Deuterium excess in the atmospheric water vapour of a Mediterranean coastal wetland: regional vs. local signatures

Stable isotopes of water vapour represent a powerful tool for tracing atmospheric vapour origin and mixing processes. Laser spectrometry recently allowed high timeresolution measurements, but despite an increasing number of experimental studies, there is still a need for a better understanding of the isotopic signal variability at different time scales. We present results of in situ measurements of δO and δD during 36 consecutive days in summer 2011 in atmospheric vapour of a Mediterranean coastal wetland exposed to high evaporation (Camargue, Rhône River delta, France). The mean composition of atmospheric vapour (δv) is δO=−14.66 ‰ and δD=−95.4 ‰, with data plotting clearly above the local meteoric water line on a δO-δD plot, and an average deuterium excess (d) of 21.9 ‰. Important diurnal d variations are observed, and an hourly time scale analysis is necessary to interpret the main processes involved in its variability. After having classified the data according to air mass back trajectories, we analyse the average daily cycles relating to the two main meteorological situations, i.e. air masses originating from North Atlantic Ocean and Mediterranean Sea. In both situations, we show that diurnal fluctuations are driven by (1) the influence of local evaporation, culminating during daytime, and leading to an increase in absolute water vapour concentration associated to a δv enrichment and d increase; (2) vertical air mass redistribution when the Planetary Boundary Layer collapses in the evening, leading to a d decrease, and (3) dew formation during the night, producing a δv depletion with d remaining stable. Using a twocomponent mixing model, we calculate the average composition of the locally evaporated vapour (δE). We find higher d(E) under North Atlantic air mass conditions, which is consistent with lower humidity conditions. We also suggest that δv measured when the PBL collapses is the most representative of a regional signal. Strong, cold and dry winds coming from the north bring an isotopically depleted vapour, while light, warm and wet winds coming from the south bring an isotopically enriched vapour. Under northern conditions, a strong advection rate dilutes the contribution of the locally evaporated vapour (δE) to the ambient moisture (δv). The higher d values measured under northern conditions, compared to the Mediterranean situation, thus results from the combination of a higher d in both local and regional vapour. This depiction of typical daily cycles of water vapour isotopic composition can be used as a framework for further quantitative analyses of vapour sources during specific days.


Introduction
At the global scale, an acceleration of the hydrological cycle is expected in response to climate change, with an intensification of both precipitation and evaporation. This effect has been recognized for the ocean (Dai et al., 2009), but less conclusive patterns appear for continental surfaces, especially because of the complex behaviour of land evapotranspiration (Ohmura and Wild, 2002;Roderick and Farquhar, 2002;Brutsaert, 2006;Roderick et al., 2007;Fu et al., 2009;Jung et al., 2010). The response of evapotranspiration to climate change is controlled both by climatic and by hydrological parameters. Climate determines the atmospheric evaporative demand (potential evapotranspiration), while hydrology controls surface water availability, thus limiting actual evapotranspiration. In addition, continental vapour is involved in land-atmosphere feedbacks such as atmospheric Published by Copernicus Publications on behalf of the European Geosciences Union.
water recycling; high evapotranspiration may contribute to regional rainfall in contexts of tropical rainforest (Brubaker et al., 1993;Eltahir and Bras, 1996;van der Ent et al., 2010) or large scale irrigation (Boucher et al., 2004;Tuinenburg et al., 2012), or may even reduce potential evapotranspiration by lowering temperature and increasing humidities near the earth surface (Destouni et al., 2010;Tuinenburg et al., 2012). A better understanding of the atmospheric component of the water cycle and of the potential contribution of wetlands evaporation to the regional water budget is thus important to anticipate the impact of global change, in particular in Mediterranean regions where more frequent droughts are expected.
Isotopic composition of atmospheric water vapour (subsequently called δ v ) represents a great opportunity to explore land-atmosphere interactions, as the addition of moisture originating from evapotranspiration induces a modification of the isotopic content of the background vapour (Salati et al., 1979;Gat and Matsui, 1991;Gat et al., 1994;Gat, 2000;Worden et al., 2007;Yamanaka and Shimizu, 2007;Vallet-Coulomb et al., 2008;Risi et al., 2013). In addition, isotope tracers are now often incorporated into climate models (Sturm et al., 2005;Werner et al., 2011;Risi et al., 2012Risi et al., , 2013 in order to improve the representation of landatmosphere interactions and feedbacks. A major challenge is thus to propose robust databases, and associated methodology for data acquisition and interpretation. Laser spectrometers recently opened the door towards the high temporal resolution analysis of δ v variations, which was not possible with the traditional cold trap method. Whereas cryogenic trapping provides grab samples generally averaging a couple of hours, laser spectrometry performs continuous high frequency (up to ∼ 1 Hz) measurements of δ v (Kerstel and Gianfrani, 2008). This technology offers new insights into processes that affect the isotopic composition of atmospheric water vapour and the number of studies based on continuous ground level isotope measurements over multiweek periods is continuously increasing, either using a tunable diode laser (TDL) (Lee et al., 2006;Wen et al., 2010Wen et al., , 2012Griffis et al., 2011;Welp et al., 2012), off-axis integrated cavity output spectroscopy (OA-ICOS) (Sturm and Knohl, 2010;Sunmonu et al., 2012;Farlin et al., 2013;Steen-Larsen et al., 2013) or wavelength-scanned cavity ring-down spectroscopy (WS-CRDS) Tremoy et al., 2012;Steen-Larsen et al., 2013. Among these experiments are some low latitude studies which have focused on the tracing of tropical or subtropical convective activity in West Africa (Tremoy et al., 2012) or in South America . In mid latitudes Farlin et al., 2013), or high latitudes (Steen-Larsen et al., 2013), several studies have explored atmospheric mixing processes at different time scales. Isotopes have also been used for partitioning evapotranspiration into plant transpiration and direct evaporation, as the associated fractionations are different (Yakir and Sternberg, 2000;Lai et al., 2006;Wang et al., 2010;Griffis et al., 2011;Sun et al., 2014), for a better understanding of the role of ecosystems in the hydrological cycle in the context of climate change (Wang et al., 2013).
However, the number of studies remains limited because of technical difficulties associated with field-deployed longterm measurements. Except for some work based on cold trap sampling (Jacob and Sonntag, 1991;Williams et al., 2004;Angert et al., 2008), there are to our knowledge only three high time-resolution studies published from European sites (Iannone et al., 2010;Kurita et al., 2012;Aemisegger et al., 2014) and none at all treating Mediterranean areas. In addition, there is a need for dual tracer databases, in order to use the deuterium excess signal (d = δD − 8 × δ 18 O; Dansgaard, 1964) as an additional indicator of atmospheric processes. Indeed, d is often considered as a indicator of evaporation conditions, but its interpretation in continental vapour remains complex since it is affected by the multiple vapour sources, combining the initial oceanic vapour and evapotranspiration from different continental water sources (Gat et Matsui, 1991;Vallet-Coulomb et al., 2008;Lai and Ehleringer, 2011;Welp et al., 2012;Jouzel et al., 2013;Aemisseger et al., 2014).
In this paper, we provide results obtained in Camargue on 36 consecutive days during summer 2011, using WS-CRDS technology. The Camargue region is emblematic of Mediterranean wetlands, with important water inflow requirements and strong evaporation losses, making it highly sensitive to climate change and human pressure on water resources. Our experimental site is located in the centre of Camargue, close to the main lagoon (Vaccarès lagoon, 65 km 2 ). Considering the location of the site, and the important availability of open water in the surroundings, we expect a dominant contribution of evaporation upon transpiration fluxes. This study will provide the opportunity to investigate how local evaporation combines and interact with the other regional vapour sources. We will focus on identifying the main drivers of deuterium excess variability at different time scales, and explore the relevance of using either relative humidity, as an indicator of evaporation conditions, or specific humidity (mixing ratio) as a proxy of mixing between different vapour sources. Before analysing our results, we present technical aspects of measurement calibration and validation. We then analyse the day-to-day variations of δ v and d in relation to climatic parameters and the air mass back trajectories. Finally, average hourly variations are explored and we interpret the typical daily cycles according to the main regional meteorological conditions in order to depict their driving factors and the influence of local processes. 2 Data acquisition

