Ozonesonde profiles from the West Pacific Warm Pool: measurements and validation

We present a series of ozonesonde profiles measured from Manus Island, Papua New Guinea, during February 2014, with new insights on the calibration of ozonesondes for measurements in the tropical troposphere. The experiment formed a part of a wider airborne campaign involv5 ing three aircraft based in Guam, to characterise the atmospheric composition above the tropical West Pacific in unprecedented detail. Thirty-nine ozonesondes were launched between February 2 and 25, of which 34 gave good ozone profiles. Particular attention was paid to evaluating the back10 ground current of the ozonesondes, as this can amount to half the measured signal in the tropical tropopause layer (TTL). An unexpected contamination event affected the measurements and required a departure from standard operating procedures for the ozonesondes. The most significant depar15 ture was not exposing the sondes to ozone during preparation, which meant that the background current remained stable before launch. Comparison with aircraft measurements allows validation of the measured ozone profiles and confirms that for well-characterized sondes (background current 20 ∼ 50 nA) a constant background current could be assumed throughout the profile, equal to the minimum value measured during preparation just before launch. From this set of 34 ozonesondes, the minimum reproducible ozone concentration measured in the TTL was 12-13 ppbv; no exam25 ples of ozone concentrations < 5 ppbv, as reported by other recent papers, were measured. The lowest ozone concentrations coincided with outflow from extensive deep convection to the east of Manus, consistent with uplift of ozone-poor air from the boundary layer. However, these minima were lower 30 than the ozone concentration measured through most of the boundary layer, and were matched only by measurements at the surface in Manus. Correspondence to: G. Vaughan geraint.vaughan@manchester.ac.uk

ture was not exposing the sondes to ozone during preparation, which meant that the background current remained stable before launch. Comparison with aircraft measurements allows validation of the measured ozone profiles and confirms that for well-characterized sondes (background current 20 ∼ 50 nA) a constant background current could be assumed throughout the profile, equal to the minimum value measured during preparation just before launch. From this set of 34 ozonesondes, the minimum reproducible ozone concentration measured in the TTL was 12-13 ppbv; no exam-25 ples of ozone concentrations < 5 ppbv, as reported by other recent papers, were measured. The lowest ozone concentrations coincided with outflow from extensive deep convection to the east of Manus, consistent with uplift of ozone-poor air from the boundary layer. However, these minima were lower 30 than the ozone concentration measured through most of the boundary layer, and were matched only by measurements at the surface in Manus.

