Introduction

Ecosystem respiration (R eco) is one of the major determinants of the C balance in most terrestrial ecosystems (Valentini et al. 2000). Global warming impacts on ecosystem moisture and temperature patterns might induce substantial changes in R eco (e.g., Huxman et al. 2003; Xu et al. 2004; Davidson et al. 2006). Thus, having a process-based understanding of R eco, particularly in response to a changing climate, is desirable although difficult to achieve since different C pools and respiratory sources (e.g., autotrophic and heterotrophic respiration) respond differentially to changes in the abiotic environment.

Stable C isotopes are used extensively to examine respiratory processes and their environmental drivers (e.g., Bowling et al. 2008), adding invaluable information on substrate turnover, C metabolism as well as photosynthetic and post-photosynthetic fractionation processes imprinted in the isotopic signature of respired CO2. However, so far a complete understanding of the isotopic composition of C pools and fluxes in terrestrial ecosystems has not been achieved (Bowling et al. 2008). The stable C isotopic composition (δ13C) of ecosystem respired CO2 (δ13CR) is an important parameter for application in isotope-based C budget models and understanding the controls over terrestrial C exchange processes (e.g., McDowell et al. 2004a). δ13CR is known to vary seasonally with changes in photosynthetic discrimination and respiratory activity (e.g., Bowling et al. 2001, 2002; Griffis et al. 2004; Ponton et al. 2006; Werner et al. 2006, 2007a; Alstad et al. 2007; Schaeffer et al. 2008). Recent studies showed small (Ogée et al. 2003; Still et al. 2003; Schnyder et al. 2004) to large (>6‰; e.g., Bowling et al. 2003; Werner et al. 2006; Kodama et al. 2008) variation in δ13CR even within a single night. Still, a general understanding of the causes of this variability in δ13CR has not yet been achieved and the interplay of processes driving R eco remains largely unknown. Changes in either the magnitude of component fluxes or their isotope composition (δ13Cres) may drive the observed variations in the isotopic signature of δ13CR. Substantial temporal variations in isotopic composition of CO2 respired by leaves (e.g., Hymus et al. 2005; Mortazavi et al. 2005; Kodama et al. 2008; Werner et al. 2007b, 2009; Priault et al. 2009; Sun et al. 2009), roots (e.g., Klumpp et al. 2005), and soils (e.g., McDowell et al. 2004a; Ekblad et al. 2005; Kodama et al. 2008) have been observed. Variability in photosynthetic discrimination is thought to largely determine the isotopic composition of plant as well as soil and even R eco (e.g., Ekblad and Högberg 2001; Bowling et al. 2002; Mortazavi et al. 2005; Werner et al. 2006). Another reason for the variation in δ13Cres and δ13CR might be temporal changes in C supply by assimilation that could bring about changes in the utilization of different respiratory substrates (e.g., old vs. new C; Nogués et al. 2006). Additionally, there is now evidence for substantial metabolic fractionation in the dark respiratory pathways of leaves (e.g., Rossmann et al. 1991; Gleixner and Schmidt 1997; Tcherkez et al. 2003; Ghashghaie et al. 2003; Mortazavi et al. 2006; Werner et al. 2007b, 2009; Barbour and Hanson 2009; Priault et al. 2009) and during C transport and allocation to stem and roots (Gleixner and Schmidt 1997; Tcherkez et al. 2004; Brandes et al. 2006; Gessler et al. 2009a; Cernusak et al. 2009). Thus, apart from the differential climatic dependencies that bring about changes in magnitude and isotope composition of plant, soil and root respiration fluxes, metabolic processes occurring in each of these components will also have isotope effects which may influence δ13CR. Determining how stable isotope compositions vary in component fluxes and how they affect δ13CR can provide a better process-based understanding of ecosystem C exchange dynamics. No study has yet measured the δ13C values of R eco and all components simultaneously, while only a few have measured several of these fluxes at once (Bowling et al. 2008).

Due to their marked seasonality and assorted periods of rapid changes in net C exchange and their simple two-layer structure, Mediterranean oak woodlands (‘montado’) provide excellent conditions for the examination of diurnal and short-term temporal dynamics in ecosystem component δ13Cres and their effects on δ13CR. Drought is the major constraint for plant productivity in the Mediterranean (e.g., Tenhunen et al. 1990; Pereira et al. 2007) and the rapid climatic transition from spring to summer drought induces a large shift in ecosystem CO2 exchange in a relatively short time frame. During this period the rapidly declining contribution of the understory is of special relevance for the net ecosystem C flux, as we previously showed in an ecosystem-scale flux partitioning study (Unger et al. 2009). We concluded that the R eco flux depends on the interplay of all contributing fluxes [trees, understory, soil microorganisms (SMO) and roots] and thus that its isotopic composition (δ13CR) should vary in accordance with changes in the contributions and isotopic signatures of the component fluxes. The present study aims firstly, to explain variation in δ13CR with variation in both isotopic signatures and respiratory fluxes of the different ecosystem components (soil, roots, understory and tree foliage) and secondly, to achieve a process-based interpretation of drought-induced and diurnal variation in the isotopic signature of these components. We monitored variability in both δ13Cres and δ13CR to add this information to the C flux partitioning presented in Unger et al. (2009). We addressed the following hypotheses: (1) drought-induced processes such as altered photosynthetic discrimination and respiratory demands will cause substantial changes in both ecosystem and component respired δ13CO2, and (2) the feedbacks of the isotopic compositions of the component fluxes (soils, foliage, roots) on δ13CR and δ13CS can be reproduced in an isotopic mass balance.

