The mole fraction and isotopic composition of molecular H2 was determined using the experimental setup developed by Rhee et al. (2004) and described in detail in Walter et al. (2012, 2013) and Batenburg et al. (2011). The D / 1H molar ratio in a sample, Rsample(D / H), is quantified as the relative deviation from the D / 1H molar ratio in a standard, Rstandard(D / H), as isotope delta δD value, and reported in per mill (‰): δD=Rsample(D/H)Rstandard(D/H)-1. The isotopic standard is Vienna Standard Mean Ocean Water (VSMOW). H2 mole fractions are reported as mole fractions in nmol mol-1, abbreviated ppb (10-9, parts per billion) and linked to the MPI2009 calibration scale for atmospheric hydrogen (Jordan and Steinberg, 2011). As working standards, atmospheric air from laboratory reference air cylinders and synthetic air mixtures were used (Walter et al., 2012, 2013; Batenburg et al., 2011); the H2 mole fractions of the air in these cylinders were determined by the Max Planck Institute for Biogeochemistry, Jena, Germany. The atmospheric reference air and the synthetic isotope reference air were measured daily (atmospheric reference air at least twice) and results were used for correction of the sample measurements. The uncertainties reported here reflect random (i.e., repeatability) errors only and do not include possible systematic errors (Batenburg et al., 2011; Walter et al., 2012, 2013). Samples were measured in random order and analyzed within 3 months (ANT-XXIV/4, ANT-XXV/5, ANT-XXVI/1) up to 2 years (ANT-XXVI/4) after sampling. Storage tests indicate that glass flasks equipped with Kel-F valves are stable for H2 (Jordan and Steinberg, 2011). The mean measurement repeatability between the two measurements on the same flask was between ±3.2 (ANT-XXV/5, n=14) and ±6.4 ppb (ANT-XXVI/4, n=108) for the mole fraction and ±3.4 ‰ (ANT-XXVI/4, n=108) and ±5.0 ‰ (ANT-XXV/5, n=14) for the isotopic composition.

H2 and CO mole fractions were also measured by using a Peak Performer 1 RGA (Reduced Gas Analyzer) with synthetic air as a carrier gas, either continuously on-board (ANT-XXVI/4, see Sect. 2.4.2) or from discrete flasks in the laboratory (ANT-XXV/5 and ANT-XXVI/1). The discrete RGA measurements were performed from the same glass flasks after measurement of the isotope system (see above). Due to a remaining slight overpressure in the flasks, an active pumping of the air into the RGA was not necessary and the flasks were simply connected to the RGA inlet by Teflon tubing. The remaining pressure was mostly sufficient to perform 8 to 10 measurements. A slight memory effect was observed and thus only the last 5 measurements were taken into account when stable. Samples with only three or less valid measurements were not used for evaluation. The standards were the same as those used for the isotope system. For both cruises (ANT-XXV/5 and ANT-XXVI/1), the mean measurement repeatability was better than ±0.8 (H2) and ±2 % (CO). A comparison between the H2 mole fractions measured with the Peak Performer 1 RGA and the isotopic experimental setup reveals on average slightly lower RGA values of (7.5 ± 23.8) ppb (see Fig. 3).

For the on-board continuous measurements of H2 mole fractions a Peak Performer 1 RGA was used. The atmospheric air was drawn from the bridge deck to the laboratory in 1/4 inch Decabon tubing. The CO mole fraction was also measured in the same measurement and will be reported here for information, but without further discussion.

In alternating order, 10 air samples and 10 aliquots of reference air were measured, using synthetic air as carrier gas. Due to small memory effects, only the last 5 measurements of each were taken into account when the values were stable. The mole fractions of H2 and CO were calculated by using the mean of the enclosing standard measurements, with an estimated maximal error of ±5 %. For more details see Popa et al. (2014). The mean measurement repeatability for the air samples was ±1.7 % for H2 and ±3.6 % for CO in ambient air, respectively ±0.8 (H2) and ±0.9 % (CO) for the reference air. Comparing the H2 mole fractions measured continuously on the RGA with discrete samples measured on the isotope system and collected close in time, we found a mean offset of (-18.8 ± 16.4) ppb for the RGA results.

The discrete samples of extracted dissolved H2 were measured as described for the discrete atmospheric samples in Sect. 2.4.1. Details about assumptions and calculations to determine dissolved H2 concentrations and isotope delta values and quantity symbols are given in detail in the Supplement.

