Examination of the parameters controlling the triple oxygen isotope composition of grass leaf water and phytoliths at a Mediterranean site: a model–data approach

. Triple oxygen isotopes ( 17 O-excess) of water are useful to trace evaporation at the soil–plant–atmosphere interface. The 17 O-excess of plant silica, i.e., phytoliths, inherited from leaf water, was previously calibrated in growth chambers as a proxy of atmospheric relative humidity (RH). Here, using a model–data approach, we examine the parameters that control the triple oxygen isotope composition of bulk grass leaf water and phytoliths in natura , at the O 3 HP experimental platform located in the French Mediterranean area. A grass plot was equipped to measure for 1 year, all environmental and plant physiological parameters relevant for modeling the isotope composition of the grass leaf water. In particular, the triple oxygen and hydrogen isotope composition of atmospheric water vapor above the grass was measured continuously using a cavity ring-down spectrometer, and the grass leaf temperature was monitored at plot scale us-ing an infrared (IR) radiometer. Grass leaves were collected in different seasons of the year and over a 24 h period in June. Grass leaf water was extracted by cryogenic vacuum distillation and analyzed by isotope ratio mass spectrometry (IRMS). Phytoliths were analyzed by IR–laser ﬂuorination– IRMS after chemical extraction. We showed that the traditional Craig–Gordon steady-state model modiﬁed for grass leaves reliably predicts the triple oxygen isotope composition of leaf water during daytime but is sensitive to uncertainties on the leaf-to-air temperature difference. Deviations from isotope steady state at night are well represented in the triple oxygen isotope system and predictable by a non-steady-state model. The 17 O-excess of phytoliths conﬁrms the applicability of the 17 O-excess phyto vs. RH equation established in previous growth chamber experiments. Further, it recorded average daytime RH over the growth period rather than daily RH, related to low transpiration and siliciﬁcation during the night. This model–data approach highlights the utility of the triple oxygen isotope system to improve the understanding of water exchange at the soil–plant–atmosphere interface. The in natura experiment underlines the applicability of 17 O-excess of phytoliths as a RH proxy.


Introduction
Continental atmospheric relative humidity (RH) is a key factor of soil evaporation, transpiration, dryness stress and ecosystem productivity (Grossiord et al., 2020;Liu et al., 2021;López et al., 2021).However, RH is estimated with low precision in the Earth system models (IPCC, 2013;Tierney et al., 2020).Long-term data beyond the instrumental period is needed to improve the representation of RH in these models.Leaf organic and mineral compounds formed during plant growth, such as cellulose, n-alkanes of leaf waxes, or phytoliths are used as past climate indicators when preserved in soils, sediments or peat (Helliker and Ehleringer, 2002a, b;Kahmen et al., 2011Kahmen et al., , 2013;;Zech et al., 2014;Tuthorn et al., 2015;Alexandre et al., 2018Alexandre et al., , 2019;;Outrequin et al., 2021;Garcin et al., 2012Garcin et al., , 2022)).To accurately interpret the isotope signal of these compounds in terms of paleoclimate, their relationship with that of leaf water and the factors driving their isotope variability need to be determined.
Regarding the phytolith isotope signature, previous calibrations have often been performed in controlled environmental conditions, not representative of the diurnal, daily and seasonal climate variations encountered in the natural environment (Alexandre et al., 2018(Alexandre et al., , 2019;;Outrequin et al., 2021).Therefore, the question of the time span (seasonal vs annual, diurnal vs daily) integrated in the phytolith isotope composition remains open.
Leaf waters generally show higher δ 2 H, δ 18 O, and lower d-excess [= δ 2 H -8 δ 18 O] than meteoric waters due to significant evaporative fractionation during transpiration.The magnitude of this isotope fractionation can be predicted by the isotopeevaporation model developed by Craig and Gordon (1965), and later adapted to leaf transpiration (Dongmann et al., 1974;Farquhar and Cernusak, 2005).This model (hereafter referred to as the C-G model) considers three main processes occurring in the boundary layer of the leaf during transpiration: (i) liquid water-water vapor equilibrium at the boundary layer interface, (ii) diffusion of water vapor from the evaporative sites in the leaf to the surrounding air, and (iii) back-diffusion of atmospheric water vapor to the leaf (Craig and Gordon, 1965;Farquhar et al., 2007;Cernusak et al., 2016).The C-G model is based on the steady-state assumption, i.e. all water that is lost by evaporation is continuously replenished by xylem water.This assumption neglects small diurnal changes in leaf water content that are expected to result in only 3 % error in the predicted leaf water δ 18 O enrichment (Farris and Strain, 1978;Farquhar and Cernusak, 2005).The C-G model also assumes isotope steady state, so that the isotope composition of transpired water matches that of source (xylem) water.To take into account the advection of less evaporated stem water to the evaporation site, as well as the diffusion of the evaporating water back to the leaf lamina, a transpiration-dependent correction, called the Péclet effect, can be added to the C-G model (e.g., Buhay et al., 1996;Helliker and Ehleringer, 2000;Roden et al., 2000;Farquhar and Gan, 2003;Farquhar and Cernusak, 2005;Ripullone et al., 2008;Treydte et al., 2014).For grasses, a two-pool model, including a pristine water pool that coincides to the xylem tissues and an evaporated water pool that corresponds to leaf lamina water has been found to best represent bulk leaf water (Liu et al., 2017;Hirl et al., 2019;Barbour et al., 2021).This mixing effect is independent from transpiration, so that a two-endmember mixing equation is combined with the C-G model (Leaney et al., 1985).
affected by temperature changes and Rayleigh distillation.This is due to its low sensitivity to equilibrium isotope fractionation between liquid water and water vapor (Barkan and Luz, 2005).Consequently, 17 O-excess varies little in meteoric water, which feeds the soil water taken up by the plants and is also assumed to vary little in atmospheric water vapor (Luz and Barkan, 2010;Aron et al., 2021;Surma et al., 2021).The 17 O-excess of bulk leaf water is thus essentially controlled by the molecular diffusion of water vapor between the leaf and the atmosphere during transpiration (Barkan and Luz, 2007).The extent of this process depends mainly on the water pressure gradient between the leaf and the atmosphere.The few existing studies on 17 O-excess of bulk leaf water showed that its 17 O-excess is inversely related to RH. Discrepancies between modeled and observed 17 Oexcess values higher than 100 per meg have been reported (Li et al., 2017;Alexandre et al., 2018;Outrequin et al., 2021).
These discrepancies have been attributed to deviations from isotope steady state in the early morning hours (Li et al., 2017) and uncertainty in the estimates of leaf temperature and the isotope composition of atmospheric water vapor (Li et al., 2017;Alexandre et al., 2018).Large discrepancies observed by Li et al. (2017) may also result from neglecting potential mixing of evaporated and non-evaporative grass leaf water pools.
Phytoliths are micrometric silica particles that form in temperature-dependent isotope equilibrium with water in living plant tissues within a few hours to days (Perry et al., 1987).In grasses, the majority of phytoliths forms in sheaths and leaves, due to concentration of solutes by transpiration (e.g., Webb andLongstaffe, 2000, 2002).Phytolith morphological assemblages recovered from soils and sediments are used to reconstruct vegetation changes and qualitatively inform on climatic conditions at the time of soil formation (Bremond et al., 2005;Aleman et al., 2012;Nogué et al., 2017).Previous studies investigated the potential of δ 18 O of phytoliths as a proxy for past temperature (Webb and Longstaffe, 2000, 2002, 2006;Alexandre et al., 2012).However, accurate temperature reconstruction using this proxy requires an independent estimate of the δ 18 O of soil water, and an estimate of the effect of RH and transpiration on δ 18 O of leaf water.These studies have also shown the dependency of δ 18 O of phytoliths on RH, but its utility to reconstruct past RH has not been further explored given the large number of factors influencing δ 18 O of precipitation, soil, and leaf water.Recent studies in growth chambers and at natural sites demonstrated that unlike the δ 18 O, the 17 O-excess of phytoliths ( 17 O-excessphyto), inherited from the 17 O-excess of leaf water, is primarily controlled by RH around the plant, according to a gradient of 4.3 ± 0.3 per meg % -1 (Outrequin et al., 2021).This relationship is independent of grass leaf length and vegetation type (Alexandre et al., 2018(Alexandre et al., , 2019;;Outrequin et al., 2021).
Further, the 17 O-excessphyto is not affected by changes in air temperature or atmospheric CO2 levels (Outrequin et al., 2021).
In this study, using a model-data approach, we examined the parameters controlling the triple oxygen isotope composition of bulk grass leaf water and phytoliths at a natural site.For that purpose, a grass plot was equipped to measure for the course of one year, all environmental and plant physiological parameters relevant for modelling the isotope composition of the grass leaf water.In particular, the triple oxygen and hydrogen isotope composition of atmospheric water vapor above the grass was measured continuously over the year using a cavity ring-down spectrometer (CRDS), and the grass leaf temperature was monitored at plot-scale using an infra-red (IR) radiometer.Grass leaf blades were collected at midday on eight days in different seasons of the year and over a 24-hour period in June for triple oxygen and hydrogen isotope analysis of bulk leaf waters.In addition, grass leaf blades were harvested in spring, summer and autumn for phytolith extraction and triple oxygen isotope analysis to examine which RH average is recorded in 17 O-excessphyto of phytolith assemblages that are formed over growth periods of several months.

