We designed an evaporation protection modification for the 1 L sample bottles of the 6712 full-size portable sampler (Teledyne ISCO, Lincoln, USA; hereafter referred to as “ISCO autosampler”). The presented evaporation protection consists of a 100 mL syringe housing (i.e., BP PlastipakTM 100 mL syringe with catheter tip, without its piston and rubber piston stopper) with attached Luer tip adapter (BP PlastipakTM). On the Luer tip, we fit a 1 mm inner diameter silicone tube of approximately 9 cm length to reach below the water level of the sample in the bottle (Fig. 1a and b). The barrel flange of the syringe housing is trimmed on one side (Fig. 1a) to allow the retrofitted sample bottles to properly fit into the bottom compartment of the ISCO autosampler. This modified syringe housing is then plugged into the opening of an ISCO sample bottle (Fig. 1b). Because a small gap remains between the syringe housing and the inner rim of the sampler bottle opening (i.e., not air-tight), pressure differences due to water flowing into the bottle will equilibrate with the outside conditions. Thus our system does not require an external tube for pressure equilibration, such as the “tube-dip-in-water collector” proposed by Gröning et al. (2012). Because the end of the silicone tube is fully immersed in the sample liquid, only the cross-sectional area of the silicone tube is exposed to the ambient atmosphere (rather than the entire cross-sectional area of the water surface), minimizing vapor exchange with the surrounding atmosphere. The presented design of the evaporation-protected ISCO bottle ensures a smooth, splash-free sample flow from the syringe through the silicone tube into the bottle when filled at a flow rate of approximately 100 mL min-1. Furthermore, it is robust, cheap (< USD 5 per sample bottle), chemically inert, and easy to disassemble and to clean.

In field operation for the collection of streamwater samples, the autosampler should be programmed to not exceed the filling rate that can be accommodated by the narrow silicone tube. This can be accomplished by programming the autosampler to deliver a series of 100 mL aliquots, allowing enough time between them (about 1 min minimum) so that they can drain from the 100 mL syringe into the sample bottle. If possible, one should also limit the total sample volume so that the water line is somewhere in the narrow silicone tube and not in the syringe, in order to limit the water surface that is available for evaporation or condensation. In order to prevent debris (e.g., sediment, insects, leaves) from clogging the evaporation protection system, the streamwater intake or precipitation funnel can additionally be equipped with a screen. To transport the filled sample bottles, the syringe housing has to be removed and the bottles have to be sealed with screw caps supplied by the manufacturer. If the bottles are transported upright and leakage is unlikely to occur, the syringe housing can also stay in place and its upper opening can be sealed with the black rubber piston stoppers that are supplied together with the syringes (BP PlastipakTM 100 mL syringe with catheter tip; Fig. 1c).

We conducted three experiments to assess the effects of evaporation and vapor mixing on the isotopic composition of the liquid samples in ISCO 6712 autosamplers, comparing the retrofitted ISCO bottles to unmodified ISCO bottles. In Experiment 1, we simulated a daily sampling routine under extremely dry and warm conditions to test for evaporative fractionation effects over different storage durations. In Experiment 2, we used two contrasting reference waters to test for changes in their isotopic compositions due to vapor transfer between samples, in addition to fractionation effects under ambient conditions with diurnal fluctuations in temperature and humidity. Experiment 3 evaluated the performance of the retrofitted ISCO bottles during 61 two-to-three-week cycles over a nearly 4-year deployment at two field sites in the northern Swiss pre-Alps.

For Experiment 3, all water samples collected between 6 October 2015 and 13 December 2017 were analyzed at the laboratory of the Swiss Federal Institute for Forest, Snow and Landscape Research (WSL) with an LGR IWA-45-EP isotopic water analyzer (Los Gatos Research, ABB Los Gatos Research, San Jose, CA, USA) with a measurement precision of 0.5 ‰ for δ18O and 1 ‰ for δ2H. All of Experiment 3's samples collected after 13 December 2017, and all water samples of Experiments 1 and 2, were analyzed with a cavity ring-down spectrometer at the ETH Zurich laboratory (L2140-i liquid isotope analyzer, Picarro Inc., Santa Clara, CA, USA) with a measurement precision of 0.2 ‰ for δ18O and 1 ‰ for δ2H. All isotope values in this study are reported in δ-notation relative to Vienna standard mean ocean water (V-SMOW), and the measurement uncertainty is provided as standard deviations calculated from 2 to 3 repeated injections of each sample.

