At the top of each core the net was closed to avoid above-ground litter input. Cores were sampled a year later, and the soil from each core was separated into 0-15 cm and 15-30 cm depth layers, except for Jastrebarsko, where the entire 0-30 cm core was considered.

All soil samples were sieved to 2 mm, and root samples carefully removed and washed with deionized water. Root samples were pooled by site and depth, and each samples analyzed in triplicates. Both soil and root samples were oven-dried at 70 °C, pulverized and analyzed for %C and δ¹³C by an elemental analyzer (Flash EA 1112 NC, CE Instrument, Wingan, UK) connected to an Isotope Ratio Mass Spectrometer (IRMS, Delta Plus, Thermo-Finnigan, Bremen, Germany). Prior to C analyses, soil samples were treated with HCl to eliminate carbonates ([36]). The measured δ¹³C values were used to calculate the proportion of new C (f_new, i.e., the Net-C_root), by using a mass balance equation ([17], [12] - eqn. 4):

where δ_soil is δ¹³C of the organic matter of the C₄ soil collected from each core after one year of field incubation, δ_old is the δ¹³C of the organic matter of the C₄ soil measured before incubation, and δ_veg is the δ¹³C of the roots averaged by site and depth. The average δ_veg value across all our sites was -28.11±0.29‰, while variation (standard deviation) within a site was between 0.15 and 0.57‰ at RO1 and RO2, respectively. Knowing the f values for the new C, the soil organic C concentrations (%C), soil depth (D, m), and soil bulk density (σ, kg m^-3), Net-C_root amounts (g m^-2) were computed for all soil samples as follows (eqn. 5):

Estimates of Net-C_root using this method ([12]) rely on the assumptions that: (1) root inputs are the same inside and outside the in-growth bags and are independent of the C₄ soil properties; and (2) that there is no isotopic fractionation during the decomposition of the native SOM or formation of the new SOM from the root tissues. New studies applying this method should test those assumptions, since some fractionation could occur ([39]).

Ecosystem fluxes and primary production

Eddy covariance flux data from all five Italian sites were analyzed for the years 2006-2007 (Tab. 1). Data of net ecosystem exchange (NEE), gross primary production (GPP) and ecosystem respiration (R_eco) at monthly time steps were downloaded from the central Fluxnet database (⇒ http:/­/­gaia.­agra⇒ ria.­unitus.­it/­database/­). Specifically, we used the NEE gap-filled data using the Artificial Neural Network method (NEE_ANN from level 4 dataset - [71]). R_eco was computed according to the short-term temperature response of night-time fluxes ([84]) and GPP values were derived as sum of the absolute values of NEE_ANN and R_eco. At sites where data for the years 2006 or 2007 were incomplete even after gap-filling because of missing weather data, data for 2008 were also included in the analysis for the calculation of annual means. As for the Jastrebarsko site, 2009 eddy flux data were derived from Marjanovic et al. ([64]).

Mean annual temperature (MAT), mean annual precipitation (MAP), and soil C stocks (0-30 cm), as well as changes in wood biomass (stem and branches - ΔC_wood), were derived from ancillary data files available at the central database, updated to 2006-2007 when necessary, or using specific yield tables available at the site (e.g., Jastrebarsko). All data were checked, if necessary updated and completed by site Principal Investigators, who are co-authors of the present study. Changes in root biomass (ΔC_roots) were derived from ΔC_wood using root-to-shoot ratios reported by Mokany et al. ([68]) or using site-specific relationships as in the case of Collelongo and do not include fine root productivity.

ANPP was calculated as the sum between ΔC_biomass and NPP_leaves (foliar net primary production). The latter corresponds to litterfall in the case of broadleaved forests, and was directly measured at the site (i.e., Rocca, Jastrebarsko, Collelongo) or assessed from NPP_wood using biomass expansion factors derived at nearby sites with similar species composition and structure (i.e., La Mandria). In the case of Lecceto, where the dominant species is evergreen (Holm oak), we assumed that the system was at steady state and thus litterfall = NPP_leaves. Then the ANPP: GPP ratio was calculated.

