Field Evaluation of an Autonomous, Low-Power Eddy Covariance CO2 Flux System for the Marine Environment

Scott D. Miller aAtmospheric Sciences Research Center, University at Albany, State University of New York, Albany, New York

Search for other papers by Scott D. Miller in
Current site
Google Scholar
PubMed
Close
,
Marc Emond bUniversity of New Hampshire, Durham, New Hampshire

Search for other papers by Marc Emond in
Current site
Google Scholar
PubMed
Close
,
Doug Vandemark bUniversity of New Hampshire, Durham, New Hampshire

Search for other papers by Doug Vandemark in
Current site
Google Scholar
PubMed
Close
,
Shawn Shellito bUniversity of New Hampshire, Durham, New Hampshire

Search for other papers by Shawn Shellito in
Current site
Google Scholar
PubMed
Close
,
Jason Covert aAtmospheric Sciences Research Center, University at Albany, State University of New York, Albany, New York

Search for other papers by Jason Covert in
Current site
Google Scholar
PubMed
Close
,
Ivan Bogoev cCampbell Scientific, Inc., Logan, Utah

Search for other papers by Ivan Bogoev in
Current site
Google Scholar
PubMed
Close
, and
Edward Swiatek cCampbell Scientific, Inc., Logan, Utah

Search for other papers by Edward Swiatek in
Current site
Google Scholar
PubMed
Close
Open access

Abstract

Eddy covariance (EC) air–sea CO2 flux measurements have been developed for large research vessels, but have yet to be demonstrated for smaller platforms. Our goal was to design and build a complete EC CO2 flux package suitable for unattended operation on a buoy. Published state-of-the-art techniques that have proven effective on research vessels, such as airstream drying and liquid water rejection, were adapted for a 2-m discus buoy with limited power. Fast-response atmospheric CO2 concentration was measured using both an off-the-shelf (“stock”) gas analyzer (EC155, Campbell Scientific, Inc.) and a prototype gas analyzer (“proto”) with reduced motion-induced error that was designed and built in collaboration with an instrument manufacturer. The system was tested on the University of New Hampshire (UNH) air–sea interaction buoy for 18 days in the Gulf of Maine in October 2020. The data demonstrate the overall robustness of the system. Empirical postprocessing techniques previously used on ship-based measurements to address motion sensitivity of CO2 analyzers were generally not effective for the stock sensor. The proto analyzer markedly outperformed the stock unit and did not require ad hoc motion corrections, yet revealed some remaining artifacts to be addressed in future designs. Additional system refinements to further reduce power demands and increase unattended deployment duration are described.

© 2024 American Meteorological Society. This published article is licensed under the terms of the default AMS reuse license. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Scott D. Miller, smiller@albany.edu

Abstract

Eddy covariance (EC) air–sea CO2 flux measurements have been developed for large research vessels, but have yet to be demonstrated for smaller platforms. Our goal was to design and build a complete EC CO2 flux package suitable for unattended operation on a buoy. Published state-of-the-art techniques that have proven effective on research vessels, such as airstream drying and liquid water rejection, were adapted for a 2-m discus buoy with limited power. Fast-response atmospheric CO2 concentration was measured using both an off-the-shelf (“stock”) gas analyzer (EC155, Campbell Scientific, Inc.) and a prototype gas analyzer (“proto”) with reduced motion-induced error that was designed and built in collaboration with an instrument manufacturer. The system was tested on the University of New Hampshire (UNH) air–sea interaction buoy for 18 days in the Gulf of Maine in October 2020. The data demonstrate the overall robustness of the system. Empirical postprocessing techniques previously used on ship-based measurements to address motion sensitivity of CO2 analyzers were generally not effective for the stock sensor. The proto analyzer markedly outperformed the stock unit and did not require ad hoc motion corrections, yet revealed some remaining artifacts to be addressed in future designs. Additional system refinements to further reduce power demands and increase unattended deployment duration are described.

© 2024 American Meteorological Society. This published article is licensed under the terms of the default AMS reuse license. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Scott D. Miller, smiller@albany.edu

1. Introduction

The micrometeorological eddy covariance (EC) technique can be used to measure turbulent air–sea fluxes with high spatial (tens of square kilometers) and temporal (15–30 min) resolution. Combined with in situ measurements of phenomena such as surface waves, EC fluxes can be used to improve understanding and physics-based models over a broad range of wind and wave conditions. The use of EC for measuring air–sea CO2 exchange has been demonstrated over the last two decades for attended cruise-length research vessel deployments (McGillis et al. 2001; Miller et al. 2010; Blomquist et al. 2014) as well as unattended ship- (Butterworth and Miller 2016a) and shore-based systems (Butterworth and Else 2018). Despite these achievements, technical hurdles have hindered wider adoption of EC CO2 flux measurements on an emerging suite of smaller ocean observing platforms such as buoys (Weller et al. 2012) and Saildrones (Zhang et al. 2019), which offer potential to vastly increase spatial and temporal coverage. Key challenges include the need to integrate the EC flux package onto small, low-power platforms, to develop robust systems for harsh marine conditions, and to address performance limitations of currently available sensor technology.

The EC technique for CO2 flux requires a system that resolves atmospheric CO2 mixing ratio across the range of temporal/spatial scales contributing to the flux. For typical measurement heights and meteorological conditions, this corresponds to 10–30 min for large-scale motions and 1–10 Hz for smaller/faster scales. Of the existing technologies, infrared gas analyzers (IRGAs) have been most frequently deployed at sea. IRGAs were originally designed for terrestrial environments, where surface–atmosphere CO2 fluxes are typically orders of magnitude larger than air–sea CO2 fluxes. To measure small fluxes in the marine environment, IRGAs have been adapted, often in an ad hoc manner, to address measurement sensitivity limitations and artifacts. For example, it is necessary to pretreat the sample airstream to reduce/eliminate temperature and water vapor fluctuations (McGillis et al. 2001; Miller et al. 2010; Blomquist et al. 2014; Landwehr et al. 2014) and the “Webb correction” (Webb et al. 1980). Additional end-user innovations for the marine environment include sample-line heating and active detection and rejection of liquid water in air sampling lines (Butterworth and Miller 2016a). These techniques have been demonstrated for research vessels with access to ample power; additional engineering progress is required to adapt them to small, low-power platforms such as buoys.

An issue that was identified in early attempts to measure air–sea CO2 fluxes using EC is that commercially available IRGAs are sensitive to platform motion, resulting in spurious CO2 variations that contaminate turbulent flux estimates. To address this issue for research vessel measurements, McGillis et al. (2001) periodically sealed off the IRGA sample cell for several minutes and computed the covariance of CO2 and vertical wind to estimate the motion-induced error in the flux. Miller et al. (2010) used a different ad hoc postdeployment data correction approach, whereby multiple linear regression (MLR) was used to determine the relationship between measured CO2 and platform motion (three linear accelerations and three angular rates), and the MLR-predicted CO2 signal was subtracted from the measured CO2. This approach has since been used effectively for other shipboard CO2 flux measurements (Blomquist et al. 2014; Butterworth and Miller 2016a; Landwehr et al. 2018; Dong et al. 2021). However, attempts by the authors to apply this technique to data from a much smaller 2-m discus buoy failed. Compared to large vessels, a discus buoy’s natural motion resides at higher frequencies. Further, the lower anemometer measurement heights characteristic of buoys shifts the flux-relevant temporal scales to higher frequencies. Thus, buoy-based fluxes require high-fidelity fast-response measurements within a higher-frequency regime than for ship-based measurements. Based on these requirements, we pursued fundamental IRGA design modifications geared toward eliminating the motion sensitivity across all frequencies.

Our goal was to design, build, and test a complete EC CO2 flux package suitable for unattended operation on small platforms in the marine environment. State-of-the-art techniques proven effective for robust CO2 flux measurements from ships were adapted for a low-power buoy. A key component of this collaborative effort between academia and an instrument manufacturer was to develop and build a prototype IRGA with reduced motion-induced error (Vandemark et al. 2023). In October 2020, the complete system was deployed in the Gulf of Maine using the University of New Hampshire (UNH) 2-m air–sea interaction buoy. Data collected during the 18-day deployment were used to evaluate the effectiveness of our methods and the ability of the system to measure air–sea CO2 fluxes, and to assess further-needed improvements required for a commercial-ready product.

2. Methods

a. UNH air–sea interaction buoy

The UNH 2-m discus air–sea interaction buoy is shown in Fig. 1. A 1.93-m-tall welded aluminum frame shaped as a tapered square (side length 0.85 m at the bottom and 0.48 m at the top) supported CO2 flux system components. A fixed vane (1.73 m × 0.635 m) attached to one leg of the mounting frame oriented the buoy into the wind direction. The hull of the buoy was accessed via a hatch directly below the mounting frame. A bank of 12-VDC rechargeable batteries (306 amp hours), located within the hull, provided power for all buoy components and the EC CO2 flux system. Solar panels (400 W) attached to the vane and mounting frame were used to charge the battery bank. A digital camera (Hikvision DS-2CD2783G1-IZS 8MP) pointed upwind was mounted on the underside of a plate at the top of the frame.

Fig. 1.
Fig. 1.

University of New Hampshire 2 m discus air–sea interaction buoy. Major components of the eddy covariance air–sea CO2 flux system are labeled.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0076.1

b. EC CO2 flux system

1) 3D wind vector and platform motion

A 3D ultrasonic (sonic) anemometer (R3-50, Gill Instruments, Inc.) was mounted to a plate at the top of the mounting frame such that its measurement volume was 2.73 m above the top of the buoy and 3.14 m above the sea surface (Fig. 1a). A Global Navigation Satellite System/Inertial Navigation System (GX5-45, Microstrain, Inc.) was used to measure platform orientation and motion (Fig. 1b). The GX5-45 was mounted directly below the sonic such that the vertical displacement of the sonic measurement volume relative to the motion sensor was 0.805 m. The meteorological convention defines the sonic anemometer vertical velocity to be positive in the upward direction, while the GX5-45 motion sensor uses an aeronautical convention with the positive vertical axis downward. Therefore, the GX5-45 was mounted “upside-down” so that its default axes were coaligned with those of the sonic (x axis aligned with the vane, y axis toward starboard, and z axis upward).

