Seasonal observations of OH and HO 2 in the remote tropical marine boundary layer

Introduction Conclusions References


Introduction
The hydroxyl radical, OH, is the dominant daytime oxidant in the troposphere.A major pathway for its formation is via the reactions OH is closely coupled to the hydroperoxy radical, HO 2 , so that they are often referred to collectively as the family HO x (= OH + HO 2 ).The key process for the formation of HO 2 in this region is the reaction which is also one of the main loss processes for OH in a clean environment.In the absence of NO, OH can then be reformed from HO 2 via the reaction Another important process for the removal of tropospheric OH is through its reactions with CH 4 and other volatile organic compounds (VOCs) to form peroxy radicals, RO 2 In environments where the levels of NO are very low, the concentration of HO 2 is controlled by the loss processes HO 2 can also undergo heterogeneous loss with atmospheric aerosol, and this process has been shown to be important in the MBL (see, for example, Sommariva et al., 2004;Smith et al., 2006;Sommariva et al., 2006;Whalley et al., 2010).In environments with high concentrations of NO, the reactions play an active role in the local chemistry.It is worth noting that the alkoxy radical, RO, generated in Reaction (R11) can react with O 2 to form HO 2 , as in Reaction (R7).
It has also been shown that the reaction of HO 2 with halogen oxides, XO (where X = Br, I) is significant in marine environments (Bloss et al., 2005b;Smith et al., 2006;Sommariva et al., 2006;Kanaya et al., 2007).The HOI and HOBr produced from Reaction (R12) can then either be removed through heterogeneous reaction with aerosol or photolysis to produce OH and a halogen atom, X.This halogen atom can then regenerate XO via the reaction Read et al. (2008) and Mahajan et al. (2010) showed that the inclusion of the chemistry of halogen oxides, present at only a few pptv, was vital in order for model simulations to reproduce their observations of ozone -the key precursor for OH in the remote MBL -at the Cape Verde Atmospheric Observatory (CVAO) in 2007.
The rate of primary production of OH from Reactions (R1) and (R2) is given by where J (O 1 D) is the photolysis rate of ozone and where the rate constants k O 1 D+H 2 O (2.1 × 10 −10 cm 3 molecule −1 s −1 ), k O 1 D+N 2 (3.1 × 10 −11 cm 3 molecule −1 s −1 ) and k O 1 D+O 2 (4.0 × 10 −11 cm 3 molecule −1 s −1 ) are for the respective reaction and quenching processes of O( 1 D) (Atkinson et al., 2004) at T ∼ 298 K, if all other removal processes for O( 1 D) are insignificant.The rate of production of OH, in the absence of NO, can thus be defined as where i represents the other photolytic processes that may lead to generation of OH (v i is the stoichiometry of the process) and the last term is the total loss of OH to its sinks, such as the major reactions with CO and CH 4 and the minor contributions from reactions with other VOCs, such as acetaldehyde.For remote environments containing low concentrations of alkenes and other non-methane hydrocarbons, such as at the CVAO, the production of OH from the reaction of O( 1 D) with water vapour has been shown to dominate the production of OH.Using the Master Chemical Mechanism, Whalley et al. (2010) performed a rate of production analysis for OH for this site and calculated that P (OH) constitutes at least 75 % of the total rate of production of OH.Assuming that that the rate of OH production is dominated by P (OH) leads to the following steady-state expression: [OH] = where τ OH is the lifetime of OH with respect to its loss to all sinks.For a constant lifetime, a plot of [OH] against P (OH) should be linear with slope τ OH .This equation can also be rewritten in terms of J (O 1 D) as where a represents the influence of all chemical sources and sinks, b accounts for the effect of combining all photolytic processes that produce OH (i.e.photolysis of O 3 as well as NO 2 , HOI, H 2 O 2 , etc.) into a single power function of J (O 1 D) (see Ehhalt and Rohrer, 2000), and c is the contribution from all light-independent processes.Equations ( 4) and ( 5) have been used successfully to explain the variability of OH in different chemical schemes (e.g.Creasey et al., 2002Creasey et al., , 2003;;Berresheim et al., 2003;Smith et al., 2006).
In the absence of NO, the rate of production of HO 2 is given by As the rate of Reaction (R4) is slow compared to Reactions (R8) and (R9), and the rate of loss of CH 3 O 2 through reaction (R6) is slow compared to the rate of production, the steadystate concentration of HO 2 can be expressed as 2k 5 + k 7 α × P (OH)×τ OH (7) where α = [CH 3 O 2 ]/[HO 2 ].For constant values of [CO], α and τ OH , a plot of [HO 2 ] against √ P (OH) should be linear.As can be seen in Eqs. ( 1) and ( 2), the rate of production of OH is controlled by the concentration of water vapour and J (O 1 D).Thus, the high solar irradiance and the warm, humid conditions in the tropics lend themselves favourably to OH generation.OH can then react rapidly with the many trace VOCs found in the troposphere, initiating their oxidation and, ultimately, the removal of these species from the atmosphere.This process is of particular importance when considering the role of OH in constraining the atmospheric budget of methane, the third most abundant greenhouse gas and second only to CO 2 of the long-lived greenhouse gases in terms of radiative forcing (Forster et al., 2007).Lawrence et al. (2001) used a global model to estimate that ∼75 % of atmospheric methane is oxidized between 30 • N and 30 • S. Bloss et al. (2005a) used the GEOS-CHEM model to estimate that 80 % of methane is removed in the tropical troposphere through OH-initiated oxidation, with as much as 25 % of the total occurring in the MBL.Therefore, reliable measurements of OH in tropical areas are of crucial importance for understanding the global oxidizing capacity of the troposphere and future climate change.
Simultaneous measurements of OH and HO 2 in the tropical boundary layer are still relatively sparse compared to the number of studies at mid-latitudes in the Northern Hemisphere, for example (see Heard and Pilling, 2003, and references therein).In recent years, the number of measurement campaigns in the tropics has increased, and Table 1 provides a brief summary of the results of previous studies in tropical and remote MBL regions.Ground-based measurements of OH have been made at the Mauna Loa Observatory in Hawaii (Tanner and Eisele, 1995;Hoell et al., 1996;Eisele et al., 1996), although that station is ∼3.5 km above sea level and, as such, the conditions are closer to free tropospheric than typical of the boundary layer.Airborne measurements of HO x were made during the Pacific Exploratory Missions (PEM) (Hoell et al., 1999;Mauldin et al., 1999;Raper et al., 2001;Tan et al., 2001;Mauldin et al., 2001) and the Transport and Chemical Evolution over the Pacific (TRACE-P) campaign (Jacob et al., 2003;Mauldin et al., 2003;Cantrell et al., 2003;Olson et al., 2004), but measurements in the tropical boundary layer in both studies were limited.HO xmeasurements have also been made over the tropical Atlantic Ocean and the pristine Amazon rainforests of Suriname, Guyana and French Guiana (Lelieveld et al., 2008); on two separate campaigns in the Mexico City Metropolitan Area (Shirley et al., 2006;Dusanter et al., 2009a); at the Pearl River Delta, China (Hofzumahaus et al., 2009); over Western Africa as part of the African Monsoon Multidisciplinary Analyses (AMMA) campaign (Commane et al., 2010;Stone et al., 2010); and in Malaysian Borneo as part of the Oxidant and Particle Photochemical Processes (OP3) project (Hewitt et al., 2010;Pugh et al., 2010;Whalley et al., 2011;Stone et al., 2011a).However, these studies were predominantly in or over either forested or polluted regions, so that the chemistry of HO x was heavily influenced by processes other than Reactions (R1)-(R12).There have been several studies in the remote MBL outside of the tropics -for example, at Cape Grim, Tasmania, as part of the Southern Ocean Photochemistry Experiment (SOAPEX-2) (Creasey et al., 2003), and on remote Japanese islands (see Kanaya and Akimoto, 2002 and the references therein; Kanaya et al., 2007).However, measurements of OH and HO 2 in the remote tropical MBL are still very limited.Concentrations of OH were measured using Differential Optical Absorption Spectroscopy (DOAS) onboard the R/V Polarstern in the tropical Atlantic as part of the Air Chemistry and Lidar Studies of Tropospheric and Stratospheric Species on the Atlantic Ocean (ALBATROSS) project in late 1996 (Brauers et al., 2001).In that study, it was found that OH followed a clear diurnal cycle, with maximum concentrations of about 7 × 10 6 molecule cm −3 at local noon, within the range observed above the tropical Pacific during the PEM.Also, [OH] and [HO 2 ] was measured by Whalley et al. (2010) at the CVAO as part of the Reactive Halogens in the Marine Boundary Layer Experiment (RHaMBLe) in 2007 -the maximum daytime concentrations of OH and HO 2 were 9 × 10 6 and 6 × 10 8 molecule cm −3 , respectively.
Almost all these studies of HO x have been "short-term"; i.e. the observation periods are typically of the order of a few weeks at one particular period of the year.For example, the measurements of Whalley et al. (2010) at the CVAO were conducted over just 11 days, with only 5 days of OH data.To date, there has been only one long-term observational study of the seasonal change in [OH] in the troposphere (Rohrer and Berresheim, 2006), a five-year dataset measured at the Meteorological Observatory Hohenpeissenberg (MOHp; 47.8 • N, 11.0 • E) at ∼1000 m above sea-level in rural southern Germany.However, there have been no studies of seasonal trends in [OH] in the tropics where the levels of H 2 O vapour and J (O 1 D) are strongly favourable for its formation and critical for constraining atmospheric methane.To that purpose, measurements of OH and HO 2 were performed at the CVAO as part of the Seasonal Oxidant Study (SOS) during three distinct periods of 2009: 27 February-8 March (SOS1), 6-16 June (SOS2) and 1-15 September (SOS3).This paper reports the observations of OH and HO 2 from this study and their respective dependencies on J (O 1 D) and P (OH).The results are compared with the previous measurements made at the same site by Whalley et al. (2010) and with other tropical locations, and the seasonal variability will be discussed.

