To evaluate the accuracy of the measurement and calibrate the humidity measured by both analysers, we compare it to an independent in-situ measurement of the atmospheric humidity between January and March 2024. Note that data from this independent measurement was not available in December 2023, so the comparison is restricted to the beginning of 2024 although the analysers were operating in December 2023.

The independent humidity sensor is installed about two meters above the surface and about twenty meters away from the inlet of the laser spectrometers. The sensor is an adapted HMP155 sensor, specifically designed to accurately measure the atmospheric humidity in dry and cold environments with frequent supersaturation conditions (Genthon et al., 2017, 2022). As in Genthon et al. (2017), Vignon et al. (2022) and Ollivier et al. (2025), we use the data from the adapted HMP155 to recalculate the relative humidity with respect to ice. The relative humidity with respect to ice is then converted to water vapour mixing ratio (in ppmv) using the equations from Murphy and Koop (2005) together with the air pressure given by ERA5. Note that the resulting water vapour mixing ratio is not sensitive to the possible mismatch between the pressure given by ERA5 and the local atmospheric pressure (not shown). We use this independent humidity measurement as the true atmospheric humidity content to correct the humidity measured by the Picarro and AP2E analysers, as follows: 1humcorr=hummeas⋅slopehum+inthum Where hum_meas is the raw humidity given by the analyser (either AP2E or Picarro), hum_corr is the humidity corrected on the independent measurement and the coefficients slope_hum and int_hum are determined by a linear regression between the hum_meas and the independent humidity measurement. The results of the linear regressions are presented in Sect. 3.1.1.

For continuous water vapour isotopic measurement, and in particular in the East Antarctic plateau where mixing ratios are often below 500 ppmv, both OF-CEAS and CRDS techniques are affected by the dependency of isotopic measurements on the water vapour mixing ratio (e.g. Lauwers et al., 2025; Weng et al., 2020). We refer to this effect as the humidity-isotope response. This humidity-isotope response is instrument-specific (e.g. Steen-Larsen et al., 2013) and is dependent on the isotopic composition of the laboratory standard used to perform the calibrations (e.g. Lauwers et al., 2025; Weng et al., 2020). A calibration of this dependency is therefore required in the humidity range of the site and using laboratory standards with a known isotopic composition close to what is observed on site.

We determined the humidity-isotope response curves by performing one series of nine calibrations in January 2024. The calibration curves for both analysers are determined using a single custom laboratory standard (FP5, δ18O = -50.52 ± 0.05 ‰ and δD = -394.7 ± 0.7 ‰), calibrated against the VSMOW-SLAP scale. We assume that the humidity-isotope response of both analysers (AP2E and Picarro) is stable in the range of isotopic values measured on site, which was validated for a Picarro analyser in Leroy-Dos Santos et al. (2021). The standard FP5 has an isotopic composition close to the atmospheric water vapour isotopic composition measured on site (varying between approximately -50 ‰ and -80 ‰ in δ18O and between approximately -400 ‰ and -550 ‰ in δD during summertime, Leroy-Dos Santos et al., 2021) and it has been previously used to calibrate a Picarro laser spectrometer at the same site (Leroy-Dos Santos et al., 2021). The calibration steps were performed from high to low humidity (humidities ranging from 1100 to 50 ppmv). Each calibration lasts approximately two hours, and the data point correspond to the average of the last 10 min, in order to minimize the memory effect. We assume that at 200 ppmv, the memory effect is negligible compared to the measurement uncertainty. The humidity levels are generated using the newest version of the custom calibration unit (LHLG, Leroy-Dos Santos et al., 2021), which enables the generation of a steady water vapour flux with a known and stable isotopic composition.

The reference humidity for the calibration curves is set to 500 ppmv (see also Sect. 2.2.3). The results of the different calibration steps are fitted with inverse functions (in combination to a linear function), as done in previous studies (e.g. Lauwers et al., 2025). The coefficients of the inverse fits are used to correct the raw isotope data for the humidity-isotope response, as follows: 2δi,humcorr=δi,meas-1c1⋅hummeas+c2⋅hummeas+c3 Where δi,meas is the raw isotope data given by the instruments (subscript i is for any isotope species, δ18O or δD), δi,humcorr is the isotope data corrected for the humidity-isotope response of the instruments and the coefficients c1, c2, and c3 correspond to the coefficients of the inverse functions fitted to the data of the calibration steps. Equation (2) is determined for each isotope species and each analyser. The results of the calibration steps, the inverse fits and the coefficients are presented in Sect. 3.1.2.

