Understanding how the ocean absorbs anthropogenic CO2 is critical for predicting climate change. We designed Sniffle, a new autonomous drifting buoy with a floating chamber, to measure gas transfer velocities and air–sea CO2 fluxes with high spatiotemporal resolution. Currently, insufficient in situ data exist to verify gas transfer parameterizations at low wind speeds (<4 m s–1), which leads to underestimation of gas transfer velocities and, therefore, of air–sea CO2 fluxes. The Sniffle is equipped with a sensor to consecutively measure aqueous and atmospheric pCO2 and to monitor increases or decreases of CO2 inside the chamber. During autonomous operation, a complete cycle lasts 40 minutes, with a new cycle initiated after flushing the chamber. The Sniffle can be deployed for up to 15 hours at wind speeds up to 10 m s–1. Floating chambers often overestimate fluxes because they create additional turbulence at the water surface. We correct fluxes by measuring turbulence with two acoustic Doppler velocimeters, one positioned directly under the floating chamber and the other positioned sideways, to compare artificial disturbance caused by the chamber and natural turbulence. The first results of deployment in the North Sea during the summer of 2016 demonstrate that the new drifting buoy is a useful tool that can improve our understanding of gas transfer velocity with in situ measurements. At low and moderate wind speeds and different conditions, the results obtained indicate that the observed tidal basin was acting as a source of atmospheric CO2. Wind speed and turbulence alone could not fully explain the variance in gas transfer velocity. We suggest that other factors like surfactants, rain or tidal current will have an impact on gas transfer parameterizations.
Introduction
Understanding how the ocean absorbs anthropogenic CO2 is critical for predicting climate change. Several factors affect air–sea exchange of CO2 and other factors related to climate change, including friction velocity, wind, bubbles, buoyancy fluxes, fetch, surfactants, rain, capillary and breaking waves, turbulence, and chemical enhancement (Borges et al., 2004a; Wanninkhof et al., 2009). Wind force has a dominant effect on gas transfer (Wanninkhof, 1992; Wanninkhof et al., 2009), and its quadratic and cubic relationships have been used as tools to assess air–sea CO2 fluxes on a global scale (Takahashi et al., 2009). However, the parameterizations lack data in the lower (Johnson, 2010) and upper (Krall and Jähne, 2014) ranges of wind speed. Traditionally, the upper range of wind speed lacks data due to the difficulty of obtaining field measurements during high sea states. At lower range of wind speed, data are lacking because approaches using tracer mass balance require measurements over several hours or days (Nightingale et al., 2000) to estimate a single gas transfer velocity. Over this period, wind speed varies significantly and rarely remains within the low range required for accurate measurement.
As parameterizations lack empirical data at lower wind speeds and 20% of wind speed frequencies across the world are below 4 m s–1, according to monthly wind speed distribution data obtained from the NCEP/NCAR Re-Analysis Project (www.esrl.noaa.gov, last visited May 24, 2016) (Kalnay et al., 1996), significant uncertainties likely exist regarding global CO2 fluxes. Furthermore, gas transfer velocity is often assumed to be near zero at lower wind speeds (Johnson, 2010), yet other factors may force exchange processes, such as buoyancy fluxes, and thus may be important at low sea states. The lack of data, which limits our understanding of gas exchange at low wind speeds, is due to the challenge of measuring gas transfer velocities under low turbulence conditions. In situ instruments, such as buoys, can create both an increase of turbulence due to underwater equipment and chamber structures, such as skirts that penetrate through the water surface, or a decrease of turbulence underneath the chamber, as the water surface is sheltered from the wind force.
The floating chamber (FC) technique has been used in previous studies (Borges et al., 2004a; Tokoro et al., 2007; Vachon et al., 2010) to provide estimates of gas transfer velocity at short time scales (~20 minutes). The FC has been criticized by several researchers because the chamber covers the water surface and eliminates wind stress (Raymond and Cole, 2001) and because interference from the chamber itself results in artificially higher estimates of transfer velocity (Borges et al., 2004b). Tokoro et al. (2007) pointed out that comparison of turbulent conditions inside and outside of the chamber’s perimeter is needed to determine gas exchange velocities quantitatively using the FC technique. Tokoro et al. (2008) measured turbulent kinetic energy (TKE) inside and outside a chamber’s perimeter using a single acoustic Doppler velocimeter (ADV). Their ADV was fixed on the seafloor at a depth of <3 m, but TKE dissipates quickly within a few centimeters from the sea surface, and towing the chamber over the ADV led to uncertainties regarding compensation of the biases. Vachon et al. (2010) compensated for biases resulting from the chamber by measuring near-surface TKE directly inside and outside of the FC’s perimeter using a single ADV that was alternately repositioned. Vachon et al. (2010) investigated a lake and assumed the TKE at each position to be constant for two consecutive measurements. Higher TKE at sea and microbreaking of waves at low sea states require simultaneous measurement of TKE at both positions.
Bluteau et al. (2016) noted that TKE measurements can become error-prone due to motion-induced contamination. They applied two approaches to reduce such contamination, but the complex mathematical computations required led to unsatisfactory results. Inertial motion units (IMUs) are commonly used to avoid motion-induced errors in the aerospace industry; however, they are rarely used in oceanographic studies despite an ability to obtain motion-sensitive measurements from drifting platforms. For example, Kilcher et al. (2017) demonstrated that an IMU reduces motion contamination and improves spectral estimates of ADV measurements.
In this work, we describe a state-of-the-art autonomous drifting buoy that takes in situ measurements of air–sea gas transfer velocities (k) and CO2 fluxes based on the floating chamber. We improved compensation for biases by implementing two ADVs for TKE measurements, one of which is equipped with an IMU. We focused on completely autonomous operation to allow higher spatiotemporal resolution of CO2 fluxes. In this paper, we present the first pCO2 and k values obtained in Jade Bay (German Bight) based on a recent set of CO2 flux measurements taken using the described buoy. The main purpose of the Sniffle is to enable measurements of a large set of CO2 fluxes in different regions, which would ultimately allow for improvement of gas transfer parameterization, particularly at low sea states and in the presence of surfactants.