Measurement of OH and HO 2 radicals
Fluorescence Assay by Gas Expansion (FAGE) has been well-demonstrated as a powerful tool for atmospheric measurements of HO x (Heard and Pilling, 2003).OH and HO 2 were monitored at the CVAO using the University of Leeds' aircraft-FAGE instrument in a ground configuration.The instrument has been described elsewhere (Commane et al., 2010), but a brief description will be provided here.The instrument sampled ambient air through a pinhole and the gas flowed through detection cells for OH and HO 2 in series.The laser-induced on-resonance fluorescence following excitation of the transition 2) of OH at λ = 308 nm was used as the basis for the detection of the hydroxyl radical.A reference cell, containing a heated filament used to thermally decompose water vapour to yield OH, was used to identify the wavelength at which the fluorescence of OH at that transition was greatest.Between the two detection cells was an injection port for NO, so that HO 2 is chemically converted to OH, which is subsequently detected at λ ∼ 308 nm.The instrument was run at a low internal pressure (typically, P = 2.2 Torr) in order to reduce the effect of quenching on the fluorescence lifetime of OH and scattering of the laser light.This methodology allowed the use of temporal gating of the detector, a channel photomultiplier (CPM; Perkin-Elmer C943P), so that the fluorescence could be discriminated from the laser pulse.The CPM was gated off until ∼110 ns after the laser pulse (full-width half-maximum = 35 ns) to avoid detector saturation.Note that an in-house computer program automatically corrected the timing of the CPM-gates for any changes in the timing of the laser pulse.Photons from fluorescence and scattered light were then recorded using a two-channel gated photon counter.The first gate (Gate A, width 1 µs) recorded both fluorescence and background light (laser, solar and dark counts).The second gate (Gate B, width 20 µs) was set to switch on beyond the lifetime of the fluorescence at a delay of 5 µs and hence counted only background light (solar and dark counts).Summing over 1 s, the signal (count s −1 ) due to fluorescence and laser scatter was calculated as signal A -(signal B/20).The laser scatter was removed from that signal by subtracting the observed signal when the laser is tuned off the OH resonance.Typically, 300 online and 60 offline onesecond data points were collected for OH.Scanning the laser wavelength to the peak OH transition at the beginning of each measurement cycle typically took ∼120 s, so that each measurement cycle lasted usually about 540 s.Because the HO 2 detection will inherently observe OH from the sampled ambient air as well as that generated in situ from chemical conversion of HO 2 to OH, the injection of NO started 60 s after the instrument began recording online fluorescence signals, and that signal was subtracted from the subsequent signal with NO flowing to give only the signal due to converted HO 2 .Thus, there are typically 240 online and 60 offline one-second data points for HO 2 .The fluorescence signals are then normalised with respect to the laser power entering each detection cell (typically 10-30 mW in the OH cell, 3-6 mW in the HO 2 cell).
The observed fluorescence signal, S OH (count s −1 mW −1 ), were related to [OH] by where C OH is the sensitivity of the instrument (count s −1 mW −1 molecule −1 cm 3 ) with respect to OH and can be defined empirically by the expression where D is a function of pressure-independent parameters (e.g.laser power, collection efficiency of the optics, quantum yield of detector), T OH is the transmission of OH within the instrument, f gate is the fraction of light sampled within the timing gate and f is the fluorescence quantum yield from excited OH.The equations linking S HO 2 , C HO 2 and [HO 2 ] are analogous to Eqs. ( 8)-( 9).The sensitivity at a given cell pressure and water vapour concentration was calibrated by measuring the observed fluorescence of OH and HO 2 generated in a flow of air from the photolysis of a known concentration of water vapour at λ = 184.9nm (see Commane et al. (2010) for more details).Calibrations were performed regularly under the same conditions (i.e.laser power, instrument pressure) as for ambient sampling when possible and verified in the laboratory after the campaign.
The mixing ratio of ambient water vapour during the measurement periods (2-3 %) was greater than could be achieved in a field calibration (∼1.2 %).Hence, the calibrated sensitivities were corrected for the quenching of the OH fluorescence by water vapour using known rate parameters that contribute to the terms f gate and f where τ is the radiative lifetime of the excited OH state (A 2 + ) in the absence of quenchers (688 ns; German, 1975), is the total rate of removal of excited OH via radiative and quenching processes (calculated using the quenching rate constants of N 2 , O 2 and water vapour at T ∼ 293 K reported in Bailey et al. (1997) and Bailey et al., 1999) and t 1 and t 2 are the start and cut-off times for the photonsampling gates, respectively.By comparing the relative values of f gate and f for the conditions of the calibration and atmospheric sampling, one can correct the sensitivity of the instrument to OH and HO 2 .These corrections lowered the instrument sensitivity by 10-30 %, raising the calculated [OH] and [HO 2 ] by the same relative proportion.The quenching-corrected instrument sensitivity (ct s −1 mW −1 molecule −1 cm 3 ) with respect to OH was found to decrease through SOS, from ∼1.5 × 10 −7 to ∼0.6 × 10 −7 , possibly because of reduced transmission of OH through the instrument or contamination/aging of the optics.The uncorrected sensitivity towards HO 2 was generally consistent at ∼1.9 × 10 −7 ct s −1 mW −1 molecule −1 cm 3 throughout the campaigns, but increasing concentrations of water vapour from SOS1-3 led to effective instrument sensitivities (ct s −1 mW −1 molecule −1 cm 3 ) of ∼1.8 × 10 −7 in SOS1 to ∼1.3 × 10 −7 in SOS3.Recent work by Fuchs et al. (2011) has highlighted a possible interference towards HO 2 -detection by LIF from the chemical conversion of alkene-and aromatic-derived peroxy radicals to OH by NO which is added inside the fluorescence cell, with alkanederived peroxy radicals exhibiting negligible interference.The concentrations of alkenes such as ethene and propene at the CVAO have been observed to be only a few pptv (see Carpenter et al. (2010) and Table 2).It is therefore expected that peroxy radicals derived from such alkenes generate only a very small HO 2 interference.Calculations using a box model incorporating the full Master Chemical Mechanism and constrained using the VOC measurements at the site (Table 2) show that ∼90 % of the RO 2 species are HO 2 and CH 3 O 2 , with other significant species being CH 3 C(O)O 2 (5 %) and C 2 H 5 O 2 (0.9 %), all of which showed no HO 2 interference during laboratory experiments.The RO 2 species OHC 2 H 4 O 2 (0.6 % of RO 2 total) and OHC 3 H 6 O 2 (0.6 %), derived from ethene and propene, respectively, do give some HO 2 interference (measured in the laboratory as 40 % with the experimental configuration used here), but their very low abundance leads to only a very small HO 2 interference.However, for other environments, such as forested or urban, where the concentrations of alkenes and aromatics may be considerably higher, HO 2 interferences from such RO 2 species may need to be taken into account.
The limits of detection (LOD) of OH and HO 2 , were defined as where S/N is the signal-to-noise ratio (taken as 1 for both species), C is the instrument sensitivity, P is the laser power (mW), m is the number of online 1 s samples (typically 300 for OH and 240 for HO 2 ), n is the number of 1 s offline samples (generally 60 for both species) and σ b is the standard deviation of the offline signal (ct s −1 ) calculated using Poisson statistics and includes the contributions due to solar scatter, laser scatter and dark counts.The limits of detection of OH (five-minute averaging) and HO 2 (four-minute averaging) over the whole SOS were in the ranges (2-11) × 10 5 and (6-13) × 10 5 molecule cm −3 , respectively.A range of LODs was observed for both species because of the variability in both the offline signal and the quenching-corrected sensitivities across the whole SOS period.The uncertainty (2σ ) in the measurements of [OH] and [HO 2 ] can be calculated as the square root of the sum of squares of the errors in the contributing variables, and was estimated to be ∼32 % for both species.
The CVAO is located on the island of Sao Vicente (16.86 • N, 24.87 • W), approximately 10 m above sea level and about 50 m from the coastline; a full description of the site is given in Carpenter et al. (2010).Levels of NO are typically a few pptv (Lee et al., 2009), so the observatory is an ideal location for monitoring atmospheric composition in a clean, remote tropical MBL.The laser system and dataacquisition electronics were located inside an air-conditioned standard shipping container.The sampling inlet of the instrument was on the roof of this container inside an insulated aluminium box, to protect against sea salt and water.To reduce the amount of scattered light detected, and also improve the transmission of OH and HO 2 through the instrument, a tube made of black nylon was placed inside the inlet between the sampling pinhole and the detection cell for OH, thus improving both the sensitivity of the instrument and the limit of detection.