In a second step, we perform the absolute calibration of both analysers to convert the raw isotopic compositions measured by the instruments (and corrected for humidity dependence beforehand) to isotopic values calibrated against the VSMOW-SLAP scale. Regular and automatic calibrations of both analysers are performed with two laboratory standards calibrated against VSMOW-SLAP (FP5: δ18O = -50.52 ‰ and δD = -394.7 ‰; NEEM: δ18O = -33.5 ‰ and δD = -257.2 ‰). Note that the very depleted standard VSAEL was not available for calibrations in the field when our instruments were deployed. The calibrations are performed every 48 to 72 h with the LHLG, injecting both standards at a target humidity level of 500 ppmv. We use the isotopic ratios measured by both analysers during the calibrations between 11 January and 6 June 2024 to establish the linear equations for the absolute calibration of each instrument. To remove the influence of the humidity measured during each calibration on the measured isotopic ratios during the calibration step, we correct the isotopic ratios for the humidity-isotope dependence (Sect. 2.2.2). In addition, we discard the calibrations with a humidity outside of two standard deviations around the mean humidity and outside of two standard deviations around the mean isotopic ratio of all calibrations during the period. Because we do not observe any significant drift in the calibration data, we then average, for each laboratory standard and each analyser, the measured water isotopic composition of all the selected calibrations over the period and establish the linear equations against the true value of the standards. The linear functions for each analyser are used to calibrate the measurements against the VSMOW-SLAP scale, as follows: 3δi,VSMOW-SLAP=δi,humcorr⋅slopeVSMOW-SLAP+intVSMOW-SLAP Where δi,humcorr is the isotope data corrected for humidity-isotope response (subscript i is for each isotope species, δ18O or δD, see Sect. 2.2.2 and Eq. 2), δi,VSMOW-SLAP is the final isotope data corrected and calibrated against VSMOW-SLAP and the coefficients slope_VSMOW-SLAP and int_VSMOW-SLAP are determined by the linear regression between the measured and true values of the two laboratory standards. The results of the absolute calibration step are presented in Sect. 3.1.3. The corrected and calibrated time series of the water vapor isotopic composition from both analysers are presented in Sect. 3.2.

It should be noted that the two laboratory standards used to perform the absolute calibration both have an isotopic composition above the one usually measured on site in the atmosphere. We therefore assume that the linear relationships between the true and measured δ18O and δD values can be extrapolated beyond the isotopic composition of both standards to be able to calibrate the in-situ measurements. Such assumption was validated for a Picarro analyser in Casado et al. (2016).

We present two approaches to estimate the uncertainty on the water vapour isotopic measurements. First, we propagate the uncertainty related to the measurement noise driven by low humidity measurements and day-to-day instrumental drift, which is manifested in the regular measurements of the two laboratory standards. Secondly, we carry out a Monte-Carlo simulation propagating the uncertainty of the absolute calibration against VSMOW-SLAP into the uncertainty estimate on the final calibrated water vapour isotope measurements.

We consider two sources of uncertainty associated with the δ18O and δD measurements. The first source of uncertainty follows a power law with respect to humidity due to the increase in measurement noise at lower humidity levels for both Picarro and AP2E analysers (Lauwers et al., 2025). The second uncertainty originates from the instrumental instabilities at hourly to daily time scales caused by the sensitivity of the optical signal of laser spectrometers to several environmental factors, such as temperature or mechanical perturbations. We refer to the latter uncertainty as the “drift” uncertainty. We group the two uncertainties (noise at low humidity and drift) into the following formulation to estimate the combined uncertainty on δ18O and δD measurements: 4σih=σi,drift⋅href/h With href is the reference humidity of the calibration steps (href = 500 ppmv, Sect. 2.2.2 and 2.2.3) and h is the humidity measured by the laser spectrometers. σi,drift corresponds to one standard deviation of the measured isotopic ratios (subscript i is for any isotope species, δ18O or δD) of all the calibration steps performed over six months with two laboratory standards (selected calibrations steps, see Sect. 2.2.3).