Materials and Methods
Description of the Sniffle
The Sniffle has an aluminum structure with a length of 1.20 m, a width of 1.20 m, and a height of 2.15 m (Figure 1). The Sniffle drifts freely, and a GPS logger (GT-730FL-S, Canmore, Taiwan) records its position at 10-second intervals. A control unit on top of the Sniffle controls the different cycles of a single flux measurement (described below) and provides the required power (with a lithium-ion battery for completely autonomous measurements for up to 15 hours).
The Sniffle, an autonomous drifting buoy with floating chamber to measure air–sea CO2 fluxes. Upper panel: The Sniffle before deployment with its main components: control unit, floating chamber, Acoustic Doppler velocimeters and CO2 sensor. Lower panel: the Sniffle during drift operation. DOI: https://doi.org/10.1525/elementa.275.f1
The Sniffle, an autonomous drifting buoy with floating chamber to measure air–sea CO2 fluxes. Upper panel: The Sniffle before deployment with its main components: control unit, floating chamber, Acoustic Doppler velocimeters and CO2 sensor. Lower panel: the Sniffle during drift operation. DOI: https://doi.org/10.1525/elementa.275.f1
Gas transfer velocity (k) calculations
The chamber (diameter: 33 cm; surface area: 855 cm2; volume: 6 L) floats in the center of the Sniffle structure (Figure 1) and is tethered loosely at each corner, which allows for free drifting within the structure. The chamber becomes sealed at the water surface and penetrates 4 cm into the water. The chamber is made of stainless steel. We monitored air temperature, air pressure, and humidity inside the floating chamber every 30 seconds with a UNI-T UT330C USB data logger (Table 1). An infrared gas analyzer (IRGA) (SubCtech OceanPackTM, LI-COR LI-840x) was mounted on the bottom of the Sniffle and connected to the chamber via Swagelok tubing (inner diameter: 6.4 mm) in a closed recirculating loop with an in-line moisture trap (dierite) at the inlet of the IRGA. The IRGA records pCO2 every 30 seconds and monitors increases or decreases in CO2 inside the chamber over a period of 15 minutes. The flux is calculated based on the positivity or negativity of the slope (dpCO2/dt), volume (V) and surface (S) of the floating chamber, temperature (T), and gas constant (R) using the following equation:
Manufacturers and specifications of the sensors incorporated into the Sniffle, Sea Surface Scanner (S3) or ashore on the roof of the building of the Institute of Chemistry and Biology of the Marine Environment. DOI: https://doi.org/10.1525/elementa.275.t1
Parameter . | Manufacturer . | Model . | Location . | Range, unit, and resolution . | Sensitivity . | Accuracy . | Sample . |
---|---|---|---|---|---|---|---|
Air temperature | UNI-TREND | UT 330C | Sniffle | –40°C to +80°C | –a | ±0. 5°C | Air inside FC |
Humidity | UNI-TREND | UT 330C | Sniffle | 0% to 100% RH | – | ±3.0% RH | Air inside FC |
Air pressure | UNI-TREND | UT 330C | Sniffle | 750 hPa to 1100 hPa | – | ±3 hPa | Air inside FC |
pCO2 | SubCtech | OceanPackTM, LI-COR LI-840x | Sniffle | 0 to 3000 μatm | 0.01 μatm | <1.5% | Air inside FC Atmosphere Water |
Turbulent kinetic energy (ADV) | Nortek | Vector | Sniffle | – | – | – | Water |
Salinity | VWR | MU 6100 H | S3 | 0.0 to 70.0 | – | ±0.2% | Water |
Water temperature | VWR | MU 6100 H | S3 | –5.0 to 105.0°C | – | ±0.1°C | Water |
Photosynthetic quantum efficiency | Turner Designs | PhytoFlash | S3 | 0.000 to 1.000 | – | – | Water |
Wind speed | ThiesCLIMA | Compact | Ashore | 0.5 m s–1 to 50 m s–1 | 0.1 m s–1 | ±0.5 m s–1 or ±3% | Atmosphere |
Wind direction | ThiesCLIMA | Compact | Ashore | 0° to 360° | 2° | ±5° | Atmosphere |
GPS | Canmore | GT-730FL-S | Sniffle | Latitude and longitude in degree | – | 3 m | Water |
Parameter . | Manufacturer . | Model . | Location . | Range, unit, and resolution . | Sensitivity . | Accuracy . | Sample . |
---|---|---|---|---|---|---|---|
Air temperature | UNI-TREND | UT 330C | Sniffle | –40°C to +80°C | –a | ±0. 5°C | Air inside FC |
Humidity | UNI-TREND | UT 330C | Sniffle | 0% to 100% RH | – | ±3.0% RH | Air inside FC |
Air pressure | UNI-TREND | UT 330C | Sniffle | 750 hPa to 1100 hPa | – | ±3 hPa | Air inside FC |
pCO2 | SubCtech | OceanPackTM, LI-COR LI-840x | Sniffle | 0 to 3000 μatm | 0.01 μatm | <1.5% | Air inside FC Atmosphere Water |
Turbulent kinetic energy (ADV) | Nortek | Vector | Sniffle | – | – | – | Water |
Salinity | VWR | MU 6100 H | S3 | 0.0 to 70.0 | – | ±0.2% | Water |
Water temperature | VWR | MU 6100 H | S3 | –5.0 to 105.0°C | – | ±0.1°C | Water |
Photosynthetic quantum efficiency | Turner Designs | PhytoFlash | S3 | 0.000 to 1.000 | – | – | Water |
Wind speed | ThiesCLIMA | Compact | Ashore | 0.5 m s–1 to 50 m s–1 | 0.1 m s–1 | ±0.5 m s–1 or ±3% | Atmosphere |
Wind direction | ThiesCLIMA | Compact | Ashore | 0° to 360° | 2° | ±5° | Atmosphere |
GPS | Canmore | GT-730FL-S | Sniffle | Latitude and longitude in degree | – | 3 m | Water |
a Not applicable.
