On instrumental errors and related correction strategies of ozonesondes: possible effect on calculated ozone trends for the nearby sites Uccle and De Bilt

. The ozonesonde stations at Uccle (Belgium) and De Bilt (Netherlands) are separated by only 175 km, but use different ozonesonde types, different operating procedures, and different correction strategies. As such, these stations form a unique test bed for the Ozonesonde Data Quality Assessment (O3S-DQA) activity, which aims at providing a revised, homogeneous, consistent dataset with an altitude-dependent estimated uncertainty for each revised proﬁle. For the Electrochemical Concentration Cell (ECC) ozonesondes at Uccle mean relative uncertainties in the 4 − 6 % range are obtained. To study the impact of 5 the corrections on the ozone proﬁles and trends, we compared the Uccle and De Bilt average ozone proﬁles and vertical ozone trends, calculated from the operational corrections at both stations and the O3S-DQA corrected proﬁles. In the common ECC 1997-2014 period, the O3S-DQA corrections effectively reduce the differences between the Uccle and De Bilt ozone partial pressure values with respect to the operational corrections only for the stratospheric layers below the ozone maximum. The upper stratospheric ozone measurements at both sites are substantially different, regardless the used 10 correction methodology, the origin of which is not clear. The discrepancies in the tropospheric ozone concentrations between both sites can be ascribed to the problematic background measurement and correction at De Bilt, especially in the period


Introduction
Although being a minor constituent, ozone is present throughout whole the lower atmosphere. Depending on the location in the atmosphere, the molecule is involved in different chemical reactions and therefore has a different impact on life on Earth.
For instance, ozone absorbs both infra-red and ultraviolet (UV) radiation, but the former reaction is more dominant in the tropopause region, where ozone acts as a greenhouse gas with an estimated globally-averaged radiative forcing of 0.40 ± 0.20 5 Wm −2 (IPCC, 2013). On the contrary, the higher ozone amounts in the stratosphere effectively block the harming solar UV radiation and act as a UV-filter for the living beings on earth. At the surface, ozone is an air pollutant that adversely impacts human health, natural vegetation and crop yield and quality (e.g., Cooper et al., 2014).
Since ozone at different (vertical) atmospheric layers is formed and destroyed by different photochemical reactions -and with precursor emissions from both natural and anthropogenic sources -the time variability of the ozone abundance (on seasonal, 10 inter-annual and decadal time scales) highly depends on the location (height) of ozone molecules in the atmosphere. This is illustrated in Fig. 1 and should recover in the next decades (Newman et al., 2009;WMO, 2014). Tropospheric and especially boundary layer ozone concentrations increased significantly since 1969, see Fig. 1. This increase is caused by growing emissions of e.g. nitrogen oxides (NO x ), methane, carbon monoxide and hydrocarbons in particularly the first (two) decades (e.g., Logan et al., 2012).
Thereafter, a levelling off of the ozone amounts took place due to declining anthropogenic ozone precursor emissions (e.g., 20 Cooper et al., 2014).
The observations used in Fig. 1 to construct the integrated ozone amount time series are gathered with ozonesondes, lightweight instruments attached to weather balloons and electronically coupled with a standard meteorological radiosonde for data transmission to a ground receiver. Ozonesondes provide the vertical distribution of ozone at very high vertical resolution (typically a few 100 metres), up to altitudes in the range 30-35 km. They have been launched worldwide already more than 25 half a century, and therefore constitute the most important data source to derive long-term ozone trends with sufficient vertical resolution up to about 20 km (SPARC-IOC-GAW, 1998). A major concern for any research with ozonesonde measurements is the data homogeneity and consistency, because every profile is obtained with a unique instrument, and different types of ozonesondes exist. Consequently, every ozonesonde needs to be calibrated thoroughly prior to launch. To have consistency between different ozonesonde stations, it is essential to have agreement on procedures for preparation as well as agreement on miniature piston pump, two iodide ions (I − ) are oxidised to form iodine (I 2 ), which is subsequently reduced back to I − at the electrodes, generating an electric current of two electrons. This current can directly be related to the number of moles of ozone, sampled per second and cm 3 , by the formula (Smit et al., 2011): with I M and I B respectively the measured electric cell current and background current (both in µA), η c the conversion ef-5 ficiency, F the Faraday's constant (=9.6487 × 10 4 C mole −1 ), and Φ p the pump flow rate in cm 3 s −1 . The factor 2 in the denominator points to the number of electrons produced in the sensor cell per ozone molecule. The pump flow rate Φ p and the background current I B are measured prior to launch. By applying the ideal gas law the corresponding partial pressure of ozone can be expressed as 10 with T p the measured pump temperature (K) and R the universal gas constant (=8.314 J K −1 mole −1 ).
