Warming and elevated CO2 intensify drought and recovery responses of grassland carbon allocation to soil respiration

Photosynthesis and soil respiration represent the two largest fluxes of CO2 in terrestrial ecosystems and are tightly linked through belowground carbon (C) allocation. Drought has been suggested to impact the allocation of recently assimilated C to soil respiration; however, it is largely unknown how drought effects are altered by a future warmer climate under elevated atmospheric CO2 (eT_eCO2). In a multifactor experiment on managed C3 grassland, we studied the individual and interactive effects of drought and eT_eCO2 (drought, eT_eCO2, drought × eT_eCO2) on ecosystem C dynamics. We performed two in situ 13CO2 pulse‐labeling campaigns to trace the fate of recent C during peak drought and recovery. eT_eCO2 increased soil respiration and the fraction of recently assimilated C in soil respiration. During drought, plant C uptake was reduced by c. 50% in both ambient and eT_eCO2 conditions. Soil respiration and the amount and proportion of 13C respired from soil were reduced (by 32%, 70% and 30%, respectively), the effect being more pronounced under eT_eCO2 (50%, 84%, 70%). Under drought, the diel coupling of photosynthesis and SR persisted only in the eT_eCO2 scenario, likely caused by dynamic shifts in the use of freshly assimilated C between storage and respiration. Drought did not affect the fraction of recent C remaining in plant biomass under ambient and eT_eCO2, but reduced the small fraction remaining in soil under eT_eCO2. After rewetting, C uptake and the proportion of recent C in soil respiration recovered more rapidly under eT_eCO2 compared to ambient conditions. Overall, our findings suggest that in a warmer climate under elevated CO2 drought effects on the fate of recent C will be amplified and the coupling of photosynthesis and soil respiration will be sustained. To predict the future dynamics of terrestrial C cycling, such interactive effects of multiple global change factors should be considered.


| INTRODUC TI ON
During the mid-to late 21st century, continued consumption of fossil fuels is expected to increase atmospheric CO 2 concentrations by 300-400 ppm above current levels, leading to global warming of 1-3.7°C and an increase in the frequency and intensity of climate extremes such as drought and heatwaves (IPCC, 2014). These global change factors can cause perturbations of the two largest CO 2 fluxes between terrestrial ecosystems and the atmosphere, that is, gross primary productivity (GPP) and ecosystem respiration (ER), with potential feedbacks to climate change. Of particular importance is the close link between GPP and soil respiration (SR; i.e., the largest component of ER), as plants allocate a significant amount of recently assimilated C to root respiration and rhizo-microbial respiration, often referred to as the autotrophic component of SR Kuzyakov, 2006;Pausch & Kuzyakov, 2018). This link is especially strong and rapid for grassland ecosystems (Bahn et al., 2009;Kuzyakov & Gavrichkova, 2010). However, it is largely unknown whether and how the allocation of recently assimilated C to SR is affected by multiple interacting global change factors such as elevated CO 2 (eCO 2 ), warming (eT) and drought.
Various experiments and modeling studies suggest that in the future, the combination of eCO 2 and eT can have a complex impact on ecosystem C dynamics. Under eCO 2 , plants assimilate more C Wang et al., 2012), and allocate more of the photosynthesized C belowground, leading to increases in SR (Figure 1; Adair et al., 2011;Black et al., 2017;Kuzyakov et al., 2019;Song et al., 2019;Wan et al., 2007). Indirectly, eCO 2 also affects SR by preserving soil moisture due to increased plant water use efficiency (Franks et al., 2013;Morgan et al., 2011). On the other hand, warming can increase GPP in cold climates  and SR (Black et al., 2017) due to accelerated metabolic activity, but can also decrease SR when warming leads to a critical reduction of soil water content (SWC, Figure 1; Fang et al., 2018). Therefore, eCO 2 and eT in combination can either accelerate or reduce C fluxes (Pendall et al., 2013;Wan et al., 2007), depending on the degree to which negative effects of eT on SWC are counterbalanced by positive effects of elevated CO 2 (Figure 1; Blumenthal et al., 2018;Heisler-White et al., 2008;Huxman et al., 2004;Schwinning et al., 2004). Thus, the relative effects of eCO 2 and eT on SWC will also determine whether drought effects on GPP and soil and ecosystem respiration are diminished or exacerbated in a future warmer and CO 2 -rich environment ( Figure 1; Albert et al., 2011;Gray et al., 2016;Obermeier et al., 2017;Roy et al., 2016).
It is well established that GPP and SR are strongly reduced by drought, and that drought effects are more pronounced for GPP than for respiration (Frank et al., 2015;Reichstein et al., 2013;Schwalm et al., 2010). In consequence, the autotrophic component of SR, which relies strongly on C supply from photosynthesis, was shown to be more strongly affected by drought than the heterotrophic component, which derives from the turnover of soil organic matter (Zhou et al., 2016). Isotopic pulse labeling studies suggest that indeed drought reduces the amount of photosynthetically fixed recent C allocated to roots and soil microorganisms (Fuchslueger et al., 2014;Hasibeder et al., 2015;Karlowsky et al., 2018;Sanaullah et al., 2012) and in consequence to SR (Barthel et al., 2011;Blessing et al., 2016;Burri et al., 2014;Hagedorn et al., 2016;Ingrisch et al., 2020; Figure 1). Drought has also been shown to increase the mean residence time of C in the plant-soil system (Chomel et al., 2019;Karlowsky et al., 2018;Ruehr et al., 2009;Zang et al., 2014). However, there is scarce and conflicting evidence to what degree the reduction in the belowground respiratory release of recent C under drought is caused by a reduced supply of recent C (from GPP) or by reduced belowground metabolic C demand. The latter would result in altered belowground partitioning of recently assimilated C and change the fraction of assimilates respired belowground and their residence times (Burri et al., 2014;Ingrisch et al., 2020). While the effects of drought on the coupling of GPP and SR are comparatively well understood under current environmental conditions, there is a major gap in our understanding of how the fate of recent C in the plant-soil system, and in particular soil respiration, during drought and drought recovery is modulated by future conditions, through direct and indirect effects of eCO 2 and eT on SWC and C supply versus demand in belowground respiration (Figure 1).
The overall aim of this study was to understand the effects of drought on grassland C dynamics and the coupling of GPP and SR, and how these drought effects are altered by warming and elevated CO 2 . We exposed sections of a managed temperate grassland to F I G U R E 1 Conceptual framework and hypothesized effects of elevated CO 2 (eCO 2 ), warming (eT) and drought on the transfer of recently assimilated carbon (green boxes) to soil respiration. The signs denote the direction of effects. Measured variables are indicated by solid boxes combined experimental manipulations of temperature, atmospheric CO 2 concentrations and drought, and conducted in situ 13 CO 2 pulse labeling campaigns during the peak drought and the recovery periods to trace the allocation of newly assimilated C from plants to soil and to SR. We tested the hypotheses that (1) under ambient conditions drought would reduce ecosystem C uptake, the absolute and relative amount of C allocated to SR and its mean residence time; (2) future conditions with eT and eCO 2 would lead to increased C uptake and SR through increased allocation to belowground fluxes and (3) eT and eCO 2 would reduce drought effects on SWC and the coupling of GPP and SR and thus mitigate the drought impact on the C cycle, and enhance post-drought recovery of SR by increased plant C supply. Furthermore, we expected that (4) the fraction of recent C remaining in the plant-soil system would be increased by drought, more strongly under ambient than under future conditions.

