•The potential for elevated [CO2]-induced changes to plant carbon (C) storage, through modifications in plant production and allocation of C among plant pools, is an important source of uncertainty when predicting future forest function. Utilizing 10 yr of data from the Duke free-air CO2 enrichment site, we evaluated the dynamics and distribution of plant C.
•Discrepancy between heights measured for this study and previously calculated heights required revision of earlier allometrically based biomass determinations, resulting in higher (up to 50%) estimates of standing biomass and net primary productivity than previous assessments.
•Generally, elevated [CO2] caused sustained increases in plant biomass production and in standing C, but did not affect the partitioning of C among plant biomass pools. Spatial variation in net primary productivity and its [CO2]-induced enhancement was controlled primarily by N availability, with the difference between precipitation and potential evapotranspiration explaining most interannual variability. Consequently, [CO2]-induced net primary productivity enhancement ranged from 22 to 30% in different plots and years.
•Through quantifying the effects of nutrient and water availability on the forest productivity response to elevated [CO2], we show that net primary productivity enhancement by elevated [CO2] is not uniform, but rather highly dependent on the availability of other growth resources.
Forested ecosystems are the dominant terrestrial sink for carbon (C) (Schimel et al., 2000; Pacala et al., 2001; Liski et al., 2003). Therefore, determining current and future global C sinks necessitates accurate information on forest responses to increased atmospheric [CO2] and climate change. Of particular interest is whether forest growth and C storage will increase under future conditions, possibly mitigating some of the human-induced rise in [CO2]. Presently, there is growing recognition that plant growth responses to increased [CO2] will not be as large as initially expected, and enhancements will not be uniform across the landscape, as a result of the uneven distribution of other growth resources, such as water and nutrients (e.g. Oren et al., 2001; Gill et al., 2002; Nowak et al., 2004; Körner, 2006). However, there is inadequate quantification of how elevated [CO2] may interact with the spatial and temporal heterogeneity of other growth resources, making it difficult to include such interactions in models that project future forest responses.
Net primary productivity (NPP) is generally expected to increase with increasing concentrations of atmospheric CO2. A recent synthesis of NPP across four temperate closed-canopy forest free-air CO2 enrichment (FACE) sites found a surprisingly consistent enhancement, averaging 23% (Norby et al., 2005). However, this uniform response of the average NPP did not take into account the considerable variability that occurred within each site (Norby et al., 2005), possibly caused by variation in resource availability (e.g. Finzi et al., 2002). Previous studies at the Duke FACE site demonstrated that the [CO2]-induced enhancement of NPP was strongly related to nitrogen (N) availability through the positive effect of N availability on leaf area (McCarthy et al., 2006a). Efforts to characterize how interactions between [CO2] and N availability affect NPP during the second rotation of Populus species in Italy (POP-EUROFACE) failed because previous usage as agricultural land rendered the site N unlimited (Liberloo et al., 2006). Overall, few studies have examined C dynamics under elevated [CO2] within the context of stand development, or have explicitly taken advantage of natural variability in climate and resource availability to increase the relevance of FACE research for predicting future forest function (Osmond et al., 2004).
Determining the effects of elevated [CO2] on forest ecosystem function and C storage requires not only knowledge of the degree to which elevated [CO2] will enhance tree biomass, but also where the additional C will be allocated. Carbon partitioning (i.e. the fraction of total production used by a given component; Litton et al., 2007) affects C storage in two ways: first, partitioning to leaves or fine roots determines the ability of plants to capture additional resources; and, second, different plant parts turn over at different rates, so that partitioning to slow turnover parts (e.g. stems), or to parts that contribute to very slow-turning pools (e.g. soil C) would increase the residence time and C storage. The traditional view of allocation is that plants allocate their resources (C and nutrients) in such a way as to optimize their gain of further resources (including water), in response to their growth environment (e.g. Thornley, 1972; Dewar, 1993; McConnaughay & Coleman, 1999). According to these principles, an optimal allocation strategy would suggest that plants growing under elevated concentrations of atmospheric CO2 should allocate proportionally more C to root formation, in order to exploit soil resources more fully (i.e. to increase water and nutrient uptake). Furthermore, the magnitude of this response should be driven by soil resource availability, with more nutrient or water-limited systems showing a greater increase in fine root allocation (e.g. Palmroth et al., 2006; Litton et al., 2007).
Free-air CO2 enrichment sites allow empirical testing of these theoretical principles, by exposing an ecosystem to a new set of conditions and observing whether the theoretical behavior (or optimum) is achieved (Dewar et al., 2009). Thus far, experimental results from the FACE sites at both Duke and the Oak Ridge National Laboratory (ORNL) have been used in such tests. At the Duke FACE site a majority of the extra NPP was allocated to wood production (DeLucia et al., 2005). Similarly, in the sweetgum (Liquidambar styraciflua L.) plantation at the ORNL FACE site (stand age 10 yr at the initiation of [CO2] enrichment), a majority of the extra NPP was partitioned to wood production during the first year of [CO2] enhancement; however, after 3 yr, 75% of the [CO2]-induced enhancement was partitioned to short-lived fine roots and foliage (Norby et al., 2002, 2004). The disparity in allocation to fine roots occurred even though both Duke and ORNL FACE sites are considered to be N-limited sites (Finzi et al., 2007). However, a recent application of a growth-optimization model correctly predicted these differing fine root allocation patterns at both sites based on changes in ‘apparent’ available soil N and differing root longevity (3 vs 0.53 yr; Franklin et al., 2009). A recent synthesis of the largely accurate performances of several optimization models also suggests that models of this type may be useful for predicting global change effects on allocation patterns (Dewar et al., 2009). However, additional data are needed to validate these models.
