Presence of nitrous oxide hotspots in the coastal upwelling area off central Chile: an analysis of temporal variability based on ten years of a biogeochemical time series

A fixed station, known as Station 18 (St. 18) at the COPAS center, was sampled regularly at approximately 30 day intervals from July 2002 to August 2012 (figure 1). St. 18 is located over the continental shelf off central Chile at 10 nm (∼18 km) from the shoreline (36°30.80'S; 73°7.75'W) and 92 m depth (http://copas.udec.cl/). Continuous profiles of temperature, salinity, O₂, fluorescence, and photosynthetically active radiation (PAR) were obtained using a CTD to measure conductivity, temperature and depth (CTD model SBE-25). Seawater was sampled using a 10 L Niskin bottle mounted on a 10 bottle rosette. Samples for gases (O₂, N₂O), nutrients and pigments (in this correlative temporal order) were obtained at nine depths distributed between the surface and 92 m depth (2/5, 10, 15, 20, 30, 40, 50, 65 and 80 m depth). Triplicate samples were taken for O₂ (in iodimetric bottles immediately fixed) and for N₂O (in 20 mL gas chromatography (GC) vials). Preservation was carried out through the addition of 50 μL of saturated mercuric chloride (Tilbrook and Karl 1995) and the GC vials were immediately sealed with a butyl-rubber septum and aluminum cap avoiding air bubbles and, then stored in darkness until analysis. From each sample depth syringes were connected directly to the spigot of the Niskin bottles to obtain triplicate nutrient samples (${\rm N}{{{\rm O}}_{3}}^{-},$ ${\rm N}{{{\rm O}}_{2}}^{-},$ ${\rm P}{{{\rm O}}_{4}}^{3-}$ and Si(OH)₄), filtered through a 0.45 μm Uptidisc adapted to the syringe, and then stored at −20 °C, whereas for Chl-a, 100 mL of seawater was filtered (triplicate) onto a glass-fiber filter (GF/F) and the filter was immediately frozen.

Zoom In Zoom Out Reset image size

O₂ was analyzed by the Winkler method using an automatic measurement system. For N₂O analysis, within each GC vial, a headspace was created adding 5 mL of ultrapure helium (He) using a 5 mL gas-tight syringe and N₂O dissolved in the seawater was measured through gas–liquid equilibration in the vial at 40 °C for 15 min under agitation (using a headspace autosampler device, HP Agilent), followed by quantification via GC. N₂O was analyzed in a Varian 3380 GC with an electron capture detector (ECD) at 350 °C, and using a capillary column and injector operated at 60 °C and 250 °C, respectively, and connected to the mentioned autosampler (Farías et al 2009). Five-point calibration curves were constructed using He and 0.1, 0.5, 1 and 100 ppm N₂O standards (Mathison gas mixture). Additionally, for the calibration curve, prepared compressed dry air was used with a 0.320 mole dry fraction of N₂O from NOAA (http://www.esrl.noaa.gov/gmd/ccl/). The ECD detector lineally responded to this concentration range, which was equivalent to a range of dissolved N₂O in seawater of about 7–1500 nmol L⁻¹ (T ∼ 18 °C and S ∼ 35). The analytical error for the N₂O measurements was about 3%. The uncertainty of the measurements was calculated from the standard deviation of the triplicate measurements by depth. Samples with a variation coefficient higher than 10% were not taken into account for the N₂O database. Nutrients were analyzed by standard manual (from 2002 to 2009) or automated (from 2009 to 2012, autoanalyzer SEAL Analytical) colorimetric methods (Grasshoff et al 1983). Chl-a was analyzed by fluorometry (Turner Design AU-10) according to standard procedures (Parsons et al 1984).

In order to interpret the vertical and temporal variation of N₂O, the water column was divided as follows into two layers according to density and O₂ gradient (more details in Farías et al 2009): 1) the mixed and illuminated layer delimited by the base of the mixed layer (ML); 2) the subsurface waters from the base of the ML to the water overlying the sediments (90 m depth). This layer included the oxycline (a layer where O₂ levels widely fluctuated but always remained higher than 4.4 μmol L⁻¹) and bottom water (a layer where levels of O₂ were as low as 4.4 μmol L⁻¹, generally from 65 to 90 m depth).

