The AnaEE (European Research Infrastructure “Analyses and Experimentation on Ecosystems”) in natura facility O₃HP (Oak Observatory of the Upper-Provence Observatory) is situated in a deciduous Mediterranean forest about 70 km inland, north of Marseille (France) at an altitude of 680 m above sea level (43.935° N, 5.711° E, Fig. 1a). The forest is dominated by downy oak (Quercus pubescens Willd.) with co-dominant Montpellier maple (Acer monspessulanum L., 25 % of canopy leaf mass). The tree canopy height is between 2 and 6 m. In the understory, typical sub-Mediterranean grasses, forbs, and shrubs are present. In 2021, the stand density was about 3550 trees per hectare with a stand basal area 18.8 m2ha-1, an average leaf area index of about 2 m2m-2, ranging between 1 and 3 m2m-2, and a canopy gap fraction of about 25 % (LAI-2000, LI-COR Inc., Nebraska, USA). In February 2021, a 5.5 m2 plot of the C3 grass species Festuca arundinacea Schreb. was set up in the understory of the forest (Voigt et al., 2023).

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 0.6 m height next to the grass plot and at 1.5, 1.9, 4.4 and 6.15 m height in the oak forest canopy (HMP155, Vaisala Oyj, Vantaa, Finland). Furthermore, we used RH and wind speed measured at 10 m height and Tair measured at 10, 50 and 100 m height on the on-site ICOS Tower. Each parameter was extracted in hourly resolution from the COOPERATE database (Reiter et al., 2015).

Photos of the forest canopy were taken every day from a fixed point to provide information on the oak phenology. Stomatal conductance (gs) and transpiration were continuously measured on a single oak leaf in the sun-crown of the canopy using a gas exchange system (Li-6400 XT, LI-COR Inc., Nebraska, USA) over 24 h in the early, middle and last stage of the oak growing season (on 22–23 June, 27–28 August and 22–23 October, respectively). Simultaneously, the spatial variability of gs, was assessed by hourly gs measurements on the abaxial side of 18 oak leaves in the shade- and sun-crown of six different trees (i.e. 3 leaves per tree), using a porometer (AP4, Delta-T Devices LTD, Cambridge, UK).

All isotopic data presented herein are expressed in delta-notation and normalized on the Vienna Standard Mean Ocean Water – Standard Light Antarctic Precipitation (VSMOW-SLAP) scale. Throughout this study, δ-values, δ′-values, and d-excess are reported in per mil (10-3, ‰), while 17O-excess is reported in per meg (i.e., 0.001 ‰). The isotopic compositions of atmospheric water vapor and precipitation are denoted by the subscripts V and P, respectively.

Figure 1 illustrates the experimental setup for in-situ isotopic monitoring of atmospheric water vapor. The inlets of four 1/4-inch wide PFA tubes (PFA-T4-062-100, Swagelok, Ohio, US) were positioned at 0.4 m above the grass plot, and at 1.5, 3.5, and 12.5 m above ground level (m a.g.l.) – below, within and above the downy oak forest canopy, respectively. From bottom to top, the tubes were approximately of 11.5, 15, 20, and 32 m length. Each tube was continuously pumped at a flow rate of ∼5 Lmin-1 using oil-free diaphragm pumps. Along the entire length, the tubing was insulated and heated to 40–50 °C using self-regulating heat wire to prevent condensation. A funnel covered by a net was placed at each inlet for protection from rain and suction of insects and large aerosol particles. A split of each line was passed to a 16-Port Distribution Manifold (A0311; Picarro Inc., California, USA) coupled to a high-precision vaporizer (A0211, Picarro Inc., California, USA) and a CRDS (L2140-i; Picarro Inc., California, USA). The instrument was installed in an air-conditioned building on the experimental site and was operated in 17O Dual Liquid/Vapor mode. It subsampled air from a selected tube with a flow rate of about 35 mLmin-1 for 70 min before switching to the next tube. During data processing, the first 10 min of each line measurement were discarded to minimize memory effects, and the remaining 60 min were averaged. Such long integration times were necessary to achieve high precision in 17O-excess (see below). From January to May 2021, measurements were alternated between the two tubes positioned at 0.4 m above the grass plot and 12.5 m above the forest canopy, resulting in 9–10 measurements per height and day. From June to December 2021, measurements were alternated between all four heights, resulting in 4–5 measurements per height and day.

The calibration protocol of atmospheric water vapor isotopic data is described in detail in Voigt et al. (2023). Three liquid water standards that covered the expected isotopic range of atmospheric water vapor at the study site were analyzed weekly using an autosampler system (A0325, Picarro Inc., California, USA) coupled to the high-precision vaporizer and the analyzer (Fig. 1). The liquid standards were injected in a dry air stream, produced by a lubricated mobile air compressor (MONTECARLO FC2, ABAC air compressors, Italy) and further dried using two drierite columns and a dry ice trap. Using dry ambient air instead of synthetic air as carrier gas for calibration of atmospheric water vapor isotopic measurements is crucial to avoid any potential matrix effect (Voigt et al., 2022). From January to May, standards were analyzed at four water mixing ratios (3000, 7000, 11 000 and 17 000 ppmv) to assess temporal variability of the mixing ratio dependency. As the observed dependency remained consistent over time, measurements from June to November were conducted only at 11 000 ppmv for VSMOW-SLAP normalization. The mixing ratio dependency functions for the three liquid standards were determined based on three higher-resolved water mixing ratio dependency assessments performed in May 2021, October 2021 and January 2022. From the end of November to December, standards were additionally analyzed at 5000 ppmv as atmospheric water mixing ratios were low. For details on the run architectures see Table A1.