Studies of trees or ecosystems under elevated [CO2] have primarily focused on average responses, seldom proceeding further to assess how responses to changes in [CO2] may depend on other growth resources, such as water and N (c.f. Finzi et al., 2002). Furthermore, most studies on the effect of elevated [CO2] on trees and forests are too brief to permit separating treatment effects on partitioning via the effect on the rate of development from the direct effects at comparable developmental stages (Körner, 2006). Our study, performed at the Duke FACE site, combines the results of 10 yr of studies on plant C pools and fluxes in order to assess the long-term effects of elevated [CO2] on stand development, as reflected in C production and partitioning. These data were collected during a time period with a broad range of climatic conditions and from a site with a variation of nearly 2.5-fold in natural N availability, allowing us to quantify how much productivity and partitioning are affected by water and N availability, and how the [CO2]-induced response is affected by these two factors. The objective of this study was to examine the long-term effects of elevated [CO2] on stand C dynamics, focusing on the following questions. Does the elevated [CO2]-induced enhancement of NPP change over time? Does elevated [CO2] change the proportion of C in different plant C pools, or the partitioning of new C, as the stand ages? And, how does elevated [CO2]-induced enhancement of NPP vary with soil N and water availability?
Materials and Methods
The Duke FACE experiment is located within a loblolly pine (Pinus taeda L.) plantation on moderately low-fertility, acidic clay-loam of the Enon Series, in the Blackwood Division of Duke University’s Duke Forest, North Carolina (35°58′N, 79°06′W; elevation 130 m). The plantation was established in 1983. Broadleaf species present include Liquidambar styraciflua, Acer rubrum, Ulmus alata and Cornus florida. Summers are warm and humid, and winters are moderate. The average annual precipitation is 1145 mm, distributed fairly evenly throughout the year (in years receiving average or greater precipitation).
This study was based on combined information from the FACE prototype and reference (plots 7 and 8, respectively) and the replicated FACE experiment (plots 1–6). Plots 7 (30 m in diameter) and 8 were established in 1993; plot 7 has received elevated concentrations of CO2 (550 parts per million) during daylight hours of the growing season, according to the FACE protocol (Hendrey et al., 1999), since 1994. Plots 1–6 were established near plots 7 and 8 at the end of August 1996. Three of these six 30-m-diameter plots receive elevated [CO2], according to the FACE protocol (targeted at +200 ppm; Hendrey et al., 1999). The average CO2 concentration in these plots during 1996–2004 was 571 ppm, with 92% of 1-min [CO2] averages within 20% of the target (average target = 573 ppm). In June 1998, plots 7 and 8 were both split in half by an impermeable barrier, and one half of each plot began receiving yearly N fertilization (11.2 g of N m−2 yr−1); four additional reference plot pairs without FACE infrastructure (auxiliary plots; 10 × 10 m) were established nearby, and one member of each pair was fertilized (Oren et al., 2001). A full description can be found at http://face.env.duke.edu.
Net primary productivity for pines, hardwoods and the entire stand was calculated as the sum of the production of coarse wood (stems, branches, coarse roots), leaf litter (lagged for pines), fine roots and reproductive structures. Standing biomass was calculated as the sum of standing coarse wood, fine roots and foliage. Details are presented in the following sections.
Coarse wood pools The stem and branch biomass values of individual trees were calculated allometrically, based on tree diameters and heights. Annual diameter increments of 31–34 pines and 10–26 hardwoods per plot were determined from dendrometer bands, as described previously by Moore et al. (2006). From these measurements, relationships were developed between the individual tree basal area and the basal area increment, and were applied to unbanded trees. Relationships were developed separately for pines and hardwoods, and were plot- and year-specific. Bands were placed on all pines and hardwoods that were > 8 cm in 2004. The heights of banded pines were measured every 5 yr (1996, 2001 and 2006) using either a height pole or a survey laser (Criterion 400; Laser Technology Inc., Englewood, CO, USA). Plot-specific height–diameter relationships were created to calculate the height of all trees for the year they were developed, and height increments (between 1996 and 2001 and between 2001 and 2006) were calculated as the difference between consecutive height values. Height increments were distributed among years without height measurements, using the same relative proportions seen in diameter increments. Based on these measurements we concluded that the relationship of height vs diameter did not change under elevated [CO2] (repeated-measures ANCOVA of the average height vs average diameter of dominant trees reported the [CO2] and [CO2] × year effect as P >0.650; Fig. 1), although trees were taller under elevated [CO2] (Fig. 2a; P =0.025).