Protocols
Continuous in situ measurements of the isotopic composition of atmospheric water vapour (δ 18 O and δD) were performed during summer 2011 between 20 July and 24 August, at 1.75 m height and approximately 170 m from the eastern border of the Vaccarès lagoon ( Fig. 1), using Wavelength-Scanned Cavity Ring-Down Spectroscopy (WS-CRDS). The instrument we use is the Picarro L1102-i isotopic liquid water and water vapour analyser (Picarro Inc., Sunnyvale, California, USA), which measures the isotopic composition of atmospheric water vapour every 5 to 7 s. Installed in an airconditioned room, the analyser is connected to an outside air intake. As water vapour may be sticky on the walls of any tubing, a bypass configuration is used to bring the air in quickly, at a rate of 6 L min −1 (with a Laboport vacuum pump N86 KN.18) through PVC tubing, in order to minimise wall effects that lead to fractionation in the inlet. The analyser then subsamples this air at a rate lower than 0.04 L min −1 . Calibration of laser measurements is performed according to liquid laboratory standards, and a three-way valve allows switching between the introduction of ambient air and vapour from the liquid sample vaporizer (vaporization module V1102-i adjusted to 110 • C for flash vaporization to avoid fractionation). We used three laboratory standards whose isotopic compositions span a range of values including the com-   position expected in the atmosphere of the Camargue (Table 1). 1.8 µL of water standards is injected into the vaporizer using an autosampler (CTC Analytics LEAP Technologies HTC PAL autosampler) with a SGE 5 µL syringe. We used synthetic air as the dry carrier gas (water content < 45 ppmv) delivered at a pressure of 2.5 ± 0.5 psi. The analyser takes about 9 min to perform approximately 40 measurements per injection, and these results are then averaged by the software. Between each injection, the syringe is cleaned with 1methyl-2-pyrrolidinone (NMP) solvent wash. Our laboratory successfully participated in the IAEA 2011 proficiency test on routine analysis of δ 18 O and δD in liquid water (Wassenaar et al., 2012). Previous studies have shown that optical spectrometric methods can induce a dependence of isotopic measurements on water vapour concentration (Gupta et al., 2009;Schmidt et al., 2010;Johnson et al., 2011;Tremoy et al., 2011). This does not affect liquid sample measurements, since the water quantity introduced into the analyser through the vaporizer is nearly constant, but important variations of vapour content do occur when analysing ambient atmosphere, making it necessary to correct optical measurements. We have evaluated the concentration dependence of our instrument for a large range of water concentrations, and for different isotopic compositions using our three standards (Fig. 2). In addition, for routine analyses, an evaluation of the water concentration effect is performed at intervals of approximately 24 h with the more depleted standard (standard 1) whose isotopic composition is close to the atmosphere of the Camargue. The autosampler is set to perform six injections of 0.9, 1.6 and 2.0 µL in the vaporization module to obtain three water vapour concentrations ranging from about 8000 to 28 000 ppmv, comprising values expected in the study area. In addition, calibration for isotopic composition is performed at least every 24 h with our three standards. For each standard, the autosampler is set to perform six injections of 1.8 µL in the vaporization module, corresponding to a water vapour concentration of approximately 20 000 ppmv. The first two injections are disregarded to remove the memory effect, and the last four injections are averaged to obtain the δ 18 O and δD measurements of each standard. Gaps in time series correspond to calibration periods, but may also reflect accidental power cuts or data eliminated during episodes of condensation in the tubing. This latter problem was detected thanks to the use of transparent air tubing, and was easily recognizable a posteriori through the abnormally smooth shape of the water vapour concentration curve. A heating cable was then used to avoid any condensation such as may occur when a negative temperature gradient exists between the sampling point and the inlet of the analyser.