World forest sites data

In order to test if the relationship between ANPP:GPP and soil C:N, observed across our study sites, was generalizable across forest ecosystems, we searched published datasets ([55], [58], [114]) for forest sites that provided the data suitable to our analyses. Twenty-three additional sites were found including ANPP and GPP data, as well as soil C:N (determined for a depth up to 45 cm) were found (Tab. 2). Fertility classification followed Vicca et al. ([114]). More details are given in Appendix 1.

Data analysis

At each site the annual change in net soil C (ΔC_soil - g C m^-2 y^-1) was calculated starting from eddy covariance NEE data and measured changes in aboveground wood biomass (ΔC_wood) and coarse roots (ΔC_roots) by re-arranging eqn. 1 (eqn. 6):

Statistical analyses were performed using the package SIGMA PLOT® 11.0 (Systat® Software, San José, CA, USA). Data were tested for normal distributions, using the Shapiro-Wilk’s test, and homogeneity of variance, and log transformed when necessary. To assess differences in Net-C_root among sites, a one way analysis of variance (one-way ANOVA) was used. Significant treatment (site) effects (P<0.05) were further explored via a treatment (site) comparison using the Least-Squares means test with Tukey’s adjustment for multiple comparisons. For sites where data for 0-15 and 15-30 cm depths were available, a two-way ANOVA with site and depth as fixed factors was also performed.

A correlation analysis between all available variables was performed using the Spearman’s rank method through a correlation matrix in Stata10® (StataCorp®, College Station, TX, USA). For variables that were correlated with p<0.10, linear models were fitted to measured data.

  Results 

Net root-derived C input to soil

Total Net-C_root in the top 30 cm soil layer ranged between 420 g C m^-2 year^-1 at Collelongo and 818 g C m^-2 year^-1 at Jastrebarko (Fig. 1). Mean annual Net-C_root across sites was 606 ± 164 g C m^-2 year^-1 (mean ± standard deviation). A significant difference in total Net-C_root (0-30 cm) was detected among sites (ANOVA, p = 0.007). In particular, post-hoc Tukey’s tests showed a significant difference between Jastrebarsko and Collelongo (p = 0.013) and Lecceto and Collelongo (p = 0.041 - Fig. 1). For sites where data for 0-15 and 15-30 cm depths were available (i.e., all sites except Jastrebarsko), the two-way ANOVA applied showed significant differences among sites (p = 0.004), between depths (p = 0.024) and for site × depth interaction (p = 0.035). No differences among sites were detected at 0-15 cm depth (Tukey’s test: p > 0.05), while Net-C_root at 15-30 cm in Lecceto was significantly different from Rocca1 (p = 0.002), Collelongo (p = 0.003) and Rocca2 (p = 0.021). Climate (i.e., MAT, MAP, soil water content) did not explain significant variability in Net-C_root among the different sites (see Tab. S1 in Appendix 1). Moreover, total Net-C_root was not significantly correlated to soil C:N.

Fig. 1 - Net annual root-derived carbon input (Net-C_root) to soils (0-15 and 15-30 cm) quantified using isotope-labelled (e.g., C₄) soil in-growth cores at the six study sites. Vertical bars indicate standard deviation. Different letters indicate significant difference in total Net-C_root at p < 0.05. For site labels, see Tab. 1. As for the Jastrebarsko site (JA), only total net derived carbon is reported (grey bar).

  Enlarge/Shrink   Download   Full Width  Open in Viewer

Ecoystem C sink partitioning

All six sites were net C sinks with similar NEP values (average NEP was 547 ± 25 g C m^-2 year^-1) but with large differences in annual GPP (Tab. 1). They actively sequestered C both aboveground and in the soil: ΔC_wood represented between 10 and 48% of annual GPP (RO1 and LM, respectively), ANPP (=ΔC_wood + litterfall) was between 13 and 77% (RO1 and LM, respectively) and ΔC_soil was positive for all sites representing between 6 and 20% of annual GPP (CO and RO1, respectively).