Ancillary measurements
The photolysis rate of ozone to produce O( 1 D), J (O 1 D), was measured using a 2π-filter radiometer mounted on the roof of the container, about 2 m from the FAGE inlet, and was at no time in the shade of local influences (e.g.other instruments or structures at the CVAO).The signal from the radiometer was recorded as a voltage on the PC used to control the FAGE instrument and was later converted to a photolysis rate (s −1 ) using the parameters (including factors to correct for solar zenith angle, ozone column density and temperature) obtained during the intercomparison study in Julich (Bohn et al., 2008).These data showed almost 1:1 agreement (slope = 1.02;R 2 = 0.97) with the output of the University of Leicester's spectral radiometer, which was positioned within a few metres at the same altitude, when the two radiometers were running simultaneously during SOS3.
Other ancillary measurements were made at the site, including those of NO x (i.e.NO and NO 2 ), CO, O 3 , VOCs (including short-chain alkanes and alkenes, acetaldehyde, acetone, methanol), relative humidity and wind direction, to allow characterisation of the atmospheric composition at the site.The majority of these measurements are part of a longterm dataset that has been running since 2006.Further details of the relevant instrumentation can be found in Carpenter et al. (2010).Table 2 shows average measured values and standard deviations of important species for the periods with simultaneous measurements of OH and HO 2 during SOS1 and 2.

Analysis of the data
The data were filtered for unfavourable meteorological conditions (e.g.heavy rainfall, as experienced during SOS3) and obvious influence of local pollution sources.Figure 1 shows examples of how pollution sources, such as the site power generator or passing fishing boats, distort the HO 2 :OH ratio on short timescales because of the conversion of HO 2 to OH by NO.Data were filtered out if the concurrent wind direction and speed data suggested that the sampled air mass was from an obviously polluted source, such as generator exhaust fumes.The fast (i.e. 1 s) NO x data would be required to allow identification of short pollution episodes from other sources, such as fishing boats; unfortunately, that data were not available for the bulk of the HO x -measurement period.The predominant wind direction was NE from the open Atlantic ocean (∼94 % of the time), with the generator located southerly to the HO x -instrument, so the wind direction (i.e. between 100 • and 300 • ) could be used to identify data affected by local pollution influences and then removed from the final analysis.
The OH-and HO 2 -signals were corrected for any instability in the wavelength of the laser; the full-width halfmaximum of the Q 1 (2) transition at λ = 308 nm is very narrow (calculated using the procedure described in Dorn et al. (1995) to be less than 1 pm at P ∼ 2 Torr), so that the magnitude of the fluorescence is highly sensitive to the incident wavelength.Figure 2 shows two examples of the HO 2 -signals (black lines) as well as the corresponding values of the scaled reference signal (red).Note that laser  power, instrument and reference cell pressures and J (O 1 D) were constant (albeit for instrument noise) within each run and the early peaks in the reference signal correspond to where the laser wavelength was scanned to find the maximum of the OH fluorescence.Figure 2a shows a run where the laser wavelength appears to be stable, as can be seen by the flat profile of the reference signal after ∼100 s. Figure 2b clearly shows that the HO 2 signal appeared to follow the same pattern as the reference signal, such that the linear correlation between the two signals was reasonably strong (R 2 = 0.80).Because there was no evidence of significant changes in J (O 1 D), laser power or the pressures inside the instrument and the reference cell, this effect was most likely as a result of small fluctuations in the laser wavelength.By normalising the OH-and HO 2 -signals to the reference signal, one is able to correct for the small variations in λ, as can be seen by the green line in Fig. 2b.A similar correction applied to the data in panel (a) suggested little change.Thus, this correction for small changes in wavelength, applied to the raw signals for both species, allowed data to be included in the analysis that may have otherwise been excluded.

Summary of data and synoptic conditions
There were 33 days of OH and HO 2 observations, made in three intensive periods over seven months, constituting 500 014 and 413 205 one-second data points for OH and HO 2 , respectively.Figures 3-5 show the time-series for each of the three campaigns.The shaded areas in the top panels represent the contributions to the air mass from five possible sources, as calculated using the NAME dispersion model (Ryall et al., 2001) using the technique described in detail in Carpenter et al. (2010) -Atlantic continental air from over North America (yellow), Atlantic marine (blue), polluted marine air from over Europe (red), African coastal (green) and Saharan dust (brown).The middle two panels show the five-minute averaged [OH] (black dots) and four-minute averaged [HO 2 ] (blue dots), respectively.The bottom panels show the supporting measurements of J (O 1 D) (red line), O 3 (blue dots), CO (green dots), NO (black dots) and NO 2 (dark yellow dots) for each 5-min HO x -data point.It should be noted that gaps in the time-series are either because the measuring instruments were not operational, the meteorological conditions were highly unfavourable during those times (e.g.heavy rainfall, very calm wind or strong wind from the direction of local pollution sources), or the HO x -data failed certain tolerances (e.g.laser instability, lack of NO flow).There were no measurements of NO x from 5-8 March, 11-16 June and no data for either NO x or CO for the whole of SOS3.It can be seen that HO x was observed for 8, 11 and 14 days for SOS1, 2 and 3, respectively, and that both OH and HO 2 followed clear diurnal cycles.