Dependence on water vapour concentration
It has been shown that the dependence of WS-CRDS isotopic measurements on water vapour concentration is specific to the individual analyser (e.g. Johnson et al., 2011;Tremoy et al., 2011;Aemisegger et al., 2012). The concentration dependence of our instrument, evaluated for a large range of concentrations ( Fig. 2), showed a linear response in the range of water vapour concentration measured in the atmosphere of the Camargue (from 9000 to 28 000 ppmv), with the relationship deviating slightly from the linear trend at very low and high water vapour concentrations. For δ 18 O of standards 2 and 3 there is no concentration effect, but a small dependence did appear for the more depleted standard (−0.33 ‰ 10 000 ppmv −1 ; R 2 = 0.75). For δD, we found significant regression coefficients (R 2 = 0.98, 0.99 and 1.00 for standards 1, 2 and 3 respectively), and the slope of the dependence was only slightly different between standards (between 7.8 and 8.1 ‰ 10 000 ppmv −1 ).
This justifies the use of only one standard (standard 1) to perform the very time-consuming daily calibration of the water concentration effect (160 min for a three-point cali-bration). Measurements are corrected for concentration dependence before applying the isotopic calibration in order to drive back atmospheric measurements to the water vapour concentration w reference at which standards are measured as follows (Schmidt et al., 2010): where δ reference (‰) is the reference δ value at w reference (20 000 ppmv in our case); δ observed (‰) is the observed δ value at w observed (ppmv) the observed concentration; m (‰ ppmv −1 ) is the slope of the concentration dependence ( Table 2).

Instrument stability
To evaluate the long-term stability of our instrument, statistics were calculated on raw delta values measured on our three standards during the entire field measurements (Table 1). Results show that the long-term variability is very low, but a regular calibration is nevertheless performed. Variability of both calibrations (for water vapour concentration and isotopic composition) is shown in Table 2. For δ 18 O, the calibration for concentration appears highly variable but often flat (no concentration effect), resulting in a linearity which is not always significant (0.26 ≤ R 2 ≤ 1.00). Thus, δ 18 O was corrected only when the concentration dependence was significant. For δD, the regression is slightly variable in the long-term, and its linearity is always significant (0.89 ≤ R 2 ≤ 0.99). For daily calibration in isotopic composition (normalization to the VSMOW2-SLAP2 scale), we used a three-point calibration, which allows to check for the linearity of the calibration. Linear regressions between raw measured values and absolute standard compositions are established as follows for δ 18 O and δD: δ calibrated = slope · δ measured + intercept. (2) Regressions are stable in the long-term (Table 2), and their linearity is always significant (R 2 = 1.00). Accuracy and precision of our instrument, estimated by considering standard 2 as a sample which has been calibrated with standards 1 and 3, show good results (Table 3). These performances are estimated on liquid measurements, and are probably lower for vapour measurements.