| Study site
The study was conducted as part of the multifactor manipulation experiment "ClimGrass," which is located at the Agricultural Research and Education Centre, Raumberg-Gumpenstein, Austria (47°29′38″N, 14°06′03″E) at an elevation of 700 m (Piepho et al., 2017). The geographic location and climatic conditions (mean annual temperature: 7.2°C; mean annual precipitation: 1000 mm) are representative of a larger part of montane grasslands in the Alps. The study site is composed of a nutrient-rich, managed mountain grassland on a Cambisol consisting of 44.2% sand, 47.6% silt and 8.3% clay (Deltedesco et al., 2019(Deltedesco et al., , 2020. The vegetation is dominated by tall oat grass (Arrhenatherum elatius L.), orchard grass (Dactylis glomerata L.), golden oat grass (Trisetum flavescens L.), meadow fescue (Festuca pratensis L.), bird's-foot trefoil (Lotus corniculatus L.) and white clover (Trifolium repens L.). The grassland is cut three times during the growing season (end of May, end of July and early October) and fertilized using mineral fertilizer to replace the amount of nutrients removed by the harvests. In 2017, the year of the pulse labeling experiment, the annual precipitation at the site was 1281 mm and the mean air temperature was 8.8°C.

| Experimental setup and drought simulation
The experiment was performed on eight plots exposed to four treatments: (i) control (i.e., ambient conditions), (ii) drought, (iii) eT and eCO 2 (eT_eCO 2 ) and (iv) eT_eCO 2 combined with drought. Air temperature and atmospheric CO 2 concentration were increased by 3°C and 300 ppm above ambient, which corresponds to the climate scenario for the end of the 21st century following to the RCP 8.5 model trajectory (IPCC, 2014). CO 2 enrichment was achieved via a mini free air CO 2 enrichment (FACE) system (Miglietta et al., 1997;Obermeier et al., 2017;Reich et al., 2014), which fumigates the plot during daytime through a circular tube surrounding the plots ( Figure   S1; Piepho et al., 2017;Pötsch et al., 2019). Infrared lamps were used to heat the canopy surface temperature 3°C above ambient temperature. Mock frames and fumigation rings and mock IR-lamps were installed in all control plots to ensure that environmental conditions in plots differed only concerning the tested environmental drivers.
In each of the eight plots, 12 stainless steel cylinders of 30 cm diameter and 60 cm height were inserted into the soil during 2014 ( Figure S1), which served as intact but physically distinct grassland units ("mesocosms") precluding any lateral transfer of labeled C.
Mesocosms were larger than the ones previously used successfully in a grassland pulse labeling experiment testing for effects of drought on C dynamics Ingrisch et al., 2020;Karlowsky et al., 2018). In the study year 2017, we tested for potential plot effects of the mesocosm experiment using aboveground biomass of all 12 mesocosms contained in each plot (n = 4 treatments*2 plots*12 mesocosms = 96). Biomass was sampled during the first annual harvest (cut 1, 2017), which occurred prior to the drought experiment. To test whether "plot" had an effect on mesocosms productivity, we used linear mixed-effect models (R-package lme4 and lmerTest) with "treatment" as fixed effect and "plot" as random effect. The variance of the random effect was very small (2.6) in comparison to the residual variance (148.8) and had only minor effects on the standard errors of means (in comparison to linear model without random effects). Thus, we concluded that plot effects are negligible in this experiment and that the individual mesocosms used in the pulse-labelling study constitute independent units and can therefore be considered "true" replicates. For our study, each of the four treatments (control, drought, eT_eCO 2 , eT_eCO 2 × drought) was replicated with three mesocosms randomly selected from the two plots per treatment. Ecosystem CO 2 fluxes and soil respiration measurements were made on different sets of mesocosms to avoid disturbances through overlapping measurements and samplings.
Soil temperature and volumetric soil water content (SWC) were measured using probes (S-TMB and 10HS, Onset computer corporation) installed in the main rooting horizon 5 cm below the soil surface in two mesocosms located in two different plots per treatment and were recorded (HOBO Micro Station H21-002; Onset computer corporation). The SWC dataset was calibrated using gravimetric soil water content ( Figure S2), which was determined by oven drying of fresh soil samples repeatedly collected in all pulse-labelled mesocosms. Air temperature, photosynthetically active radiation (PAR) and relative humidity were continuously measured all around the year by a climate station based at the ClimGrass facility ( Figure S3).