Experimental setup
The AnaEE in natura experimental platform O3HP is located about 100 km north of Marseille (France) at an altitude of 680 m above sea level (43.935°N, 5.711° E).On 14 February 2021, seeds of the C3 grass species Festuca arundinacea, also referred to as tall fescue, were sown (8 g m -2 ) on a 5.5 m 2 plot in the understory of an oak-dominated forest.The same grass species was used for the calibration of the relationship between 17 O-excessphyto and RH in growth chamber experiments (Alexandre et al., 2018(Alexandre et al., , 2019;;Outrequin et al., 2021).Potting soil was added to the shallow calcaric leptosol (IUSS Working Group WRB, 2015;Belviso et al., 2016) and supplied with ~ 50 g m -2 organic fertilizer (Engrais Gazon, Neudorff, Emmerthal, Germany) and 2.7 g m -2 SiO2 (General Hydroponics Mineral Magic, Terra Aquatica, Fleurance, France) to ensure a sufficient amount of nutriments and bio-available silica.
The experimental plot was automatically irrigated with tap water (30 mm d -1 ) from 04 March 2021 until the end of the experiment on 23 November 2021 to avoid water stress in the grasses.The potential evaporation from the grass plot (2-4 mm d -1 ) estimated using the Penman-Monteith equation (Monteith, 1965) was an order of magnitude lower than the irrigation rate.Therefore, we assume that soil water evaporation was negligible and had no impact on the isotope composition of leaf water.An aliquot of the irrigation water was collected in an evaporation-free water collector (Rain Sampler 1, Palmex d.o.o., Zagreb, Croatia;Gröning et al., 2012), that was sampled weekly.Precipitation was collected on an event-based interval using a second water collector of the same type.Both collectors were emptied and dried after sampling.For isotope analysis of atmospheric water vapor, the air at 0.4 m above the grass plot was pumped continuously (N 86 KN.18, KNF DAC GmbH, Hamburg, Germany) to a Picarro L2140-i CRDS (Picarro Inc., California, USA), installed in an air-conditioned cabin on the experimental site.The air was passed through a 11.5 m long and 1/4 " wide PFA tube (PFA-T4-062-100, Swagelok, Ohio, USA), at a flow rate of 5 L min -1 .The tubing was insulated and heated to prevent condensation of the water vapor.A funnel covered by a net was placed at the inlet for protection from rain and suction of insects and large aerosol particles.
The following climate parameters were measured on the experimental site: Global solar radiation at 6 m above ground (LI-200, LI-COR Biosciences Inc., Nebraska, USA), precipitation amount (15189 H, LAMBRECHT meteo GmbH, Göttingen, Germany), RH and atmospheric temperature (Tair) at 60 cm height next to the grass plot (HMP155,Vaisala Oyj,Vantaa,Finland), atmospheric temperature at 5 cm above the ground (Tground) (DTS12, Vaisala Oyj, Vantaa, Finland), soil water content and soil temperature at ~ 5 cm depth (CS655, Campbell Scientific Inc, Logan, Utah, USA), plot-scale grass leaf temperature (Tplot) (IR radiometer SI-411-SS, Apogee Instruments Inc., Utah, USA), and sky temperature (Tsky).Tplot is the temperature integrated over the field of view of the IR radiometer that covered ~ 90 % of the grass plot surface.Each parameter was extracted in hourly resolution from the COOPERATE database (COOPERATE database, 2022).
On sampling days (Table 1), stomatal conductance (gs) and transpiration, were monitored continuously over the day on a single grass leaf of 4-5 mm width using a Li-6400 XT gas exchange system (LI-COR Biosciences Inc., Nebraska, USA).To assess the spatial variability of gs, this parameter was additionally measured hourly on the adaxial side of ten leaves of at least 3 mm width, randomly selected on the plot, using an AP4 porometer (Table S1; Delta-T Devices LTD, Cambridge, UK).In addition, leaf temperature (Tleaf) was measured in situ on the adaxial side of ten grass leaves, randomly selected, in one-hour intervals using an Optris CT IR thermometer (Table S2; Optris GmbH, Berlin, Germany).Tplot and Tleaf measurements were corrected for emissivity of the grass canopy, considering the tree canopy gap fraction: where ε is the emissivity of the grass canopy (ε = 0.95; Apogee Instruments Inc, 2022) and α is the tree canopy gap fraction, which is estimated to be 0.3 throughout the experimental period.Traw is the temperature recorded by the sensor, Tsky is the sky temperature and Tcanopy is the canopy temperature, which is assumed to equal to Tair.

Sampling
Leaf blades of F. arundinacea were collected at midday on eight days in May, July, August, October, and November 2021 (Table 1), as well as every ~ 1.5 h over a 24-hour period from 14-15 June 2021.About ten fully developed, not senescent leaf blades from different tillers evenly distributed over the grass plot were immediately transferred to 12 mL Exetainer vials (Labco, High Wycombe, UK), and stored in a fridge until water extraction and isotope analysis.2).Each regrowth started after the grasses had been cut above the sheath at 2-4 cm height.Grass heights were measured at monthly intervals.At the end of each regrowth, the grass leave blades from the entire plot were harvested and dried at 50 ºC.Between 120 and 150 g of dry matter were obtained for phytolith extraction and analysis.The isotope composition and mixing ratio of water vapor in the air at 0.4 m above the grass plot was measured for 70 min every 140 min during the spring monitoring and every 280 min during the monitoring in summer and autumn.In between these measurements, the instrument was used for another experiment.The atmospheric water vapor data from the first 10 minutes of each measurement cycle were removed to account for memory effects and provide sufficient time to establish a stable baseline.The remaining 60 minutes were averaged.During the 24-hour monitoring, air sampling was performed continuously without interruption.Liquid water standard measurement runs were performed on a weekly basis.The mean of four measurement runs of liquid water standards was used to normalize the atmospheric water vapor isotope data to VSMOW-SLAP scale.The calibration protocol is described in detail by Voigt et al. (2022).In brief, three liquid water standards that covered the expected isotope range of atmospheric water vapor at the study site were analyzed at a water mixing ratio of 11000 ppmv using a Picarro autosampler system (A0325, Picarro Inc., California, USA) coupled to a high-precision vaporizer (A0211, Picarro Inc., California, USA).The liquid standards were injected in a dry air stream, produced by a lubricated mobile air compressor (MONTECARLO FC2, ABAC air compressors, Italy), further dried using two drierite columns combined with a dry ice trap (Voigt et al., 2022).Raw isotope compositions of the liquid standards of four consecutive measurement runs were averaged and then corrected to the water mixing ratio of the measured atmospheric water vapor, using the mean of three mixing ratio dependency functions that were determined on site for water mixing ratios between 3000 and 30000 ppmv in May 2021, October 2021 and January 2022 (Fig. A1).The precision of calibrated and integrated atmospheric water vapor data was determined using a Monte Carlo simulation (Voigt et al., 2022).Precision was better than ± 0.1 ‰, ± 0.2 ‰, ± 1.8 ‰ and ± 14 per meg, and ± 0.9 ‰ for δ