Local atmospheric data
Hourly air temperature and relative humidity were measured at the study site. Hourly wind speed and daily PET (Potential Evapotranspiration calculated from the Penman-Monteith formula (Monteith, 1965)) were obtained from a meteorological station located 400 m far from the site of isotopic measurements. In addition, surface temperature was also measured in the main pond of the Vaccarès lagoon system at hourly time step.
1.00 1.00 Table 3. Results of standard 2 measurements (δ std2−m ) normalized to the VSMOW2-SLAP2 scale by using standards 1 and 3 as working standards, following the IAEA Reference Sheet (IAEA, 2009). Statistics are calculated from the set of 40 calibrations. Standard deviation provides the reproducibility, and the root mean square deviation from the known value (RMSE) provides the accuracy of liquid measurements. Regional advection conditions were determined for each individual day by computing backward three-dimensional trajectories with the Internet-based NOAA HYSPLIT Trajectory Model (http://ready.arl.noaa.gov/HYSPLIT.php). The HYSPLIT (Hybrid Single-Particle Lagrangian Integrated Trajectory) model is a complete system, which allows computing simple air parcel trajectories using GDAS (Global Data Assimilation System) meteorological data. Trajectories were initialized at the station location every 24 h at 12:00 UTC, at three heights (50, 100 and 500 m) above ground level. A minimum back-run time of 72 h was computed, and extended, when necessary, in order to reach either the Mediterranean Sea, or the Atlantic ocean (depending of the wind regime, a maximum of 144 h was computed). The air mass trajectories were consistent between the three different initial altitudes, and hourly data were then classified, on a daily basis (i.e. from 00:00 to 23:00 UTC), according the three air mass origins represented in Fig. 1: Mediterranean, North Atlantic, or Bay of Biscay.
Planetary boundary layer heights were obtained from GDAS meteorological data (3 h time step). They were extracted and interpolated to our specific station using the Me-teoInfo software (www.meteothinker.com).
In addition, we collected and analysed three rainfall samples corresponding to four small rainfall events (25, 26, 27 July, 7 August 2011). These samples account for a total of 7.6 mm of the 9.6 mm of cumulated rainfall during the measurement campaign; the remaining 2 mm correspond to eight very small events between 14 and 23 August (not sampled).

Results and discussion
The mean composition of atmospheric water vapour during the campaign is δ 18 O = −14.66 ‰, δD = −95.4 and d = 21.9 ‰, with significant daily variations, from δ 18 O = −19.22 to −9.96 ‰ and δD = −125.2 to −61.7 ‰, while d varies between 7.8 and 31.2 ‰ (Fig. 3). These deuterium excess values are higher than those of our three rainfall samples: 5.6, 4.5 and 7.0 ‰, (for δ 18 O = −3.89, −3.7, and −0.66 ‰ respectively), and also higher than the longterm average composition of regional precipitation in the neighbouring GNIP (Global Network of Isotopes in Precipitation) station (Avignon, ≈ 50 km far from the experimental site; d = 9.2 ‰). The composition of vapour in equilibrium with rainfall is plotted for comparison (Fig. 3). Values slightly enriched in 18 O compared to measured δ v suggest a probable evaporation of rainfall in the atmosphere, a classical feature for small summer rainfall events (Celle-Jeanton et al., 2001), while values slightly depleted in deuterium compared to measured δ v points to the influence of an enriched vapour at ground level.
In measured atmospheric vapour composition, substantial day-to-day variations of δ v are observed, while d presents high hourly variations (Fig. 3). In order to explore the relevant time scales to interpret isotopic data in terms of synoptic and mesoscale meteorology vs. local influences, we first analyse the relations of our data to daily climatic variables and air mass origins. Then, we focus on average diurnal variations to identify the local influences. Table 4. Determination coefficients (R 2 ) and slopes (S) of linear correlations between daily means of water vapour composition (δ 18 O, δD, d) and air temperature (T a ), mixing ratio (q), relative humidity at air temperature (RH a ) and relative humidity at water surface temperature (RH s ).

Correlations between isotope data and local climatic parameters
Quite good correlations are found between daily values of δ v and local climatic parameters (Table 4), the best fit being with the mixing ratio q, (R 2 = 0.72 and 0.62 for δ 18 O and δD respectively). Such a correlation could result from the case of simple Rayleigh rainout processes, where condensation phases progressively remove water from the atmosphere and deplete the remaining atmospheric vapour in heavy iso-topes (Dansgaard, 1964;Lee et al., 2006;Wen et al., 2010). The progressive rainout follows a temperature decrease, and should also result in a correlation between δ v and T , and subsequently between rainfall isotope content (δ p ) and T , which has led to the establishment of the isotopic thermometer (Dansgaard, 1964;Jouzel et al., 1997). Our data show a relation between δ v and T (Table 4, Fig. 3) close to the isotopic thermometer: δ 18 O/ T = 0.53 ‰ • C −1 (Jouzel et al., 1997), but the correlation is relatively poor (R 2 = 0.30), indicating more complex processes, as was observed during a long-term German survey in which δ v -T correlation was degraded during summer because of the admixture of vapour originating from plant transpiration (Jacob and Sonntag, 1991). Since the correlation between δ v and q is much stronger than the δ v -T correlation, it indicates a mixing between air masses characterized by different vapour concentrations, as could result from different marine origins and/or from the addition of continental vapour into the atmosphere.
With respect to deuterium excess, we found that q is also the best predictor of d variations at a daily time step (R 2 = 0.54, Table 4). In addition, we observed a correlation between d and RH a (R 2 = 0.51), consistent with Welp et al. (2012), but the correlation becomes very low (R 2 = 0.19) when using relative humidity at surface temperature (RH s ). The significance of deuterium excess in terms of a proxy for the conditions at the vapour sources comes from the fact that it is determined by the kinetic isotopic fractionation that occurs during evaporation. The kinetic fractionation mainly depends on relative humidity at the water surface, which represents the vapour concentration gradient between the liquidvapour interface and free air (Craig and Gordon, 1965;Gonfiantini, 1986;Pfahl et Wernli, 2008;Uemura et al., 2008). Relative humidity is not conservative during air mass transport, as temperature changes modify the saturated vapour pressure, and the value measured at the meteorological station has to be reported at surface temperature.
Starting from a value around d = 0 ‰ in ocean water, the kinetic fractionation associated to the initial seawater evaporation produces an increase in d in oceanic vapour inversely related to humidity conditions (Merlivat and Jouzel, 1979;Armengaud et al., 1998;Pfahl et Wernli, 2008;Uemura et al., 2008;Jouzel et al., 2013;Steen-Larsen et al., 2013). . Phys., 15, 10167-10181  During the air mass trajectory over land, d may be further increased by the addition of vapour of continental origin (Gat and Matsui, 1991;Gat et al., 1994;Angert et al., 2008;Lai and Ehleringer, 2011;Welp et al., 2012). This "secondary" vapour is also expected to carry a d value inversely related to humidity conditions at the vapour source (Aemisseger et al., 2014). However, the lack of correlation between d and RH s in our data, whilst the d-q correlation is stronger, suggests, as was the case for the δ v -q correlation, that mixing processes between air masses characterized by different vapour origins and concentrations have weakened the d-RH s relation.