Materials and methods

Field site and climatic conditions

The experimental site Mitra (CarboEurope–IP) near Évora in southern Portugal (38°N; 8°W) has a typical Mediterranean climate, with hot and dry summers and mild wet winters. The mean annual temperature is 15–16°C and average annual precipitation is ca. 669 mm. The site is located in a homogeneous landscape (220–250 m a.s.l.) dominated by savannah-like woodlands with drought-adapted evergreen Quercus ilex ssp. rotundifolia L. and Quercus suber L. (Werner and Correia 1996) growing on sandy soils derived from granite rock. The understory consists of grazed pasture dominated by drought-sensitive herbaceous annuals [e.g., Tuberaria guttata (L.) Fourr.], some Gramineae and a few drought-deciduous shrubs (Cistus salvifolius L.). Tree canopy cover is about 21% and leaf area index ca. 0.55 (Carreiras et al. 2006). The vegetative period of the understory lasts from the autumn rains until May–June when soil water content (SWC) generally decreases rapidly. Measurements were conducted between 20 May and 3 June 2006. For C flux partitioning between understory and tree canopy we used understory flux data from Tojal, a grassland adjacent to the Mitra field site (8 km distance, CarboEurope–IP) at the same altitude, with the same soil type and climate as Mitra (Pereira et al. 2007). Vegetation was mainly C3 grasses that were also found in the oak site and one additional C4 species (Cynodon dactylon L.; Pereira et al. 2007). Furthermore, seasonality and leaf duration are similar between the oak woodland understory and grassland (Pereira et al. 2007) and annual courses of gross primary production (GPP) and R eco are highly correlated between the sites. The grassland had been mown on 24 May 2006. However, the cut biomass was mainly dead dry leaf matter (Aires et al. 2008) thus had little influence on the understory respiration flux. Further, the cut dry leaf matter was left on the field, thus the shade effect on the soil was not significantly reduced.

Eddy covariance and micrometeorological measurements

Continuous records of CO2 and H2O fluxes and climate variables were taken on a 28-m-high metal tower (at the Mitra site) equipped with a sonic anemometer (Gill R3; Gill Instruments, Lymington, Hampshire, UK) and a gas analyzer (LI-7000; LI-COR, Lincoln, NE, USA). Weather conditions were continuously recorded by a meteorological station (datalogger CR10X; Campbell Scientific, Logan, UT, USA), with a Q7 REBS net radiometer (Campbell Scientific), aspirated psychrometer H301 (Vector Instruments, Rhyl, Denbighshire, UK) and a rainfall recorder (tipping-bucket rain gauge; Casella, Bedford, UK). Understory fluxes at the Tojal grassland site were measured by a small eddy covariance system at a height of 2.5 m (3D sonic anemometer, 1210R3; Gill Instruments; IRGA, LI-COR) at Tojal, a grassland adjacent to the Mitra field site (CarboEurope–IP).

Eddy covariance data were processed using Eddyflux software (Meteotools, Jena, Germany). Following the CarboEurope-IP recommendations, a planar fit coordinate rotation for wind components was performed. The quality of all primary data was guaranteed by a routine equipment calibration and, for meteorological data, a comparison with data from other nearby stations. Flux-partitioning methods proposed by Reichstein et al. (2005) were applied to separate the net ecosystem exchange into GPP and R eco, respectively. During the 2-week study period there was no missing eddy covariance or climate data. For a detailed description of the eddy covariance data processing and flux partitioning see Unger et al. (2009).

Plant-scale observations

Leaf gas exchange (LI-6400; LI-COR) was recorded in 24-h cycles every 2–4 h on marked branches, for at least three sun-exposed leaves on three Q. ilex trees and on ten sun exposed T. guttata plants. During daytime dark leaf respiration was measured by covering the cuvette with a black plastic shield. This procedure did not affect leaf temperature. Predawn and midday leaf water potentials (pressure chamber; Manofrigido, Portugal) were also obtained for the same species.

Soil respiration measurements and trenching

Soil respiration (R S) was measured in 24-h cycles every 2–4 h using a closed path chamber system (PP-System EGM2 soil respiration system with SRC-1 chamber; PP-Systems, Amesbury, MA, USA). Measurements were conducted with three replicates on three plots of untreated bare soil and three plots with root exclusion. Root exclusion was achieved by inserting metal rings (30 cm diameter) from 0 to 20 cm depth (trenching) and covering them with black foil (solarization) for 2 weeks, 1 year prior to measurements to avoid effects of disturbance or decomposition of decaying root material (e.g., Lee et al. 2003). We regarded trenching up to 20 cm depth as sufficient for the studied site, since the upper soil layer contains the largest fraction of root biomass (Kurz-Besson et al. 2006). Plots were installed 20 m from the nearest tree. Thus, the contribution of respiration from deep tree roots was regarded as negligible. Three weeks before measurements a few new seedlings inside the trenching plots were carefully removed and solarization was repeated to ensure root-free plots. Soil temperature (T soil) and SWC were recorded at 5- to 10-cm depth alongside R S using the temperature sensor of the soil respiration system and a moisture probe (Theta Meter HH1; Delta-T Devices, Cambridge, UK).

Air sampling and isotopic analysis

We used 12-ml soda glass vials (Exetainer; Labco, High Wycombe, UK) capped with pierceable septa for atmospheric air sampling (Werner et al. 2007b). Samples for Keeling plot analyses were collected during four 24-h cycles with one measurement series every 2–3 h. Air was collected at nine heights: 24, 20, 16, 12, 8, 4, 2, 1 and 0.5 m (the heights between 16 and 24 m were above the tree canopy). Air was pumped at 10 l min−1 (pump Capex V2X; Charles Austen Pumps, Byfleet, Surrey, UK) through Dekarbon tubing (25 m length, inner diameter 6.9 mm; Sertoflex, Serto Jacob, Fuldabrück, Germany), which was fixed at the tower. A simple sampling system consisting of an Exetainer vial pierced by two needles connected to the pump and an infrared gas analyzer (IRGA; BINOS 100 4P, Rosemount Analytical, Hanau, Germany; precision for CO2 ±1 p.p.m.) was mounted, and pumped air through the tubing until stable CO2 concentrations were reached; it then sampled for 1 min. Samples were repeatedly collected from the top to the bottom for 30 min resulting in two to three samples per height.