SOS1 (28 February-8 March 2009)
The conditions during the HO x -measurements for SOS1 (28 February-8 March) were warm (T ∼ 294 K) with very little rainfall.The average relative humidity was ∼77 %, corresponding to a concentration of water vapour of ∼4.7 × 10 17 molecule cm −3 .Winds were typically north- easterly with speeds greater than 4 m s −1 .The air mass had strong contributions from Atlantic marine air and from the African coast for most of SOS1.Polluted marine air, having passed over Europe a few days previously, contributed up to ∼20 % of the air mass in the first half of this period.Saharan dust contributed up to 10 % in the first half of the HO x -measurement period, and Atlantic continental air (having passed over North Africa a few days previously) made similar contributions in the latter half.CO and O 3 were relatively constant at ∼110 and ∼40 ppbv, respectively, although these levels were ∼20 % higher on 4 and 5 March.There were generally clear skies, although conditions were overcast from 4-6 March, as can be seen in the behaviour of J (O 1 D).Consequently OH and HO 2 showed more structure in their diurnal profiles on those days compared to the rest of the measurement period.Data were recorded for 27 February and 1 March, but were excluded from the final analysis because there may have been local pollution sources those days.Very few measurements of [HO 2 ] were made on 6 March because the supply for the injection of NO was closed to confirm that the cylinder was not in any way a contaminating source of NO at the site.

SOS2 (6-16 June 2009)
The HO x -measurement period for SOS2 was characterised by warmer (T ∼ 297 K), with very few periods of cloud cover or rainfall.The mean relative humidity was ∼82 %, so that the average concentration of water vapour was ∼5.6 × 10 17 molecule cm −3 .As for SOS1, winds were typically north-easterly with speeds more than 4 m s −1 .This period experienced the cleanest conditions, with Atlantic marine air as the main source of the sampled air mass for the whole period, although coastal African air contributed up to ∼40 % on some days.There was little contribution from polluted marine air, Saharan dust or Atlantic continental air.The noontime peak in J (O 1 D) was quite consistent at ∼3.7 × 10 −5 s −1 for the whole period.CO and O 3 were higher, at ∼100 and ∼30 ppbv, respectively, from 7-10 June compared to the rest of SOS2.These dates correspond to the highest peak [OH] and [HO 2 ] for SOS2 of ∼8 × 10 6 and ∼4 × 10 8 molecule cm −3 , respectively.From 11 June to the end of SOS2, there was ∼80 ppbv of CO and ∼20 ppbv of ozone, and the daytime peak [OH] and [HO 2 ] were relatively constant at ∼5 × 10 6 and ∼2.7 × 10 8 molecule cm −3 , respectively.

SOS3 (1-15 September 2009)
The meteorological conditions were the most variable during SOS3 with periods of strong daylight interspersed with heavy rainfall, although the temperature remained relatively constant at ∼300 K.It was impossible to make HO xmeasurements during the heavy rainfall, which caused flooding at the CVAO, so that there are large gaps in the timeseries of [OH] and [HO 2 ] for those periods.There was also significant electrical power disruption at the site, affecting the ancillary measurements severely -there were no data for NO x or CO and limited coverage of O 3 and relative humidity.Nevertheless, there were ∼211 200 and 169 920 one-second data-points for OH and HO 2 , respectively, corresponding to ∼700 individual five-minute measurement runs and almost 60 h of HO x -observations.Winds were typically north-easterly and above 4 m s −1 from 1st to 4th, with the relative humidity ∼85 % (∼4.7 × 10 17 molecule cm −3 of water vapour).The daytime peak concentrations of OH and HO 2 for these days were ∼4 × 10 6 and ∼2 × 10 8 molecule cm −3 , respectively.The conditions at the site were very different on the 5th.The relative humidity ranged from a maximum of 89 % in the morning to a minimum of 69 % at 03:00 p.m. local time.Also, the wind swirled at speeds below 2 m s −1 , which led to the HO x instrument sampling air from the exhaust of the generator.Fast conversion of ambient HO 2 to OH by NO led to observations of [OH] exceeding 10 8 molecule cm −3 .Measurements where such chemical influence was obvious -over half the data from that daywere excluded from the final analysis, but the remaining data after filtering suggested higher concentrations of OH and HO 2 compared to the rest of SOS3.From the 6th-8th, the north-easterly winds at speeds more than 4 m s −1 returned and the relative humidity was again typically ∼85 %, and there is good data coverage for those days -the peak daytime concentrations of OH and HO 2 were about 2.5 × 10 6 and 3 × 10 8 molecule cm −3 , respectively.An almost full night of measurements of [OH] and [HO 2 ] was made from 7th-8th, the only time possible during the whole SOS; the observations for this night will be discussed further in the following sections.There was then a marked change in the relative humidity at the site.The concentration of water vapour from 1st to 9th was in the range 7-8 × 10 17 molecule cm −3 , compared to 6-7 × 10 17 molecule cm −3 (relative humidity ∼75 %) for 10th-12th.The skies were generally clear, and the daytime peak [OH] and [HO 2 ] were in the ranges 6-10 × 10 6 www.atmos-chem-phys.net/12/2149/2012/and 3-4 × 10 8 molecule cm −3 , respectively.The wind speed dropped below 2 m s −1 in the late afternoon of the 12th and remained low through the 13th, similar to the conditions of the 5th, with the relative humidity rising to ∼92 %.However, there was no evidence of ambient HO 2 -OH conversion.Heavy rainfall again prevented measurements of OH and HO 2 on the 14th and most of the 15th.

Seasonal behaviour
Figure 6 shows the median half-hourly averaged diurnal profiles of OH, HO 2 and P (OH) for SOS.The error bars represent the 1σ day-to-day variability in the half-hour averaged data and, although the diurnal profiles for the three measurement periods do agree within the combined 1σ limits, the data around local noon (i.e.10:00 to 14:00) for SOS1 is statistically different (at 95 % confidence limit of a Student's t-test) to the data in the same timeframe for both SOS2 and 3.The peak values of both [OH] and [HO 2 ] followed the trend SOS1 < SOS2 ∼ SOS3, consistent with the trends in J (O 1 D) and water vapour concentration.Similar qualitative trends were observed for the sum of [HO 2 ] and [RO 2 ] recorded by the University of Leicester's PERCA instrument (see Carpenter et al., 2010).However, P(OH) appears to follow the opposite trend (i.e.SOS1 > SOS2 ∼ SOS3) because the decrease in [O 3 ] from SOS1 is larger than the combined increases in J (O 1 D) and water vapour concentration.Possible reasons for this discrepancy will be discussed in Sect.4.1.
The median ratio [HO 2 ]/[OH] around local noon (10:00 to 14:00) was ∼75 for all three periods.During daylight, it may be expected for this ratio to vary inversely with [NO] because of conversion of HO 2 to OH via Reaction (R4).Simultaneous coverage of [NO] (15 min averaging), [HO 2 ] and [OH] was limited to only 22 % of the total HO x coverage -4 days in SOS1, 5 days in SOS2, with no NO x measurements in SOS3.Nevertheless, by taking half-hour averages of concurrent [NO], [OH] and [HO 2 ] between 08:00 and 17:00, it was found that [HO 2 ]/[OH] was ∼112 in air containing less than 5 pptv of NO and ∼74 for greater than 5 pptv of NO.This trend is consistent with behaviour observed in other environments (e.g.Carslaw et al., 2001;Creasey et al., 2002;Kanaya et al., 2001b), although it must be noted that the dataset in this case is limited to only 92 half-hour averages.to the whole dataset.Tables 3 and 4 show the results of forced fits of Eq. ( 13) to the observations (i.e.where [OH] is set to a linear relation (b = 1) and [HO 2 ] is set to a squareroot relation (b = 0.5) -as predicted by the simple steadystate Eqs. ( 4) and ( 7), respectively, -and a and c are allowed to vary), with the results of floating fits (i.e.a, b and c allowed to vary) given in parentheses.For clarity, only the standard errors in b for the floating fits are shown, and it can be seen that the values of b from the forced and unforced fits to each dataset do not necessarily agree within error.This disagreement suggests deviations from the simple approach used here caused by, for example, the reactions of HO 2 with O 3 or the photolysis of HO x -precursors other than ozone, or a change in the relative importance of HO x -sinks.It is worth noting that the values of R 2 were virtually the same for the two types of fitting conditions.For ease of comparison with other studies, the following discussion will focus only on the results of the forced fits.