The drought treatment was applied using automated rainout shelters, which covered the respective plots during rainfall ( Figure   S1). The drought treatments started on doy-112 and ended on doy-209 by re-wetting the drought treatment plots with 40 mm of previously collected rainwater for 2 h. Corresponding to the common management of the grassland, the vegetation in all the treatments was cut on doy-208, which was also the last day of the drought treatment. Hence, the post-drought recovery phase concurs with post-cut regrowth phase. Thus, three distinct periods were defined ( Figure S4): peak drought (doy-190 to 208), rewetting (doy-209) and recovery (after doy-209).

| Ecosystem CO 2 flux measurements
Net ecosystem productivity (NEP), ecosystem respiration (ER) and gross primary productivity (GPP) were measured using closed dynamic chambers as described by . Briefly, NEP was measured using a cylindrical transparent plexiglass chamber with a 30 cm diameter and 100 cm height. The chamber was ventilated with fans fitted inside. The chamber had a hole in the top to avoid any pressure effects while placing the chamber, which can be closed. The chambers had sensors inside to measure CO 2 concentration (GMP 343, Vaisala), air temperature and water vapor (HMP 75, Vaisala) for 1 min at 5-s intervals. Mesocosms were always first measured under light conditions to assess NEP, followed by a measurement of ER, for which the chamber was darkened with an opaque cloth. Measurements were quality controlled visually (Pirk et al., 2016) and the CO 2 flux rates were calculated by linear regression, as described by similar studies (e.g., Schmitt et al., 2010). GPP was calculated as the difference between paired measurements of NEP and ER. To ensure comparability across measurements, we only present light-saturated GPP max and NPP max , defined as fluxes at a photon flux density (PFD) greater than 1400 µmol m −2 s −1 . The measurements were made on three mesocosms per treatment during peak drought (doy-191), and at two time points during recovery (doy-213, 221). The measurements were made in random order of mesocosms on days with clear sky during late morning hours (9:00-13:00 CEST).

| 13 CO 2 pulse labeling
Two 13 CO 2 pulse labeling campaigns were performed, the first during peak drought and the second during the recovery period. For the "peak drought" campaign, 12 mesocosms (three per treatment) were labeled on three consecutive days during doy-197 to 199 ( Figure S4). During the recovery campaign, different sets of 12 mesocosms were labeled on three consecutive days during doy-218 to 220. On both campaigns, labeling was done on paired (drought-control) sets of mesocosms within ambient and eT_eCO 2 treatment.
The pulse labeling procedure was similar to previous studies (Bahn et al., 2009;Hasibeder et al., 2015;Ingrisch et al., 2020). Pulse labeling was performed on days with clear sky between 09:00 and 15:00 CEST. Transparent plexiglass chambers were placed airtight on each mesocosm. Chambers were ventilated and temperature stabilized using fans and by circulating cold water through tubes inside the chamber. Air temperature, CO 2 concentration (Li-840A, Li-Cor Inc) and 13 C/ 12 C ratio (G2201i Analyzer, Picarro Inc) inside the chamber were continuously monitored and were used to calculate the 13 CO 2 uptake during labeling. This yielded similar results compared to two alternative approaches to assessing 13 CO 2 uptake (1) through GPP measured on the plots immediately prior to labeling and (2) through the amount of 13 C recovered in plant biomass immediately after labeling ( Figure S5). Photosynthetically active radiation (PQS1, Kipp & Zonen) was monitored outside the chamber. Before labeling, the CO 2 concentration inside the chamber was reduced to ~200 ppm by plant photosynthesis in the closed chamber and by scrubbing using soda-lime. Labeling started with injecting 15 ml of 99%-13 CO 2 (Sigma-Aldrich) into the chamber with a syringe.
Consecutive pulses of labels were added to maintain 40-70 atom% 13 C inside the chamber over the complete labeling time of 50-70 min. During labeling, the mean CO 2 concentration in the chamber was 600 ppm ± 200 ppm and mean air temperature was 25°C ± 5°C.