Grass leaf water
Grass leaf water was extracted by cryogenic vacuum distillation (static pressure < 10 Pa) with sample vials placed in the vacuum line and immersed in a heated water bath for 3 h with a final target temperature set to 80 °C (attained within 45 min of extraction).A detailed description of the system design is given by Barbeta et al. (2022).Water extraction yield was derived by comparing the volume of water collected (in mL) and the difference of sample weights before and after water extraction (with the exetainer and converted in equivalent mL of water).For our sample set, the average water extraction yield was 103 ± 5 % (102 ± 3 % without one outlier) and average extracted volume was 0.5 ± 0.2 mL, with only one extraction volume below 0.3 mL.Thus, methodological uncertainties linked to cryogenic vacuum distillation should be negligible (Diao et al. 2022).Isotope analysis of grass leaf waters was performed at the University of Cologne.For triple oxygen isotope analysis, pure O2 liberated from grass leaf waters by fluorination was introduced in a Thermo Fisher Scientific MAT 253 dual-inlet mass spectrometer (Massachusetts, USA), following the procedure described by Surma et al. (2015).The reproducibility (1 SD, n = 2) of δ 17 O, δ 18 O and 17 O-excess measurements was better than ± 0.15 ‰, ± 0.30 ‰ and ± 11 per meg, respectively.
Hydrogen isotope ratios were determined by high-temperature carbon reduction in a pyrolysis elemental analyzer (HEKAtech GmbH, Wegberg, Germany), coupled to the mass spectrometer.The reproducibility (1 SD, n = 3) of δ 2 H measurements was always better than 1.1 ‰.An intercomparison of water analysis at CEREGE and the University of Cologne was performed.

Phytoliths
The silica contents of harvested grass leaf blades were determined by inductively coupled plasma-atomic emission spectroscopy (Ultima C, Horiba Jobin Yvon, Longjumeau, France).Phytoliths were extracted following the 'wet digestion'protocol detailed in Table 2 of Corbineau et al. (2013).The protocol involves treatment of the sample with different chemical agents (HCl, H2SO4, H2O2, HNO3) to remove organic and carbonate compounds.The pure phytolith cocnentrates were mounted on microscope slides in Canada Balsam and the morphological types were counted using light microscopy at a 600X magnification.The epidermal silicified intercoastal long cells were quantified relative to the silicified short cells to obtain information on the silicification process (Alexandre et al., 2019).
The phytolith samples (1.6 mg) were dehydrated at 1100 °C under a flow of N2 (Chapligin et al., 2010) to prevent the formation of siloxane from silanol groups during dehydroxylation.Molecular O2 was extracted using the IR laser-heating fluorination technique (Alexandre et al., 2006;Crespin et al., 2008;Outrequin et al., 2021).At the end of the procedure, the gas was passed through a -114 °C slush to refreeze any molecule interfering with the mass 33 (e.g., NF potentially remaining in the line).The gas was directly sent to a ThermoQuest Finnigan Delta V Plus dual-inlet mass spectrometer (Massachusetts, USA) for triple oxygen isotope analysis.Each gas sample was run twice with each run consisting of eight dual-inlet cycles.A third run was performed when the standard deviation on the first two averages was higher than 12 per meg for 17 O-excess.The reproducibility for δ 18 O and 17 O-excess measurements of the quartz laboratory standard was 0.16 ‰ and 8 per meg, respectively (1 SD, n = 5).For the phytolith samples, the precision for δ 18 O and 17 O-excess was always better than 0.5 ‰ and 12 per meg (1 SD), respectively.The sample measurements were corrected using a quartz laboratory standard analyzed at the beginning of the day until a 17 O-excess plateau was reached and again at the end of the day.The isotope composition of the reference gas was determined against NBS28.For robust comparisons between silica and water isotope compositions, the phytolith data are normalized to VSMOW-SLAP scale (Outrequin et al., 2021).

Modelling
According to the C-G isotope steady state model (Craig and Gordon, 1965;Dongmann et al., 1974;Farquhar et al., 2007;Cernusak et al., 2016), the isotope ratio of the evaporated water pool in the leaf (Re) is: where RV and RS denote the isotope ratios ( 2 H/ 1 H, 17 O/ 16 O and 18 O/ 16 O) of atmospheric water vapor and source water, respectively.h is the ratio of the actual vapor pressure in the atmosphere to the saturation vapor pressure inside the leaf (i.e. at leaf temperature, Tleaf).When the leaf-to-air temperature gradient is small, h is equal to RH.The isotope fractionation during water vapor diffusion in air through the leaf stomata and boundary layer (αdiff) was estimated as: where gs and gb (mol m -2 s -1 ) denote the stomatal and leaf boundary layer conductances, and αkin denotes the kinetic isotope fractionation during molecular diffusion of water vapor in air.We took 18 αkin = 1.028 and 2 αkin = 1.025 from  (Barkan and Luz, 2005) and θkin = 0.5185 for the kinetic fractionation during molecular diffusion (Barkan and Luz, 2007).
The bulk grass leaf water at isotope steady state (Rleaf,ss) represents a mixture of an evaporated water pool in the lamina mesophyll whose isotope composition is predicted by the C-G model (Re, Eq. ( 2)), and an unevaporated pool in the leaf veins and associated ground tissues, whose isotope composition matches Rs (Leaney et al., 1985;Yakir et al., 1994;Hirl et al., 2019): where f represents the water volume fraction of the unevaporated pool and was set to 0.2 in our study.Similar values were used in previous studies on grass leaf water (Wang et al., 2018;Alexandre et al., 2019;Hirl et al., 2019).Instead of a mixing equation, the Péclet effect can be considered to estimate the bulk leaf water isotope composition (Farquhar and Lloyd, 1993;Farquhar et al., 2007;Holloway-Phillips et al., 2016): With p [= EL/CD] the Péclet number, where L is the effective path length, E is the grass leaf transpiration rate, C is the molar density of liquid water (55500 mol m -3 ), and D is the diffusivity of water (2.3 10 -9 m 2 s -1 at 25 ºC).One single value of L was applied for the data set and adjusted to fit the observed grass leaf water isotope composition.
When the steady state cannot be reached, non-steady state enrichment of bulk leaf water (Rleaf,nss) can be modelled using the following equation (Dongmann et al., 1974;Farquhar and Cernusak, 2005;Hirl et al., 2019): where g = gs gb/(gs+gb), wi is the mole fraction of water vapor in air in the intercellular spaces, W is the leaf water content and Rleaf,ss denotes the isotope composition of bulk leaf water at steady state, as predicted by Eq. ( 4).Similar to Farquhar & Cernusak (2005)
The annual average isotope composition of atmospheric water vapor was -17.4 ± 3.1 ‰ for δ 18 O, -126 ± 24 ‰ for δ 2 H, 13.0 ± 1.7 ‰ for d-excess and 28 ± 5 per meg for 17 O-excess.These values coincide with δ 18 O, δ 2 H, d-excess and 17 O-excess values estimated for a water vapor in isotope equilibrium with the amount-weighted precipitation (Table S4).As for precipitation, the atmospheric water vapor monthly averages in δ 18 O and δ 2 H increase from winter to summer, whereas averages in d-excess and 17 O -excess decrease (Fig. 1; Table S4).S5).
315   2 show changes in RH, h, Tair, Tplot, F. arundinacea leaf transpiration and stomatal conductance averaged over 30 minutes before the 8 grass leaf samplings at midday.RH is always equal or lower than h (by less than 9 %) but covaries with h from low values in spring and summer (30-40 %) to high values in autumn (ca.64 %).Tplot is 1-3 ºC lower than Tair but changes along with Tair from a measurement day to another, with high values in summer (ca. 25 °C), and lower values in spring (ca.18 °C) and autumn (ca.14 °C).Figure A3 shows five daily variations of Tair, Tplot and Tleaf.Although Tleaf varies spatially within the plot, its spatial average around midday is close to Tplot (Fig. A3), supporting that Tplot can be considered as an approximation of Tleaf.Transpiration and stomatal conductance are relatively stable from a measurement day to another, varying from 1.1-3.7 mmol m -2 s -1 and 50-130 mmol m -2 s -1 , respectively (Fig. 2).
The isotope composition of F. arundinacea leaf water sampled at midday is also shown in Table 1 and Figure 2. The grass leaf water has δ 18 O (-0.05 ‰ to 20.1 ‰) and δ 2 H (-31 ‰ to 18 ‰) that are higher than irrigation water, and d-excess (-31.0 ‰ to -142.8 ‰) and 17 O-excess (17 per meg to -165 per meg) that are lower than irrigation water, as can be expected for an evaporation signal.The changes in δ 18 O, δ 2 H, d-excess and 17 O-excess observed from a sampling day to another follow the changes in RH and h (Fig. 2).Evaporative isotope enrichment is highest in May and July when RH is low and lowest in November when RH is high.Samples from October and November have similar d-excess as expected from little variation in RH (64 ± 2 %).However, their 17 O-excess values differ by 66 per meg.The reason for this difference in 17 O-excess remains unclear.
Table 1 and Figure 3 show the 24-hour evolution of the isotope composition of grass leaf water from 14-15 June 2021 in relation to RH, h, Tair, Tplot, F. arundinacea transpiration and stomatal conductance.Tair and RH range from 14 °C to 31 °C and 38 % to 97 %, respectively.Tplot is ca. 1 °C higher than Tair at night, and up to 4 °C lower than Tair during daytime.During daytime, stomatal conductance measured continuously on a single leaf, ranges from 60 to 120 mmol m -2 s -1 and co-varies with transpiration (1.3-3.9 mmol m -2 s -1 ).However, stomatal conductance varies greatly (by 200-500 mmol m -2 s -1 ) between different leaves in the grass plot (Table S1, Fig. A4).At night, stomatal conductance is never higher than 20 mmol m -2 s -1 , while transpiration remains lower than 0.5 mmol m -2 s -1 .The isotope variability of grass leaf water on this diurnal scale is of the same order of magnitude as the changes observed among samples collected at midday in different months.The evolution of the isotope composition of grass leaf water follows RH and h, except for samples collected at night and in the early morning when transpiration is low.During this time, stomatal closure impeded exchange between the leaf and the atmosphere, decoupling the isotope composition of grass leaf water from RH.
Figure 2: F. arundinacea transpiration (E) and stomatal conductance (gs) measured on a single leaf blade using the LI-COR gas exchange system, atmospheric temperature (Tair), plot-scale grass leaf temperature (Tplot), relative humidity (RH), water vapor pressure ratio between leaf and the atmosphere (h) and measured (circles) and predicted (+) isotope composition of F. arundinacea leaf water (δ 18 O, δ 2 H, 17 O-excess, d-excess) for midday samples over the year 2021 (see Table 1 for sampling dates).Error bars of isotope data represent analytical precision (see method section).The modeled isotope composition of bulk grass leaf water is predicted by the C-G steady state model combined with the mixing equation (Eq.( 4)) using average environmental conditions over 30 minutes before sampling (Table 1, S5).The model uncertainty (1 SD) was estimated using a Monte Carlo simulation accounting for uncertainty of input variables (RH ± 1 %, Tplot ± 2 °C, δ 18 OS ± 0.2 ‰, δ 2 HS ± 0.7 ‰, d-excessS ± 0.6 ‰, 17 O-excessS ± 6 per meg, δ 18 OV ± 0.2 ‰, δ 2 HV ± 1.8 ‰, d-excessV ± 0.9 ‰, 17 O-excessV ± 14 per meg, gs ± 100 mmol m -2 s -1 , and the fraction of unevaporated water pools (f) ± 0.1).Gray dashed circles indicate the sample that has been likely affected by evaporation during sampling.Light blue dash circles indicate samples with anomalously high 17 O-excess relative to d-excess.humidity (RH), water vapor pressure ratio between leaf and the atmosphere (h), and the observed (circles) and predicted steady state (pale gray curve, Eq. ( 4)) and non-steady state (orange curve, Eq. ( 6a)) isotope composition of F. arundinacea leaf water (δ 18 O, δ 2 H, 17 O-excess, d-excess) from 14-15 June 2021.Error bars of isotope data represent analytical precision (see method section).Shaded areas mark nighttime intervals.The isotope composition of grass leaf water is predicted using average environmental conditions over 30 minutes before sampling (Table 1, S5, S6).The pale gray shaded area represents model 370 uncertainty (1 SD) of the predicted steady-state leaf water isotope composition estimated using a Monte Carlo simulation (see caption Figure 2).The dashed part of the steady state prediction represents the time when grass leaf water isotope composition deviates from steady state due to low transpiration and long leaf water residence times (see discussion for details).