Uncertainties may change during flight as the pump efficiency degrades with increasing altitude, or due to inaccurate pump temperature measurements or the presence of a background current that is subtracted from the measured current (Staufer et al., 2014, and references therein). Within the O3S-DQA initiative, an uncertainty analysis has been developed and the overall relative uncertainty of P O3 is expressed as a composite of the contributions of the individual uncertainties of the different listed 15 instrumental parameters above (Smit et al., 2011): As some of the contributions depend on the air pressure, the overall uncertainty of the ozone measurement is a function of pressure or altitude. The O3S-DQA initiative therefore provides this uncertainty estimate for each ozone measurement of the vertical profile. It should however be noted that this uncertainty estimation does not take into account the uncertainty due to (Tarasick et al., 2002;Stübi et al., 2008). Therefore, the results from previous comparisons of BM sondes with other types of sondes, either on dual flights or in the laboratory, or with other instrument types are not consistent (Smit and Kley, 1998;Stübi et al., 2008). The performance of the BM sondes in the troposphere is even more problematic than in the stratosphere, and the quality of tropospheric data from earlier European BM sondes has been questioned by Schnadt Poberaj et al. (2009)  be very similar, as assessed in an environmental simulation chamber (Smit et al., 2007) and on a balloon experiment (Deshler et al., 2008). It should however be noted that the amount of the sensing solution at De Bilt changed from 2.5 to 3.0 cc on the 23rd of November, 1994.
The largest difference in the operating procedures between the Uccle and De Bilt stations (see also

Data correction methods
The sonde data are processed according to Eq. 2, but design changes (e.g. the presence and location of the pump temperature sensor), differences in pre-flight operating procedures, evolving guidelines following inter-comparison campaigns led to wide variety of post-processing algorithms applied in the ozonesonde network. For instance, the background current is measured at different times during pre-flight preparation, e.g. before or after the sonde is exposed to a sampling flow with about 100 ppbv. This BGC can be assumed constant during the flight, equal to 0 for BM sondes, or alternatively, a pressure dependent BGC correction can be used 2 , assuming a small oxygen dependence with a gradual decline that is proportional with decreasing pressure and is negligible in the upper troposphere and stratosphere (Komhyr, 1986). In this latter case, the BGC is assumed to be caused by a small interference of oxygen reacting with KI in the cathode and therefore generating a small additional current (Smit et al., 2007).

10
Furthermore, it has been observed that at reduced air pressure, the pump flow rate Φ p in Eq. 2 declines due to pump leakage, dead volume in the piston of the pump, and the back pressure exerted on the pump by the cathode cell solution (Komhyr, 1967;Steinbrecht et al., 1998). This decrease in pump efficiency is corrected by multiplying the pump flow rate in Eq. 2 with a pump correction factor C PF as function of air pressure, based on laboratory measurements of the pump efficiency at reduced pressures (Smit et al., 2007). The different pump efficiency correction profiles C PF used worldwide for the BM and ECC 15 sondes are e.g. shown in Fig. 2 of Stübi et al. (2008). They all smoothly increase with decreasing pressure and predominantly affect the upper part of the ozone profile.
Another common practice is the normalisation (linear scaling) of the ozonesonde profiles to an independently determined total ozone amount (measured by e.g. a Brewer or Dobson spectrophotometer). This is in particular important for BM sondes, because they have a typical response equivalent to about 80−90% of the actual ozone amount (SPARC-IOC-GAW, 1998). 20 Therefore, the partial ozone column above the balloon burst altitude has to be estimated, either by the assumption of a constant mixing ratio or by applying satellite climatologies (e.g. McPeters and Labow, 2012).

O3S-DQA corrections
From the discussion in the previous paragraphs, it is obvious that there is a need for a standardisation of the operating procedures and a homogenisation of the ozonesonde time series (not only between different stations, but also for a given station), which is 25 the aim of the already mentioned O3S-DQA activity. This activity is however restricted to ECC sondes only, not for BM sondes.
Consequently, for Uccle, the time series of ozonesonde measurements homogenised according to the O3S-DQA principles, starts with the introduction of ECC sondes in 1997.
