A high‐resolution submersible oxygen optode system for aquatic eddy covariance

The aquatic eddy covariance technique is increasingly used to determine oxygen (O2) fluxes over benthic ecosystems. The technique uses O2 measuring systems that have a high temporal and numerical resolution. In this study, we performed a series of lab and field tests to assess a new optical submersible O2 meter designed for aquatic eddy covariance measurements and equipped with an existing ultra‐high speed optical fiber sensor. The meter has a 16‐bit digital‐to‐analog‐signal conversion that produces a 0–5 V output at a rate up to 40 Hz. The device was paired with an acoustic Doppler velocimeter. The combined meter and fiber‐optic O2 sensor's response time was significantly faster in O2‐undersaturated water compared to in O2‐supersaturated water (0.087 vs. 0.12 s), but still sufficiently fast for aquatic eddy covariance measurements. The O2 optode signal was not sensitive to variations in water flow or light exposure. However, the response time was affected by the direction of the flow. When the sensor tip was exposed to a flow from the back rather than the front, the response time increased by 37%. The meter's internal signal processing time was determined to be ~ 0.05 s, a delay that can be corrected for during postprocessing. In order for the built‐in temperature correction to be accurate, the meter should always be submerged with the fiber‐optic sensor. In multiple 21–47 h field tests, the system recorded consistently high‐quality, low‐noise O2 flux data. Overall, the new meter is a powerful option for collecting robust aquatic eddy covariance data.