δ13Cres of R S was estimated by chamber Keeling plots. The same pump and IRGA were connected with Dekarbon tubing to a well-ventilated and gas-tight cylindrical Plexiglas chamber (17 l, diameter 30 cm) in a closed system. Temperature and light levels inside the chamber were monitored with two thermocouples fixed at different heights and a quantum sensor, connected to a datalogger (CR10; Campbell Scientific). The chamber was tightly fitted on a metal ring inserted in the soil and sealed with acryl rubber mastic. Air was sampled every minute, while CO2 concentrations inside the chamber were rising; a total of ten samples per soil Keeling plot was collected.

For collection of leaf and root respired CO2, ten leaves of Q. ilex as well as ten leaves and ten root fragments of T. guttata were cut and immediately placed into exetainers which were flushed with CO2-free air (passed through a soda lime column and monitored by IRGA) and incubated for 10 min in a temperature-isolated plastic box (see Werner et al. 2007b). A validation of the method with online-gas exchange measurements on intact leaves revealed no difference in δ13Cres in comparison to cut leaves (see Werner et al. 2007b). Respired CO2 was collected with a 25-ml gas-tight syringe (SGE International, Victoria, Australia) from the exetainer and injected into a second CO2-free exetainer, with rapid needle removal following injection. Tests revealed no significant differences in δ13CO2 before and after the syringe transfer (SD < 0.15‰). Additionally, five vials of known δ13C and CO2 concentration were filled in the field at each sampling date and used as references. Samples were analyzed (UAI, ICAT, Lisbon, Portugal) within 12–20 h after collection. Isotopic compositions were measured with a stable isotope ratio mass spectrometer (IsoPrime; GV Instruments, Manchester, UK) operating in continuous-flow mode coupled to a Multiflow prep-system (GV Instruments). A laboratory gas of known isotopic composition was measured to correct for any drift. Repeated measurement precision was <0.1‰.

Keeling plot calculation

The Keeling plot approach was used to assess the isotopic composition of ecosystem and soil respired CO2. Assuming that both source and background CO2 remain constant during the sampling period (<30 min), the isotopic signature of R eco (δ13CR) can be calculated as the y-intercept of a linear regression of δ13C versus the inverse of the CO2 mixing ratio obtained from vertical profiles (Keeling 1958):

$$ \delta^{13} {\text{C}}_{\text{a}} = c_{\text{T}} (\delta^{13} {\text{C}}_{\text{T}} - \delta^{13} {\text{C}}_{\text{R}} )( 1/c_{\text{a}} ) + \delta^{13} {\text{C}}_{\text{R}} $$
(1)

where δ13C denotes isotopic composition and c denotes CO2 concentrations of the mixing ratios. The subscripts indicate air samples from several heights above and within the canopy (a), tropospheric air (T) and air respired from the ecosystem (R). We calculated ordinary least square (model I; Zobitz et al. 2006) and geometric mean (model II) regressions (Pataki et al. 2003) and did not find significant variation between model I and model II results, when CO2 gradients were above 30 p.p.m. Presented data are from model I regressions following current practice (Zobitz et al. 2006). To remove outliers, residual analyses were performed. Data points were removed from the regression when the residual of an individual data point was greater than 3 times the SD. Regressions were rejected when not significant (α = 0.01). Uncertainties in the Keeling plot intercepts were expressed as SEs of the intercept. Statistical and regression analyses were performed using Statistica (StatSoft, Tulsa, OK, USA) and JMP (SAS Institute, Cary, NC, USA).

Flux partitioning

For nighttime flux partitioning a mass balance equation was applied (Eq. 2; henceforth called the ‘ecoflux’ approach), which assumes that the sum of component fluxes equals the ecosystem flux and that no other component fluxes than soil flux (roots and SMO) and canopy flux (leaves of trees and herbs) contribute to the ecosystem flux:

$$ R_{\text{eco}} = R_{\text{S}} + R_{\text{C }} = R_{\text{r }} + R_{\text{SMO}} + R_{\text{T}} + R_{\text{U}} $$
(2)

where R denotes the respiratory CO2 flux per square meter ground. Subscripts indicate respired CO2 from ecosystem (eco), soil (S), canopy (C), roots (r), SMO as well as from tree (T) and understory leaves (U). Respiration from non-mycorrhizal fungi and microbes is included in heterotrophic R S (R SMO), whereas autotrophic R S (R r) flux comprises root respiration and associated rhizospheric components (e.g., mycorrhizal respiration). Nighttime R eco (measured) minus the soil flux (R S, measured) yields the canopy respiration flux (R C; comprising trees and understory). Subtracting R S of trenching plots (R SMO, measured) from the total soil flux (R S, measured) yields the root flux (R r). The nighttime respiration flux from understory eddy covariance data (see Eddy covariance and micrometeorological measurements) was used to partition R S from understory respiration (R U). Tree respiration (R T) was then calculated by subtracting R U from R C. We then added stable isotope information to Eq. 2 yielding an isotopic flux mass balance (‘eco-isoflux’ approach, Eq. 3):

$$ \delta^{13} {\text{C}}_{\text{R}} \times R_{\text{eco}} = \left( {\delta^{13} {\text{C}}_{\text{S}} \times R_{\text{S}} } \right) + \left( {\delta^{13} {\text{C}}_{\text{C}} \times R_{\text{C}} } \right) = \left( {\delta^{13} {\text{C}}_{\text{r}} \times R_{\text{r}} } \right) + \left( {\delta^{13} {\text{C}}_{\text{SMO}} \times R_{\text{SMO}} } \right) + \left( {\delta^{13} {\text{C}}_{\text{T}} \times R_{\text{T}} } \right) + \left( {\delta^{13} {\text{C}}_{\text{U}} \times R_{\text{U}} } \right) $$
(3)

where δ13C denotes isotopic composition (subscripts as in Eq. 2).