Previously at the Duke FACE site (e.g. DeLucia et al., 1999; Hamilton et al., 2002; Schäfer et al., 2003; Finzi et al., 2006; McCarthy et al., 2006a), pine stem, branch and root standing biomasses were calculated as a function of diameter only, based on the site-specific equations of Naidu et al. (1998). Quantification of actual tree heights revealed that the fixed (i.e. invariant) height–diameter relationship of Naidu et al. (1998) represented measured values reasonably well at the beginning of the study (1996), but increasingly diverged from measured values, such that by 2006 most trees were underestimated in height by several meters (Fig. 2a,b). When an alternative approach (described below), which explicitly accounted for the effect of both diameter and height in determining standing biomass, was compared to the approach of Naidu et al. (1998), the two methods yielded similar standing biomass values at the beginning of the study, which then diverged over time because the equations of Naidu et al. (1998) progressively underestimated actual tree heights (Fig. 2c). This comparison suggests that the former equations worked reasonably well at the stand near the time they were developed, but were not flexible enough to describe the Duke FACE stand as it developed. Therefore, it was imperative to use an allometric approach which accounted for both height and diameter.
Stem biomass of individual trees was derived from stem volume multiplied by wood density. Inside-bark stem volume was calculated from measurements (or estimates) of individual tree heights and diameters by integrating a taper equation based on more than 1000 trees in the coastal plain and Piedmont regions of the southeastern USA (Fang et al., 2000). The relationship between bark volume (as a fraction of total stem volume) and relative height on stems, presented in Schultz (1997), was scaled to the site-measured bark fraction at breast height and was used to calculate absolute bark volume. Wood and bark volumes were converted to biomass using a specific gravity of 0.427 g cm−3 for wood and a specific gravity of 0.279 g cm−3 for bark (Naidu et al., 1998). In December 2002, a severe ice storm caused substantial tree breakage (23% of pines lost tops; McCarthy et al., 2006b). After this event, subsequent standing biomass estimates of broken pines were corrected by subtracting the biomass associated with broken tops (measured after the storm; McCarthy et al., 2006b), and subtracting biomass increment overestimates that resulted from the changed allometry. Lacking better information, individual tree branch biomass was calculated from stem biomass using the same proportionality between these components as presented in Naidu et al. (1998). Plot-level coarse root biomass (> 2 mm) was calculated as a function of aboveground biomass (Johnsen et al., 2004).
Biomass increments were calculated by subtracting consecutive standing biomass pools. Woody biomass increments of P. taeda from fertilized plots were reduced by 8% to account for fertilization-induced reductions in specific gravity (Oren et al., 2001). Conversions from biomass to C were made using C contents of 0.48 for stems and branches and 0.44 for roots (Oren et al., 2001).
For hardwoods, individual tree stem and branch biomass estimates were calculated directly, based on measured or calculated diameters. Species-specific equations were available for A. rubrum, L. styraciflua, Liriodendron tulipifera, Quercus alba, Quercus falcata, Ulmus spp. and Carya spp. (Clark et al., 1986). For species without specific equations, the equation of the most similar species was used. Plot-level hardwood coarse roots were calculated from aboveground biomass, using the same proportion observed in pines. As with pines, biomass increments were calculated by subtracting consecutive standing biomass estimates. Conversions from biomass to C were made using C contents of 0.48 for stems and branches and C contents of 0.44 for roots (Schäfer et al., 2003).
Foliage According to standard procedures (e.g. Ryan, 1991), foliage production for P. taeda was determined by lagging collected leaf litterfall masses by 2 yr in order to account for foliage longevity, correcting when necessary for the effects of droughts and storms (see McCarthy et al., 2007); for hardwood species, foliage production was based on that year’s litterfall mass (McCarthy et al., 2007). Leaf litter was collected twice-monthly during peak litterfall (September to December) and monthly otherwise from either 12, 0.16-m2 baskets per plot (Finzi et al., 2002, 2006) or four, 0.5-m2 litter baskets (plots 7, 8; starting in 2001). A period of parallel sampling, positioning both collectors in the same plots, found no difference in the sampled litterfall between the two different types of collectors (McCarthy et al., 2007). Samples were dried (65°C, 4 d), separated into pine needles or hardwood leaves, and weighed. Standing leaf biomass was calculated at its maximum (mid-September) by dividing leaf area index (McCarthy et al., 2007) with litter specific leaf area (SLA) for hardwood or green needle SLA for pines (DeLucia et al., 2002). Litter-specific values were used for hardwoods, as repeated measurements at this site (Finzi et al., 2001; McCarthy et al., 2007) have shown no differences between litter and green leaf SLA. Calculations for standing biomass of pine foliage use canopy averaged green needle SLA in order to determine leaf biomass at the maximum leaf area (September) when both current year and previous year foliage is present. Foliage biomass was converted to C using a C content of 0.48 for pine foliage, and a C content of 0.46 for hardwood foliage (Schäfer et al., 2003).
Fine roots The fine root standing biomass (diameter < 2 mm) for 1997–1998 was taken from a previous publication (Matamala & Schlesinger, 2000). The fine root biomass in 2002–2005 was determined from four soil cores per plot, taken at near-maximum fine root biomass (Jackson et al., 2009). At each measurement time point, the organic litter layer (O horizon) was collected first, and then the top 30 cm of the mineral soil (A horizon) was sampled using a 5-cm-diameter corer (AMS Inc., American Falls, ID, USA). Coring depths of 30 cm were chosen based on early profile measurements, finding no fine roots below 25 cm (Matamala & Schlesinger, 2000), and subsequent measurements showing few roots between 15 and 30 cm. Live roots were manually extracted from the O and A horizon samples and cleaned in deionized water. Fine roots were removed and dried to constant weight at 55°C. Fine root production (g m−2 yr−1) was estimated by determining the proportion of average annual root length standing crop (assuming that this is equal to mass turnover rate) that was produced during a given year (using 12 mini-rhizotrons per plot in plots 1–6) and multiplying this value by the average annual standing crop quantified using soil cores (Matamala & Schlesinger, 2000; Jackson et al., 2009), as described in Pritchard et al. (2001, 2008a). Fine root biomasses were corrected for soil rock volume by multiplying the estimates from coring with the nonrock fraction of soil, and were then converted to C using a C content of 0.41 (Matamala & Schlesinger, 2000). Fine roots were not separated between pine and hardwood. For the purpose of calculating total pine and hardwood NPP and standing biomass, fine roots were partitioned based on the ratio of foliage biomass of each species to the stand foliage biomass.