| Soil respiration and isotopic composition
Soil respiration (SR) and its isotopic composition were measured continuously on the mesocosms that were subjected to labeling. Initially, 12 mesocosms labeled during drought were measured on the last 10 days of drought and 5 days after the rewetting. Afterwards, the soil respiration chambers were moved to the second set of mesocosms that were labeled during the recovery period, where they measured from doy-217 to 228.
A custom made steady-state measurement setup as described by Ingrisch et al. (2020) was used to measure soil respiration and isotope composition ( Figure S1). Briefly, it includes 12 cylindrical steady-state flow-through chambers made from white PVC tubes with 4.5 cm diameter. Each chamber had two connections that is, inlet and outlet. The outlet connection had an inner diameter of 4 mm diameter and the inlet tube had a diameter of 3 cm. The inlet of each chamber was connected to a 25 L polyethylene cylinder which acted as buffer volume to stabilize the concentration of CO 2 entering the chamber. The chambers were constantly flushed with air from the buffer volume at a flow rate of 200 ml min −1 to ensure steady-state conditions. The inlet/buffer volume and outlet of all the chambers were connected to an online isotope analyzer (G2201i Analyzer, Picarro Inc) through a custom-made valve multiplexing system. The valve multiplexer switches between the inlet and outlet of each chamber and flushes all chambers continuously with air from the buffer volume to maintain steady-state conditions in the chambers. To assess potential effects of chamber size and setup on soil respiration fluxes, we compared soil respiration data during peak drought with those obtained with using larger (21 cm diameter) automated chambers (Li 8100-104, Li-Cor Inc.) from the same experiment, but on different larger (3 m 2 ) plots (Reinthaler et al, in preparation). We found that the soil respiration data obtained with the custom-made smaller chamber setup used for the present study tended to be higher (on average 14%-25%), but overall covered a similar range as observed with the Licor 8100-104 chambers in the larger plots and yielded comparable treatment effects.
The isotope analyzer measures the concentration of isotopologues of CO 2 ( 12 CO 2 and 13 CO 2 ). Each chamber measurement took 12 min: the inlet was measured for 180 s, the outlet was measured for 360 s and then the inlet was measured again for 180 s. A measurement cycle covering all 12 chambers took 2.4 h. Three calibration gases (400, 1500 and 3000 ppm) with known isotopic composition were measured at the end of each measurement cycle to allow individual span-offset calibration of the analyzer for the two CO 2 isotopologues. The calibration gas with 3000 ppm and δ 13 C of −-6.35‰ contained >30 ppm of 13 CO 2 and exceeded the maximum amount of 13 CO 2 measured during the chase period, thus allowing us to calibrate the analyzer across the full range of 13 CO 2 observed in soil respiration (Bowling et al., 2003).
To assess the effect of physical back diffusion of 13 CO 2 from the soil, two additional mesocosms were pulse-labeled under dark conditions. The same pulse labeling protocol as described above was followed, but the chambers were closed with an opaque cloth to avoid photosynthetic tracer uptake. SR and its isotopic composition were continuously measured for 2 days from labeling. The cumulative amount of 13 CO 2 efflux from back diffusion ranged from 3.6% to 4.3% of the mean cumulative 13 CO 2 respired from the soil after labeling under ambient conditions ( Figure S6). The correction based on the 13 CO 2 back diffusion data increased the 13 CO 2 mean residence times during peak drought by 8.9% and 30.6% in control and drought treatments, respectively, and during recovery by 13.2% and 7.5% in control and drought treatments, respectively ( Figure S7).
However, the correction for back diffusion did not alter the treatment effects on 13 CO 2 respiration dynamics and mean residence times. These results were added in the supplementary section.

| Sampling of plant and soil material
Aboveground and belowground plant material and soil were sampled before (i.e., natural 13 C abundance controls, collected immediately next to the mesocosms) and 192 hours after each pulse labeling.
From each mesocosm, a composite sample of two soil cores with a diameter of 2 and 10 cm depth was taken after sampling plant shoots above this coring area. Soil samples were immediately sieved to 2 mm and fine roots were picked out manually. All samples were immediately frozen and stored in the field using liquid nitrogen. In the laboratory, samples were then freeze-dried for 48 hours, ground and analyzed for bulk δ 13 C and C concentration in each pool using an elemental analyzer (EA 1110: CE Instruments), coupled to a Finnigan MAT Delta Plus isotope ratio mass spectrometer (Thermo Fisher Scientific).