We define the extraction efficiency η as η=chVhcw0Vw, where Vh and Vw are the volume of the headspace and the water fraction, and ch the concentration of H2 in the headspace. The initial concentration of H2 in seawater, cw0, can be calculated from cw0=chVhηVw. The concentration in the headspace, ch, was not measured directly, but can be derived from the measured H2 mole fraction in the sampling flask. The sampling procedure following gas extraction under vacuum can be broken into three steps (see Methods section):

expansion of the headspace into the gas transfer system

As shown in the Appendix, the original concentration of H2 in seawater (in nmol L-1) can be calculated using the following equation ηcw0=yf1+VtVhphtm-ph(H2O)-ym1+VtVhphtm-phVwRT, where yf is the dry mole fraction of the air in the flask and ym the mole fraction of the makeup gas = (543.9 ± 0.3) ppb.

The extraction efficiency, η can be calculated from the following mass balance Vwcw0=Vhch+αVwch. Assuming that headspace gas phase and water phase are in equilibrium, the ratio of the H2 concentration in water and in the headspace is given by the Ostwald coefficient (Battino, 1984) (where the concentrations refer to in situ temperature): α=cwch. This gives for the extraction efficiency as defined in Eq. (2) η=1+αVwVh-1. In the present case, α=α(H2) was equal to 0.0163 ± 0.0001, which gives η=92.12 (±0.013) % for Vw/Vh=8.4/1.6 = 5.25.

Two alternative scenarios were considered to derive the δD of the dissolved H2, with scenario 1 assuming equilibrium isotopic fractionation between headspace and water, and scenario 2 assuming kinetic isotopic fractionation during extraction from Niskin bottle to glass vessel. Scenario 1:δw0=δh+ε1-η1+δh The equilibrium isotope fractionation between dissolved phase and gas phase is ε=(37 ± 1) ‰ at 20 ∘C (Knox et al., 1992). Scenario 2:δw0=1+δhη1-1-η1+εk-1≈δh+εk1-η1+δhln1-ηη The kinetic isotope fractionation during gas evasion is εk = (-18 ± 2) ‰ at 20 ∘C (Knox et al., 1992). The approximation is not used and only shown to illustrate the small difference between δw0 and δh when η≈1.

The temperature dependences of ε and εk are unknown and were neglected here.

The air saturation equilibrium concentration, csat(H2), was determined using the parameterization of Wiesenburg and Guinasso (1979). The H2 saturation anomaly, Δ(H2), was calculated as the difference between the measured H2 concentration, c(H2), and csat(H2): Δ(H2)=c(H2)-csat(H2). Meteorological and oceanographic parameters (radiation, air and water temperatures, salinity, relative humidity) were measured using standard instrumentation and recorded and provided by the data system of the ships. More information about devices and sensor documentation can be found on the website of the Alfred Wegener Institute http://dship.awi.de/. Backward trajectories were calculated using the backward “Hybrid Single Particle Lagrangian Integrated Trajectory” (HYSPLIT, Schlitzer, 2012) model of the National Oceanic and Atmospheric Administration (NOAA, http://ready.arl.noaa.gov/HYSPLIT.php).

We performed simulations of H2 mole fractions and isotopic composition with the global chemistry transport model TM5 (Krol et al., 2005), and compared them with our measurement data (Fig. 5). The simulation setup was similar to the one of Pieterse et al. (2013) and only a short description is given here. The model version used employs the full hydrogen isotopic scheme from Pieterse et al. (2009) and uses ERA-Interim meteorological data. The chemistry scheme is based on CBM-4 (Houweling et al., 1998), which has been extended to include the hydrogen isotopic scheme (that is, for all chemical species that include hydrogen atoms, HH and HD are treated separately and have different reaction rates). The H2 sources and isotopic signatures are given as input; these and also the H2 soil deposition velocities are identical to Pieterse et al. (2013).

The model has a relatively coarse spatial resolution of 6∘ longitude by 4∘ latitude, and a time step of 45 min. Daily average mole fraction fields are used for comparison to observations. The model results were interpolated to the time and location of the observations.

The climatological global oceanic emissions were calculated using the protocol of Pieterse et al. (2013), based on the GEMS database and an assumed mean oceanic H2 source of 5 Tg a-1 as given from global budget calculations (see Ehhalt and Rohrer, 2009, and references therein, Pieterse et al., 2013). The spatial and temporal variability of oceanic H2 emissions caused by N2 fixation are adopted from the spatial and temporal distribution of oceanic CO (Erickson and Taylor, 1992).