Reproductive organs The production of reproductive organs was determined using two different methods. First, the production of pine and hardwood reproductive materials (seeds, cones and fruits) was determined from the same litterfall traps used to collect foliage. Reproductive organ biomass was converted to C using a C content of 0.5. Additionally, for pine, estimates of seed and cone production were also made using cone and seed counts. Cones on all P. taeda trees within plots 1–6 were counted from above-canopy towers using binoculars, starting in September 1999 (LaDeau & Clark, 2001, 2006). The average number of seeds per cone was determined from cones collected in September 2000 (LaDeau & Clark, 2006). Combining the total number of seeds and cones per ring with the average mass (0.021 g and 21 g) and C fraction (56% and 49%) of seeds and cones yielded estimates of production of seeds and cones for each year (Way et al., 2009; S. L. LaDeau, unpublished).
Atmosphere and soil variables
Atmospheric measurements Vapor pressure deficit (VPD) was calculated according to Buck (1981), from upper canopy level air temperature and relative humidity (HMP35C and HMP45C; Vaisala, Helsinki, Finland), which were monitored continuously in each plot, starting in 1997. Above-canopy precipitation (P; tipping bucket TI; Texas Instruments, Austin, TX, USA) was measured in plots 4 and 7. Photosynthetically active radiation was measured in plot 1 (LI190SB; Li-Cor, Lincoln, NE, USA). Sensors were sampled every 30 s, and 30-min averages were logged (21X or CR23X; Campbell Scientific, Logan, UT, USA). Monthly averages of air temperature and VPD, and monthly totals of photosynthetically active radiation and precipitation, are shown in Fig. 3. Growing season potential evapotranspiration (PET) was derived from adjusted pan evaporation values (Kohler et al., 1955), from a station approx. 7 km from the study site. These data were obtained from the National Climatic Data Center (http://www.ncdc.noaa.gov).
Soil water content Beginning in 1997, volumetric soil water content (θ; m3 m−3) of the upper 30 cm was measured continuously in four locations (CS615; Campbell Scientific) in plots 1–6. Within plots 7 and 8, θ at 30-cm depth was measured continuously at two locations over the entire study period; an additional two probes at 30 cm and four probes at 10 cm were added in 2001 (ThetaProbe ML1x or ML2x; Delta-T Devices, Cambridge, UK). Sensors were sampled every 30 s, and 30-min averages were logged. Moisture values were corrected for soil rock volume by multiplying the sensor outputs with the nonrock fraction of soil. Monthly averages of soil moisture are shown in Fig. 3e.
Nitrogen availability Annual net N-mineralization rates (g of N m−2 yr−1; Nmin) were measured during 3 yr: 1998, 2003 and 2005 (Finzi et al., 2006; A. Gallet-Budynek, unpublished). According to Finzi et al. (2006), rates of annual N mineralization were measured in the top 15 cm using the buried bag technique (Eno, 1960), wherein soil cores were analyzed for initial NH4 and NO3 concentrations, field incubated and then remeasured for NH4 and NO3 concentrations. In 1998, 15 cores (4.78 cm diameter × 15 cm depth) were taken per plot per sample period; in 2003 and 2005 four cores were taken per plot per sample period. The cores were extracted and placed in polyethylene bags. Initial concentrations of NH4 and NO3 were determined on 20-g subsamples of soil from each polyethylene bag. Samples were incubated in the field for 1 month, and then removed to the laboratory to determine the NH4 and NO3 concentrations. At this time a new set of cores was collected for incubation during the following month. The annual rate of net mineralization was then calculated by summing the differences between the concentration of inorganic N in incubated and initial samples across the 12 months. The concentrations of inorganic N were measured using an autoanalyzer (Lachat QuickChem FIA+ 8000 Series; Zellweger Analytics, Milwaukee, WI, USA). Nitrogen mineralization values were corrected for soil rock volume by multiplying the estimates from cores with the nonrock fraction of soil.
Mineralization rates were pooled across years to estimate an average mineralization rate for each plot. For plots 7 and 8, N-mineralization rates were measured only in 2005; therefore the average rates were calculated based on the relationship of 2005 N mineralization with the average mineralization derived from plots 1–6. Considering that losses of N from this system have been shown to be negligible (Phillips et al., 2001; Finzi et al., 2002, 2006), available N was defined as the sum of N mineralization, N deposition (from Sparks et al., 2008) and N fixation (from Hofmockel & Schlesinger, 2007).
Soil rock fraction The average rock fraction of the soil in plots 1–8 was determined in 2005 by excavating two pits (0.3 × 0.3 × 0.3 m) within each plot and two pits (0.5 × 0.5 × 0.3 m) directly adjacent to the plots. The volume of rocks extracted (estimated using water displacement) was divided by the volume of the pit (estimated by metering fine sand to refill each pit). The rock fraction varied from 4 to 18% between the plots, with an average of 9% (Table 1).