Atmos. Chem
Since it is hardly modified by equilibrium condensation, d can be considered as a conservative tracer of vapour sources during mixing processes. Nevertheless, the conservative behaviour of d during rainout is perturbed by the non-linearity in the definition of deuterium excess: the delta-notation approximation induces a small shift in d when there is a large decrease in δ 18 O and δD Angert et al., 2008;Welp et al., 2012). In the observed range of δ 18 O (from −11 to −18 ‰), we have estimated this effect to induce a 2.5 ‰ increase in d, while the substantial variations in d observed in our daily data are significantly higher (from 15 to 26 ‰), allowing the use of d as a tracer of different vapour pools.

Regional isotopic signatures
In order to examine the link between the isotopic characteristics of vapour and the air mass origin, we analyse the data classified according to the three main backward trajectories schematically represented in Fig. 1 (see Sect. 3). Corresponding hourly data are scattered in the δ 18 O-δD plot distinctly above the local meteoric water line (LMWL, Fig. 4), and show isotopically depleted compositions for air masses coming from the North Atlantic, while air masses coming from the Mediterranean Sea display isotopically enriched vapour (Table 5, Fig. 4). For air masses coming from Bay of Biscay, hourly data encompass the entire isotopic range. Local Table 5. Number of days (n) associated with the different origins of air masses (Fig. 1). Corresponding mean values of δ 18 O, δD, deuterium excess (d), air temperature (T a ), mixing ratio (q), relative humidity (RH), wind speed (V ) and potential evapotranspiration (PET) measured in the experimental site. climatic data associated with the three meteorological situations reflect synoptic weather conditions related to air mass origins. Air masses coming from the North Atlantic are associated with strong, cold and dry winds, while those from the Mediterranean are associated on the contrary with light, warm and wet winds ( Table 5). The Bay of Biscay is an intermediate situation. It thus appears that the greater the distance over land (Fig. 1), the more depleted δ v . The northern trajectory corresponds to a "Mistral" situation: a typical strong north-northwest wind that affects the north of the occidental Mediterranean basin 130 days a year on average. The air mass is gradually dried out by rainout processes over land and accelerated in the Rhône River valley and acquires a depleted isotopic signature. On the contrary, an air mass travelling over the Mediterranean Sea with the slight thermal wind or sea breeze coming from the south is moistened and acquires and maintains an enriched isotopic signature. With respect to d, North Atlantic air masses bring higher values (23.2 ‰) than Mediterranean air masses (18.6 ‰), but the main feature, which appears on the δ 18 O-δD plot (Fig. 4), is the distinct separation between nocturnal and diurnal hourly data, with the latter plotting higher above the meteoric lines. This feature is observed for each of the three air mass origins.
Within these three classes of data, strong linear relationships are observed (Fig. 4), with slopes lower than eight, a value attributed to thermodynamic equilibrium, as it approximately represents the average ratio between D and 18 O liquid-vapour equilibrium fractionation. Such low slopes have also been observed in atmospheric vapour from Hawaii (7.02 ), Greenland (between 6.47 and 7.44 (Steen-Larsen et al., 2013)), western Siberia (between 5.6 and 7.7 (Bastrikov et al., 2014)), and in North America (7.5 ). The previously described shift in d during Rayleigh-type rainout, due to the nonlinearity in the deuterium excess definition, is not sufficient to explain these low slopes, and non-equilibrium processes such as evaporation are also probably involved (Gibson et al., 2008).