In total, 16 headspace samples were taken during the RV Polarstern cruise in April/May 2010 along the transect 32.53∘ W/18.79∘ S to 13.00∘ W/36.54∘ N and 6 samples during the RV L'Atalante cruise in February 2008 between 23.00–17.93∘ W to 16.9–19.2∘ N to analyze the H2 mole fraction and the isotopic composition (see Table 4).

Although our setup was a prototype with possibilities for improvement, the mole fractions are in line with previously published data. The H2 excess, Δ(H2), exceeds 5 nmol L-1, the saturation differ from close to equilibrium to 15-fold supersaturation. Highest supersaturation was found in the Southern Hemisphere between 16 and 11∘ S and in the Northern Hemisphere around the Cape Verde islands and the coast of Mauritania (Fig. 7a, Table 4).

Herr et al. (1984) reported patchy enhanced H2 concentrations in the surface water with up to 5-fold supersaturation in the subtropical south Atlantic (18–31 and 29–42∘ W). This is comparable to what Conrad and Seiler (1988) found in the southern Atlantic, on a similar cruise track as the RV Polarstern. Around the equator they measured H2 surface water concentrations up to 12-fold supersaturation. In the southern Pacific, Moore et al. (2009) combined H2 surface water measurements with N2 fixation measurements. They reported a strong correlation between these parameters, a patchy distribution and a steep maximum of H2 concentrations up to 12.6 nmol L-1 around 14∘ S.

The recently published data by Moore et al. (2014) show similar patterns across the Atlantic as we found, with highest values around the southern and northern subtropics. However, our saturations are lower than the ones given by them, especially in the Northern Hemisphere. Such differences might be caused by experimental issues such as overestimated extraction efficiency or can be due to real temporal variability as the sampling seasons differed. The extraction efficiency has been estimated as 92.12 (±0.013) % (see Supplement) and was incorporated into the calculation of the original seawater concentration. With respect to the assumption of biological production as main production pathway it is more likely that due to the different sampling seasons less H2 was produced in April than in October/November because of less microbial activity especially on the Northern Hemisphere in boreal winter.

Additional information about H2 sources comes from the analysis of the H2 isotopic composition. In the literature, only one experimental value of dissolved marine δD exists, δD = -628 ‰ (Price et al., 2007; Rice et al., 2010), but the origin of this value is unclear and it is based on unpublished data. Nevertheless, this value has been used as representative for oceanic emission in several global budget calculations (e.g., Price et al., 2007; Pieterse et al., 2011). Other authors (e.g., Rahn et al., 2003; Rhee et al., 2006) used a theoretical value of -700 ‰, as expected for thermodynamic isotope equilibrium between H2 and H2O based on the calculations of Bottinga (1969). The results presented here are the first well-documented experimental results for isotope analysis of dissolved H2 in seawater.

From the measurement of the isotopic composition of H2 in the headspace we calculate the isotopic composition of H2 that was originally dissolved in the sea water as described in Sect. 2.4.3 and in the Appendix, using two different assumptions for fractionation between dissolved H2 and H2 in the gas phase. The results shown in Table 4 reveal δD values for the dissolved H2 that vary within a wide range of -112 to -719 ‰ for both fractionation scenarios. Interestingly, δD shows two distinct groups of samples that can be separated by the water temperature (Fig. 7b). In water masses with a temperature above 21 ∘C the δD values are (-629 ± 54) ‰ (n=14), in water masses with a temperature of 20 ∘C or below δD values are (-249 ± 88) ‰ (n=8). There is no correlation of δD with salinity (Fig. 7c), but the high temperature (and low δD) waters show also a generally higher saturation than the low temperature (high δD) waters (Fig. 7d).

The very depleted isotope signature of the H2 in the warmer water masses is consistent with the values expected for biological production. The slight enrichment compared to the value of ≈-700 ‰ that is expected for biologically produced H2 in equilibrium with ocean water (Bottinga, 1969; Walter et al., 2012) may be caused by a partial consumption within the water, which would enrich the remaining fraction. The relatively smooth distribution of the isotopic composition of H2 in the atmosphere strongly indicates that the contribution from atmospheric variability cannot be a main contributor of the isotope variations observed in dissolved H2, even within the group of the depleted samples.