Table 1. Rock volume (%) by plot
Rock volume (%)
Plots with elevated [CO2] are underlined.
The Duke Forest FACE experiment includes the six plots of the ‘replicated’ portion of the experiment (n =3) and the FACE prototype complex. The FACE prototype complex (plots 7 and 8, and four auxiliary plots) includes five ambient plots receiving no fertilization and five fertilized replicates. Given the small size of the auxiliary plots, we treated the ambient plots of the complex collectively as one plot, resulting in four replicates each for ambient and elevated [CO2] conditions, and one each for fertilized and elevated [CO2] × fertilized conditions. The fertilization treatments were used only to generate information on the upper limits of responses under ambient and elevated [CO2] conditions, through one-sample t-tests (Sokal & Rohlf, 1995). The effects of elevated [CO2] on tree height, plant biomass pools and fluxes and C partitioning were analyzed through repeated-measures ANOVA with plots blocked according to the pairing of plots established at the onset of the experiment (n =4 blocks). The effects of [CO2] during each year of the experiment were tested through linear contrasts within the repeated-measures ANOVA. The relationships of wood production and NPP with available N were evaluated using ordinary least-squares regression. The relationship of the residuals of stand NPP vs available N, against growing season precipitation minus potential evapotranspiration (P-PET) was evaluated using ordinary least-squares regression within each treatment. As treatment with elevated [CO2] did not begin in plots 1–6 until nearly the end of the growing season in 1996, data from 1996 are considered pretreatment. All statistical analyses were conducted in sas (Version 8.0; SAS Institute Inc., Cary, NC, USA).
Plant carbon pools and annual net primary productivity fluxes
The NPP of pine before treatment (1996) was not significantly different between the elevated and ambient [CO2] treatments (Table 2; P =0.293; n =3). Elevated [CO2] increased pine biomass production, starting in 1997 and continuing every year thereafter (Fig. 4a; maximum P <0.017). There was no significant interaction effect of elevated [CO2] and time on NPP ([CO2] × year; P =0.626), indicating that the [CO2]-induced enhancement remained fairly consistent as the stand developed. Elevated [CO2] increased stand (pine plus all other species) biomass production every year from 1997 onwards (Fig. 4b; maximum P <0.003), with no trend over time ([CO2] × year; P =0.713). The average enhancement was 277 g of C m−2 yr−1 or 28% greater stand level NPP. Relative to ambient [CO2] without fertilization, fertilizing significantly enhanced pine NPP under ambient [CO2] in 1999, 2000 and 2003, and under elevated [CO2] in all years except 2002 (P <0.05). Fertilization effects on stand NPP were similar to those for pine alone (Fig. 4b).
Table 2. Net primary productivity (NPP) and standing biomass for 1996 (pretreatment) to 2004
Elevated × fertilized
n =4 for ambient and elevated; n =1 for fertilized and elevated × fertilized.
Pine NPP (g of C m−2)
Pine biomass (g of C m−2)
Stand NPP (g of C m−2)
Stand biomass (g of C m−2)
Although elevated and ambient treatments had similar standing biomass at the start of treatment (1996 in Table 2; P =0.536 and P =0.383 for the pine and the entire stand, respectively), by 2000, plots under elevated [CO2] had accumulated significantly more standing biomass (P =0.023 for stand) and by 2002 plots under elevated [CO2] had accumulated significantly more pine biomass (P =0.039). The mean rate of change in pine and stand biomass under elevated [CO2] was greater than under ambient [CO2] (Fig. 4c,d; [CO2] × year P =0.002 for the pine and P =0.004 for the entire stand). Notably, the variation in plant biomass among plots was reduced when considering the entire stand rather than pine alone, reflecting a somewhat compensating effect of higher hardwood biomass in plots with the lowest pine biomass. Standing biomass was calculated using the annual maximum value for foliage and September values for fine roots. Using minimum values would have reduced the standing biomass by approx. 4%. More than half of the standing biomass present in 2004 was produced in the 8 yr since [CO2] enrichment began (57% under ambient [CO2] and 62% under elevated [CO2]). Lower initial biomass in fertilized plots under ambient and elevated [CO2], combined with low statistical power, precluded detection of any significant effects of these treatments on the standing biomass of the pine or the entire stand (P >0.05).
Plant carbon partitioning
Repeated-measures analysis on the partitioning of NPP (into pine stems, branches, coarse roots, foliage; hardwood stems and branches, coarse roots, foliage; and combined fine roots) revealed a significant change in partitioning over the course of the experiment (maximum P =0.002), but no effect of elevated [CO2] (minimum P >0.262; Fig. 5a–d). In particular, a greater fraction of NPP was partitioned to foliage, and a lesser fraction to stems, as the stand aged. However, the actual magnitude of the temporal change was small, with allocation to stems decreasing from 50 to 46% of NPP and to foliage increasing from 18 to 23% (excluding the severe drought year of 2002). Partitioning to reproductive structures was measured, but not included in this analysis, as it comprises < 1% of the total stand NPP. For more information on the distribution of NPP within specific plots see Supporting Information Tables S1,S2.