2) Fast-response carbon dioxide

Carbon dioxide was measured using two Campbell Scientific, Inc. (CSI), closed-path IRGAs. One IRGA was a stock unit (EC155, hereafter referred to as “stock”) and the other IRGA was a modified EC155 (prototype, hereafter referred to as “proto”; Vandemark et al. 2023). The stock and proto IRGAs were plumbed in series, and air was drawn at 8 liters per minute (Lpm) from an inlet near (15.6 cm below and 15 cm behind) the sonic anemometer measurement volume to the IRGAs through 3.5 m of 6.4 mm ID Dekoron tubing (Fig. 2). CSI’s novel “vortex” inlet (Burgon et al. 2016) was used to reduce the risk of fouling of the sample cell source and detector windows due to liquid water ingested into the air sample line. The vortex inlet includes an air sample line and a separate “bypass line” with its intake offset and oriented tangentially to the centerline axis of the sample inlet. The bypass flow generates a swirling motion (i.e., vortex) that propels heavier particles and liquid water outward. Meanwhile, the sample-line inlet, centered at the middle of the vortex, draws air with fewer particles/droplets. A dedicated pump (TD-2N, Brailsford, Inc.) was used to draw air through the bypass line. The sample and bypass lines were exhausted to the atmosphere after exiting the pumps.

Fig. 2.
Fig. 2.

Eddy covariance air–sea CO2 flux system plumbing and flows. Green and gray lines represent sample airflow upstream and downstream of the IRGAs, respectively; orange lines represent bypass airflow used to create the inlet vortex; and red lines represent compressed airflow during sneeze events.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0076.1

The proto IRGA was designed and built specifically for this project to reduce motion-induced CO2 measurement error that has been observed in commercially available IRGAs (McGillis et al. 2001; Miller et al. 2010; Blomquist et al. 2014). We hypothesize that these errors are caused by thermally induced convection within the hermetically sealed infrared detection transducer housing. More specifically, the detector element, which is cooled to maintain a temperature near −40°C, develops temperature gradients within its cavity, which, in the presence of motion, contaminate measurements at the spectral bands used by the analyzer for CO2 and H2O and result in errors in the sensor outputs. The errors change nonlinearly with motion and are not fully compensated by the integrated detector thermoelectric cooler control circuitry, resulting in CO2 errors that scale with the magnitude of platform motion, with magnitudes more than 1 ppmv commonly observed (Miller et al. 2010). Whereas the stock IRGA detector cavity was filled with nitrogen gas, the proto cavity was evacuated to test whether removing the thermal oscillations reduced motion sensitivity. Initial laboratory and dive tank testing confirmed reduced motion sensitivity compared to the stock (Vandemark et al. 2023). The proto was subsequently reinforced and ruggedized prior to installation on the UNH buoy for a field trial in the Gulf of Maine.

3) Dampening of water vapor fluctuations

Previous studies demonstrated the necessity of active removal of water vapor fluctuations from the sample airstream using a Nafion counterflow moisture exchanger, where sample air passes through the inside of a length of water-permeable Nafion tubing with a counterflow of relatively drier air (Fig. 2; Miller et al. 2010; Blomquist et al. 2014; Landwehr et al. 2014). Differences in water vapor potential across the Nafion membrane result in moisture transport, while it remains relatively impermeable to CO2 (Ma et al. 2005). On research ships, the counterflow is typically generated using a dry-air generator supplied with compressed air at roughly 100 psi (Miller et al. 2010). For a buoy system, an onboard compressor is often not feasible due to power limitations. We tested several low-power approaches to plumbing the Nafion moisture exchanger (PD-200T, PermaPure) that do not require compressed air, such as no counterflow, counterflow using an external pump, and “reflux” mode where the downstream flow from the IRGAs provided the counterflow, with a primary goal to dampen high-frequency water vapor variations and a secondary goal to remove moisture from (i.e., dry) the sample airstream. We determined the reflux mode was the most effective for the buoy deployment.

4) Liquid water rejection (the “sneeze”)

A key consideration for closed-path gas IRGAs in a marine environment is to maintain a clean and dry sample cell, as liquid water ingest (e.g., during rain, sea spray, or wave splash events) and/or buildup of salt deposits on the source and detector windows could require IRGA replacement or repair and recalibration. The vortex inlet (described above) was designed to reduce liquid water entering the air sample line; however, since any water ingestion could be catastrophic, we also implemented an active purge capability (the “sneeze”) to expel water that passed beyond the vortex inlet. The sneeze was originally designed for autonomous ship-based EC measurements in the Southern Ocean (Butterworth and Miller 2016a), and was adapted to operate on a low-power buoy for this project.

A custom-made sneeze module was composed of a 1.6 mm plastic spacer that separated two 6.4-mm-diameter stainless steel tube fittings (Swagelok, Inc.). Electrical leads attached to each metal fitting were routed to resistive sensor circuits (MK108, Velleman, Inc.) installed in a watertight enclosure on the underside of the sonic mounting plate (Fig. 1b). One sneeze module was installed in the sample line and another in the bypass line. The resistive sensor circuits were connected to a datalogger (CR6, CSI) and powered by the buoy 12 VDC supply. The output of the circuit was compared to a 5 VDC reference voltage that was continuously monitored. When liquid water in the sample line (or bypass line) bridged the gap created by the plastic spacer, the circuit closed and the resistance and output voltage changed. When liquid water was detected in either line, the datalogger switched relays (CD16AC, CSI) that controlled solenoid valves (7000 series, Parker, Inc.) and caused a reverse flow of compressed air that purged ingested water back out the inlet. A sneeze was triggered when the output voltage dropped from above 4.5 VDC to below 1 VDC, and the sample line was determined to be free of liquid water when the voltage increased from below 1 VDC to above 4.5 VDC. During a sneeze the sample and bypass line pumps were turned off to prevent drawing a vacuum in the IRGA sample cells. Data collected during a 15-min interval when a sneeze occurred were removed from subsequent processing and flux calculations.

An 11 L compressed nitrogen cylinder (AirGas 60 HP) with an attached pressure regulator was mounted inside the buoy hull and used to supply backflush air for the sneeze. Laboratory testing was designed to determine a sneeze pressure and duration that expelled the liquid water while minimizing the volume of cylinder gas required. Sneeze tests were run for pressures of 20, 30, and 40 psi, which required 4.5, 2.25, and 1.25 s, respectively, to purge the line as detected by the circuit. The volume of air required to purge the line varied inversely with the input pressure, with 20 psi requiring 3.75 L, 30 psi requiring 2.6 L, and 40 psi requiring 2.5 L. Based on these tests, the sneeze cylinder pressure on the buoy was regulated to 40 psi and the sneeze duration was set to 2 s for an unscheduled (actual water intrusion) sneeze and 125 ms for scheduled hourly sneezes that were executed prior to start of each 20-min “flux interval” (see below). To conserve sneeze cylinder gas, if more than 5 sneezes were detected within a 1-h period the pumps would shut down for the remainder of that hour. For the 11 L tank initially at 2000 psi, we estimated that roughly 1500 sneezes could be executed. The cylinder pressure was continuously monitored using a pressure transducer (PX309-3KG5V, Omega, Inc.) that was logged by the CR6 datalogger and transmitted to shore.

5) Electronics, data acquisition, and communications

The EC CO2 flux system plumbing and electronics were integrated into a watertight “system box” constructed of welded aluminum (0.52 m × 0.52 m × 0.22 m) and bolted midway up the tapered mounting frame (Figs. 1b and 2). The system box housed the stock and prototype IRGAs and their electronics (EC100, CSI), Nafion moisture exchanger, solenoid valves, sample and counterflow pumps, datalogger (CR6, CSI), relay controller (CD16AC, CSI), and cellular modem (CELL210, CSI). The CR6 was programmed using CRBasic (CSI) to configure sensors, collect sensor data, control and monitor the system, report data to shore, and allow for real-time monitoring and on-the-fly software updates.

The high-rate sensors (sonic, IRGAs, motion sensor) were recorded at 10-Hz sample frequency. The Gill 3D sonic anemometer and Microstrain GX5-45 inertial measurement sensor digital data streams were sampled using the CR6 RS232 digital ports. The stock and proto IRGAs were sampled by the CR6 using CSI’s synchronous device measurement (SDM) digital protocol. The CR6 merged the high-rate sensor data into a single stream. The EC155 IRGA electronics (EC100) applied a 5-Hz low-pass filter to the IRGA outputs to reduce high-frequency noise and facilitate analysis of turbulence spectra (Campbell Scientific 2018). Slower-rate data (pressure, flow, system diagnostics) were recorded at 1 Hz, and the forward-facing camera (Fig. 1a) recorded one image every 5 min that was used to assess ambient conditions in the upwind fetch. All data were written to files on the datalogger microSD card. The cellular modem was used to transfer data and image files to a shore-based server using FTP protocol, and also facilitated real-time connection to the datalogger using LoggerNet software (CSI) for troubleshooting and updating data acquisition code. Data on the shore-based server were used to create plots that were posted to an interactive website dashboard that monitored system status and performance throughout the deployment.

