The Penlee Point Atmospheric Observatory (50∘19.08′ N, 4∘11.35′ W; http://www.westernchannelobservatory.org.uk/penlee/) was established in May 2014 by the Plymouth Marine Laboratory (PML) on the south-west coast of the United Kingdom for long-term observations of air–sea exchange and atmospheric chemistry. PPAO is in close proximity to two nearby long-term marine stations that form the Western Channel Observatory (http://www.westernchannelobservatory.org.uk). Meteorological variables (wind, temperature, humidity, and pressure), sea surface temperature (SST), salinity, chlorophyll, oxygen, and dissolved organic matter are measured continuously from buoys stationed at L4 (50∘15.0′ N, 4∘13.0′ W) and E1 (50∘02.6′ N, 4∘22.5′ W), which are about 6 and 18 km south of PPAO. Seawater pCO2 is measured on weekly cruises to the L4 station and biweekly cruises to the E1 station (Kitidis et al., 2012).

PPAO is situated on an exposed headland on the western edge of the Plymouth Sound, which is primarily fed by the Tamar Estuary from the north-west and is open to the Atlantic Ocean to the south-west (Fig. 1). South-south-west of PPAO, the water depths increase steadily to ∼ 8, 15, 22, and 24 m (relative to mean sea level) at horizontal distances of 100, 300, 1000, and 1300 m (www.channelcoast.org). North-easterly wind comes over the Plymouth Sound to PPAO and is limited to a fetch of about 5 km. Air from the south-east is affected by pollution from the European continent as well as shipping emissions (Yang et al., 2016). In the south-westerly direction, the wind fetch is up to thousands of km and the wind speed sometimes exceeds 20 m s-1. This brings in air that has much less anthropogenic influence and is more representative of the background Atlantic atmosphere (see Sect. 4.1).

The stone PPAO building (length, width, height of 3.5, 3.5, 3.0 m) is approximately 11 m a.m.s.l., mains powered, vehicle-accessible, and uses line-of-sight radioethernet to communicate with PML (6 km to the north-north-east). A small strip of land and a narrow, rocky intertidal zone separate the building from the sea. South-west and north-east of PPAO, the horizontal distance to the water's edge is 30–60 m, depending on the tide. South-east of PPAO, the distance to water is greater (about 70–90 m) due to an exposed rocky outcrop. The local tidal amplitudes (semi-diurnal) are ∼ 5 m during spring tide and ∼ 2 m during neap tide. The intertidal zone is only sparsely covered by macroalgae (< 10 % by area), likely due to frequent exposure to large waves.

During May–June 2014, a sonic anemometer (Gill Windmaster Pro) and a meteorology station (Gill Metpak Pro) were mounted on a metal pole about 1.4 m above the PPAO rooftop. A telescopic mast (retracted length of 2.8 m and fully extended length of 12.3 m; Clark Masts) was installed on top of the observatory roof (Fig. 1) on 17 June 2014. The Windmaster Pro anemometer and the meteorology station were then moved to a cross bar on top of the mast. In February 2015, another sonic anemometer (Gill R3) was installed at the same height as the Windmaster Pro, about 80 cm apart in the horizontal. The sonic anemometers measure 3-dimensional wind velocities (u, v: the two horizontal components; w: the vertical component) at 10 Hz (Windmaster Pro) and 20 Hz (R3). Table 1 summarises measurement periods and corresponding sensor heights.

We deployed the Windmaster Pro and the R3 sonic anemometers side by side for two reasons. First, signal dropouts at high frequencies were common for the Windmaster Pro during moderate-to-heavy precipitation, which tended to coincide with strong south-westerly winds. Valid flux measurements from the Windmaster Pro, limited to mostly dry periods, may thus be biased towards low-to-intermediate wind speeds. Second, initial drag coefficient measurements from the Windmaster Pro at PPAO were lower than expected compared to published results for air–sea momentum flux. The manufacturer Gill report a firmware bug in the Windmaster Pro and recommend a bias correction to the w axis (+16.6 % for positive w; 28.9 % for negative w; see technical key note: http://gillinstruments.com/data/manuals/KN1509_WindMaster_WBug_info.pdf). This correction is not necessary for the R3 anemometer, which has individually calibrated u, v, and w components. Simultaneous deployments of these two anemometers allow us to evaluate the effectiveness of the Windmaster Pro correction (Sect. 3.3).