The uncertainty is calculated for the whole dataset for both analysers and is valid from 50 to 1100 ppmv (i.e. corresponding to the upper and lower limit of the humidity-response curves, see Sect. 2.2.2). With this method, the uncertainty on the data incorporates both the instrumental drift over six months, similarly as done by Casado et al. (2016), and the dependency of the uncertainty on the measured humidity (i.e. larger uncertainties at lower humidities). This measurement uncertainty is probably overestimated, as σi,drift integrates both the drift from the LHLG and from each isotope analyser over a six-month period. The uncertainty σ(h) for d-excess is calculated by propagating the uncertainties on δ18O and δD, as follows: 5σd-excessh=σδDh2+8×σδ18O(h)2 Alternatively, we propose to compute the uncertainty on the final δ18O and δD values from the Picarro and AP2E analysers by performing a Monte Carlo test with 1000 resamples of the linear regression coefficients within their uncertainty range to calibrate the δ18O and δD values against the VSMOW-SLAP scale (as described in Sect. 2.2.3 but including uncertainty on the linear equation coefficients in Eq. 3). The uncertainty (referred to as σMC) is computed as one standard deviation of the 1000 Monte Carlo calibrated time series and should be similar to σdrift, since the same dataset of calibration steps is used for both methods. We compute the uncertainty for d-excess by propagating the uncertainties on δ18O and δD, using Eq. (5). Results of this analysis are presented in Sect. 3.1.4.

Figure 2 shows the evaluation of the atmospheric mixing ratio (or humidity, in ppmv) measured by the two analysers (Picarro and AP2E) against an independent humidity sensor (Sect. 2.2.1). The humidity measured by both analysers agree very well with the independent humidity measurement, with linear regression slopes close to the one-to-one line for both analysers (Fig. 2a and b). Overall, the Picarro analyser measures a lower humidity content than the independent sensor (average difference of 20 ppmv between 1 January and 15 March 2024), especially at higher humidity levels (Fig. 2a). On the other hand, the AP2E analyser gives similar humidities than the independent sensor (average difference of 2 ppmv between 1 January and 15 March 2024) in the whole range of humidities (Fig. 2b). The humidity measured by both analysers also compare very well together, with an overall positive bias of the AP2E compared to the Picarro (Fig. 2c).

Even if the difference between the humidity measured by the Picarro and AP2E analysers and the independent humidity sensor is small, the linear regression coefficients slope_hum and intercepts at origin int_hum (Fig. 2, Table 1) can be used to calibrate the humidity measured by both analysers, as described in Sect. 2.2.1.

During the period of interest (December 2023 to 15 March 2024), the humidity measured and calibrated by the two laser spectrometers ranges from 15 to 1100 ppmv (see also Fig. 6, Sect. 3.2). Note that the lowest humidity measured by the modified HMP155 system during this period is about 1 ppmv, however the two laser spectrometers didn't record this low humidity due to gaps in the dataset (Sect. 3.2).

Figure 3 shows the humidity-isotope calibration curves determined with the laboratory standard FP5 (δ18O = -50.52 ‰ and δD = -394.7 ‰, Sect. 2.2.2), for three laser spectrometers (described in Sect. 2.1): (1) the Picarro HIDS2319 analyser from Leroy-Dos Santos et al. (2021), (2) the Picarro HIDS2308 analyser (this study) and (3) the AP2E analyser (this study). For the Picarro HIDS2319 analyser, the calibration steps were performed with the initial version of the LHLG while for this study (Picarro HIDS2308 and AP2E analysers), the calibration steps were performed with the newest version of the LHLG (Sect. 2.2.2). Each point on the humidity-isotope response curves of all three analysers corresponds to the average isotopic composition of the calibration step over a ten-minute stable period. Note that each calibration step lasted from 40 min to 1 h.