As with NPP partitioning, the distribution of C in plant biomass pools depended strongly on time (maximum P <0.006), but not on [CO2] ([CO2] effect minimum P =0.138 and [CO2] × year effect minimum P =0.055; Fig. 5e–g). The exception to this was a minimal, but significant, decrease in the proportion of standing fine root biomass under elevated [CO2] (0.97% vs 1.08% by 2004; [CO2] × year effect P =0.016). Additionally, linear contrasts indicated that the proportion of standing pine coarse roots was significantly lesser under elevated [CO2] during 1997–2002 and the proportion of standing pine foliage was significantly lesser in 1998–1999. However, the actual magnitude of these changes was small (< 1%), well within the measurement or estimate uncertainty. The vast majority of stand C was in pine biomass (> 90%), particularly in stems (> 60%). Although the [CO2] treatment did not affect the distribution of C in live biomass pools, there was a significant temporal shift in the distribution of C in several of the pools as the stand developed. From 1997 to 2004 the fraction of standing C in pine stems increased by 4% (P =0.001), while the fraction of C in short-lived tissues (pine and hardwood foliage and fine roots) decreased (P <0.001, P =0.018, and P <0.001, respectively). Mycorrhizal C was not measured until late in the experiment, but its omission earlier in the study is not expected to have much of an impact on the assessment of standing biomass, as mycorrhizal C represented < 1% of the live biomass at the end of the study period (K. K. Treseder, unpublished). For more information on the distribution of standing biomass within specific plots see Tables S3,S4.
Annual net primary productivity fluxes – interaction effects of [CO2] with nitrogen and water availabilities
Thus far, the results have been presented as treatment-level averages. However, even before the onset of [CO2] treatment, there was large variation in plant biomass among plots, reflecting differences in site resources. Although other factors may also vary between plots, most of the difference in initial (1996) standing biomass was related to our index of available N (Fig. 6; R2 = 0.94, P =0.001). Subsequently, variation of temporally averaged stand NPP among plots under treatment with ambient and elevated [CO2] was related to N (Fig. 7a; R2 = 0.63 and R2 = 0.74 for ambient and elevated [CO2] respectively), with a significant interaction of [CO2] and N (P = 0.003). Similar results were obtained for pine NPP, and for pine and stand wood production (data not shown). Accounting for spatial variation in productivity among plots did not address the appreciable interannual variability in NPP. After testing a number of environmental factors (mean growing season VPD, growing season length, mean growing season temperature, mean growing season θ, growing season precipitation, growing season PET and P-PET), P-PET was found to have the most explanatory power (R2 = 0.48, P <0.001). When the residual variation in stand NPP (after accounting for N) was related to P-PET (Fig. 7b), the data from 2 yr (2002 and 2003) were found to fall well below the rest of the data. Because NPP in these years represents the carryover effects of the previous year’s disturbance (as reflected in substantially reduced leaf area index; McCarthy et al., 2007), rather than the effect of weather conditions during the year of observation, analyses were made excluding these 2 yr (Fig. 7b, thick line). Elevated [CO2] did not affect the relationship between the residuals of the NPP vs N relationships and P-PET (P =0.637). Similar results were obtained for pine NPP, and for pine and stand wood production (data not shown).
Multivariate regression models accounting for the combined effects of N and water availabilities on NPP (for ambient and elevated [CO2] treatments separately) were used to generate a continuous surface of absolute and relative [CO2]-induced enhancement, demonstrating the dependence of the [CO2]-induced enhancement of stand NPP on available N and P-PET; (Fig. 8). The regression models, having a significant N-availability index (P =0.014 and P <0.001 for ambient and elevated [CO2], respectively) and P-PET terms (P =0.009 and P =0.013 for ambient and elevated [CO2], respectively) captured much, but not all, of the variability in NPP (R2 = 0.48 and 0.58 for ambient and elevated [CO2], respectively; P <0.001 for both). The interaction of N × P-PET was not significant in either treatment (P =0.170 and 0.907 for ambient and elevated [CO2], respectively). From the surface, the absolute [CO2]-induced enhancement of stand NPP increased with both increasing N and P-PET (Fig. 8a). However, it was evident that the [CO2] interaction is stronger with N than with P-PET, as the enhancement changed more over the N gradient (approx. 130 g of C m−2 from the lowest N to the highest N) than the P-PET gradient (approx. 30 g of C m−2 from the lowest P-PET to the highest P-PET), even though the ranges in P and PET over the study period encompassed much of their respective ranges in the longest record available in this area. This pattern was also confirmed by the multivariate regression of all data combined, testing for [CO2], N and P-PET effects and their potential interactions. Stepwise regression revealed only three significant effects: N, P-PET and N × [CO2] (P =0.003, P <0.001, P <0.001, respectively). The interaction terms of P-PET × [CO2], P-PET × N and the three-way interaction of P-PET × N × [CO2] were not significant (minimum P =0.227). The relative [CO2]-induced enhancement was nearly constant across the P-PET gradient, except at very high concentrations of N, where the relative enhancement increased with decreasing P-PET (Fig. 8b). The average enhancement (using the average N-availability index of the plots and the average P-PET observed during the study) was 271 g of C m−2 or 26% (Fig. 8a,b). This is similar to the treatment-level enhancement, of 277 g of C m−2 (or 28%), calculated as the average ratio of elevated NPP/ambient NPP.