35
The Tropical Tropopause Layer (TTL) is the region of the tropical atmosphere between the top of the main convective outflow and the base of the stratosphere (approximately 13-17 km altitude) (Holton et al., 1995;Highwood and Hoskins, 1998;Folkins et al., 1999;Gettelman and Forster, 2002;40 Fueglistaler et al., 2009;Ploeger et al., 2011;Pan et al., 2014). It is a transition layer between the convectivelydominated mid-troposphere beneath and the statically-stable (and convection-free) stratosphere above, with composition dependent both on convective uplift and large-scale trans-45 port. Since the TTL is the main source region for air entering the stratosphere in the Brewer-Dobson circulation, the concentrations of source gases within it determine the stratospheric burden of ozone-destroying radicals such as Cl x and Br x . Furthermore, the temperature of the cold point deter-50 mines the concentration of water vapour in the stratosphere, while clouds in the TTL, especially near the cold point, affect the radiation budget. The TTL is therefore a region of considerable importance both for global stratospheric chemistry and for climate. 55 The region of the tropics from the Maritime Continent to the International Date Line is known as the Tropical Warm Pool, where very warm sea surface temperatures (>28 • C) support widespread deep convection (Wang and Mehta, 2008). The tropopause is higher and colder here 60 than in other regions of the tropics, especially in northern hemisphere winter, making this region of particular importance for the dehydration of air as it enters the stratosphere (Fueglistaler et al., 2009). The West Pacific region is also noted for very low ozone concentrations. Satellite measure-mum in the same region, generally attributed to photochemical destruction of ozone in the very clean marine boundary layer followed by rapid vertical mixing by deep convection (Thompson, 2003). Folkins et al. (2002) noted that tropical ozone profiles typically exhibit an 'S' shape with height, with a minimum concentration in the boundary layer (where ozone is destroyed photochemically), a maximum in the mid-troposphere due to long-range transport, and a further minimum at around 11 km before increasing into the stratosphere. They argued that this 80 is consistent with the effect of deep convection lifting air from the boundary layer to the outflow region. Closer examination of this process however suggests a more complex explanation. Heyes et al. (2009) analysed a series of ozonesondes launched from Darwin, Australia as part of the ACTIVE 85 campaign (Vaughan et al., 2008) and concluded that the lowest TTL concentrations of ozone occurred above the level of convective outflow. Back-trajectories suggested that the origin of this air lay to the north-east of Darwin, to the east and north-east of New Guinea. Uplift of air in large convective 90 complexes over the warm ocean in this region was proposed as the source region for the lowest ozone concentrations measured over Darwin. This suggests that there may be preferred locations or 'hot spots' for lifting material to the TTL. A controversial question regarding ozone measurements 95 in the TTL is whether the concentrations can fall to nearzero values (< 10 ppbv) in the outflow of deep convection. Ozonesonde observations during the CEPEX cruise over the central Pacific frequently measured ozone concentrations less than 10 ppbv, and occasionally close to zero 100 in the TTL (Kley et al., 1996). The authors suggested that lifting of near-surface air (where ozone is often strongly depleted in the tropics) essentially unmodified to the outflow of the convection could explain these observations, but they also pointed out that near-zero ozone in the TTL was en-105 countered more frequently than near the surface during the cruise, and postulated that there may be a hitherto-unknown mechanism to destroy ozone in clouds. Model simulations by Lawrence et al. (1999) showed that minima in ozone concentration in the TTL over the West Pacific result from con-110 vective uplift, but could not replicate the very low ozone concentrations found by Kley et al. Such near-zero ozone values in ozonesonde profiles were also reported by Solomon et al. (2005) and Rex et al. (2014), again in the West Pacific region. Doubts about the validity of these very low ozone concen-115 trations were raised by Vömel and Diaz (2010), who examined in detail how the ozonesonde measurement is made. In particular they examined the background current-an interfering signal that must be allowed for when deriving ozone concentrations from the raw data. Vömel and Diaz (2010) 120 pointed out that the ozonesondes in Kley et al. (1996) and Solomon et al. (2005) measuring the lowest TTL ozone concentrations also had the highest background current. A reexamination of the ozonesonde profiles of Heyes et al. (2009) shows that the same issue may have arisen there with the 125 minimum value of 4 ppbv occurring in a sonde with a higher background current than the others in that series (Section 3). We discuss the issue of the background current in detail in Section 2, but there is clearly uncertainty in the literature on the best way to account for it when calculating ozone profiles 130 from the raw ozonesonde data. One of the aims of this paper is to shed light on this uncertainty.
We present a series of ozonesonde profiles measured from Manus Island, Papua New Guinea (2·07 • S, 147·4 • E) during February 2014 as part of the CAST/CONTRAST/ATTREX 135 (Coordinated Airborne Studies in the Tropics/Convective Transport of Active Species in the Tropics/Airborne Tropical Tropopause Experiment) campaign to investigate the composition of the atmosphere above the West Pacific Warm Pool. The campaign featured three aircraft based in Guam 140 (13·5 • N, 144·8 • E), to the north of the Warm Pool: the NASA Global Hawk, the NCAR Gulfstream V and the UK Natural Environment Research Council's BAe146 (figure 1). The ground campaign took place at the Atmospheric Radiation Measurement (ARM) site next to the airport on Manus, and 145 comprised an ozonesonde campaign with supporting groundlevel observations from a TECO-49C UV photometric ozone monitor, a Picarro G-2401 cavity ring-down spectrometer to measure CO 2 , CH 4 and CO, and a home-built gas chromatograph to measure halogenated compounds (Gostlow et al.,150 2010). Support with both logistics and meteorological data were provided by ARM and the Papua New Guinea Meteorological Service. The ground-based dataset was collected between 1 and 25 February 2014, with 39 ozonesonde ascents (34 of which gave good data) between 2 and 25 February. As 155 we show in this paper, overflights of the NCAR Gulfstream V provided an opportunity to validate ozonesonde measurements in the TTL during conditions of low ozone concentration.
A key result of the CONTRAST campaign, reported by 160 Pan et al. (2015), is the bimodal distribution of free tropospheric ozone concentration measured over the tropical Western Pacific. Gulfstream V in situ measurements indicate that vertical mixing and uplift of near-surface air maintains a primary mode, narrowly distributed around 20 ppbv, from the 165 surface to 15 km. A secondary mode below 10 km, broadly distributed around 60 ppbv, was identified as incursions of midlatitude air based on the low humidity and layered structure. The minimum ozone concentration measured during CONTRAST between 12 and 15 km was 13 ppbv, consistent 170 with Vömel and Diaz (2010)'s contention that ozonesonde measurements of much lower concentrations are not reliable.
In section 2, experimental details of the ozonesonde campaign are presented, including the procedure to correct for the background current. Section 3 presents the aircraft mea-175 surements used to validate the ozonesonde profiles. Section 4 presents a summary of the ozonesonde and ground-level ozone measurements, and the conclusions are in section 5. The ozonesonde technique relies on an electrochemical reaction between ozone and potassium iodide (eqn (1)), followed by half-cell reactions in the anode (eqn (2)) and cathode (eqn (3)) (Komhyr, 1969).
The anode half-cell contains a saturated potassium iodide 185 solution and the cathode an unsaturated KI solution; as the ozonesonde ascends, a teflon pump bubbles air through the cathode cell. The current produced is proportional to the flow of ozone through the cathode cell, with each ozone molecule assumed to generate two electrons (Komhyr, 1969).  ever, this is not the only reaction that produces a current within the ozonesonde: other reactants produce a residual background current (Thornton and Niazy, 1982), which increases the measured signal and which must be accounted for when calculating the ozone concentration.