6) Low-power considerations

The flux system was designed to operate in a low-power environment. The battery bank inside the buoy hull provided 306 amp hours of energy storage. The CSI EC155 was chosen for the stock IRGA and as the starting point for the prototype IRGA since it requires roughly half the power (12 W) compared to other commercially available fast-response IRGAs (Campbell Scientific 2018). The use of a compressed gas cylinder for the sneeze backflush air (described above) eliminated the need for an onboard air compressor (Butterworth and Miller 2016a). During “flux mode” with all instruments turned on, the entire CO2 flux system used 28 W, while during “idle mode” with IRGAs and pumps turned off the system used 10 W. To reduce power usage and extend the deployment length, the system was run in flux mode for 25 min each hour and in idle mode for the remaining 35 min, such that the average power consumption was roughly 18 W. To preserve battery health, the battery bank voltage was monitored and datalogger programmed to switch to idle mode in the event that the voltage dropped below 12.54 V (80% of maximum battery voltage).

7) Eddy covariance data processing

The data during the first 5 min after switching to flux mode were disregarded to allow for sensors to warm up and stabilize. The subsequent 20 min of raw 10-Hz data were written to files that were postprocessed to compute air–sea momentum, buoyancy, and CO2 fluxes. The sonic anemometer, motion sensor, and IRGA data were first time shifted to remove sample delays introduced by the plumbing configuration, sensor electronics, signal conditioning, and datalogger processing. The 5-Hz digital filter applied to the IRGA signals by the EC100 electronics introduced a known (800 ms) sample delay. Additional time delays were introduced due to travel time of the airflow from the air inlet near the sonic anemometer to reach the IRGA measurement volumes. The total IRGA delays were estimated pre- and postdeployment by ejecting puffs of compressed air with known CO2 concentration close to the sample air inlet and measuring the delay before the CO2 perturbation was recorded by the IRGAs. The puffs were initiated by solenoid valves controlled by the CR6. The sample delay of the upstream proto IRGA was 1.65 s and the downstream stock IRGA was 1.85 s. In addition, the response time (e-folding) of both IRGAs was 0.13 s, estimated as the time for CO2 concentration to reach 63% of the step change introduced by the puff. After aligning the data streams to remove instrument and tube delays, the flux interval data were trimmed to 15 min for further processing.

The time-synchronized sonic, motion, and IRGA signals were quality controlled prior to calculating statistics, flux covariances, and spectra. For each 15-min flux period, spikes in the u, υ, and w wind components, sonic temperature, platform accelerations, angle rates, roll, pitch, yaw, and CO2 and H2O mixing ratios were identified as individual records outside of an expected range specified with minimum and maximum threshold values. The threshold values were determined empirically (u, υ, w magnitude less than 30 m s−1; 350 < CO2 < 550 ppm; acceleration magnitude < 10g; angle rate magnitude < 1 rad s−1). In addition, spikes in 10-Hz turbulence variables within a 15-min record were identified as points more than a specified threshold number of standard deviations from the record mean (Vickers and Mahrt 1997). The thresholds were determined empirically by inspection of the 10-Hz data: 4.5 for CO2, 7 for u, υ, and w, and 8 for sonic temperature. For each variable, if the ratio of the number of spikes to the total number of records in the 15-min flux interval was less than 0.03, the spikes were replaced using linear interpolation or extrapolation. If any variables had more than 3% missing data, statistics and fluxes for that interval were not computed. The measured wind vector was corrected for the effects of platform motion (Miller et al. 2008) and rotated into a natural coordinate frame prior to calculating fluxes.

3. Results

a. Gulf of Maine deployment

On 3 October 2020, the buoy was deployed in the Gulf of Maine (43.02°N, 70.54°W; 70 m water depth) from the UNH research vessel R/V Gulf Challenger. The mooring location was roughly 400 m from an existing air–sea pCO2 measurement buoy (Vandemark et al. 2011). The EC CO2 flux buoy was moored for 18 days and was recovered on 20 October 2020. The two-way communications with the buoy enabled remote monitoring, troubleshooting, reboots, and datalogger program modifications. For example, during the first few days of the deployment it was determined that the pressure in the nitrogen cylinder was decreasing too quickly. The sneeze duration was subsequently reduced from 2 s to 125 ms such that the nitrogen cylinder lasted for the remainder of the deployment (final tank pressure = 1000 psi). The solar panels and battery bank were able to keep the system powered for the entire deployment, with one close call at approximately 1100 UTC 14 October when the battery bank voltage was within 0.01 V of the 12.54 V threshold for switching to idle mode.

The 18-day deployment included collection of 405 h of data. The mean water temperature (13.9°C) was higher than the mean air temperature (12.9°C, Fig. 3a, blue curve) and decreased from 15.8°C at the beginning of the campaign to 13.5°C at the end (maximum = 16.1°C, minimum = 11.8°C). The air temperature showed more variability than the water temperature, and was higher than water temperature during 117 intervals (29% of the time, Fig. 3a, red curve). The mean wind speed ranged from 0.2 to 13.8 m s−1, with a deployment average of 6.3 m s−1 (Fig. 3b, blue curve). The mean wind direction showed considerable variability (Fig. 3b, gray dots). The mean significant wave height was 0.9 m, with a maximum of 2.7 m (Fig. 3c, blue curve). The dominant wave forcing period, estimated from the dominant peak in the buoy heave spectra, varied between 2.2 and 6.9 s (mean = 4.2 s, Fig. 3c, red curve). The air–sea dpCO2 was positive (suggesting upward gas transfer from the sea surface to the atmosphere) during roughly 90% of the campaign, and 52 intervals had dpCO2 > 50 ppm (Fig. 3d).

Fig. 3.
Fig. 3.

Summary of Gulf of Maine eddy covariance air–sea CO2 flux buoy deployment data collected between 3 and 20 Oct 2020. (a) Air temperature (°C; red) and relative humidity (%; orange) at 2 m above water surface, and sea surface temperature 75 cm below the surface (°C; blue dashed); (b) wind speed (m s−1; blue curve) and direction (deg; gray dots) at 2 m above the sea surface; (c) significant wave height (m; blue) and average wave period identified as zero crossings of the wave signal (s; red); (d) atmospheric CO2 partial pressure at 2 m above sea surface (ppm; red), and seawater pCO2 75 cm below the surface (ppm; blue), measured at a 3-h time step using the UNH CO2 monitoring buoy (NDBC station 44073) located 200 m from the flux buoy.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0076.1

The UNH buoy response to wave forcing occurs at higher frequencies than those typical of research vessels. For comparison, power spectra of platform vertical acceleration, roll, and pitch are compared for the UNH buoy in the Gulf of Maine and for the Research Vessel Ice Breaker Nathaniel B. Palmer (NBP) during a 2013 cruise (Butterworth and Miller 2016b) in the Southern Ocean (Fig. 4). For the buoy, most of the spectral energy was between 0.1 and 1.5 Hz, with the peak amplitude at 0.5 Hz (2-s period) corresponding to the buoy’s natural frequency. A smaller peak between 0.1 and 0.5 Hz likely corresponded to the dominant wave forcing frequency during this period. In comparison, the natural frequency of the research vessel was much lower (0.15 Hz or 7-s period). The magnitude of the vertical acceleration spectral peak was higher and narrower for the ship than for the buoy during the periods chosen. In contrast, the roll and pitch spectra magnitudes were higher and narrower for the buoy than for the ship. The ship roll showed a low-frequency peak around 0.07 Hz (roughly 14 s) that was not seen in the vertical acceleration.

Fig. 4.
Fig. 4.

Frequency-weighted motion spectra for the UNH 2-m discus air–sea interaction buoy (red), and Research Vessel Ice Breaker (RVIB) Nathaniel B. Palmer (NBP; blue). The buoy data were collected during the October 2020 Gulf of Maine deployment (DOY 290.8, wind speed = 8.6 m s−1, significant wave height = 0.8 m, and peak wave period = 3.5 s), and the NBP data were collected at the ship’s bow during a 2013 cruise (NBP 1210) from Puntas Arena, Chile, to McMurdo Station, Antarctica (wind speed = 12.1 m s−1). (a) Vertical acceleration (m2 s−4), (b) roll angle (deg2), and (c) pitch angle (deg2). The vertical acceleration variances (represented by the integral beneath the curves in (a) for the chosen periods were approximately equal. The NBP spectrum also shows several distinct higher-frequency peaks between 0.7 and 5 Hz, likely due to vibration of the mounting hardware.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0076.1

b. Sneeze performance

There were 416 scheduled hourly maintenance sneezes during the campaign. During high wind conditions on 13 October, breaking waves and spray combined with rain to induce water ingest in the sampling lines (Fig. 5a). During this event there were 10 water-induced (real) sneezes following water detection in the bypass line. During the 6-min period shown in Fig. 5d, the water detector voltage dropped below the 1.5 V cutoff voltage four times, each resulting in a sneeze. The first two sneezes were triggered by rapid voltage drops (i.e., instantaneous), while the last two were characterized by gradual voltage drops over a period of roughly 1 min, presumably reflecting a gradual buildup of the liquid water “bridge” between the two metal tube sections that formed the water detector. In all cases, the voltage and the IRGA signals showed rapid recovery to presneeze levels (Figs. 5d,e). The combined vortex inlet and sneeze were effective at keeping the IRGA sample cells free of saltwater during the 18-day campaign. Postdeployment photos of the source and detector windows at either end of the proto IRGA sample cell showed that they remained free of particles and salt deposits (Figs. 5b,c).

Fig. 5.
Fig. 5.

(a) Photo taken from a buoy-mounted camera during high winds at 1345:00 UTC 13 Oct 2020 (day 287.54), showing an upwind breaking wave, sea spray, and water on the camera lens. (b) Photo of the IRGA source window at the end of the 18-day deployment. (c) As in (b), but for detector window. (d) Voltage (mV) measured by the in-line water detector circuit in the sample line (red curve) and bypass line (blue curve) during a 6-min interval when multiple sneezes were triggered (indicated by gray vertical lines). High- and low-voltage limits used to determine if liquid water was ingested are shown as gray dashed curves at 1.0 and 4.5 V. (e) Stock IRGA cell H2O (mmol mol−1, left axis, blue curve) and CO2 (ppm, right axis, red). (f) Nitrogen tank pressure (psi).