Analysis of average daily cycles
In order to observe the diurnal variations suggested by the night-day separation, average daily cycles of isotopic and climatic data are plotted for the three air mass origins (Fig. 5). A well-pronounced cyclicity appears for d in each meteorological condition. For the Bay of Biscay origin, the average d cycle is very similar to the North Atlantic case, with larger standard deviations, reflecting a more variable climatic situation. For Mediterranean conditions, the d cycle is higher in amplitude than for North Atlantic conditions, and lower in absolute values, especially during the night. The high standard deviations observed during the last hours of the day correspond to calibration periods for which there are fewer data, but may also be due to variations in the timing of the PBL evening transition. The increase in d during the morning is mainly associated with a δD increase, while δ 18 O variations are smoother. Amplitudes of daily isotopic variations are higher for Mediterranean conditions (9.7, 16.2 and 1.3 ‰ for d, δD and δ 18 O respectively) than for North Atlantic ones (7.0, 13.1 and 1.1 ‰ for d, δD and δ 18 O respectively). From a climatic point of view, we observe high diurnal variations in RH a , mainly driven by the large amplitude of air temperature variations between night and day (Fig. 5). As discussed above, the significance of relative humidity conditions in terms of evaporation conditions -and deuterium excess control -is only relevant at surface temperature. The variations are smoother for RH s , and it more or less follows the absolute amount of water (Fig. 5). Under Mediterranean conditions, a distinct increase of q occurs in the morning, indicating a net addition of vapour in the atmosphere. Under North Atlantic conditions, the q increase is lower, but still detectable.
The daily cyclicity of the planetary boundary layer height (PBLH) is specifically related to the different atmospheric conditions. The PBLH attains around 1100 m on average during the afternoon under North Atlantic and Bay of Biscay conditions, with a significant standard deviation. Under Mediterranean conditions, the PBLH is less variable and remains at almost 700 m. Nocturnal values are also very low -less than 100 m -in Mediterranean conditions, compared to 400-500 m in the other weather situations (Fig. 5). Low PBLH in Mediterranean conditions corresponds to weak turbulence and air stability resulting from light winds, while northerly advection transports a relatively cold air mass over a warmer surface, leading to a weaker stratification.

Daytime processes: contribution of local evaporation
The diurnal increase in d is a widely observed feature, which reflects the diurnal variation of the water and air mass balances of the planetary boundary layer (Lai and Ehleringer, 2011). This increase can be caused by the entrainment of free atmosphere into the boundary layer Welp et al., 2012;Berkelhammer et al., 2013), but also by the addition of locally evaporated vapour. The direction of vapour concentration changes associated with these d variations can help to identify which of these two processes dominates the PBL water mass budget. Entrainment brings a dryer air into the PBL, and thus causes a daytime decrease in vapour concentration, while an increase in mixing ratio indicates the addition of an evaporation flux. Note that the transpiration component of evapotranspiration is not expected to carry high d, as it has the same isotopic composition as the  (Yakir and Wang, 1996;Williams et al., 2004). In addition, the free atmospheric air is characterized by a isotopically lighter vapour (He and Smith, 1999;Bailey et al., 2013;Berkelhammer et al., 2013), and the dominance of entrainment during daytime is generally associated with a δ v depletion (Lai et al., 2006;Lai and Ehleringer, 2011;Tremoy et al., 2012;Bailey et al., 2013;Berkelhammer et al., 2013). Welp et al. (2012) compared six measurement locations and found that both local evapotranspiration and entrainment were involved: in some of their sites, the d daily increase was associated with a δ 18 O decrease during the early morning due to entrainment of free atmosphere into the boundary layer during convective mixing, while the subsequent slight δ 18 O increase came from evapotranspiration. Our data display a simultaneous increase in q, δ 18 O and δD (Fig. 5), and thus indicate that the addition of surface vapour is the dominant process governing the d cyclicity during the day.

Nocturnal processes: dew formation
In each meteorological situation, d displays an abrupt decrease at around 20:00 UTC, which corresponds to the timing of sunset. Afterwards, during the 20:00-06:00 UTC period, d remains low, while δ v follows a gradual depletion associated with a q decrease (Fig. 5). This corresponds to the effect of a progressive Rayleigh-condensation process, and indicates the formation of dew induced by the nocturnal cooling of soil and/or leaves surfaces, as was observed by . A small nocturnal evaporation of the lagoon water could nevertheless occur since the water surface temperature remains high and maintains RH s below saturation (Fig. 5), but it is expected to remain low compared to daytime fluxes. During the night, the PBL subsidence corresponds to the reduction of both the surface heat fluxes and entrainment, and the addition of water vapour to the atmosphere, if any, remains low. The large-scale motion becomes dominant and, without changing the total PBL mass, redistributes it horizontally (Medeiros et al., 2005). The nocturnal isotopic values thus represent the average composition of the PBL, onto which are superimposed daytime fluxes. More precisely, since dew formation is a local process, the value measured just after the sunset (20:00 UTC), when the PBL collapses, is probably the most representative composition of the average PBL, with the greater proportion of regional water vapour. The high d values observed during the night (on average 15.1 ± 0.5 ‰ under Mediterranean conditions and 20.9 ± 0.7 ‰ under North Atlantic conditions), thus result from the remaining imprint of local evaporation, combined with the signature of the regional atmosphere.