In Leroy-Dos Santos et al. (2021), the humidity-isotope response curves (for both δ18O and δD) of the Picarro HIDS2319 are described with polynomial fits (their equations 4 and 5, light green dashed lines in Fig. 3a and b). Their results show a divergence of the measured isotopic composition below 500 ppmv, especially strong for δD (light green dashed line and dots in Fig. 3b). For the Picarro analyser HIDS2308, the humidity-isotope response curves are described with inverse fits (Sect. 2.2.2, dark green dotted lines in Fig. 3a and b). In comparison to the HIDS2319 analyser, the response curves show a similar strong divergence in δ18O and a much weaker divergence in δD. In addition, the HIDS2308 curves don't show any humidity-isotope dependence above 500 ppmv for both δ18O and δD (dark green dotted lines and dots in Fig. 3a and b). The difference in humidity-isotope response of the two Picarro analysers (HIDS2319 and HIDS2308) is not surprising since different spectrometers will have a different humidity-isotope response (e.g. Steen-Larsen et al., 2013).

For the AP2E analyser, the humidity-isotope response curves are also described with inverse fits (Sect. 2.2.2, blue dotted lines in Fig. 3a and b). As already identified and described in Lauwers et al. (2025), the AP2E analyser humidity-isotope response curves show two different regimes. Below 500 ppmv, both δ18O and δD diverge as humidity levels decrease, but in the opposite direction observed in both Picarro analysers (blue dotted lines in Fig. 3a and b). Above 500 ppmv, δ18O shows a positive linear dependency to increasing humidity (blue dotted line in Fig. 3a), while a weaker dependency is observed for δD (blue dotted line in Fig. 3b).

These results show that the isotope-humidity response of the Picarro analyser presented in this study is better constrained compared to the previous Picarro analyser, with a calibration curve determined down to a lower humidity than in Leroy-Dos Santos et al. (2021; 50 ppmv in this study, 110 ppmv previously). In addition, the new Picarro shows a weaker isotope-humidity dependence in the range of observed humidities at Dome C over the period of interest (15 to 1100 ppmv, Sect. 3.1.1), which leads to a better constrain on the correction for the isotope-humidity response and improves the reliability of the dataset. These results also show a well constrained isotope-humidity dependence for the AP2E analyser in the range of observed humidities at Dome C over the period of interest, which similarly to the Picarro analyser, improves the reliability of the dataset.

It should still be noted that the isotope-humidity calibration only goes down to 50 ppmv, although the minimum humidity recorded by the instruments is 15 ppmv during the period of interest (and the overall minimum humidity recorded by the HMP155 is 1 ppmv, Sect. 3.1.1). To correct the dataset, we therefore extrapolate the calibration curve down to 15 ppmv. This can lead to abnormal isotopic values after correction, leading to the increase of the uncertainty on the data at low humidities. This point is further developed in Sect. 3.1.4.

Table 2 summarises the coefficients of the inverse fits shown in Fig. 3 for both the Picarro HIDS2308 and AP2E analysers. As described in Sect. 2.2.2, these coefficients are used in Eq. (2) to calibrate the isotope measurements from both analysers for the humidity-isotope dependence (following Eq. 2, positive values in Fig. 3 correspond to a negative correction).

As described in Sect. 2.2.3, the absolute calibration against the VSMOW-SLAP scale of the isotope data given by the Picarro and the AP2E analysers relies on the results of regular calibrations over six months of two laboratory standards with known isotopic composition. Figure 4 shows the results of these regular calibrations performed between January and June 2024.

We first see that, despite a target humidity of 500 ppmv, the humidity measured during these regular calibrations varies slightly, from 250 to 450 ppmv, depending on which instrument and standard is measured (Fig. 4a). We also see that some of the calibrations are associated with very low humidities (red markers in Fig. 4a), which we exclude in the pool of calibrations used for the absolute calibration of both analysers (Sect. 2.2.3). These low humidity calibrations can be explained by the LHLG, which failed to generate the target humidity level.