The rationale, recommendations and guidelines of the O3S-DQA activity are described in Smit et al. (2012) and can be consulted there. We here shortly give an overview of the proposed corrections for Uccle and De Bilt, also summarized in , where I B0 is the background current measured during pre-flight preparations at surface pressure P 0 (Smit et al., 2011). The Vaisala manual however proposes a second order correction for the SPC ECC sensor:  (Davies et al., 2003). Therefore, for these data, the conversion efficiency η c is not longer equal to one and its composite, the absorption efficiency, was processed by a pressuredependent expression for pressures above 100 hPa.
For the O3S-DQA correction, both Uccle and De Bilt stations subtracted the background current (BGC) measured prior to 10 launch from the measured electrical currents, i.e. the BGC is kept constant. As at Uccle the recommended BGC measurement after ozone exposure is only since recently available, the value recorded before ozone exposure is used. The former is higher than the latter, but never exceeds 0.1 µA at Uccle, because this is the established upper limit for accepting the ozonesonde for launch. In De Bilt, to reduce the I B , the following strategy has been adopted: after exposure to ozone, the chemicals in the cell were changed (refreshed) as many times as necessary to get the I B to a small value (< 0.2 µA from 1998 onwards, < 0.1 15 µA from 2003, see Fig. 3). The value for the BGC that is actually used for the correction, is measured through the radiosonde system, at the end of the calibration procedure. This is typically one or two hours after the rest of the procedure to condition and calibrate the ozone sensor. Normally the I B has gone down significantly in this period. Before 1998, this value was measured in the laboratory, immediately after the calibration of the radiosonde. From November 2005 onwards the I B was measured at the launch field during the inflation of the balloon. Between 1998 and 2005 the I B was measured both in the lab and on the field, 20 see Fig. 3. The value that is used for correcting the ozone profile changed in 2003 from "lab" to "field". The "field" values are typically lower than the "lab" values. Although the constant BGC subtraction with the measured value has been applied for the O3S-DQA correction, this remains questionable for the De Bilt record, as the measured BGCs are too high in the early years.
Instead, the O3S-DQA guidelines recommend to use a climatological value of 0.045 µA±0.03µA for the BGC after exposure of ozone. In this paper, we nevertheless use the measured BGC at De Bilt. 25 The ECC sondes now used in Uccle and De Bilt are equipped with a thermistor, mounted in a hole drilled in the pump body, to measure the pump temperature T p . However, the pump temperature needed in Eq. 2 is the actual temperature inside the cylindrical housing of the moving piston of the pump, which is about 1-3 K smaller than the measured T p , depending on the pressure (Smit et al., 2012). Within O3S-DQA, a correction (with an uncertainty of about ±0.5 K) is proposed, based on simulation chamber measurements. For the periods during which the thermistor was located only in the box (and not in the 30 pump) at Uccle and De Bilt (see Table 1), an additional pressure-dependent correction is applied (Eq. 9 in Smit et al., 2011), because the frictional heating of the moving piston of the pump gives an internal temperature within the pump base that is higher than the external pump temperature. Measurements in the simulation chamber pointed out that the differences between both temperatures were between 0.5 and 2 K at ground pressure, but increased to a maximum in the range 7−10 K at 50 hPa and then slightly decreasing towards lower pressures (Smit et al., 2007). In Uccle, the pump flow rate is measured in the laboratory with a Brooks volume calibrator with a mercury ring. In De Bilt, a bubble flow meter is used for this measurement. However, this latter technique is susceptible to an offset due to the evaporation of water from the detector cell, which is positioned between the pump and the bubble flow meter: this is called the "humidification effect". The proposed correction method for this effect (Smit et al., 2011) is based on the temperature and relative humidity at laboratory conditions. These have been recorded in De Bilt for the majority of the flights. In the few 5 cases when these conditions have not been recorded, they have been estimated from the meteorological conditions during the preparations of the sensor. More in general, the equilibrium pump temperature turns out to be about 2 K higher than the room temperature in which the volume calibrator is located (Komhyr et al., 1995;Smit et al., 2012). As a consequence, the actual pump flow rate at ground will be larger than the measured one by a factor of 1.007, and is corrected for accordingly for both stations. This value is then multiplied in Eq. 2 with the already mentioned pressure-dependent pump correction factor C PF , 10 obtained from the laboratory measurements described in Komhyr (1986) for SPC (De Bilt), and described in Komhyr et al. (1995) for ENSCI (Uccle). These two curves differ by about 1 % at 10 hPa and 3 % at 5 hPa.
