Characterising water vapour concentration dependence of commercial cavity ring-down spectrometers for continuous onsite atmospheric water vapour isotope measurements in the tropics

. The recent development and improvement of commercial laser-based spectrometers have expanded in situ continuous observations of water vapour (H 2 O) stable isotope ratios (e.g., δ 18 O, δ 2 H, etc.) in a variety of sites worldwide. However, we still lack continuous observations in the Amazon, a region that significantly influences atmospheric and hydrological cycles on local to global scales. In order to achieve accurate on-site observations, commercial water isotope analysers require regular in situ calibration, including H 2 O concentration dependence ([H 2 O]-dependence) of isotopic 15 accuracy. Past studies have assessed [H 2 O]-dependence for air with H 2 O concentrations up to 35,000 ppm, a value that is frequently surpassed in tropical rainforest settings like the central Amazon where we plan continuous observations. Here we investigated the performance of two commercial analysers (L1102i and L2130i models, Picarro, Inc., USA) for measuring δ 18 O and δ 2 H in atmospheric moisture at four different H 2 O levels from 21,500 to 41,000 ppm. These H 2 O levels were created by a custom-built calibration unit designed for regular in situ calibration. Measurements on the newer analyser model 20 (L2130i) had better precision for δ 18 O and δ 2 H and demonstrated less influence of H 2 O concentration on the measurement accuracy at each moisture level compared to the older L1102i. Based on our findings, we identified the most appropriate calibration strategy for [H 2 O]-dependence, adapted to our calibration system. The best strategy required using two pairs of a two-point calibration with four different H 2 O concentration levels. The smallest uncertainties in calibrating [H 2 O]-dependence of isotopic accuracy of the two analysers were achieved using a linear-surface fitting method and a 28 h 25 calibration interval, except for the δ 18 O accuracy of the L1102i analyser for which the cubic fitting method gave best results. The uncertainties in [H 2 O]-dependence calibration did not show any significant difference using calibration intervals from 28 h up to 196 h; this suggested that one [H 2 O]-dependence calibration per week for the L2130i and L1102i analysers is sufficient. continuous in-situ observations. Over a two week periods, we examined the effects of H 2 O concentration on isotopic measurement precision and accuracy for an older (L1102i) and a newer CRDS models (L2130i). We used a custom-made calibration system that 35 regularly supplied standard water vapour samples at four different H 2 O concentrations covering high moisture conditions (21,500 to 41,000 ppm) expected based on past measurement at the ATTO site. Standard water vapour samples were made from two standard waters, almost covering the previously reported isotopic ranges ( δ 18 O = -19.4 to -6.7 ‰ and δ 2 H = -151 to -42 ‰) for water vapour samples in Manaus, located near the ATTO site, or in the Ducke Reserve near Manaus (Matsui et al., 1983; Moreira et al., 1997; IAEA/WMO, 2020). We also assessed which [H 2 O]-dependence calibration strategy, based 40 on the two-week operation, can reduce measurement uncertainty of the two CRDS models the most. According to the determined best calibration strategy, we discussed whether the CRDS analysers can sufficiently detect natural signals of stable water vapour isotopes, expected at the ATTO site. [H O]-dependence 2 2 The variations are higher than the corrected deviation of each CRDS analyser This supports that both the analysers will detect diel or probably seasonal/interannual variations in vapour in rainforest. adjusted to our custom-made calibration system, and then evaluated which [H 2 O]-dependence calibration procedure best improved the accuracy of δ 18 O and δ 2 H measurements for both the L2130i and L1102i analysers. The best [H 2 O]-dependence strategy was the DI1-2*2Pairs strategy that required two pairs of a two-point calibration with different moisture levels from 21,500 to 41,000 ppm. The 28 h interval strategy with the linear-surface fitting method leads to the most accurate measurements for both the CRDS analysers, except δ 18 O accuracy of the L1102i analyser that required the cubic fitting method. In addition, [H 2 O]-dependence calibration uncertainties hardly changed at any interval over 8 days. That indicates one [H 2 O]-dependence calibration per week is sufficient for correcting moisture-biased isotopic accuracy of