195
The background current is of particular importance in the TTL where it can be a substantial fraction of the total current measured by the sonde. The best way to correct for the background current is the subject of much debate (e.g. Komhyr and Harris, 1971; Niazy, 1982, 1983;Reid 200 et al., 1996;Smit and Sträter, 2000;Smit et al., 2007), and the two main manufacturers of ozonesondes, Droplet Measurement Technologies and Science Pump Corporation, recommend two different methods: either a constant value mea-sured before launch or a value that scales linearly with ambi-205 ent pressure. The practice of using a pressure-dependent correction arises from early suggestions that the ozonesonde reacts with oxygen (Komhyr and Harris, 1971), but later studies ruled out this mechanism and suggested that the background current should be taken as constant with altitude, at 210 least in the troposphere (Thornton and Niazy, 1982;Reid et al., 1996). However, Johnson et al. (2002) found that a background reaction with the phosphate buffers of a standard electrolyte solution did lead to a time dependence.
This confusion led Vömel and Diaz (2010) to examine in 215 detail the issue of background current. In the normal preparation of an ozonesonde, the sonde is exposed to stratospheric concentrations of ozone to check that it is responding correctly. The background current is measured as the sonde is drawing in ozone-free air before and after exposure to 220 ozone. Reid et al. (1996) recommended that the first of these measurements be adopted as the background current and removed (as a constant value) from current measurements in flight. However the standard procedure for ozonesonde preparation (Smit et al., 2007) uses a value measured 10 min-225 utes after exposure to ozone. Vömel and Diaz (2010) found that the background current continues to decrease after exposure to ozone, even for periods of hours -suggesting that a value measured 10 minutes after exposure to ozone will be an overestimate by the time a sonde reaches the TTL, leading 230 to an underestimate of the ambient ozone concentration when subtracted from the measured current. This decrease in background current is strongly dependent on the strength of the phosphate buffer concentration in the cell solution. Vömel and Diaz (2010) recommended the use of a background cur-235 rent I bg = 0 · 09I + 0 · 014 µA for the 1% KI, full-buffer cathode cell solutions used in this paper, regardless of the measurements made during sonde preparation; the dependence of I bg on the current I suggesting that the assumption of two electrons per ozone molecule passing through the cathode 240 cell is not correct. Reprocessing past soundings with this formula for background current was shown to remove all the cases of near-zero ozone -not surprising as the background current of ∼0·025 µA that this gives in the TTL is well below the 0·065 µA used for example in the original analysis 245 of the CEPEX data. Independent verification of Vömel and Diaz (2010) has however not been performed to date, and we examine below the application of this recommendation to the Manus dataset and the comparison with aircraft data.
It is clear from previous work that the background current 250 is not a well-defined quantity, and that there is uncertainty on the best way to measure it and its possible variation during flight. This is acknowledged by the Global Atmospheric Watch (GAW) report on ozonesondes (Smit et al., 2013) which calls for more fundamental research on this topic. We 255 now describe in detail the ozonesonde preparation method in Manus, which departed from GAW-standard procedures in a number of ways.

Ozonesonde preparation
The ozonesondes used here were EnSci Model Z sondes supplied by Droplet Measurement Technologies, coupled to Väisälä RS92G radiosondes which provided pressure, temperature, humidity and wind profiles. All were from the same batch of sondes supplied just before the campaign. The cathode solution comprised 1% KI with 25 g/l of KBr, 5 g/l Na 2 HPO 4 ·12H 2 O and 1·25 g/l NaH 2 PO 4 ·H 2 O as buffers. Standard procedures for preparing ozonesondes follow a two-stage process aimed at reducing the background current to less than 50 nA at the time of launch and measuring the sonde's pump flow rate. In this work the background current was obtained by drawing air into the sonde through a charcoal filter in an air-conditioned cabin where RH < 50% at all times. The current was measured with a Keithley 6485 picoammeter, and the pump flow rate F (in ml min −1 ) with a Sensodyne Gilibrator unit. Repeated measurements of pump flow rate generally agree to 1-2%. The ozone partial pressure p O3 (in mPa) was derived from the measured sonde current as: where T box was measured by taping a thermistor to the inlet 260 tube as it entered the ozonesonde pump. In this equation I and I bg are both measured in µA and T box in K. A pump correction following Komhyr et al. (1995) was also applied to the data but this is negligible for the altitude range considered in this paper.