O 2 flux measurements at the sediment-water interface are typically used to quantify benthic ecosystem health and as a proxy for carbon cycling in aquatic environments such as seagrass meadows, oyster reefs, and mudflats (Gazeau et al. 2005;Glud 2008;Eyre et al. 2011;Berg et al. 2022). However, representative in situ fluxes, determined under naturally varying environmental conditions, are difficult to obtain due to various limitations of standard flux methodologies (Cook et al. 2007;Glud 2008). Benthic flux chambers separate an enclosed area from the surrounding environment, which excludes in situ hydrodynamics and disturbs in situ light conditions (Tengberg et al. 2004;Amo-Seco et al. 2021;Polsenaere et al. 2021). In situ microelectrode profiling and subsequent profile interpretations exclude the effects of local macrofauna (Archer and Devol 1992;Glud 2008) and may produce artifacts in permeable sediments due to interactions with bottom flows (Huettel and Gust 1993). Laboratory sediment core incubations introduce additional artifacts by removing the sample from the natural environment (Khalil et al. 2013).
Over the last two decades, the aquatic eddy covariance (AEC) technique (Berg et al. 2003) has improved the quality of benthic O 2 flux measurements for many ecosystems, including seagrass meadows (Lee et al. 2017;Berger et al. 2020;Koopmans et al. 2020), oyster and mussel reefs (Attard et al. 2019b;Volaric et al. 2020), macroalgal and Maerl beds (Attard et al. 2019a;Polsenaere et al. 2021), permeable sands (Chipman et al. 2016;Merikhi et al. 2021), freshwater systems (McGinnis et al. 2008;Koopmans and Berg 2015), and coral reefs (Long et al. 2013;de Froe et al. 2019). It has also recently been used for upside-down measurements of gas exchange at the air-water interface Long and Nicholson 2018;. The technique measures in situ O 2 fluxes without altering light and flow conditions or excluding the sampling area from the surrounding environment. It accounts for small-scale spatial variability and benthic heterogeneity by integrating the flux over an area of 10-100 m 2 (Berg et al. 2007;Rheuban and Berg 2013). Deployments typically last for 24-72 h, and fluxes are usually resolved over time intervals of 15 min to 1 h. The technique derives fluxes from simultaneously measured in situ water column velocities and associated water column O 2 concentrations. The velocities are measured at rates between 8 and 64 Hz by an acoustic Doppler velocimeter (ADV), while O 2 concentrations are recorded near the ADV measuring volume by a fast-responding O 2 measuring system .
Fast-responding O 2 measuring systems used for AEC include systems based on microelectrodes, fiber-optic optodes, or microplanar optodes. Thin Clark-type glass microelectrodes typically have short response times (< 0.3 s) but consume O 2 , which makes their signal sensitive to current velocity changes (stirring sensitivity) (Attard et al. 2016;Reimers et al. 2016). Their fragile sensor tips (outer tip diameter 10-200 μm) break more easily than other sensors, which can result in a loss of data (Chipman et al. 2012;Attard et al. 2014). Fiber-optic optodes also typically have short response times (< 0.3 s), and use luminescence quenching by O 2 to quantify O 2 concentrations, a process that does not consume O 2 (Chipman et al. 2012;Koopmans et al. 2020). While some studies report no or negligible stirring sensitivity (Chipman et al. 2012;Holtappels et al. 2015), others have shown that fiber-optic sensors can exhibit stirring sensitivity . Although fiber-optic sensors are less susceptible to signal drift and typically have a more robust tip (e.g., outer tip diameter 430 μm) than microelectrodes (Chipman et al. 2016), they will experience a loss of signal strength over time due to bleaching of the fluorophore that coats the tip. Biofouling of both fiberoptodes and microelectrodes can cause erroneous signals (Huettel et al. 2020), and collisions with debris can lead to signal spikes, sensor defects and data loss (Koopmans et al. 2020(Koopmans et al. , 2021. To increase robustness and thus, deployment success, some studies use multiple microelectrodes or fiber-optic optodes (Attard et al. 2014(Attard et al. , 2019bMerikhi et al. 2021). An alternative to these sensing systems is the microplanar optode sensor, which is more robust because of its larger tip (8 mm diameter) that does not break and is rarely affected by floating debris Amo-Seco et al. 2021). However, its relatively large tip can disturb the current flow, so recorded data may be compromised when the current flow comes from behind the sensor (Huettel et al. 2020). The microplanar optode must be placed farther away from the ADV so that its tip does not disturb velocity measurements ).
Due to their complexities, O 2 measuring systems used for AEC must undergo rigorous testing under well-defined lab and field conditions to assess performance, reliability, and potential limitations (McGinnis et al. 2011;Chipman et al. 2012;Berg et al. 2016). Typically, these O 2 measuring systems are tested in the lab in water that is close to atmospheric saturation or O 2 -undersaturated. These conditions are indeed representative of some ecosystems where AEC has been used, such as unvegetated muddy sediments at depths below the photic zone (Berg et al. 2009;Attard et al. 2014;Amo-Seco et al. 2021). However, the water column of ecosystems dominated by photosynthesizing vegetation such as seagrass meadows (Rheuban et al. 2014b;Berger et al. 2020) or macroalgal canopies (Volaric et al. 2019;Attard et al. 2019a) can be highly supersaturated with O 2 during the daytime. To best represent such field conditions, new O 2 measuring systems for AEC should be evaluated under both O 2 -undersaturated and O 2 -supersaturated conditions.
In this study, we assessed the suitability of a new fiber-optic O 2 measuring system for AEC. We provide a review of the system's characteristics and performance in the lab over a broad spectrum of conditions and in the field in a shallow, temperate seagrass meadow.

Materials and procedures
New O 2 measuring system The AquapHOx-LX logger (hereafter termed "AquapHOx") is a new optical submersible meter designed and optimized for fast response times and high-resolution O 2 readings (PyroScience, GmbH). The AquapHOx has a measuring frequency up to 40 Hz, a 16-bit digital-to-analog signal conversion, and a 0-5 V analog output. To measure O 2 concentrations, the AquapHOx is connected to a manufacturer-calibrated, ultra-high speed retractable fiberoptic minisensor (OXR430-UHS-SUB; PyroScience). The AquapHOx automatically compensates for O 2 signal drift due to the temperature dependency of the optode and changes in mean water temperature via a temperature sensor located on the meter's housing (response time = 0.5 s, accuracy = 0.05 C). The AquapHOx produces analog output signals of both the O 2 concentration and temperature at frequencies up to 40 Hz.
For our AEC applications, the AquapHOx was also connected to a standard ADV through a one-cable plug and play connection (Vector; Nortek; firmware version 3.44). The ADV measured velocities (x, y, and z) at 64 Hz, powered the AquapHOx, and recorded the analog O 2 concentration and mean temperature output signals. It is advantageous to record all data on the same data logger to effectively identify and account for time offsets between signals .
We thoroughly tested the PyroScience system (AquapHOx and O 2 sensor) in the lab and the field (Table 1). In all tests, O 2 concentrations were measured by the PyroScience system at 40 Hz. The ADV recorded velocity data at 64 Hz in the lab experiments and at 16 Hz in the field experiments (note that the ADV sampled at 64 Hz but internally averaged these data to provide output at 16 Hz). These high-frequency ADV and O 2 signals were reduced to 8 Hz through averaging before eddy fluxes were extracted (Berg et al. 2009;Lorrai et al. 2010;Berg et al. 2016).