N₂O solubility was estimated according to Weiss and Price (1980) based on in situ temperature and salinity. Delta or excess of N₂O, a measure to diagnose the apparent production of N₂O in different water masses (Nevison et al 1995, 2003) was calculated as ΔN₂O = [N₂O]_{in situ} − [N₂O]_eq, where [N₂O]_{in situ} is the measured concentration of N₂O, and [N₂O]_eq is the concentration of N₂O in equilibrium with the atmosphere at the time of the last atmospheric contact. Atmospheric values, at the time in which samples were taken, came from the NOAA/ESRL program register of N₂O hemispheric and global monthly means (http://www.esrl.noaa.gov/gmd/hats/combined/N2O.html).

Monthly anomalies of environmental variables, for example N₂O, were estimated as

$$\begin{eqnarray*}&&{\rm Anomaly}=\frac{x{{N}_{2}}O-\;{{{\bar{X}}}_{{{N}_{2O}}}}}{\sigma {{N}_{2}}O}\end{eqnarray*}$$

where, xN₂O is the discrete value at a certain depth and time, and ${{\bar{X}}_{{{N}_{2}}O}}$ is the average value for the whole 2002–2012 period, and σN₂O is the respective standard deviation of the dataset.

N₂O hotspots were defined as exhibiting a ΔN₂O three times higher than the average monthly ΔN₂O anomaly at each depth range, i.e.

$$\begin{eqnarray*}&&\frac{\Delta {{N}_{2}}O}{{{{\bar{X}}}_{\Delta {{N}_{2}}O}}}\gt 3\end{eqnarray*}$$

where, ΔN₂O is the monthly estimated N₂O anomaly, and ${{\bar{X}}_{\Delta {{N}_{2}}O}}$is the average N₂O anomaly for the entire water column.

N₂O flux through the air–sea interface was determined using the following equation, modified by Wanninkhof (1992):

$$\begin{eqnarray*}&&F=kw\;\left( T{}^\circ ,{\rm salinity} \right)\cdot \left( {{C}_{w}}-{{C}_{{\rm eq}}} \right)\end{eqnarray*}$$

where, kw is the transfer velocity from the surface water to the atmosphere, as a function of wind speed, temperature and salinity from the ML, C_w is the mean N₂O concentration in the ML and C_eq is the gas concentration in the ML expected to be in equilibrium with the atmosphere, according to Weis and Price (1980). The ML depth was calculated using a potential density-based criterion of Kara et al (2003).

Gas transfer velocity was calculated according to Wanninkhof (1992) or W92, estimated to be the best parametrization for the study area. Wind speed and direction based on a six hourly register were obtained from a permanent meteorological station located at Carriel Sur (http://www.meteochile.gob.cl/) that is considered to be a coastal station and meets with international standards. These data were compared with those obtained from satellite observations of ocean winds based on QuikSCAT (2002–2009) and ASCAT (2010–2012).

Cumulative alongshore (south–north) wind stress was obtained using a running mean (10-day) wind stress calculated to eliminate the intra- and inter-daily variability (Barth et al 2007). Wind stress was calculated according to Nelson (1977):

$$\begin{eqnarray*}&&{{\tau }_{y}}={{C}_{d}}\cdot {{\rho }_{{\rm air}}}\cdot \left| V \right|\cdot \nu \end{eqnarray*}$$

where C_d is the drag coefficient of wind at 10 m above sea level and corresponds to the value of 0.0013 (Kraus 1972), ρ_air is the air density and corresponds to the value 1.22 kg m⁻³. $\left| V \right|$ is the wind speed in m s⁻¹ and ν is the meridional component of the wind speed in m s⁻¹.

N₂O, O₂, ${\rm N}{{{\rm O}}_{3}}^{-},$ ${\rm N}{{{\rm O}}_{2}}^{-}$ and Chl-a inventories were calculated through numerical integration of data at 1 m increments (linear interpolation) based on at least 4–6 sampled depths per layer. Spearman correlations (Rho) were determined for N₂O, Chl-a, and nutrient inventories estimated in the surface and subsurface layers. The threshold value for statistical significance was set at p < 0.05. Multiple comparison procedures for all pairwise comparisons among seasonal means were made with a t-student test, with a previous application of the Shapiro–Wilk test to check a normally distributed population. The threshold value for statistical significance was set at p < 0.05.