The flux measurements were rejected when the slope (dpCO2/dt) was associated with an R2 of less than 0.95. The calibration of the IRGA was checked before and after the sampling campaign with five CO2 standard gases, and the accuracy was better than 1.5%. The RMS noise at 370 ppm with 1-second signal filtering is <1 μatm (threshold for detection), as specified by the manufacturer. During autonomous operation, the chamber is flushed with ambient air prior to each measurement (Figure 2). To flush the chamber with ambient air, four air pumps (NMP 830 KNDC-B, KNF, flow rate = 2.5 L min–1) are integrated into the control unit. To ensure complete flushing, two pumps inject fresh ambient air into the chamber, and the remaining two pumps aspirate air out of the chamber.
Sequence of pCO2 measurements during the Sniffle deployment. Sequence of pCO2 measurements, including (1) aqueous pCO2, (2) the first measurement of atmospheric pCO2, (3) flux inside chamber, (4) the second measurement of atmospheric pCO2, and (5) the second flux measurement. Aqueous pCO2 was averaged after reaching equilibrium across the sensor’s membrane. DOI: https://doi.org/10.1525/elementa.275.f2
Sequence of pCO2 measurements during the Sniffle deployment. Sequence of pCO2 measurements, including (1) aqueous pCO2, (2) the first measurement of atmospheric pCO2, (3) flux inside chamber, (4) the second measurement of atmospheric pCO2, and (5) the second flux measurement. Aqueous pCO2 was averaged after reaching equilibrium across the sensor’s membrane. DOI: https://doi.org/10.1525/elementa.275.f2
To calculate k, aqueous pCO2 (pCO2 of water) is measured at a depth of 1.2 m by a 1–2 μm flat-silicone-membrane-equilibrator integrated into the IRGA as described above. Aqueous pCO2 is measured for 40 minutes (Figure 2). Laboratory tests indicated a half-equilibration time (T50) of ~40 seconds and a T90 of 15 minutes at water temperatures between 16 and 18°C. Atmospheric pCO2 (pCO2 of air) was recorded for 3 minutes while the chamber was flushed with ambient air. To calculate kCO2 the following equation was used:
where K is CO2 solubility defined by Weiss (1974). Then, kCO2 was standardized with a Schmidt number of 660 using the following equation:
where ScCO2 is the CO2 Schmidt number at a given temperature (Wanninkhof, 1992). We used nSc = 2/3 for wind speeds of <3.7 m s–1 and nSc = 1/2 for wind speeds of >3.7 m s–1 (Guérin et al., 2007; Vachon et al., 2010). A single measurement cycle takes 80 minutes (Figure 2) and includes enough data to calculate two values for k660 (i.e., a single k value every 40 minutes).
TKE dissipation rate
To correct for potential biases caused by the floating chamber (Borges et al., 2004b), we integrated two ADVs (Nortek Mhz) to measure TKE using the method described by Vachon et al. (2010). One ADV was positioned 10 cm underneath the floating chamber inside the Sniffle’s structure, and the other was positioned about 55 cm outside the chamber’s perimeter (Figure 1). The two measuring heads of the ADVs are located about 1.1 m apart. The ADVs recorded velocity at a frequency of 32 Hz within the local XYZ coordinate system. The burst interval was set to measure 29.5 minutes every 30 minutes. The nominal velocity range was set at 2 m s–1. The ADVs had a sampling volume of 14.9 mm3. The ADV located inside the chamber’s perimeter was equipped with a Microstrain 3DM-GX3-25-OEM inertial motion sensor (IMU), and triggers the second ADV for synchronized data acquisition. The IMU corrects TKE measurements according to the Sniffle’s movement. We corrected for motion (including the Earth’s rotation) in line with Kilcher et al. (2016) and Kilcher et al. (2017).
Data obtained from the ADVs were processed using the Python dolfyn toolbox (http://lkilcher.github.io/dolfyn/). Data were screened using a 3D phase space algorithm (Goring and Nikora, 2002), and linear interpolation was used to fill the screened points. Less than 5% of the data within each dataset was screened due to poor quality. The energy dissipation rate was calculated by a spectrum analysis of the inertial subrange according to Kolmogorov’s law (Kolmogorov, 1941). We cannot consider the three components of the turbulence to be isotropic close to the surface because the three components were significantly different (χ = 6.621; p < 0.05). For this reason we added up the three dimensions for one estimated TKE. Velocity component spectra, , were calculated using the Fourier transform with a 14.5-minute Hanning window time series. The ensemble average of the spectra were determined with 50% overlap in the time domain and normalized to preserve variance. Then, the TKE was calculated by integrating spectra from 1–2 Hz to avoid interference by wave-induced movement of the ADV at lower frequencies or doppler noise at higher frequencies (Figure 3).