265
The ozonesonde preparation procedures normally involve, at different stages, purging the electrochemical cell and/or the pump with high concentrations of ozone, characterising the cell response to expected atmospheric concentrations of ozone and drawing ozone-free air through the cell. For this 270 a Science Pump TSC01 ozone calibration unit was available. Normally, each ozonesonde would be first prepared 3-5 days before flight, in a four-step process: (i) passing high ozone through a new cell to remove organic traces; (ii) filling the anode and cathode cells and waiting for the current to fall 275 to 0·5 µA while drawing in ozone-free air; (iii) exposing the cell to atmospheric concentrations of ozone to verify its response; (iv) again drawing ozone-free air, measuring the time response of the cell and the background current after ∼10 minutes. Then, on the day of flight, a second preparation 280 would follow basically the same steps except that high ozone was only passed through the pump rather than the cathode cell. Standard ozonesonde procedures specify a change of solution once, at the beginning of the second preparation. We found in Manus, however, that repeated changes of so-285 lution were needed to reduce the background current to an acceptable value (the number of changes varied from sonde to sonde according to its requirements). The background current was measured both at the beginning of the second preparation and as the minimum value recorded by the Väisälä 290 software after the sonde package was finally assembled (but before taking it out of the air-conditioned environment -in the humid tropical atmosphere outside the cabin the charcoal destruction filter does not work correctly).
The procedures used in Manus departed, as already men-295 tioned, from the GAW recommendations. The most important deviation (a consequence of the malfunctioning calibration unit, see below) was that the majority of sondes were not exposed to ozone during preparation. This turns out to have been advantageous, as it avoided the decay in I bg re-300 ported by Vömel and Diaz (2010). Smit et al. (2007) report that the background current measured 10 minutes after exposure to ozone in the final preparation exceeded that measured before exposure to ozone by 34 nA on average for a sample of five EnSci sondes. By contrast, for the uncontaminated 305 sondes in Manus the average difference in I bg measured at the beginning and end of the final preparation was only 6 nA ( Figure 2). Together with changes in solution to ensure that I bg fell to around 50 nA, not exposing the cell to ozone resulted in a stable I bg during preparation, lending confidence 310 to the subsequent assumption that it remained constant during flight. We examine this assumption further in the next section.
Other departures from GAW recommendations were: the use of a 1% solution rather the 0.5% which leads 315 to an oversensitivity to ozone and a bias of ∼ +5% in ozone concentration (Smit et al., 2013) measurement of T box rather than the pump temperature, leading to an underestimate of ozone by ∼ 3% since the pump temperature is higher by around 10 • C 320 (Smit et al., 2013) use of a charcoal filter to provide ozone-free air rather than an ozone-free gas supply. The effect of this is difficult to quantify, but will be most serious in a laboratory with humid air and measurable concentrations of ozone.

325
In this case the relative humidity of cabin air was around 50%, within the expected operational range of the filter. On occasion a sonde was allowed to sample laboratory air without the filter attached, but this made no difference to the measured current. This means either that the 330 laboratory was essentially ozone-free or that the filter was not working. When the sonde was taken outside and the filter removed, an increase in signal was measured, so we conclude that the filter was working correctly and that laboratory air was essentially ozone-free.

335
correction to pump flow rate measurement for humidification of air. For a laboratory at 20 • C and 50% RH this correction reduces F in equation 4 by around 1.5% (Smit et al., 2013), increasing ozone by the same amount -in other words equation 4 underestimates 340 ozone by ∼ 1.5%.
The overall effect of departures from the GAW recommendations is therefore small -much smaller than the error due to the background current uncertainty for tropical tropospheric ozone concentrations.

Contamination
A complication encountered during this experiment was the sudden appearance of a contamination source inside the TSC01 which produced a large signal from the ozonesonde. This badly affected the first two sondes, rendering their data unusable. These sondes were extensively exposed to air drawn through the TSC01 during their second preparation (the first having been completed normally before the contamination appeared). Contamination also rendered the calibration cell on the TSC01 unusable. Sondes 3 and 4 were again clean on first preparation but were briefly exposed to the TSC01 on second preparation, after an initial measurement of the background current. The remaining sondes were not exposed to the TSC01 at all during the second preparation -the sonde's response to ozone was assumed to be nor-360 mal and the background current was measured by drawing air through an external charcoal filter. Sondes 5-14 were briefly exposed to the TSC01 on first preparation and were subsequently found to have elevated background currents. Sondes 15 onwards were not exposed at all to the TSC01 and the 365 background currents from these sondes were around 50 nA before launch. Figure 2 shows how the background currents measured for each sonde varied during the campaign, compared to the minimum current measured by the cell in the TTL (taken from 370 the Väisälä raw data telemetry). During the latter part of the campaign the background current was around half the minimum measured in the TTL, but during the early part the minimum current was close to or even lower than the background -implying an impossible negative ozone. Clearly the con-375 tamination did not remain constant during a flight.  On return from Manus a series of laboratory experiments were conducted to ascertain the properties of the contamination. These are summarised in the Appendix, but the salient result is that for lightly contaminated sondes (such as 3-15) the effects of the contamination tended to disappear over a similar timescale -1 hour -to that taken by a sonde to reach the TTL. Based on this, and the evidence in figure 2 that the minimum ozonesonde current in the TTL was remarkably stable over the campaign, we assume that in flight the contamination disappeared and the background current returned to a value of 50 nA, consistent with the uncontaminated sondes. A hybrid background current correction was thus devised: where I meas bg was the measured background current before launch, p the pressure and p 0 the surface pressure.
The spread in measured background current for the uncontaminated sondes was around 10 nA (0·01 µA, figure 2, son-380 des 15 onwards), with a similar difference between the values measured at the beginning and the end of the preparation, so it is reasonable to estimate an uncertainty in I bg measured before flight of ±10 nA. If I bg were constant during flight this would correspond to an uncertainty of ±3.4 ppbv in the 385 TTL. According to Thornton and Niazy (1983), I bg should remain constant up to 100 mb, then decline logarithmically with pressure. Our laboratory investigations on an uncontaminated sonde ( fig A3) suggest a small decrease of around 5 nA in going from lab pressure to 100 mb, consistent with 390 Thornton and Niazy's result within error limits. Taking this as an uncertainty (rather than a bias) in the variation of I bg we estimate the uncertainty in TTL ozone below 100 mb to be 5 ppbv. The cold-point tropopause during the campaign at Manus was always between 90 and 110 mb, with the ozone 395 concentration increasing rapidly in this range: the minimum concentration was always found below 110 mb. Above 100 mb the use of a constant I bg will tend to lead to an underestimate of ozone, but as ozone was generally > 50 ppbv above 100 mb, and increasing rapidly with height, this effect is only 400 manifest in the stratosphere.
The error in TTL ozone for the contaminated sondes cannot be assessed quantitatively but will certainly be greater than that for the uncontaminated sondes. We can only get an estimate of this error from a comparison with another tech-405 nique, so we now turn to a comparison of ozonesonde profiles with aircraft measurements.