Apart from δ13CC we measured all isotopic compositions entering Eq. 3 directly: δ13CR was measured by the Keeling plot method at the eddy flux tower (see above), δ13CS and δ13CSMO were measured on separate plots (trenched and not trenched, see above), while δ13CT, δ13CU and δ13Cr were measured from leaves (trees and understory) and roots (understory) by the in-tube incubation method (see above). Flux measurements were obtained as described above for Eq. 2. We used these measurements to rearrange Eq. 3 in order to compare the results of eco-isoflux and ecoflux approaches (Eq. 2). For example, to obtain R eco from the eco-isoflux approach we rearranged Eq. 3 by division with δ13CR and compared this value with the measured R eco.

Since root respiration was calculated exclusively from R S, both flux and isoflux partitioning could also be applied independently from R eco at the soil scale (‘soilflux’, Eq. 4 and ‘soil-isoflux’, Eq. 5):

$$ R_{\text{S}} = R_{\text{r}} + R_{\text{SMO}} $$
(4)
$$ \delta^{13} {\text{C}}_{\text{S}} \times R_{\text{S}} = \left( {\delta^{13} {\text{C}}_{\text{r}} \times R_{\text{r}} } \right) + \left( {\delta^{13} {\text{C}}_{\text{SMO}} \times R_{\text{SMO}} } \right) $$
(5)

To quantify the degree of uniformity between eco-isoflux (soil-isoflux) and ecoflux (soilflux) mass balances, we linearly correlated data from the different approaches and calculated an average offset (AO; %) between the measured and the isotope mass balance data. Correlation coefficients (R 2) give information on similar trends, while the AO indicates differences in the data range.

Results

Variation in climatic variables and respiration fluxes

Meteorological conditions during spring 2006 were typical for the Mediterranean climate with highest precipitation in March (104 mm) and April (57 mm) and continuously increasing air temperature (T air), net radiation and vapor pressure deficit (VPD). The last rainfall was on 22 April. Hence, when measurements were conducted at the end of May, the herbaceous vegetation already showed signs of water deficit.

Most days were sunny, except for 22 May with rather cloudy conditions (Fig. 1a). Maximum T air and VPD were 31°C and to 36 hPa, respectively (Fig. 1b, d). T soil followed the same pattern as T air, but revealed slightly higher values (Fig. 1c).

Fig. 1
figure 1

Diurnal variation in a net radiation (NR), b air temperature (T air), c soil temperature (T soil; 5-cm depth) and d vapor pressure deficit (VPD) for 20 May (black circles), 22 May (dark-grey diamonds), 30 May (light-grey triangles) and 3 June (white squares) 2006

Full size image

Leaf gas exchange and water potential measurements on trees (Q. ilex) and understory plants (T. guttata) reflected a differential response to increasing water deficit and temperature (Table 1; Fig. 2a–h). While Q. ilex was able to preserve predawn leaf water potentials around −0.5 MPa and maintained net photosynthesis rates of ~3–4 μmol m−2 s−1 (Table 1), understory plants like T. guttata were severely affected by seasonal drought and died off during the study period. Predawn leaf water potentials became increasingly similar to midday values (−1.5 MPa) and generally low net photosynthesis turned into net respiration towards 3 June (Table 1). Dark respiration rates of leaves revealed a clear diurnal cycle and increased strongly with drought in both species (Fig. 2a–h), particularly in T. guttata (up to 6 μmol m−2 s−1; Fig. 2h). Please note that leaf gas exchange measurements were always performed on live green foliage. Thus, the increasing respiration rates do not contrast with ecosystem-scale measurements, where understory respiration decreased due to strong wilting effects causing a decline in leaf area.

Table 1 Net photosynthesis (NP) rates, predawn and midday leaf water potentials (WP) of trees (Quercus ilex) and understory plants (Tuberaria guttata) during late spring 2006. n = 3–10 (± SD)
Full size table
Fig. 2
figure 2

Temporal variation in a–d single-leaf respiration (R leaf) of Quercus ilex (grey triangles), e–h R leaf of T. guttata (grey inverted triangles), i–l respiration fluxes from soil (R soil; black circles), soil microorganisms (SMO) (R SMO; grey hexagons), roots (R root; white diamonds) and m–p ecosystem respiration (R eco; grey squares) for four diurnal cycles with increasing summer drought (May–June 2006). Note: foliage flux values (a–h) are given in m2 leaf area. Black bars indicate nighttime; n = 3–5 (± SE)

Full size image

RS showed a slight decrease (~1 μmol m−2 s−1) in response to seasonal drought and exhibited higher rates during daytime than during nighttime (Fig. 2i–l). Soil microorganism respiration showed slightly smaller rates but a similar pattern to total soil efflux, whereas root respiration exhibited little variation over the observed period with consistently low rates of about 0.2 μmol m−2 s−1 (Fig. 2i–l).

Reco decreased from 3.5 to 1.5 μmol m−2 s-1 (Fig. 2m–p) and showed similar diurnal dynamics to RS.

Variation in isotopic composition of respired CO2

Respired δ13CO2 of plant material was substantially more enriched than R S. δ13Cres of Q. ilex leaves varied around −19.8 ± 1.2‰ on 20 May and exhibited a continuous decrease to −22.1 ± 0.6‰ throughout the experiment (Fig. 3a–d). The same tendency was observed in leaves of T. guttata but the decrease in δ13Cres was even stronger, from −21.3 ± 1.3 to −25.9 ± 1.7‰, whereas roots of T. guttata exhibited only a small decrease, from −21.8 ± 1.5 to −22.9 ± 1.7‰ (Fig. 3e–h).