Power spectra from the two Acoustic Doppler Velocimeters measuring inside and outside Sniffle’s perimeter. Power spectra of the north (A, B), east (C, D), and top (E, F) based on the fluctuations of currents measured by primary (A, C, E) and secondary (B, D, F) ADVs on August 1. The primary ADV (equipped with an inertial motion unit) measures inside Sniffle’s perimeter, and the secondary ADV measures outside the perimeter, in client-server communication mode. Red lines represent raw and uncorrected data; blue lines represent processed data, including corrections for the Sniffle’s movement. The frequency range used in our calculations ranged from 1 to 2 Hz. The dashed line is the Doppler noise level, and the peaks around 0.5–0.8 Hz correspond to the wave motion, which was excluded from TKE calculations. DOI: https://doi.org/10.1525/elementa.275.f3
Power spectra from the two Acoustic Doppler Velocimeters measuring inside and outside Sniffle’s perimeter. Power spectra of the north (A, B), east (C, D), and top (E, F) based on the fluctuations of currents measured by primary (A, C, E) and secondary (B, D, F) ADVs on August 1. The primary ADV (equipped with an inertial motion unit) measures inside Sniffle’s perimeter, and the secondary ADV measures outside the perimeter, in client-server communication mode. Red lines represent raw and uncorrected data; blue lines represent processed data, including corrections for the Sniffle’s movement. The frequency range used in our calculations ranged from 1 to 2 Hz. The dashed line is the Doppler noise level, and the peaks around 0.5–0.8 Hz correspond to the wave motion, which was excluded from TKE calculations. DOI: https://doi.org/10.1525/elementa.275.f3
Study area
Field measurements were obtained during three campaigns between 18 June and 1 August 2016 in Jade Bay. The locations of field measurements are shown in Figure 4. Jade Bay is one of the largest tidal basins in the Wadden Sea, which is one of the largest tidal flats in the world and stretches along the coastline of the southern North Sea (Figure 4). The average speed for a deployment is approximately 2 km h–1 in the Jade Bay region. An example of the drifting trajectory can be seen in Figure 4. Depending on the tidal forces, the Sniffle drifted approximately 0.5 km during a single measurement cycle inside the floating chamber (~15 minutes) and 1.3 km during a complete measurement of a single k-value (~40 minutes). Therefore, the spatial resolution depends on the drifting speed, e.g., prevailing wind speed and currents. The region is divided into several tidal basins, each of which is connected to the North Sea. The study area is influenced by semi-diurnal tides with a tidal range of up to 3.8 m (Flemming and Davis, 1994). Therefore, a large volume of water (4 × 108 m3) flows in and out of Jade Bay during each rising and falling tide, according to Götschenberg and Kahlfeld (2008). During high tide, the water surface area is 158 km2, while at low tide, it is 44 km2 (Schückel et al., 2013). Freshwater discharge causes salinity to vary from 26 to 36, and occasionally to 18 in southern Jade Bay (Beck et al., 2013). The growth of phytoplankton in the Wadden Sea is strongly regulated by nutrients and irradiance (Colijn and Cadée, 2003). The first spring blooms in the Wadden Sea can usually be observed in March, and they peak in April or May (Colijn and Cadée, 2003).
Sampling area in the coastal North Sea. The dotted blue line in the inset of the enlarged map is an example of the Sniffle drifting trajectory, which was similar for the three deployments. DOI: https://doi.org/10.1525/elementa.275.f4
Sampling area in the coastal North Sea. The dotted blue line in the inset of the enlarged map is an example of the Sniffle drifting trajectory, which was similar for the three deployments. DOI: https://doi.org/10.1525/elementa.275.f4
Meteorological and hydrographical data
We used wind data from the meteorological station on the rooftop of the Institute of Chemistry and Biology of the Marine Environment building (#101280 Wilhelmshaven). The station is located approximately 2 km from the study site at a height of 27 m. The measurements range from 0.5 to 50 m s–1, with accuracy of <0.1 m s–1 and precision of 3% or 0.5 m s–1 (whichever is greater; Table 1). Wind speed data were referenced at 10 m above the sea surface (U10), in line with Kleemann and Meliss (1993). We determined photosynthetic quantum efficiency (Fv/Fm), salinity, and water temperature based on measurements from active fluorimeter PhytoFlash (Turner Designs, Sunnyvale, CA, USA) and multi-meter MU6100H (VWR, Darmstadt, Germany) taken by the research catamaran Sea Surface Scanner (S3) at a depth of 1 m (Ribas-Ribas et al., 2017) that sailed along the Sniffle trajectory at a distance of less than 500 m.
Statistical analysis
Statistical analysis of the dataset was performed using R (R Development Core Team 2008). Null hypothesis testing was considered significant when p < 0.05. Normal distribution of the data was tested using the Shapiro-Wilk and Anderson-Darling tests. The hypotheses were tested using nonparametric tests as stated in the text. Unless otherwise indicated, the results are presented as mean ± 1 standard deviation. We analyzed correlations between the gas transfer velocity (response variable) and wind speed and TKE (explanatory variables) using a multiple linear regression model. The variables maintained homogeneous variance and normal distribution. The significance of the coefficients within models was analyzed using an analysis of variance, and significant differences from the null model were assessed using the F-statistic. Prior to analysis, multicollinearity between the explanatory variables was assessed using Pearson’s correlation. Variables with a correlation of >0.95 were removed. We compared the regression equations with an analysis of covariance (ANCOVA) and t-tests.
Results
Table 2 summarizes the data obtained from different campaigns in the Jade Bay in July and August 2016. The sea state during these campaigns was usually low, and the mean wind speed at 10 m (U10) ranged from 2.8 m s–1 to 4.7 m s–1. Atmospheric pCO2 varied within a narrow range, with a mean value of 401.6 ± 7.7 μatm (n = 60). Aqueous pCO2 was significantly higher during the campaign on August 1 (939.7 ± 10.1 μatm, n = 240) compared to the average pCO2 during the campaigns on July 18 and 19 (830.8 ± 14.2 μatm, n = 560) (Mann-Whitney test, p < 0.001).