We observe that the measured δ18O by both analysers varies throughout the period, but no drift is observed (Fig. 4b). Since the δ18O values shown in Fig. 4b are corrected for the humidity-isotope response (Sect. 2.2.3), variations around the mean δ18O over the whole period cannot be explained by the variations of the humidity measured by the analysers (Fig. 4a). Instead, these variations can be explained by variations of environmental conditions, such as the temperature in the room where the spectrometers are installed, or instability of the humidity generated by the LHLG during the calibration step. Despite these variations, the standard deviation of the ensemble of δ18O values associated to the calibration of the two laboratory standards is low for both instruments (1.0 ‰ for the standard NEEM measured by the AP2E analyser and 0.8 ‰ for FP5; 0.8 ‰ for the standard NEEM measured by the Picarro analyser and 0.6 ‰ for FP5; Fig. 4b) compared to results from Lauwers et al. (2025) obtained at Dumont d'Urville station over a year. We further exclude the few calibrations which appear as outliers (outside of two standard deviations around the mean δ18O, red markers in Fig. 4b) to establish the absolute calibration of both analysers (Sect. 2.2.3).

As for δ18O, we do not observe any drift in δD over the period, for neither analyser (Fig. 4c). The standard deviation of the ensemble of δD values associated with the calibration of the two laboratory standards is low for both instruments (7.4 ‰ for the standard NEEM measured by the AP2E analyser and 6.5 ‰ for FP5; 6.9 ‰ for the NEEM standard measured by the Picarro analyser and 2.4 ‰ for FP5; Fig. 4c). These results are comparable with the results from Lauwers et al. (2025). We observe that for both laboratory standards, the variations in δD over the period are higher for the AP2E analyser than Picarro. One reason that could explain this difference is that the OF-CEAS technique used in AP2E spectrometers is particularly sensitive to noise associated with optical absorption, compared to the CRDS technique used in Picarro spectrometers (Lauwers et al., 2025). This effect is more visible when the absorption peak is very close to the baseline: for example at low humidity, or when looking at the deuterium absorption peak which shows an amplitude one order of magnitude smaller than the ¹⁸O peak. We further exclude the few calibrations which appear as outliers (outside of two standard deviations around the mean δD, red markers in Fig. 4c) to establish the absolute calibration of both analysers (Sect. 2.2.3).

As described in Sect. 2.2.3, the results of the regular calibrations over six months are used to calibrate the data against the VSMOW-SLAP scale (selected calibrations from Fig. 4). Figure 5 shows the result of the linear regressions between the true and humidity-corrected δ18O and δD. The coefficients of the linear regressions (used in Eq. 3) for both analysers and both isotope species are summarised in Table 3.

Both the Picarro and AP2E analysers have an absolute calibration slope for δ18O close to one (respectively 0.98 and 0.97, Fig. 5a and Table 3). The intercepts of the linear relations for the two analysers are of the same magnitude, however opposite signs (-2.2 ‰ for Picarro and 3.6 ‰ for AP2E, Fig. 5a and Table 3). This indicates that the absolute calibration of the AP2E analyser will be of opposite sign and slightly larger than the Picarro, which is also visible in Figs. 6 and 7. The associated error on both linear coefficients from the two analysers are also comparable, despite the ones for the AP2E analyser being slightly higher (Fig. 5a and Table 3).

For δD, the Picarro shows an absolute calibration slope also close to one (0.98), while the AP2E analyser shows a lower slope (0.94, Fig. 5b and Table 3). This indicates that the AP2E analyser requires a stronger correction to calibrate the data against VSMOW-SLAP. Similarly, the intercepts of the linear relations are very different and of opposite signs between the two analysers (3.6 ‰ for Picarro and -22.9 ‰ for AP2E, Fig. 5a and Table 3). This shows that the AP2E analyser is measuring further away than the true isotopic composition compared to the Picarro, and therefore that the correction to calibrate the AP2E analyser will be stronger than the one for the Picarro. This is also visible in Figs. 6 and 7. The results of the linear regressions for δD also show that the errors associated to the coefficients for the AP2E analyser are twice as high than the ones for the Picarro (Fig. 5 and Table 3). This means that the error on the absolute calibration of the AP2E analyser is higher, as also described in the following section (Sect. 3.1.4).