Atmospheric mixing ratios of CO2 and CH4 were measured by a Picarro cavity-ringdown analyzer (G2311-f) at a frequency of 10 Hz (flux mode). The inlet to this analyzer was mounted ∼ 30 cm below the centre volume of the Windmaster Pro anemometer. An external dry vacuum pump drew sample air via a ∼ 18 m long, 3/8′′ OD Teflon perfluoroalkoxy (PFA) tubing at a flow rate of initially ∼ 30 L min-1. The pump performance deteriorated over time due to constant exposure to sea salt. A high-performance particulate arrestance (HEPA) filter was installed immediately upstream of the pump in late 2014, which resulted in a ∼ 15 L min-1 reduction of the main flow. The Picarro instrument subsampled from the main flow via a ∼ 2 m long, 1/4′′ OD Teflon PFA tubing at a rate of ∼ 5 L min-1. Airflow was fully turbulent throughout the inlet.

The presence of water vapour (H2O) degrades the measurements of CO2 and CH4 via dilution, spectral interference and line broadening (Rella, 2010). Miller et al. (2010) and Blomquist et al. (2014) found that ambient variability in H2O mixing ratio causes significant bias to the EC measurements of air–sea CO2 flux. We followed the recommendation of Blomquist et al. (2014) and dried the sampled air using a high-throughput dryer (Nafion PD-200T-24M). H2O efficiently permeates through the Nafion membrane while CO2 and CH4 essentially do not. Set up in counter-flow mode (reflux configuration), the dryer utilises the lower pressure of the Picarro exhaust air to dry the sample air. The ambient H2O mixing ratio is typically on the order of 1 % at PPAO. With the dryer inline the measured H2O mixing ratio was reduced by 5 to 10-fold. The Picarro instrument reports mixing ratios of CO2 and CH4 in sample air based on precisely controlled cavity temperature and pressure. An internal, point-by-point correction by the instrument for residual humidity yields the dry mixing ratios (CCO2 and CCH4), which we use for flux computations. Air density fluctuations (i.e. Webb et al., 1980) should thus not affect our measurements. Tuned by the manufacturer prior to our first use, we checked the instrument calibration with CO2 and CH4 gas standards (BOC) and occasionally determined the instrument backgrounds with nitrogen gas. CO2 and CH4 measurements were unavailable between August 2014 and March 2015 due to faults in the Picarro instrument.

In the eddy covariance method, flux is determined from the correlation between the vertical wind velocity (w) and the variable of interest (x): w′x′‾. Here the primes indicate fluctuations from the means while the overbar denotes temporal averaging. The coastal environment near PPAO is complex and heterogeneous in both air and water phases. Shifts in air mass and wind direction result in substantial changes in air temperature and gas-mixing ratios. Thus we chose a relatively short averaging interval of 10 min (as used by e.g. Miller et al., 2010) to more easily satisfy the homogeneity/stationarity requirements for eddy covariance (see Appendix A for flux quality control).

For the computations of CO2 and CH4 fluxes (w′CCO2′‾, w′CCH4′‾), an lag correlation analysis is performed hourly to determine the time delay between the instantaneous vertical winds and gas-mixing ratio measurements. Most of the atmospheric variability in CO2 and CH4 is caused by horizontal transport rather than the air–sea flux. Detrending the gas-mixing ratios to remove low-frequency variability improves the accuracy of the lag-time determination. Between May and July 2014, a delay of 1.9 ± 0.1 s was found between w (Windmaster Pro anemometer) and CCO2. After the installation of the HEPA filter, the delay increased to 3.3 ± 0.1 s. Lag times derived from w and CCH4 are much noisier due to the smaller magnitude of the CH4 flux. We apply the lag correction determined from the w : CCO2 analysis to the CH4 flux calculation. Ten-minute segments of CO2 and CH4 fluxes that pass the quality control criteria (see Appendix A) are further averaged to hourly intervals, which reduces random noise by a factor of ∼ N0.5, where N is the number of valid flux segments. Only hours with at least three 10 min flux intervals are considered for further analysis.