Main variables (mean ± standard error) measured in Jade Bay during the summer of 2016a. DOI: https://doi.org/10.1525/elementa.275.t2
. | Sampling date in 2016 . | ||
---|---|---|---|
Variable (units) . | 18 July (n = 6) . | 19 July (n = 7) . | 1 August (n = 6) . |
FCO2·10–3 (mmol m–2 min–1)b | 21.6 ± 5.4 | 21.1 ± 5.6 | 24.4 ± 11.3 |
ΔpCO2 (μatm)c | 440.0 ± 18.2 | 418.3 ± 5.1 | 542.0 ± 11.6 |
k660 (cm h–1)d | 9.1 ± 2.4 | 9.2 ± 2.4 | 7.9 ± 3.8 |
U10 (m s–1)e | 4.7 ± 0.6 | 2.8 ± 0.7 | 4.1 ± 1.9 |
Salinityf | 35.7 ± 0.1 | 34.3 ± 0.1 | NAg |
Water temperature (°C)f | 19.0 ± 0.2 | 19.7 ± 0.3 | 21.47 ± 0.0 |
FC temperature increase (°C)h | 0.6 ± 0.4 | 2.5 ± 1.9 | 1.0 ± 1.0 |
TKEin (m2 s–2)i | 0.8 ± 5.9 × 10–4 | 1.0 ± 1.9 × 10–4 | 1.8 ± 4.0 × 10–4 |
TKEout (m2 s–2)j | 5.9 ± 21.1 × 10–4 | 5.4 ± 7.0 × 10–4 | 5.0 ± 10.7 × 10–4 |
. | Sampling date in 2016 . | ||
---|---|---|---|
Variable (units) . | 18 July (n = 6) . | 19 July (n = 7) . | 1 August (n = 6) . |
FCO2·10–3 (mmol m–2 min–1)b | 21.6 ± 5.4 | 21.1 ± 5.6 | 24.4 ± 11.3 |
ΔpCO2 (μatm)c | 440.0 ± 18.2 | 418.3 ± 5.1 | 542.0 ± 11.6 |
k660 (cm h–1)d | 9.1 ± 2.4 | 9.2 ± 2.4 | 7.9 ± 3.8 |
U10 (m s–1)e | 4.7 ± 0.6 | 2.8 ± 0.7 | 4.1 ± 1.9 |
Salinityf | 35.7 ± 0.1 | 34.3 ± 0.1 | NAg |
Water temperature (°C)f | 19.0 ± 0.2 | 19.7 ± 0.3 | 21.47 ± 0.0 |
FC temperature increase (°C)h | 0.6 ± 0.4 | 2.5 ± 1.9 | 1.0 ± 1.0 |
TKEin (m2 s–2)i | 0.8 ± 5.9 × 10–4 | 1.0 ± 1.9 × 10–4 | 1.8 ± 4.0 × 10–4 |
TKEout (m2 s–2)j | 5.9 ± 21.1 × 10–4 | 5.4 ± 7.0 × 10–4 | 5.0 ± 10.7 × 10–4 |
a More details given in Table S1.
b CO2 flux across the air-water interface.
c Difference in CO2 partial pressures between the water and atmosphere.
d Gas transfer velocity normalized by a Schmidt number of 660.
e Wind speed at 10 m above the water.
f At 1 m depth.
g Not available.
h Air temperature increase inside floating chamber during a slope measurement (15 minutes).
i TKE measured inside the perimeter of Sniffle’s structure.
j TKE measured outside the perimtere of Sniffle’s structure.
Gas transfer velocity (k)
Fluxes measured in July and August did not significantly differ (Mann-Whitney test, p > 0.05), but the two campaigns in July were slightly lower (average FCO2 = 21.3 ± 5.3 × 10–3 mmol m–2 min–1; n = 13) than those in August (FCO2 = 24.4 ± 11.3 × 10–3 mmol m–2 min–1; n = 6). Five percent of the flux measurements were rejected because the measured slope (dpCO2/dt) was associated with an R2 of <0.95. The average k values were lowest (7.9 ± 3.8 cm h–1) and highest (9.2 ± 2.4 cm h–1) on July 18 and 19, respectively, within the same tidal cycle. Wind speed was always lower than 7 m s–1, and thus no large breaking waves were observed, which can potentially break the seal between the FC and the water surface. We thus deployed the Sniffle in suitable conditions for the FC technique (Kremer et al., 2003), as at higher winds the seal between the FC and water surface potentially breaks due to wave actions.
TKE
TKE was integrated from 1–2 Hz in the power spectra of the three components of the velocities based on a slope of –5/3 (Figure 3) according to Kolmogorov’s law (Kolmogorov, 1941). The power spectra were corrected for the ADVs’ movement (blue line in Figure 3). The spectra show a clear peak around 0.5–0.8 Hz, which represents the wave energy. For the whole spectrum, the correction for the Sniffle’s movement differs by a maximum of 20% compared to the uncorrected spectrum. However, the corrected spectrum aligns with the uncorrected one within the frequency range required for calculating TKE (i.e., 1–2 Hz). Nevertheless, for further calculations, we used the corrected spectrum (Figure 3). The TKE outside the Sniffle’s perimeter (TKEout) ranged from 5.0 × 10–4 to 5.9 × 10–4 m2 s–2 (Table 2). These values are one to two orders of magnitude higher than the TKEout values measured in lakes (Vachon et al., 2010). The TKE inside the structure, located under the floating chamber (TKEin), ranged from 0.8 × 10–4 to 1.8 × 10–4 m2 s–2 (Table 2) and is comparable to the TKEin reported by Vachon et al. (2010). Tokoro et al. (2008) reported TKEin with a wider range covering five orders of magnitude (from 10–6 to 10–1 m2 s–3). Our measurements fit in this range.
Floating chamber method: biases and corrections
We assume a maximum uncertainty of 10% for all parameters (i.e., volume and surface of FC, temperature and salinity of seawater, and pCO2) affecting the calculation of k. We then propagated these errors on our calculations according to the method used by Taylor (1997), resulting in 13.7% variation for kCO2 and 10.8% variation for CO2 fluxes.
The bias caused by unnatural turbulence can be quantified by comparison of TKE measured inside (TKEin) and outside (TKEout) of the Sniffle structure’s perimeter (Tokoro et al., 2008; Vachon et al., 2010). We improved this approach by taking the first TKE measurements inside and outside of the structure’s perimeter simultaneously. To measure the biases induced by the chamber, diversion from the equal relationship between TKEin and TKEout (i.e., TKEout = TKEin; Figure 5) can be used to correct the flux.