To quantify isotopic effects in the water samples, we calculated the isotopic difference (Δδ18O and Δδ2H, ‰) between the water sample at the end of the storage period and the reference water at the beginning of the storage period: 4ΔδiE=δiEsample-δiEreference, where δiEsample and δiEreference are the delta values of the isotope iE in the sample water or the reference water, respectively. For Experiment 1, we compared the isotope composition of the water samples from the open and retrofitted bottles (δ2Hsample, δ18Osample) to the isotope composition of the reference water from the storage tank (δ2Hreference, δ18Oreference). Because each second day we collected one reference water sample from the tank and filled one open and one retrofitted bottle with reference water, the comparison of the isotopic differences (ΔδiE) of samples from bottles with and without evaporation protection assesses the effectiveness of the retrofitted sampler bottles in protecting against evaporative enrichment. For Experiment 2, we compared the isotopic composition of the RefA and RefB water samples from the various open and retrofitted bottles (δ2Hsample, δ18Osample) to the isotopic composition of the closed sample bottles with the corresponding reference water RefA or RefB (δ2Hreference, δ18Oreference) in order to quantify isotope effects due to vapor mixing and evaporation. For Experiment 3, we compared the isotopic composition of water samples from the open and retrofitted bottles (δ2Hsample, δ18Osample) to the isotope values of the reference water in the tightly sealed bottles (i.e., δ2Hreference, δ18Oreference) for each sampling period.

During Experiment 1, humidity outside the ISCO autosampler stayed relatively constant at approximately 11±3 % (mean ± 1 standard deviation), while it continuously increased inside the ISCO bottom compartment from 33 % to 100 % between day 1 and 13 and then remained at 100 % until the end of the experiment (Fig. 2a). Air temperature outside the ISCO autosampler was around 35±1 ∘C with distinct diurnal variations (a 1.2 ∘C temperature drop at the beginning of the fourth day was caused by moving the humidity–temperature logger from a position close to the heater to a higher position near the sampler's control unit to better represent the conditions inside the ventilated chamber). The air temperature inside the ISCO housing was 36±1 ∘C and did not exhibit strong diurnal patterns.

The sample bottles stored outside the autosampler (i.e., at 11 % relative humidity and 36 ∘C temperature) experienced different degrees of evaporative fractionation between the start and end of the monitoring period (Table 1): while evaporative fractionation was insignificant in the closed bottle, we observed isotopic enrichment in both the retrofitted and the open bottle. Enrichment was substantially stronger for the sample in the open bottle (a change of roughly 100 ‰ in δ2H and 22 ‰ in δ18O within the first 12 d) compared to the sample in the retrofitted bottle (a change of 9 ‰ in δ2H and 2 ‰ in δ18O over 24 d). We sampled the open bottle already on day 12 of the experiment because we had observed substantial evaporation by then. At the end of the monitoring period (day 24), the water from the open bottle had evaporated completely, while the loss of water volume was small in the retrofitted bottle (Fig. 3).

The δ2H values of the water samples from inside the ISCO autosampler show that evaporative fractionation differed between samples from open and retrofitted bottles and also varied with storage duration (Fig. 2b and c). The isotopic differences (Eq. 4) were generally smaller for samples in retrofitted bottles compared to those in the open bottles. For δ2H, the isotopic differences (Δδ2H) in the retrofitted bottles were mostly close to 0 ‰ independent of storage duration, while the isotopic enrichment in the open bottles ranged up to 5 ‰ (Fig. 2b). For δ18O, we obtained less clear fractionation signals, and while the enrichment was always greater in samples of open bottles compared to those in retrofitted bottles, we also observed isotopic depletion of up to -0.3 ‰ for samples in retrofitted bottles that were filled early on in the experiment (data for δ18O are presented in Fig. S1 in the Supplement).