In Eq. (4) (Sect. 2.2.4), σi,drift is estimated as one standard deviation of the selected calibrations over six months, combining both laboratory standards (Fig. 4b and c). Table 4 summarises the values of σi,drift found for δ18O and δD and for each analyser. Associated with the measured atmospheric humidity, this provides the measurement uncertainty on the final δ18O, δD and d-excess from both analysers presented along the data in the following section. For comparison, the uncertainty on δ18O estimated by Leroy-Dos Santos et al. (2021) is approximately 2.5 ‰ when the humidity is maximal (between 400 and 600 ppmv, their Fig. 7) and close to 4.5 ‰ when the humidity is minimal (around 200 ppmv, their Fig. 7). Although our estimation method differs from their study, we find here the uncertainty on the δ18O measured by the Picarro to be between 0.3 ‰ at 1000 ppmv (approximate maximum value during the studied time period) and 1.5 ‰ at 200 ppmv. Both values were calculated with Eq. (4) and the values of σi,drift in Table 4.

Besides, the Monte Carlo tests show that between December 2023 and January 2024, the uncertainty (σMC) of δ18O from the Picarro is 0.5 ‰ and 2.7 ‰ for δD, which leads to an uncertainty of 3.0 ‰ on d-excess. The AP2E analyser shows higher uncertainties, with σMC=0.8 ‰ for δ18O, 4.9 ‰ for δD, and 5.3 ‰ for d-excess. As expected, the errors σMC on δ18O and δD are in the same order of magnitude as the corresponding σdrift (Table 4), since they are computed with the same set of calibrations (Sect. 2.2.4).

The raw humidity measured by both analysers show the same variations over the whole period (Figs. 6a and 7a), except for a bias already identified in Sect. 3.1.1. After the calibration against the independent humidity sensor, the humidities are in excellent agreement over the whole period (Table 5). The calibrated humidities are showing the same diurnal variations for both analysers, synchronous with the temperature diurnal cycle on site (Fig. 7a). In addition, both instruments record the decrease of the humidity from the beginning of February, coinciding with the onset of the winter at Dome C (Fig. 6a).

Contrary to the humidity, the calibration of the raw data has a significant effect on the δ18O time series of both analysers. For the AP2E analyser, the calibration of the raw δ18O time series shifts it towards higher values (Figs. 6b and 7b), with a mean difference of 9.2 ‰ over the whole period between the raw and calibrated time series (Table 5). This shift is expected from the absolute calibration curve (Sect. 3.1.3). The amplitude of the diurnal cycle is also slightly reduced after applying the calibration (Fig. 7b), due to the humidity- δ18O response of the analyser (i.e. positive correction for low humidities and negative correction for high humidities, Sect. 3.1.2). For the Picarro analyser, the raw and calibrated δ18O time series show a mean difference of -9.3 ‰ (Table 5), which is mostly due to the part of the time series from beginning of February onwards (Fig. 6b). The amplitude of the diurnal cycle is larger after calibration, as expected from the humidity-δ18O response of the Picarro which shows negative correction for lower humidities (Fig. 3a, Sect. 3.1.2). This is further visible on the period from the end of January onwards, where the diurnal cycles show an opposite behaviour between the raw and calibrated data: the raw data is in opposite phase to the humidity (minimum δ18O associated with maximum humidity) and the calibrated data is in phase with the humidity (minimum δ18O associated with minimum humidity). This is an effect of the large humidity-δ18O response of the Picarro at low humidities (Fig. 3a, Sect. 3.1.2).

Compared to δ18O, the raw and calibrated δD time series from both instruments are rather similar, at least during the period where the humidity is above 200 ppmv (mid-December to end of January, Fig. 6c). The calibration of both analysers modifies the average δD values (mean difference of 7.4 ‰ for the Picarro and 19.1 ‰ for the AP2E analyser over the whole period, Table 5). The calibration of the δD time series does not affect the amplitude of the diurnal cycle for neither analyser (Fig. 7c). Both raw δD time series compare relatively well from mid-December to the end of January (dashed lines in Fig. 6c), with the same in-phase relationship between δD and the mixing ratio as for the calibrated δ18O time series. This in-phase relationship between δD and the humidity is preserved after calibration (plain lines in Figs. 6c and 7c).

