Interactive comment on “ Propagation of radiosonde pressure sensor errors to ozonesonde measurements ”

We found that the first part of this paper about pressure offsets statically calculated for various types of radiosondes shows very interesting results and conveys a important message to radiosonde user community particularly for the stratosphere. However, we suspect that the latter part about estimation affecting on ozonesonde measurements is somewhat misleading in view of the altitude coordinate when they make a comparison.

R. M. Stauffer et al.: Propagation of radiosonde pressure sensor errors to ozonesonde measurements Sun et al., 2010), humidity Yoneyama et al., 2008;Miloshevich et al., 2006;Sun et al., 2010) and pressure (De Muer and De Backer, 1992;Inai et al., 2009;Hurst et al., 2011) measurements and typically have been associated with the adoption of new sonde models. The performance of electrochemical concentration cell (ECC) ozonesonde instruments, of which there have been three manufacturers since the 1970s, has been compared with various compositions of sensing solution type in laboratory conditions (Smit and Kley, 1998;Smit et al., 2007;Smit and Berg, 2011), and field conditions (Komhyr et al., 1995a, b;Deshler et al., 2008). The discrepancies among the ozonesonde instrument-sensing-solution combinations are ∼ 5-15 % relative to an absolute O 3 measurement, depending on ECC manufacturer, and are pressure (and thus, altitude) dependent. The O 3 community has made many attempts to homogenize standard operational procedures (Deshler, 2012;WMO, 2013) for station pre-flight preparations and intercomparison of different ECC cells, so some of the performance characteristics of the ozonesonde prior to launch are well understood. At present, the global ozonesonde community is reprocessing thousands of O 3 profiles from dozens of stations to produce a more accurate profile data set for trend analysis (Ozonesonde Data Quality Assessment, O3S-DQA; Smit et al., 2012). In this effort, the pressure measured by the radiosonde to which the O 3 partial pressure is referenced, has been taken as free of biases.
The relatively recent widespread use of GPS-enabled radiosondes has shown that pressure sensors often differ from the pressure derived from the GPS data. These errors propagate to errors in the calculated O 3 mixing ratio (O 3MR ).

Efforts to quantify radiosonde errors and biases
Numerous intercomparison studies investigate biases in the pressure, temperature, humidity and GPS measurements amongst various radiosonde types. da Silveira et al. (2006) launched five types of GPS-enabled radiosondes in groups to analyze GPS measurements in addition to meteorological measurements. They found the reproducibility and comparisons of GPS altitude in the stratosphere were within ±20 m. Similar results were obtained by Nash et al. ( , 2011, who, in two different studies, found pressure sensors to be superfluous based on excellent performance of GPS radiosondes. Hurst et al. (2011) compared RS92 and iMet pressure measurements and found that paired RS92 radiosondes all compared to within ±0.3 hPa in the stratosphere and that iMet radiosondes averaged approximately 0.8 hPa lower than the RS92s between 25-30 km, an error of > 5 %. Inai et al. (2009) studied individual RS80 radiosonde launches to compare pressure derived from GPS measurements with the radiosonde pressure sensor and found pressure sensor biases of −0.5 hPa above 20 km. These pressure errors need to be considered in the context of O 3MR measurements and total column O 3 integration.
Lately, radiosonde manufacturers (e.g., Lockheed Martin Sippican, Inc., GPS Mark II Microsonde) have been producing radiosondes without pressure sensors, relying on GPS altitude, temperature, and humidity measurements and the hypsometric equation to derive pressure data. This same technique is used in this study and will be described below.