For calibration, raw δ17O, δ18O and δ2H of the liquid standards from four consecutive runs (four weeks) were averaged and subsequently corrected to the water mixing ratio of the measured atmospheric water vapor described by a hyperbolic-linear function: 1δstd,corr=δstd,meas+a1H2Oatm-1H2Ostd+b(H2Oatm-H2Ostd)+c where δ denotes δ17O, δ18O or δ2H (in ‰), and H2O is the measured water mixing ratio (in ppmv) of the standard (std) and the atmosphere (atm), respectively. The coefficients a, b and c describe the mixing ratio dependency functions and are specific for δ17O, δ18O and δ2H and for each standard (Table A2). The mixing ratio corrected values of the standards with the lowest and the highest isotopic values were then used for two-point calibration on VSMOW-SLAP scale, while the third standard with an intermediate isotopic composition served as quality control. The standard deviation of the calibrated quality control standard over the experimental period (n=41) was 0.01 ‰, 0.02 ‰, 0.1 ‰, 5 per meg and 0.1 ‰ for δ17O, δ18O, δ2H, 17O-excess, and d-excess, respectively.

The precision of raw isotopic data was estimated from Allan deviation analysis of 24 h in-situ measurements of atmospheric water vapor at O₃HP in June 2021. While the optimal integration time for δ17O, δ18O, δ2H and d-excess is on the order of a few minutes, the Allan deviation of 17O-excess reaches a minimum (∼7 per meg) at an averaging time of about 1 h (Fig. A1). To assess the precision of calibrated data, a Monte Carlo simulation was applied following Voigt et al. (2022). In 100 000 iterations, random normally distributed values were generated accounting for the standard deviations of (i) raw δ18OV, δ2HV and 17O-excess_V, (ii) the coefficients of the mixing ratio dependency functions of each standard (Table A2) and (iii–iv) the measured and reference values of δ18O, δ2H and 17O-excess of the two standards. In each iteration, raw atmospheric values were calibrated following the above-described procedure.

The Monte Carlo simulation was run for 2 min integration intervals for δ18O, δ2H and d-excess and for 1 h integration intervals for 17O-excess. Over 2 min intervals, the raw Allan deviations were 0.11 ‰ for δ18O, 0.65 ‰ for δ2H and 0.5 ‰ for d-excess, yielding simulated standard deviations of calibrated 2 min mean values of 0.12 ‰, 1.2 ‰ and 1.6 ‰, respectively. Over 1 h intervals, the raw Allan deviations were 0.2 ‰ for δ17O, 0.4 ‰ for δ18O, 2.5 ‰ for δ2H, 7 per meg for 17O-excess and 1.6 ‰ for d-excess, yielding simulated standard deviations of calibrated 1 h mean values of approximately ±0.2 ‰, ±0.4 ‰, ±2.7 ‰ and ±15 per meg, and ±4.2 ‰ for δ17O, δ18O, δ2H, 17O-excess, and d-excess, respectively. For consistency, all isotopic tracers were integrated over 1 h, which also determined the frequency of alternation between measurement heights. This is limited by the resolution achievable for 17O-excess. The simulated standard deviations account for both the measurement precision and the natural variability in the isotopic composition of atmospheric water vapor over the integration interval.

Precipitation was collected in an evaporation-free rainfall collector (Rain Sampler 1, Palmex d.o.o., Zagreb, Croatia; Gröning et al., 2012), sampled monthly from June 2019 to December 2020 and on an event-based interval in 2021. The latter resulted in two snow and 53 rain samples. In addition, an on-site well, an underground spring, two springs in the village Saint-Michel-l'Observatoire and the nearby river Largue were sampled seasonally between May 2020 and October 2021 to constrain the isotopic composition of groundwater. The triple oxygen and hydrogen isotopic composition of precipitation, river, well and spring samples was determined using a second Picarro L2140-i CRDS, operated in 17O-High Precision mode at CEREGE (Aix-en-Provence, France). The measurement protocol was described in detail previously (Vallet-Coulomb et al., 2021). The external reproducibility of a quality control standard (1 SD, n=10) measured along with the samples in each sequence was ±0.02 ‰, ±0.04 ‰, ±0.3 ‰ and ±5 per meg, and ±0.1 ‰ for δ17O, δ18O, δ2H, 17O-excess, and d-excess, respectively.

To evaluate the significance of monthly variability in the isotopic compositions of precipitation and atmospheric water vapor, we performed pairwise comparisons of monthly values using Student's t-test, and applied a Bonferroni correction (Bland and Altman, 1995). For each pair of months, we compared the average isotopic composition (δ-values) with their standard deviations to determine statistically significant differences.

The isotopic composition of water vapor in isotopic equilibrium with precipitation (hereafter equilibrium water vapor) is estimated from: 2RVeq=RPαeq,l-v where RVeq is the isotopic ratio in water vapor in equilibrium with precipitation, RP is the measured isotopic ratio in precipitation and αeq,l-v is the liquid-vapor equilibrium isotopic fractionation factor. We used temperature-dependent equilibrium fractionation factors for 2H1HO/1H2O and 1H218O/1H216O from Majoube (1971). The equilibrium fractionation factor for 1H217O/1H216O was derived from 17αeq,l-v=18αeq,l-v0.529 (Barkan and Luz, 2005). We used Tair measured on the ICOS tower at 10 m a.g.l. Note that using Tair measured at other heights between 0.4–100 m a.g.l. does not significantly influence the results.