Introduction 1 30
δ 2 H) of the DI1 and DI2 standards were analysed at the stable isotope laboratory (BGC-IsoLab) of the MPI-BGC using Isotope Ratio Mass Spectrometry (IRMS). For details on the IRMS technique, we refer readers to Gehre et al., (2004). The DI1 and DI2 standards were calibrated against VSMOW and SLAP via in-house standards: DI1-δ 18 O = -25.07 ± 0.16‰, DI1-δ 2 H = -144.66 ± 0.60‰, DI2-δ 18 O = -3.69 ± 0.15‰, DI2-δ 2 H = -34.30 ± 1.00‰ (also see the section S1 in the Supplement). The isotopic span of the DI1 and DI2 almost covers the previously reported range of δ 18 O (-19.4 to -6.7 ‰) 5 and δ 2 H (-151 to -42 ‰) for water vapour samples in the Ducke Reserve near Manaus or in Manaus, located near the ATTO site (Matsui et al., 1983;Moreira et al., 1997;IAEA/WMO, 2020). For calibration, we alternated between the two standard waters. One calibration run required 75 minutes, of which the first 30 minutes were used for stabilizing the produced standard water vapours at the highest concentration level and delivering it to the CRDS analysers. Subsequently the calibration system created stepwise lower concentration levels of the standard water vapour every 15 minutes by regulating 10 the dilution flow rate. One calibration run consisted of a 4-point concentration calibration at approximately 41,000 ppm, 36,000 ppm, 29,000 ppm, and 21,500 ppm. The actual measured mean and standard deviation of H 2 O concentration at the respective four moisture level for all the calibration cycles during the two-week operation are shown in Table 1. The dilution flow rates for the different moisture levels were set to 9 sccm (41,000 ppm), 14 sccm (36,000 ppm), 21 sccm (29,000 ppm) and 28 sccm for 21,500 ppm. The set-point values were not changed during the 2-week test, thus simulating remote 15 automatic onsite calibration runs. We used the last 7-minutes of data collected at each concentration level for the calibration assessment of the CRDS analysers. Immediately after a calibration cycle, the syringe-pump drained the remaining standard water inside the tube between the vaporization chamber and 3-way solenoid valve 1 (SV1) through the waste line and then washed the inner space between the vaporization chamber and the SV2 valve 3 times with the standard water scheduled for the next calibration cycle (Fig. 1). Subsequently, the rinsed vaporization chamber was fully dried with air from the dry-air 20 unit for 2 to 4 hours. These rinsing and drying steps prevented any residual memory effect from the last calibration cycle on standard water vapour isotopic compositions during the next calibration run. We started the next calibration cycle seven hours after the start time of the last calibration cycle (i.e., the interval of the same working standard was 14 hours).
Throughout the entire experimental period, the calibration system conducted automatically 24 calibration runs for each standard water, which used 160 mL each standard water in total.

Calibration for water vapour concentration dependence
We devised four strategies, referred to here as DI1, DI2, DI1-DI2*1Pair, DI1-2*2Pairs, to use the automated calibration system to determine and correct for [H 2 O]-dependence, and used the two-week operation to assess which calibration strategy decreased the uncertainties in δ 18 O and δ 2 H measurements the most. Figure 2 summarises the overview of the four calibration strategies. D11 and D12 refer to the two standard waters, measured at the four different [H 2 O]. For example, the dimensional (2D) and a three-dimensional (3D) fitting methods (i.e., linear, quadratic, cubic, quartic, and linear-surface fitting methods) to obtain [H 2 O]-dependence calibration fitting parameters.