In accordance with our hypothesis, GPP and ΔC_soil were correlated to soil C:N (p = 0.0048 and p = 0.07, respectively - Tab. S1 in Appendix 1) and decreased linearly as soil C:N increased (Fig. 2). Moreover, GPP was linearly and inversely related to soil clay content (p = 0.05) but, similarly to ΔC_soil, it was not correlated with either MAT, MAP or stand age. ΔC_wood was weakly related with soil C:N (p = 0.07), while ΔC_wood-to-GPP ratio (Fig. 3.a) and ANPP-to-GPP ratio significantly increased with soil C:N (p < 0.0001 and p = 0.005, respectively - Tab. S1 in Appendix 1). In contrast to GPP, Spearman’s correlation analysis revealed a significant relationship between ΔC_wood and both MAT and MAP (p = 0.05 and p =0.008, respectively - Tab. S1 in Appendix 1). Similarly, ANPP was significantly correlated with MAT and MAP (p = 0.001 and p = 0.0003, respectively). Finally, the fate of root C input appeared to depend on soil C:N stoichiometry, with proportionally more Net-C_root being allocated to C sequestration with decreasing soil C:N (ΔC_soil vs. Net-C_root p = 0.07) and soil C:N was able to explain 40% of the variation in the ratio of ΔC_soil to Net-C_root + litterfall (Fig. 3.b).

Fig. 2 - Relationships between soil C:N stoichiometry and gross primary productivity (GPP - panel A) or net soil C sequestration (panel B) as related to soil C:N ratio. Dashed lines represent 95% confidence interval; the reported R² is the adjusted R².

  Enlarge/Shrink   Download   Full Width  Open in Viewer

Fig. 3 - Relationships between soil C:N stoichiometry and ΔC_wood-to-GPP ratio (panel A), or ΔC_soil : (Net-C_root + litterfall) ratio (panel B). Dashed lines represent 95% confidence interval; the reported R² is the adjusted R².

  Enlarge/Shrink   Download   Full Width  Open in Viewer

World forest sites

The positive relationship between ANPP-to-GPP ratio and soil C:N found across our six study sites was confirmed also when including additional fertile forests from different regions (Tab. 2). In particular, a positive relationship was found between ANPP:GPP and soil C:N for high fertility (sensu [114]) sites (adjusted-R² = 0.64; p = 0.03; Fig. 4). Conversely, no significant relationship was detected for low and medium fertility sites.

Fig. 4 - Relationships between soil C:N stoichiometry and the ANPP:GPP ratio at forest sites with different fertility levels. For the complete list of sites see Tab. 2. Dashed lines represent 95% confidence intervals. The six sites considered in the present study are labeled as reported in Tab. 1. Eucalyptus saligna plantations (Tab. 2) have been averaged by stand age and vertical bars indicate standard deviation. The reported R² is the adjusted R².

  Enlarge/Shrink   Download   Full Width  Open in Viewer

  Discussion and conclusions 

To our knowledge, this study is the first to quantify Net-C_root in a range of forest ecosystems. The measurement of Net-C_root in situ is difficult, thus measured values are lacking and modeled estimates cannot be validated. However, the in-growth core isotope technique has already been shown to allow detection of changes in Net-C_root in CO₂ and climate manipulation experiments ([40], [12]), even though it does suffer from several caveats related to the use of an exogenous soil and high spatial variability. Steingrobe et al. ([103]) reviewed the in-growth core method for measuring gross root growth: a first shortcoming associated with this method is achieving the soil conditions inside the bag similar to the bulk soil. Moreover, soil texture has also been shown to significantly influence rhizodeposition rates ([94]), although it is difficult to determine whether soil texture influenced rhizodeposition rates in our study.