A double rotation (Tanner and Thurtell, 1969; Hyson et al., 1977) streamline correction is applied to wind data in 10 min blocks prior to flux computation. Tilt angles between the horizontal and vertical planes from this calculation for sampling heights of 15, 18, and 27 m a.m.s.l. are shown in Fig. 2. During onshore airflow, the mean tilt angle is positive as air is forced upwards. The magnitude of this tilt for the south-westerly wind, which blows perpendicularly across the Penlee headland and makes contact with water again to the north-east, is comparable to shipboard measurements. The tilt angle is negative in the north-west sector due to the presence of a small hill behind the observatory building in that direction. A comparison of horizontal wind speed between Penlee and the L4 buoy when the wind is from the south-west does not show, within measurement uncertainties, a significant acceleration in the Penlee measurement (e.g. as might be expected when air is forced over a large superstructure). Thus the hill to the north-west of the site should not have a major influence on our measurements during south-westerly conditions. A peak in tilt angle near 120∘, more apparent at low sampling heights, is likely caused by the exposed rocky outcrop in that direction. The impact of this local topography is reduced with increasing sampling height.

From the friction velocity u∗=(u′w′‾2+v′w′‾2)1/4 and wind speed (Utrue), we compute the drag coefficient CD=(u∗/Utrue)2. Bin-averaged CD at the three sampling heights as a function of wind direction is shown in Fig. 3. At 15 and 18 m a.m.s.l., measured CD from about 80 to 150∘ are clearly elevated compared to open-ocean values (which typically range between 0.5 × 10-3 and 2.5 × 10-3 depending on the wind speed; Edson et al., 2013). This is likely because a part of the flux footprint overlapped with the rocky outcrop in that direction, which has a greater roughness length than the surface ocean. Likewise, high CD values between 250 and 40∘ are caused by land. The impact from the rocky outcrop to the south-east is no longer obvious at a sampling height of 27 m a.m.s.l., when the flux footprint shifts further away from the observatory. For winds blowing from the north-east and south-west, measured CD is lower and much closer to values expected for the open ocean. North-easterly winds are relatively infrequent (∼ 8 % of the time) and limited in fetch; also the air mass from that direction is affected by terrestrial pollution and ship emissions. We thus focus on the more frequent (∼ 20 % of the time) south-west wind sector (180–240∘) for most of this paper. In Appendix B, we compute the theoretical flux footprints at different sampling heights and during various atmospheric conditions/tidal cycles. For south-westerly winds, land influence is predicted to be only a few percent when the mast height is ≥ 18 m a.m.s.l.

Here we compare the 10-m neutral drag coefficient (CD10N=(u*/U10N)2) and sensible heat fluxes to the fairly well established open-ocean bulk formulae predictions. The 10-m neutral wind speed U10N is determined using Businger–Dyer relationships (Businger, 1988) from the wind speed and air temperature at PPAO, tidal-dependent sampling height, and SST from L4. EC sensible heat flux is derived from the sonic temperature and further corrected for humidity using the bulk latent heat flux. To avoid sheltering by Rame Head to the west and near-shore processes, we limit our CD10N observations to a narrower wind sector of 180–220∘. Figure 4 shows the relationship between CD10N and U10N from the Windmaster Pro sonic anemometer. Also shown are the predicted CD10N from the COARE (Coupled Ocean-Atmosphere Response Experiment) model version 3.5 (Edson et al., 2013) and Smith (1980). When the sensors were initially placed at 15 m a.m.s.l., measured CD10N values were significantly higher than the open-ocean parameterisations at moderate wind speeds, probably because land/foreshore was within the flux footprint. At 18 m a.m.s.l., the mean CD10N at intermediate-to-high wind speeds was in close agreement with bulk predictions. Measured CD10N are sometimes elevated at wind speeds less than ∼ 5 m s-1, possibly due to increased flow distortion or minor land influence.