As an example, Figure 3 illustrates five calibration fittings for [H 2 O]-dependence of δ 18 O accuracy for each CRDS analyser, acquired from two pairs of DI1 and DI2 calibration cycles (i.e., 3,4 and 7,8) following the DI1-2*2Pairs strategy at a 28 h interval (also c.f., Fig. 2 https://doi.org/10.5194/amt-2020-305 Preprint. Discussion started: 2 September 2020 c Author(s) 2020. CC BY 4.0 License. Figure 1 Schematic diagram of the calibration system. MFC, MFM and SV denote mass flow controller, mass flow meter, and 3-way solenoid valve, respectively. The profile system, prepared for in situ observation at the Amazon Tall Tower Observatory site (c.f., Andreae et al. 2015) in the Amazon tropical forest, is not described in this article. The diagram is not to scale. 5     (Fig. 4).
The larger variation in H 2 O concentration at 41,000 ppm also implies instability of the calibration system, which possibly 20 induces a decline in measurement precision of δ 18 O and δ 2 H values for the CRDS analysers at high humidity. Table 1 summarizes the precision of δ 18 O and δ 2 H for each [H 2 O] on each standard water for the L2130i and L1102i analysers. The L2130i analyser has the highest precision of δ 18 O measurement for the DI1 standard water at 29,000 ppm even though variability in H 2 O concentration measurement was higher (H 2 O-σ = 578.6 ppm at 29,000 ppm versus H 2 O-σ = 243.6 at 21,000 ppm). Additionally, the L1102i analyser had higher δ 18 O and δ 2 H measurement precision at higher moisture 25 conditions (≥ 29,000 ppm) than at the lowest moisture condition (= 21,500 ppm) for both the standard waters. The increase in measurement precision of L2130i and L1102i with [H 2 O], despite larger [H 2 O] variability, indicates that the instability of the calibration unit did not inherently exert a large influence on the measurement precision of the L2130i and L1102i analysers. Furthermore, the L1102i's δ 18 O and δ 2 H precision at 21,500 ppm (δ 18 O -σ = 0.11-0.14‰ and δ 2 H -σ = 0.99-1.01‰) were similar or better than those reported by Delattre et al. (2015) at 20,000 ppm for the L1102i (δ 18 O and δ 2 H 30 precision of 0.08-0.19‰ and 1.5-2.0‰ respectively based on 40 calibration data over 35 days). This proves that the calibration system has a negligible effect on the isotopic measurement precision for L2130i and L1102i analysers.
At all H 2 O concentration levels, and for both standard waters, the L2130i analyser had higher δ 18 O (σ ≤ 0.11‰) and δ 2 H (σ ≤ 0.59‰) precision under 30,000 ppm than over 30,000 ppm: δ 18 O -σ ≥ 0.12‰ and δ 2 H -σ ≥ 0.68‰ (Table 1). This indicates that the L2130i analyser can measure stable water isotopes more precisely for water vapour samples below 30,000 ppm. 35 Compared to the L2130i analyser, the L1102i analyser had higher precision for only δ 18 O below 30,000 ppm relative to over 30,000 ppm (Table 1). The different behaviour of both analysers described above would mainly be due to the old fitting algorithm used for the L1102i analyser (Aemisegger et al., 2012).

Accuracy of isotope values for water vapour concentration dependence
For the L2130i analyser, the δ 18 O deviation of both standard waters from reference values at 21,500 ppm gradually increased with H 2 O concentration, reaching a maximum median value of 0.32‰ for DI1 and of 0.28 ‰ for DI2 (Fig. 5a). For δ 18 O, these differences were significant for both standard waters between 41,000 and 36,000 or 29,000 ppm (Fig. 5a, Welch's ttest, p<0.01), but differences between 36,000 and 29,000 ppm were not significant (Fig. 5a, Welch's t-test, p>0.09). As with 10 δ 18 O, the values of δ 2 H measured with the L2130i for both standard waters differed significantly between 41,000 ppm and the lower H 2 O concentrations (Fig. 5b, Welch's t-test, p<0.05), without a significant difference between 29,000 ppm and 36,000 ppm (Fig. 5b, Welch's t-test, p>0.86). These results indicate accurate measurement of both δ 18 O and δ 2 H using the L2130i analyser require correction for [H 2 O]-dependence under high moisture conditions (>36,000 ppm H 2 O).