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0076.1

c. Prototype IRGA performance

Figure 6 shows an example of the performance of the proto IRGA compared to the stock IRGA for a 15-min period during the Gulf of Maine deployment (1800 UTC 14 October). The mean wind speed was 6 m s−1 and SST − Tair was −2.5°C, corresponding to an expected downward heat flux and a stable atmospheric surface layer. The sea–air pCO2 difference (dpCO2) was small (16 ppm), such that the expected upward CO2 flux was small. A 30-s time series of the IRGA thermoelectric cooler temperature (TEC; or “detector temperature”) is shown in Fig. 6a, where the mean TEC temperature of the stock IRGA (−40°C) and proto IRGA (−27°C) was removed. Note that the lower detector temperature of the stock IRGA is desirable to reduce sensor noise; however, the prototype IRGA could not achieve temperatures as low as the stock IRGA due to a trade-off with the prototype detector that can be remedied in future designs. The stock IRGA TEC (Fig. 6a, blue curve) showed motion-induced temperature oscillations with amplitude of ±0.05°C. In comparison, the proto IRGA TEC showed a dramatic reduction in motion sensitivity (range ∼0.001°C, Fig. 6a, red curve). The power spectral density (Fig. 6b) confirmed that the amplitude of the motion-induced stock IRGA TEC spectral peak near the wave frequency (∼0.5 Hz) was at least four to five orders of magnitude larger than the proto IRGA TEC spectral amplitude.

Fig. 6.
Fig. 6.

Stock (blue) and proto (red) IRGA performance during EC air–sea CO2 flux buoy deployment in the Gulf of Maine at 1800 UTC 14 Oct 2020. The mean wind speed during this 15-min period was 6 m s−1, dpCO2 was 16 ppm, and SST − Tair was −2.5°C. (a) 30-s time series of infrared detector temperature (TEC) with mean removed (stock IRGA mean TEC = −40°C, proto IRGA mean TEC = −27°C). (b) Power spectral density of TEC (°C2 Hz−1). (c) Time series of CO2 mixing ratio (ppm) with 15-min mean removed. (d) Power spectral density of CO2 mixing ratio (ppm2 Hz−1). (e) Instantaneous product of motion-corrected vertical wind and CO2 mixing ratio with means removed (ppm m s−1). (f) Frequency-weighted CO2 flux cospectra (ppm m s−1).

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0076.1

The improved performance of the proto IRGA detector temperature led to improvements in the reported CO2 mixing ratio. The CO2 mixing ratio time series for the stock IRGA (Fig. 6c, blue curve) shows features that appear correlated with the motion-induced oscillations in the TEC temperature (Fig. 6a, blue curve). For both the stock and proto IRGAs, the slope of the CO2 power spectrum follows the theoretical f−5/3 slope between 0.01 and 0.1 Hz (Fig. 6d, gray dashed curve). During this 15-min interval there was additional spectral energy at lower frequencies, presumably associated with slow background changes in atmospheric CO2 concentration. At higher frequencies, the stock IRGA CO2 power spectrum shows peaks at the buoy motion frequencies (0.15 and 0.5 Hz; Fig. 6d, blue curve). In contrast, for the proto IRGA, the spectrum “flattens out” at frequencies above 0.1 Hz, with no discernable spectral peak at 0.15 Hz, and a small residual peak at 0.5 Hz (Fig. 6d, red curve). Above 1 Hz, both the stock and proto IRGAs were flat, reflecting that the amplitude of the CO2 noise was greater than the amplitude of the atmospheric or motion-induced CO2 fluctuations. The increased noise of the proto IRGA CO2 for frequencies above 1 Hz is reflected in both the time series (Fig. 6c, red curve compared to blue curve) and spectrum (Fig. 6d) and attributed to the reduced cooling of the proto IRGA detector compared to the stock IRGA detector, as discussed above.

The motion-correlated features in the stock CO2 signal translated into more distinct structures in the instantaneous wCO2 product time series that were not found in the proto IRGA (Fig. 6e, blue curve compared to red curve). The small dpCO2 (16 ppm) during the interval shown in Fig. 6 suggests that the CO2 flux was expected to be small, corresponding to low expected “real” amplitudes of the wCO2 cospectrum. Choosing such a period facilitates examining motion-induced effects, which are not expected to scale with the true CO2 flux. The motion-induced features shown in Fig. 6e appear as prominent positive and negative peaks in the frequency-weighted wCO2 cospectrum (0.15 and 0.5 Hz; Fig. 6f, blue curve) that align with the peaks in the TEC and CO2 spectra (Figs. 6b,d). During this 15-min interval, the lower-frequency peak due to platform motion effects on the stock IRGA was positive, while the higher-frequency peak was negative. The 0.5-Hz peak was not discernable in the proto IRGA wCO2 cospectrum, while the 0.15-Hz peak was considerably reduced, reinforcing that the proto modifications were effective in reducing the IRGA motion sensitivity.

d. Motion contamination of IRGA CO2 mixing ratio

Ship-based studies have used empirical approaches to attempt to remove the effects of motion on IRGA-reported CO2 (Miller et al. 2010). These approaches assume that correlations between the IRGA CO2 signal and platform motion are entirely artifacts of instrument error rather than reflecting real variations in atmospheric CO2. Following Miller et al. (2010), we attempted to apply multiple linear regression (MLR) to determine a relationship between measured CO2 mixing ratio and eight predictor variables: three measured platform linear accelerations, three measured platform angle rates, the detector TEC temperature, and IRGA cell pressure. Whereas Miller et al. (2010) used only the accelerations and angle rates as predictor variables, here we added the TEC temperature based on the hypothesis that TEC temperature variations are more directly correlated with the IRGA CO2 motion sensitivity (Vandemark et al. 2023), and IRGA cell pressure was added to evaluate whether the onboard pressure correction was adequate.

An example of the performance of the MLR is shown in Fig. 7 for a 15-min interval (0400 UTC 14 October 2020). During this interval, wind speed was 8.4 m s−1, SST minus air temperature was 2.3°C, and dpCO2 was 17 ppm. The stock CO2 time series showed obvious motion-induced signal (Fig. 7a, blue curve) and visual inspection suggests that the MLR captured some of these peaks (Fig. 7a, red curve). The CO2 power spectrum indicates that at the buoy natural frequencies (roughly 0.5 Hz), the MLR captures the amplitude of the motion-induced signal; however, at lower frequencies of motion (0.1–0.5 Hz) the MLR peak amplitudes (Fig. 7b, red curve) were lower by roughly a factor 2 or more, suggesting that substantial residual motion artifacts remained in the stock “corrected” CO2 power spectrum (Figs. 7a,b, green curve). In fact, the wCO2 cospectra during this interval showed a large negative peak at the buoy frequency (Fig. 7c, blue curve) that was captured by the MLR (red curve) and substantially, though not completely, removed from the corrected cospectrum (green curve). In contrast, the MLR was not effective in eliminating the lower-frequency peaks (0.1–0.5 Hz). Thus, the MLR was not effective in eliminating motion contamination of the stock IRGA CO2 signal during this interval.

Fig. 7.
Fig. 7.

Attempt to apply multiple linear regression technique previously used on ship-based data (e.g., Miller et al. 2010) to characterize and remove IRGA CO2 channel sensitivity to buoy motion. The data shown are for a 15-min interval collected on board the UNH buoy in the Gulf of Maine at 0400 UTC 14 Oct 2020. (a) 30-s segment of the 15-min time series of measured stock IRGA CO2 (blue curve). Carbon dioxide computed from multiple linear regression (MLR) between measured CO2 and 8 predictor variables, as described in the text (red curve). “Corrected” CO2 signal (green curve) calculated as difference between the measured and MLR signals (blue curve minus red curve). (b) Fifteen-minute power spectral density of the three CO2 time series for the stock IRGA shown in (a). (c) 15-min wCO2 flux cospectrum between motion-corrected vertical wind and the three CO2 time series shown in (a). (d)–(f) As in (a)–(c), except, prior to calculating the MLR, all signals were bandpass filtered with a pass band of 0.1–2.0 Hz to reduce impacts of noise and turbulence on the MLR. (g)–(i) As in (d)–(f), but for prototype IRGA.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0076.1

The r2 of the stock IRGA MLR shown in Fig. 7a was relatively low (0.19), presumably since the total CO2 signal included a mix of true atmospheric fluctuations (i.e., turbulence) and the effects of noise and motion. Since the motion frequencies were contained in a distinct frequency range, we attempted to improve the MLR approach by bandpass filtering the measured CO2 mixing ratio and predictor variables using a sixth-order Butterworth filter with passband between 0.1 and 2 Hz. The bandpass-filtered approach (referred to as “bandpass MLR”) results are shown in Fig. 7 (middle row). The impact of the bandpass filter was clearly seen in the MLR term with the “energy” at low frequencies (<0.1 Hz, Fig. 7b, red curve) removed from the bandpass MLR (Fig. 7e), and slightly increased MLR signal within the passband. The r2 of the bandpass MLR (0.4) was modestly higher. Despite the improvement of the bandpass MLR, it did not translate to a marked improvement in the wCO2 cospectrum, where the large spikes within the motion frequency range remained (Fig. 7f, green curve).