| Data analysis and statistics
The SR (in µmol m −2 s −1 ) was calculated using time-averaged inlet and outlet CO 2 concentrations, flow rate and the base area of the chamber.
The atom fraction of 13 CO 2 was, The 13 C atom fraction χ( 13 C) of SR was calculated as, The χ 13 C of SR was corrected for χ 13 C in natural abundance (na) in SR which was derived from pre-labeling measurements to calculate the fraction of 13 C label in SR The absolute amount of 13 CO 2 (Excess 13 CO 2 (abs)) respired (in µmol m −2 s −1 ) from soil was calculated as, The cumulative amounts of soil-respired 13 CO 2 over a period of 192 h after each pulse labeling were calculated by integrating 13 CO 2 efflux rates following the trapezoid rule, that is, by linear interpolation between adjacent data points. Therefore, adjacent data points were averaged, multiplied by elapsed time between those data points T i − T i−1 and summed over the cumulation period.
The 13 CO 2 label uptake was calculated using ( 13 C) of CO 2 measured inside the labeling chamber during labeling and GPP measured during pulse labeling ( Figure S5). Here, the "T" represents the time elapsed during labeling.
The amount of label 13 C recovered in plant biomass (shoot + root) and soil (excess 13 C biomass ) was calculated as, Here 13 C na and 13 C sample are the 13 C atom fraction measured in plant and soil samples before and after labeling, respectively. C pool represents the amount of C in plants and soil on a per m −2 soil surface basis, respectively.
The coefficient of variation of each chamber-outlet measurement was calculated for CO 2 concentration and χ 13 C as the ratio of standard deviation and mean value. Measurements with a coefficient of variation greater than 0.1 were considered as outliers and removed prior to further analysis. The proportion of flux estimates that were removed based on this filter was 6.2%. A spline function (R function, "smooth.spline," span = 0.5 day) was fitted to SR and abs 13 C time series of every mesocosm measurement, to fill data gaps.
Then the values were predicted at a 2.4-h interval (representing the measurement interval) based on individual splines to allow grouping between replicates.
An exponential decay function (R function "nls") was used to calculate the mean residence times of 13 C tracer in SR as the amount of time required to reduce abs 13 CO 2 respired to 1/e times its initial value (Bahn et al., 2009;Kuzyakov & Gavrichkova, 2010). The abs 13 CO 2 respiration rates were fitted with an exponential decay model y = a + e bx (R function "nls"), where b corresponds to the reciprocal of the mean residence time.
Multiple linear regression analysis was used to analyze the influence of environmental drivers, namely PAR, soil temperature and SWC on soil respiration and on soil-respired 13 CO 2 . To account for the overarching exponential decay in the latter, the residuals of the exponential model were used (hereafter referred to as " 13 CO 2 rsd ").
This transformation focused our analysis on diel patterns by removing variation in the data caused by the exponential decay effect of time (Bahn et al., 2009). This analysis was performed using linear that the environmental driver leads before response variable and that, therefore, the flux lags after the environmental driver. To examine the time delay in the heat transfer from sunlight to different soil depths, the lag of soil temperature at two soil depths (3 and 5 cm) to PAR was also tested using cross-correlation analysis (see Figure S8).
The effects of drought, eT_eCO 2 and their interaction on CO 2 fluxes (GPP, ER, NEP and SR) and excess 13 C (absolute and relative) in plant, soil and soil respiration were tested for each measurement campaign separately using permutational ANOVA (R-package

| Ecosystem CO 2 fluxes
Under warming combined with elevated CO 2 (eT_eCO 2 ), soil water content (SWC) was reduced on average by 14% compared to ambient conditions ( Figure S2c,d). Drought reduced SWC to less than 15% vol under both ambient and eT_eCO 2 conditions. The simulated rain event ending the drought treatments did not fully restore SWC and only after a natural rain event a week after the rewetting, SWC in drought-treated plots recovered to values observed in the control treatment.
eT_eCO 2 did not significantly affect CO 2 fluxes on the ecosystem scale ( Figure 2, Table S1). During peak drought (doy-192), GPP max was reduced by 48-50% and ER was reduced by 28% relative to the controls, irrespective of ambient or eT_eCO 2 conditions (Table S1).
Phytomass in all mesocosms was harvested the day before the rewetting, which strongly reduced GPP max . After rewetting, drought had negative legacy effects on the recovery of GPP max from harvest under ambient, but not under eT_eCO 2 conditions, where recovery was generally faster than under ambient conditions. ER was not affected by cutting, but increased with progressing recovery under eT_eCO 2 .

| 13 C tracer in plant, soil and soil respiration
Drought reduced the amount of excess 13 C remaining in plants 8 days after pulse labeling under both ambient and eT_eCO 2 ( Figure 3a; Table S2). The amount of excess 13 C recovered in soil was generally small and was not significantly affected by drought under ambient conditions, but was reduced to below detection limit under eT_eCO 2 (Figure 3b). Under drought, the total amount of 13 C respired belowground was strongly reduced, the effect being stronger under eT_eCO 2 than under ambient conditions (Figure 3c).
Drought did not affect the proportion of 13 C label (relative to the total amount taken up) in plant biomass (Figure 3d; Table S2), but it reduced the proportion of 13 C respired in soil and remaining in soil under eT_eCO 2 (Figure 3e,f).