Response time evaluation
System response time was assessed by measuring the t 90% response time, which is the time a system requires to register 90% of an abrupt change in O 2 concentration. To create this abrupt concentration change, the fiber-optic sensor tip was quickly dipped from the air into an O 2 -undersaturated (< 10% saturation) water bath at a perpendicular angle to the main current flow (90 ). The water bath (diameter = 30 cm, depth = 20 cm) was continuously stirred with a stir bar to ensure no O 2 concentration gradients were present. The stable air and water O 2 concentrations, as well as 90% of the change between the two, were marked with dashed lines on a graph (Fig. 1). The number of time intervals between data points for the signal to get from the "air-line" to below the "t 90% -line" were counted and divided by the ADV frequency (64 Hz). Because the AquapHOx updated its analog signal 40 times per second (40 Hz) while the ADV recorded data at 64 Hz, two data points sometimes occurred side-by-side (Fig. 1). Turbulent frequencies less than 8 Hz typically contribute little to the O 2 flux, so these side-by-side points would not compromise flux estimates. Three fiber-optic sensors were used, and each sensor was tested nine times. The entire test was repeated in O 2 -supersaturated water ($ 200% saturation) (Fig. 1b).
The effect of sensor angle relative to the main current flow on the t 90% response time was also assessed. The response time test was repeated in O 2 -undersaturated water while the sensor was dipped at two more angles: first, at a 45 angle pointing into the main flow direction (45 ), and then at a 45 angle pointing away from the main flow direction (À45 ). Response times were compared to the response times determined using perpendicular (90 ) dips into O 2 -undersaturated water. Differences in t 90% response times were analyzed using a one-way ANOVA.

System precision
We evaluated system precision by comparing the PyroScience system's O 2 concentration readings to those of two reference planar optode systems (t 90% = $30 s) (miniDOT; PME) that are characterized by minimal drift. The fiber-optic sensor and the miniDOTs were submerged in an O 2 -undersaturated, rotating water bath that was open to the atmosphere. The sensors remained in the water until the O 2 concentration in the water bath had re-equilibrated with the atmosphere. This test was repeated in an O 2 -supersaturated water bath.

Time delay
Photon collection and subsequent signal processing within the AquapHOx can introduce a time lag between the ADV velocity data and the O 2 concentration data. Unless corrected for, such lag time can result in the systematic underestimation of fluxes ). To assess this potential time lag, the fiber-optic sensor and the ADV transmitter were simultaneously dipped into an O 2 -undersaturated water bath that was seeded with particles to ensure adequate backscatter for acoustic velocity measurements. The optical fiber tip was attached to the ADV a few millimeters away from and at the same height as the transmitter to ensure that they penetrated the water surface at the same time. We compared the times that the PyroScience system and the ADV recorded the transition from air into water by analyzing the ADV signal-to-noise ratio (SNR) and the PyroScience system's O 2 concentration reported by the ADV ("counts"). The internal time lag was found by averaging the time difference over 21 trials. To ensure that no time delay existed within the ADV itself, this experiment was repeated where the PyroScience system was replaced by an electric circuit. The circuit was comprised of a wire with a non-insulated ending that was mounted similarly to the optical fiber sensor next to the ADV transmitter, and two 1.5 V batteries. The electric circuit was closed when this wire touched the water surface.