As shown in Fig. 6, the proto IRGA CO2 was much less sensitive to motion, though some small residual motion-correlated signal appeared in the measured CO2 power spectrum (see Figs. 6d and 7g, red curve). The MLR approach performed poorly for the proto IRGA (r2 = 0.23), presumably due to the small motion-correlated CO2 signal and increased noise of the proto IRGA. Compared to the MLR, the bandpass MLR r2 for the proto was less (0.09, Fig. 7, bottom row). The decreased amplitude of the proto CO2 response to motion was evident in the time series (Fig. 7g, red curve) and the power spectrum (Fig. 7h compared to Figs. 7b,e). The removal of the bandpass MLR CO2 signal had a relatively small impact on the wCO2 cospectrum (Fig. 7i, green curve compared to blue curve), somewhat reducing the amplitude of the cospectral peak near the wave frequency (0.1–0.2 Hz).

e. Averaged spectra and flux cospectra

Averaged spectra and flux cospectra are shown in Fig. 8 for fifty-two 15-min periods when dpCO2 was greater than 50 ppm and wind speed was greater than 5 m s−1. The median measured spectrum of the along-wind and vertical velocity components (Fig. 8a, black and red curves, respectively), showed large motion-induced peaks between 0.2 and 1 Hz that were mostly removed by the platform motion correction (Fig. 8a, blue and magenta curves). A small residual bump appears in the motion corrected vertical velocity spectrum at the buoy natural frequency (0.5 Hz), presumably the result of incomplete motion correction, flow distortion, or interaction between motion and flow distortion. We note that the u and w velocity component spectra slopes were less than the theoretical −5/3 (Figs. 8a–c, gray dashed curve), presumably due to aliasing of higher-frequency noise. The sonic aliasing resulted from the instrument and data acquisition configuration and affects the wind component spectra and momentum and buoyancy flux cospectra. The measured uw momentum flux cospectrum showed a large positive peak due to the platform motion (Fig. 8d, black curve) which is removed by the motion correction (Fig. 8d, red curve). The small nonzero uw cospectrum at high frequency may have been due to the aliasing.

Fig. 8.
Fig. 8.

Median spectra for fifty-two 15-min intervals during the October 2020 Gulf of Maine deployment of the UNH air–sea interaction buoy when dpCO2 was greater than 50 ppm and wind speed was greater than 5 m s−1. (a) Wind vector component power spectra in natural coordinates (m2 s−2 Hz−1; u measured—black, u corrected—blue; w measured—red; w corrected—magenta); f−5/3 slope shown as the gray dashed line. (b) Stock IRGA CO2 mixing ratio frequency-weighted power spectra (ppm2; measured—black; after attempting to remove motion sensitivity using MLR—blue; after attempting to remove motion sensitivity using bandpass MLR—red). (c) As in (b), but for proto IRGA. (d) Measured (black) and motion-corrected (red) frequency-weighted uw cospectra (m2 s−2). (e) Frequency-weighted wCO2 cospectra (ppm m s−1) for stock IRGA (w measured—black; w motion corrected—blue) and prototype IRGA (w measured—red; w motion corrected—magenta). (f) Frequency-weighted wCO2 cospectra (ppm m s−1) using motion-corrected w for stock IRGA (MLR subtracted from CO2 signal—black; bandpass MLR subtracted from CO2 signal—blue) and prototype IRGA (MLR subtracted from CO2 signal—red; bandpass MLR subtracted from CO2 signal—magenta).

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0076.1

Similar to the individual 15-min CO2 spectra shown in Fig. 6d and Figs. 7b, 7e, and 7h, the median stock IRGA CO2 spectra for 52 intervals showed signatures of IRGA motion sensitivity for frequencies between 0.1 and 1 Hz (Fig. 8b). The averaged stock spectra showed distinct motion-induced peaks at 0.3 and 0.5 Hz, and a narrow higher-frequency peak at 1.5 Hz. Applying the MLR reduced the motion-correlated CO2 signal between 0.15 and 0.8 Hz; however, excess energy was added at frequencies below 0.15 Hz (Fig. 8b, blue curve), presumably due to noise/turbulence affecting the MLR. The bandpass MLR was effective at reducing this low-frequency effect, though was no more effective than the MLR in reducing the motion-induced peaks. Compared to the stock IRGA, the averaged proto IRGA CO2 spectra showed little motion-correlated signal, with a relatively small bump in the measured spectrum at 0.5 Hz (Fig. 8c, black curve). The bump was effectively removed by both the MLR and bandpass MLR. Similar to the stock IRGA, the MLR added energy to the power spectrum, while the bandpass MLR was effective at removing the motion-induced peak while not distorting the spectrum across a broad frequency range.

The impact of motion on the wCO2 cospectra is shown in Figs. 8e and 8f. The median raw wCO2 cospectra calculated using the measured wind vector and measured CO2 mixing ratio showed large negative spikes near 0.3 Hz and large positive spikes near 0.5 Hz for both the stock (Fig. 8e, black curve) and proto (red curve), reinforcing that motion-induced errors in w can correlate with motion-induced errors in CO2 and have large impacts on their cospectra. The stock IRGA peak amplitudes were much larger than the proto IRGA. For both IRGAs, the motion correction of the wind vector appeared to eliminate the 0.3-Hz peak, changing the sign of the cospectrum from negative to positive, as expected for positive dpCO2 (Fig. 8e, blue and magenta curves). For the stock IRGA, the 0.5-Hz peak changed sign from positive to negative, while the amplitude remained high. For the proto IRGA, the sharp 0.5-Hz peak appeared to be removed; however, a broader negative “dip” was revealed between 0.3 and 1 Hz. The dip was apparent in the average wCO2 cospectra computed as the median or as the mean, and its cause is revisited in the discussion.

The attempts to remove IRGA motion artifacts using MLR showed mixed results. For the stock IRGA, the MLR reduced but did not eliminate the spurious negative peak at the 0.5 Hz (Fig. 8f, black curve, compared to Fig. 8e, blue curve). Applying the bandpass MLR to the stock IRGA provided no added benefit compared to the MLR (Fig. 8f, blue curve, compared to Fig. 8f, black curve), as the increased noise introduced by the MLR (Fig. 8b, blue curve compared to black curve) was presumably eliminated by calculation of the cospectrum with the vertical wind. Meanwhile, using the proto IRGA CO2 after applying the MLR showed no benefit compared to using the proto IRGA measured CO2 (Fig. 8f, red curve, compared to Fig. 8e, magenta curve). The bandpass MLR caused significant changes to the wCO2 cospectrum between 0.1 and 1 Hz, switching from positive to negative between 0.1 and 0.5 Hz, and from negative to positive between 0.5 and 1 Hz. For the proto, the negligible impact of the MLR and the poor performance of the bandpass MLR reflected the effectiveness of the sensor modifications in reducing motion sensitivity; i.e., the regressions were driven by noise as there was little motion-correlated signal.

The ability of the stock and proto IRGAs to resolve air–sea CO2 fluxes for smaller bulk pCO2 differences across the interface was assessed by computing average wCO2 cospectra for six dpCO2 ranges (Fig. 9). In addition to the measured stock CO2, the MLR stock IRGA CO2 is shown since it showed the “best” performance for the stock (Fig. 9, black curves). For the proto IRGA, only the measured CO2 is shown as the MLR and bandpass MLR provided no clear benefit. The dpCO2 > 50 ppm range (Fig. 9a) corresponds to the curves shown in Fig. 8f (black and red curves) which showed general agreement between the stock MLR and proto cospectra. For smaller dpCO2 ranges there was less agreement. In particular, spurious negative peaks remained in the stock MLR wCO2 cospectra at 0.3 and 0.5 Hz, even after application of MLR, for most of the dpCO2 ranges (0–10, 10–20, 20–30, 30–40, and 40–50 ppm). The 20–30 ppm range also showed a large positive peak near 0.1 Hz, corresponding to swell frequency. The “dip” in the wCO2 cospectra between roughly 0.3–1 Hz is apparent for both stock and proto IRGAs for all dpCO2 ranges.

Fig. 9.
Fig. 9.

Median frequency-weighted wCO2 cospectra (ppm m s−1) during the October 2020 Gulf of Maine deployment of the UNH air–sea interaction buoy. Cospectra were averaged according to air–sea dpCO2 bins: (a) dpCO2 > 50 ppm (fifty-two 15-min intervals); (b) 40 < dpCO2 < 50 ppm (20 intervals); (c) 30 < dpCO2 < 40 ppm (18 intervals); (d) 20 < dpCO2 < 30 ppm (41 intervals); (e) 10 < dpCO2 < 20 ppm (83 intervals); and (f) 0 < dpCO2 < 10 ppm (24 intervals). Measured stock IRGA CO2 (blue), stock IRGA CO2 after applying multiple linear regression (black), and measured prototype IRGA CO2 (red). Only periods with mean wind speed above 5 m s−1 were included in the averages.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0076.1

The wCO2 cospectra were sensitive to heave. Figure 10a shows mean cospectra during low heave conditions, defined as Hs < 0.75 m, dpCO2 > 20 ppm, and wind speed above 5 m s−1. Thirty-two 15-min flux intervals satisfied this criterion (mean dpCO2 = 78 ppm, mean wind speed = 7.6 m s−1, mean Hs = 0.5 m). Similar to the results shown in Fig. 9, the measured stock cospectrum showed large motion-correlated spikes (Fig. 10a, black curve), the stock MLR cospectrum showed much smaller spikes (blue curve), and the proto did not show sharp spikes but did feature the cospectral dip between 0.3 and 1.5 Hz (red curve), discussed above. For high heave conditions (Fig. 10b), defined as Hs > 1 m, dpCO2 > 20 ppm, and wind speed above 5 m s−1 (thirty-four 15-min intervals, mean dpCO2 29 ppm, mean wind speed = 8.4 m s−1, mean Hs = 1.6 m), the stock cospectra (black curve) showed large peaks at the swell frequency (∼0.12 Hz, 8 s) and buoy resonant frequency (0.5 Hz) that were not significantly reduced by the MLR (blue curve). The measured proto cospectrum (red curve) does not exhibit the sharp features found in the stock and stock MLR curves.

Fig. 10.
Fig. 10.