Elevated [CO2] has the potential to change many aspects of forest ecosystem development and function. We found that elevated [CO2] resulted in a sustained increase in plant biomass production, the magnitude of which was determined by water and, particularly, N availability. The more available these resources were, the greater was the [CO2]-induced enhancement in NPP. Notably, on a relative basis, [CO2]-induced enhancement was mostly invariant with changing P-PET. Thus, the anticipated increase in the relative [CO2]-induced enhancement with increasing moisture limitation (i.e. the amelioration of drought effects by increased [CO2] (e.g. Strain & Bazzaz, 1983) was observed only where N availability was very high. Elevated [CO2] led to increases in NPP (averaging 277 g of C m−2 yr−1) and in standing plant C (averaging 217 g of C m−2 yr−1 greater rate of accumulation), yet did not change the distribution of C among biomass pools or fluxes.
Plant carbon pools and average annual net primary productivity fluxes
The majority of living plant biomass currently in the study ecosystem has been produced since the onset of the experiment (averaging 57% in plots under ambient [CO2] and 62% in plots under elevated [CO2]; Fig. 4d). This means that the vast majority of actively functioning plant tissue (where older wood is nonfunctional or less functional than newer wood) was formed since the commencement of the experiment. Qualitatively, our findings regarding the effects of elevated [CO2] on standing plant biomass and NPP are not fundamentally different from previous assessments at this site (e.g. Hamilton et al., 2002; Schäfer et al., 2003; Finzi et al., 2006). At the treatment level, elevated [CO2] significantly increased NPP and standing plant C, and the enhancement of NPP was maintained over time. However, correction of the systematic underestimation of tree height, and the consequently substantial underestimates of standing biomass and NPP (e.g. DeLucia et al., 1999, 2002; Hamilton et al., 2002; Schäfer et al., 2003; McCarthy et al., 2006a), bring the stand in line with observations of NPP and standing biomass in similar stands in the North Carolina Piedmont region (e.g. Kinerson et al., 1977), as well as moderating what was believed to be a sharp decline in NPP with stand age (Finzi et al., 2006). Furthermore, accurate quantities of plant C pools and fluxes are critical for closing ecosystem C and N budgets in both [CO2] treatments, and are essential for testing and constraining models intended to predict the effect of elevated [CO2] on forest ecosystems.
Interaction effects of [CO2] with nitrogen and water availabilities
In the absence of additional information, initial standing biomass has sometimes been used as a covariate to account for the impact of initial conditions on later effects of elevated [CO2]; however this is likely to be less appropriate for long-term studies in which trees are no longer experiencing exponential growth. It is more desirable to identify the causes of spatial differences and to employ those to interpret the effects of [CO2]. Thus, by extending earlier work linking NPP to net N mineralization (Finzi et al., 2002), this study shows that the N-availability index – statistically interchangeable with initial plant biomass (Fig. 6) – explains much of the spatial variation in NPP at this site and strongly determines the response of NPP to elevated [CO2] (Fig. 7a). The positive interaction with elevated [CO2] results in progressively smaller absolute enhancements of stand NPP as available N decreases. In the extreme, very low availability of nutrients (e.g. in P. taeda on sandy soils and Picea abies on sandy glacial till), have resulted in no observable response to elevated [CO2] (Oren et al., 2001; Ward et al., 2008). In both of these studies, significant [CO2]-induced responses were observed in fertilized trees.
Unsurprisingly we also found that water availability, here represented as growing season P-PET, explained much of the remaining interannual variability in productivity within each treatment (Fig. 7b). This is consistent with previous analysis of factors driving inter-annual variation in pine basal area increment (BAI), as well as previous multivariate analyses showing significant, positive effects of precipitation on NPP under ambient and elevated [CO2] at this site (Finzi et al., 2006; Moore et al., 2006). When we combined the responses of NPP to N and P-PET to assess the effect of varying both N and water availability on the elevated [CO2]-induced enhancement of stand NPP (Fig. 7), we found on an absolute basis, [CO2]-induced enhancement increased with increasing N and P-PET, while on a relative basis, [CO2]-induced enhancement was mostly invariant with changing P-PET. This relative response contrasts with the finding that [CO2]-induced relative enhancement of BAI was strongly related to growing season P, mean temperature and VPD (Moore et al., 2006), the latter two variables positively related to PET. Basal area increments are just one component of stand level NPP, so the disparity in results between BAI and NPP could be caused by the nonlinear increase in tree biomass with increasing tree diameter, and the nonstem biomass components, which are likely to have differential sensitivity to environmental drivers. Two years of our study (2002 and 2003) exhibited NPP that was seemingly decoupled from the current year’s water availability (P-PET). While we have no measure of nonstructural carbohydrates, it seems likely that depletion of these reserves (through drought and ice storm) could have resulted in the below-average NPP observed in these years.
Early conceptual models predicted that relative [CO2]-induced enhancement would increase with increasing nutrient availability and decreasing water availability, such that ecosystems with high water availability and low nutrient availability should be virtually unresponsive to elevated [CO2], and ecosystems with low water availability and high nutrient availability should have the greatest response (e.g. Strain & Bazzaz, 1983). In forests, subsequent data has supported the predictions regarding N availability (Oren et al., 2001; Nowak et al., 2004). By contrast, the projections of greater relative enhancement with decreasing water availability have largely not been borne out by data, with enhancement in forests actually increasing with precipitation (Nowak et al., 2004). Our results add to the very few that quantify the important role that both nutrient and water availability are expected to have in facilitating forest productivity response to elevated [CO2].