Local isotopic signatures
The concomitant increase of q and d during daytime in our average diurnal cycles indicates that the d cyclicity is mainly driven by the addition of surface vapour fluxes. As stated above, the location of the experimental site close to the main lagoon, and the availability of large open water areas in the surroundings point to a dominant contribution of evaporation over transpiration. In order to calculate the isotopic composition of the locally evaporated vapour (δ E ), we use a twocomponent mixing model based on the linear correlation between 1/q and δ v , derived from the "Keeling plot" method. The ability of this simple model to describe our data and the possible contribution of regional advection is then discussed. The original Keeling approach (Keeling, 1958) was initially proposed to describe the addition of CO 2 into the atmosphere, and then further used to determine the isotopic composition of evapotranspired vapour flux (Yakir and Sternberg, 2000;Yepez et al., 2003;Williams et al., 2004;Zhang et al., 2010;Noone et al., 2011Noone et al., , 2013Griffis et al., 2013). Assuming a background represented by an initial air quantity with a mixing ratio q BG (mmol mol −1 ), and taking n A (mol) the corresponding dry air amount, the addition of water vapour n E (mmol) leads to the following mixing equation: where q is the mixing ratio (mmol mol −1 ) of the resulting humid air (note that n E /n A does not represent a mixing ratio despite same units). Attributing an isotopic composition to each of these vapour pools, δ BG for the background vapour and δ E for the added vapour, the corresponding isotopic mass balance is where δ v is the vapour composition of the resulting atmosphere. In a data set tracing the progressive addition of pure vapour in a closed system, and assuming an isotopically stable background, the δ v vs. 1/q relationship displays a linear trend. The composition of the added vapour δ E is therefore deduced from the intercept (for q →∝). However, when the increase in water vapour concentration results from the mixing with a humid air mass, Eq. (3) becomes where x is the fraction of the initial background air, and q E is the mixing ratio of the added air mass (mmol mol −1 ). Thus, the corresponding expression for δ v is  The composition of the added vapour is then deduced from the δ v vs. 1/q plot for x = 0, i.e. for q = q E , and the mixing ratio of the added air (q E ) is a prerequisite for applying this two-component mixing model. For a humid air mass coming from open water evaporation, a mixing ratio corresponding to saturation conditions at the liquid-vapour interface (q E = q s ) can be considered as representing the limit condition (Craig and Gordon, 1965;Noone et al., 2011;Noone, 2012).
Our discussion will focus on the two most contrasted meteorological situations, i.e. Mediterranean and North Atlantic air mass origins. Strong correlations are found between average values of δ v and 1/q for the period of q increase, i.e. 05:00-13:00 UTC for North Atlantic situations and 05:00-10:00 UTC for Mediterranean situations (R 2 values are between 0.85 and 0.96, see Fig. 6). The local endmembers are calculated for the average value of saturation mixing ratio at the water surface temperature (q s ), during the daily maximum air water concentrations (at 13:00 or 10:00 UTC). Uncertainties are determined as resulting from error propagation of the linear model in the whole range of q s standard deviation. The resulting compositions of local endmembers (δ E ) are δ 18 O = −12.5 ‰ (−11.0 to −13.9 ‰); δD = −48 ‰ (−30 to −64 ‰) for North Atlantic conditions, δ 18 O = −11.8 ‰ (−11.0 to −12.5 ‰); δD = −58 ‰ (−46 to −68 ‰) for Mediterranean conditions (Fig. 7). High deuterium excess values characterize the surface vapour for North Atlantic (d = 52 ± 29 ‰) and Mediterranean conditions (37±17 ‰), which is consistent with previous estimates from the Craig and Gordon model (Craig and Gordon, 1965;Gat et al., 1994). Such high d is a characteristic of vapour coming from continental evaporation and allowed to detect the contribution of continental vapour to regional precipita- tion (Gat et Matsui, 1991;Gat et al., 1994;Vallet-Coulomb et al., 2008). In our case, these high d values confirm that evaporation is largely dominant over transpiration, and that the higher d under North Atlantic air mass conditions is consistent with lower RH s , compared to Mediterranean conditions (Fig. 5, Table 5).
Going on with the two end-member mixing assumption, this δ E composition could be used to interpret the amplitude of diurnal δ v variations in terms of the daily addition of locally evaporated vapour to the nocturnal background. The lower amplitude of diurnal variations observed under North Atlantic condition (Fig. 7) would then indicate a lower proportion of locally evaporated vapour in ambient moisture. However, this is not consistent with the stronger average potential evapotranspiration rates under North Atlantic conditions, compared to the Mediterranean situation: 5.8 vs. 3.8 mm day −1 (Table 5). This apparent contradiction indicates that the isotopic imprint of local evaporation is diluted by advection, with a greater extend under Northern conditions than under Mediterranean conditions. This is fully consistent with the high wind speed characterizing North Atlantic conditions: an average value of 2.0 m s −1 compared to 0.7 m s −1 for Mediterranean conditions. Local vapour is thus flushed by advection of dry northward air masses, and since this advected vapour may carry a different isotopic signature than the nocturnal background, a simple two-component mixing model is not sufficient to describe the PBL water balance.
We have evaluated the impact of regional advection on the determination of δ E , by testing different simulations of a three-component mixing process. Starting from the initial background (δ BG = average δ v value at 05:00 UTC), we simulate the progressive addition of both regional advected vapour (δ R ) and locally evaporated vapour (δ E ), and we compare the resulting composition with observed δ v during the q-increase period of the day. For each simulation, we choose a hypothetical value of δ R , and we adjust δ E in order to match the observed data on the δ v vs. 1/q relationship, using a simple trial-and-error procedure, successively for δ 18 O and δD. We have tested a range of δ R values compatible with the shift Atmos. Chem. Phys., 15, 10167-10181, 2015 www.atmos-chem-phys.net/15/10167/2015/ observed during sunset, on the δ 18 O-δD plot (Fig. 7). As discussed previously, the δ v value measured at 20:00 UTC is expected to carry the lowest proportion of locally evaporated vapour. We thus postulate that the shift in d observed at 20:00 ("B" arrow in Fig. 7) is "dragged" by the composition of a more regional vapour. We have then explored a domain of δ R values limited by (1) the 20:00 UTC value as an upper limit for δD; (2) the 20:00 UTC (05:00 UTC) value as a lower limit for δ 18 O under Mediterranean (North Atlantic) conditions and (3) the LMWL. For each δ R value, different mixing ratio were tested (q R ), chosen to remain compatible with the midday-20:00 UTC trend observed on the 1/q − δ v plot. We found that in all cases, the δ E values remain in the range of the error bars shown in Fig. 7, with slightly lower δ 18 O and higher d.
The processes involved in the diurnal isotopic behaviour of atmospheric vapour can be summarized in the δ 18 O-δD plot, which is a good mean to understand the drivers of d variations (Fig. 7). The two clusters of average vapour data are distributed along two linear trends defined by the q-increase period of the day ("A" arrow in Fig. 7), which also meet the calculated value of δ E . These trends result from the addition of locally evaporated vapour into the ambient air, but, as discussed above, are also influenced by the input of (and flushing by) regional advection. The daytime vapour composition then oscillates, but stays around the maximum δ v value, until the abrupt shift that occurs at 20:00 UTC, during sunset, consistently for both meteorological situations ("B" arrow in Fig. 7). The d decrease observed when the PBL collapses, during sunset, corresponds to the weakening of surface evaporation, and to an increasing proportion of a more regional atmosphere. The night-time δ v evolves closer to the meteoric water line, compared to daytime compositions. The progressive condensation occurring between 20:00 and 05:00 UTC maintains the δ v along a regression line roughly parallel to the LMWL ("C" arrow in Fig. 7).
The stronger advection rate which prevails under North Atlantic conditions, and which smoothed the diurnal amplitude of δ v variation, suggests that the higher d signature of water vapour (Table 5) results from the combined influences of higher d in the locally evaporated vapour and in the regional vapour, consistently with the longer continental trajectory. A more precise determination of the regional vapour isotopic composition would rely on the quantification of advection fluxes and on a full PBL mass balance estimate.