At 27 m a.m.s.l., CD10N measurements from the Windmaster Pro within the wind sector of 180–220∘ are limited (valid flux segments N=42), which appear to be lower than the open-ocean parameterisations by about 0.2 × 10-3. These low CD10N values may partly be due to remaining uncertainties in the Windmaster Pro sonic anemometer even after applying the bias correction to the w axis. Our coastal measurements show that at a tilt angle of 5∘, the recommended w correction increases u∗ from the Windmaster Pro by 6 % (and increases scalar fluxes by 14 %). Relative to the R3 sonic anemometer, this reduces the low bias in the Windmaster Pro u∗ from 9–10 to 3–4 %. The remaining 3–4 % bias can account for an approximate 0.1 × 10-3 underestimation of CD10N by the Windmaster Pro.

Figure 5 shows a comparison between the EC sensible heat flux and the bulk sensible heat flux. The latter is computed from SST from the L4 buoy (1 m depth), potential air temperature and U10N from PPAO, and the heat transfer rate from the COARE model (Fairall et al., 2003). Measurement and prediction are not far from the 1 : 1 line at a sampling height of 27 m a.m.s.l. (slope = 0.82; r2=0.72). A perfect agreement is not expected here, as any spatial heterogeneity in SST along the 6 km between L4 and PPAO (e.g. due to the Tamar Estuary outflow) or near-surface vertical gradient in seawater temperature would contribute to the discrepancy between measured and predicted sensible heat flux. At the initial sampling height of 15 m a.m.s.l., measured sensible heat flux is often very large and shows no correlation with the bulk flux estimate, most likely due to the terrestrial influence within the flux footprint. At 18 m a.m.s.l., a better coherence is observed but significant scatter remains, probably because the largest horizontal variability in SST is close to shore (and occupies more of the footprint at 18 m than at 27 m). Overall, the comparisons above suggest that the mean measured fluxes at a sampling height > = 18 m during south-westerly winds are within 20 % of the expected open-ocean air–sea transfer rates.

Mixing ratios of CO2, and CH4 (CCO2 and CCH4) varied at PPAO depending on wind direction (Fig. 6). On average between May and July 2014, CCO2 and CCH4 were generally higher for winds blowing from land than for winds blowing from the sea, likely due to the much greater terrestrial emissions of these gases and also different boundary layer dynamics. Mean CCO2 and CCH4 from the south-west sector (180–240∘) are similar to “well-mixed” atmospheric observations from sites such as Mauna Loa and Mace Head, consistent with the long atmospheric lifetime of these gases. Mean diel cycles in CCO2 and CCH4 between May and July 2014 during onshore (110–240∘) and offshore (300–60∘) wind flows are shown in Fig. 7. CCO2 and CCH4 for onshore winds show little diel variability, consistent with the relatively small air–sea CO2 and CH4 fluxes (on a per area basis). CCO2 and CCH4 for offshore winds increased at night and peaked in the early morning. Night-time wind speeds tend to be low in offshore flow, with an average of ∼ 3 m s-1 during these months. The resultant low atmospheric turbulence favours the formation of a shallow nocturnal boundary layer, which traps surface emissions. Between about 11:00 and 20:00 UTC, CCO2 was lower for offshore winds than for onshore winds, probably due to terrestrial photosynthesis. Similar diel cycles in CCO2 and CCH4 are often observed at terrestrial sites (e.g. Winderlich et al., 2014). Clear day/night differences were also apparent in the mixing ratios of oxygenated volatile organic compounds measured from the rooftop of PML (Yang et al., 2013). While not the focus of this work, it is worth noting that the elevated atmospheric CO2 and CH4 in the early morning will influence their air–sea fluxes in coastal regions during offshore conditions.