Dependence of [OH] and [HO
The results show that [HO 2 ] has excellent correlations with J (O 1 D) and √ P (OH), for each measurement period (R 2 = 0.88-0.95) and the whole SOS (R 2 = 0.88 and 0.87, respectively), suggesting that ∼90 % of the variability of that species can be explained by a balance between its production from the reaction of OH and CO and its loss through selfreaction and reaction with CH 3 O 2 .These results also imply that the behaviour of HO 2 at the CVAO can be broadly described by the equations  The units of a are molecule cm −3 s (for b = 1) and molecule cm −3 s 1/2 (for b = 0.5), for OH and HO 2 , respectively.The units of c are molecule cm −3 .For ease of comparison, the values displayed in bold are those for forced fits of b = 1 and b = 0.5 for [OH] 13) to the observed five-minute averaged [OH] and four-minute averaged [HO 2 ] using P (OH).
The units of a are s (for b = 1) and molecule 1/2 cm −3/2 s 1/2 (for b = 0.5), for OH and HO 2 , respectively.The units of c are 10 6 molecule cm −3 .For ease of comparison, the values displayed in bold are those for forced fits can be slightly better described by a linear relation with J (O 1 D) (R 2 = 0.59) rather than P (OH) (R 2 = 0.53).Similarly, averaging the data across longer time-periods only slightly improves the fit of Eq. ( 13) to the complete OH dataset.Also, the parameters a, b and c for the raw 5-min data and the increased averaging agree within the standard errors of the fits; for example, two-hour averaging of the OH data results in fit parameters of b = 0.94 (±0.14), c = 0.43 (±0.19) × 10 6 molecule cm −3 and R 2 = 0.62, where the values in parentheses are the standard errors on the results of the fit.Thus, the fit results to [OH] suggest that 50-70 % of the variability of OH in each measurement period can be explained by the primary production process (R1-R2).
Comparing the three campaigns, the values of a do not agree within the standard errors of the fits (for clarity, these are not shown in Tables 3 and 4) and follow the general trend SOS1 < SOS2 ≤ SOS3 for both OH and HO 2 , suggesting that the sources of HO x are greater and/or the sinks are less in the summer months than in winter.In fact, a is equal to the lifetime of OH for a linear fit of [OH] to P (OH).Thus, the results suggest τ OH in the winter (∼0.45 s) is approximately half that in the summer, implying that the sinks of OH were stronger in SOS1 compared to SOS2 and 3.In terms of the relationship between HO 2 and P (OH) (Eq.7) in a low-NO x environment, the parameter a provides information of the mean ratio of [CH 3 O 2 ] to [HO 2 ], α, i.e.
Using average values of [CO] of 2.7 (SOS1) and 2.2 (SOS2) × 10 12 molecule cm −3 and the rate constants k 3 = 2.3 × 10 −13 cm 3 molecule −1 s −1 , k 5 = 6.4 (SOS1) and 6.9 (SOS2) × 10 −12 cm 3 molecule −1 s −1 (the values are different because of the dependence of k 5 on the concentration of water vapour and temperature) and k 7 = 5.3 × 10 −12 cm 3 molecule −1 s −1 , calculated using the recommendations in Atkinson et al. (2004) and Atkinson et al. (2006), provides values of α of ∼4.1 and 2.1 for SOS1 and 2, respectively.These results suggest that HO 2 could constitute up to 20 % of the total budget of peroxy radicals in winter compared to up to a third in the summer, if CH 3 O 2 is the dominant organic peroxy radical.
The reason for the trend in concentrations of OH and HO 2 not following the trend in P (OH) is not immediately clear.
One possibility would be that the trend is a result of errors in the field calibrations -for example, the true sensitivity of the instrument for OH and HO 2 during SOS1 would have to be lower than that from derived from the calibration.The instrument was calibrated by observing the signal from known concentrations of OH and HO 2 generated by the photolysis of water at λ = 184.9nm using light from a mercury pen-ray lamp.An error or deviation in the calibration of the flux from that lamp would lead to a systematic error in the observed instrument sensitivity.However, two different pen-ray lamps were used in the field, and the results of the instrument calibrations were in good agreement.The field calibrations of the fluxes of these lamps at λ = 184.9nm also agreed well with laboratory tests.These observations would suggest that the difference in the trends of [OH], [HO 2 ] and P (OH) was not a result of errors in the instrument calibration.
The most likely explanation would be that there is seasonality in the sinks of HO x .Long-term measurements at the CVAO show that there is a tendency in the levels of VOCs to be higher in winter than in summer months, as shown in Table 2 (see also Read et al., 2009;Carpenter et al., 2010), which may explain, to some extent, the lower OH observed in February compared to June and September.However, one would still expect the reactions with methane and CO to be the dominant sinks for OH, and modelling studies suggest that CO represents 36.7 % and 38.1 % of the total OH loss during SOS1 and SOS2, respectively, while CH 4 represents 14.2 % and 18.0 % of the total OH loss during SOS1 and SOS2, respectively.It is known that the export of Saharan dust across the Atlantic Ocean exhibits strong seasonal behaviour (see, for example, Schepanski et al., 2009).In winter, the dust layer is transported within the trade-wind layer in a south-west direction and at near-surface levels, frequently depositing in the region of the Cape Verde archipelago, while in the summer, the dust remains elevated above the boundary layer and is transported westwards (i.e.there is little deposition to Cape Verde).Figures 3-5 show there was a small but significant contribution to the air mass from Saharan dust for SOS1 (up to ∼15 %) and 3 (up to ∼30 %), but little in SOS2.Also, aerosol sampling measurements at the CVAO revealed approximately half the samples were of high mass concentration (i.e.greater than 60 µg m −3 ) in the winter months compared to very few (∼5 %) in the summer months (see Carpenter et al., 2010).Whalley et al. (2010) used models to show that heterogeneous loss of HO 2 on aerosols is an important process in this region, constituting ∼23 % of the total loss of HO 2 at noontime and leading to a reduction in daytime HO 2 of ∼30 % compared to the modelled scenario with no aerosol uptake.Those authors also showed that the halogen oxides, IO and BrO, were important instantaneous sinks for HO 2 (∼19 %), despite being present at only a few pptv, and that the combined effects of heterogeneous losses and halogen oxide chemistry reduced modelled daytime HO 2 by ∼50 % compared to the modelled scenario without such processes; it should be noted, however, that the impact on OH was only a few percent.It is conceivable, therefore, that the observed winter-summer trends observed in HO 2 in this study may be influenced by the seasonality of the concentrations of both aerosols and halogen oxides.Unfortunately, no measurements of halogen oxides were possible during SOS.The measurements by Mahajan et al. (2010) at the CVAO in 2007 suggested that the concentrations of both IO and BrO were relatively consistent throughout the year, although Whalley et al. (2010) suggested that the day-to-day variability in their measurements of OH and HO 2 could be explained by variability in concentrations of halogen oxides.A detailed modelling study is required to fully assess the chemical influences on the seasonal behaviour of OH and HO 2 , and that will be the focus of a future paper.