Conclusion
Our results show that both local and regional isotopic signatures are recorded in the ground level atmospheric vapour composition, and that a sub-daily analysis is necessary to decipher these influences. After having classified the data according to air mass origins and trajectories, we found strong diurnal variations of d, with lower amplitude varia-tions under the influence of northern air masses than under Mediterranean influences. The higher average d observed under northern conditions (23.2 ‰ compared to 18.6 ‰ for the Mediterranean situations) is mainly due to higher nocturnal values, while the daytime maxima are similar. However, the drivers of these diurnal variations have to be understood to interpret the isotopic signal with respect to its relevant spatiotemporal scales, and it is essential to understand and compare the behaviour of both δ 18 O and δD (referred to as δ v ) to understand d.
A diurnal increase in d can be caused by entrainment of free atmospheric air, by surface evaporation, or by a combination of both processes. Unlike evaporation, the contribution of free atmospheric air would induce a decrease in absolute vapour concentration. Therefore, our average daily cycles clearly indicate that the diurnal increase in d is essentially driven by local evaporation. Based on the robust alignment of average hourly data in a δ v vs. 1/q plot for the q-increase period of the day, we applied a two-component mixing model to estimate the composition of locally evaporated vapour (δ E ). Then, since regional advection modulates the amplitude of diurnal δ v variations, we discussed the twocomponent mixing assumption and evaluate the influence of a third component on the determination of δ E . We show that it remains low, because the composition of regional advection is expected to be close to the nocturnal values. We found higher d for the locally evaporated vapour under North Atlantic air mass conditions, which is consistent with lower humidity conditions, while the corresponding δ 18 O compositions are very similar between the two meteorological situations, with a difference of about 0.7 ‰.
In contrast, there is an important difference in the δ 18 O compositions between the two meteorological situations when considering the nocturnal values. More specifically, it is interesting to consider the ambient vapour composition measured at 20:00 UTC when the PBL collapses, and before the progressive influence of dew formation during the night. This 20:00 UTC composition is the least influenced by locally evaporated vapour, and the most representative of a regional signal. The difference in δ 18 O between the two main meteorological situations (upper than 3 ‰) reflects the influence of synoptic and mesoscale weather conditions, related to air mass origin and trajectory. Strong, cold and dry winds coming from the north bring an isotopically depleted vapour. Conversely, light, warm and wet winds coming from the south bring an isotopically enriched vapour. In addition, the strong advection rate that prevails under North Atlantic conditions suggests that, in this situation, the high average d of water vapour at 20:00 UTC also reflects the signature of regional vapour.
In general, our data showed that the mixing ratio is a better predictor of deuterium excess variations than relative humidity, at both daily and sub-daily time scales, since mixing processes between different vapour sources weakened the relevance of relative humidity as an indicator of evaporation conditions. At the daily time step, the mixing ratio is a proxy for the air mass origin, since there is a huge contrast between water content of northern and Mediterranean air masses. At the sub-daily time step, a δ v -q correlation, and subsequently a d-q correlation, during the q-increase period of the day, results from the addition of locally evaporated vapour. The general framework proposed in this study, for typical daily cycles of water vapour isotopic composition under different meteorological conditions, could be used in further quantitative analyses of vapour sources during specific days.