Like Graf et al. (2019), we denote the differences between equilibrium water vapor and atmospheric water vapor measured at 12.5 m a.g.l. as: 3Δδ18OVeq-V=δ18OVeq-δ18OV4Δd-excessVeq-V=d-excessVeq-d-excessV5Δ17O-excessVeq-V=17O-excessVeq-17O-excessV. A negative value of Δδ18OVeq-V (Δd-excess_Veq-V, Δ17O-excess_Veq-V) indicates that δ18O (d-excess, 17O-excess) of precipitation is lower than expected from isotopic equilibrium with near-surface atmospheric water vapor. Inversely, positive values of Δδ18OVeq-V (Δd-excess_Veq-V, Δ17O-excess_Veq-V) indicate that precipitation has higher δ18O (d-excess, 17O-excess) compared to the expected equilibrium value.

Synoptic atmospheric variability over Europe can be characterized using weather regimes (Cassou, 2008; Cassou et al., 2005; Michelangeli et al., 1995; Vautard, 1990). Previous studies have demonstrated strong positive correlations between δ18OP and δ2HP and the NAO index in the temperate and Mediterranean area (Baldini et al., 2008; Field, 2010; Deininger et al., 2016; Zannoni et al., 2019). In addition, Zannoni et al. (2019) reported a negative correlation between both d-excess_V and d-excess_P and the NAO index at a coastal lagoon in Venice. These correlations are strongest in central Europe but are often weaker in mountainous and circum-Mediterranean regions, where local orographic effects and moisture recycling modify the isotopic composition of precipitation (Baldini et al., 2008).

Here, we identify the dominant weather regimes during the study period to assess whether these relationships persist. This may be of interest for paleoclimate reconstructions of NAO circulation patterns based on isotopic records. To this end, large-scale atmospheric circulation fields for the North-Atlantic-European region (20–80° N and 80° W–30° E) were obtained from the European Centre for the Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis data and classified over 1989–2022. The classification was based on daily-averaged 500 hPa geopotential height (Z500) fields, interpolated onto a 2.5°×2.5° grid and clustered using the k-means algorithm (Hersbach et al., 2020). Each day was assigned to a specific weather regime based on the criterion of minimum Euclidian distance.

To identify oceanic moisture sources of the local atmospheric water vapor, backward trajectory analysis was performed using the Hybrid Single Particle Lagrangian Integrated Trajectory model (HYSPLIT) (Stein et al., 2015). Meteorological input data were obtained from the Global Data Assimilation System (GDAS) dataset at a spatial resolution of 1°×1°. As most of the atmospheric moisture resides in the lowermost 2000 m a.g.l. (Bershaw et al., 2012), back-trajectories were initiated at 1500 m a.g.l. Note that back-trajectories initiated at 500 and 3000 m showed similar paths. Trajectories were computed hourly throughout 2021 extending 168 h (7 d) backward. This resulted in a total of 8760 trajectories covering 365 measurement days.

Moisture uptake zones were quantified using a Lagrangian moisture source diagnostic following Sodemann et al. (2008). Along each backward trajectory, positive increments in specific humidity (Δq>0) calculated at 6 h intervals were interpreted as moisture uptake from the underlying surface. Each uptake location was assigned to the midpoint of the corresponding trajectory segment. Only uptake events exceeding 0.2 gkg-1 and occurring below 1.5 times the planetary boundary layer height were considered in order to reduce the influence of numerical noise and restrict moisture contributions to boundary-layer exchange processes. Each uptake event was weighted by its relative contribution to the final specific humidity at the study site. If a decrease in specific humidity (Δq<0), interpreted as precipitation loss, occurred downstream of one or more uptake events, the contributions of the preceding uptakes were proportionally reduced to account for moisture removal. This procedure yields, for each trajectory, the fractional contribution of a grid cell to q at the study site.

In order to assess the potential relation between the isotopic composition of atmospheric water vapor at the study site and its oceanic moisture source region, we applied a modified Concentration Weighted Trajectory (CWT) approach. A spatial grid of 1°×1° resolution was defined over the study domain. Instead of weighting the isotopic composition measured at the study site by the residence time in each grid cell, as done by Salamalikis et al. (2015), we weighted it by the previously derived fractional moisture contribution τ(ij)k of the corresponding hourly trajectory. The isotopic signature assigned to each grid cell (i,j) was computed as: 6Cij=∑k=1NfijkCk∑k=1Nfijk where Ck is the isotopic composition (δ18O, d-excess, or 17O-excess) measured upon arrival of trajectory k at the study site, and fijk is the fractional moisture contribution of grid cell (i,j) to trajectory k. Thus, Cij represents the moisture-contribution-weighted isotopic composition associated with evaporation from grid cell (i,j). For precipitation, all air mass back-trajectories of hours during which rain was detected on-site were compiled and weighted by rainfall amount.

For regional interpretation, we aggregated grid cells into eight predefined source regions after Sodemann and Zuber (2010): NW = Northwest North Atlantic, NE = Northeast North Atlantic, SW = Southwest North Atlantic, SE = Southeast North Atlantic, A = Arctic and Nordic Seas, CE = Central Europe, MW = Western Mediterranean, ME = Eastern Mediterranean. To focus on oceanic moisture sources, only grid cells located over ocean surfaces were considered. Continental areas were masked using the GLDAS Land/Sea Mask Dataset at 1° resolution (NASA/GES DISC, 2026).

Monthly averages of sea surface temperature (SST), Tair and dew point temperature at 2 m above the ocean surface (Tdew) were obtained from ERA5 hourly data for the region spanning 20–70° N latitude and -50–40° E longitude (Hersbach et al., 2020). RH over the ocean was derived from Tair and Tdew and subsequently normalized to SST (RH_SST).