Fig. 3
figure 3

Diurnal cycles of stable C isotopic composition (δ13C) of respired CO2 (δ13Cres) from a–d Q. ilex leaves (grey triangles), e–h T. guttata leaves (grey inverted triangles) and roots (white diamonds), i–l soil (black circles) and SMO (grey circles) and m–p ecosystem respiration (δ13CR) derived from Keeling plots (grey squares) and the eco-isoflux approach (crosses) during May and June 2006. Black bars indicate nighttime; n = 10 (± SE)

Full size image

On a diurnal scale, leaf dark δ13Cres was more enriched during the day than at night in both species. Maximum observed variation in δ13Cres was 1.7, 4.2 and 4.8‰ during the daytime and 3.3, 4.6 and 3.8‰ during the nighttime in Q. ilex leaves, T. guttata leaves and roots, respectively. Maximum differences between day and night were 3.6 and 5.3‰ in Q. ilex and T. guttata, respectively. The diurnal amplitude in Q. ilex δ13Cres decreased with drought.

With increasing drought roots of T. guttata exhibited a pattern opposite to that of leaves, showing progressive enrichment during night (Fig. 3e–h). The largest differences between δ13Cres of leaves and roots (up to 7‰) were found on the night of 3 June (Fig. 3h). δ13Cres of R S did not change significantly over time and was on average −26.8 ± 1.2‰ (Fig. 3i–l), with no distinct diurnal pattern. Heterotrophic R S constituted the largest fraction of overall R S and δ13Cres from root-free plots was similar to that from intact soil. However, under progressive drought there was some diel variability in δ13Cres of SMO with a nocturnal increase in δ13Cres on the last 2 sampling days (Fig. 3k, l), similar to the pattern observed for root respiration (Fig. 3g, h).

Isotopic composition of R eco (δ13CR) showed, apart from δ13Cres of soil and SMO, the most depleted values (on average −25.9 ± 2.1‰; Fig. 3m–p). Further, we observed a strong nocturnal enrichment in δ13CR ranging between 2 and 4‰ (7.2‰ on 30 May; Fig. 3o). With increasing drought mean δ13CR values became slightly enriched (−27.3 ± 1.4 to −25.3 ± 1.4‰) and the amplitude in the nocturnal cycle decreased from on average 5.7 ± 1.7 to 3.6‰. However, given the large SEs of the intercepts these trends must be interpreted with care. δ13CR estimated by the eco-isoflux approach was in the range of measured δ13CR but in comparison exhibited slightly enriched values (on average −24.6 ± 1.2‰; Fig. 3m–p). Deviations between measured and estimated δ13CR were partially large (in one case up to 6‰; Fig. 3n) and on average ca. 2‰. There was a weak, though not significant, correlation between the SEs in the Keeling plot intercepts and the differences of measured and eco-isoflux estimated δ13CR (R 2 = 0.13).

Partitioning of isotopic fluxes and comparison of different partitioning approaches

To evaluate changes in the contribution of component fluxes to total R eco with increasing water deficit and temperature, we used two different partitioning approaches, based on measured fluxes (ecoflux approach, Eq. 2) and their respective isotopic signatures (eco-isoflux approach, Eq. 3). A main outcome of the ecoflux partitioning is that the drought-sensitive understory species play the most important role in causing the rapid decrease in R eco, the latter thus becoming increasingly dominated by heterotrophic R S. The results of the ecoflux approach have been discussed in Unger et al. (2009). Including stable isotopes in the mass balance (eco-isoflux approach), however, revealed partially different results.

To test the eco-isoflux approach for consistency, we solved the mass balance presented in Eq. 3 for the flux rates of all individual components (Fig. 4). We found that not all measured fluxes and corresponding isotope ratios led to a sensible solution of the mass balance. The most reasonable solution was achieved using isofluxes from δ13Cres of soil, herbs and tree respiration and their respective measured fluxes and measured δ13CR data to solve Eq. 3 for R eco. In contrast, when using measured δ13CR in the eco-isoflux approach we detected significant deviations in the calculated component fluxes from those obtained by the ecoflux approach. This deviation was larger when the estimated component flux was small (Table 2; Fig. 4). Thus, the highest agreement (i.e., highest correlation and lowest offset) between the eco-isoflux and the ecoflux approach was achieved for R eco (R 2 = 0.72; AO = 8%; see Table 2; Fig. 4a), while largest deviations were found for the smallest component fluxes (R 2 = 0.33 and 0.02 for root and tree respiration see Table 2; Fig. 4c, d). Further, non-congruencies of summed mass balance estimates from R root and R SMO magnified the difference between ecoflux and eco-isoflux mass balance for R S, resulting in low correspondence between the approaches (R 2 = 0.21; AO = 31%; see Table 2; Fig. 4b).

Fig. 4
figure 4

Temporal changes in measured and scaled fluxes from a R eco (squares), b R soil (circles), c tree respiration (triangles), d R root (diamonds), e understory respiration (inverted triangles) and f R SMO (hexagons); measured and ecoflux values, grey symbols with lines; eco-isoflux values, black symbols; soil-isoflux values, white symbols. Note: apart from R soil, all respiration flux partitioning was generally during nighttime and foliage flux values are given in m2 ground and not leaf area. All isotope data entering the eco-isoflux mass balance are measured values. Black bars indicate nighttime. For abbreviations, see Fig. 2

Full size image
Table 2 Correlation coefficients (R 2) and average offset between correlated data points (AO; %) for correlations of measured and ecoflux estimated respiration fluxes (R) from ecosystem and ecosystem components [trees, understory, soil, soil microorganisms (SMO), roots] with the respective isotopic flux mass balances (eco-isoflux and soil-isoflux); n = 20
Full size table

Since respiration of soil, roots and SMO was measured independently of the ecosystem fluxes we were able to compare the different partitioning approaches exclusively at the soil scale (soil-isoflux approach; Eq. 5; Fig. 4b, d, f): root respiration calculated by soil-isoflux partitioning was in the same range and followed nicely the pattern of the ecoflux approach (Fig. 4d; Table 2). Outliers were only observed for the midday value of 20 May, where isoflux partitioning overestimated the root flux. Given that the root respiration flux was very small (<0.5 μmol m−2 s−1) these outliers are negligible. Respiration fluxes from total soil and SMO as estimated by the soil-isoflux approach revealed high agreement with the soilflux approach (Eq. 4; Fig. 4b, f; Table 2).