F I G U R E 2
Gross primary productivity (GPP max ) and net ecosystem productivity (NEP max ) at photon flux density >1400 µmol m −2 s −1 and ecosystem respiration (ER) measured under ambient and eT_eCO 2 conditions during (a-c) peak drought (doy-191) and (d-i) at two time points during recovery. Recovery 1 and Recovery 2 were assessed 4 (doy-213) and 12 days (doy-221) after rewetting, respectively. The grassland was cut one day prior to rewetting. Error bars indicate the standard deviation of the mean (n = 3). Asterisks denote significant differences between control and drought treatments within ambient and eT_eCO 2 treatments (permutational ANOVA, p value: 0.05 > * >0.01 > ** >0.001)

F I G U R E 3
Absolute (a-c, g-i) and relative amount of 13 C label (d-f, j-l) in plants, soil and cumulative soil respiration 8 days after pulse labeling during drought (left) and recovery (right). Relative amounts of 13 C have been normalized to 13 C uptake in the respective mesocosms.
Error bars indicate the standard deviation of the mean (n = 3). Asterisks denote significant differences between control and drought treatments within ambient and eT_eCO 2 treatments (permutational one-way ANOVA, p value: 0.01 > **) During recovery from drought, the absolute amount and proportion of 13 C in plant biomass and in soil respiration remained largely reduced under ambient conditions, but less so under eT_eCO 2 (Figure 3g-l). Across all treatments, during the recovery period, a distinctly larger amount of 13 C remained in the soil compared to the drought period, amounting to up to 108 mg 13 C m −2 and up to 16% of the 13 C label taken up by the plants.

| Dynamics of soil respiration and respired 13 C in relation to environmental drivers
In the absence of drought, eT_eCO 2 enhanced diel dynamics and doubled SR (Figure 4b and Figure S9; Table S1). Under drought, cumulative SR was significantly decreased under both ambient and eT_ eCO 2 conditions ( Figure 4c; Table S1) We tested for potential effects of environmental drivers, including soil temperature, soil water content (SWC) and photosynthetically active radiation (PAR) on the diel dynamics of soil respired CO 2 and 13 CO 2 (i.e., the residuals from the background decay trend, 13 CO 2 rsd ) using time-series-regression analysis ( Figure 5; SWC is not displayed as it did not show any distinct diel variation; Tables S3   and S4). Under ambient conditions soil respired CO 2 and 13 CO 2 rsd lagged after PAR by 2.4 h (Figure 5a,b; Tables S3 and S4). By contrast, soil temperature lagged after soil respired CO 2 and 13 CO 2 rsd (Figure 5a,b; Tables S3 and S4). Drought dampened the diel dynamics of soil respired CO 2 and 13 CO 2 rsd (Figure 4a,d) and increased the time lags of soil respired CO 2 following PAR (Figure 5a,b; Tables   S3 and S4).
Under eT_eCO 2 , the diel variation of soil respired CO 2 and 13 CO 2 rsd was higher (Figure 5a,b) and the lag of 13 CO 2 rsd after PAR was shorter compared to ambient conditions ( Figure 5; Tables S3   and S4). Under eT_eCO 2 , drought did not affect the diel dynamics of soil respired CO 2 and 13 CO 2 rsd (Figure 4) but increased the explanatory power of PAR on soil respired CO 2 when soil respired CO 2 was lagged after PAR by 2.4 h (Figure 5a,b; Table S3). Under eT_eCO 2 , soil temperature was generally lagged after 13 CO 2 rsd (Figure 5b).
During recovery, the regression coefficients of environmental drivers on soil respired CO 2 and 13 CO 2 rsd in both drought treatments were not significantly different from the respective control treatments, indicating that the post-drought effects on the temporal lags diminished (Figure 5c,d; Tables S3 and S4).

| DISCUSS ION
While it is well established that photosynthesis can play a major role in driving SR and its dynamics (Kuzyakov & Gavrichkova, 2010;Vargas et al., 2011;Zhang et al., 2018), the coupling of photosynthesis and SR in response to interacting global change drivers is so far poorly understood. Here, we combined time-series analysis and in situ isotopic pulse labeling to assess the effects of drought and drought recovery under ambient versus future conditions of warming combined with elevated CO 2 on ecosystem C dynamics and the linkage between plant C uptake and SR. We found that the combined effects of warming and eCO 2 increased SR and the proportion of recently assimilated C respired in soil. At the same time, it altered the drought-and post-drought responses of ecosystem CO 2 fluxes and the partitioning of recently assimilated C between plants, soil and SR, highlighting significant interactive effects between the three global change factors.