Finally, the O3S-DQA initiative recommends not to use the total ozone normalisation for ECC ozonesondes, but still to calculate and report the scaling factor when distributing the data through international databases. It can be used as an additional quality indicator of the ozone sounding data. Furthermore, although both Uccle and De Bilt switched from RS80 to RS92 15 radiosondes and the corresponding change in the pressure sensor affects the vertical ozone profile (Steinbrecht et al., 2008;Stauffer et al., 2014;Inai et al., 2015), we follow the O3S-DQA recommendation not to apply any altitude correction to the profile. Additionally, this radiosonde change also caused a change in the Vaisala interface card, and hence the pump temperature sensor, so that an effect on the recorded pump temperatures cannot be excluded (see Fig. 2 in Van Malderen et al., 2014, which shows a 2 • C decrease at 700 hPa). Since this effect is not quantified, no correction can be applied. 20

The Uccle corrections (PRESTO)
In Uccle, after using BM sondes for about 25 years, the transition was made to ENSCI ECC sondes in 1997. Therefore, the operational post-flight algorithms at Uccle are developed primarily to construct a homogeneous time series, without any break caused by this transition. The details of these corrections can be found in De Backer (1999) and are presented in Table 2. The main aim of the correction strategy is to combine the pump efficiency correction with the total ozone normalisation, as 25 the latter is required for BM sondes. Therefore, we will use the acronym PRESTO (PRESsure and Temperature dependent total Ozone normalisation) for this correction method in the remaining of the paper. This method is operationally applied only at Uccle, but could also be adapted to other ozonesonde site datasets. In practice, the following steps are taken (see also  (1998a) found that the efficiency of the miniature pumps is not only a function of pressure, but 30 is also dependent on the temperature of the pump, especially for BM sondes, and the following temperature correction was derived k(T ) = a 0,0 + a 0,1 · T + a 1,0 · log 10 (p) + a 1,1 · T · log 10 (p) where k(T ) represents the factor by which the time to pump 100 ml (∝ 1/Φ P in Eq. 2) of air at 20 • C must be multiplied to obtain the time at temperature T (in • C), and a i,j regression coefficients. These factors are visualized for different pressures and temperatures in Fig. 2 (Komhyr and Harris, 1965;Komhyr et al., 1995, for BM and ENSCI ECC sondes respectively).
Both sets of the obtained measurements could be fitted by a similar equation for the time needed to pump 100 ml of air at pressure p: with p 0 the ground pressure and b a parameter depending on the sonde type. Inspired by this equation, De Backer et al. (1998b) 10 proposed the following empirical shape for the pressure dependency of the pump flow rate (= the pump flow correction factor C PF ): with c 0 the ground calibration factor determined with a calibrated ozone source (320 µgm −3 running through the ozone sensor during 10 minutes) before launch and b a parameter depending on the performance of the sensor, determined in such a way 15 that the integrated amount of ozone in the profile (increased with the residual amount of ozone), is equal to the total ozone measured with a spectrophotometer at the same site. In other words, the pump flow correction factor C PF , determined after the temperature dependency correction of the pump flow rate in Eq. 4, is adjusted for each pump individually as to match both the single point calibration of the ozone sensor at the laboratory and the total ozone column value measured on site. For completeness, we add that the residual amount of ozone is calculated with either the constant mixing ratio assumption or the 20 McPeters and Labow (2012) satellite climatology, depending on the balloon burst altitude, as prescribed by WMO (1987).
When the ground calibration factor c 0 is not available (i.e. before May 1992), the value c 0 in Eq. 6 is estimated from a relation between c 0 and the total ozone scaling factor, depending on the quality of the pumps. there. First, the background current, measured in the laboratory before exposure to ozone since May 1992, is subtracted from the measured cell currents over the whole altitude range for the ECC ozonesondes. For the BM sondes, no correction for the BGC is made (I B = 0). However, before October 1981, the ozone concentrations imposed to the sensor during the preconditioning phase in the laboratory were much lower than recommended (WMO, 1987), causing too low ozone concentrations in the lower tropospheric ascent profiles, as found by comparing the ascent to descent ratios of ozone profiles before and after 5 that date (De Backer, 1994). Therefore, a pressure-dependent amount of ozone partial pressure is added to the ascent profiles from ground to 70 hPa, as proposed by De Backer (1999), which can be interpreted as a correction for "a negative BGC caused by impurities in the sensor".