TKE outside versus inside the perimeter of the Sniffle structure. The solid line shows equal TKEs (i.e., TKEout = TKEin), and the dashed line shows the least-square linear regression, with corresponding equation and statistics also given. DOI: https://doi.org/10.1525/elementa.275.f5
TKE outside versus inside the perimeter of the Sniffle structure. The solid line shows equal TKEs (i.e., TKEout = TKEin), and the dashed line shows the least-square linear regression, with corresponding equation and statistics also given. DOI: https://doi.org/10.1525/elementa.275.f5
Figure 5 shows the significant relationship between TKEin and TKEout. Least-squared linear regression analysis produced the predictive equation:
where R2 = 0.851, n = 42, and p < 0.0001. The equation indicates that the TKE for measurements taken inside and outside the Sniffle’s perimeter differed significantly (Mann-Whitney test, p < 0.0005). In addition, the diversion between TKEout and TKEin regarding gas transfer velocity (k) indicates that biases were induced by the buoy and FC. We examined these relationships on a daily basis.
Figure 6 represents the relationships of TKE on a log scale. As logarithmic scale was used to transform the data, the slope and intercept should be interpreted differently: k660 will not be 0 when TKE is 0. We found a statistically significant relationship between k660 and both TKEin and TKEout only on 1 August 2016:
Relationships between k660 and TKE on a logarithmic scale for each deployment day. Data are from Sniffle deployments on (A) 18 July, (B) 19 July 19, and (C) 1 August 1 in 2016. Only statistically significant linear regressions are shown, with corresponding equations and statistics. The solid line represents the relationship between TKEin and k660. The dashed line represents the relationship between TKEout and k660. DOI: https://doi.org/10.1525/elementa.275.f6
Relationships between k660 and TKE on a logarithmic scale for each deployment day. Data are from Sniffle deployments on (A) 18 July, (B) 19 July 19, and (C) 1 August 1 in 2016. Only statistically significant linear regressions are shown, with corresponding equations and statistics. The solid line represents the relationship between TKEin and k660. The dashed line represents the relationship between TKEout and k660. DOI: https://doi.org/10.1525/elementa.275.f6
We performed an ANCOVA to compare the slopes and intercepts of the two linear regressions, which were not significantly different (F-test of slopes = 0.703, and F-test of elevations = 0.812). We calculated the overestimation index by dividing k660(TKEin) by k660(TKEout) This index ranged from 0.8 to 1.2 with a mean of 0.98 ± 0.1, which indicates that the FC created no bias in the estimation of gas transfer velocity. However, the relationships were identified based on a relatively small number of observations and should be interpreted with caution. For example, we did not observe significant relationships between the TKEs and k660 on July 18 and 19, indicating that TKE was not the sole determinant of CO2 exchange across the sea surface. We also found no significant correlation between k660 and either U10 (Figure 7A) or the TKEs (Figure 7B), although k660 tended to decrease with higher U10 values (Figure 7A). This tendency is in disagreement with many gas transfer velocities parameterization (Wanninkhof et al., 2009). We suggest that multiple factors control k660, with higher impacts on k660 than wind forces. For example, periods of high wind speeds could coincide with the high tide (and therefore low tidal currents) causing lower k660. In addition, we performed a multiple linear regression between k660, the TKEs, and U10 for all observations. In doing so, we found significant relationships with both TKEin and TKEout (Figure 8):
Relationship between gas transfer velocity and wind speed. Relationship of gas transfer velocity, normalized by a Schmidt number of 660 (k660), to wind speed (U10), normalized at 10 m (A) and to turbulent kinetic energy (B). DOI: https://doi.org/10.1525/elementa.275.f7
Relationship between gas transfer velocity and wind speed. Relationship of gas transfer velocity, normalized by a Schmidt number of 660 (k660), to wind speed (U10), normalized at 10 m (A) and to turbulent kinetic energy (B). DOI: https://doi.org/10.1525/elementa.275.f7
Relationship between observed and predicted gas transfer velocities. Relationship between observed gas transfer velocity, normalized by a Schmidt number of 660 (k660), and the k660 predicted by multiple linear regression model (MLR) for all observations. Black dots represent MLR using U10 and TKEout; white dots represent MLR using U10 and TKEin. The solid and dotted lines show the least-square linear regressions for each MLR; the dashed line represents equal k660 values (i.e., observed k660 = predicted k660). DOI: https://doi.org/10.1525/elementa.275.f8
Relationship between observed and predicted gas transfer velocities. Relationship between observed gas transfer velocity, normalized by a Schmidt number of 660 (k660), and the k660 predicted by multiple linear regression model (MLR) for all observations. Black dots represent MLR using U10 and TKEout; white dots represent MLR using U10 and TKEin. The solid and dotted lines show the least-square linear regressions for each MLR; the dashed line represents equal k660 values (i.e., observed k660 = predicted k660). DOI: https://doi.org/10.1525/elementa.275.f8
The residuals are normally distributed for both relationships and range from –3.82 to 4.24 (mean 0.28 ± 2.36). According to an ANCOVA, the slopes and intercepts are very similar (F-test is 0.235 for the intercept, and 1.531 for the slope), indicating that the buoy and FC caused no biases in the estimation of k660.
TKE and its relationship to wind and tides
Figure 9 shows the correlation between U10 and TKEin as well as the tidal cycle on 19 July 2016. On all sampling days, we deployed the Sniffle two hours before high tide. We observed similar trends for the remaining days of observations (18 July and 1 August 2016). TKEin is at its minimum during high tide with still water (i.e., between the inflowing and outflowing tide), and increases with the outflowing tide. TKEout showed a similar trend (data not shown in Figure 9), as indicated by its relationship to TKEin (Figure 5). U10 reached its maximum at high tide during all observations. Note that the wind data were not taken locally but within a distance of 2 km from the study site, which could lead to a mismatch to values close to the FC and to missing some local variability.