The triple oxygen and hydrogen isotopic composition of water vapor evaporated from the ocean was estimated using a simple evaporation model (Craig and Gordon, 1965), assuming that all water vapor above the ocean is derived from local evaporation (i.e. the closure assumption; Merlivat and Jouzel, 1979). Seawater δ18O and d-excess values were taken from the CISE-LOCEAN seawater isotopic database (waterisotopes-CISE-LOCEAN, 2025) and range from 0.3 ‰–1.3 ‰ and -0.9 ‰ to -2.3 ‰, respectively. For 17O-excess of seawater, the global average value of -5 per meg from the only existing compilation is used (Luz and Barkan, 2010). The turbulence coefficient is an empirical parameter that quantifies the contribution of molecular diffusion to the total kinetic fractionation during evaporation. Several studies have attempted to constrain this parameter empirically (e.g., Pfahl and Wernli, 2008; Uemura et al., 2010; Zannoni et al., 2022) and to develop theoretical frameworks (Horita et al., 2008; Xia et al., 2023; Dütsch et al., 2026). The turbulence coefficient has been adjusted to 0.33, which is at the upper end of previously reported values for open-ocean evaporation (see, e.g., Gat, 1996; Pfahl and Wernli, 2008; Uemura et al., 2010; Dütsch et al., 2026). Lower turbulence coefficients correspond to lower kinetic fractionation during evaporation and therefore lower d-excess and 17O-excess values in the evaporated water vapor.

In 2021, annual average Tair and RH at O₃HP were 12 °C and 63 %, respectively. The annual precipitation was 730 mm, which is about 150 mm lower than the average annual precipitation from 2008–2020 (Reiter et al., 2015). As typical for a Mediterranean climate, more than 70 % of precipitation occurred from April to May and from September to December, while summer months were the driest (Fig. 2a). The monthly atmospheric water mixing ratio (χ) ranged from 7 mmolmol-1 in winter to 22 mmolmol-1 in summer. Despite the increase in χ during summer, the lowest monthly average RH was recorded in August (53 %), while the highest monthly average RH was observed in February (75 %), due to higher monthly average Tair in summer (JJA, 21 °C) than in winter (DJF; 6 °C). The daily average maximum planetary boundary layer height increased from 600 m a.g.l. in December to 1700 m a.g.l. in July. The average wind speed at 10 m height over daily mean values was 2.6 ± 1.1 ms-1 in 2021. One to eight windy days (daily average wind speed >4 ms-1; Obermann et al., 2018) were observed per month. Windy days are mostly indicative for Mistral conditions – a strong northerly wind typical of the Provence region – and occurred most frequently in winter and spring.

In 2021, daytime average χ was 2 to 6 mmolmol-1 higher than at night. Daytime average Tair was 6 to 16 °C higher than at night, while average RH increased by 26 % to 50 % from day to night. The largest day-night differences occurred near the ground and in summer (Figs. B1–B3). During the day, χ and Tair were highest near the ground, whereas RH was vertically homogenous (Figs. B1–B3). At night, the vertical gradient in Tair and χ weakened. Nighttime cooling resulted in higher RH that decreased from the surface to 10 m a.g.l. (Figs. B1–B3).

The first oak buds appeared in mid-April 2021. The oak leaves matured in May and senesced in November. Their stomatal conductance and transpiration followed the diurnal cycle of global solar radiation and RH (Fig. B4). During the three separate days of continuous measurement, daytime stomatal conductance and transpiration ranged from 20–60 mmolmol-1 and 0.5–1.8 mmolmol-1, respectively. During the night, transpiration and stomatal conductance were always lower than 20 mmolmol-1 and 0.1 mmolmol-1, respectively. Transpiration was highest at the beginning of the growing season in June, and the lowest in August, mainly due to lower oak leaf stomatal conductance in response to water stress (high vapor pressure deficit).

The annual amount-weighted average isotopic values of precipitation in 2021 were -7.5 ‰ ± 2.3 ‰ for δ18OP, -48.1 ‰ ± 20.2 ‰ for δ2HP, 11.6 ‰ ± 3.5 ‰ for d-excess_P and 27 ± 9 per meg for 17O-excess_P (Fig. 3). These values are close to those obtained for 2020 and indistinguishable from those obtained for the whole precipitation dataset from June 2019 to December 2021 (Table B1). The springs, the well and the river Largue sampled seasonally between December 2020 and October 2021 showed little spatial and seasonal isotopic variability (Table B1) and averaged to -7.3 ‰ ± 0.5 ‰ for δ18O, -47.0 ‰ ± 3.6 ‰ for δ2H, 11.2 ‰ ± 0.6 ‰ for d-excess and 29 ± 5 per meg for 17O-excess. These values coincide with the annual amount-weighted average isotopic composition of precipitation.

Monthly amount-weighted average isotopic values of precipitation varied between -11.8 ‰ and -1.1 ‰ for δ18OP, between -78.8 ‰ and -6.5 ‰ for δ2HP, between -1.4 ‰ and 18.2 ‰ for d-excess_P and between 1 per meg and 41 per meg for 17O-excess_P (Table B1). In general, monthly values of δ18OP and δ2HP were low in winter and varied in their upper range from spring to autumn. Inversely, monthly values of d-excess_P and 17O-excess_P were low in summer and tended to higher values in autumn and winter.

The isotopic values of event-based rain samples ranged from -18.4 ‰ to -0.2 ‰ for δ18OP, from -142.5 ‰ to -7.8 ‰ for δ2HP, from -13.0 ‰ to 20.5 ‰ for d-excess_P and from -15 per meg to 51 per meg for 17O-excess_P (Table B1). The two snow samples had the lowest δ18OP (-22.3 ‰ and -18.4 ‰, respectively) and δ2HP (-162.9 ‰ and -142.5 ‰, respectively) of the whole data set, while d-excess_P (15.6 ‰ and 4.0 ‰) and 17O-excess_P (68 per meg and 17 per meg) were variable. The seasonal variability of 17O-excess_P was generally higher than its variability from one event to another. In contrast, variability of δ18OP and d-excess_P was often higher on event-scale than on seasonal scale (Student's t-test).