The introduction of Vaisala radiosondes in 1990 allowed to measure the temperature in the Styrofoam box containing the ozone sensor pump ("box temperature"). Since December 1998, the pump temperature T P is measured with a thermistor in 10 a hole in the pump. For the ECC sondes, this measured value (either box or pump temperature) is used in Eq 2. For the BM sondes launched after 1990, we use the measured box temperature as an approximation of T P , instead of a recommended fixed value of 300 K (WMO, 1987), that is known to produce an overestimation of the ozone partial pressure of about 8 % near the burst altitude. From May 1989 to December 1989, the mean box temperature as derived from the soundings during 1990 and 1991 is used. As the insulating capacity of the Styrofoam boxes used before 28 April 1989 was higher, a modified average box 15 temperature profile is used for this period, with a slower temperature decrease adjusted to reach the measured 7 • C at 10hPa (instead of 3 • C thereafter).
With the replacement of the VIZ radiosondes by Vaisala radiosondes in 1990, the accuracy of the pressure measurements increased substantially, which has an impact on the (BM) ozone profile measurements. At Uccle, between 1985 and 1989, more than 450 soundings were used to calculate the differences between the altitudes from the VIZ radiosondes and the 20 altitudes deduced from the tracking of the balloon train with a primary wind-finding radar (De Muer and De Backer, 1992).
They showed that a systematic bias of up to 1.5 km was present near the top of the soundings, caused by the slow response time of the VIZ pressure sensor. Furthermore, the differences seemed to have changed during the campaign period, probably due to an additional calibration error in the period 1985−1988. Consequently, different altitude corrections have been made for these different periods, see De Backer (1999). For the period before 1985, when no radar information is available, the 25 more conservative altitude correction of the period 1988−1989 is applied. Although smaller differences between the radar and Vaisala altitudes were observed during a small campaign in the period September−December 1989, no altitude correction is made for this type of sondes.
The electrochemical sensors of both BM and ECC ozonesondes are also sensitive to other atmospheric trace gases, such as SO 2 . The decrease in the ozone readings of BM sondes is proportional to the SO 2 concentration with a proportionality factor a fictitious (Dobson) total ozone trend has been induced (De Muer and De Backer, 1993) and the lower tropospheric ozone trends calculated from the BM sondes would be overestimated. Therefore, for Uccle, corrections for the SO 2 interference on the BM ozone soundings (and on the Dobson spectrophotometer) are applied (De Backer, 1994, 1999, making use of the in situ measurements of the SO 2 density near the ground in the urban area of Brussels (and even at Uccle itself). Since the Z-ECC sondes were not used in Uccle before 1996, when the SO 2 concentrations in Brussels had already stabilized at low values, the 5 impact of these concentrations on the ozone soundings is negligible and no correction for SO 2 needs to be applied to the ozone profiles obtained with this type of sensor.
As already mentioned, this complete set of corrections, operationally applied at Uccle, will be referred to as the PRESTO corrections.

10
The focus of the ozonesonde programme of the De Bilt station, now 22 years long, lies more on the satellite validation and the Match campaign for the determination of stratospheric polar ozone losses 3 , rather than the creation of a homogeneous long term data record. As a consequence, small changes in their procedures and data processing have occurred several times.
However, the data from the ozone sensor has been digitised on board the sonde, and all original raw data are still available.
It is not our purpose to discuss here all changes that have been made over time, but concentrate on the ones that affect the 15 homogeneity of the data series, also presented in Table 2.
The most significant changes took place in late 1998, when the participation of De Bilt in the Match campaign started, and an agreement on standardisation of operating procedures and data processing was reached among the participating ozonesonde stations. Therefore, from November 1998 onwards, the environmental conditions in the laboratory were recorded, the pump temperature instead of the box (or sensor) temperature was measured, another pump efficiency correction table was used 20 (Komhyr et al. (1995) instead of Komhyr (1986)), the background current value was reduced by adopting a new measurement strategy (see above), and the constant BGC subtraction was applied.
Most critical for the homogeneity in the De Bilt dataset is the BGC. Before late 1998, the measured BGC values were too high (see Fig. 3), so that the BGC subtraction leads to an underestimation of the total ozone column from the integrated profile with respect to the co-located Brewer instrument's value, which is even noted until the year 2000. Because a pressure-25 dependent correction subtracts a smaller BGC through the profile than the subtraction with a constant value -the subtracted BGC equals the measured one at ground pressure and then decreases with increasing pressure, see Sect.2.2 -the pressuredependent correction with the measured BGC for the period before the end of 1998 is still preferred. As the BGC values in De Bilt decreased over time (see Fig. 3), this trend will have an impact on the calculated trends of (in particular tropospheric) ozone, see Sect. 4. But also the change of the BGC subtraction method might generate an artificial trend in the ozone profile 3 Impact on the average ozone profiles The different possible post-processing steps described in the previous section all have an impact on the final ozone profile. In this section, we will quantify these impacts on the average ozone profiles, first for Uccle and De Bilt separately. Thereafter, we will compare the resulting average ozone profiles of both stations.