Effect of buoy heave on wCO2 cospectra (ppm m s−1) during the October 2020 Gulf of Maine deployment of the UNH air–sea interaction buoy. Cospectra were averaged according to significant wave height (Hs, m): (a) Hs < 0.7 m (thirty-two 15-min intervals, mean Hs = 0.5 m); (b) Hs > 1 m (34 intervals, mean Hs = 1.6 m). Measured stock IRGA CO2 (black), stock IRGA CO2 after applying multiple linear regression (blue), and measured prototype IRGA CO2 (red). Only periods with mean dpCO2 > 20 ppm and wind speed above 5 m s−1 were included in the averages.

Citation: Journal of Atmospheric and Oceanic Technology 41, 3; 10.1175/JTECH-D-23-0076.1

f. wCO2 covariance

The wCO2 covariance, wC O2 (ppb m s−1), was calculated for the stock and proto IRGAs and tabulated in Table 1. For each of the six ranges of (positive) dpCO2 included in Fig. 9, the median covariance from the proto IRGA was positive (column 3) corresponding to a net upward air–sea CO2 flux. While the sign of the covariance for the proto was as expected, the persistent negative wCO2 cospectral dip had a spurious negative contribution to the covariance (Fig. 9). In addition, required corrections for high-frequency attenuation and sensor separation were not applied, such that the covariances reported are not believed to be reliable estimates of the kinematic CO2 flux. The spurious features in the stock and stock MLR cospectra shown in Fig. 9 indicate that they are also not the true fluxes. We therefore present a qualitative description of the calculated wCO2 covariances.

Table 1.

Covariance between vertical wind and CO2 (wC O2, ppb m s−1) for six air–sea dpCO2 bins during the October 2020 Gulf of Maine deployment of the UNH air–sea interaction buoy. The number of 15-min periods used to compute the statistics is shown in the second column. Only periods with mean wind speed above 5 m s−1 were included, and the vertical wind was corrected for platform motion. The “proto” and “stock” statistics were calculated using CO2 measured by the proto and stock IRGAs, respectively, and “stock MLR” statistics were calculated using stock IRGA CO2 after subtracting the multiple linear regression between CO2 and platform motion, as described in the text. Rightmost columns show the percent of 15-min intervals when the sign of the wCO2 covariance was opposite to that expected for the measured air–sea dpCO2.

Table 1.

The sign of the mean wCO2 covariance computed using the proto IRGA was positive as expected; however, negative 15-min covariances were found in each of the dpCO2 bins (Table 1, column 6). While the presence of some “countergradient” fluxes is expected due to turbulence variability and sampling uncertainty, especially when dpCO2 and air–sea CO2 fluxes are small, the persistent dip in the wCO2 cospectra (Fig. 9) suggests that measurement error was also a factor. For the proto IRGA, the percentage of 15-min intervals for which the sign of the wCO2 covariance was not consistent with the measured dpCO2 varied between 4% and 39% (average 23%). For the stock IRGA, the mean wCO2 covariance (column 4) was negative (countergradient) for five of the six bins, reflecting the degraded performance shown in the cospectra (Fig. 9, blue curves), and the number of 15-min wCO2 covariances with unexpected sign was much higher (range = 46%–94%, average = 77%). The stock MLR performed slightly better than the stock, with two of the six dpCO2 bins showing countergradient mean wCO2 covariances, and the percentage of 15-min countergradient fluxes ranging from 13% to 65% (average = 43%). These results indicate that the motion-induced errors in the stock IRGA were not random errors; instead, they caused a bias in the measured wCO2 covariances.

4. Discussion

The 18-day deployment in the Gulf of Maine demonstrated the system’s capabilities for a moderate range of conditions. The codeployment of a stock IRGA was necessary to evaluate the performance of the proto, and having two closed-path IRGAs on a low-power platform required considerable engineering. Campbell Scientific IRGAs were chosen for their lower power demand, novel vortex inlet, and CSI’s interest to collaborate on the development of the prototype IRGA. To our knowledge, this is the first use of CSI IRGAs for air–sea EC CO2 flux measurements. Additional engineering was required to implement strategies for low-power removal of water vapor fluctuations in the sample stream, the sneeze capability for liquid water rejection, and a sampling schedule that included flux and idle modes to reduce power consumption. These efforts led to collection of 15-min EC flux data for over 400 hourly intervals.

The postprocessing approaches that proved successful in mediating IRGA motion sensitivity for ship-based EC CO2 flux systems were generally not effective for the stock unit on the discus buoy as they showed dependence on conditions such as dpCO2 (Fig. 9) and heave (Fig. 10). In controlled experiments, Vandemark et al. (2023) demonstrated the complexity of the stock IRGA CO2 response to distinct pitch and roll motions. In the Gulf of Maine deployment with mixed linear and angular accelerations, the stock IRGA was found to be more sensitive to heave than to pitch and roll, with error increasing nonlinearly for significant wave heights exceeding 0.75 m (Fig. 10). We suspect that the poor performance of the MLR and bandpass MLR approaches for the stock IRGA were due to the higher-frequency motion of the buoy compared with research vessels (Fig. 4), the higher frequency of flux-carrying eddies for the lower buoy measurement height compared to ships, and potential nonlinear IRGA response to motion. Further, any residual motion sensitivity of the stock IRGA that was not removed in postprocessing can be amplified by residual errors in the vertical wind due to incomplete motion correction or flow distortion.

The degradation of stock IRGA performance for important conditions such as low dpCO2 (Fig. 9) or high winds/seas (Fig. 10) suggests that CO2 fluxes are more uncertain during these regimes where progress in understanding and parameterization of air–sea CO2 exchange are critically needed. We also note that our “high-heave” criterion (Hs > 1 m, Fig. 10b) is relatively low compared to open-ocean seas and swell, such that the divergence between stock and proto performance may become more prominent in larger seas. The uncertainty in the performance of the stock IRGA for a broad range of conditions reinforces the value of improving the sensor rather than more elaborate ad hoc postprocessing correction techniques.

The improved performance of the prototype IRGA is clear (e.g., Fig. 6) and encouraging. Of particular note is that the prototype wCO2 cospectra appear plausible, even for the lowest dpCO2 bin considered (0–10 ppm). Similar to the stock results (Fig. 9), our previous ship-based results indicated noisy/erratic cospectra during low dpCO2 conditions, and dpCO2 thresholds of ∼40 ppm have routinely been used to filter and reject flux data. Rejecting large amounts of data makes deployments less efficient and cost-effective, and degraded IRGA performance near the cutoff limits increases uncertainty in the fluxes and resulting parameterizations. The improved performance of the prototype IRGA suggests that more data of higher quality can be collected in shorter time than with stock models combined with ad hoc corrections.

While the proto IRGA CO2 measurements are markedly less motion sensitive compared to the stock unit, residual performance issues remain. The prototype IRGA was essentially a benchtop unit, not ruggedized for field deployment. The residual dip in the proto wCO2 cospectrum between 0.3 and 1.0 Hz (Fig. 9) requires further investigation. While it may have been associated with prototype sample cell modifications, this feature was also found in the stock IRGA. Either way, it should be addressed in the next design iteration. Further improvements to the overall air–sea CO2 flux system can also be realized. For example, the compressed gas tank used in the sneeze configuration reduced power requirements, but may not be feasible for all platforms and for long deployments. Design modifications are needed to make the water rejection system sustainable. As noted in Vandemark et al. (2023), this prototype’s water vapor measurement improvement was not as dramatic as for CO2. Future improvements in the water vapor channel may result in improved CO2 measurements via better pressure broadening and absorption crosstalk mixing ratio corrections.

5. Conclusions

Progress has been demonstrated toward developing an autonomous EC CO2 flux system suitable for small platforms in the marine environment. Published state-of-the-art techniques proven on research vessels (e.g., airstream drying, liquid water rejection) were adapted for a low-power buoy. The system featured both a stock IRGA and a prototype IRGA, developed in a collaborative effort with an instrument manufacturer, that was designed to reduce motion-induced contamination of the CO2 signal. The system was integrated onto the UNH 2 m discus buoy and deployed for 18 days in the Gulf of Maine in October 2020. The deployment demonstrated the system capabilities during a moderate range of conditions. Data from the stock IRGA demonstrated that empirical postprocessing techniques to account for IRGA motion sensitivity, previously found to be effective for ship-based measurements, were less effective on the small buoy platform, particularly for small values of the sea–air CO2 partial pressure difference, dpCO2, and for higher amounts of heave. Meanwhile, the prototype IRGA showed markedly less motion sensitivity and outperformed the stock IRGA without the need for empirical postprocessing steps. Additional development is required to understand and mitigate the cause of a residual persistent dip found in the prototype IRGA wCO2 cospectrum between 0.3 and 1 Hz, and to further refine the low-power techniques to increase the capability for longer deployments.

Acknowledgments.

We thank John Sicker for discussions and advice on implementing and testing the sneeze electronics, the crew of the UNH R/V Gulf Challenger for support of buoy and mooring deployment and recovery efforts, and editor Gilberto Fochesatto and reviewers Mingxi Yang, Eric Saltzman, and Brent Else for thoughtful comments and suggestions. This work was supported by NSF Ocean Technology and Interdisciplinary Coordination Program Awards 1737328 and 2319150 (SDM) and 1737184 (DV). The Southern Ocean NBP cruise was supported by NSF Office of Polar Programs Award 1043623 (SDM).

Data availability statement.

Study test datasets are available from the authors upon request.

REFERENCES

  • Blomquist, B. W., B. J. Huebert, C. W. Fairall, L. Bariteau, J. B. Edson, J. E. Hare, and W. R. McGillis, 2014: Advances in air–sea CO2 flux measurement by eddy correlation. Bound.-Layer Meteor., 152, 245276, https://doi.org/10.1007/s10546-014-9926-2.

    • Search Google Scholar
    • Export Citation
  • Burgon, R., S. Sargent, and B. Conrad, 2016: Performance evaluation of an innovative sampling system for closed path eddy covariance measurements. 32nd Conf. on Agricultural and Forest Meteorology, Salt Lake City, UT, Amer. Meteor. Soc., 140, https://ams.confex.com/ams/32AgF22BLT3BG/webprogram/Paper295682.html.