To our knowledge this is the first time that oceanic production of H2 has been directly attributed to biological processes by using isotope techniques. For the samples collected from warm surface waters, our results verify the general assumption of a biological production process as a main source of oceanic H2 to the atmosphere rather than photochemical or other sources (Herr et al., 1981; Conrad, 1988; Punshon and Moore, 2008; Moore et al., 2009). The dominance of biological formation at higher temperatures is qualitatively consistent with the general understanding of the temperature dependence of N2 fixation rates for N2 fixers such as e.g., Trichodesmium sp., which exhibit highest N2 fixation rates within a temperature range between 24 to 30 ∘C (Breitbarth et al., 2007; Stal, 2009). In fact, the saturations also show a correlation with temperature, but less clear than for δD (Fig. 7d), presumably due to simultaneous uptake and consumption processes in a complex microbial community.

However, this clear attribution is only valid in water masses with higher temperatures and the unexpectedly high δD values in cooler waters indicate the influence of other processes. The isotopic enrichment that is expected for removal of H2 (Chen et al., 2015; Rahn et al., 2003; Constant et al., 2016) is highly unlikely to cause a shift of almost 400 ‰ in δD from an assumed pure biological source, because in this case the removed fraction would have to be unrealistically large, as also recently argued for soil emitted H2 (Chen et al., 2015). We suggest that a source of H2 must exist in these surface waters, which produces H2 that is out of isotope equilibrium with the water. This can be either one single source with an isotopic signature of approximately -250 ‰, or an even more isotopically enriched source that mixes with the depleted biological source.

Punshon and Moore (2008a, and references therein), reported abiotic photochemical H2 production from CDOM and small organic compounds such as acetaldehyde or syringic acid. Walter et al. (2013) indicated, that biologically active regions such as the Banc d'Arguin at the coast of Mauritania could act as a pool of precursors such as VOCs for atmospheric H2 with high δD values. It is thus possible that abiotic photochemical production in the surface water might be an alternative source of H2 excess, which is not isotopically equilibrated with water, especially in regions with high radiation and biological activity, and less N2 fixation. Given the fact that the two groups of warm and cold waters are relatively well separated and there is not a continuous mixing curve between two end members, the explanation of a single different source seems more straightforward. Isotope analyses are a powerful tool to distinguish this source from biological production. Additional measurements are needed to determine the isotopic signature of such a source and investigate to which extend photochemical production contributes to the oceanic H2 budget in colder water masses, and also update the current models. However, with an isotopic signature of approximately -250 ‰, or an even more isotopically enriched, such a source would not significantly impact the current models.

Based on their H2 measurements, Moore et al. (2014) suggested a substantial underestimation of oceanic N2 fixation, especially due to high H2 supersaturations measured in the Southern Hemisphere. By using direct measurements of N2 fixation rates, a systematic underestimation by approximately 60 % was also proposed by Großkopf et al. (2012) who suggested a global marine N2 fixation rate of (177 ± 8) Tg N a-1. In order to identify a possible significant mismatch between N2 fixation rates and total marine H2 production, we calculated the climatological global oceanic emissions from the GEMS database using the protocol of Pieterse et al. (2013), and an assumed mean oceanic H2 source of 5 Tg a-1 as given from global budget calculations. The estimated emission rates and distributions in the Atlantic Ocean (Fig. 8) are in line with the calculations of Moore et al. (2014), who reported H2 sea-to-air fluxes mostly in the range of (10 ± 5) mmol, m-2 a-1 and an almost equal distribution between the hemispheres.

Westberry and Siegel (2006) estimated the global nitrogen fixation rate by Trichodesmium blooms by using satellite ocean color data at 42 Tg N a-1 and an additional 20 Tg N a-1 under non–bloom conditions, suggesting that Trichodesmium is likely the dominant organism in the global ocean new nitrogen budget. The good agreement between our measurements of H2 concentrations and δD and the model results from the TM5 model indicate that the oceanic emissions of H2 to the atmosphere are actually well represented in current atmospheric models (Pieterse et al., 2013 and references herein). The proposed underestimate of oceanic N2 fixation and a possible additional H2 release during this process seems already be incorporated in the current atmospheric budgets of H2. Thus, supposing that both an assumed total oceanic H2 source of 5 Tg a-1 to the atmosphere and a total global nitrogen fixation rate of approximately 177 Tg N a-1 are correct, our calculations clearly support the suggestion of Großkopf et al. (2012) that N2 fixers other than Trichodesmium have been severely underestimated in the global picture and that the oceanic release ratio of H2 to fixed N2 clearly needs more attention. Besides Trichodesmium, several other N2-fixing organisms are known for their potential to produce hydrogen (Wilson et al., 2010a; Falcón et al., 2002, 2004; Zehr et al., 2001; Kars et al., 2009; Barz et al., 2010), and even non-N2-fixing organisms might play a role (Lilley et al., 1982).