5
As two types of ozonesondes have been used at Uccle, we will treat them separately in this section.

The BM 1969-1996 time series
To visualize the influence of the different steps in the Uccle PRESTO corrections on the average ozone profile obtained by BM sondes, we show in Fig. 4 the relative differences to the profile obtained by applying only the correction of the pump efficiency decrease with the standard correction factors (Komhyr and Harris, 1965). A first thing to note is that applying the total ozone 10 normalisation by multiplying the profile with a scaling factor (gold dotted curve in Fig. 4)  The combination of the correction for the pump efficiency decrease with decreasing pressure and the total ozone normal-15 isation leads to a smaller relative difference in the troposphere (around 10−15 %), and higher relative differences above the ozone maximum (see black curve in Fig 4). This can be explained by the fact that the used pump correction factors, determined in the vacuum chamber at Uccle, are higher than the standard correction factors on one hand, and on the other hand by the redistribution of the total ozone amount over the entire profile. With our combined method, layers contributing hardly to the total ozone amount, like e.g. the troposphere, will be exposed to smaller ozone normalisation scaling factors, as should be 20 obvious from the figure. Additionally, also the poorer performance of the pump for decreasing temperatures is corrected for at Uccle. However, as without any box temperature correction a constant value of 300 K is assumed for BM soundings before 1990, we show in Fig. 4 the combined effect of both contributions (red curve). These pump temperature effect corrections seem to have the largest impact on tropospheric ozone, if we compare with the previous described correction (black curve). At first sight, this seems contradictory, because it is in the upper parts of the atmosphere that the pump efficiency is most affected by 25 the lower temperatures and the box temperature deviates most from the 300 K standard value. Once more, we should keep in mind that these effects have been smeared out over the entire profile by the redistribution of the total amount of ozone. Because of a detected change in quality of the Styrofoam boxes and pumps used after April 1989, alternative corrections for both the box temperature and the pump efficiency were extrapolated for the period before April 1989. As can be seen from Fig. 4 (cyan curve), the effect of these corrections is quite large, especially in the upper part of the profile, where their impact is largest (see 30 also the comparison with SAGE II data in Lemoine and De Backer, 2001 The following two additional corrections for BM ozonesondes that are investigated, especially affect the tropospheric ozone. The first one, the correction for a negative background current because of the sensor being exposed to too low ozone concentrations during the preconditioning before October 1981, enhances the tropospheric ozone by about 10 to 15 % (magenta curve in Fig. 4). The correction for SO 2 interference adds another 5−10% of ozone in the boundary layer (see green curve in Fig. 4), and even around 25 % at the surface (not shown here). When we finally add the altitude correction for radiosondes launched before 5 1990, which affects especially the ozone profile at and above the ozone maximum (see grey curve in Fig. 4), the complete set of the PRESTO correction is in use (blue curve in Fig. 4). With respect to the standard pump correction, all these corrections give a roughly 30 % ozone increase in the free troposphere and even between 30 and 40 % in the lower troposphere/boundary layer. The impact of the PRESTO correction is lowest in the lower stratosphere (around 20% ozone increase), and increases again from the ozone maximum to reach again 30% in the upper parts of the sounding. The PRESTO post-processing steps 10 have been developed based on simulation chamber tests, double soundings, the comparison of ascent and descent profiles, etc.
and have been validated against reference satellite data (SAGE II).