Validation
One of the aims of the CAST and CONTRAST campaigns was to investigate the accuracy of ozonesonde measurements  (Ridley et al., 1992). A small flow of pure NO is added to a flow of ambient air and the resulting photons are counted using a dry-ice cooled photomultiplier tube. The instrument is periodically calibrated against a Thermo Scientific 49i-PS 420 primary ozone standard on non-flight days. The overall uncertainty is 5%, or 1 ppbv at 20 ppbv. The precision of the measurements at 20 ppbv is 0·1 ppbv (0·5%), or better, for the 10 second averages used here.
On 5 February Gulfstream V flight RF09 flew to the west 425 of Manus Island and profiled from the surface to ∼11 km. Figure 3 shows  Kley et al. (1996) and Rex et al. (2014). We conclude 440 that the hybrid correction provides a satisfactory estimate of I bg but reiterate the point made in the previous section that a quantitative error estimate in TTL ozone for the contaminated sondes is not possible. the west of Manus on two occasions-on an outbound journey towards Australia and then on the return journey back to Guam. On both occasions, an ozonesonde was launched so as to reach aircraft altitude as the aircraft made closest approach to Manus. Ozonesonde 34, coincident with the 450 outbound leg, was launched at 0131 UTC (11:31 local) and reached the Gulfstream V cruising altitude of 13·5 km at 0211 UTC. In the flight-path map in figure 3, the red line is the outbound leg of RF14. Figure 5 shows the ozone profiles from ozonesonde 34 and the co-located measurements 455 of RF14. Likewise ozonesonde 35, launched at 0449 UTC (14:49 local) coincided with the return leg of RF14, reaching the Gulfstream V cruising altitude of 180 hPa (13·5 km) at 0529 UTC. On this leg the aircraft executed a profile between 13·1 and 14·7 km as it passed by Manus. Figure 3 shows the 460 flight-path of the return leg in green. Figure 6 shows the profiles from ozonesonde 35 and the co-located measurements from RF14. Ozonesondes 34 and 35 were uncontaminated, so constant background currents of 61 nA and 54 nA respectively were 465 used in the data analysis. In both cases, the agreement between the ozonesonde and the aircraft data is within 3 ppbvconsistent with the uncertainty in the background currents. By contrast, the pressure-dependent correction and that rec- the ozone concentration. We therefore conclude that for a well-conditioned ozonesonde not exposed to ozone at all in the pre-flight preparation, where the background current at the end of the preparation is around 50 nA or less, subtraction of this constant background produces an ozone measure-475 ment in the TTL within a few ppbv of the correct value. We also conclude that our method of applying a hybrid correction produces sensible results for the contaminated sondes.
What we cannot be sure of is whether the hybrid method applies only to this particular batch of sondes, or whether 480 it can be applied more generally to sondes where the background current in the preparation is substantially larger than 50 nA. To check this, we reanalysed an ozonesonde profile from the ACTIVE campaign in Darwin, launched on 22 January 2006. This had a background current of 85 nA which, 485 when subtracted from the measured currents, resulted in an ozone concentration minimum of 4 ppbv in the TTL. A sonde the following day with a very similar ozone profile but a background current of 55 nA measured a minimum ozone mixing ratio of 12 ppbv. Applying the hybrid correction to 490 the sonde on 22 January increased the minimum value in the TTL to 12 ppbv, in line with the other sonde. This suggests that the hybrid method may have wider validity than the Manus dataset and may be worth investigating further. (We should emphasise that not all ozonesonde measurements 495 < 10 ppbv in the TTL are artifacts of elevated background currents: the lowest measured in Darwin was 8 ppbv on 15 February 2006 with a background current of 37 nA).
We have therefore applied the following background current correction to the Manus dataset, after discarding the first 500 two profiles: for sondes 3 and 4, a hybrid correction was applied using I bg measured at the beginning of the second preparation, before exposure to the TSC01 ozoniser. This value was considerably smaller that that measured after expo-505 sure to the ozoniser, but higher than the ∼50 nA typical of the uncontaminated sondes. for sondes 5 to 14, a hybrid correction was applied using I bg measured just before launch. for sondes 15 on, a constant value of I bg was applied 510 equal to that measured just before launch.
Note that for sondes 15 onwards I bg measured at the beginning and end of the second preparation were very similar ( Figure 2).