| Carbon dynamics and drought responses of recently assimilated C in soil respiration under ambient conditions
Our time-series analysis indicates that the diel variation of SR was directly related to PAR, which was followed by SR by c. 2 h ( Figure 5). Radiation can affect respiration by increasing photosynthesis and thereby C supply to the autotrophic component of respiration (roots and the associated rhizosphere microbiome), or by warming the soil and therefore enhancing metabolic rates (Kuyzakov & Gavrichova, 2010). Our analysis indicates that temperature was not the primary driver for SR dynamics, because even in the uppermost soil layer (3 cm) temperature lagged behind PAR and respiration by several hours (Figure 5; Figure S8). This suggests that diel SR dynamics were primarily driven by photosynthesis, and thus by the autotrophic rather than the heterotrophic component of SR (Vargas et al., 2011).
Through 13 CO 2 pulse labeling and analyzing the temporal variability in tracer efflux from soil, we were able to investigate the autotrophic component of SR further. We found that soil respired 13 CO 2 was coupled to PAR similarly as the overall SR, but with higher explanatory power (Tables S3 and S4), which supports the notion of a preferential use of recently assimilated C in autotrophic soil respiration of temperate grassland under non-water limiting conditions (Bahn et al., 2009. Drought under ambient conditions reduced GPP more than ecosystem and soil respiration, leading to reduced net C uptake (Figure 2), and reduced allocation of recently photosynthesized C to SR (Figure 4c,f; hypothesis 1), which is consistent with findings from earlier C allocation studies (Barthel et al., 2011;Hagedorn et al., 2016;Ingrisch et al., 2020;Ruehr et al., 2009). Our understanding of drought effects on the partitioning of recent C to SR is still controversial: previous studies found that the proportional allocation of C to soil and root respiration grasslands may be increased or reduced under drought (Burri et al., 2014;Hasibeder et al., 2015;Ingrisch et al., 2020). In our study, the relative amount of F I G U R E 4 Temporal dynamics of soil respiration and environmental drivers measured for 120 hours after pulse labeling during the peak drought (a-j) and the recovery period (k-t). (a, b, k, l) Dynamics and (c, m) cumulative of soil-respired CO 2 . (d, e, n, o) Dynamics and (f, p) mean residence time of soil respired 13 CO 2 . The mean residence time was calculated from the decay rate of the exponential model (y = a + e bx ; shown as line and colored ribbon) fitted on soil respired 13 CO 2 . (g, h, q, r) Residuals from the fitted exponential model ( 13 CO 2 rsd ). (I, j, s, t) Dynamics of soil temperature and PAR. Shaded areas in the background denote nighttime (8 pm-5 am). Error bars indicate the standard deviation of the mean (n = 3). Asterisks denote significant difference between control and drought treatments within ambient and eT_eCO 2 conditions (permutational one-way ANOVA, p value: 0.01 > **) 13 C respired in soil was reduced by drought (Figure 3f), which could reflect changes in source-sink relationships and in C demand for osmolytes and storage versus catabolic C demand (Hasibeder et al., 2015), as discussed on more detail in the subsequent section.

| Carbon dynamics and drought responses of recently assimilated C in soil respiration under warming and elevated CO 2
The warming and elevated CO 2 scenario tested in our study corresponds to a likely scenario projected for the Alpine region in the coming decades (+300 ppm CO 2 , +3°C; Gobiet et al., 2014;IPCC, 2014). While the combination of eT and eCO 2 had no clear effects on GPP (Figure 2a), it resulted in a significant increase in SR compared to ambient conditions (Figure 4b,c; Table S1). This was expected because eCO 2 increases belowground C input and turnover (Kuzyakov et al., 2019;van Groenigen et al., 2017;Yue et al., 2017;Zhou et al., 2016), which is also reflected by a higher absolute amount of recent C respired ( Figure 3c) and a reduced lag of 13 CO 2 efflux to PAR ( Figure 5b); furthermore, under nonmoisture limiting conditions, eT increases microbial activity and respiration (DeAngelis et al., 2019;Melillo et al., 2002;Yanni et al., 2020; but see Alvarez et al., 2018; Figure 1). Correspondingly, the proportion of recent C (relative to uptake) in SR increased ( Figure 3f; Table S2; hypothesis 2), which suggests that in a future environment the increased source strength caused by eCO 2 overrides possible effects of eT on increased C demand for belowground metabolism (sink strength).
The future eT_eCO 2 scenario also led to a decrease in SWC in control plots, but less so under drought conditions, where SWC was similar under both ambient and eT_eCO 2 conditions ( Figure   S2). This indicates that the negative effect of eT on SWC was more pronounced than the water-saving effect of eCO 2 under nondrought conditions (see also Morgan et al., 2011), but that the antagonistic effects of eT and eCO 2 cancelled each other out during drought. Interestingly, drought effects on SR and on the amount and the proportion of recent C allocated to SR were nevertheless more severe under eT_eCO 2 compared to the ambient conditions (Figure 3c,f; Table S2). However, strikingly, under drought the diel dynamics of SR persisted under eT_eCO 2 (Figure 4b), and diel peaks remained strongly related to PAR (Figure 5a), suggesting that the enhancing effect of eCO 2 on autotrophic respiration persisted also during conditions of severe drought. This apparent discrepancy between the reduced use of 13 C in SR and the sustained coupling of photosynthesis and SR in drought under eT_eCO 2 can be reconciled if one considers that the allocation of recent C could have shifted dynamically between storage and respiration as drought progressed: at the point of pulse labeling a significant portion of recent C could have been preferentially allocated to starch and/ or osmotically active fructans and sugars (Barthel et al., 2011;Hasibeder et al., 2015;Karlowsky et al., 2018), reducing the fraction of 13 C in SR. When demand for these compounds was satisfied as drought progressed, an increasing portion of subsequently photosynthesized C might again have been preferentially used for immediate catabolic processes, indicated by the sustained coupling of PAR and SR during peak drought. This hypothesized dynamic response of the use of freshly assimilated C for belowground F I G U R E 5 Time-series regression analysis of soil respiration in relation to photosynthetically active radiation (PAR) and soil temperature. The regression coefficients (β) of PAR and soil temperature on (a, c) soil respired CO 2 and (b, d) residuals from the exponential model of soil respired 13 CO 2 ( 13 CO 2 rsd ) in dependence of time lags between variables during (a, b) peak drought and (c, d) during the recovery period. Closed symbols denote whether the regression coefficient is significantly different from zero (p < 0.05), open circles indicate the lack of a significant difference from zero. Asterisks denote a significant interaction of env driver and drought treatment and the on soil fluxes (p value: 0.05 > * >0.01 > ** >0.001 > ***). See Tables S3 and S4 for standardized regression coefficients to compare the relative effect size of drivers processes under drought is consistent with our finding concerning the mean residence time of 13 C (Figure 4f), whose increase under eT_eCO 2 conditions indicates a reduced 13 C turnover, possibly caused by initial preferential storage and subsequent slow release of small amounts of 13 C for fuelling respiration.