Temperature correction bypass
Submersible O 2 measuring systems are sensitive to changes in water temperature (Gundersen et al. 1998;Berg et al. 2022). As a result, most newer systems have a built-in temperature correction. The PyroScience system applies the correction using a temperature sensor fixed to the housing of the AquapHOx (Fig. 2b). We quantified the PyroScience system's sensitivity to temperature changes when the AquapHOx was not submerged in the same body of water as the fiber-optic sensor. The AquapHOx was placed on a lab bench in a room with a constant air temperature and the sensor was submerged in a water-filled flask along with a dual temperature and O 2 planar optode sensor (RINKO; JFE Advantech) (Fig. 2). The flask was capped to ensure a constant molar O 2 concentration and placed on a hot plate with a stir bar (Fig. 2). The RINKO sensor was chosen as a reference because its temperature and O 2 sensors are mounted next to one another ($ 3 mm apart). Both systems recorded O 2 concentrations and temperatures for 15 min at room temperature, and then for 45 min as the water was gradually heated. This test was performed in O 2undersaturated water, O 2 -saturated water, and O 2 -supersaturated water.

Sunlight sensitivity
Some fiber-optic sensors are sensitive to changes in light, which can affect the accuracy of O 2 concentration measurements during the diurnal cycle. We submerged the fiber-optic sensor in an O 2 -saturated, rotating water bath that was positioned in direct sunlight. O 2 concentrations were recorded as the water bath was exposed to alternating 5-min periods of direct sunlight and complete darkness by placing a black box over the entire setup.

Stirring sensitivity
To assess stirring sensitivity, the fiber-optic sensing tip was exposed to varying tangential velocities in an O 2 -saturated, rotating water bath. No O 2 concentration gradients were present due to the mixing applied to the bath, so any observed changes in the O 2 concentration could be attributed to changes in velocity. The water was rotated at tangential velocities of 0, 4, and 15 cm s À1 .
The PyroScience system was similarly evaluated for stirring sensitivity in the presence of wave motion using the setup and data analysis described by Berg et al. (2016). In short, the PyroScience system and the ADV were mounted on a frame used for AEC field measurements (see below) and submerged in a wave tank with seeding particles added to facilitate acoustic flow measurements. Velocities and O 2 concentrations were measured by the AEC system.

Field test
An AEC system comprised of an ADV and the PyroScience system was deployed in a restored eelgrass (Zostera marina) meadow in South Bay, a shallow subtidal lagoon located within the Virginia Coast Reserve Long-Term Ecological Research site (VCR-LTER). The lagoon has a mean water depth of 1.2 m and a tidal range of 1 m (Safak et al. 2015). South Bay is an ideal location to evaluate the PyroScience system because it is the site of 15 years of AEC measurements (Hume et al. 2011;Rheuban et al. 2014b;Berger et al. 2020;Juska and Berg 2022).
Four continuous deployments occurred between 1 and 18 June 2021. Deployments ranged from 21 to 47 h in length, resulting in approximately 6 full days of data. In all deployments, the ADV and the AquapHOx were mounted on a thin, light, stainless steel frame (Fig. 3) (Berg and Huettel 2008).

Fig. 2.
To test the temperature sensitivity of the PyroScience system when the built-in temperature correction was bypassed, (a) the PyroScience sensor and the RINKO sensor were submerged in a waterfilled, capped flask while (b) the AquapHOx was placed on the lab bench. The temperature sensor mounted on the AquapHOx is contained in the copper-colored small cylinder located next to the fiber connector.
The ADV measured velocity (x, y, z) continuously at 16 Hz in a $ 2 cm 3 measuring volume located 30 cm above the benthic surface, which is the average eelgrass canopy height at slack tide. The fiber-optic sensor tip was positioned $ 0.7 cm away from the edge of the ADV measuring volume to avoid interfering with the velocity measurements (Fig. 3c). To calibrate O 2 concentrations, two reference miniDOTs were positioned at 2 and 30 cm above the sediment surface. No significant vertical stratification was detected. Photosynthetically active radiation (PAR) was measured at 5-min intervals by two planar PAR loggers (Odyssey; Dataflow Systems) mounted 30 cm above the sediment surface. All instruments were deployed and retrieved during low tide.
Eddy fluxes were extracted following protocols that have been described in detail by Berg et al. (2022). Briefly, 15-min O 2 fluxes were extracted from the data using EddyFlux3.2 software (Peter Berg, unpublished). In this process, a standard time shift correction (Fan et al. 1990;McGinnis et al. 2008;Lorrai et al. 2010) and storage correction (Rheuban et al. 2014b) were applied. Flux data with signal spikes, which indicate collisions with or temporary attachments of debris in the flow, were removed. The remaining 15-min fluxes were binned into hourly fluxes.