In this section, we examine the eddy covariance flux detection limit of CH4 and its dependence on instrumental noise as well as ambient variability. Blomquist et al. (2014) estimated an hourly CO2 flux detection limit of ∼ 1 mmole m-2 d-1 for a prototype version of the Picarro analyzer (G-1301-f) with a Nafion dryer at a wind speed of 8 m s-1 and in a neutral atmosphere. This represents an order of magnitude improvement over previous CO2 sensors (e.g. Licor) and is lower in magnitude than the typical air–sea CO2 flux. Based on terrestrial eddy covariance measurements, Peltola et al. (2014) estimated the CH4 flux detection limit using the Picarro analyzers G-1301-f and G-2311-f to be ∼ 170 µmole m-2 d-1 for an averaging interval (T) of 30 min (∼ 120 µmole m-2 d-1 at T=60 min). In comparison, the expected emissions of CH4 (FCH4) based on dissolved CH4 in the open ocean are generally less than 10 µmole m-2 d-1 (e.g. Forster et al., 2009).

We estimate the air–sea CH4 flux detection limit using an empirical and a theoretical approach. First, following Spirig et al. (2005), we compute the variability in the CCH4 : w covariance at a time lag far away from the true lag (i.e. +300 s). During periods of consistent south-westerly winds, the 1σ of this null CH4 flux is 15 µmole m-2 d-1 at T=10 min. The flux detection limit (defined as 3σ) should thus be 18 µmole m-2 d-1 (= 3 × 15/60.5) for an hourly average and 4 µmole m-2 d-1 for a daily average.

Based on theory and scalar flux observations, Blomquist et al. (2010, 2012) attributed total uncertainty in eddy covariance flux (δFC) to ambient variance (σCa2) and sensor noise (σCn2): δFC=2σWTσCa2τWC+σCn2τCn1/2=2σWTσCa2τWC+φCn41/2. Here τWC and τCn are the integral timescales for ambient variance and noise variance. The noise term in Eq. (1) relates to φCn, the band-limited noise. According to the manufacturer the precision of the Picarro G2311-f is ≤ 3 ppb for CH4 at a sampling rate of 10 Hz. The variance spectra of CH4 during two periods of south-westerly winds are shown in Fig. 8. Variance below ∼ 0.025 Hz largely follows the expected -5/3 slope for atmospheric transport. At frequencies above ∼ 0.025 Hz, the Picarro shows a “pink” background noise that approximately scales to a -1/5 slope. The integrated variance from 0.025 to 5 Hz is ∼ 1.1 ppb2, while the average φCn between 1 and 5 Hz is ∼ 0.23 ppb2 Hz-1. Considering noise alone (i.e. σCa2=0), for a neutral atmosphere at a wind speed of 10 m s-1 and a sampling height of 20 m a.m.s.l., Eq. (1) predicts an uncertainty in hourly CH4 flux of 11 µmole m-2 d-1 (Fig. 9). From the expected air–sea CH4 flux, using similarity theory we can estimate the variability in CCH4 caused by air–sea exchange in a neutral atmosphere as 3|FCH4|/u∗ (e.g. Fairall et al., 2000; Blomquist et al., 2010). For FCH4=2–20 µmole m-2 d-1 and u∗=0.3 m s-1, this corresponds to a predicted variability of 0.006–0.057 ppb. Figure 9 shows that if the ambient variability in CCH4 were in this range, the hourly flux uncertainty would be dominated by sensor noise.

The observed ambient variability in CCH4 tends to be more than an order of magnitude greater than is predicted from similarity theory, which is likely related to processes other than air–sea flux (e.g. spatial heterogeneity and horizontal atmospheric transport). We estimate σCa2 as the second point of the autocovariance of CCH4 (the difference between the first and second points of the autocovariance equates to σCn2 of ∼ 1 ppb2). At PPAO, the minimum CH4 ambient variability during onshore flow is 0.2 ppb (σCa2=0.04 ppb2), which corresponds to a predicted hourly flux uncertainty of 20 µmole m-2 d-1 (Fig. 9). This is close to our empirical estimate of the CH4 flux detection limit above. With increasing σCa (i.e. more variable CCH4), the flux uncertainty increases substantially and becomes much greater than FCH4, while the relative importance of σCn2 decreases. Thus, we expect the 10-fold greater CH4 flux detection limit estimated by Peltola et al. (2014) to be due to the higher variability in CCH4 over land than at our marine site (for onshore winds only). Over the open ocean where σCa in CH4 is likely to be even lower than at PPAO, the flux detection limit for CH4 should slightly decrease.