Plant carbon partitioning
Our findings of pine standing biomass distribution of 67% to stem, 10% to branches, 18% to roots and 4% to foliage (at age 21 yr), closely match the expected proportions for loblolly pine of ≥ 15 yr, of 65–70% to stem, 10% to branches, 15–20% to roots and 3–5% to foliage (Schultz, 1997). Unlike many assessments of ecosystem C pools and fluxes, this study also captured partitioning to reproductive organs. Although reproductive structures currently represent less than 1% of the total annual C partitioned at the Duke FACE site Table S2), reproduction in loblolly pine was significantly enhanced in plots under high [CO2] and may indicate long-term ecosystem consequences for forest composition (e.g. LaDeau & Clark, 2001, 2006). Similarly, although allocation to mycorrhizae has been estimated to be only approx. 5% of NPP in this forest (K. K. Treseder, unpublished data), significant increases in ectomycorrhizal colonization under elevated [CO2] (Garcia et al., 2008; Pritchard et al., 2008b) are likely to be an important factor contributing to increased NPP under elevated [CO2].
The results from other FACE studies have been mixed with regard to whether elevated [CO2] causes substantial shifts in the proportion of C in various plant pools and whether there is a shift in the partitioning patterns of new C. The L. styraciflua plantation at the ORNL FACE site showed a distinct shift, from up to 80% of the extra NPP induced by [CO2] partitioned to slow pools (wood) during the early phase of the experiment, to only 25% after 3 yr (with the remainder partitioned to fine roots; Norby et al., 2002). Similarly, in the last year of the first rotation in the POP-EUROFACE experiment on Populus species, root pools were increased relatively more under elevated [CO2] than were aboveground woody components (Gielen et al., 2005). However, the root-to-shoot ratio was unchanged, and the fraction of NPP allocated to woody aboveground biomass was high, ranging (among species) from 53 to 67% (Calfapietra et al., 2003; Gielen et al., 2005). Unlike the first rotation, in which the relative accumulation of biomass in stems and roots did not change, during the second (coppice) rotation of Populus species at POP-EUROFACE, elevated [CO2] resulted in greater C accumulation in branches and less accumulation in stems as compared to trees under ambient [CO2]; the ratio of aboveground and belowground biomass remained the same (Liberloo et al., 2006). In contrast, when grow from the start under elevated [CO2], neither the fraction of standing biomass in various pools, nor the partitioning of NPP of Populus tremuloides was changed by elevated [CO2] (King et al., 2005). These differing outcomes suggest that elevated [CO2] does not have a uniform effect on allocation, but rather must be considered in conjunction with other site factors, as reflected, for example, in leaf area index (Palmroth et al., 2006). Additionally, practical approaches to estimating biomass may also constrain the ability to detect some changes in partitioning: in our present study, we can only identify resource-induced changes in partitioning to branches and coarse roots when they are reflected in changes in height or diameter.
Plant C partitioning is the outcome of many processes that are influenced by both internal and external factors (Dewar, 1993; Cannell & Dewar, 1994), and is frequently poorly resolved, even under ambient [CO2] conditions (Litton et al., 2007). However, modeling and empirical advances have been made in some aspects of ecosystem C partitioning. A recent resurgence in optimization models has shown that these models can reproduce measured allocation patterns in current and global change settings (Dewar et al., 2009; Franklin et al., 2009). These studies accurately predicted the quantities of fine root allocation at Duke and ORNL FACE sites, capturing their different partitioning patterns (Franklin et al., 2009). Furthermore, a recent analysis of total belowground C allocation (TBCA) across these four forest FACE sites demonstrated significant, inverse relationships between TBCA and leaf area index and between TBCA and NPP (Palmroth et al., 2006), with a higher fraction of the [CO2]-enhanced NPP allocated belowground where leaf area and NPP are low.
A recent review of plant responses to elevated [CO2] calls for forest FACE experiments to be analyzed with respect to factors driving variation in intrasite growth signals, over as many years as possible, with the recognition that broad averages of elevated [CO2] effects on the C cycle are dependent primarily on the frequency of representation of particular conditions in the literature (Körner, 2006). Here we show that elevated [CO2] may significantly increase the rate of biomass production and the rate of ecosystem-level C storage. However, we also show that NPP enhancement is highly dependent on the availability of other growth resources. Thus, we add our findings to a growing body of literature suggesting that the rate at which extra C will be sequestered with increasing atmospheric [CO2] would greatly depend on the spatial and temporal distributions of other growth resources.
We thank Sari Palmroth and A. Christopher Oishi for early comments on this manuscript. We also thank Evan DeLucia, David Moore and Jeff Pippen for early raw data on tree growth in plots 1-6. This study was supported by the US Department of Energy (Grant No. DE-FG02-95ER62083) through the Office of Biological and Environmental Research (BER) and its National Institute for Global Environmental Change (NIGEC), Southeast Regional Center (SERC) at the University of Alabama, and by the US Forest Service through both the Southern Global Climate Change Program and the Southern Research Station. A.C. Finzi and R.B. Jackson acknowledge ancillary support from the US NSF (DEB0236356 and DEB0235425) and S.G. Pritchard acknowledges ancillary support from the National Institute for Global Environmental Change (NICCR Grant No. DE-FC02-06ER64156).