Time series vertical distributions of O₂, Chl-a, ${\rm N}{{{\rm O}}_{2}}^{-}$ and N₂O are shown in figure 2. Chl-a in surface waters fluctuated between 0.1 and 53.15 μg L⁻¹ (mean ± SD = 3.63 ± 5.65) and revealed a marked seasonal cycle with maximum values in spring–summer (upwelling period) and minimum values (<1.00 μg L⁻¹) in wintertime (non-upwelling period) (figure 2(a)). Temporal variability of O₂ stands out as the predominant seasonal signal of all oceanographic variables (figure 2(b)). Each year, a winter oxygenation throughout the water column is followed by a gradual deoxygenation (undersaturated levels) in the oxycline and the bottom layers, commencing in September/October and lasting until April. Each spring–summer, the oxycline and the bottom waters showed a decrease in the O₂ content reaching values as low as 1 μmol L⁻¹, creating a suboxic/anoxic environment in the bottom waters (close to the sediments), and also a marked oxycline in depths as shallow as 20 m (figure 2(b)).

Zoom In Zoom Out Reset image size

${\rm N}{{{\rm O}}_{2}}^{-},$ which is a nutrient with a specific sensitivity to O₂ levels, showed highly variable concentrations, ranging from 0 to 11.7 μmol L⁻¹ (mean ± SD = 0.56 ± 0.99) (figure 2(c)). Extremely low values were detected in the surface layer, and some middle values up to 2.5 μmol L⁻¹ at the oxycline (figure 2(c)); whereas during the summer higher values (up to 11.7 μmol L⁻¹) were observed in the bottom layer (associated with suboxic waters). N₂O varied from 2.92 nmol L⁻¹ or 28.3% saturation (bottom water, 90 m depth) to 492 nmol L⁻¹ or 4849% saturation, having a mean ± SD of 39.4 ± 29.2 nmol L⁻¹ in the oxyclines (20–64 m depth) and a mean ± SD of 37.6 ± 23.3 nmol L⁻¹ in the bottom waters (65–90 m depth) (figure 2(d)). During mid-autumn and winter, N₂O levels were relatively low with occasional undersaturation in the surface layer. Conversely, in spring–summer a large accumulation of N₂O was observed as O₂ decreased at the oxyclines and the bottom layers (figure 2(d)); however in late summer and early autumn N₂O levels as low as 30% saturation were observed in the bottom waters close to the sediments, coinciding with an extremely low O₂ concentration (anoxia), and indicating that N₂O is being consumed in this layer.

Besides the seasonal cycles described for these variables, an inter-annual variability is also evident, particularly for N₂O. Monthly anomalies for Chl-a, O₂, ${\rm N}{{{\rm O}}_{2}}^{-},$ and N₂O are illustrated in figure 3. Inter-annual variability was clearly visible during the occurrence of N₂O hotspots (figure 2(d)), with very high saturation during December 2004 (up to 2442%), January 2006 (up to 1583%), January 2008 (up to 2097%), December 2008 (up to 2066%), January 2010 (up to 1574%) and November 2011 (up to 4849%). In addition, N₂O excess (ΔN₂O), the difference between observed N₂O and the expected N₂O concentration at equilibrium with the atmosphere, was analyzed to eliminate the majority of N₂O variability caused by changes in seawater temperature and salinity (figure 4(a)). The monthly anomalies of ΔN₂O highlighted the presence of N₂O hotspots which contained at least ΔN₂O three times higher than the average monthly ΔN₂O (figure 4(b)).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Oceanographic/biogeochemical profiles obtained during the presence of hotspots are shown in figure 5. In general, the N₂O hotspots developed at the oxycline layer, below the ML until 50 m depth and under variable O₂ conditions but always in a range of hypoxic conditions (higher than 4.4 μmol L⁻¹, except in November 2011) and with some ${\rm N}{{{\rm O}}_{2}}^{-}$ accumulation (from 0.53 to 2.53 μmol L⁻¹). Furthermore, the N₂O hotspots were always immersed in a layer with high ${\rm N}{{{\rm O}}_{3}}^{-}$ and ${\rm HP}{{{\rm O}}_{3}}^{2-}$ content, ranging from 7.94 to 35.2 and from 1.03 to 2.32 μmol L⁻¹, respectively.