Comparison with other measurements
The levels of OH and HO 2 observed during the SOS are similar to measurements in other clean, remote environments (see Table 1).The maximum concentrations of both species observed here are lower than the highest values reported in polluted tropical locations (Shirley et al., 2006;Dusanter et al., 2009a, b;Hofzumahaus et al., 2009), where the chemical regime is much more complex.The most direct comparison can be made to the 2007 measurements at the CVAO.The measurements of OH and HO 2 during RHaMBLe and SOS2 were taken at roughly the same time of year -late May to early June.Figure 9 shows the hourly-averaged median diurnal profiles of OH, HO 2 and P (OH) for those two campaigns (calculated using the typical values of relative humidity and the mixing ratio of O 3 reported in Whalley et al., 2010).The black lines represent the new dataset, the red lines the 2007 data and the error bars represent the 1σ day-to-day variability in the hourly-averaged data.Instrument problems during the 2007 study meant that only HO 2 was measured for the early part of the campaign, with simultaneous observations of OH and HO 2 possible during the last 5 days, so the solid red line in one-second values for the panel C shows P (OH) when OH was recorded and the dashed line is for when HO 2 was measured.There appears to be reasonably good agreement between the profiles of P (OH) for SOS2 and the 5 days that OH was measured during RHaMBLe, at least before local noon.A Student t-test of the OH data for 10:00-14:00 shows no significant statistical difference between the two campaigns at the 95 % confidence level, despite the limited dataset for OH from RHaMBLe.The hourly-averaged concentrations of HO 2 between 10:00 a.m. and 02:00 p.m. are statistically higher for SOS2 than RHaMBLe (at the 95 % confidence level), although they do agree within the 1σ day-to-day variability.Interestingly, P (OH) was higher for RHaMBLe than SOS2, but levels of NO x were lower in 2007 than in 2008 and 2009 (see Carpenter et al., 2010), suggesting that sinks for HO 2 such as heterogeneous loss through interaction with aerosols may have been stronger in the 2007 study.et al., 2001).The values of a from these two studies are close the average SOS value, suggesting that the effects of the chemical processing during the three different campaigns -separated by 13 yr -produced similar dependences of [OH] on J (O 1 D).By taking an average of the a-values, the contribution to OH in the tropical Atlantic MBL from photolytic processes may be described by the expression However, the results shown in Table 3 suggest that there appears to be a seasonal dependence of OH on J (O 1 D), so this expression should only be used as a crude approximation when examining the long-term influence of OH in the tropical MBL.
It should be noted that the correlations between [OH] and J (O 1 D) in the tropical Atlantic MBL are not as strong as the result of Rohrer and Berresheim's five-year study in rural Germany (R 2 = 0.88).Nevertheless, the correlations between [OH] and J (O 1 D) in the tropical Atlantic MBL are stronger than those yielded from measurements in tropical forested and urban environments, where the chemical complexity leads to deviations from the simple steady-state approximation.For example, the airborne study over the Suriname rainforest by Martinez et al. (2008) found that correlations of OH with P (OH) and HO 2 with √ P (OH)were poor in the boundary layer over the forest (R 2 = 0.19 and 0.24, respectively), compared to in the boundary layer over the Atlantic ocean and the free troposphere over both regions (R 2 = 0.47-0.76).Those authors suggested that the poor correlation over the forest resulted from the recycling of OH through its chemistry with isoprene.In an example of an urban study of tropical OH, Dusanter et al. (2009a) reported that only ∼20 % of the variance in OH in Mexico City during MCMA-2006 could be attributed to the variation in J (HONO), and that the correlation between OH and J (O 1 D) was worse.Thus, while Eq. ( 17) may hold reasonably well for OH in the tropical Atlantic MBL, it seems clear that the influence of local chemistry is prohibitive in deriving a uniform expression for the behaviour of [OH] across the tropics.
This work represents the first study of the long-term behaviour of HO 2 and its dependence of J (O 1 D) in any environment.The relationship between [HO 2 ] and the squareroots of both J (O 1 D) and P (OH) agrees well with observations in other clean environments.For example, Creasey et al. (2003) found that the slope of a plot of log [HO 2 ] against log J (O 1 D) for one day of baseline conditions (i.e.very low NO, <2 pptv) during SOAPEX was equal to 0.49, compared to a slope of 0.48 for a similar plot of all the HO 2 data in this study.Deviations from the square-root relation have been observed in regions where NO x chemistry becomes more influential.For instance, Kanaya et al. (2001b) found during ORION99 that the power law dependence of HO 2 on J (O 1 D) was b ∼ 0.5 when NO was below 300 pptv, b ∼ 1 when NO was more than 1000 pptv, and b was between 0.5 and 1 in the intermediate range of NO.Also, this study at the CVAO has shown that ∼90 % of the variability of HO 2 can be explained by J (O 1 D) in a remote environment, compared to only 36 % reported by Holland et al. (2003) for a site in rural Germany, where levels of NO were significant (i.e.several ppbv).

Nighttime measurements of OH and HO 2
Measurements of OH and HO 2 at night (here defined as J (O 1 D) < 10 −7 s −1 , typically before 06:00 a.m. and after 06:00 p.m. local time) were limited, and a prolonged study of nighttime OH and HO 2 was only possible during one night of the SOS -7 September 2009.The solid black lines in Fig. 10 represent the time-series of OH and HO 2 for that night, the dashed line the respective LODs and the red line represents J (O 1 D).It can be clearly seen that HO 2 persists into the night above the LOD, with an average concentration of ∼10 7 molecule cm −3 .The values of [OH] for the night of 7 September 2009 are normally distributed about ∼8 × 10 4 molecule cm −3 , with a standard deviation close to the five-minute LOD, suggesting that OH does not persist at measurable levels at least for that night.Whalley et al. (2010) were not able to perform nighttime measurements of OH at CVAO in 2007, so no comparison can be made.However, they did observe ∼10 7 molecule cm −3 of HO 2 at night, similar levels to those observed on the night of 7 September 2009.In fact, the mean concentration of HO 2 when J (O 1 D) < 10 −7 s −1 was ca. 10 7 molecule cm −3 for SOS1-3, above the averaged LOD, suggesting that there was some persistence of HO 2 at night in each of the three measurement periods.It is worth noting that (RO 2 + HO 2 ) was also observed to persist through the night at levels of almost 10 pptv (ca.2.5 × 10 8 molecule cm −3 ) (see Carpenter et al., 2010).Whalley et al. (2010) observed that HO 2 followed the nighttime profile of O 3 , suggesting that entrained air during the night was providing the source of radicals.On the night of 7 September 2009, O 3 remained constant at ∼23 ppbv, so it was impossible to identify a link between [HO 2 ] and ozone levels in this study.Whalley et al. (2010) also suggested that the source was the decomposition of peroxyacetyl nitrate (PAN) and used a box model to show that ∼100 pptv of PAN was sufficient to reproduce their nighttime levels of HO 2 .Nighttime levels of NO y , which would include PAN, of the order of hundreds of pptv were measured, so that process may have been an important source of nighttime HO 2 during the SOS.Rohrer and Berresheim (2006) assessed which factors influenced the variability of OH over five years by calculating the contributions to V OH , the mean of the variances in [OH] divided into a range of timebins.For example, for a 5-yr dataset divided into timebins of 24 h, V OH was calculated as the average of the daily variances of [OH].Those authors then suggested that this total variance was a combination of three individual variances, such that where the first term in this expression represents the mean variance of OH common to J (O 1 D), where R 2 (J O 1 D) is the square of the correlation coefficient for a power fit of [OH] to J (O 1 D) (i.e.Eq. 5), V inst is the mean variance due to instrument noise, and V other is the remaining variance due to other sources, including chemical influences.On subtracting V inst from V OH , Rohrer and Berresheim found that the variance of OH across five years was dominated by the diurnal link to J (O 1 D) (76 %) and the seasonal cycle (23 %).It was shown that there was no observable long-term trend, such that there was a strong degree of seasonal stability in the relationship between [OH] and J (O 1 D) across the five-year study.This behaviour suggested that competing chemical processes influencing the fit parameters in Eq. ( 5) were compensating for each other across the seasons.This simple relationship was also shown to better describe the observed [OH] than a detailed chemical mechanism.The observations of [OH] and [HO 2 ] at the CVAO during 2009 were treated in a similar fashion and the methodology of the variance analysis is provided in the Appendix.Once the total variances and the contributions due to instrument noise and J (O 1 D) were calculated for each timebin, the weighted mean value for each of these three variances across the whole dataset was calculated.The difference between the mean total variance and the sum of the mean variances due to instrument noise and dependence on J (O 1 D), is the variance due to unattributed sources, be that from factors influencing instrument noise not accounted for by V inst or chemistry controlling OH and HO 2 .