Our estimates of Net-C_root using the in-growth core isotope technique were on average 606 g C m^-2 y^-1, which is higher than values reported by Cotrufo et al. ([12]) for a Arbutus unedo L. coppice in dry Mediterranean conditions, but lower than values reported by Hoosbeek et al. ([40]) for an irrigated and fertilized poplar plantation in central Italy. A possible overestimation of Net-C_root can be also related to the fact that a certain amount of fine root fragments could have passed through the 2 mm sieve. Such an amount is a function of root integrity as affected by plant age and sample processing. Being aware of this possible overestimation and of the above-mentioned limitations associated with the in-growth core isotope technique, in this study we used Net-C_root estimates solely as an indicator of differences in the effect of root-derived C on SOC sequestration through the calculation of the ratio ΔC_soil : (Net root-derived C + litterfall C).

Many factors have been suggested to affect soil C sequestration, including the characteristics of input material, soil texture and mineralogy, climatic factors, and soil nutrient status ([27], [1], [52]). We found that the proportion of root C input resulting in C sequestration at these high fertility sites was related to soil C:N ratio, and soil C sequestration was greater at low C:N (Fig. 2b) therefore confirming our hypothesis. Recently, Manzoni et al. ([62]) suggested a C-to-nutrient stoichiometric control on microbial C use efficiency (CUE), which would increase with increasing nutrient availability. The importance of CUE as a determinant of the fate of plant inputs to soils has also been recognized by other recent studies ([97], [13]) and some models have suggested that low nutrient availability, particularly N, might limit soil C storage through mechanisms that are still not completely understood ([83], [41]). Recently, Kirkby et al. ([47]) hypothesized that the sequestration of C-rich crop residue material into SOM could be improved only by adding supplementary nutrients, as the more stable SOM fraction has more N, P and S per unit of C than the plant material input due to microbial reprocessing. Thus, the increase in soil C sequestration at lower soil C:N values observed in this study may be explained by a higher microbial CUE of root C inputs.

Soil C:N exerted a strong control on GPP across our six forests and GPP increased with decreasing soil C:N (Fig. 2a). This relationship is based on six forest sites and we cannot exclude the possibility that other factors influenced this relation. At the ecosystem scale, variation in global plant productivity across ecosystems has often been related to environmental factors ([24], [86]), but also to nutrient availability ([114]). In this context, Zha et al. ([120]) reported a strong positive relationship between GPP or NPP and total soil N. Across our sites, ΔC_wood and ANPP increased slightly, but not significantly with increasing soil C:N, and showed significant relations with MAT, MAP and stand age, thus confirming previous studies ([15], [42], [88], [37]). How such different behavior between GPP and ΔC_wood or ANPP with respect to soil C:N could be explained? We suggest that this result is due to the lower demand for N by woody tissues (which comprise the largest fraction of the tree and are characterized by very high C:N) as compared to green leaves (which control GPP, and have much lower C:N than woody tissues). As a result of these variations in both GPP and ANPP, the ratio between ANPP and GPP varied substantially among our six forest sites.

Following the distinct patterns of ANPP and GPP versus soil C:N, the ANPP-to-GPP ratio significantly increased with increasing soil C:N (Fig. 2a). At first sight, this seems to contradict the current understanding that partitioning of photosynthates into aboveground biomass increases with increasing nutrient availability across a wide range of forests ([114]). However, all six forest sites had high nutrient availability but, at those sites where soil N presumably exceeded tree demand for wood growth (i.e., sites with low C:N), root C inputs were probably responsible for the higher net soil C sequestration. We speculate that the link between soil C:N stoichiometry and microbial activity controls C sequestration belowground, as well as for the increase in ANPP-to-GPP ratio with increasing soil C:N across the high fertility forests in our dataset. At soil C:N below 15, CUE is expected to be high, and more of the fresh C input is used for microbial products, resulting in the net formation of new SOM. Conversely, when C:N is high, microbes have a low C use efficiency and therefore they respire more of the fresh C inputs and prime SOM decomposition ([26]), which increases N availability and supports a higher allocation of fixed C (GPP) to ANPP. Our observations of increasing ANPP-to-GPP ratio, and the tendency for a decrease in soil C sequestration with increasing soil C:N (Fig. 2b), support this hypothesis.