The differences in the deviation of each isotope from reference values at 21,500 ppm were similar for the two standard 15 waters at all concentration levels except 41,000 ppm (Figs. 5a and 5b, Welch's t-test, p>0.05), where the L2130i indicated differences in δ 2 H deviation for the two different standard waters (Figs. 5a and 5b, Welch's t-test, p<0.05). This finding indicates that the L2130i's δ 2 H accuracy for high moisture like 41,000 ppm is dependent on the isotopic composition, thus more than one standard water needs to be used in the field. The L1102i also had strong [H 2 O]-dependence for both isotopes, larger than that of the L2130i (Figs. 5a-b and 5d-e). The larger variations also led to large deviations in the d-excess values (Fig. 5f) Tremoy et al. (2011) observed negative δ 18 O deviations at 39,000 ppm with a range between -2 and 0 ‰, different from this study. In addition, they confirmed a smaller increase in δ 2 H deviations with H 2 O concentration from 20,000 to 39,000 ppm than this study. The above differences in δ 18 O and δ 2 H deviations between Tremoy et al. (2011) and this study shows that [H 2 O]-dependence of δ 18 O and δ 2 H accuracy must be evaluated for each individual analyser (Aemisegger et al., 2012;Bailey et al., 2015). 15 In summary, the measurement accuracy for in δ 18 O and δ 2 H is more dependent on H 2 O concentration for the L1102i than the L2130i, mainly due to the older fitting algorithm for the initial version of the L1102i. In other words, the accuracy of [H 2 O]dependence of δ 18 O and δ 2 H for the L2130i has been improved due to the updated/corrected fitting algorithm (Aemisegger et al., 2012), but our results still remind us of the importance of correcting for the [H 2 O]-dependence of δ 18 O and δ 2 H accuracy for the L2130i analyser, particularly for high moisture condition at 36,000 ppm and above. 20   Moreira et al. (1997), water vapour isotope values in Amazon rainforest is expected to change diurnally by up to 2‰ (δ 18 O) or 4-8‰ (δ 2 H) with H 2 O concentration. The diel isotope variations are higher than the corrected deviation values of each CRDS analyser (Fig. 8). This supports that both the CRDS analysers will detect diel or probably seasonal/interannual 5 variations in water vapour isotopes in Amazon rainforest.  Assuming continuous in situ observation together with regular calibration in tropical Amazon rainforest, we devised four calibration strategies, adjusted to our custom-made calibration system, and then evaluated which [H 2 O]-dependence calibration procedure best improved the accuracy of δ 18 O and δ 2 H measurements for both the L2130i and L1102i analysers. 20 The best [H 2 O]-dependence strategy was the DI1-2*2Pairs strategy that required two pairs of a two-point calibration with different moisture levels from 21,500 to 41,000 ppm. The 28 h interval strategy with the linear-surface fitting method leads to the most accurate measurements for both the CRDS analysers, except δ 18 O accuracy of the L1102i analyser that required the cubic fitting method. In addition, [H 2 O]-dependence calibration uncertainties hardly changed at any interval over 8 days.
That indicates one [H 2 O]-dependence calibration per week is sufficient for correcting moisture-biased isotopic accuracy of 25 https://doi.org/10.5194/amt-2020-305 Preprint. Discussion started: 2 September 2020 c Author(s) 2020. CC BY 4.0 License. the CRDS analysers. Nevertheless, to stay on the safe side, we decided to conduct the [H 2 O]-dependence calibration at 28 h or less interval. The best calibration strategy at 28 h interval supported that both the CRDS analysers can sufficiently distinguish temporal variations of water vapour isotopes in the aimed ATTO site.

Data availability
All the data used in this publication are freely available at the "https://dx.doi.org/10.17617/3.4n." 5

Author contribution
The calibration system was designed and developed by SK, JL, TS and US. The laboratory experiments were conducted by SK with assistance from JL, TS, US and FK. The IRMS analysis was done by HM and HG. DW helped us to check water vapour concentration in the ATTO site. SK conducted the data-analysis and wrote the manuscript with assistance from JL, FK, and HM.