Linear regression through monthly amount-weighted average precipitation data for 2020 and 2021 (n=22) gives (Fig. 3): 7δ2H=7.9(±0.4)δ18O+9.7(±2.7)8δ′17O=0.5263(±0.0009)δ′18O+0.0128(±0.0064). In the δ2H-δ18O system, the slope of the regression line is indistinguishable from that for Avignon GNIP station (IAEA/WMO, 2025), located 70 km to the west of our study site (Fig. 3). It is close to that of the global meteoric water line (GMWL) (8; Rozanski et al., 1993), but higher than the range of previously reported values for the western Mediterranean area (7.15–7.56; Celle-Jeanton et al., 2001; Giménez et al., 2021; Saighi, 2005). In the δ′17O-δ′18O system, the slope of the regression line is lower than that of the GMWL (0.528; Luz and Barkan, 2010; Terzer-Wassmuth et al., 2023) but similar to the value obtained for temperate and intertropical areas (0.5265; Sharp et al., 2018; Terzer-Wassmuth et al., 2023). Notably, the slope of the regression line varies strongly from 2020 to 2021 for both the δ2H-δ18O system (7.3–8.5) and the δ′17O-δ′18O system (0.5254–0.5271). A longer precipitation isotopic record is required to establish a robust local meteoric water line for the study site.

The annual average isotopic composition of atmospheric water vapor at 12.5 m a.g.l. was -17.7 ‰ ± 2.8 ‰ for δ18OV, -128.1 ‰ ± 21.9 ‰ for δ2HV, 13.8 ‰ ± 1.7 ‰ for d-excess_V and 30 ± 5 per meg for 17O-excess_V, and indistinguishable from that measured at lower heights. Similar to precipitation, δ18OV and δ2HV were low in winter (-23.6 ‰ and -174.2 ‰, respectively) and high from spring to autumn, peaking in June (-14.6 ‰ and -105.2 ‰, respectively) (Table B2). Values of d-excess_V were high in October and November (>16 ‰) and low from May to July (<12 ‰). In contrast, 17O-excess_V varied little throughout the year (from 24 per meg to 40 per meg). Low values (25–27 per meg) were recorded between May and August 2021, while highest values were observed in December (35 per meg) and January (40 per meg).

Day-to-day isotopic variability of atmospheric water vapor was high (up to 9.3 ‰, 16.0 ‰ and 27 per meg for δ18OV, d-excess_V and 17O-excess_V, respectively) and exceeded monthly and seasonal variability for d-excess_V and 17O-excess_V (Fig. 2). There is no statistically significant difference in the daily average isotopic composition of atmospheric water vapor on rainy and non-rainy days, indicating that precipitation had no significant influence on the isotopic composition of near-surface atmospheric water vapor on daily and monthly scale.

Diurnal patterns of d-excess_V and 17O-excess_V changed from a season to another (Figs. B6 and B7). They can be illustrated by two examples (Fig. 4): In June, the month with the most active leaf gas exchange in the Mediterranean (Bartsch et al., 2020), d-excess_V and 17O-excess_V were lowest at sunset and highest at noon, coinciding at noon with minimum RH, maximum transpiration and maximum χ measured under the forest canopy. In December, a month characterized by the absence of alive tree foliage and transpiration and a thin planetary boundary layer, the amplitude of the diurnal cycle was three times lower than in June for the d-excess_V and more than 15 times lower for the 17O-excess_V. Diurnal isotopic patterns were consistent for different heights, but the magnitude of variations decreased with distance from ground. Vertical isotopic gradients were generally stronger during night, in line with limited atmospheric turbulent convection (Fig. B5).

Figure 5 presents the relationship between the measured isotopic composition of near-surface atmospheric water vapor and that expected from isotopic equilibrium with precipitation. Following the framework of Graf et al. (2019), deviations from isotopic equilibrium can indicate cloud-dominated conditions resulting from incomplete equilibration between falling raindrops and ambient water vapor, or evaporation-dominated conditions associated with sub-cloud rain re-evaporation.

In winter, spring and late autumn, δ18OVeq was up to 5.5 ‰ lower than δ18OV, while in summer δ18OVeq was up to 3.7 ‰ higher than δ18OV. Values of d-excess_Veq were generally close to d-excess_V in winter, but up to 16 ‰ lower than d-excess_V during the oak growing season. Values of 17O-excess_Veq were slightly lower than 17O-excess_V in winter (∼15 per meg) and up to 44 per meg lower during the oak growing season (Fig. 5).

However, on annual scale, δ18OV (-17.8 ‰ ± 2.8 ‰), δ2HV (-128.3 ‰ ± 21.9 ‰) and d-excess_V (13.8 ‰ ± 1.7 ‰) were similar to their values expected from isotopic equilibrium with precipitation (-17.8 ‰ ± 2.7 ‰, -130.1 ‰ ± 21.6 ‰ and 12.3 ‰ ± 3.1 ‰, respectively). In contrast, 17O-excess_V (30 ± 5 per meg) was in annual average 14 per meg higher than 17O-excess_Veq (16 ± 10 per meg).

Monthly average δ18OV was strongly correlated with Tair and χ (r2=0.67 and 0.63, respectively). In contrast, d-excess_V was only weakly correlated with these variables (r2=0.26 and 0.29, respectively), and 17O-excess_V showed no significant relationship (Fig. B8). There was no correlation between the monthly average isotopic composition of atmospheric water vapor and RH or wind speed.