The ECC 1997-2014 time series
For the ECC time series, we again chose to confront the corrected profiles with the standard pump corrected (average) Uccle ECC profile in Fig. 5. The alternative correction methods at Uccle produce average profiles within ±2 % of this reference 15 profile, a number even smaller than the estimated uncertainties for the Uccle ECC profiles (see Fig. 2). These smaller relative differences compared to the average BM profiles shown in Fig. 4 are due to the nearly 100 % response equivalent of the actual ozone amount of ECC sondes. Indeed, the total ozone normalisation by simple linear scaling increases the ozone relatively by less than 1 % throughout the profile (see gold dotted curve in Fig. 5). Consequently, the relative differences for the average profiles processed by the Uccle pump efficiency correction method, with a pressure-dependent total ozone normalisation (in 20 black in Fig. 5), are within the same range. They increase with decreasing pressure, because the measured pump efficiency correction factors in the vacuum chamber in Uccle are higher than the standard correction factors, see e.g. Fig 2 in Stübi et al. (2008). Introducing the temperature dependence of the pump efficiency in the corrections (to complete the PRESTO correction, blue curve in Fig. 5) adds another 1 % relative difference in the troposphere and the upper stratosphere. A similar vertical behaviour of this temperature dependency correction as in Fig. 4 is observed. For ECC sondes, the relative differences 25 only due to this correction (hence applying only Eq. 4) increase from around 0 % in the troposphere to a maximum of 4 % at balloon burst altitudes.
The Uccle O3S-DQA corrected profile is also included (grey curve in Fig. 5), and resembles the chosen reference most (within ±1 %), as could be expected from methods using the same standard pump efficiency correction factors (and applying no total ozone normalisation). The difference is largest at high altitudes, due to the pump temperature correction applied in correction blows up the differences with the O3S-DQA correction (compare the black and blue curves in Fig. 5), also at altitudes above the ozone maximum.

De Bilt
Now we compute for the entire observation period of De Bilt (1993Bilt ( -2014 the average profiles of the two different correction strategies: one generated according to the O3S-DQA guidelines, and another one corrected by the De Bilt operational 5 algorithms (see Table 2). In Fig. 6 (green line), we compare the vertical profile of the relative differences between both those average profiles. A first important note is that the O3S-DQA average profile has smaller ozone amounts at all altitude levels.
The relative differences between both corrections are smallest at the surface and the ozone maximum (around 2 %), and largest at the tropopause (about 6 %). Above the ozone maximum, the relative differences increase to a 4 % at burst altitude. The variation of these relative differences in altitude is caused by the differences in the correction and operating procedures before the 10 end of 1998 (black curve in Fig. 6). From November 1998 on, the MATCH standard operating procedures were applied in the operational chain at De Bilt, resulting in an average profile differing by only 2 % at all altitudes with the O3S-DQA corrected profile for the same period (red curve in Fig. 6). Before 1998, the large relative differences in especially the free troposphere (even more than 15 %) can be ascribed to the different background current correction strategies applied in the O3S-DQA and operational dataset. In both cases, the same (relatively high) value for the BGC is used, but this (constant) value is subtracted at 15 all pressure levels for the O3S-DQA correction and a pressure-dependent BGC subtraction is applied for the operational correction. Because the subtracted BGC value decreases with increasing pressure in the latter case, the O3S-DQA correction results in lower ozone partial pressures at all pressure levels. The relative differences between the two average profiles are therefore largest in this period at those levels where the impact of the BGC on the measurements is highest (the free troposphere, see

Comparison of Uccle and De Bilt
As already mentioned, in the previous figure  the upper stratosphere is not clear to us, in particular because the agreement in the lower stratosphere is fairly good, around 5 % at most. In any case, measuring the ozone concentrations above 25 km is the most challenging for ozonesondes due to e.g.
the pump efficiency decrease and the evaporation of the sensing solutions, but the relative differences found at those layers are also well above the quoted relative uncertainties of 5−6 % in Fig. 2. A relative difference around 5 % is achieved for the troposphere, somewhat less for the boundary layer (2−5 %), and somewhat more for the upper troposphere (5−8 %). 4 Impact on the vertical ozone trends 15 Looking back at the ozone monthly means for three distinct (vertical) atmospheric layers in Fig. 1, the similar seasonal behaviour in the Uccle and De Bilt time series stands out. In this section, we will study the long-term time behaviour of the Uccle and De Bilt ozone series, which span different time periods. In particular, we will analyse the impact of the different correction strategies on the resulting vertical ozone trends, for different periods. To determine these trends, we first calculate the monthly anomalies of ozone partial pressures in layers of 1 km height, relative to the tropopause height. Then, for each of 20 these layers, (robust) trends are estimated from the monthly anomaly time series by simple linear regression. We did not apply a multiple linear regression model (e.g., Harris et al., 2015) to calculate trends, because the focus is here on differences between trends rather than on the trend values themselves. Compared to the average ozone profile calculation, we chose a lower vertical resolution, because the trend estimation is more sensitive to the number of available measurements per layer. Nevertheless, we stress that the results are comparable when using the identical vertical resolution as for the average profiles. For the different correction steps present in Fig. 4, the estimated relative trends are shown in the same colour coding in Fig. 8. in the stratosphere, hereby inducing negative trends in the bulk of the stratosphere, and by around 5 % dec −1 in the troposphere.