Model-data comparison
For six of eight midday samplings, the isotope composition of bulk grass leaf water predicted by the C-G steady state model combined with the mixing equation (Eq.( 4)) using boundary conditions averaged over 30 minutes before sampling (Table S5) agrees with the measured isotope values within model uncertainty, that is in average ± 2.8 ‰, ± 7.9 ‰, ± 15 ‰, and ± 24 per meg for δ 18 O, δ 2 H, d-excess and 17 O-excess, respectively (Fig. 2).Samples collected on 20 May 2021 and 23 November 2021 show larger discrepancies between observed and predicted values.The May sample has significantly higher δ 18 O (> 8 ‰) and δ 2 H (> 10 ‰), and lower d-excess (59 ‰) and 17 O-excess (33 per meg) than respective steady state values predicted by the two-pool mixing model (Eq.( 4)) (Fig. 2).These large deviations are indicative of stronger evaporation than expected.In view of the large magnitude of the deviation, we suppose that this sample was affected by evaporation during sampling.We therefore exclude this sample from further discussion.For the November sample, δ 18 O, δ 2 H and d-excess agree within 1.1 ‰, 1 ‰ and 8 ‰ with the predicted steady state values, respectively.However, the 17 O-excess is 66 per meg higher than the predicted steady state value (Fig. 2).The reason for this discrepancy remains unclear.
For the 24-hour monitoring, the C-G steady state model combined with the two-pool mixing equation (Eq.( 4)) reproduces the evolution of the isotope composition of grass leaf water during the day, but not at night and in the early morning, when stomatal conductance and transpiration are low (Fig. 3).During daytime, best agreement between predicted and observed grass leaf water is found for samples collected on the morning of 15 June 2021 until midday, with deviations lower than ± 0.6 ‰ for δ 18 O, ± 5 ‰ for δ 2 H, ± 6 ‰ for d-excess and ± 8 per meg for 17 O-excess.However, on the afternoon of 15 June 2021, when transpiration is highest, observed δ 18 O and δ 2 H are 1.5-4 ‰ and 3-9 ‰ lower, and d-excess and 17 O-excess are 9 ‰ and 34 per meg lower than predicted values, respectively.In contrast, on the evening of 14 June 2021, observed δ 18 O are 1-2 ‰ higher, whereas δ 2 H, d-excess and 17 O-excess are respectively 4-6 ‰, 18 ‰, and 38 per meg lower than respective steady state values predicted by the two-pool mixing model (Eq.( 4)).The non-steady state equation (Eq.( 6)) was applied for night predictions to match the data (Fig. 3).Differences between predicted non-steady state and observed values at night range from 0.2-3.6 ‰ for δ 18 O, 5-12 ‰ for δ 2 H, 3-19 ‰ for d-excess and 1-31 per meg for 17 O-excess (Table S6).Note that a grass leaf water content of 6 mol m -2 is required for the model to fit the data (Table S6).This value is higher than leaf water contents reported for grasses in previous studies (2-4 mol m -2 ; Hirl et al., 2019;Barbour et al., 2021).
Figure 4: Sensitivity of δ 18 O, δ 2 H, 17 O-excess and d-excess of leaf water to changes in environmental and plant physiological parameters.Green circles represent measured F. arundinacea leaf water isotope composition over a 24-hour period from 14-15 June 2021.The black line shows the steady state leaf water isotope composition predicted by the C-G steady state model combined with the mixing equation (Eq.( 4)) using mean boundary conditions over 30 minutes before sampling (Table 1).Shaded areas indicate the sensitivity of the predicted leaf water isotope composition for relative humidity (RH) (± 5 %) (a-d), the fraction of unevaporated water pools (f) (± 0.1) (e-h), leaf temperature (Tleaf) (± 2 °C) (i-l), the isotope composition of source water (± 0.2 ‰ for δ 18 OS, ± 0.7 ‰ for δ 2 HS, ± 0.6 ‰ for d-excessS, ± 6 per meg for 17 O-excessS) (m-p), the isotope composition of atmospheric water vapor (± 0.2 ‰ for δ 18 OV, ± 1.8 ‰ for δ 2 HV, ± 0.9 ‰ for d-excessV, ± 14 per meg for 17 O-excessV) (q-t) stomatal conductance (gs) (± 100 mmol m -2 s -1 ) (u-x).Coloured curves show the isotope composition of leaf water predicted by the C-G steady state model combined with the mixing equation (Eq.( 4)) (i) when assuming leaf temperatures being equal to atmospheric temperature (panel i-l, orange), (ii) when assuming leaf temperatures being 2 °C lower than atmospheric temperature (panel i-l, light blue), and (iii) when estimating atmospheric water vapor from isotope equilibrium with source water (irrigation water) (panel q-t, orange).