Importance of accurate O 3 measurements
The importance of long-term, accurate O 3 profile records is well documented in climate reports (IPCC, 2007), O 3 assessment reports (WMO, 2011), and numerous studies of trends in tropospheric (Logan et al., 1994;Logan et al., 1999Logan et al., , 2012IPCC, 2007), stratospheric (Miller et al., 1995;Froidevaux et al., 1996;Liu et al., 2006;Rault and Taha, 2007;Jiang et al., 2007;Kroon et al., 2011) and total column O 3 (Thompson et al., 2003, Osterman et al., 2008. Furthermore, ozonesondes provide the highest vertical resolution (∼ 100 m) O 3 measurements from the surface to over 30 km. For this reason, the satellite remote-sensing community continues to use ozonesonde profile data for validation and improvement of O 3 profile retrievals (e.g., Nalli et al., 2013). The absolute accuracy of radiosonde measured pressure profiles themselves also has ramifications in the validation of satellite-derived pressure-profile Environmental Data Records (EDRs; Nalli et al., 2013).
Biases in O 3 measurements from the use of several different types of ECC ozonesonde manufacturers, as well as different potassium iodide sensing solution strengths and sonde preparation techniques have made the homogenization of the historical ozonesonde record a necessity. The goals of the homogenization process performed through collaborative activities from WMO's O3S-DQA (Smit et al., 2012) and SI2N (this special issue) are to compile the highest accuracy O 3 profile records for more robust trend studies and satellite comparisons (Deshler, 2012). With the ongoing reprocessing of ozonesonde data, it is vital to identify every potential bias or error in the O 3 measurements. A goal of this paper is to contribute to a consensus-based recommendation on the handling of these radiosonde errors. Note, however that the data we present here and recommendations made apply only to radiosondes launched in the GPS era.
In the present investigation a series of 624 ozonesonderadiosonde instrument packages and 107 RS92-SGP radiosondes flown solo have been analyzed. In this paper, we address the following questions: 1. What are the statistical characteristics for pressure differences ("offsets") between the pressure sensor and that derived from the GPS? How do the offsets vary as a function of pressure (altitude)?
2. How do the offsets vary between radiosonde models?
In this study, we analyze the RS92-SGP, three versions of International Met Systems (iMet) radiosondes, all of which have GPS units integrated into the radiosonde hardware, and the RS80-15N flown with a separate Garmin GPS unit attached inside the ozonesonde styrofoam box.
3. In addition to pressure offsets, some of the radiosondes demonstrate highly variable pressure measurements during ascent, especially in the stratosphere. What are the statistical characteristics of this variability?
4. How do the radiosonde pressure offsets propagate to the O 3 profiles? How is integrated total O 3 to either the balloon burst altitude or a pressure cut off (e.g., ∼ 11 hPa/∼ 30 km, as recommended in Dobson, 1973 or 10 hPa, as utilized in Thompson et al., 2003, and an extrapolated add-on determined from a climatology like McPeters and Labow (2012) affected?
The soundings were taken in the 2005-2013 period in a range of locations from the northern mid-latitudes through the subtropics and tropics to southern subtropics (Table 1).

Site and instrument descriptions
A total of 731 radiosondes were analyzed for this study, with ozonesonde/radiosonde pairs accounting for 624 of those profiles. Our analysis includes data from twelve different launch sites (including two simultaneously operated, closely located sites in Houston, TX) launching five types of radiosondes, and spanning the years 2005-2013 (Table 1). The locations range from the southern subtropics (Irene, South Africa) to the northern mid-latitudes (Sapporo, Japan) with every month of the year represented. Stations include both those making regular ozonesonde launches (Irene, Houston, Beltsville) and those making intensive launches for specific campaigns (see Table 2 for campaign details), as well as other profiling missions at other sites.
Two radiosonde types from Vaisala (Vantaa, Finland; RS80-15N, RS92-SGP; herein RS80 and RS92) and three from International Met Systems (Grand Rapids, MI, USA; iMet, iMet-P, iMet-S) were launched at the various locations. Analyses are presented for each radiosonde type. The number of launches of each radiosonde type and the manufacturer-quoted pressure accuracies/uncertainties are given in Table 3. International Met Systems uses a piezoresistive silicon device to measure pressure and quotes only one pressure accuracy throughout the manufacturing of their radiosondes from 2009-2013. The analyses are still presented by each series type (based on serial numbers that are, in general, temporally partitioned) to determine any differences throughout the evolution of iMet radiosonde production. The RS92 radiosondes received a significant pressure sensor upgrade from the RS80s, moving from an aneroid capacitor, which is observed to have a low bias in the stratosphere (Steinbrecht et al., 2008), to a more accurate solid-state silicon barocap sensor. Note that the quoted Vaisala pressure accuracies are valid only if a ground check with an independent surface pressure measurement and calibration are performed prior to launch. The RS92 radiosondes used in this study underwent this check.