Discussion

The presented approach combines understory versus overstory C flux partitioning with the additional information provided by stable isotopes. We will discuss the potential mechanisms underlying the observed drought effects on R eco by disentangling the possible causes of variation in δ13Cres of the main ecosystem components. Further, we will discuss the benefits and limitations of including stable C isotopes in a flux-based mass balance at both soil and ecosystem scales.

Variation in δ13CS

Many studies found that R S and its isotopic composition are largely controlled by photosynthesis through its impact on root respiration (Steinmann et al. 2004; Göttlicher et al. 2006; Högberg et al. 2008; Bahn et al. 2009). Most of these studies, however, were conducted in dense forests with a high relative root contribution to soil CO2 efflux. Inversely, lower C allocation to the roots may explain smaller autotrophic respiration fluxes in less productive ecosystems (Subke et al. 2006). Our results for the oak-savannah fit this theory: low R S rates coincided with a low contribution of roots (15–28%) to R S, and heterotrophic organisms dominated overall R S (up to 85%; see Fig. 2i–l).

δ13Cres of both SMO and soil were substantially more depleted than foliage and root respiration (Fig. 3) and exhibited little variation in response to drought. At the end of the observation period δ13Cres values increased slightly. While R SMO declined with drought by ca. 1 μmol m−2 s−1, root respiration remained stable (see Fig. 4), and thus increased its relative contribution to total soil efflux explaining the slight enrichment in δ13Csoil with enhanced drought. However, overall changes in soil δ13Cres were not significant, thus we assume the utilization of a stable C source and only minor influence of photosynthetic discrimination. With increasing drought, diurnal variations in δ13Cres of SMO were found, similar to the pattern in root respiration (see Fig. 3). Soil microbes are known for their ability to assimilate C by anaplerotic CO2 fixation, where phosphoenolpyruvate carboxylase (PEPc) assimilates CO2 from the soil air (e.g., Wingler et al. 1996). Through mixing of atmospheric and respired C in the soil air spaces, anaplerosis-derived CO2 is substantially enriched compared to C input from C3 plants (Tu and Dawson 2005). Furthermore, through decreased input of photosynthetic assimilates, the C cycle of deeper soil layers is increasingly sustained by soil microbial biomass (Tu and Dawson 2005) and old SOM, thus leading to a more enriched 13C signature (e.g., Paterson et al. 2009). However, it has to be considered that anaplerotic C fixation by PEPc is not an efficient mechanism to sustain respiration under increased drought, but rather might contribute to the generally observed enrichment with deeper soil layers. Under drought conditions, when SMO respiration in the upper soil layers is restrained by low soil moisture, the influence of R S from deeper soil layers, where soil microbial communities have less access to plant-derived C and are supplied to a larger extent by anaplerotic CO2 fixation, might become higher and thus result in more enriched δ13Cres of SMO. However, there was no substrate input from plants in the trenching plots and there were no indications as to why the magnitude of anaplerosis-derived CO2 should be increased at night. More research will be needed to clarify the observed pattern.

Variation in δ13Cres of foliage

With progressing drought R eco and R S became increasingly similar, indicating a large decrease in canopy respiration (Davidson et al. 2006). In spite of the canopy-scale decrease, dark respiration of single leaves of the understory species T. guttata increased strongly (see Fig. 2a–d), which can be attributed to accelerated metabolism in dying tissue. The ~90% decrease in understory respiration (Fig. 4e; see also Unger et al. 2009) at the canopy scale was therefore entirely due to a drought-induced decrease in leaf area. Q. ilex on the other hand was able to extract sufficient water from deeper soil layers (e.g., Kurz-Besson et al. 2006; Pereira et al. 2007) and did not respond markedly to drought (Fig. 4c).

During the course of drought δ13Cres decreased by 2.4 ± 1.4 and 5.0 ± 2.0‰ in Q. ilex and T. guttata leaves, respectively (see Fig. 3). In contrast, controlled experiments found more enriched δ13Cres in drought-stressed plants as compared to controls, as drought-induced decreases in photosynthetic discrimination resulted in enriched substrates and thus increasing δ13Cres (Duranceau et al. 1999). However, under natural conditions drought is accompanied by increasing temperatures and VPD. In accordance with our findings, Tcherkez et al. (2003) showed that increasing temperatures led to more depleted δ13Cres. Further, old C sources may contribute more than 50% to overall respiration (Nogués et al. 2006). When respiratory demand exceeds C input during drought, C from more depleted C sources will be respired, particularly at night when assimilate pools decline, which was probably the case in T. guttata. Thus, increased respiratory substrate demand in combination with decreasing C fixation may induce the utilization of more depleted sources (e.g., fatty acids; see also Duranceau et al. 1999).

Further, substantial post-photosynthetic fractionation during dark respiration of foliage has been observed (Ghashghaie et al. 2003), which can be explained by heterogeneous distribution of heavy C within hexose molecules (Rossmann et al. 1991; Gleixner and Schmidt 1997). This intramolecular heterogeneity of 13C results in an ‘apparent fractionation’ during dark respiration depending on the balance between pyruvate dehydrogenase and Krebs cycle metabolic reactions due to the deviation of C to secondary metabolism (Tcherkez et al. 2003; Ghashghaie et al. 2003; Werner et al. 2007b; 2009; Priault et al. 2009). The observed depletion in foliage δ13Cres with drought (Fig. 3a–h) might, therefore, be explained by decreasing investment into secondary metabolism in response to decreasing C supply versus increasing C demand. Post-photosynthetic fractionation may also be responsible for the marked diurnal dynamics in foliage δ13Cres (especially in T. guttata) with increasing enrichment during the light period and subsequent depletion after dark (Hymus et al. 2005; Prater et al. 2006; Werner et al. 2007b, 2009; Priault et al. 2009). It was shown that strongest diurnal enrichment occurred in species with enhanced investment into secondary metabolism as well as increased maintenance versus growth respiration (Werner et al. 2007b, 2009; Priault et al. 2009). In contrast to these studies under controlled conditions, diurnal variability was larger in the fast-growing T. guttata than in the evergreen Q. ilex and increased with enhanced drought. Thus, we hypothesize that changes in δ13Cres of T. guttata were due to substantial substrate changes rather than a response in post-photosynthetic fractionation. On the other hand, in Q. ilex drought might have caused decreasing investment in secondary metabolism, explaining the decrease in diurnal amplitude of δ13Cres variations. Hence, under drought and low C assimilation, substrate changes and post-photosynthetic fractionation effects masked the rather small effects of photosynthetic discrimination on δ13Cres of the studied Mediterranean species.