  • Butterworth, B. J., and S. D. Miller, 2016a: Automated underway eddy covariance system for air–sea momentum, heat, and CO2 fluxes in the Southern Ocean. J. Atmos. Oceanic Technol., 33, 635652, https://doi.org/10.1175/JTECH-D-15-0156.1.

    • Search Google Scholar
    • Export Citation
  • Butterworth, B. J., and S. D. Miller, 2016b: Air-sea exchange of carbon dioxide in the Southern Ocean and Antarctic marginal ice zone. Geophys. Res. Lett., 43, 72237230, https://doi.org/10.1002/2016GL069581.

    • Search Google Scholar
    • Export Citation
  • Butterworth, B. J., and B. G. T. Else, 2018: Dried, closed-path eddy covariance method for measuring carbon dioxide flux over sea ice. Atmos. Meas. Tech., 11, 60756090, https://doi.org/10.5194/amt-11-6075-2018.

    • Search Google Scholar
    • Export Citation
  • Campbell Scientific, 2018: EC155 CO2 and H2O closed-path gas analyzer. Campbell Scientific Doc., 94 pp., https://s.campbellsci.com/documents/au/manuals/ec155.pdf.

  • Dong, Y., M. Yang, D. C. E. Bakker, V. Kitidis, and T. G. Bell, 2021: Uncertainties in eddy covariance air–sea CO2 flux measurements and implications for gas transfer velocity parameterisations. Atmos. Chem. Phys., 21, 80898110, https://doi.org/10.5194/acp-21-8089-2021.

    • Search Google Scholar
    • Export Citation
  • Landwehr, S., S. D. Miller, M. J. Smith, E. S. Saltzman, and B. Ward, 2014: Analysis of the PKT correction for direct CO2 flux measurements over the ocean. Atmos. Chem. Phys., 14, 33613372, https://doi.org/10.5194/acp-14-3361-2014.

    • Search Google Scholar
    • Export Citation
  • Landwehr, S., S. D. Miller, M. J. Smith, T. G. Bell, E. S. Saltzman, and B. Ward, 2018: Using eddy covariance to measure the dependence of air–sea CO2 exchange rate on friction velocity. Atmos. Chem. Phys., 18, 42974315, https://doi.org/10.5194/acp-18-4297-2018.

    • Search Google Scholar
    • Export Citation
  • Ma, S., M. Odgaard, and E. Skou, 2005: Carbon dioxide permeability of proton exchange membranes for fuel cells. Solid State Ionics, 176, 29232927, https://doi.org/10.1016/j.ssi.2005.09.024.

    • Search Google Scholar
    • Export Citation
  • McGillis, W. R., J. B. Edson, J. E. Hare, and C. W. Fairall, 2001: Direct covariance air-sea CO2 fluxes. J. Geophys. Res., 106, 16 72916 745, https://doi.org/10.1029/2000JC000506.

    • Search Google Scholar
    • Export Citation
  • Miller, S. D., T. S. Hristov, J. B. Edson, and C. A. Friehe, 2008: Platform motion effects on measurements of turbulence and air–sea exchange over the open ocean. J. Atmos. Oceanic Technol., 25, 16831694, https://doi.org/10.1175/2008JTECHO547.1.

    • Search Google Scholar
    • Export Citation
  • Miller, S. D., C. Marandino, and E. S. Saltzman, 2010: Ship-based measurement of air-sea CO2 exchange by eddy covariance. J. Geophys. Res., 115, D02304, https://doi.org/10.1029/2009JD012193.

    • Search Google Scholar
    • Export Citation
  • Vandemark, D., J. E. Salisbury, C. W. Hunt, S. M. Shellito, J. D. Irish, W. R. McGillis, C. L. Sabine, and S. M. Maenner, 2011: Temporal and spatial dynamics of CO2 air-sea flux in the Gulf of Maine. J. Geophys. Res., 116, C01012, https://doi.org/10.1029/2010JC006408.

    • Search Google Scholar
    • Export Citation
  • Vandemark, D., M. Emond, S. D. Miller, S. Shellito, I. Bogoev, and J. M. Covert, 2023: A CO2 and H2O gas analyzer with reduced error due to platform motion. J. Atmos. Oceanic Technol., 40, 845854, https://doi.org/10.1175/JTECH-D-22-0131.1.

    • Search Google Scholar
    • Export Citation
  • Vickers, D., and L. Mahrt, 1997: Quality control and flux sampling problems for tower and aircraft data. J. Atmos. Oceanic Technol., 14, 512526, https://doi.org/10.1175/1520-0426(1997)014<0512:QCAFSP>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Webb, E. K., G. I. Pearman, and R. Leuning, 1980: Correction of flux measurements for density effects due to heat and water vapour transfer. Quart. J. Roy. Meteor. Soc., 106, 85100, https://doi.org/10.1002/qj.49710644707.

    • Search Google Scholar
    • Export Citation
  • Weller, R. A., S. P. Bigorre, J. Lord, J. D. Ware, and J. B. Edson, 2012: A surface mooring for air–sea interaction research in the Gulf Stream. Part I: Mooring design and instrumentation. J. Atmos. Oceanic Technol., 29, 13631376, https://doi.org/10.1175/JTECH-D-12-00060.1.

    • Search Google Scholar
    • Export Citation
  • Zhang, D., and Coauthors, 2019: Comparing air-sea flux measurements from a new unmanned surface vehicle and proven platforms during the SPURS-2 field campaign. Oceanography, 32 (2), 122133, https://doi.org/10.5670/oceanog.2019.220.

    • Search Google Scholar
    • Export Citation
Save
  • Blomquist, B. W., B. J. Huebert, C. W. Fairall, L. Bariteau, J. B. Edson, J. E. Hare, and W. R. McGillis, 2014: Advances in air–sea CO2 flux measurement by eddy correlation. Bound.-Layer Meteor., 152, 245276, https://doi.org/10.1007/s10546-014-9926-2.

    • Search Google Scholar
    • Export Citation
  • Burgon, R., S. Sargent, and B. Conrad, 2016: Performance evaluation of an innovative sampling system for closed path eddy covariance measurements. 32nd Conf. on Agricultural and Forest Meteorology, Salt Lake City, UT, Amer. Meteor. Soc., 140, https://ams.confex.com/ams/32AgF22BLT3BG/webprogram/Paper295682.html.

  • Butterworth, B. J., and S. D. Miller, 2016a: Automated underway eddy covariance system for air–sea momentum, heat, and CO2 fluxes in the Southern Ocean. J. Atmos. Oceanic Technol., 33, 635652, https://doi.org/10.1175/JTECH-D-15-0156.1.

    • Search Google Scholar
    • Export Citation
  • Butterworth, B. J., and S. D. Miller, 2016b: Air-sea exchange of carbon dioxide in the Southern Ocean and Antarctic marginal ice zone. Geophys. Res. Lett., 43, 72237230, https://doi.org/10.1002/2016GL069581.

    • Search Google Scholar
    • Export Citation
  • Butterworth, B. J., and B. G. T. Else, 2018: Dried, closed-path eddy covariance method for measuring carbon dioxide flux over sea ice. Atmos. Meas. Tech., 11, 60756090, https://doi.org/10.5194/amt-11-6075-2018.

    • Search Google Scholar
    • Export Citation
  • Campbell Scientific, 2018: EC155 CO2 and H2O closed-path gas analyzer. Campbell Scientific Doc., 94 pp., https://s.campbellsci.com/documents/au/manuals/ec155.pdf.

  • Dong, Y., M. Yang, D. C. E. Bakker, V. Kitidis, and T. G. Bell, 2021: Uncertainties in eddy covariance air–sea CO2 flux measurements and implications for gas transfer velocity parameterisations. Atmos. Chem. Phys., 21, 80898110, https://doi.org/10.5194/acp-21-8089-2021.

    • Search Google Scholar
    • Export Citation
  • Landwehr, S., S. D. Miller, M. J. Smith, E. S. Saltzman, and B. Ward, 2014: Analysis of the PKT correction for direct CO2 flux measurements over the ocean. Atmos. Chem. Phys., 14, 33613372, https://doi.org/10.5194/acp-14-3361-2014.

    • Search Google Scholar
    • Export Citation
  • Landwehr, S., S. D. Miller, M. J. Smith, T. G. Bell, E. S. Saltzman, and B. Ward, 2018: Using eddy covariance to measure the dependence of air–sea CO2 exchange rate on friction velocity. Atmos. Chem. Phys., 18, 42974315, https://doi.org/10.5194/acp-18-4297-2018.

    • Search Google Scholar
    • Export Citation
  • Ma, S., M. Odgaard, and E. Skou, 2005: Carbon dioxide permeability of proton exchange membranes for fuel cells. Solid State Ionics, 176, 29232927, https://doi.org/10.1016/j.ssi.2005.09.024.

    • Search Google Scholar
    • Export Citation
  • McGillis, W. R., J. B. Edson, J. E. Hare, and C. W. Fairall, 2001: Direct covariance air-sea CO2 fluxes. J. Geophys. Res., 106, 16 72916 745, https://doi.org/10.1029/2000JC000506.

    • Search Google Scholar
    • Export Citation
  • Miller, S. D., T. S. Hristov, J. B. Edson, and C. A. Friehe, 2008: Platform motion effects on measurements of turbulence and air–sea exchange over the open ocean. J. Atmos. Oceanic Technol., 25, 16831694, https://doi.org/10.1175/2008JTECHO547.1.

    • Search Google Scholar
    • Export Citation
  • Miller, S. D., C. Marandino, and E. S. Saltzman, 2010: Ship-based measurement of air-sea CO2 exchange by eddy covariance. J. Geophys. Res., 115, D02304, https://doi.org/10.1029/2009JD012193.