Water samples filled on and before day 14 of Experiment 1, i.e., samples with 10 or more days of storage time, showed substantially larger Δδ2H values in the open bottles than in the retrofitted bottles: samples in open bottles experienced stronger enrichment (Fig. 2b). Conversely, samples collected on day 16 and later, and thus stored for 8 d or less, experienced little or no evaporative fractionation, independent of the bottle type (i.e., Δδ2H was not significantly different from zero). This decrease in evaporative fractionation in the later samples may have been caused by the increase in relative humidity to approximately 90 % on day 10 and 100 % on day 12 inside the autosampler housing (Fig. 2a). Surprisingly, samples filled on days 10 and 12 showed stronger enrichment than those filled on adjacent days, both for open and retrofitted bottles. Because this isotope effect occurred in both bottle types it cannot be attributed to a specific process; it may have been related to interferences in the isotope analyzer.

Figure 2c compares the Δδ2H values of samples from open and retrofitted bottles against the average relative humidity inside the ISCO autosampler. While the change in relative humidity inside the ISCO autosampler did not seem to affect the samples in the retrofitted bottles, we obtained a nearly linear relationship for the open bottles indicating a 2.8 ‰ enrichment per 10 % decrease in relative humidity (p<0.005). Such a relationship is expected, because the vapor phase in the open bottle is in exchange with the vapor phase inside the ISCO housing, and evaporation from the liquid phase is generally faster when water vapor concentrations in the gas phase are lower (assuming constant temperature). Due to the much smaller contact area between the liquid and vapor phases in the retrofitted ISCO bottles, vapor exchange was reduced and evaporation from the liquid sample was much smaller (even when relative humidities were below 90 % inside the autosampler housing; Fig. 2c). Retrofitting made little difference at ≈100 % humidity, when vapor pressure deficits, and thus evaporation rates, were minimal. Relative humidity outside the ISCO autosampler and temperatures inside and outside the ISCO autosampler remained nearly constant throughout the monitoring period, so their relationship with the observed isotopic composition could not be identified and their effect on evaporative fractionation in this laboratory experiment could not be assessed.

In spite of the high temperatures and low relative humidity inside the ISCO autosampler, the observed fractionation effects during Experiment 1 were not large. This may have been due to some limitations in the setup. For one, we did not use the built-in peristaltic pump of the ISCO system to fill our samples, which follows the protocol for collecting precipitation samples but is not suitable for streamwater grab sampling. In case of streamwater sampling, the pump is used and the tubing between the sampling location and pump is flushed with air before sampling. This process likely results in an intake of air into the ISCO and consequently enhanced vapor exchange with the surrounding atmosphere, which may enhance isotopic fractionation of the collected water sample. In addition, we could not measure the water volumes in the sample bottles at the time of filling and at the end of the experiment because we did not want to open the ISCO autosampler during the experiment period. While we took care to fill exactly 400 mL reference water into each sample bottle through the distributor arm, we cannot exclude that some spillage occurred during the filling procedure or that some residual water remained in the sampling tube. It is therefore not possible to assess the exact amount of sample volume that was lost due to evaporation.

Ambient conditions during Experiment 2 were colder and more humid compared to Experiment 1 and substantially more variable. Outside the ISCO autosamplers, air temperature (mean ± standard deviation) was 13.3±6.2 ∘C in the outdoor setting and 18.6±4.7 ∘C indoors, while relative humidity was 73.5±23.0 % outdoors with distinct daily fluctuations and 44.8±8.9 % indoors (values of temperature and relative humidity measured outside the ISCO autosamplers are not shown but are provided in the Supplement). Temperature and relative humidity measured inside the autosampler housings exhibited similar but damped diurnal patterns (see Fig. 4a and b). The temperature and relative humidity inside the outdoor ISCO were 16.7±6.7 ∘C and 86.0±13.6 %, respectively. For the indoor ISCO autosampler, the respective values were 18.2±2.9 ∘C and 96.7±2.7 %. In contrast to Experiment 1, the relative humidity inside the autosampler housing did not increase gradually to 100 % over several days but remained high throughout the experiment, probably because all sample bottles were filled from the start, instead of successively as in Experiment 1.