Response time evaluation
The O 2 saturation state of the water had a significant effect on the t 90% response time of the PyroScience system (Fig. 4). The t 90% response time was significantly longer in O 2 -supersaturated water than in O 2 -undersaturated water (p = 0.0015). Specifically, the mean t 90% response time was 0.088 AE 0.0034 s (n = 27, mean AE SE) in O 2 -undersaturated water and 0.12 AE 0.0067 s (n = 27, mean AE SE) in O 2supersaturated water (Fig. 4).
The t 90% response time was not significantly different when the sensor was dipped at a 90 angle (t 90% = 0.088 AE 0.0034 s, n = 27, mean AE SE) and at a 45 angle (t 90% = 0.087 AE 0.0038 s, n = 27, mean AE SE) (p = 0.91) in O 2 -undersaturated water (Fig. 5). However, the t 90% response time was significantly longer when the sensor was dipped at a À45 angle (t 90% = 0.14 AE 0.010 s, n = 27) (p = 0.00001) (Fig. 5). Thus, the PyroScience system t 90% response time is shortest when the sensor is positioned at a 90 angle or 45 angle facing into the main current flow.

System precision
There was a good overall agreement between O 2 concentrations measured by the PyroScience system and by the two miniDOTs used as references (Fig. 6). There were no noticeable discrepancies between the sensor readings in O 2undersaturated water or in O 2 -saturated water (Fig. 6a). In O 2supersaturated water, the PyroScience system reported lower O 2 concentrations than the miniDOTs (Fig. 6b). This relatively small concentration difference occurred largely outside of the manufacturer-specified calibration range of the miniDOTs (0-150% O 2 saturation, AE 10 μmol L À1 accuracy), and thus, is difficult to address.

Time delay
The ADV and electric circuit recorded the dip into water at the same time (Fig. 7a,b). Thus, no time delay was detected within the ADV. On the contrary, a distinct time delay was  identified between the input and the output sides of the AquapHOx (Fig. 7c,d). The average time delay was 0.046 AE 0.0023 s (n = 21, mean AE SE). This time delay can be corrected in post-processing by applying a standard time shift correction.

Temperature correction bypass
When the temperature correction of the O 2 signal was bypassed in the lab, the PyroScience system showed temperature sensitivities comparable to other O 2 measuring systems used for AEC Berg et al. 2022) (Fig. 8). Specifically, when the water temperature was increased, the PyroScience system had a temperature sensitivity of 2.5% per C in O 2 -saturated water, 1.7% per C in O 2undersaturated water, and 2.4% per C in O 2 -supersaturated water (Fig. 8).

Sunlight sensitivity
The fiber-optic sensor's O 2 concentration readings were not affected by exposure to strong sunlight. The PyroScience system correctly reported consistent O 2 concentrations when the sensor was exposed to periods of strong sunlight and periods of complete darkness (Fig. 9).

Stirring sensitivity
We detected no measurable stirring sensitivity. When the fiber-optic sensor was submerged in the rotating water bath, concentration readings were not significantly different at different tangential velocities (Fig. 10). The minimal upward drift likely occurs because the water bath was slightly O 2 -undersaturated with respect to the air at the onset of the test. Similarly, O 2 concentration readings recorded by the AEC system in the wave tank were not affected by wave motion.

Field test
All field deployments produced high-quality AEC data from which benthic O 2 fluxes were extracted. In the 43-h sample deployment, O 2 concentrations recorded by the PyroScience system and the two miniDOTs agreed well (Fig. 11b). The hourly O 2 fluxes derived from these data were well-correlated with PAR (Fig. 11c,d). The maximum positive O 2 flux was 599 mmol O 2 m À2 d À1 and the maximum negative O 2 flux was À284 mmol O 2 m À2 d À1 (Fig. 11c). These values are within the range of O 2 fluxes previously recorded in South Bay with Clark-type microelectrodes Berger et al. 2020). A smaller flux was observed in the middle of day 2 (hour 37), likely due to the effect of a decrease in flow velocity on the benthic flux (Fig. 11a,c). This relationship has been previously documented in seagrass meadows (Hume et al. 2011;Long et al. 2015).