Zoom In Zoom Out Reset image size

Table 1 presents the annual mean inventories of N₂O, O₂, Chl-a, and ${\rm N}{{{\rm O}}_{2}}^{-}$ estimated in the surface and subsurface layer from September to March during the sample period. Cumulated surface wind stress for the whole upwelling period is also included. Significant differences were found among mean inventories for all environmental variables estimated for the yearly spring–summer period (p < 0.05) and indicated differences between upwelling seasons. Additionally, table 1 shows that the majority of N₂O was accumulated in the oxyclines, which on average accounts for 72% of the total N₂O inventory in the water column over the study period, contrasting with 21% of the total inventory that accumulates in the bottom layer.

The wind speed and alongshore wind speed ranged from 0.51 to 26.46 m s⁻¹ (figure 6(a)) and −15.47 to 9.36 m s⁻¹, respectively. The meridian component of the wind, which provoked favorable upwelling events, was observed more frequently in spring–summer and it occurred in total for approximately 61.26% of the whole year. The cumulated wind stress (10-day running mean) varied from −1.43 to 0.68 N m⁻² 10 days and clearly showed that the cumulative upwelling-favorable wind stress occurred during the spring–summer time (figure 6(b)). Sea surface temperature (SST in °C), measured at 10 m depth to avoid the effect of solar radiation, presented a similar pattern to the cumulated wind stress, indicating that cold subsurface water is being upwelled to the surface layer (figure 6(c)). N₂O fluxes varied between −10.47 and 260.2 μmol m⁻² d⁻¹ (mean ± SD = 27.88 ± 44.5) (figure 6(d)). Temporal trends in N₂O fluxes also showed a marked seasonal pattern, being lower (influx) during the non-favorable upwelling periods and higher (efflux) during the favorable upwelling periods; this coincided with supersaturated N₂O conditions in the ML, with levels ranging from 71.85 to 1428% saturation over the whole study period. The highest N₂O fluxes coincided with the presence of the N₂O hotspots (figure 6(d)).

Zoom In Zoom Out Reset image size

Table 2 shows Spearman correlations carried out among several environmental variables as cumulative wind stress, SST, Chl-a and inventories of N₂O, ${\rm N}{{{\rm O}}_{2}}^{-}$ and ${\rm N}{{{\rm O}}_{3}}^{-}$ in the surface and subsurface layer. Cumulative wind stress correlated negatively with SST, whereas it correlated positively with Chl-a and N₂O inventories, but no correlations were found between wind stress and nutrient inventories (table 2). These correlations imply that more intensive upwelling favorable winds provoke a rise of cold and nutrient- and gas-rich subsurface water to the surface, fertilizing and allowing the accumulation of phytoplankton biomass in the surface water and accelerating the gas exchange across the air–sea interface (figure 6(d)). However, these nutrients were more rapidly consumed and assimilated by phytoplankton than the rates at which they were supplied at the surface by upwelling events. In addition, N₂O inventories in surface and subsurface layers correlated reasonably well with Chl-a inventories in the surface layer (table 2). Since the N₂O hotspots were only present during upwelling-favorable winds, this pattern would suggest that these structures were controlled by the wind. However, no significant correlation was found between the N₂O inventory at surface layer and SST, suggesting that the dynamics of N₂O was not just controlled by physical processes (table 2). By contrast, the content of N₂O in surface and subsurface layer was related positively to Chl-a level and the NO₂- inventory at surface waters.

Table 2.  Spearman correlation coefficients among cumulative wind stress, N₂O inventories and several biogeochemical variables in the COPAS time series station.

	Cumulative Wind Stress (Nm−2 10 days)		sur N2O (μmol m−2)		sub N2O (μmol m−2)
	Rho	p	Rho	p	Rho	p
SSTa	−0.40	0.00	−0.07	0.45	−0.20	0.04
Chl-ab	0.51	0.00	0.47	0.00	0.24	0.01
sur N2Ob	0.43	0.00	—	—	—	—
sub N2Ob	0.24	0.01	—	—	—	—
sur ${\rm N}{{{\rm O}}_{2}}^{-}$ b	0.00	0.97	0.32	0.00	−0.33	0.00
sub ${\rm N}{{{\rm O}}_{2}}^{-}$ b	−0.03	0.76	−0.03	0.77	0.12	0.20
sur ${\rm N}{{{\rm O}}_{3}}^{-}$ b	0.00	0.92	0.39	0.00	−0.43	0.00
sub ${\rm N}{{{\rm O}}_{3}}^{-}$ b	0.17	0.06	−0.17	0.06	0.23	0.10

^aSST in °C. ^bBiogeochemical variables are estimated as inventories in surface (sur) and subsurface (sub) layer.