From the analysis above, it seems that an improvement in the precision of the CH4 instrument will only marginally reduce the uncertainty in CH4 flux. Blomquist et al. (2010) arrived at a similar conclusion in an analysis of air–sea carbon monoxide flux. At present, the relative CH4 flux uncertainty is best minimised by measuring in regions of large flux (i.e. high seawater supersaturation and strong winds) and minimal ambient variability (i.e. spatially homogenous environment).

Blomquist et al. (2010) and Yang et al. (2011) estimated the high-frequency loss in dimethylsulfide flux of typically less than 5 % from the same type of Nafion dryer as used in this study. Flux attenuation by the tubing itself should be negligible given the turbulent flow. Considering the other larger random uncertainties in our CO2 and CH4 fluxes, we present the measured fluxes without any attenuation correction in this paper.

Air–sea CO2 fluxes measured at sampling height of 27 m a.m.s.l. between June and July 2014 were generally small (Fig. 10). Diurnal land–sea breezes were common and durations of onshore winds tended to be short during this period. CO2 fluxes from the south-west (negative = into the ocean) ranged between 3 and -9 mmole m-2 d-1 (mean of -3 mmole m-2 d-1) during the relatively windy periods on 27 June and 4 July. Seawater pCO2 at the L4 station ranged between 326 and 345 µatm (mean of 337 µatm) from 9 June to 7 July 2014. The atmospheric CO2 mixing ratio at L4 agrees well with Picarro measurements at PPAO during onshore flow (Fig. 10). Using the air–sea difference in partial pressure of CO2 (ΔpCO2), SST and salinity at L4, as well as wind speed at PPAO, we compute the expected air–sea CO2 flux as kW⋅α⋅ΔpCO2, where α is the solubility of CO2 and kW is the gas transfer velocity from Nightingale et al. (2000) adjusted for Schmidt number. The expected air–sea CO2 flux of -1 to -5 mmole m-2 d-1 (mean of -3 mmole m-2 d-1) on 27 June and 4 July are of the same magnitude as our EC measurements. The mean EC CO2 flux could not be distinguished from zero in the second half of July, consistent with the increase in seawater pCO2 at L4. The spring algal bloom ended abruptly in early July 2014, with chlorophyll a concentration dropping from ∼ 3 to less than 1 mg m-3 (http://www.westernchannelobservatory.org.uk/buoys.php). The rapid warming of seawater from ∼ 13 ∘C in June to ∼ 18 ∘C in July aided a rapid approach towards air–sea CO2 equilibrium by the middle of July 2014.

Air-to-sea CO2 fluxes as substantial as -90 mmole m-2 d-1 were observed between April and June 2015 (sampling height of 18 m a.m.s.l., Fig. 11). For the south-west sector, the mean fluxes (standard errors) computed from the Windmaster Pro and the R3 sonic anemometers were -19.3(±1.4) and -23.7(±1.4) mmole m-2 d-1 respectively during this period. The reduced mean flux from the Windmaster Pro was primarily caused by signal dropouts in this anemometer during moderate-to-heavy precipitation, which tended to coincide with high wind speeds (and greater air–sea transfer). When both sonic anemometers were functional, CO2 fluxes computed from the Windmaster Pro and the R3 demonstrate excellent agreement (slope = 0.98, r2=0.95). Example CO2 cospectra over about half a day from 24 April (wind speed of 8 m s-1) and 10 May 2015 (wind speed of 6 m s-1) are shown in Fig. 12. The observed cospectra are fairly well described by theoretical fits for a neutral atmosphere (Kaimal et al., 1972). Minimal (< 10 %) flux loss at high frequencies is evident, as expected. Hourly CO2 flux (reversed in sign for clarity) during this period clearly increased with wind speed (Fig. 13). Unfortunately seawater pCO2 was not measured during this period for comparison. For reference, pCO2 measurements from L4 in May 2014 had a mean (1σ) of 306(26) µatm, implying a ΔpCO2 close to -100 µatm. We compute the predicted CO2 fluxes at SST of 12.5 ∘C (mean from the E1 station) and ΔpCO2 of -50 and -100 µatm. During most of this period, EC CO2 flux is fairly close to prediction using ΔpCO2=-100 µatm. Towards late May/beginning of June, the magnitude of CO2 flux appeared to be smaller at high wind speeds. A reduction in ΔpCO2 as occurred in 2014 could explain the declining CO2 fluxes in 2015.