The changes in isotopic composition, Δδ2HRefA and Δδ2HRefB, were calculated following Eq. (4) with sample being RefA or RefB water in the open or retrofitted bottles and reference being RefA or RefB water in the closed bottles. We observed no significant change in the isotopic composition of samples in retrofitted bottles (both Δδ2HRefA and Δδ2HRefB≈0 ‰, red filled markers in Fig. 4c–f; results for Δδ18O were similar, see Fig. S2), whereas the isotopic composition of samples in open bottles changed over the course of the experiment by up to 10 ‰ (blue open markers in Fig. 4c–f). The observed isotopic differences of samples in open bottles were more pronounced for smaller sample volumes (comparison of circles and diamonds in Fig. 4c–f). It was also larger in the outdoor setting compared to indoor conditions (comparison of Fig. 4c and e with Fig. 4d and f), even though the average temperature was lower in the outdoor setting. This may indicate that the average temperature is less important for causing isotope effects than the magnitude of the temperature fluctuations that may trigger gas volume exchanges. In contrast to Experiment 1 (ventilated chamber), Experiment 2 was characterized by significant daily temperature fluctuations (20 ∘C and more inside the autosampler that was situated outdoors; Fig. 4a). During the daytime when the air temperature increased, the air inside the autosampler expanded and some of the (moist) air was pushed out of the device; when temperatures dropped (at night), the opposite happened: the air inside the autosampler contracted and sucked in fresh air from outside. This “sampler breathing” probably happened on a daily basis, resulting in a greater air exchange with the outside, which in turn may have reduced the humidity inside the autosampler and resulted in more evaporation. In addition, the greater potential for wind-driven ventilation of the autosampler located outdoors may also have enhanced evaporative fractionation effects in the open sample bottles. Samples which underwent a stronger change in isotopic composition also experienced a greater loss of sample volume between the start and end of the experiment. We observed a larger decrease in sample volumes in open bottles compared to the retrofitted sample bottles, a larger decrease in sample volumes in the outdoor setting compared to the indoor setting, and a larger relative decrease in the 200 mL samples compared to the 400 mL samples (see Fig. S3).

In the outdoor setting, the samples in the open bottles became isotopically heavier, with larger changes observed in the 200 mL samples than in the 400 mL samples (e.g., δ2H in RefB water increased by 9.6 ‰ and 3.9 ‰ in 200 and 400 mL samples, respectively; Fig. 4e), likely due to evaporative fractionation. For the open bottles in the indoor setting, however, RefA samples became isotopically lighter by about 1 ‰ in δ2H (-1.2 ‰ and -0.9 ‰ for 200 and 400 mL, respectively; Fig. 4e), while RefB samples became roughly 2 ‰ heavier (2.6 ‰ and 1.6 ‰ for 200 and 400 mL, respectively; Fig. 4f). The isotopic lightening of RefA samples may be explained by mixing in the vapor phase of isotopically heavier RefA water with isotopically lighter RefB water and subsequent condensation in both samples. This isotopic exchange should make RefA samples isotopically lighter and RefB samples isotopically heavier, in addition to any isotopic fractionation due to net evaporative losses from both samples. Thus, a large part of the observed enrichment in RefB water in the indoor setting may have been due to isotopic exchange with the heavier RefA water, in addition to any evaporative fractionation. A likely reason why the mixing effect was only visible in the indoor setting may be that evaporation was smaller compared with the outdoor setting; the greater evaporative losses (and thus evaporative fractionation) in the outdoor setting may have overprinted the vapor mixing effect. In either case, mixing and/or evaporative fractionation only affected the isotopic composition in the open sample bottles, while no measurable effect was observed in samples from the retrofitted bottles.