Discussion
AEC is a cutting-edge technique that is increasingly used to quantify in situ benthic O 2 fluxes under naturally varying environmental conditions (Berg et al. 2003Long 2021). While the principle of the technique is simple-it relies on rapid measurements of the vertical water velocities and O 2  concentrations above the benthic surface-there are challenges associated with the technologies applied to secure such data. In this study, we tested a new optical submersible meter (the AquapHOx-LX; PyroScience), which was designed and optimized by the manufacturer to fit the criteria for highquality AEC measurements of O 2 fluxes. We present a series of rigorous lab and field tests that evaluate the performance of the AquapHOx. In all tests, the AquapHOx was connected to an ultra-high speed fiber-optic O 2 sensor (OXR430-UHS-SUB; PyroScience) to measure O 2 concentrations and to an ADV (Vector; Nortek) to provide power to and record the output from the AquapHOx.
The t 90% response time of the PyroScience system was 27% slower when the sensor was dipped into O 2 -supersaturated water ($ 200%) compared to into O 2 -undersaturated water (< 10%) (Fig. 4). Typically, sensor response times are not assessed in O 2 -supersaturated water, even though O 2supersaturated conditions have been reported in shallow water environments (Attard et al. 2014(Attard et al. , 2019aLong et al. 2020). For example, Berger et al. (2020) reported frequent O 2 saturation levels of over 200% in a temperate seagrass meadow. The substantial difference in response times found here may be caused by the diminishing phosphorescence decay of fiber-optic measuring systems associated with increasing O 2 concentrations or the low luminescence at high O 2 concentrations, which reduces system sensitivity.
Similarly, the dipping angle between the sensor and the water surface affected response time. The sensor response times were, on average, 37% slower when the sensor was dipped at a À45 angle (flow approaching the back of the sensor) compared to a 45 angle (flow approaching the front of the sensor) (Fig. 5). Dip angles of 45 or 90 (vertical) gave statistically identical response times. The identified difference is presumably due to a thicker diffusive boundary layer between the bulk flow and the O 2 sensing coating on the tip at the À45 angle. Regardless of these differences, all identified response times (t 90% = 0.087-0.14 s) are well within the range of those reported for other O 2 sensors successfully used for AEC (t 90% = 0.25-0.5 s) Long et al. 2019;Attard et al. 2019b;Koopmans et al. 2020). It should also be noted that such differences will be partially corrected for if a standard time shift correction (Fan et al. 1990;McGinnis et al. 2008) is applied to the flux calculation. This data processing step is increasingly used in AEC work (Lorrai et al. 2010;Koopmans et al. 2020;Berg et al. 2022). An added advantage of using this correction is that it will also automatically correct for the internal time delay (0.046 s) associated with signal processing in the AquapHOx found in this study (Fig. 7).
A fundamental principle of AEC measurements is that recorded variations in O 2 concentrations are not attributed to changes in other environmental variables. This condition can be challenging to achieve. For example, some O 2 sensors are subject to stirring sensitivity, a phenomenon where changes in velocity alone affect the recorded O 2 concentration. Microelectrode sensors are inherently subject to stirring sensitivity due to their internal consumption of O 2 (Gust et al. 1987;Holtappels et al. 2015). For unknown reasons, optical sensors are sometimes affected by stirring sensitivity as well ), but not always (Holtappels et al. 2015). These conflicting results still need to be investigated. In this study, which included sensor exposure to both varying currents and wave action in separate tests, the O 2 sensors showed no sign of stirring sensitivity (Fig. 10). In an independent test, we similarly documented that the O 2 sensors did not show any sensitivity to varying light exposure (Fig. 9), which is another dynamic environmental variable that can potentially bias the concentration reading.
All electrochemical and optical O 2 sensors are sensitive to changes in temperature (Gundersen et al. 1998;. Typically, when exposed to a constant molar O 2 concentration, O 2 sensor readings will change 2-3% per C of temperature change. For that reason, newer O 2 measuring systems, including the PyroScience system, have a built-in temperature correction of the O 2 signal. We documented that the PyroScience system has a temperature sensitivity that ranges from 1.7% to 2.5% when this correction is bypassed (Fig. 8). PyroScience system (c, d). No time delay was observed between (a) the ADV signal-to-noise ratio (SNR) and (b) the ADV output of the electric circuit. An average time delay of 0.046 AE 0.0023 s (n = 21, mean AE SE) was observed in the PyroScience system when (c) the ADV SNR was compared to (d) the PyroScience system.