    • Search Google Scholar
    • Export Citation
  • Vandemark, D., J. E. Salisbury, C. W. Hunt, S. M. Shellito, J. D. Irish, W. R. McGillis, C. L. Sabine, and S. M. Maenner, 2011: Temporal and spatial dynamics of CO2 air-sea flux in the Gulf of Maine. J. Geophys. Res., 116, C01012, https://doi.org/10.1029/2010JC006408.

    • Search Google Scholar
    • Export Citation
  • Vandemark, D., M. Emond, S. D. Miller, S. Shellito, I. Bogoev, and J. M. Covert, 2023: A CO2 and H2O gas analyzer with reduced error due to platform motion. J. Atmos. Oceanic Technol., 40, 845854, https://doi.org/10.1175/JTECH-D-22-0131.1.

    • Search Google Scholar
    • Export Citation
  • Vickers, D., and L. Mahrt, 1997: Quality control and flux sampling problems for tower and aircraft data. J. Atmos. Oceanic Technol., 14, 512526, https://doi.org/10.1175/1520-0426(1997)014<0512:QCAFSP>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Webb, E. K., G. I. Pearman, and R. Leuning, 1980: Correction of flux measurements for density effects due to heat and water vapour transfer. Quart. J. Roy. Meteor. Soc., 106, 85100, https://doi.org/10.1002/qj.49710644707.

    • Search Google Scholar
    • Export Citation
  • Weller, R. A., S. P. Bigorre, J. Lord, J. D. Ware, and J. B. Edson, 2012: A surface mooring for air–sea interaction research in the Gulf Stream. Part I: Mooring design and instrumentation. J. Atmos. Oceanic Technol., 29, 13631376, https://doi.org/10.1175/JTECH-D-12-00060.1.

    • Search Google Scholar
    • Export Citation
  • Zhang, D., and Coauthors, 2019: Comparing air-sea flux measurements from a new unmanned surface vehicle and proven platforms during the SPURS-2 field campaign. Oceanography, 32 (2), 122133, https://doi.org/10.5670/oceanog.2019.220.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    University of New Hampshire 2 m discus air–sea interaction buoy. Major components of the eddy covariance air–sea CO2 flux system are labeled.

  • Fig. 2.

    Eddy covariance air–sea CO2 flux system plumbing and flows. Green and gray lines represent sample airflow upstream and downstream of the IRGAs, respectively; orange lines represent bypass airflow used to create the inlet vortex; and red lines represent compressed airflow during sneeze events.

  • Fig. 3.

    Summary of Gulf of Maine eddy covariance air–sea CO2 flux buoy deployment data collected between 3 and 20 Oct 2020. (a) Air temperature (°C; red) and relative humidity (%; orange) at 2 m above water surface, and sea surface temperature 75 cm below the surface (°C; blue dashed); (b) wind speed (m s−1; blue curve) and direction (deg; gray dots) at 2 m above the sea surface; (c) significant wave height (m; blue) and average wave period identified as zero crossings of the wave signal (s; red); (d) atmospheric CO2 partial pressure at 2 m above sea surface (ppm; red), and seawater pCO2 75 cm below the surface (ppm; blue), measured at a 3-h time step using the UNH CO2 monitoring buoy (NDBC station 44073) located 200 m from the flux buoy.

  • Fig. 4.

    Frequency-weighted motion spectra for the UNH 2-m discus air–sea interaction buoy (red), and Research Vessel Ice Breaker (RVIB) Nathaniel B. Palmer (NBP; blue). The buoy data were collected during the October 2020 Gulf of Maine deployment (DOY 290.8, wind speed = 8.6 m s−1, significant wave height = 0.8 m, and peak wave period = 3.5 s), and the NBP data were collected at the ship’s bow during a 2013 cruise (NBP 1210) from Puntas Arena, Chile, to McMurdo Station, Antarctica (wind speed = 12.1 m s−1). (a) Vertical acceleration (m2 s−4), (b) roll angle (deg2), and (c) pitch angle (deg2). The vertical acceleration variances (represented by the integral beneath the curves in (a) for the chosen periods were approximately equal. The NBP spectrum also shows several distinct higher-frequency peaks between 0.7 and 5 Hz, likely due to vibration of the mounting hardware.

  • Fig. 5.

    (a) Photo taken from a buoy-mounted camera during high winds at 1345:00 UTC 13 Oct 2020 (day 287.54), showing an upwind breaking wave, sea spray, and water on the camera lens. (b) Photo of the IRGA source window at the end of the 18-day deployment. (c) As in (b), but for detector window. (d) Voltage (mV) measured by the in-line water detector circuit in the sample line (red curve) and bypass line (blue curve) during a 6-min interval when multiple sneezes were triggered (indicated by gray vertical lines). High- and low-voltage limits used to determine if liquid water was ingested are shown as gray dashed curves at 1.0 and 4.5 V. (e) Stock IRGA cell H2O (mmol mol−1, left axis, blue curve) and CO2 (ppm, right axis, red). (f) Nitrogen tank pressure (psi).

  • Fig. 6.

    Stock (blue) and proto (red) IRGA performance during EC air–sea CO2 flux buoy deployment in the Gulf of Maine at 1800 UTC 14 Oct 2020. The mean wind speed during this 15-min period was 6 m s−1, dpCO2 was 16 ppm, and SST − Tair was −2.5°C. (a) 30-s time series of infrared detector temperature (TEC) with mean removed (stock IRGA mean TEC = −40°C, proto IRGA mean TEC = −27°C). (b) Power spectral density of TEC (°C2 Hz−1). (c) Time series of CO2 mixing ratio (ppm) with 15-min mean removed. (d) Power spectral density of CO2 mixing ratio (ppm2 Hz−1). (e) Instantaneous product of motion-corrected vertical wind and CO2 mixing ratio with means removed (ppm m s−1). (f) Frequency-weighted CO2 flux cospectra (ppm m s−1).

  • Fig. 7.

    Attempt to apply multiple linear regression technique previously used on ship-based data (e.g., Miller et al. 2010) to characterize and remove IRGA CO2 channel sensitivity to buoy motion. The data shown are for a 15-min interval collected on board the UNH buoy in the Gulf of Maine at 0400 UTC 14 Oct 2020. (a) 30-s segment of the 15-min time series of measured stock IRGA CO2 (blue curve). Carbon dioxide computed from multiple linear regression (MLR) between measured CO2 and 8 predictor variables, as described in the text (red curve). “Corrected” CO2 signal (green curve) calculated as difference between the measured and MLR signals (blue curve minus red curve). (b) Fifteen-minute power spectral density of the three CO2 time series for the stock IRGA shown in (a). (c) 15-min wCO2 flux cospectrum between motion-corrected vertical wind and the three CO2 time series shown in (a). (d)–(f) As in (a)–(c), except, prior to calculating the MLR, all signals were bandpass filtered with a pass band of 0.1–2.0 Hz to reduce impacts of noise and turbulence on the MLR. (g)–(i) As in (d)–(f), but for prototype IRGA.

  • Fig. 8.

    Median spectra for fifty-two 15-min intervals during the October 2020 Gulf of Maine deployment of the UNH air–sea interaction buoy when dpCO2 was greater than 50 ppm and wind speed was greater than 5 m s−1. (a) Wind vector component power spectra in natural coordinates (m2 s−2 Hz−1; u measured—black, u corrected—blue; w measured—red; w corrected—magenta); f−5/3 slope shown as the gray dashed line. (b) Stock IRGA CO2 mixing ratio frequency-weighted power spectra (ppm2; measured—black; after attempting to remove motion sensitivity using MLR—blue; after attempting to remove motion sensitivity using bandpass MLR—red). (c) As in (b), but for proto IRGA. (d) Measured (black) and motion-corrected (red) frequency-weighted uw cospectra (m2 s−2). (e) Frequency-weighted wCO2 cospectra (ppm m s−1) for stock IRGA (w measured—black; w motion corrected—blue) and prototype IRGA (w measured—red; w motion corrected—magenta). (f) Frequency-weighted wCO2 cospectra (ppm m s−1) using motion-corrected w for stock IRGA (MLR subtracted from CO2 signal—black; bandpass MLR subtracted from CO2 signal—blue) and prototype IRGA (MLR subtracted from CO2 signal—red; bandpass MLR subtracted from CO2 signal—magenta).

  • Fig. 9.

    Median frequency-weighted wCO2 cospectra (ppm m s−1) during the October 2020 Gulf of Maine deployment of the UNH air–sea interaction buoy. Cospectra were averaged according to air–sea dpCO2 bins: (a) dpCO2 > 50 ppm (fifty-two 15-min intervals); (b) 40 < dpCO2 < 50 ppm (20 intervals); (c) 30 < dpCO2 < 40 ppm (18 intervals); (d) 20 < dpCO2 < 30 ppm (41 intervals); (e) 10 < dpCO2 < 20 ppm (83 intervals); and (f) 0 < dpCO2 < 10 ppm (24 intervals). Measured stock IRGA CO2 (blue), stock IRGA CO2 after applying multiple linear regression (black), and measured prototype IRGA CO2 (red). Only periods with mean wind speed above 5 m s−1 were included in the averages.

  • Fig. 10.

    Effect of buoy heave on wCO2 cospectra (ppm m s−1) during the October 2020 Gulf of Maine deployment of the UNH air–sea interaction buoy. Cospectra were averaged according to significant wave height (Hs, m): (a) Hs < 0.7 m (thirty-two 15-min intervals, mean Hs = 0.5 m); (b) Hs > 1 m (34 intervals, mean Hs = 1.6 m). Measured stock IRGA CO2 (black), stock IRGA CO2 after applying multiple linear regression (blue), and measured prototype IRGA CO2 (red). Only periods with mean dpCO2 > 20 ppm and wind speed above 5 m s−1 were included in the averages.

All Time Past Year Past 30 Days
Abstract Views 72 72 0
Full Text Views 299 299 71
PDF Downloads 158 158 47