Ozonesonde measurements
Each of the Science Pump Corporation (SPC) and EN-SCI/Droplet Measurement Technologies (DMT) ozonesondes in this study operate using the electrochemical concentration cell (ECC; Komhyr, 1969) technique where ambient air is bubbled through a potassium iodide solution. The subsequent reactions generate two electrons per O 3 molecule, so the current measured through an attached circuit board is converted to O 3 partial pressure (p O3 ) via this equation: where the constant 0.043085 is derived from the ratio between the gas constant R and the Faraday constant, T p the measured temperature of the ozonesonde pump, ε the pump efficiency, F the volumetric flow rate through the pump, I M the measured electrical cell current, and I B the background cell current quantified in lab testing prior to launch (WMO, 2013).
Since O 3MR is calculated from p O3 and total air pressure, p air : any bias or error in the radiosonde pressure measurement introduces error in O 3MR . The p O3 measurements have typical tropospheric accuracies on the order of −7 to +17 %, improving to ±5 % in the low to mid-stratosphere with decreasing accuracy above 10 hPa, provided standardized and accepted ozonesonde conditioning and launch procedures are followed (Komhyr et al., 1995b;WMO, 2013).

Calculation of GPS pressure
The pressure altitude reported by the radiosonde is given in geopotential altitude (Z), using standard gravity (g 0 = 9.80665 m s −2 ). Conversely, the GPS altitude is reported as a geometric altitude (H ), and the latitude (g φ ≈ 9.78-9.83 m s −2 ) and altitude-dependent gravity is used to calculate pressure. The equation for gravity with latitude (φ) is estimated from the WGS-84 ellipsoid (National Imagery and Mapping Agency, 2000, p. 42): The gravity with altitude is then given by  where g φ is the surface gravity at a given location, G the gravitational constant (6.67428 × 10 −11 N m 2 kg −2 ), m E the mass of Earth (5.9736 × 10 24 kg), r E the average radius of Earth (6.371 × 10 6 m), and H the GPS altitude from the radiosonde. This process is the reverse of obtaining a geopotential altitude from the radiosonde pressure measurements, but with a geometric altitude. We note that the reported GPS altitude is actually an ellipsoidal altitude, though the difference between that and altitude AMSL (geoidal altitude; National Imagery and Mapping Agency, 2000, p. 68) is reconciled with the input of the station AMSL altitude as the initial GPS altitude prior to launch. Surface pressure from the radiosonde (often set at the launch site from a high-precision barometer) is used to initialize the GPS pressure calculation from the hypsometric equation: Here, p GPS , is the pressure calculated from g φ,H , the latitude and altitude-dependent gravity, H , the change in geometric GPS altitude from consecutive measurements, R d , the specific gas constant for dry air (287.05 J kg −1 K −1 ), and T vavg , the average virtual temperature of the consecutive measurements. Calculating pressure in iterative fashion from measurement to measurement throughout the profile reduces the error that use of a standard atmosphere or scale height would introduce. Since the uncertainty in the GPS altitude is small (Nash et al., , 2011, usually within ±20 m (Vaisala RS92 technical specifications testing) to ±30 m (iMet radiosonde 2σ error specifications), the uncertainty resulting from GPS altitude measurements in the calculated p GPS will be negligible in the stratosphere. An additional source of uncertainty in p GPS results from errors and biases in radiosonde temperature and humidity measurements (Richner and Viatte, 1995;Hurst et al., 2011). Large systematic biases in radiosonde temperature measurements can cause some errors in the calculated p GPS profile. However, the characteristic pressure errors resulting from temperature biases bear no resemblance to the errors seen in this paper -therefore we rule out this factor as the cause of the pressure offset. The p GPS calculation assumes that the atmosphere is in hydrostatic balance with pressure dependence only in the vertical and with negligible changes horizontally Figure 1. Edgewood, MD iMet-S profile from 14 July, 2011 of GPS and pressure altitude 1 differences (black), and pressure differences after recalculation of pressure data from GPS 2 measurements (grey). The red dashed line marks the zero line for reference. The median 26 3 km pressure offset for this launch is -1.17 hPa. 4 5 Fig. 1. Edgewood, MD iMet-S profile from 14 July, 2011 of GPS and pressure altitude differences (black), and pressure differences after recalculation of pressure data from GPS measurements (grey). The red dashed line marks the zero line for reference. The median 26 km pressure offset for this launch is −1.17 hPa. and in time. The radiosonde makes the same assumption when deriving a geopotential altitude from the pressure measurements through use of the hypsometric equation.
An example of the differences in radiosonde pressure (herein p) and p GPS (treated as the reference), as well as the pressure altitude and GPS altitude differences, are shown in Fig. 1. Large differences, on the order of hundreds (and sometimes thousands) of meters, between pressure altitude and GPS altitude are an indication of systematic errors in reported pressures. For the remainder of this paper, we define the pressure offset to be p-p GPS .
Variability in the pressure offset appears in the lower troposphere since a difference of just a few meters between the GPS and the pressure altitude can cause several tenths of 1 hPa difference between the calculated and measured pressures. The noise in the pressure offset stabilizes in the stratosphere and tends to remain somewhat constant until balloon burst.