515
We present here an overview of the measurements made at Manus during CAST. The campaign experienced two distinct weather regimes-a dry period from around 1-10 February with little precipitation (figure 7) when deep convection was well to the south of Manus, and a wetter period from 11 generally around 370 K from 1-12 Feb and rather lower, 530 around 364 K, from 13 Feb onwards. Tropopause heights and pressures for the whole campaign (not shown) ranged from 15·7 to 17·2 km, and 89 to 115 hPa respectively. Double tropopauses were found from 9 to 16 February; the tropopause shown in figure 9 corresponds to the first steep in-535 crease in ozone concentration as the balloon ascended. (The cold point during this period was around -86 • C). Following the period of double tropopauses, on 18 February, the tropopause was at 370 K (17 km), but as the very wet conditions became more established it descended to reach 354 K 540 (15·7 km) on the 22nd. At the same time a distinctive feature became established in the wind field (figure 10): from 16 February onwards, and especially from 20-23 February, an easterly jet with wind speed up to 40 m s −1 was found in the TTL, just below the tropopause. This jet was confined to the 545 troposphere -by 1·5 km above the tropopause the wind had backed round to westerly, and remained westerly between 18 and 26 km. A corresponding minimum in wind speed (of ≤2 m s −1 in most cases) was measured 700-1200 m above the tropopause from 16 February onwards. This easterly jet 550 is consistent with convective outflow from the large convective complexes to the east of Manus (figure 8) reaching up to the tropopause during this period but not extending into the stratosphere.
The corresponding contour plot of ozone concentration is 555 shown in figure 11. This clearly shows the 'S' shape expected of tropical ozone soundings, with low values near the surface and in the TTL, and a maximum in the mid-troposphere. Minimum values of < 20 ppbv are frequently shown in the TTL, around 14 km during the first meteorological period 560 and then up to 16·5 km during the second period. The periods of precipitation in Manus (figure 7) both correspond to ozone concentrations < 20 ppbv reaching the tropopause, and indeed in the very wet period between 20 and 22 February, when the TTL easterly jet was at its most intense, ozone 565 minimum concentrations fell to <15 ppbv. The lowest measured value was 8·2 ppbv on 21 February -a similar mini- mum to that measured in Darwin during ACTIVE. This may have been an outlier (its background current was 60 nA), but five sondes reached between 12 and 13 ppbv (e.g. sonde 570 34 on 22 Feb, figure 5) and a further four between 13 and 15 ppbv. These values are entirely consistent with the minimum ozone concentration of 13 ppbv measured by the Gulfstream V during CONTRAST (Pan et al., 2015).
To confirm that the very low ozone measured in the TTL the HYSPLIT trajectories cannot represent ascent in individual cloud systems, and so cannot determine whether the air is really of boundary-layer origin, but they do confirm that the meteorological conditions at this time were consistent with widespread deep uplifting of air.  I I II II II I II II I I II I I I I II I II I II I II I   310  15-20 ppbv, more than the minima measured in the TTL. However, the ground-level measurements from the TECO-49 ozone monitor (figure 14) tell a rather different story. The dry early period of the CAST campaign, from 1 to 12 February, was characterized by a strong diurnal variation in ozone, 600 with maxima of ∼8-10 ppbv during the day and minima ∼2- If the lower tropospheric ozone in the uplift region to the east of Manus was similar to that over the island, this would sug-615 gest that the air reaching the very top of the TTL in the wet period originated very near to the surface and was lifted to the tropopause without significant mixing with surrounding air, consistent with the suggestion of Kley et al. (1996).

620
One of the aims of this paper was to determine the best way to correct ozonesonde profiles from a tropical station for the effect of the background current. We were very fortunate that the Gulfstream V flight RF14 was able to fly by Manus during the period when very low ozone concentrations were 625 observed in the TTL by the sondes. Ozonesondes 34 and 35 were free of contamination, and when using a constant background current measured just before launch their measurements agreed with the Gulfstream V to within 3 ppbv (the realistic limit on the accuracy of the ozonesonde at 100 mb 630 is ± 5 ppbv due to background current uncertainty). We conclude that for a well-prepared sonde -i.e. (for the batch used here) one where I bg ∼ 50 nA -a constant background current correction is the best choice.
In preparing these sondes we found it necessary to change 635 solutions in the cells up to three times during a day-of-flight preparation in order to ensure a sufficiently low background current. Other than for sondes 3 and 4, we also did not expose the sondes to ozone during the day-of-flight preparation, which removes the problem of the slow decay in I bg after 640 such exposure (Vömel and Diaz, 2010). Both these changes in standard procedures are recommendations from this work. For the sondes exposed to contamination during first preparation a hybrid background current correction was adopted after the laboratory investigation. Using this, the profile for 645 sonde 6 was found to agree remarkably well with the aircraft profile from RF09 (figure 4), lending confidence to this somewhat arbitrary correction. Care must be taken not to generalise this result too far, but we can conclude (both from the CAST sondes from Manus and the ACTIVE sondes from 650 Darwin) that a background current in excess of 70 nA is too high for a constant I bg correction -as shown by Vömel and Diaz (2010) this leads to a substantial underestimate of the TTL ozone and even to negative ozone in some cases (e.g. Rex et al., 2014).