Hourly means of δ18OV, d-excess_V and 17O-excess_V averaged by month (Figs. B6 and B7) were strongly correlated with RH, Tair and transpiration during the oak forest growing season (r2>0.55, >0.73 and >0.64, respectively). Correlations between 17O-excess_V and d-excess_V were also strong during the oak forest growing season (r2=0.52–0.93), peaking in June, whereas they were not correlated in winter.

For precipitation, monthly average δ18OP and d-excess_P were weakly correlated with χ, Tair and RH (0.28≤r2≤0.38), whereas 17O-excess_P showed stronger correlations with these variables (0.44≤r2≤0.61). No correlation was observed between the monthly average isotopic composition of precipitation and precipitation amount.

At the event scale, δ18OP and 17O-excess_P remained weakly correlated with Tair (r2=0.41 and 0.43, respectively) and χ (r2=0.34 and 0.29, respectively), while d-excess_P showed little or no correlation (r2<0.23) (Fig. B9). Event-based isotopic values of precipitation showed no correlation with RH or precipitation amount.

The moisture source contributions for atmospheric water vapor and precipitation are illustrated in Figs. C1 and C2. For atmospheric water vapor, the NE North Atlantic (33 %) and the Western Mediterranean (29 %) are the dominant moisture sources for our study site (Table C1). For precipitation, most of the moisture is derived from the Western Mediterranean (41 %), followed by the NE North Atlantic (25 %) and the SE North Atlantic (18 %) (Table C2). Changes in the relative contributions of each moisture source over the year are illustrated in Figs. C3 and C4 for atmospheric water vapor and precipitation, respectively.

The isotopic compositions of atmospheric water vapor and precipitation clustered by moisture sources were generally similar (Tables C1 and C2). In particular, d-excess and 17O-excess of atmospheric water vapor or precipitation coming from the North Atlantic sectors and the Western Mediterranean were not significantly different.

We also compared the monthly average isotopic composition of atmospheric water vapor observed at O₃HP with those estimated above the four oceanic moisture sources (Fig. 6). The latter shows seasonal isotopic variability which co-varies with SST and RH_SST. Interestingly, from July to December, RH_SST over the Mediterranean Sea is systematically lower than over the NE North Atlantic but does not differ from RH_SST over the SE North Atlantic. For all sources, δ18OV is higher and d-excess_V and 17O-excess_V are lower in summer than in winter. A similar seasonal trend is evident in the isotopic composition of the water vapor observed at O₃HP (Fig. 6). However, the δ18OV values at O₃HP were 3 ‰–11 ‰ lower than those estimated for the ocean sources. In contrast, the d-excess_V at O₃HP was of the same order of magnitude as that over the ocean sources, while the observed 17O-excess_V was slightly higher (by 11 ± 5 per meg).

Four weather regimes were identified for wintertime (November–March) 2021 in the North Atlantic area: the negative North Atlantic Oscillation (NAO⁻), the positive North Atlantic Oscillation (NAO⁺), the Scandinavian Blocking regime, and the Atlantic Ridge regime (Fig. C5). The Atlantic Ridge and NAO⁻ regimes were dominant, whereas NAO⁺ occurred only sporadically. In summer 2021 (May–September), the NAO⁻, Icelandic Low, Atlantic Low, and Scandinavian Blocking/NAO+ regimes were identified. The Blocking/NAO+ regime was clearly the dominant, followed by NAO⁻. Precipitation occurred most frequent during NAO⁻ phases (Fig. C6c and d). The isotopic compositions of water vapor and precipitation grouped by weather regime did not show significant differences (Fig. C6).

The three principle oceanic moisture sources show lower RH_SST during winter than in summer, resulting in stronger kinetic fractionation during evaporation from the ocean and higher d-excess_V and 17O-excess_V in the evaporated water vapor in winter than in summer (Fig. 6). This seasonal change predicted for water vapor over the oceanic moisture sources is also visible in the d-excess_V measured at O₃HP and to a lesser extent in the 17O-excess_V. This consistency suggests that seasonal changes in evaporative conditions at the oceanic moisture sources contribute to seasonal changes in d-excess_V and 17O-excess_V at the study site. This is in agreement with previous interpretations of the triple oxygen isotopic composition of water vapor and precipitation (Affolter et al., 2015; Landais et al., 2012; Merlivat and Jouzel, 1979; Pfahl and Sodemann, 2014; Tian et al., 2018; Uechi and Uemura, 2019). However, with regard to δ18OV, seasonal changes in the evaporative conditions at the oceanic moisture sources are limited to 1.5 ‰ and cannot explain the ∼7 ‰ seasonal variations observed at the study site. This indicates that δ18OV is likely modified by further fractionation processes.

In previous studies, evaporative conditions at the moisture sources were invoked to explain d-excess_P in precipitation from the Western Mediterranean (∼14 ‰) to be higher than in precipitation from the NE North Atlantic (∼10 ‰) (Casellas et al., 2019; Celle-Jeanton et al., 2001; Cruz-San Julian et al., 1992; Delattre et al., 2015; Natali et al., 2021). Theoretical predictions indeed show that d-excess_V over NE North Atlantic is about 3.7 ‰ lower than over Western Mediterranean due to higher RH_SST (Fig. 6). Similarly, 17O-excess of water vapor evaporated from the Western Mediterranean is predicted to be ∼6 per meg higher than from the NE North Atlantic, although the magnitude remains within measurement precision. However, d-excess and 17O-excess measured in atmospheric water vapor and precipitation at O₃HP grouped by their primary moisture source (NE North Atlantic vs. Mediterranean) were not significantly different. Several factors may have contributed to diminish the isotopic contrast between these source regions, including frequent mixing of moisture from multiple sources, local orographic and boundary-layer dynamics or moisture recycling during atmospheric transport (Aemisegger et al., 2014; Natali et al., 2022).