There is a good agreement between the δ18O calibrated time series from the AP2E and Picarro analysers over the period, until they start to diverge mid-February (Fig. 6b). Considering the entire period, the two δ18O calibrated time series show a mean absolute difference of 3.6 ‰ (Table 6), which is within the range of uncertainties of the calibrated time series (shaded areas in Fig. 6b), and a squared Pearson correlation coefficient R2 of 0.58 (Table 6). When considering only the period before mid-February, the mean absolute difference is reduced to 1.8 ‰ and R2 is improved to 0.8 (not shown). The good agreement between the two analysers confirms that the calibration is valid for the range of humidities encountered over this period.

As for δ18O, we observe that the calibrated δD time series from both instruments agree well between mid-December to mid-February, when similarly to δ18O they start to diverge (Fig. 6c). Over the entire period, the mean absolute difference between the two calibrated δD time series is 22.1 ‰ (Table 6), which is also within the uncertainty of both calibrated time series (shaded areas in Fig. 6c), and R2 is 0.59 (Table 6). When considering only the period before mid-February, the mean absolute difference is reduced to 10.7 ‰ and R2 is improved to 0.85 (not shown).

Finally, the raw time series of d-excess are very different between the two analysers (Figs. 6d and 7d). However, after the calibration of both analysers, the two d-excess time series are comparable within their uncertainty range (Figs. 6d and 7d), with a mean absolute difference of 27.7 ‰ (Table 6). As for δ18O and δD, the calibrated d-excess time series of the two analysers diverge from mid-February onwards (Fig. 6d).

The divergence from mid-February in both δ18O and δD between the two instruments is probably due to the increase of instantaneous measurement noise of the analysers when the humidity decreases. It is also related to the difficulty of calibrating the instruments for very low humidity levels (Sect. 3.1.2). This is reflected in the uncertainty of the measurements which increases significantly for both instruments from mid-February onwards (Fig. 6b and c), when the mixing ratio is consistently below 200 ppmv (Fig. 6a). In addition, as stated above, the mean absolute difference and R2 values calculated between the two calibrated time series are improved when considering the period before mid-February. We therefore restrict the comparison between the observations and the model in Sect. 3.3 to the period before mid-February, and make available in public access only this part of the dataset in Landais et al. (2024b) for future work.

After mid-February, it seems that this particular Picarro analyser best captures the atmospheric water vapour isotopic composition at these low humidities. As shown in Fig. 6b and c, the diurnal cycle in δ18O and δD measured by the Picarro follows the diurnal cycle in the atmospheric humidity (minimum δ18O and δD associated with minimum humidity), which is what we expect. In contrast, the AP2E analyser shows an opposite phase between the humidity and the δ-values. Nevertheless, this does not permit concluding on the ability of the AP2E or Picarro analyser to measure at very low humidity levels, since the data shown here are only calibrated down to 50 ppmv (Sect. 2.2.2). Further efforts should be made to better constrain the calibration in order to use the instruments at very low humidities.

In addition, the amplitude of the mean diurnal cycle in δ18O (calibrated data, calculated over the period 11 January to 15 January 2024 shown in Fig. 7b) is similar for both instruments: 5.7 ‰ for the Picarro analyser (from -73.4 ‰ to -67.7 ‰, not shown) and 4.7 ‰ for the AP2E analyser (from -72.6 ‰ to 67.9‰, not shown). The mean diurnal cycle in δD over the same period is also comparable for both analysers: 34.9‰for the Picarro analyser (from -523.2 ‰ to -488.3‰, not shown) and 29.5 ‰ for the AP2E analyser (from -509.6 ‰ to -480.1‰, not shown). As for δ18O and δD, both instruments show a similar mean diurnal cycle in d-excess: 11.6 ‰ (from 52.9 ‰ to 64.5 ‰, not shown) for the Picarro analyser and 13.5 ‰ (from 63.2 ‰ to 76.7 ‰, not shown) for the AP2E analyser. Considering the uncertainties on the δ18O, δD and d-excess values of both instruments, we conclude that both analysers compare well and that the AP2E captures well the diurnal cycle measured by the Picarro analyser.