655
The minimum reproducible ozone concentration measured in the TTL during CAST was 12 ppbv, consistent with the minimum of 13 ppbv measured between 12 and 15 km by the Gulfstream V during CONTRAST (Pan et al., 2015). This is also consistent with the minimum measured in Darwin 660 with well-prepared sondes (12 and 11 ppbv on 23 January and 14 February 2006 respectively) in air whose origin, according to back-trajectory calculations, lay in deep convective uplift east and north-east of New Guinea. In both campaigns an isolated example of a lower concentration, around 665 8-9 ppbv, was also measured. The CAST measurements confirm Vömel and Diaz (2010)'s conclusions that ozonesonde measurements < 5 ppbv in the TTL are artifacts of the background current correction.
The lowest ozone concentrations measured in the TTL 670 above Manus occurred around 16 km during a period when widespread deep convection was occurring near and to the east of the island. This is consistent with the 'hot spot' idea proposed by Heyes et al. (2009)  convection was able to lift air from the lower boundary layer into the upper TTL without significant mixing -a hypothesis we cannot pursue further here but which will be the subject of future investigations. rel Chapman, for extensive logistical support in enabling operations in Manus, and to Mae S. Chu for organising the shipping. Hymson Waffi and his staff, and the ARM technicians Mike Ryczek and Matt Gould, provided invaluable support on-site. We also thank Herman Smit, Neil Harris and Dale Hurst for valuable discussions, and an anonymous reviewer for pointing out that the contaminant could be H2O2. The project was supported by NERC grant NE/J006173/1, and the National Centre for Atmospheric Science Atmospheric Measurement Facility (AMF) provided the radiosonde 695 and ground-level ozone equipment. Richard Newton is a NERCsupported research student.

Appendix A Laboratory experiments
When the pattern found in figure 2 was discovered, the records of the CAST field campaign were examined (section 700 A1) and a series of laboratory experiments devised to ascertain the reasons why the background current generally decreased between sondes 5 and 14 yet the minimum measured current in the TTL remained reasonably constant. It was observed that when an ozonesonde drew air from the TSC01 705 ozonizer unit, a high current was registered. This was identified in the laboratory experiments as being contamination, rather than high concentrations of ozone, as explained in section A2.
Neither the source nor the identity of the contamination 710 was known, and so an experiment was devised to determine the response of an ozonesonde to pressure with various degrees of contamination, by placing it into a bell jar and varying the pressure. The results of this experiment are described in section A3. The contamination gradually disappeared over 715 time, so the bell jar experiments were neither reproducible nor did they replicate exactly the conditions that were experienced on Manus, but they serve as a check on the validity of the hybrid background current correction.

A1 Examination of records 720
The first five ozonesondes were normal on first preparation. Between the fifth ozonesonde being prepared for the first time and the following day when the first ozonesonde was being prepared for flight, the ozonizer was found to be causing the cell current to increase dramatically even 725 when it was supplying 'no-ozone' air. This affected the first two ozonesondes' day-of-flight preparations, and their background current remained well above that of a normal working ozonesonde. Thenceforth, an external ozone destruction filter was used instead of the ozonizer to produce no-ozone 730 air and the sonde was not exposed to the ozoniser during the day-of-flight preparation. However, ozonesondes 6 to 14 were briefly exposed to the ozonizer during their first preparation to check the response of the sonde to ozone. Since exposure to the ozoniser was resulting in elevated background 735 currents, the ozonesonde sample tube was only connected to it for a few seconds before being removed. However, this turned out to be long enough to allow the contaminant to get into the ozonesonde where it remained throughout the preparations.

740
Ozonesondes 15 onwards were not exposed to the ozoniser at all, and were therefore the most reliable ozonesondes launched during CAST.

A2 Source of contamination
In order to investigate the cause of the high background cur-745 rents in the first fourteen ozonesondes, laboratory investigations were conducted after the equipment was returned from Manus to Manchester, some two months after the campaign ended.
First, the response of an ozonesonde was compared 750 with that of the TECO-49 ultraviolet photometric ozone monitor. When sampling laboratory air, both TECO-49 and the ozonesonde measured comparable concentrations (∼22 ppbv), and when drawing air through the external char-measured 12 ppbv. However, when sampling supposedly ozone-free air from the ozoniser (air drawn through an internal charcoal filter) the sonde measured 189 ppbv while the TECO-49 again measured 12 ppbv. Clearly, therefore, the ozoniser was acting as a source of some contaminant 760 which produced a positive signal in the ozonesonde but not in the photometric ozone monitor -i.e. this substance was not ozone. (The 12 ppbv signal measured by the TECO through the filters is understandable as the flow rate of the TECO-49 is much higher than the ozonesonde and exceeds the capacity of the filters). Further investigation, dismantling the ozoniser and examining different parts, identified the source of the contamination as the tube which is illuminated by a mercury lamp to generate ozone. However, contamination was found even on the PTFE manifold at the outlet of the 770 ozoniser. A plausible explanation for the contamination, pointed out by one of the reviewers, is that condensation of water occurred inside the tube at some point, which, when irradiated by ultraviolet light, led to the production of hydrogen perox-775 ide. H 2 O 2 is known to react with KI in the cathode cell with a very slow response time (Cohen et al., 1967), consistent with the behaviour of the contaminant, and to stick to surfaces for a long time. The contaminant appeared first thing in the morning when the equipment had been enclosed in the 780 air-conditioned laboratory overnight.