As a result, care must be taken to ensure that the AquapHOx and sensor are submerged in the same body of water at all times during deployments. Even then, the distance between the AquapHOx and the sensor can present a challenge in environments with substantial vertical heat fluxes, such as at the air-water interface or in benthic systems where tides or currents drive pronounced temperature changes . In such settings, rapid turbulent fluctuations in temperature associated with the heat flux can be falsely recorded as turbulent fluctuations in O 2 concentration and can bias the O 2 flux. The only means to avoid this is to measure the temperature with a fast-responding temperature sensor located right next to the O 2 sensor .
The AEC technique requires that data segments corrupted by particle interference or sensor breakage are excluded to reduce the error margin of derived fluxes Rheuban and Berg 2013;Attard et al. 2016). The amount of data that is excluded varies with site conditions and the sensor used. Ecosystems with turbid water, floating debris, and frequent high flow rates and wave action such as seagrass meadows sometimes require high percentages of data to be excluded (Rheuban and Berg 2013). For example, no full 24-h data record was collected in South Bay in June 2018 due to breakage of the Clark-type microelectrodes used (Berger Fig. 8. A fiber-optic sensor and a RINKO planar optode sensor were submerged in a capped flask filled with O 2 -saturated water (see Fig. 2). The AquapHOx was placed on a lab bench at room temperature. (a) The PyroScience system and the RINKO recorded O 2 concentrations and (b) the RINKO recorded water temperature. The water was at room temperature for 15 min and then was gradually heated for 45 min. Fig. 9. The PyroScience system was evaluated for sensitivities to strong sunlight. There was no significant change in O 2 concentration readings when the sensor was exposed to periods of strong sunlight and periods of complete darkness. et al. 2020). Similarly, and despite using the more robust optical fiber sensors used in this study, 42% of recorded data were excluded from measurements in a eutrophic freshwater embayment (Koopmans et al. 2021). Also, sometimes 50% of the data are excluded when planar optode sensors are used in tidal systems as the larger sensor tip may disrupt velocity measurements Amo-Seco et al. 2021). In comparison with these studies, 17% of the recorded data were excluded in our four deployments. Overall, the data were characterized by low noise and less disturbances, and no sensors broke or were compromised by fouling.
The extracted O 2 fluxes from our deployments were fully in line with previous flux measurements recorded at our site using microelectrodes (Rheuban et al. 2014a;Berg et al. 2019;Berger et al. 2020). In our example deployment, hourly O 2 fluxes ranged from À295 to 590 mmol m À2 d À1 (Fig. 11). In comparison, hourly O 2 fluxes from the same site ranged from À267 to 359 mmol m À2 d À1 in June 2015  and from À500 to 550 mmol m À2 d À1 in June 2014 (Berger et al. 2020). Overall, our lab and field tests demonstrate that the PyroScience system resolves high quality O 2 fluxes, and is an excellent choice for future AEC measurements.

Comments and recommendations
We recommend that future studies of O 2 measuring systems for AEC include an evaluation of t 90% response times at different O 2 saturation states, perhaps tailored using sitespecific knowledge of where the sensors will be used. Because such tests, to our knowledge, have not been done before, it is unknown if other sensors may show an even stronger O 2 concentration level dependency. Although it can be a challenge in shallow water systems, the AquapHOx and O 2 sensor must be submerged in the same body of water at all times during deployments to ensure an accurate temperature correction of the O 2 signal. Finally, we propose that the test protocols developed during this study should be used as guidelines or as a starting point for future evaluations of new O 2 systems that are going to be used for AEC measurements.

Data availability statement
The data that support the findings of this study are available from the corresponding authors, KEG or PB, upon request.