In order to further test this hypothesis, we analyzed a larger dataset. Also in this case, ANPP-to-GPP ratios were quite variable (average ANPP:GPP = 0.28 with SD = 0.10; Tab. 2) and our analysis confirmed the relationship between ANPP-to-GPP ratio and soil C:N at sites with high fertility (Fig. 4). At sites where overall nutrient availability was low, this relationship did not hold. Variation in partitioning of GPP to ANPP at these sites is probably driven by the need for plants to invest in the nutrient acquiring system (i.e., roots and root symbionts - [114]). When nutrient availability is limited, belowground input by plants may be the dominant control of microbial activity and SOM mineralization ([33], [116], [16], [18]), thereby influencing mineral nutrient availability for plant uptake. Our speculation is also consistent with other recent findings. At the Duke Free Air CO₂ Enrichment (FACE) experiment, the increase in the belowground C flux stimulated microbial activity, accelerated SOM decomposition, and stimulated tree uptake of N bound to this SOM, sustaining ANPP ([21], [22], [75]). Yin et al. ([119]) found that an increase in the release of root exudates into the soil was an important physiological mechanism to sustain growth responses of plants to experimental warming.

At our study sites, soil C:N stoichiometry appeared to be weakly controlled by the soil clay content (p = 0.15 - Tab. S1 in Appendix 1), decreasing with increasing %clay in soil. This is consistent with our knowledge of soil primary organo-mineral particles, which describes clay-associated SOM as the fraction with the highest microbial contribution and lowest C:N ratio ([9], [32]).

In conclusion, our results suggest that a specific site property, such as soil texture, could drive soil C:N stoichiometry which in turn would control ecosystem C uptake and partitioning within forests of high nutrient availability. While GPP strongly and linearly increased with increasing soil N, aboveground tree biomass demand for N appeared to saturate, possibly because of the higher C:N of wood vs. green leaves, and, at high nutrient availability, NPP became limited by other environmental factors. When this occurs, more C is sequestered by soil (Fig. 5), where the high N availability promotes CUE efficiency and new SOM formation.

Fig. 5 - Schematic representation of the proposed dynamics of gross (GPP) and aboveground net primary production (ANPP) with changes in nutrient availability. At high fertility sites (such as the sites considered in the present paper), GPP, ANPP and soil C sequestration (ΔC) changes are controlled by soil C:N stoichiometry. At low soil C:N ratio, C sink allocation shifts from NPP to soil C sequestration.

  Enlarge/Shrink   Download   Full Width  Open in Viewer

  Acknowledgements 

This work was financially supported by the Italian government through the FIRB project “CarboItaly”. We thank all site investigators, their funding agencies, the various regional flux networks (Afriflux, AmeriFlux, AsiaFlux, CarboAfrica, CarboEurope-IP, CarboExtreme, ChinaFlux, Fluxnet-Canada, KoFlux, LBA, NECC, OzFlux, TCOS-Siberia, USCCC), development of measurement and data submission protocols (funded by Office of Science (BER), US Department of Energy), and the Fluxnet project, whose work and support was essential for obtaining the measurements without which the type of integrated analyses conducted in this study would have not been possible. SV is a postdoctoral research associate of the Fund for Scientific Research - Flanders. Collelongo is a research site of the Italian network of Long Term Ecological Research (LTER-Italy).

  References

  Supplementary Material

Box S1 - Nutrient classification. Tab. S1 - Spearman correlation matrix for the six Mediterranean forests considered in the study.