Recently, δ13Cres has been shown to be markedly 13C enriched during the first minutes of darkness after illumination (e.g., Mortazavi et al. 2006; Gessler et al. 2009b; Werner et al. 2009), mainly in light-acclimated leaves, compared with the more depleted steady state of dark-adapted leaves. This was explained by light-enhanced dark respiration (LEDR) that would originate from the use of organic acids (e.g., malate) as respiratory substrates (e.g., Gessler et al. 2009b; Barbour and Hanson 2009; Werner et al. 2009). Hence, LEDR might play a role in the observed enrichment of daytime respiration in foliage. The extent to which rapid post-illumination changes are involved in diurnal dynamics in δ13Cres is largely unclear (Werner et al. 2009). However, Werner et al. (2009) showed that diurnal dynamics in δ13Cres occur independently from the rapid post-illumination dynamics. Moreover, they showed that time series of post-illumination δ13Cres measurements in Q. ilex leaves undergo diurnal enrichment of comparable magnitudes, when comparing the same times of dark adaptation. Incubation time of our samples in the dark was in all cases 10 min, thus providing comparable conditions. Furthermore, main shifts in leaf δ13Cres occurred during nighttime, when LEDR does not occur. Therefore, we suggest that LEDR only played a minor role explaining the observed daytime dynamics in foliage δ13Cres.

Variation in δ13Cr

Root respiration contributed only marginally to R eco (<10%), and remained stable even with senescence of aboveground biomass, which might be due to the C storage function of roots (e.g., Pregitzer et al. 2000). Metabolic processes of root respiration are not yet well understood (Barbour and Hanson 2009). Many laboratory studies (Klumpp et al. 2005; Badeck et al. 2005; Bathellier et al. 2008, 2009; Gessler et al. 2009b; Barbour and Hanson 2009) under well-watered conditions have shown that root respiration is slightly more depleted than shoot respiration. This was generally explained with anaplerotic CO2 fixation. Like SMO, roots contain PEPc, which fractionates in favor of 13C (Klumpp et al. 2005) and thus metabolizes enriched CO2, re-fixed from the root respiration flux, resulting in more depleted respiration. In contrast to these studies, root respired δ13CO2 of T. guttata became increasingly enriched under severe drought, whereas the opposite pattern was observed in leaves. As PEPc does not use gaseous CO2 but HCO3 − in the liquid phase (Mohr and Schopfer 1992) and thus is directly moisture dependent, a drought-induced reduction in PEPc activity might be a likely explanation for the increased enrichment in root δ13Cres. However, the influence of PEPc on respiratory metabolism in roots may be highly species dependent. Bathellier et al. (2009) found no marked effect of PEPc on δ13Cres of roots in Phaseolus vulgaris, whereas Gessler et al. (2009b) reported large effects in Ricinus communis.

Similar to our results, other studies found root (Gessler et al. 2007) as well as trunk respired CO2 (Damesin and Lelarge 2003; Brandes et al. 2006) to be 13C enriched as compared to the putative substrates (Cernusak et al. 2009). Gessler et al. (2007) found a marked nocturnal enrichment in δ13C of Eucalyptus phloem sap, probably due to fractionation during nocturnal starch remobilization. Fructose produced from starch break-down is isotopically heavier than sucrose produced in the light (Gleixner et al. 1998). Gleixner et al. (1998) also showed that tuber sucrose of potato plants appeared to be significantly enriched compared with leaf sucrose. T. guttata also produces tubers for enhanced carbohydrate storage. Thus, a nocturnal substrate change from freshly produced hexose to stored starch might explain the observed nocturnal enrichment.

Variation in δ13CR

Despite the strong decrease in R eco, δ13CR did not undergo marked temporal variations with drought (e.g., McDowell et al. 2004b; Lai et al. 2005; Ponton et al. 2006), though exhibiting large diurnal shifts (e.g., Bowling et al. 2003; Werner et al. 2006; Kodama et al. 2008). Flux partitioning indicated that changes in isotopic composition of R eco occur mainly upon changes in the relative contribution of component fluxes with different isotopic signatures. Similar to Tang and Baldocchi (2005) the pattern of R eco was mainly driven by soil moisture but the diurnal pattern was controlled by temperature (see Unger et al. 2009). A main contributor to R eco was R S (ca. 60–80% of R eco), while canopy respiration constituted a smaller flux (20–40%); these ranges are comparable to those of other studies (e.g., Goulden et al. 1996; Lavigne et al. 1997; Law et al. 2001; Davidson et al. 2006). The R S flux (particularly that of SMO) thus had the largest influence on δ13CR, with both δ13CR and δ13Csoil being more depleted compared to other ecosystem components.

Seasonal variations in δ13CR and δ13Csoil have been linked to meteorological events via their effects on canopy discrimination and substrate supply to root respiration (Ekblad and Högberg 2001; Bowling et al. 2002; McDowell et al. 2004b; Knohl et al. 2005; Mortazavi et al. 2005; Werner et al. 2006; Alstad et al. 2007; Högberg et al. 2008) and even to a drought-induced increase in ecosystem-scale water use efficiency (Ponton et al. 2006). However, such relations were not observed in this study: mean nocturnal δ13CR remained relatively stable with increasing VPD and T air, which was in agreement with gas exchange measurements, indicating only a minor drought response of tree photosynthetic discrimination (Q. ilex; Table 1). Assimilates of the vanishing understory (e.g., T. guttata) probably also played a minor role in δ13CR as the fluxes declined rapidly.