Seasonal variance analysis
The results of the analysis are shown in Fig. 11.As would be expected, the influence of J (O 1 D) becomes more apparent at timebins where its value shows more variation (for instance, above 3 h).The variance of [OH] on a 1 s timescale is dominated by instrument noise (up to 75 % at 5-min binning), unsurprising given that the low concentration of this species leads to fluorescence signals just above the offline signal due to scattered light and dark counts (a 1 s limit of detection for OH would be ∼3.5 × 10 6 molecule cm −3 ).This dependence on the instrument noise would also explain the reduced sensitivity of the total variance in 1 s-[OH] to changes in the length of timebin compared to the 1 s-HO 2 data, where the mean total variance in [HO 2 ] increases by a factor of ∼22 from a timescale of 5 min to 200 days (the total number of days spanned from the start of SOS1 to the end of SOS3).The influence of instrument noise on the observed [OH] can be reduced by averaging the 1s-data into 30 s averages, and then binning these calculated means.However, even with prior 30 s-averaging, instrument noise still contributes ∼35 % of the observed total variance in [OH] at 5-min and ∼6 % at 200 days.Nevertheless, the averaging has now made the temporal variability of [OH] across different timeframes more apparent; for instance, one can now clearly see that the variance in [OH] across a day is a factor of ∼3 more than across 5 min.Averaging across 20 min reduces the contribution of noise further, and it can be seen that the variances tend to more constant values with increasing time-averaging and that the variance of [OH] becomes more clearly controlled by J (O 1 D).
The variance due to undefined sources, V other , which includes the variability in the chemical parameters controlling OH and HO 2 other than J (O 1 D), gradually increases with timebin-length for OH, but remains relatively constant for timebins greater than 6 days for HO 2 , suggesting that timescale as the order of which the air mass is changing.This result may not be unreasonable given the changes in O 3 , CO and air mass contribution observed during the SOS (see Figs. 3-5), and is particularly evident for SOS2.The day-to-day diurnal profiles of J (O 1 D) were very similar for this period.Figure 4 clearly shows that the levels of OH and HO 2 were relatively constant from 7-10 June, when the mixing ratios of CO and O 3 were also reasonably consistent at ∼100 ppbv and ∼30 ppbv.The air mass was different for the remaining days of SOS2, with a smaller contribution from air originating from the African coast and lower levels of CO and O 3 , and the levels of OH and HO 2 are reduced compared to 7th-10th.
After subtracting the effects of instrument noise, the analysis suggests that ∼70 % of the variance of both OH and HO 2 across the SOS can be explained by diurnal behaviour, and about 30 % from changing air mass and seasonal behaviour.It must be remembered that the SOS data are from three short, discrete periods, as opposed to the continuous 5-yr dataset of Rohrer and Berresheim.The sharp rises in variances at the order of 2-3 days may be a result of using a noncontinuous dataset containing measurements that were frequently only practical between 06:00 a.m. and 09:00 p.m. local time.Although not shown here, similar variance analysis within each of the three SOS measurement periods showed similar patterns as those seen in Fig. 11.Continuous, longterm measurements at the CVAO would provide better evidence of the existence of any genuine seasonal trends in [OH] and [HO 2 ] and the relative contributions to the variance in those two species from diurnal and seasonal behaviour.

Conclusions
The Leeds aircraft-FAGE system, in its ground configuration, was successfully used to measure the concentrations of OH and HO 2 radicals at the Cape Verde Atmospheric Observatory for a total of 33 days over three periods of 2009 as part of the Seasonal Oxidant Study.This study was the first time that both OH and HO 2 have been measured in a tropical location in order to assess the seasonal variability of these species.
The concentrations of both OH and HO 2 followed the trend September ∼ June > February-March, with maximum concentrations of ∼9 × 10 6 and 4 × 10 8 molecule cm −3 , respectively, observed in the summer months, almost double the observations in winter, when increased levels of dust may act as an enhanced sink for HO x .The diurnal profiles of the June campaign agreed well with observations at the CVAO two years previously (Whalley et al., 2010).HO 2 was observed to persist into the night at levels of 10 7 molecule cm −3 , again consistent with the earlier work of The concentrations of both OH and HO 2 showed good correlations with J (O 1 D) and P (OH) were observed, particularly for HO 2 , with some differences in behaviour observed for summer and winter months.It was found that ∼60 % and 90 % of the respective variabilities in observed OH and HO 2 , respectively, could be described by a simple steady-state approximation based on the primary production of OH from the photolysis of ozone and subsequent reaction of O( 1 D) with water vapour.The coefficients yielded from a linear fit of [OH] to J (O 1 D) are similar to those yielded from two previous studies in the tropical Atlantic MBL (Whalley et al., 2010;Brauers et al., 2001), possibly suggesting that the behaviour of OH in this region may be predicted using a simple expression.A variance analysis of the data suggested that 30 % of the variance in [OH] and [HO 2 ] across the study may be attributable to changes in the air mass, although it is recommended that a more complete dataset of observations would lend strength to this conclusion.
The seasonal behaviour of OH observed in this study could have important implications for our understanding of the oxidizing capacity of the Earth's troposphere.However, the observed seasonal trend in OH observed during SOS does not fit with the simple chemistry scheme adopted here.A study of the ability of two atmospheric models to reproduce the observed behaviour in OH and HO 2 -and hence a test of the current understanding of tropospheric chemistry -is the subject of a future paper.

Methodology of variance analysis
Variance analysis was carried out both on the raw one-second dataset and using 30 s, twenty minute and one hour averages.For the averaged data, the values of [OH], [HO 2 ] and J (O 1 D), each recorded at 1 Hz, were grouped into n arrays of length equal to the averaging time in seconds and covering the complete dataset from midnight on 28 February 2009 to midnight on 16 September 2009.Thus, for example, the first such array for the hourly-averaged OH data would contain the OH data between midnight and 01:00 a.m.local time on 28 February 2009 (Julian day 58).Some arrays were excluded from analysis if they did not contain a sufficient number of one-second data -the limits were set at 15, 300 and 900 one-second values for the 30 s, twenty minute and one hour averages, respectively (corresponding to 50 % of a full array for 30 s and 25 % for both 20 min and one hour averaging, a smaller fraction required for the latter two because these averaging times were longer than a typical run time, for which the instrument would only be online for 5 out of ∼8 min).The average values of each of those arrays was calculated and placed in a two-dimensional array that also contained the corresponding average Julian day for each array (t i ) -for example, an hour is ∼0.042 of a day, so that the corresponding array for the hourly-averaged OH data would begin assuming that each of the original arrays contained sufficient one-second data to be included in the analysis.
The timebins that were chosen for the variance analysis were 5 min, 30 min, one hour, 1 1 / 2 h, 2 h, 3 h, 4 h, 6 h, 12 h, one day, 1 1 / 2 days, 2 days, 3 days, 4 days, 6 days, 30 days and 200 days.These lengths of timebins were chosen to give the greatest spread in the variability of J (O 1 D) -and hence [OH] and [HO 2 ] -within each timebin (i.e.very little across 5 and 30 min compared to several hours and days).The timebin of 30 days was chosen so that the average variance would be the average of the variances of each of SOS1-3.The timebin of 200 days was used to calculate the variances across the whole of the SOS.
The variance analysis was carried out on each timebin length sequentially.Thus, the first variance analysis was performed using a timebin of five minutes.The campaign, starting at midnight 28 February 2009 and ending at midnight 16 September 2009, was divided into lengths of time equal to the value of the timebin.For example, there were approximately 57 000 bins of five minutes.The variances of [OH] and [HO 2 ], V OH and V HO 2 , within each timebin were calculated by using the values of the appropriate array (A1) that fall within each timebin.Then, a non-linear fit of the equation was made to the data in each timebin, and the R 2 value for that fit was multiplied by the value of V OH or V HO 2 for that timebin.Finally, the variance in [OH]  where R is the correlation coefficient for a linear fit of PD(t) to Sig(t).The first (signal) term in the large parentheses represents the relative noise of the recorded fluorescence signal.The noise on the signal is defined as shot noise (i.e.σ Sig = Sig 0.5 ), and the total raw signal before correction for background (i.e. total fluorescence + background counts per second) must be used.The second (PD) term in the large parentheses represents the relative noise of the recorded laser power (typically less than 5 %).Thus, Eq. (A4) simplifies to peak ∼ 9 pptv HO 2,night ∼ 0-3 pptv Good agreement between model and daytime HO 2 for some days, factor of 2 difference other days.Kanaya et al. (1999) Kanaya et al. (2000) peak ∼4×10 6 molecule cm −3 HO 2,peak ∼ 17 pptv HO 2,night ∼2-5 pptv [HO 2 ]:[OH]∼76; model underestimated day-time HO 2 by ∼20 % and cannot reproduce nighttime HO 2 Kanaya et al. (2001a) Kanaya et al. (2001b) HO x predicted to be larger in spring than autumn; OH seasonal changes due to changes in [O 3 ], [H 2 O] and J(O 1 D)