Variations in turbulence, wind and tide for one Sniffle deployment. Variations in turbulence (TKEin), wind speed (U10), and tide height (H) with time during the Sniffle deployment on 19 July 2016. DOI: https://doi.org/10.1525/elementa.275.f9
Variations in turbulence, wind and tide for one Sniffle deployment. Variations in turbulence (TKEin), wind speed (U10), and tide height (H) with time during the Sniffle deployment on 19 July 2016. DOI: https://doi.org/10.1525/elementa.275.f9
Discussion
Evaluation of the floating chamber method
Despite the known biases of the FC technique (Kremer et al., 2003), it is often used to estimate gas fluxes in lakes (Vachon et al., 2010; Duc et al., 2013; Erkkilä et al., 2017) and in estuarine (Borges et al., 2004a) and oceanic waters (Conrad and Seiler, 1988; Calleja et al., 2009). We re-evaluated biases related to artificially introduced turbulence with simultaneous measurements of TKE. Using error propagation, we obtained errors of approximately 14% and 11% for kCO2 and the fluxes, respectively. The FC technique is known to overestimate fluxes by creating artificial turbulence in very still water (Matthews et al., 2003), though such conditions are rare for coastal and oceanic waters. Other studies have shown that the FC technique measures fluxes in lakes with a similar range as those measured by the eddy covariance and boundary layer methods (Podgrajsek et al., 2014; Erkkilä et al., 2017). For example, Podgrajsek et al. (2014) showed similar flux estimates for FC and the eddy covariance technique, but that important differences can occur due to short-term and discontinuous FC measurements with missing episodic flux events and possible diurnal variability. We overcame this discrepancy by automating our FC’s operation using air pumps to exchange air inside the FC with ambient air after completing a measuring cycle. In addition, we followed recommendations of Cole et al. (2010) for proper design of the platform and FC in order to reduce water surface disturbances caused by the FC and therefore minimize biases. We were not able to compare estimates obtained using the Sniffle in parallel with those that could be obtained using other techniques, but the proper design and automated operation allowed us to obtain estimates of CO2 fluxes that fit well into the observed range for European coastal areas using other techniques (see below). Podgrajsek et al. (2014) also pointed out that the FC technique is suitable for local and spatially well constrained flux measurements and, therefore, the FC technique should be seen as supplementary rather than fully comparable to other techniques (e.g., the eddy covariance and dual tracer techniques).
In Figures 5 and 8, we showed that our FC does not create artificial turbulence. The Sniffle is designed to minimize artificial turbulence by, for example, minimizing the surface area of structural components that are in contact with the sea surface. In addition, our FC has no flexible skirt with which to seal the FC with the water surface (Matthews et al., 2003); there has been debate regarding the extent to which a flexible skirt may create artificial micro-turbulence under the FC. For sealing, we used a rigid ring to hold the FC so that its wall penetrates 4 cm into the water. Our seals never broke during deployment. In our study, TKEout was higher than TKEin by 60%, which can be explained by the arm holding the outer ADV and measuring TKEout (Figure 1) or by the design of the buoy and FC, which minimize wind-introduced turbulence close to or under the chamber. However, Figure 8 shows that the multiple dependencies of U10 and TKE on k660 are similar under and outside the chamber. We thus conclude that the arm holding the outer ADV creates no measurable impact on the flux measurements, although artificial turbulence outside the Sniffle was higher under the conditions of this study (e.g., high tidal forces). We extended the approaches used by Tokoro et al. (2008), Vachon et al. (2010), and Gålfalk et al. (2013) and measured TKE at both positions simultaneously. Tokoro et al. (2008) towed the FC over the seabed-fixed ADV (depths of <3 m) and reported no artificial turbulence. However, the difference between TKE measurements and FC was considerably higher (by a factor of 30) in Tokoro et al. (2008) than in Vachon et al. (2010) and our study (in which TKE was measured at a depth of 10 cm). Gålfalk et al. (2013) fixed the ADV to a pier and attached the chamber with a tether line. Vachon et al. (2010) observed higher TKE under their chamber compared to consecutive measurements of TKEout. However, Vachon et al. (2010) used a rectangular chamber approximately four times larger than our dome-like FC; their chamber may have rolled more due to its size and shape and thus created higher TKE compared to our smaller, round FC. Therefore, we reiterate the assertion of Cole et al. (2010) that properly designed FCs are required to obtain comparable results with other techniques.
pCO2 water in Jade Bay in the summer of 2016
This study presents the first data on air–sea CO2 exchange in the German Bight. The observed fluxes varied from 10.5 × 10–3 to 38.0 × 10–3 mmol m–2 min–1. Aqueous pCO2 values in our study (863.5 ± 52.7 μatm) were similar to the values obtained in the Outer Elbe Estuary, which is the transition zone from riverine to the full marine system of the German Bight (950 ± 200 μatm; Amann et al. (2015). Bozec et al. (2005) reported pCO2 values ranging from 290 to 490 μatm from the Southern Bight during the summer. These are lower than our values but are expected for water masses with more oceanic characteristics. Our fluxes fit within the range in the European coastal environment, observed by Borges et al. (2006), and globally, observed by Laruelle et al. (2010) and Chen et al. (2013). For example, air–sea CO2 fluxes observed in the Scheldt Estuary (Zeebrugge, Belgium) after spring bloom, a system that is similar to Jade Bay, ranged from 2.0 × 10–3 to 6.9 × 10–3 mmol m–2 min–1 (Borges and Frankignoulle, 2002). Overall, the comparability of our pCO2 data and fluxes to other studies indicates that Sniffle is a powerful tool for determining ocean carbon budgets.