Along the air mass trajectory, the water vapor undergoes Rayleigh distillation due to the progressive rainout. Rainout is forced by a decrease in temperature, accompanied by a decrease of χ and leads to a decrease of δ17OV, δ18OV and δ2HV and subsequent precipitation (Dansgaard, 1964; Galewsky et al., 2016). Rayleigh distillation explains why the monthly average δ18Ov is systematically lower than at the oceanic moisture sources (Fig. 6). Furthermore, enhanced rainout along the longer transport distance from the North Atlantics likely explain why the δ18OV derived from the NE North Atlantic is 1 ‰–3 ‰ lower than δ18OV from the Western Mediterranean (Delattre et al., 2015; Natali et al., 2021).

In contrast, Rayleigh distillation has limited impact on d-excess_V and 17O-excess_V as, by definition, these parameters are weakly sensitive to equilibrium isotopic fractionation (Merlivat and Jouzel, 1979; Xia et al., 2023). In mid-latitudes Rayleigh distillation is expected to cause up to 3 ‰ changes in d-excess_V, with the direction of change depending on the season and the degree of distillation (Xia et al., 2022). The effect should not exceed a 5 per meg increase for 17O-excess_V (Xia et al., 2023). Nevertheless, enhanced rainout along transport pathways from the North Atlantic may have attenuated potential moisture source differences in d-excess_V and 17O-excess_V.

On a daily scale, χ, d-excess_V and 17O-excess_V are higher and δ18OV is lower during the day (with a peak in the afternoon) than at night. Such a day-night pattern has been reported for d-excess_V across diverse environments, including forests, grasslands, wetlands, agricultural areas and urban regions (Berkelhammer et al., 2013; Delattre et al., 2015; Welp et al., 2012) and attributed to several processes, including entrainment of the residual layer or air from the lower free troposphere in the boundary layer (Lai and Ehleringer, 2011; Simonin et al., 2014), local evapotranspiration fluxes (Huang and Wen, 2014; Simonin et al., 2014; Welp et al., 2012; Zhao et al., 2014; Delattre et al., 2015), and dewfall evaporation (Bastrikov et al., 2014; Berkelhammer et al., 2013). The fact that the observed day-night pattern in d-excess_V and 17O-excess_V is more pronounced during the forest growth period and most pronounced near surface (Fig. 4) supports the key influence of surface-atmosphere processes. Assuming that the soil water has an isotopic composition close to that of precipitation, soil water evaporation decreases δ18OV and increases d-excess_V and 17O-excess during the day (Craig and Gordon, 1965; Rothfuss et al., 2021; Fig. 7). Plant transpiration, which is assumed to be non-fractionating relative to soil water (Galewsky et al., 2016, and references therein) increases δ18OV but has little impact on d-excess_V and 17O-excess_V (Fig. 7). In total, net evapotranspiration affects δ18OV, with the direction and magnitude depending on the relative contributions of evaporation and transpiration, whereas d-excess_V and 17O-excess_V are elevated only when evaporation is substantial (Fig. 7). Entrainment of a remnant of the convective boundary layer of the previous day, with a high d-excess_V and 17O-excess_V used by evapotranspiration, may amplify the increase in d-excess_V and 17O-excess_V in the morning (Lai and Ehleringer, 2011; Simonin et al., 2014, Griffis et al., 2016; Welp et al., 2012). Entrainment of the lower free troposphere may also occur. However, the low water vapor mixing ratio of the free troposphere (<4000 ppmv) likely limits its influence on the isotopic composition of atmospheric water vapor. During the night, when RH is high, water exchange between plant water subjected to incomplete stomata closure, or evaporated surface soil waters, and the atmospheric water vapor can increase δ18OV and decrease d-excess_V (Berkelhammer et al., 2013; Bastrikov et al., 2014; Lai and Ehleringer, 2011; Simonin et al., 2014; Welp et al., 2012) and 17O-excess_V. The combination of all these factors likely contributes to the observed day-night pattern in δ18OV, d-excess_V and 17O-excess_V. While evapotranspiration contributes to diurnal isotopic variability, opposing daytime and nighttime responses diminish its imprint at monthly timescales. Consequently, local vegetation-driven moisture recycling does not appear to leave a strong isotopic imprint on atmospheric water vapor and precipitation at our site.

The mixing of air masses from different origins frequently occurs during atmospheric transport. The mixing of two air masses with different δ18OV but similar 17O-excess_V and d-excess_V results in a decrease in 17O-excess_V, while d-excess_V remains largely unchanged. The effect on 17O-excess_V differs from that on d-excess_V due to its logarithmic (non-linear) δ′ notation (Li et al., 2015; Aron et al., 2021; Leuenberger and Ranjan, 2021). The decrease in 17O-excess_V due to mixing is however small, representing only 3 per meg for a δ18OV difference of 10 ‰ between two air masses with the same water content (Xia et al., 2023). As isotopic differences in atmospheric water vapor observed between the oceanic moisture sources are small, mixing of air masses from different oceanic sources is unlikely to significantly alter the isotopic composition of atmospheric water vapor observed at the O₃HP site.