Changes in climate averages, grass height, silicification rate, and triple oxygen isotope composition of phytolith assemblages
Table 2 shows daily and daytime climate averages, maximum grass height, silicification rate, ratio of long cell to short and long cell phytoliths, and the triple oxygen isotope composition of phytoliths for the three regrowth periods.Daily average Tair is from 9 °C to 22 °C and daily average RH is from 64 % to 81 %.Daytime average Tair is about 2.4 °C higher than the daily average.Daytime average RH is about 8 % lower than the daily average.Daily averages of Tplot are similar to Tair, so that RH approximates h (cf.section 2.4).During daytime, averages of RH and h differ by 1-4 % due to ΔTleaf-air ranging from -1.1 °C to 0.3 °C.Daytime average h is 61 % in spring and summer, and 76 % in autumn.The average soil water content is always higher than 0.20 ± 0.05 L L -1 , whatever the regrowth, supporting that the grass plot is always well-watered, and that water stress is excluded.
Grass height increases exponentially during spring regrowth, and linearly during summer regrowth (Fig. A5).During the autumn regrowth, the grass height increases only in the first month of the regrowth and stabilizes thereafter (Fig. A5).The silicification rate (from 2.7 to 5.9 SiO2 g -1 d -1 ), and the ratio of long cell to short and long cell phytoliths (from 30 to 70 %) increase with the number of regrowth periods, without any correlation with RH or h that varied little from a regrowth to another (Table 2).The δ' 18 O and 17 O-excess of the grass leaf phytoliths are similar in spring and summer (36.2 ± 0.5 ‰ and -260 ± 5 per meg, respectively; Table 2) and slightly different in autumn (34.3 ‰ and -234 per meg, respectively).These isotope values fall within the range of values observed in previous growth chamber calibrations (Alexandre et al., 2018(Alexandre et al., , 2019;;Outrequin et al., 2021).The 17 O-excessphyto coincide with the lower range of values reported for phytoliths extracted from soils in Western and Central Africa (Alexandre et al., 2018).

Relationship between the 17 O-excess of grass phytoliths and leaf water
Numerous studies have investigated the temperature-dependent isotope fractionation between amorphous and/or biogenic silica and their formation water ( 18 αphyto-H2O) with variable results (e.g., O'Neil and Clayton, 1964;Knauth and Epstein, 1976;Shemesh et al., 1992;Brandriss et al., 1998;Hu and Clayton, 2003;Dodd, 2011, and many more).Here we use temperaturedependent 18 αphyto-H2O obtained for the diatom-water pair by Dodd and Sharp (2010) (1.0326 at 20ºC).The triple oxygen isotope exponent between silica and water (qphyto-H2O) linking 17 αphyto-H2O to 18 αphyto-H2O ( 17 α = 18 α θ ), has been defined as 0.524 ± 0.0002 for the 5-35°C temperature range (Cao and Liu, 2011;Sharp et al., 2016).However, a different value of 0.522 ± 0.001 was consistently obtained for phytoliths, reproducible regardless of bio-climatic constraints (Outrequin et al., 2021).Using this apparent λphyto-H2O, we calculated the triple oxygen isotope compositions of the formation water (FW) in equilibrium with the phytolith samples obtained from the three regrowths (Fig. 5).The reconstructed triple oxygen isotope composition of FW is close to that estimated for daytime average climate conditions of the three regrowths using the C-G model combined with the mixing equation (Eq.( 4)) (Fig. 5).The differences are lower than 1.8 ‰ and 33 per meg for δ' 18 O and 17 O-excess, respectively.
Using the same 18 αphyto-H2O, but λphyto-H2O of 0.524, the 17 O-excess of FW is largely underestimated by 35-60 per meg compared to model predictions (Fig. 5).4)) using average daytime boundary conditions for the three regrowth periods (Table 2).Error bars represent analytical precisions (see methods section), except for precipitation, for which the amount-weighted standard deviation is indicated.

Parameters responsible for discrepancies between observed and predicted isotope compositions of grass leaf water
Overall agreement between the observed and predicted leaf water δ 18 O and 17 O-excess trends from a sampling day to another shows that the C-G steady state model combined with the two-pool mixing equation (Eq.( 4)) is appropriate for estimating seasonal scale variations in the triple oxygen isotope composition of grass leaf water at midday.The two-pool mixing model also correctly reproduces the trends in triple oxygen isotope evolution of leaf water during daytime, although observed and predicted values diverge little when transpiration is maximal in the early afternoon (Fig. 3).As shown by the sensitivity tests, ΔTleaf-air contributes largely to model uncertainty (Fig. 4).Assumptions on Tleaf equal to Tair can explain the discrepancies between predicted and observed isotope values often reported in the literature.In the present case, Tleaf was indirectly measured using Tplot and large misestimation of Tleaf (>2 °C) is unlikely.Part of the small model-data discrepancies in the afternoon on 15 June 2021 can result from RH measured at 60 cm above the ground next to the grass plot being lower than RH surrounding the grass leaf canopy, due to intense soil evaporation.Another bias may come from misestimation of the unevaporated water pool f that can drive large variations in the triple oxygen isotope composition of leaf water, as shown by the sensitivity tests.
The value of 0.2 chosen for f in the present study is at the lower limit of previously reported values selected for grass species (0.2-0.4; Hirl et al., 2019;Barbour et al., 2021).Considering a value for f of 0.4 instead of 0.2 would minimize the discrepancy between observed and predicted δ 18 O of leaf water for the samples taken in the afternoon on 15 June 2021 (Fig. 4).Some studies suggested that f may increase with increasing transpiration, due to increased advection of unevaporated xylem water, known as the Péclet effect (Farquhar and Lloyd, 1993;Cuntz et al., 2007).In contrast to a recent isotope study that found no evidence for the Péclet effect in grass leaves (Hirl et al., 2019), the data from the 24-hour monitoring shows a significant positive correlation (R 2 = 0.49) between transpiration and the difference between observed and predicted δ 18 O values of leaf water.Considering the Péclet effect (Eq. ( 5)) instead of a simple mixing significantly reduces model-data discrepancies by 50-80% and leads to deviations between predicted and observed δ 18 O and 17 O-excess of grass leaf water in the afternoon on 15 June 2021 that are lower than 1.1 ‰, and 12 per meg, respectively (Table S5).The Péclet effect can thus explain that the observed triple oxygen isotope composition of leaf water varies less than predicted when transpiration is high.
In agreement with previous studies on d 18 O and d 2 H (Farquhar and Cernusak, 2005;Cernusak et al., 2016), a non-steady state model is used to reproduce the trends in isotope evolution of leaf water at night when stomatal conductance and transpiration are low.Our results confirm the applicability of this model for the triple oxygen isotope composition of leaf water.In addition, the model-data comparison shows the advantage of 17 O-excess over d-excess in detecting isotope non-steady state in leaf water on a diurnal scale.Figure 6a illustrates the 17 O-excess vs d' 18 O evolution of leaf water from the beginning to the end of the night when transpiration is too low to reach the isotope steady state.RH of 96 ± 2 % persisting between 3:00 and 7:00 (LT) on 15 June 2021 drives the theoretical isotope steady state values to the upper end of the predicted trend on Fig. 6a.However, due to the long leaf water residence time, the observed leaf water isotope composition evolves only slowly towards these values without reaching them.This is well captured by the concave curvature of the non-steady state prediction (Fig. 6a).In contrast, linearity of evaporation trends in the d-excess vs δ 18 O space challenges the differentiation between isotope steady state and non-steady state conditions, as illustrated in Figure 6b.4), Table S5), open circles indicate the non-steady state model prediction (Eq.( 6), Table S6).Colours differentiate samples collected between 19:15 and 21:45 (LT) on 14 June 2021 (grey), between 14 June 2021 23:30 and 15 June 2021 08:15 (LT) (blue) and between 10:00 and 19:00 on 15 June 2021 (orange).The black line serves as a guide-of-the-eye for the trend in modelled isotope steady state values.The average isotope composition of the irrigation water over the experimental period is also shown.The global meteoric water line (GMWL) is shown for comparison.