| Recovery responses and ecosystem implications
When assessing the overall resilience of ecosystem processes to drought, it is essential to also evaluate post-drought recovery effects , given their importance for the ecosystem C budget (Frank et al., 2015). We found that eT_eCO 2 enhanced the recovery of GPP compared to the ambient conditions (Figure 2d,g; hypothesis 3), which supports the finding from an earlier experimental study (Roy et al., 2016) and a larger-scale assessment (Schwalm et al., 2017). Drought effects on SR recovered under both ambient and eT_eCO 2 conditions (Figure 4k-m and S9), but eT_eCO 2 accelerated the recovery of the amount and the speed of recently assimilated C respired from soil (Figures 3i,l and 4p; hypothesis 3). These results suggest that while eT_eCO 2 can increase the severity of drought effects on grassland C dynamics, it can also increase recovery rates and the amount of recent C allocated to belowground metabolic processes.
From an ecosystem C budget perspective, it is essential to quantify not only the fraction of assimilated C, which is returned to the atmosphere by soil respiration, but also the fraction and the amount remaining in the plant-soil system (Hartmann et al., 2020;Jiang et al., 2020). Our pulse-labeling study showed that by the end of the respective chase periods only a small portion (1%-16%) of recently assimilated C was recovered in the soil, while 16%-45% was retained in the plant biomass, both the fractions and the amount being larger during the recovery period (Figure 3g This shift of the use of recent C from below-to aboveground respiration under drought was unexpected. It could have been caused by increased water stress-induced metabolic demand, which has been suggested to be species-specific and related to stress intensity (Dahal & Valnerberghe, 2017;Rowland et al., 2021;Varone & Grattani, 2015), but has not yet been studied for grassland species. Interestingly, during recovery from drought a distinct drought legacy could be observed under ambient conditions, leading to a reduced allocation of recent C to plant biomass and SR, while no significant reductions were found under eT_eCO 2 . This highlights the potential for a more rapid drought recovery of C dynamics under future warmer conditions in a CO 2 -rich world (Roy et al., 2016). While we recognize that for obtaining a longer-term perspective of global change effects on the ecosystem C balance the stabilization and decomposition of plant-and microbial-derived C in soil organic matter need to be considered (e.g., Lavalee et al., 2020;Liang et al., 2017;van Groenigen et al., 2017), our findings highlight large shifts in C allocation and partitioning under climate change, which could represent an important mechanism underpinning also potential long-term effects.

| CON CLUS ION
In conclusion, compared to ambient conditions, the combination of eT and eCO 2 (1) increased the fraction and the absolute amount of freshly assimilated C in soil respiration, (2) increased the severity of drought effects on the fate of recent C in soil respiration and amplified dynamic shifts between storage and respiration, which led to sustained coupling between photosynthesis and autotrophic soil respiration, and (3) favored the recovery of gross primary productivity, plant regrowth and the amount and the proportion of recent C respired belowground. Our findings therefore indicate that a warmer climate under eCO 2 can alter drought and post-drought responses of ecosystem CO 2 fluxes and of C allocation from photosynthesis to belowground respiration. This highlights the importance of accounting for the interactions of multiple global change factors to understand and predict future dynamics of the terrestrial C cycle.

ACK N OWLED G EM ENTS
This study was financially supported by Austrian Science Fund (FWF; P28572-B22). KM was additionally supported by a PhD completion grant from the University of Innsbruck. We thank the team from the Agricultural Research and Education Centre (AREC) Gumpenstein for their support at the ClimGrass-facility. We also thank Mario Deutschmann and Herbert Wachter for their technical support, Lisa Geres for assistance with the fieldwork, Hans-Peter Piepho for statistical advice and Markus Reichstein for advice concerning the time-series analysis.

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest.

AUTH O R S' CO NTR I B UTI O N S
MB conceived and supervised the study; KM, JI, DR and LM performed the field measurements; AC, AR and WW collected the plant and soil material and measured their C isotope content; KM and JI analyzed the data; EMP and DR maintained the experimental infrastructure; KM and MB wrote the paper with significant inputs from JI, and all co-authors provided feedbacks and comments on the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available in Zenodo repository at https://doi.org/10.5281/zenodo.4643297.