Measured CO2 flux from the south-west between May and June 2014 (sampling height of 15 m a.m.s.l.) varied from a mean (±1 standard error) of about 40(±8) mmole m-2 d-1 at night to -55(±11) mmole m-2 d-1 during the day (Fig. 14). Mean wind speeds were fairly similar between day and night at around 5 m s-1 during this period. The pronounced diel variability and large magnitude of the CO2 flux suggest that these fluxes were likely affected by photosynthesis and respiration from land upwind of the observatory building and/or organisms living on the foreshore. As atmosphere–land exchange of CO2 can be more than an order of magnitude greater than air–sea CO2 flux on a per area basis (e.g. Goulden et al., 1996), a relatively small terrestrial contribution to the flux footprint (> 5 % spatially) could significantly bias the EC measurement. At sampling heights ≥ 18 m a.m.s.l., CO2 fluxes show much less diel variation, as would be largely expected for air–sea transfer (Fig. 14). However, the possibility of minor influence from land/foreshore on measurements at 18 m a.m.s.l. cannot be entirely ruled out. Such local effects might explain some of the scatter in CO2 fluxes at wind speeds below ∼ 5 m s-1, i.e. when the flux footprint was probably closer to land.

Overall, except at the lowest sampling height, air–sea CO2 fluxes by EC show the expected magnitude and direction in the mean. High resolution CO2 fluxes demonstrate significant temporal variability, which is often not well captured by the weekly seawater sampling at L4. We plan to make more regular measurements of seawater pCO2, SST, and salinity within the flux footprint in the future (e.g. as discrete water samples or using a semi-automated dissolved measurement system on Plymouth Marine Laboratory's research vessel Quest), which will enable a direct estimate of the CO2 gas transfer velocity in a coastal environment.

We use historical observations to assess the validity of the EC CH4 fluxes since dissolved CH4 was not measured during 2014–2015. Surface CH4 saturation values of around 2000 % were measured at the mouth of the Tamar Estuary in spring 2001 by Upstill-Goddard and Barnes (2016). At a SST of 10 ∘C and wind speed of 10 m s-1, CH4 saturation of 2000 % implies a predicted CH4 flux of ∼ 0.2 mmol m-2 d-1 (kW from Nightingale et al., 2000). Moving further out from the estuary mouth, dissolved CH4 concentration is expected to decrease due to dilution and oxidation. A strong inverse relationship between CH4 concentration and salinity has been demonstrated by previous investigators (e.g. Upstill-Goddard et al., 2000), with higher CH4 concentrations found in fresher waters. According to the compilation by Bange (2006), typical seawater saturations of CH4 range from 110 to 340 % in the shelf waters of the North Sea, resulting in fluxes on the order of 10 µmole m-2 d-1.

Over the three measurement periods presented here, mean EC CH4 fluxes ranged between 16 and 30 µmole m-2 d-1 in the south-west wind sector, with peak emissions above ∼ 50 µmole m-2 d-1 (Figs. 10 and 11). As with CO2, during April–June 2015 the smaller mean CH4 flux computed from the Windmaster Pro anemometer than from the R3 is primarily due to signal dropouts in the former during rainy, windy conditions (Table 1). The cospectra of CH4 are noisier than those of CO2 (Fig. 12) but demonstrate the expected spectral shape. The lowest mean CH4 fluxes were observed at a sampling height of 15 m a.m.s.l., when the flux footprint should be the closest to shore. This suggests that surface waters, rather than the foreshore/land, are the predominant source of CH4 at PPAO. In other words, the EC CH4 fluxes during the low mast period in May–June 2014 are likely underestimates of air–sea transfer.