Fig. 1 .
Fig. 1. [OH] (black) and [HO 2 ] (red) recorded at 1 Hz showing influences of pollution from passing boats on 16 June 2009 (left) and the site power generator on 5 September 2009 (right); the green line in the left-hand panel is the 10 s running average of the OH data.The HO 2 data in the right-hand panel have been offset by +2 × 10 8 molecule cm −3 for clarity.

Fig. 2 .
Fig. 2. Plots showing two examples of measured HO 2 signals (black line) without (a) and with (b) significant variation in the laser wavelength, observed as the response of the scaled reference cell signal (red line).The corrected HO 2 -signals are represented by the green lines.

Fig. 3 .
Fig. 3. Time-series of OH, HO 2 and supporting measurements for SOS1.Top panel: source-region percentage contributions to air mass from Atlantic continental (yellow area), Atlantic marine (blue area), polluted marine (red area), African coastal (green area) and Saharan dust (brown area).Middle panels: observed concentrations of OH (five-minute average) and HO 2 (four-minute average).Bottom panel: five-minute averaged J (O 1 D) (red line), and the average mixing ratios of CO (green dots), O 3 (blue dots), NO (black dots) and NO 2 (dark yellow dots); for clarity only the supporting data that are simultaneous with HO x -measurements are shown.

Fig. 4 .
Fig. 4. Time-series of OH, HO 2 and supporting measurements for SOS2.Top panel: source-region percentage contributions to air mass from Atlantic continental (yellow area), Atlantic marine (blue area), polluted marine (red area), African coastal (green area) and Saharan dust (brown area).Middle panels: observed concentrations of OH (five-minute average) and HO 2 (four-minute average).Bottom panel: five-minute averaged J (O 1 D) (red line), and the average mixing ratios of CO (green dots), O 3 (blue dots) and NO (black dots); for clarity only the supporting data that are simultaneous with HO x -measurements are shown.

Fig. 5 .
Fig. 5. Time-series of OH, HO 2 and supporting measurements for SOS3.Top panel: source-region percentage contributions to air mass from Atlantic continental (yellow area), Atlantic marine (blue area), polluted marine (red area), African coastal (green area) and Saharan dust (brown area).Middle panels: observed concentrations of OH (five-minute average) and HO 2 (four-minute average).Bottom panel: five-minute averaged J (O 1 D) (red line), and the average mixing ratio of O 3 (blue dots); for clarity only the supporting data that are simultaneous with HO x -measurements are shown.
Figures7 and 8show[OH]  and [HO 2 ] as functions of J (O 1 D) and P (OH) for each of the three measurement periods.Data from SOS1-3 are represented by black, red and green dots, respectively, and the blue lines represent the results of the non-linear fits

Fig. 6 .
Fig. 6.Half-hourly averaged median diurnal profiles of OH (top panel), HO 2 (middle panel) and primary production rate of OH, P (OH) (bottom panel); SOS1-3 are represented by the black, red and green lines, respectively.The error bars represent the 1σ dayto-day variability of the data.

Fig. 7 .
Fig. 7. Plots of five-minute averaged [OH] as a function of J (O 1 D) (upper) and P (OH) (lower).Data from SOS1-3 are represented by black, red and green squares, respectively, and the blue lines are the results of the non-linear fits to the complete datasets.

Fig. 8 .
Fig. 8. Plots of four-minute averaged [HO 2 ] as a function of J (O 1 D) (upper) and P (OH) (lower).Data from SOS1-3 are represented by black, red and green squares, respectively, and the blue lines are the results of the non-linear fits to the complete datasets.
of b = 1 and b = 0.5 for [OH] and [HO 2 ], respectively; the values in parentheses are the results of unforced fits together with the standard errors on b.The values of R 2 are virtually the same for the respective forced and unforced fits.

Fig. 9 .
Fig. 9. Plots of the hourly-averaged median diurnal profiles of OH (A), HO 2 (B) and P (OH) (C) for SOS2 (June 2009; black lines) and RHaMBLe (May-June 2007; solid red OH-measuring period, dashed red HO 2 -measuring period).The error bars represent the 1σ day-to-day variability in the hourly-averaged data.For clarity, the error bars on P (OH) for the OH-measuring period are not shown.

Fig. 10 .
Fig. 10.The time-series of the five-minute averaged [OH] and fourminute averaged [HO 2 ] (solid black lines) and one-second J (O 1 D) (red line) for the night of 7 September 2009; the dashed black lines represents the five-minute and four-minute LODs (S/N = 1) for OH and HO 2 , respectively.

Fig. 11 .
Fig. 11.The mean variances of 1 s (A), 30 s (B), 20 min (C) and one hour (D) averaged [OH] (left-hand panels) and [HO 2 ] (right-hand panels) during SOS as a function of binsize; the different coloured lines correspond to the total variance (black), the variance due to instrument noise (red), the variance due to J (O 1 D)-dependence (blue) and the contribution from other sources (green).The dashed vertical black line represents the timescale of individual measurement runs (i.e.five minutes).

Vaughan et al.: Seasonal observations of OH and HO 2
OH + CH 4 ,VOCs(+O 2 ) → RO 2 + H 2 O (R5)RO 2 can undergo self-reaction and ultimately form HO 2 .For example, in the case of the methylperoxy radical, CH 3 O 2 ,CH 3 O 2 + CH 3 O 2 → 2CH 3 O + other products (R6)Published by Copernicus Publications on behalf of the European Geosciences Union.S.

Table 1 .
Summary of measurement campaigns of HO x in tropical and remote marine regions; studies that included measurements in the tropical MBL are shown in bold.

Table 2 .
Summary of the mean and standard deviations of NO, NO 2 , NO y , CO, O 3 and VOCs at the CVAO during the HO xmeasurement periods of SOS1 and 2. Note that the NO, NO 2 and NO y data coverage was sporadic across these periods.

Table 3 .
Results of the analytical fits of Eq. (13) for J (O 1 D) to the observed five-minute averaged [OH] and four-minute averaged [HO 2 ].

Table 4 .
and [HO 2 ], respectively; the values in parentheses are the results of unforced fits together with the standard errors on b.The values of R2 are virtually the same for the respective forced and unforced fits.Results of the analytical fits of Eq. ( or [HO 2 ] due to instrument noise, V inst , within each timebin was calculated using the following technique.The values of [OH] and [HO 2 ] measured at Cape Verde are calculated from HO x represents either OH or HO 2 , not the sum of OH and HO 2 , Sig(t) is the raw HO x signal (count s −1 ) at time t, PD(t) is laser power (mW) and C is the instrument sensitivity (count s −1 mW −1 cm 3 molecule −1 ).V inst , which is equal to σ 2 HO x , can thus be defined by where