Based on the O₂/N₂O emission ratio observed in atmospheric records (Lueker et al 2003), marine N₂O is predominately being sourced from nitrification. Considering correlations between ΔN₂O versus AOU and ΔN₂O versus ${\rm N}{{{\rm O}}_{3}}^{-},$ it is possible to determine if N₂O production is related to O₂ consumption and ${\rm N}{{{\rm O}}_{3}}^{-}$ production, due to OM remineralization coupled with nitrification (Yoshinari 1976, Nevison et al 2003). Figure 7 illustrates the relationship among these parameters and variables with all data (figures 7(a), (b)) a significant correlation of ΔN₂O versus AOU and ΔN₂O versus ${\rm N}{{{\rm O}}_{3}}^{-}$ (respectively Rho = 0.40, p < 0.05 and Rho = 0.34 p < 0.05) is observed; similar results were obtained with data from the upper oxycline (i.e., 20 m depth, respectively, Rho = 0.53 p < 0.05 and Rho = 0.43p < 0.05; figures 7(c), (d)), indicating that N₂O and ${\rm N}{{{\rm O}}_{3}}^{-}$ are being produced in parallel to O₂ consumption by nitrification. However, correlations between these parameters lose statistical significance as the estimates are carried out with data from the deepest layers (i.e., 50 and 80 m depth). In fact, a ${\rm N}{{{\rm O}}_{3}}^{-}$ consumption was observed at O₂ levels as low as ∼5 μmol L⁻¹, which implies that an advected/diffused denitrification signal from the bottom layer or sediment, and/or an in situ denitrification at the oxyclines is/are occurring simultaneously with the nitrification. Indeed, when these relationships (ΔN₂O versus AOU and ΔN₂O versus ${\rm N}{{{\rm O}}_{3}}^{-})$ were estimated from 50 m (respectively Rho = 0.20 p < 0.05 and 0.30 p < 0.05; figures 7(e), (f)) and 80 m depth (respectively, Rho = 0.10 p = 0.36 and Rho = 0.23 p < 0.05; figures 7(g), (h)), they revealed a strong denitrification observable by an intense ${\rm N}{{{\rm O}}_{3}}^{-}$ and N₂O consumption.

Zoom In Zoom Out Reset image size

Regarding the processes responsible for the observed N₂O distribution in the study area, high AAO rates at the oxyclines were observed in association with high ${\rm N}{{{\rm H}}_{4}}^{+}$ regeneration rates (Fernandez and Farías 2012), however at the oxyclines and possibly also at the bottom waters, AAO appeared to be coupled to ${\rm N}{{{\rm O}}_{3}}^{-}$ and ${\rm N}{{{\rm O}}_{2}}^{-}$ reduction to N₂O (partial denitrification), which has been previously reported in the study area (Fernandez and Farías 2012, Galán et al 2014). With respect to the microbes involved in these processes, beside AAO bacteria, AAO archaea also play an important role as ammonia oxidizers at oxyclines in the central Chilean upwelling (Molina et al 2010). Some archaea have a high affinity for ${\rm N}{{{\rm H}}_{4}}^{+}$ (Martens-Habbena et al 2009), efficiently producing N₂O as a by-product (Santoro et al 2011); sulfur-oxidizing γ-proteobacteria are abundant and dominant at the oxyclines in our study area, and additionally, they have the ability to reduce dissimilative ${\rm N}{{{\rm O}}_{3}}^{-}$ and ${\rm N}{{{\rm O}}_{2}}^{-}$ and carry out a process known as chemolithoautotrophic denitrification (Murillo et al 2014), which could be producing N₂O throughout partial denitrification (Galán et al 2014).