A3 Ozonesonde behaviour at different pressures
The effect of lowering the ambient pressure on the contamination was then investigated by placing the ozonesonde in a bell jar and lowering the pressure as the sonde continu-785 ally sampled the air inside the bell jar. The bell jar was too small to admit the ozone destruction filter but ozone measurements inside the jar at ambient pressure were the same as in the laboratory with the filter attached; thus air in the bell jar was ozone-free. Three ozonesondes were exposed 790 to different amounts of contaminant by drawing air through the TSC01 unit for different times: the first was heavily contaminated, the second slightly contaminated, and the third not contaminated at all. The ozonesonde was placed in the bell jar and left to settle to a constant background current for 795 about five minutes. The bell jar was then pumped down to a target pressure using a rotary pump, and then the rotary pump was switched off. The ozonesonde was left for five minutes to settle and reach a constant background current, and then a new target pressure was chosen.

800
The first, heavily contaminated ozonesonde emulated the first two ozonesondes launched in CAST, which were prepared just after the contamination episode, but before the contamination was recognised. The slightly contaminated ozonesonde emulated ozonesondes 3 to 14, which were only were not contaminated, like the third test ozonesonde in this experiment.  The heavily contaminated ozonesonde was contaminated on both first preparation and the day-of-flight preparation 810 and had a background current of 132 nA, which is comparable to the early ozonesondes in CAST. Figure A1 shows the result of the bell jar experiment. The current was erratic, which was observed with the contaminated ozonesondes during CAST: the current occasionally spiked by ∼20 nA, possibly due to the cell picking up more contamination. The most likely behaviour of the ozonesonde was a decay of the background current from 135 nA at surface pressure to 115 nA at 20 hPa, still well above the expected value for a well-functioning sonde. This confirms that a reliable back-820 ground current estimate could not be made for the first two CAST sondes. The ozonesonde used in this experiment was subjected to a further preparation cycle (without exposure to contaminant) to investigate whether it could be cleaned. Its background current reached 40 nA after 15 minutes of no-825 ozone-air treatment, indicating that the contamination was changing its character over time: changing solutions in the Manus sondes did not remove the contamination.
The second ozonesonde was initially contaminated in first preparation, and then prepared cleanly in the day-of-flight 830 preparation, similar to ozonesondes 3-14 in Manus. However, as with the first test sonde, the contamination was found to disappear so that the 'day-of-flight' background current was 55 nA-consistent with a clean ozonesonde. It appears than that the contaminant changed its nature and became less 835 adhesive over the three-month period since the contamination event. More contaminant was therefore added at the end of the second preparation, bringing the background current to 80 nA. The bell jar experiment showed little consistency in the background current as a function of pressure, but a clear 840 decay over time ( figure A2). Since in a normal ozonesonde launch pressure decreases as a function of time, this gives weight to the idea that a decaying background current correction with pressure is appropriate for the slightly contaminated ozonesondes.

845
The uncontaminated ozonesonde was prepared cleanly both times, and had a background current of 45 nA. Figure A3 shows the result of the bell jar experiment. The experiment was split into two sections, one in which the ozonesonde remained above 200 hPa at all times, followed by 850 another in which the pressure was pumped down to 70 hPa. The current decreased slightly between 1000 and 100 hPa (45-40 nA), before decreasing to 27 nA at 70 hPa. This is similar to the result found by Thornton and Niazy (1983), which was attributed to a change in the mass transfer inside 855 the ozonesonde. Within experimental accuracy of ±10 nA, therefore, a constant background current is appropriate to the uncontaminated ozonesondes up to 100 hPa, with a possible decrease above this level. As the tropopause pressure encountered in Manus was >90 hPa, with the ozone concen-860 tration increasing rapidly into the stratosphere, we have used a constant background current throughout the profile for uncontaminated sondes. (Note that the ozonesonde in this test exhibited hysteresis when exposed to pressures lower than 100 hPa). Only the datapoints that simulate ascent are shown.

A4 Conclusions from laboratory experimentation
The laboratory experiments could not reproduce the exact conditions experienced in Manus because the contamination was gradually disappearing and becoming less adhesive over time. This is consistent with the general decrease of back-870 ground current between sondes 5 and 14 in Manus, despite their identical preparation procedure. Nevertheless, the behaviour is sufficiently similar to the CAST sondes as to provide support for the method used in equation 5 to calculate the background current.

875
The bell-jar experiments show that the background current in this batch of ozonesondes was largely constant in the absence of contamination, while that in a slightly contaminated ozonesonde reduced with time to a 'clean' value over a period of ∼30 minutes. This decay in I bg is consistent 880 with the slow timescale for the reaction of KI with peroxide identified by Cohen et al. (1967). The heavily contaminated ozonesonde did not reduce to an acceptable background current, confirming that data from the heavily contaminated ozonesondes launched in CAST should be discarded.