The relatively high aqueous pCO2 values are probably due to low primary productivity rates, which are indicated by the low Fv/Fm ratio, an indicator of photosynthetic yield, observed on 19 July 2016 (mean Fv/Fm = 0.392 ± 0.047; n = 1093). In other words, the Fv/Fm ratio indicates that the phytoplankton community was stressed, probably due to limited light (high turbidity due to strong tidal forces), limited nutrients (Liebezeit et al., 1994), and zooplankton grazing, which is typical during the summer (Behrends and Liebezeit, 1999). The Sniffle was deployed successfully to assess Jade Bay as a source of atmospheric CO2 during summer. Overall, our study reveals that Jade Bay, a region with one of the most extreme tidal changes in the world, emits 61 ± 21 tons of C d–1 during summer, based on the bay’s area during high tide and assuming that the estimation of emission from the small fraction of the bay we sampled was the same across the whole bay. This value is in line with the emissions reported for coastal regions by Chen et al. (2013) and Laruelle et al. (2010).
Factors regulating gas transfer velocity
Our results show that only 40% of the variance of k660 can be explained by U10 and TKE (Figure 8). This limited degree of explanatory power highlights the complexity of environmental factors governing air–sea CO2 exchange, which have been acknowledged for several decades and recently reviewed by Wanninkhof (2014). Parameterizations for k based solely on wind speed have been used to estimate air–sea CO2 fluxes since the 1980s (Liss and Merlivat, 1986; Wanninkhof, 1992; Nightingale et al., 2000). Such parameterizations are useful for estimating global fluxes, as wind speed is readily available for large spatial scales via remote sensing products. However, on a regional scale, such as coastal environments, parameterizations are prone to error when used to estimate k, especially in low wind regimes (Cole and Caraco, 1998; Borges et al., 2004a) and in the presence of microbreaking (Zappa et al., 2004) and surfactant film (Frew et al., 2004). Additionally, as wind speed is convenient to measure or is readily available, investigations of other factors controlling k on a regional scale have been limited.
Zappa et al. (2007) and Zappa et al. (2003) reported that TKE has a profound effect on k, which is a reason to incorporate TKE measurements into the FC technique (Tokoro et al., 2008). Our TKE measurements allowed us to increase the percentage of variance of k660 that can be explained from 12% with U10 as a single parameter to 40% with TKE. Additional factors affecting the determination of k660 that were not measured directly but were observed visually included the presence of surfactants (Broecker et al., 1978; Brockmann et al., 1982; Frew et al., 2004), tidal current (Zappa et al., 2003) and precipitation (Ho et al., 2004). For example, on 1 August 2016, we deployed the Sniffle during light rainfall and obtained relatively high k660 (average of 11.7 cm h–1; n = 2; Table S1). Rain supports the formation of surfactant films (Wurl et al., 2011), as visually observed after this rainfall; k660 was reduced to an average of 3.7 cm h–1 (n = 2, Table S1), likely due to diffusion-limited gas transfer through the films. After the films disappeared, k660 increased to an average of 8.4 cm h–1 (n = 2, Table S1). Buoyancy is another factor impacting gas fluxes across the air-water interface (MacIntyre et al., 2002; McGillis et al., 2004; Rutgersson et al., 2011), but could not be measured in this study due to technical challenges. Without water surface renewal, gas concentrations at the interface will equilibrate quickly with the overlying atmosphere and therefore reduce gas transfer velocities (MacIntyre et al., 2002). The buoyancy flux could represent up to a 40% increase from nighttime to daytime in areas with low wind and strong insolation (McGillis et al., 2004). The local climate that developed inside the FC might have modified heat flux due to shielding effect of the chamber against wind and solar radiation; however, within the measuring period (~15 minutes) and the small variations of temperature inside the FC (Table S1), we suggest that the effect was minimal.
Although the presented data are limited, they show the potential of the Sniffle to investigate underdetermined factors controlling gas exchange on sufficiently small spatiotemporal scales. Other techniques, such as the eddy covariance or dual tracer techniques, are not able to resolve the scales, justifying the use and further development of the FC technique. Autonomous operation of our Sniffle is advantageous not only from a practical and operational viewpoint but also because the lack of manpower does not affect the Sniffle’s operation; manually operating the chamber with a small boat would create turbulence and increase gas exchange into the chamber.
Conclusions
We successfully operated a newly designed, autonomous floating chamber. The Sniffle was designed for oceanic and coastal conditions and up to 15 hours of operation. The first deployments in a tidal bay in the German Bight showed fluxes similar to those observed previously in other European coastal environments using other techniques.
The known biases of the FC technique were monitored for the first time by simultaneous TKE measurements inside and outside of the Sniffle’s perimeter. Based on our results, we conclude that correction for artificially induced turbulence is not required under the circumstances of this study, probably due to strong tidal forces or free float of the chamber. Our results further indicate that wind speed and turbulence explain only 40% of the variance in gas transfer velocity (k), from which we conclude that, at least on a regional scale, other environmental factors, such as surfactant films, precipitation and currents, are important for accurately predicting air–sea gas exchange.
The FC technique is the only technique designed for the small spatiotemporal scales required for a better understanding of the factors determining air–sea gas exchange. Autonomous and properly designed FCs that are able to monitor the creation of artificial turbulence will allow for good prediction of gas transfer velocities and therefore are useful tools for studying marine biogeochemistry and climate change.
Data Accessibility Statement
Data from this study are available at https://doi.org/10.1594/PANGAEA.873428.
Acknowledgments
We thank the Scientific Committee for Oceanic Research (SCOR) for supporting the working group #139 that initiated this special feature. We thank the captain for operating the ship. We also thank our colleagues at the ICBM workshop for constructing the mechanical framework of the Sniffle and the logistic support for sampling. Further, we thank Frank Hillmann for obtaining meteorological data and maintaining the weather station, Maike Ladehoff for Picture 1b, and Daniel López for statistical help.
Funding information
This work was supported by the ERC project PASSME [grant number GA336408].
Competing interests
The authors have no competing interests to declare.
Author contributions
Contributed to conception and design: MR-R, OW
Contributed to acquisition of data: MR-R, OW
Contributed to analysis and interpretation of data: MR-R, OW, LFK
Drafted the article: MR-R, OW
Revised the article: MR-R, OW, LFK
Approved the submitted version for publication: MR-R, OW, LFK