Since this total ozone normalisation corrects for the lacking total ozone response equivalent by the BM sondes at the begin of the period, while the ECC ozonesondes have a nearly full total ozone response equivalent, it is clear that the ozone concentration trends will be smaller after this correction. With the introduction of a pressure-dependent total ozone normalisation by combining it with the pressure and temperature dependency of the pump efficiency (red curve in Fig. 8), the trends are in- Analogously to the previous time period, applying the Uccle pump efficiency correction method (in black in Fig 9), which is driven by the total ozone normalisation, leads to a significant reduction of the positive trends compared to the standard pump correction: by about 5 % dec −1 in the troposphere, and by about 5 to 10 % dec −1 in the stratosphere, with an increasing trend reduction with increasing pressure in the latter case. This significant change of trends after these corrections can be ascribed to the larger impact on the BM sonde measurements, which were launched during the first 4 years of the considered time  Figure 1-22 in WMO, 2014). The EESC is a sum of chlorine and bromine derived from ODS tropospheric abundances weighted to reflect their potential influence on ozone. 5 As, by the end of 2012, the EESC has already returned 38−41 % from its peak value (WMO, 2014), a major issue in current ozone research is if the onset of ozone recovery can be detected. Our study demonstrates that, at least for measurements with ozonesondes, caution is needed before qualifying an even significant ozone increase as the onset of ozone recovery.

Conclusions
For the nearby stations of Uccle and De Bilt, we calculated average profiles and vertical trend estimates from both the oper-10 ational and internationally agreed O3S-DQA corrections. Because typical horizontal ozone correlation lengths are generally much longer than the distance between both stations, except in the boundary layer, and because the time separation between the launches at those stations is at most one day, the comparisons of the average profiles and trends enable us to investigate the impact of the correction strategies on the ozone profiles and resulting trends.
In Uccle, where the time series is built up with both BM and ECC ozonesondes, the main feature of the operational PRESTO 15 correction is the combination of a pressure and temperature dependent pump efficiency correction with the total ozone normalisation. For the BM 1969-1996 time period, the operational corrections result in a relative ozone increase between 20-30 % in the average profile with respect to the standard pump efficiency corrections, due to the typical BM response being only equivalent to about 80% of the actual ozone amount. Because of the correction for SO 2 interference, this relative ozone partial pressure difference even increases to about 40 % in the lower tropospheric layers. For ECC ozonesondes (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014), the 20 different correction strategies produce average profiles within ±2 % of this reference. In particular, the O3S-DQA correction for ECC ozonesondes adds about 2 % ozone at the tropospheric levels compared to the operational correction, whereas above the ozone maximum, the reverse is true, but now with an amount around 1 %. For the De Bilt time series , the O3S-DQA average profile has smaller ozone amounts at all altitude levels. Here, the largest relative differences are obtained in the UTLS (about 15 %) in the period before November 1998, when different background measurement operations and corrections 25 have been applied for both corrections.
When comparing the average profiles of Uccle and De Bilt, we conclude that higher tropospheric ozone concentrations are measured at Uccle than at the Bilt, which might be ascribed to both natural (more polluted area at Uccle) and instrumental (higher background currents subtracted at De Bilt) origins. In the lower stratosphere, higher ozone amounts are present in De Bilt, while the opposite is true above the ozone maximum. At burst altitudes, the relative differences between the Uccle and 30 De Bilt ozone partial pressures can amount to up to 10 %, independently of the used correction method. This reason for these larger discrepancies is not clear to us. As a matter of fact, we found that the O3S-DQA corrections for ECC ozonesondes at both sites effectively reduce the relative differences between Uccle and De Bilt only in the lower stratosphere (below the ozone  [1993][1994][1995][1996] from about 20-30 % for the standard pump corrections to less than 5 %. The used correction method has also a large impact on the derived trends. For the entire Uccle time period, the operational corrections result in a fairly constant and consistent trend over the troposphere (+2 to +3 % dec −1 ) and stratosphere (−1 to 5 −2 % dec −1 ), which is a serious reduction of the overall positive trends estimated from the profiles corrected only by the standard pump efficiency profiles. In particular, the correction for the SO 2 interference is responsible for a reduction of the