Large shifts in δ13CR (6‰) were found within the nocturnal cycle. Unexpectedly, however, these variations did not show a strong relationship with diurnal variations in the different component fluxes, as derived from a comparison of measured and eco-isoflux calculated values (see Fig. 3m–p).

The major problem here seems to be the applicability of the Keeling plot method at the ecosystem scale. Given the large differences between the δ13Cres of the component fluxes from soil, roots and foliage the sampling method of a vertical gradient may conflict with the law of a constant source flux signature, and thus produce bias in the two-source mixing model. Further, Schaeffer et al. (2008) showed that canopy structure and differences in atmospheric mixing might introduce a bias into Keeling plot estimates sampled over a height gradient. They found that δ13CR sampled at canopy height correlated with time-lagged VPD, while δ13CR sampled at ground level correlated with soil moisture. These correlations became stronger when sampling periods with strong mixing of ground and canopy air were excluded, suggesting that atmospheric dynamics play a large role in the estimate of δ13CR. Finally, it is known that small CO2 gradients (below 30 p.p.m.) can reduce confidence in the estimate of δ13CR (Pataki et al. 2003; Zobitz et al. 2006). Indeed, CO2 gradients obtained for ecosystem Keeling plots were on average only 36 p.p.m. Hence, δ13CR results need to be considered with care. Nevertheless, the lowest nocturnal variability in δ13CR was found on 3 June, when the δ13Cres of foliage and R S were most similar. This is in agreement with the hypothesis that changes in both signatures and relative contributions from the components may result in temporal shifts in δ13CR (for a thorough discussion see Werner et al. 2006).

Validity of isotopic partitioning approaches

While no study has yet measured the δ13C values of all major ecosystem components simultaneously for δ13CR, a few have measured several of these fluxes at once (Bowling et al. 2003; McDowell et al. 2004a; Mortazavi et al. 2005; Kodama et al. 2008); however, so far none of them has achieved mass balance (Bowling et al. 2008). Here, we attempted to close an isotopic flux mass balance at both ecosystem (eco-isoflux) and soil scales (soil-isoflux). At the soil scale (soil-isoflux) both partitioning approaches revealed good agreement (see Table 2; Fig. 4). However, we did not manage to close the mass balance at the ecosystem scale: especially when calculating small fluxes, the eco-isoflux and ecoflux approach showed somewhat large deviations (see Table 2; Fig. 4). Only when calculating R eco from the ecosystem component fluxes and their isotopic compositions did we find good agreement with measured R eco.

Apart from a methodical bias potentially associated with the flux measurements (e.g., pressure gradients in the soil chambers and the trenching approach; for a thorough discussion see Unger et al. 2009), there might be several reasons for the inability to close the eco-isoflux mass balance:

Firstly, there was good agreement of measured and calculated R eco in contrast to the somewhat large deviations of the component fluxes derived from the eco-isoflux approach. This indicates that, as discussed above, the Keeling plot measurements at the ecosystem scale were a large source of error in the eco-isoflux approach. At the soil scale, on the other hand, soil-isoflux partitioning showed good agreement with the soilflux estimates, since Keeling plots were more reliable because: (1) they were sampled using a well-mixed chamber system, thus eliminating the problems involved in sampling over a height gradient with different components signatures; (2) CO2 gradients were generally large (~80 p.p.m.); and (3) at the soil scale atmospheric dynamics are less important, as the soil chamber acts as a closed system.

Another potential source of error in the eco-isoflux approach might have been a missing flux. All main ecosystem components and their respective δ13Cres were measured, apart from stem and branch respiration from the trees that were included in the tree flux. Sensitivity analyses revealed that the addition of a small flux with natural signatures expected from stem and branch respiration would not change the results of the mass balance significantly. However, only a small flux with a very negative isotopic signature (e.g., from fossil fuel combustion) might explain differences in measured and calculated δ13CR (data not shown). On the contrary, for R S, a scale with fewer contributing components, the soil-isoflux approach worked very well.

Potential bias might have been introduced by using isotopic compositions from single leaves and root respiration as representative of the isotopic compositions of the understory, tree and root fluxes, resulting in error in the fluxes estimated by the eco-isoflux approach. Further, the understory flux data might have introduced error into R u estimates as these were from a grassland site adjacent to the studied oak forest (although they had similar conditions). However, these errors are expected to be minor as trees are widely spaced in the forest site and soil measurements for the understory flux partitioning were taken in the large areas between trees, where energy balance and understory flux are expected to be similar to those in the grassland site. Further, the seasonality of understory growth patterns is comparable between the two sites (Pereira et al. 2007) and somewhat different management and species composition did not largely affect the comparability of the understory fluxes from the different systems (see “Materials and methods”) during the study period.

Conclusion

Stables isotope ratios of ecosystem component fluxes carry valuable information on regulatory processes of an ecosystem’s response to drought. Large and rapid dynamics in isotopic compositions of foliage, root and soil respiratory fluxes were most probably due to: (1) post-photosynthetic fractionation, (2) changes in respired substrates, (3) drought-induced changes in PEPc activity, and (4) an increased relative contribution of R S from deeper soil layers with drought—effects masking the imprint of photosynthetic discrimination on δ13CO2 of both components and R eco.

We did not manage to fully close an isotope-based mass balance at the ecosystem scale, mainly due to uncertainties regarding ecosystem Keeling plots. However, promising results were achieved at the soil scale, indicating that isotopic mass balance works well under conditions of high CO2 gradients and stable sampling conditions.

Nevertheless, δ13CR is an important parameter in isotope-based C budget models and for an understanding of the controls over terrestrial C exchange processes. Thus, future research should focus on the achievement of improved methodology to measure or model δ13CR for mass balance approaches and to quantify the underlying processes of short-term dynamics of the isotopic composition of R eco.