What can we learn from measurements of Tplot and triple oxygen isotope composition of atmospheric water vapor?
The sensitivity tests highlight the importance of plot-scale grass leaf temperature and the isotope composition of atmospheric water vapor for accurate prediction of the isotope composition of leaf water.
The influence of variations in Tleaf relative to Tair on the isotope composition of leaf water is two-fold.On the one hand, changes in Tleaf slightly modify the magnitude of equilibrium isotope fractionation at the liquid-vapor interface.A few degrees change in Tleaf is however of minor importance for the isotope composition of leaf water.In contrast, changes in ΔTleaf -air, associated with changes in Tleaf, modify the water vapor pressure ratio between the leaf and the atmosphere, i.e. h.For example, a decrease in Tleaf from 20 to 18 ºC at constant Tair of 20 ºC, modifies h by 5-10 % for RH ranging from 40 to 80 %.As h is the major driver of isotope variability in leaf water, even little variations in ΔTleaf -air can therefore significantly influence the isotope composition of leaf water (Fig. 4i-l).
Accurate measurement of Tleaf on plot-scale is challenging, as Tleaf can vary considerably in space and time, particularly according to soil moisture, leaf transpiration, canopy structure and position, net radiation, elevation, and latitude (Still et al., 2019).Sufficient soil moisture supports transpiration, which generally leads to leaf cooling, i.e.Tleaf lower than Tair.On the contrary, water stress is compensated by stomata closure, which stops transpiration and increases Tleaf.In this case, Tleaf may exceed Tair, as demonstrated for irrigated vs rain-fed crops (Siebert et al., 2014).The amplitude of ΔTleaf-air also increases with leaf size (Leuzinger and Körner, 2007).ΔTleaf -air lower or equivalent to -2 °C was reported, at the ecosystem scale, for tropical forests (Rey-Sánchez et al., 2016), grasslands or cold desert areas, whereas larger differences were reported for cold forests and warm desert areas (Blonder and Michaletz, 2018).In the present case, continuous irrigation of the grass plot sustained the transpiration, leading to a daytime Tleaf consistently near or below the daytime Tair (Figs.A3, A6).However, under natural conditions, estimation of Tleaf 2 °C lower than or equal to Tair may lead to significant bias in modeled leaf water isotope composition.Figure A3 shows that Tplot can be used to estimate Tleaf.The measurement of Tplot using IR radiometry as performed here is easy to set up and is strongly recommended if high accuracy is sought in the estimate of Tleaf at plot scale.
The isotope difference between source water and the atmosphere is another key determinant of the leaf water isotope composition.According to the C-G model (Eq.( 2)), the influence of atmospheric water vapor relative to source water becomes increasingly important with increasing h (or RH).While the isotope composition of source water can be often constrained by measurements, accurate estimates of the isotope composition of atmospheric water vapor are difficult to obtain.In the absence of direct measurements, the d 18 O of atmospheric water vapor is often assumed to be in equilibrium with precipitation (e.g., Cernusak et al., 2002;Voelker et al., 2014;Bush et al., 2017;Li et al., 2017;Song et al., 2011;Flanagan and Farquhar, 2014).
However, a recent comparison between modelled vapor and precipitation isotope compositions obtained from different isotope-enabled global climate models suggests that precipitation is rarely in isotope equilibrium with atmospheric water vapor (Fiorella et al., 2019).The deviation generally increases with increasing latitude.In continental areas, the d 18 O of near-surface atmospheric water vapor can be lower than suggested by isotope equilibrium with precipitation due to high evaporation fluxes from lakes (Krabbenhoft et al., 1990;Benson and White, 1994).Similarly, the d 18 O of atmospheric water vapor can be lower than suggested by isotope equilibrium, if precipitation is affected by sub-cloud re-evaporation, as has been reported for monsoon areas (Landais et al., 2010;Wen et al., 2010).Moreover, the equilibrium assumption is often not valid in semi-arid to arid regions, when precipitation is limited to a short period of the year and not representative for the annual average atmospheric conditions (Tsujimura et al., 2007;Voigt et al., 2021).The atmospheric water vapor record presented here supports the validity of the equilibrium assumption at the study site, for annual d 18 O, d 2 H, d-excess and 17 O-excess averages.The agreement remains good at the monthly scale, but significant discrepancies occur for d-excess and 17 O-excess during the summer months when RH is the lowest.Sub-cloud re-evaporation of precipitation can be invoked to explain the low d-excess and 17 O-excess in summer precipitation, whereas d-excess and 17 O-excess of atmospheric water vapor remain stable.At the diurnal scale, primary isotope ratios of atmospheric water vapor can vary strongly, often deviating from the monthly equilibrium value.This can lead to significant model-data discrepancies (Fig. 4).On diurnal scale, 17 O-excess and d-excess of atmospheric water vapor generally agree with the monthly equilibrium water vapor at daytime, when transpiration is high, but significantly deviate at night and in the early morning.Notably, the variations in 17 O-excess of atmospheric water vapor over the 24-hour monitoring are low (45 per meg) compared to its large variability observed in leaf water (120 per meg) (cf.Table 1, S5).In comparison, δ 18 O shows much higher variability in atmospheric water vapor (5 ‰) compared to leaf water (8 ‰) (cf.Table 1, S5).