CH4 fluxes from the north-east wind sector (the direction of Plymouth Sound) are on average 2–3 times higher than fluxes from the south-west (Fig. 15), suggesting higher CH4 concentrations in the Tamar Estuary outflow than in open water. CH4 fluxes from the south-west show a significant but weak relationship with wind speed (r=0.33 during June–July 2014; r=0.25 during April–June 2015; p < 0.05). The weak relationship between CH4 flux and wind speed was likely in part due to variable seawater CH4 concentrations. CH4 emissions do not obviously vary with time of day but they tend to be higher during incoming (rising) tide than during outgoing (falling) tide. In Fig. 16, CH4 fluxes from the south-westerly direction (April–June 2015) are plotted against hours after low water (low tide occurs at hour zero; high tide occurs near hour six). The median, 25, and 75 % percentiles within each hour bin are also shown. The largest average CH4 emissions are observed in the first ∼ 4 h after low tide, while CH4 fluxes during the falling tide are lower and less variable. Mean CH4 fluxes were also ∼ 50 % higher during spring tide (here limited to daily tidal amplitude > 4 m) than during neap tide (daily tidal amplitude < 3 m). These patterns are consistent with an incoming tidal current that pushes the CH4-rich surface outflow from the Tamar Estuary around the Rame peninsula (Uncles et al., 2015).

To further examine the influence of the Tamar estuarine plume, a 3-dimensional hydrodynamic Finite Volume Community Ocean Model (FVCOM, Chen et al., 2003) was run for April–June 2011 with tidal forcing at the boundaries (TPXO, Egbert et al., 2010), surface wind (Met Office Unified Model, Davies et al., 2005), surface heating (NCEP Reanalysis-2, Kanamitsu et al., 2002), and river input (E-HYPE, Donnelly et al., 2012) at variable resolution (15 km at the open boundaries near the shelf edge and 150 m near the Plymouth Sound). The model predicts that within 1 km south-south-west of Penlee, surface layer (∼ 0.2 m thick) salinity tends to be lower during rising tide (about 33.4–33.7) than during falling tide (about 33.9–34.1). This suggests a greater freshwater influence from the Tamar at the surface during rising tide, qualitatively consistent with our CH4 flux observations. Natural processes other than direct air–sea gas transfer (e.g. ebullition) could also contribute to the variability in CH4 fluxes. Quantifications of the temporal/spatial seawater CH4 distribution within the PPAO flux footprint and measurements of the pelagic/benthic cycling of CH4 is essential for addressing this uncertainty.

CH4 emissions of a few tens of µmole m-2 d-1 at PPAO are higher than estimates for the open ocean (e.g. Forster et al., 2009) and are lower than previous measurements over other aquatic systems. Kitidis et al. (2007) measured a CH4 emission of 63 µmole m-2 d-1 using a floating chamber in the Ria de Vigo (a large coastal embayment), consistent with wind-driven turbulent diffusivity models for the conditions at the time of the chamber deployment. These authors also estimated fluxes up to 170 µmole m-2 d-1 during periods when the chamber was not deployed. With an open path sensor Podgrajsek et al. (2014) recently measured CH4 emissions from a Swedish lake using the EC technique. Lake CH4 emissions range from near zero during the day to over 20 mmole m-2 d-1 at night (3 orders of magnitude higher than observations at PPAO). Aircraft mixing ratio measurements suggest that CH4 emissions from the partially ice-covered Arctic are 4–5 times larger than mean emissions at PPAO (Kort et al., 2012). Our observations and estimates of the CH4 flux uncertainty suggest that an EC system such as the one employed here should be able to quantify emissions from those CH4 hotspots.