Another observed consequence of the upwelling processes was a strong outgassing. In fact, the contribution from coastal upwelling has been estimated as 0.2 ± 0.14 Tg N₂O-N yr^–1, which represents ∼5% of the total ocean source (Nevison et al 2004), despite representing less than 1% of the total surface of the ocean. Most of these coastal upwellings, which encompass eastern boundary systems, have been recognized as significant N₂O emission areas, such as those associated with the Arabian Sea (De Wilde and Helder 1997, Naqvi et al 2005), the eastern South Pacific (Cornejo et al 2007, Charpentier et al 2010) and west Africa, including the Benguela current and Mauritanian upwelling (Wittke et al 2010, Kock et al 2012).

Given the proximity of the N₂O hotspots to surface water (figure 5), they could be shifted closer to the surface by the piston effect of the SW wind. Indeed, air–sea N₂O fluxes vary in magnitude throughout the year, changing from negative to positive values (figure 6(d)); however, for the majority of the year there is a strong efflux of N₂O, indicating that the area behaves as a strong gas source towards the atmosphere. It is important to note that fluxes were estimated with wind data that came from a continental location (close to the coast); this may indicate the occurrence of a retarding or frictional force on the wind that could reduce the wind strength magnitude. In fact, comparison of these data with those obtained from satellites indicates that these data underestimate the N₂O air–sea exchange by about 20%. Taking into consideration that wind is highly variable with bursts of intensity (up to 26 m s⁻¹), the parametrization of W-92 should be the most appropriate for the study area.

During the presence of hotspot structures, N₂O effluxes were up to one order of magnitude higher than the average flux estimated during the entire decade (figure 6(d)), and almost two orders of magnitude higher than the global ocean estimate of about 3 μmol m⁻² d⁻¹ (Nevison et al 1995). Breaking down the variation of N₂O flux across the air–sea interface into decadal, seasonal, and inter-annual variation facilitates a more precise temporal scaling to assess the impact of the gas exchange on the atmosphere. If we consider that the observed dynamic is occurring on Chile's continental shelf (41 105 km²), this coastal area contributes, based on a weighted average (upwelling and non-upwelling representative of 60% and 40% of the year, respectively), 20.7 Gg N₂O-N per year, which is a significant N₂O contribution/source that should be considered in the global balance of atmospheric N₂O. Furthermore, if we take into consideration that the N₂O hotspots may occur more frequently than observed during this study's sampling cycle (at intervals of 30 days or more), this contribution appears to be further underestimated. In fact, when the N₂O efflux data was removed for periods during the N₂O hotspots' presence, the annually weighted average was reduced to 13.5 Gg N₂O-N; therefore, the importance of the area as a source of N₂O could be scaled up. As a result transient events rather than long-term steady-state conditions have been proposed to govern N₂O production in coastal areas subjected to eutrophication and hypoxia (Naqvi et al 2000).

There exists the possibility that N₂O dynamics act as hot moments, which are defined as short periods of time exhibiting disproportionately high reaction rates relative to longer intervening time periods (McClain et al 2003). In this sense, high frequency events (at the weekly and intra-seasonal scales) could be affecting the presence of hot moments. In general, upwelling-favorable winds work with quasi-weekly upwelling pulses, consisting of an alternation in the winds and/or change in wind intensity; with respect to the southern hemisphere, this may vary from southerly and intense (active upwelling) to northern and decreased winds (upwelling relaxation) (Send et al 1987, Rutllant and Montecino 2002). Over a 5- to 10-day time scale this active-relaxing cycle affects both microbial communities and the accumulation of biomass, and primary production rates and microbial community composition (Rutllant and Montecino 2002, Wilkerson et al 2006, Du and Peterson 2009). For example, for effective Chl-a accumulation an optimal window of 3–7 days of relaxed winds was required following an upwelling pulse. Thus, N₂O hot moments may occur during the relaxing phase of the upwelling process, and therefore would require a weekly and intra-seasonal study scale in order to resolve the frequency of these events.

Within the study region there are no clear decadal trends of an increase in phytoplanktonic biomass or an intensification of hypoxic conditions in subsurface waters regarding the analysis of the COPAS time series; however, phenomena such as the cooling of surface waters and intensification of coastal wind patterns have been reposted by Falvey and Garreaud (2009) in the study area. Therefore, we recommend further study on the physical forcing (i.e. ENSO and even global warming) mechanisms that may be affecting biogeochemical processes involved on N₂O production and exchange across the air–sea interface.