Does the 17 O-excess of grass leaf phytoliths reflect daily or daytime RH?
The triple oxygen isotope composition of bulk grass leaf phytoliths is influenced by their distribution along the leaf blade in relation to the leaf water isotope gradient and to silicification patterns (Alexandre et al., 2019;Outrequin et al., 2021).The triple oxygen isotope gradient along grass leaf blades can be predicted by a string-of-lakes model (Alexandre et al., 2019).
However, the triple oxygen isotope composition of the bulk grass leaf water is independent of grass leaf length and predictable by the C-G model combined with the mixing equation (Eq.( 4)) (Alexandre et al., 2019) or a Péclet effect.The bulk phytolith FW integrates the whole grass elongation period and is thus different from the sampled bulk leaf waters that only represent a snapshot in time.Short cells are among the first cell types to be silicified, sometimes even before the leaf transpires (Motomura et al., 2004;Kumar et al., 2017).The process is metabolically controlled and does not depend on the transpiration rate.Long cell silicification occurs in a second step in relation to transpiration (Motomura et al., 2004;Kumar et al., 2017).Moreover, in grass leaves, the epidermal cells are produced at the base of the leaf and pushed upward during the growth.Hence, epidermal cells along the leaf blade gather phytoliths that were formed at short and long distances relative to the leaf base, i.e. at isotopically low and high evaporative conditions, respectively.The combination of these two processes likely causes the apparent λphyto-H2O being lower than the established θSiO2-H2O (=0.524;Sharp et al., 2016) (Outrequin et al., 2021).The consistency of λphyto-H2O equal to 0.522 ± 0.001 observed in this study and in previous calibrations (Outrequin et al., 2021), supports that the silicification patterns are systematic and similar for different climate conditions.
The relationship between 17 O-excessphyto and RH was previously investigated in two growth chamber experiments where F. arundinacea was grown under different conditions of RH (40-60-80 %) and Tair (20-24-28 °C) (Alexandre et al., 2018;Outrequin et al., 2021).The parameters were set constant for more than 10 days, without day-night cycles.Differences in δ 18 O values between source water and atmospheric water vapor were set to 0 ‰ in the first experiment (Alexandre et al., 2018) and to 10 ‰ in the second experiment (Outrequin et al., 2021).The two equations obtained from these experiments were statistically similar (Outrequin et al., 2021).Linear regression through both datasets (n = 16) gives: RH = 0.22 (± 0.01) 17 O-excessphyto + 115.2 (± 3.9) (r 2 = 0.94) Here, under natural conditions, we investigate whether the RH obtained from Eq. ( 7) reflects daytime or daily conditions.RH values reconstructed from 17 O-excessphyto obtained for the three regrowths applying Eq. ( 7) are closer to daytime averages (underestimation of RH by 4 ± 4 %) than to daily averages (underestimation of RH by 12 ± 5 %) (Fig. 7a, Table 2). 17O-excessphyto (Fig. 7).Further, the low stomatal conductance of grasses observed at night causes its leaf water to deviate from isotope steady state.Hence, the 17 O-excess of grass leaf water at night remains close to daytime values (Fig. 6).The low amount of phytoliths that may form overnight thus introduces little bias to the 17 O-excessphyto of the phytolith sample.Nighttime stomatal conductance, however, can vary across biomes, depending among others on plant functional types and soil moisture (Tobin and Kulmatiski, 2018;Resco de Dios et al., 2019).A recent data compilation reported that tropical trees show the highest stomatal conductance at night, followed by desert species (Resco de Dios et al., 2019).The lowest stomatal conductance was found for non-tropical evergreen angiosperms including Mediterranean species.Therefore, for a given case, the magnitude of night-time vs daytime transpiration must be assessed to determine whether the RH reconstructed from 17 O-excessphyto reflects day and night or only daytime conditions.
RH estimated from 17 O-excessphyto can be biased by variations in ΔTleaf-air.This is because the isotope composition of leaf water is not directly determined by RH, but rather the water vapor pressure ratio between the leaf and the atmosphere, i.e. h (cf.Eq.
2).As discussed in Section 4.2, ΔTleaf-air of -2 ºC lead to h that are 5-10% higher than RH.The calibration line obtained from growth chamber experiments is calibrated for ΔTleaf-air of -2 ºC (Outrequin et al., 2021).The lower ΔTleaf-air ranging from -1.1 ºC to 0.3 ºC observed in our study can explain the general underestimation of RH reconstructed from the calibration line (cf. Fig. 7a).The effect of ΔTleaf-air can be removed when considering h instead of RH.We used the same datasets from the growth chamber experiments as for RH (Alexandre et al., 2018;Outrequin et al., 2021;n = 16) to obtain a relationship between 17 O-excessphyto and h, assuming that Tleaf was 2 °C lower than Tair: h = 0.25 (± 0.02) 17 O-excessphyto + 130.0 (± 4.4) (r 2 = 0.94) h values reconstructed from 17 O-excessphyto obtained for the three regrowths applying Eq. ( 8) are in good agreement with corresponding observed daytime averages (Fig. 7b, Table 2).The deviations between reconstructed and measured daytime h values (1 ± 5 %) are lower than for RH (-4 ± 4 %).However, the difference is insignificant considering the uncertainty on the reconstructed values (4 %).A small amplitude of ΔTleaf-air, as observed in the present study (< 1.1 °C), has thus little impact on the RH estimates from 17 O-excessphyto.However, the possibility of larger amplitude, especially in the case of cold forests or warm desert areas, should be considered when interpreting 17 O-excessphyto in terms of RH.

Future tracks for reconstruction of past RH from 17 O-excess of phytoliths extracted from soils
Assessing the relationship between 17 O-excessphyto and RH is crucial for accurate reconstructions using phytolith assemblages extracted from sediments, which are supplied by soil phytoliths from the catchment area.Soil phytoliths likely represent several decades of phytolith production.The limited variation of 17 O-excess in meteoric water (Aron et al., 2021;Surma et al., 2021) and atmospheric water vapor, and its insensitivity of 17 O-excessphyto to temperature make it a powerful indicator of RH.The results of the present study reveal that grass leaf phytoliths record daytime RH under the studied eco-climatic conditions but emphasize that daytime vs nighttime stomatal conductance and ΔTleaf-air need to be considered when interpreting 17 O-excessphyto in terms of RH.In soils, the accurate interpretation of 17 O-excessphyto is further complicated by the mixture of phytoliths from transpiring (leaves, inflorescences) and non-transpiring plant tissues (stems).As previously reported, grass stem phytoliths contribute to less than 10 % dry weight of the above-ground grass silica content (Webb and Longstaffe, 2002;Ding et al., 2008;Alexandre et al., 2019).A simple calculation shows that this contribution should increase 17 O-excessphyto of grass phytolith assemblages extracted from soils by less than 20 per meg relative to an only grass leaf blade phytolith sample, biasing RH estimates obtained from Eq. ( 7) by less than 5 % towards higher values.
When tree phytoliths contribute to soil phytolith assemblages, globular granulate phytoliths are abundant (Alexandre et al., 2011(Alexandre et al., , 2018;;Aleman et al., 2012).This phytolith type is assumed to form in the non-transpiring secondary xylem of the wood (Collura and Neumann, 2017).However, investigation of phytolith assemblages extracted from soils under different vegetation types, including grass savannas, wooded savannas and enclosed savannas developed under similar RH conditions show the same range of 17 O-excessphyto values in agreement with the 17 O-excessphyto vs RH relationship obtained from the growth chamber calibration (Alexandre et al., 2018).This suggests that the FW of the globular granulate phytoliths can be affected by evaporation and calls for further investigation of its anatomical origin.

Conclusion
17 O-excess provides useful insights into evaporation processes at the soil-plant-atmosphere interface as it varies little in rainfall and atmospheric water vapor at the annual scale.In this study, a model-data approach was used to interpret the diurnal and seasonal evolution of the triple oxygen isotope composition of F. arundinacea bulk grass leaf water.This example shows that measuring the triple oxygen isotope composition of leaf water contributes to a better understanding of water exchange at the soil-plant-atmosphere interface.
The ability to measure the grass Tleaf showed that ΔTleaf-air is a key determinant of the isotope composition of leaf water.Under the study conditions, it is close to -2 °C at midday, which is in line with the ΔTleaf-air previously observed on F. arundinacea in climate-controlled growth chambers (Alexandre et al., 2019).To gain further insights into this parameter and its variability according to vegetation and climate types, we recommend IR radiometer measurements with spatial coverage as carried out in the present study.
The first continuous record of atmospheric water vapor including d 17 O measurement at a natural site presented here shows that although d 17 O, d 18 O and d 2 H are highly variable at the daily scale, assuming isotope equilibrium between precipitation and atmospheric water vapor is reasonable for these first order parameters at the monthly and annual scales.The second order parameters (d-excess and 17 O-excess) vary little at the daily, monthly and annual scales and are always close to the equilibrium values estimated from precipitation.Further records of the triple oxygen isotope composition of the atmospheric water vapor, facilitated by the use of laser spectrometers, and precipitation will help to generalize this result.
The measured values of 17 O-excessphyto and daytime RH fit well with the 17 O-excessphyto vs RH equation established from previous growth chamber experiments (Alexandre et al., 2018;Outrequin et al., 2021).However, we emphasize that the magnitude of night-time stomatal conductance and transpiration needs to be assessed in each study individually to evaluate if RH reconstructed from 17 O-excessphyto reflects daily or daytime conditions.Small ΔTleaf-air of less than 2 °C as observed in the present study have little impact on the RH estimates from 17 O-excessphyto.However, larger ΔTleaf -air as common in cold forests or warm desert vegetation should be considered when reconstructing RH using 17 O-excessphyto in these contexts.The insights gained from this study provide important tracks for the interpretation of 17 O-excess of phytoliths accumulated in soils and sediments in terms of RH.The study also confirms the consistency of 18 αphyto-H2O and λphyto-H2O for grasses, which implies that the distribution of phytoliths along grass leaf blades is virtually invariant.This also opens perspectives for reconstructing past changes in leaf water isotope composition from the triple oxygen isotope composition of fossil grass phytolith assemblages 675 recovered from buried soils and sediments, e.g., useful for land-surface model and data comparisons.