We can quantify the isotope effects due to mixing and evaporative fractionation in the different settings under the assumption that (a) evaporative fractionation and mixing have additive effects, (b) the per mil change due to evaporative fractionation is the same for RefA and RefB waters, and (c) mixing has an exactly inverse effect on the two waters (i.e., it results in the same degree of isotopic depletion in the heavier RefA water and enrichment in the lighter RefB water): 5mRefA+f=Δδ2HRefA, 6mRefB+f=Δδ2HRefB,7mRefA=-mRefB, where mixing-induced isotopic change is denoted by m for RefA and RefB waters, and the change in isotopic composition due to evaporative fractionation is denoted by f.

The results of this analysis are illustrated in Fig. 5 for deuterium, confirming that isotope effects due to mixing and evaporation are small in samples from the retrofitted bottles (red filled markers). In the open bottles (blue open markers), the isotope effects due to evaporative fractionation were 1.5 to 2 times larger than the isotopic change due to mixing in the outdoor setting, but fractionation was less important than mixing in the indoor setting. Both the mixing- and fractionation-induced isotope effects were roughly twice as large in the 200 mL samples as in the 400 mL samples (diamonds vs. circles, respectively). Applying Eqs. (5)–(7) to oxygen-18 yielded similar results (see Fig. S4).

In summary, the results of Experiment 2 confirmed the findings from Experiment 1 that the retrofitted ISCO bottles efficiently protected the collected water samples from undergoing isotopic changes due to both evaporative fractionation and vapor mixing.

During the field experiment (Experiment 3, October 2015 to June 2019), we observed distinct seasonality in air temperature but no seasonal pattern in relative humidity at both field sites and slightly higher temperatures and humidity at the EIN site (Fig. 6). At both field sites, the vapor pressure deficit (VPD), which is strongly correlated with air temperature, peaked around June and July and was lowest in December and January. Because these climatic variables exhibited very similar behavior at both field sites, we decided to pool the isotope data sets from both sites for analysis.

The isotope data from 8 March 2016 were excluded from this analysis because the water samples in the retrofitted bottles were isotopically lighter than the reference water (e.g., Δδ2H ranged from -4.7 ‰ to -3.5 ‰) for unknown reasons. In addition, we removed the data points from 14 February 2017 from our analysis because of an anomalous δ18O measurement of the water sample from the retrofitted bottle at the ERL site.

Experiment 3 resulted in 244 usable samples (i.e., 61 samples per site and bottle type) for which the storage duration varied between 12 and 23 d. The isotopic differences of the samples in open bottles relative to the reference water exhibited substantial scatter, with values between -2 ‰ and +3.5 ‰, but the average isotopic differences (mean ± 1 standard error were 1.45±0.22 ‰ for Δδ2H and 0.27±0.05 ‰ for Δδ18O and thus deviated statistically significantly from zero. Conversely, when the retrofitted bottles were used, the isotopic differences were not statistically significantly larger than zero, i.e., 0.10±0.11 ‰ for Δδ2H and 0.05±0.03 ‰ for Δδ18O. Figures 7 and S5 show that the isotopic differences of the samples in open bottles relative to the reference water were positively correlated with average air temperature and thus with VPD. Pearson correlation coefficients between air temperature (average, minimum and maximum) and Δδ2H were r>0.70 (p<0.001) and r>0.60 (p<0.001) for Δδ18O. For VPD, the correlation coefficients were r>0.56 (p<0.001) for both isotopes. No statistically significant relationships were evident for the samples from the retrofitted bottles (r<0.17, p>0.1 for both isotopes; see also the red data points in Figs. 7 and S5). Similarly, higher climatic variability at our field sites (represented by the maximum changes in air temperate and relative humidity) was associated with larger isotopic differences only for the open bottles (the relationship was statistically significant only for the change in air temperature, with r>0.40 and p<0.0001 for both isotopes; Figs. 8 and S6).

Overall, our results indicate that the retrofitted sample bottles significantly reduced isotopic fractionation compared to the open sample bottles when deployed over 2–3-week periods under the ambient climatic conditions at our two field sites.