Liquid precipitation forms in isotopic equilibrium with atmospheric water vapor at cloud height, while ice formation may be accompanied by an additional kinetic effect due to supersaturation of water vapor (Dütsch et al., 2017; Xia et al., 2023). Vapor at cloud height is typically depleted in heavy isotopes compared to near-surface atmospheric water vapor due to Rayleigh fractionation processes (Giménez et al., 2021, and references therein; Salmon et al., 2019; Sodemann et al., 2017). As raindrops fall through the air column, they continuously exchange water molecules with the surrounding vapour and equilibrate isotopically. This is the basis for the isotopic equilibrium assumption between precipitation and near-surface atmospheric water vapor. During precipitation events in dry atmosphere or during convective storms in warm atmosphere additional fractionation processes can occur. Rain re-evaporation in dry atmosphere increases δ18OP and decreases d-excess_P and 17O-excess_P (Fig. 5). During convective storms, multiple processes, including vertical diffusion, mixing, and rain re-evaporation can influence the isotopic composition of precipitation (Risi et al., 2012; Tremoy et al., 2014; Xia et al., 2022).

At O₃HP during winter months, δ18OV is slightly higher than δ18OVeq and the two snowfall events have the largest Δδ18OVeq-V (-12.3 ‰) (Fig. 5). These differences suggest that δ18OV at ground level is higher than at the height of rain formation, and that equilibration during raindrop falling is incomplete. Indeed, equilibration rates are hampered under cold conditions, and no equilibration occurs between vapor and solid (Graf et al., 2019). Incomplete equilibration can also occur when low χ in the atmosphere is associated with strong vertical isotopic gradients in atmospheric water vapor, when drops are large or precipitation forms at high altitude (Graf et al., 2019; Penchenat et al., 2020). This incomplete equilibration has little effect on d-excess_P (average deviation -0.8 ‰ ± 3.0 ‰) due to the generally small vertical gradients of d-excess_V in the lowermost troposphere (Salmon et al., 2019; Sodemann et al., 2017). Notably, in winter, 17O-excess_P is always close to 17O-excess_V rather than 10–15 per meg lower as expected from isotopic equilibrium (Fig. 5). The reason for this offset remains unclear so far.

During spring and summer months, the systematic correlation between Δδ18OVeq-V, Δd-excess_Veq-V and Δ17O-excess_Veq-V (Fig. 5) suggests the impact of below-cloud rain re-evaporation. Rain re-evaporation has been found to occur under unsaturated conditions in mid-latitude to tropical regions (Aemisegger et al., 2015; Aemisegger and Sjolte, 2018; Graf et al., 2019; Landais et al., 2010; Mercer et al., 2020). Indeed, positive Δδ18OVeq-V values at O₃HP occur when RH is lower than 90 %. Convective storms occurring in summer in the Mediterranean area can strengthen this correlation. The fact that there are no isotopic differences in atmospheric water vapor between rainy and non-rainy days (data not shown) implies that sub-cloud rain re-evaporation does not significantly affect near-surface atmospheric water vapor. The contribution of plant transpiration and soil evaporation to the near-surface atmospheric water vapor may emphasize the isotopic disequilibrium between precipitation and atmospheric water vapor during daytime in summer (cf. Sect. 4.3).

At annual scale, the opposing winter and summer monthly Δδ18OVeq-V values at our study site effectively cancel each other out, so that δ18OVeq closely approximates δ18OV (cf., Voigt et al., 2023). However, this may differ depending on the environmental setting. Significant deviations from isotopic equilibrium at annual scale were observed in monsoonal areas due to rain re-evaporation (Desphande et al., 2010; Landais et al., 2010; Wen et al., 2010), convective activity (Risi et al., 2012; Tremoy et al., 2014; Xia et al., 2022) or in environments with seasonal precipitation, where the isotopic composition of precipitation is not representative for annual average atmospheric water vapor (Tsujimura et al., 2007; Voigt et al., 2021).

The absence of correlation between the weather regimes and the isotopic composition of vapor and precipitation is probably due to the combined influence of multiple moisture sources weakly isotopically contrasted as presented in Sect. 3.6. In addition, local moisture recycling may further modify the isotopic signal of atmospheric water vapor and, thus, precipitation, as previously suggested for the circum-Mediterranean region (Baldini et al., 2008). However, opposing day–night vegetation-related isotopic variations that largely cancel each other out on the monthly scale, indicate a limited net influence of local moisture recycling. It is also possible that the isotopic composition of near-surface atmospheric water vapor differs slightly from that at the precipitation formation height due to ground roughness, convection, mixing processes in the planetary boundary layer (Griffis et al., 2016; Salmon et al., 2019; Tada et al., 2021).

Although monthly average d-excess_v measured at the O₃HP is close to estimated d-excess_v at the oceanic moisture sources, as expected from Rayleigh distillation, monthly average 17O-excess_V is systematically higher by about 15 per meg. This inconsistency may result from an inaccurate 17O-excess average value used for sea water (-5 per meg) in the calculation of 17O-excess_V above the oceanic moisture sources. To date, the dataset of 17O-excess in seawater is limited (Lin et al., 2021; Luz and Barkan, 2010), so that estimates of oceanic 17O-excess_V are not very accurate. Uncertainty in the turbulence coefficient may also contribute to the mismatch between observed 17O-excess_V and model predictions. Increasing the turbulence coefficient reduces the offset but simultaneously worsens the agreement for d-excess_V and contradicts previous studies (e.g., Gat, 1996; Pfahl and Wernli, 2008; Uemura et al., 2010, Dütsch et al., 2026). It therefore cannot, by itself, explain the discrepancy. More likely, the closure assumption during ocean evaporation does not fully represent actual isotopic fractionation. For example, aerodynamic conditions over oceans and mixing of water vapor from other sources into the marine boundary layer can decouple 17O-excess and d-excess signals (Xia et al., 2023).

A calibration artifact caused by inaccuracies in the mixing ratio dependency functions or a weak matrix effect cannot be completely excluded. However, it is unlikely, as the quality control standard was routinely measured and accurately calibrated (Sect. 2.2). In any case, such effects would introduce systematic biases and would not alter the main conclusions presented here.