Data availability 710
Data available within the article or its supplementary materials.Additional data (e.g., Li-COR measurement data) will be made available on request.Climate data can be accessed from the COOPERATE database: https://cooperate.eccorev.fr/db.
During the 24-hour monitoring, δ 18 O of atmospheric water vapor increased overnight from about -16 to -12 ‰ and then stabilized.The d-excess and 17 O-excess of atmospheric water vapor showed diurnal variations, reaching respective minimum values of -3.2 ‰ and -10 per meg in the early morning and respective maximum values of 18.4 ‰ and 36 per meg at noon (Table

Figure 1 :
Figure 1: Daily precipitation amount, daily (black) and monthly (red) means of relative humidity (RH) and atmospheric temperature (Tair) measured at 60 cm above the ground next to the grass plot, and the isotope composition of atmospheric water vapor (δ 18 OV, 17 O-excessV, d-excessV) measured at 40 cm height above the grass plot monitored at the O3HP platform from February to November 2021.The three regrowth periods lasting from 17 February-20 May 2021 (spring), from 15 June-27 August 2021 (summer) and from 27 August-23 November 2021 (autumn) are indicated by shaded areas.320

Figure 3 :
Figure3: 24-hour monitoring of F. arundinacea transpiration (E) and stomatal conductance (gs) measured on a single leaf blade using the LI-COR gas exchange system, atmospheric temperature (Tair), plot-scale grass leaf temperature (Tplot), relative 365 humidity (RH), water vapor pressure ratio between leaf and the atmosphere (h), and the observed (circles) and predicted steady state (pale gray curve, Eq. (4)) and non-steady state (orange curve, Eq. (6a)) isotope composition of F. arundinacea leaf water (δ 18 O, δ 2 H, 17 O-excess, d-excess) from 14-15 June 2021.Error bars of isotope data represent analytical precision (see method section).Shaded areas mark nighttime intervals.The isotope composition of grass leaf water is predicted using average environmental conditions over 30 minutes before sampling (Table1, S5, S6).The pale gray shaded area represents model 370 uncertainty (1 SD) of the predicted steady-state leaf water isotope composition estimated using a Monte Carlo simulation (see caption Figure2).The dashed part of the steady state prediction represents the time when grass leaf water isotope composition deviates from steady state due to low transpiration and long leaf water residence times (see discussion for details).

Figure 5 :
Figure 5: 17 O-excess vs δ' 18 O of amount-weighted annual average precipitation, average irrigation water, and the measured isotope composition of phytoliths extracted from F. arundinacea grass leaves harvested on 20 May 2021 (spring), 27 August 2021 (summer), and 23 November 2021 (autumn).Also shown are the formation water (FW) predicted using temperature-dependent 18 ⍺SiO2-H2O from Dodd and Sharp (2010) and λphyto-H2O of 0.522 or 0.524, and the isotope composition of bulk leaf water predicted by the C-G model for steady state conditions combined with the mixing equation (Eq.(4)) using average daytime boundary conditions for the three regrowth periods (Table2).Error bars represent analytical precisions (see methods section), except for precipitation, for which the amount-weighted standard deviation is indicated.

Figure 6 :
Figure 6: Comparison of predicted (small circles) and observed (large circles) F. arundinacea leaf water over a 24-hour period from 14-15 June 2021 in diagrams of (a) 17 O-excess vs δ' 18 O and (b) d-excess vs δ 18 O.Filled circles indicate the steady state model prediction (Eq.(4), TableS5), open circles indicate the non-steady state model prediction (Eq.(6), TableS6).Colours differentiate samples collected between 19:15 and 21:45 (LT) on 14 June 2021 (grey), between 14 June 2021 23:30 and 15 June 2021 08:15 (LT) (blue) and between 10:00 and 19:00 on 15 June 2021 (orange).The black line serves as a guide-of-the-eye for the trend in modelled isotope steady state values.The average isotope composition of the irrigation water over the experimental period is also shown.The global meteoric water line (GMWL) is shown for comparison.

Figure 7 :
Figure 7: Observed 17 O-excessphyto vs average daytime (a) relative humidity (RH), and (b) water vapor pressure ratio between the leaf and the atmosphere (h), for regrowth periods in spring, summer and autumn.The growth chamber calibration lines with 95 % confidence interval (Eqs.(6),(7)) are shown for comparison.At night, low stomatal conductance and transpiration measured on F. arundinacea likely hamper the silicification due to cell water saturation relative to silica formation during daytime transpiration, explaining that daytime RH determines The C-G steady state model associated with a two-pool mixing equation reliably predicts the triple oxygen isotope composition of grass leaf water during daytime, when all model-relevant parameters are measured.The few model-data discrepancies (up to 4 ‰, 9 ‰, 34 per meg for d 18 O, d-excess and 17 O-excess, respectively) are likely related to differences between Tplot and actual Tleaf, variations in the fraction of the unevaporated water pool with changes in transpiration (i.e.Péclet effect), and/or slight differences between measured RH close to the grass plot and actual RH right around the grass leaves.Deviations of the isotope composition of leaf water from steady state at night are well captured by a non-steady state model.These deviations from steady state can also be identified in the 17 O-excess vs d' 18 O system, whereas this is not the case in the d-excess vs d 18 O system.

Fig A2 :
Fig A2: Evolution of (a) δ 18 O, (b) 17 O-excess and (c) d-excess of the irrigation water from March to November 2021.Each data point represents the average isotope composition of the irrigation water over the period between two samples.Error bars are 1 standard deviation (SD).The solid lines and the grey shaded areas indicate mean and SD of the isotope composition of irrigation water averaged over all samples.690

Figure A3 :
Figure A3: Diurnal evolution of atmospheric temperature (Tair), plot-scale grass leaf temperature (Tplot) and mean and 1 standard deviation of leaf temperature measurements on single leaves using the Optris IR thermometer (Tleaf) measured on field days between April and November 2021. 695

Figure A4 :
Figure A4: Diurnal evolution of stomatal conductance (gs) measured on field days between April and November 2021.Black 700 lines show gs of a single grass leaf measured continuously over the day using the Li-COR gas exchange system in hourly resolution.Red points represent gs of different grass leaves measured with the AP4 porometer.

Figure A6 :
Figure A6: Monthly mean and daytime mean of the difference between plot-scale grass leaf temperature (Tplot) and air temperature (Tair).The shaded area represents 1 standard deviation.

Table 1 :
F. arundinacea leaf water isotope composition (δ 18 O, 17 O-excess, and d-excess), stomatal conductance (gs) and transpiration (E) measured on a single leaf blade using the LI-COR gas exchange system, atmospheric temperature (Tair) and 175 relative humidity (RH) at 60 cm height next to the grass plot, plot-scale grass leaf temperature (Tplot), and the ratio of atmospheric vapor pressure at 60 cm height and saturation vapor pressure at Tplot (h), averaged over 30 minutes before sampling on 8 days at midday between May and November 2021 and 14 samplings during a 24-hour period from 14-15 June 2021.The sample ID indicates 'sampling location_plant_species_sample type_sampling date_sampling time'.Plant species 'FA' denotes the C3 grass Festuca arundinacea, sampling date is in the format YYYYMMDD and sampling time in UTC.ΔTleaf-air = Tplot-180Tair.

Table 2 :
Grass and phytolith descriptors, phytolith isotope composition, atmospheric temperature (Tair), plot-scale grass leaf temperature (Tplot), relative humidity (RH) and the ratio between actual atmospheric vapor pressure and saturation vapor 185 pressure at Tplot (h) for the three regrowth periods.Grass height = grass height at the harvest day, LC = proportion of long cell phytoliths on the amount of short and long cell phytoliths in the sample.The silicification rate is inferred from the measured SiO2 concentration in grass leaf blades harvested at the end of the regrowth and the length of the regrowth period, assuming a linear production rate (av.rate).Observed RH and h values are compared to estimated values using 17 O-excessphyto and Eqs.

.3 Extractions and isotope analyses 2.3.1 Irrigation water, precipitation, and atmospheric water vapor A
Picarro L2140-i CRDS (California, USA), operated in 17 O Dual Liquid/Vapor mode was installed on-site for the experiment.
or Hirl et al. (2019), we neglected diurnal changes in W, which should result in only ~ 3 % error in predicted (Farquhar and Cernusak, 2005)(Farquhar and Cernusak, 2005).We adjusted W to fit the observed grass leaf water isotope composition.The best fit was found for W of 6 mol m -2 .Both steady state and non-steady state model calculations were performed for isotope ratios ( 2 H/ 1 H, 17 O/ 16 O and 18 O/ 16 O) independently, and the secondary isotope parameters (d-excess and 17 O-excess) were derived from predicted primary isotope values (δ 17 O, δ 18 O, δ 2 H) using the equations given in Section 1.

Table 1 and
Figure