Recalculation of pressure-dependent data
Using the pressure calculated from the GPS measurements, any pressure-dependent variables can be recalculated and compared to the original measurements. In addition to the reported altitude and pressure differences between the GPS and radiosonde measurements, we examine the effects on the O 3MR and total column O 3 . (Note that the pressure corrections implemented here also result in a need to recalculate potential temperature and, to a lesser extent, water vapor mixing ratio, but we do not discuss these modifications here). We choose to examine O 3MR rather than p O3 so we can describe statistics on the differences between "coincident" measurements on a single GPS altitude, that is, the original O 3MR and the GPS-pressure derived and recalculated O 3MR . The p O3 is 29 Figure 2. Sapporo, Japan RS80 profile from 19 August, 2008 showing original pressure (blue) 1 and recalculated GPS pressure (red) O3MR profiles. The inset figure is the same profile, 2 zoomed-in to highlight O3 differences in the stratosphere. The median 26 km pressure offset 3 for this launch is -1.94 hPa. not dependent on ambient pressure (except for minor pump efficiency corrections) so coincident measurements will not change in magnitude, and only an altitude shift in the profile will be evident.
The recalculation of O 3MR causes differences in both the O 3 magnitude and profile shape, particularly above 26 km and near the burst altitude (Fig. 2). Depending on the severity of the pressure offset, O 3MR errors can approach ±1-2 ppmv (parts per million by volume; ±10-20 % error) or greater in the stratosphere in the heart of the ozone layer. Differences between GPS altitude and pressure altitude can cause the apparent O 3MR maximum to shift by as much as ±2 km, having further consequences for stratospheric satellite measurements and comparison/validation studies with ozonesondes.
We note that a pressure-dependent pump correction factor (PCF) is applied to p O3 based on decreasing ozonesonde pump efficiency in the stratosphere, particularly above 25 km (Johnson et al., 2002). However, both the application of various PCFs in different processing software and the negligible (∼ 0.5 % difference in PCF between 20 and 18 hPa, near where statistics from this paper are presented) difference the PCF has between p and p GPS profiles lead us to neglect this small correction. This effort to quantify pressure sensor errors is separate from and not a substitute for the PCF problem.

Results
The IONS-06 campaign (  Channel and in Galveston Bay. The comparisons allow us to test confidence in the p GPS recalculation procedure, namely the reproducibility of stratospheric O 3MR using ozonesondes with different radiosonde types released closely in space and time. Nine such pairs occurred within 90 min of each other in IONS-06, with RHB launching RS92s and the UH site launching RS80s with a separate GPS unit attached. An example of one pair, 15 min and 77 km apart on 30 August 2006, is shown in Fig. 3. The two profiles show similar tropospheric O 3MR with or without correcting the pressure offset (the p and p GPS profiles are indistinguishable below ∼ 15 km). The GPS corrected pressure, however, results in better agreement in stratospheric O 3MR . Before correcting the pressure offset error, the mixing ratio differences between the two flights are greater than 1 ppmv near the UH balloon burst altitude (also note the altitude shift; Fig. 3). Those differences become markedly smaller to within 0.1-0.2 ppmv after correction of both profiles using p GPS . Both the shift in the altitude and correction of the O 3MR contribute to this improved agreement.

Statistical characteristics of the pressure offsets
The median pressure offset for each km altitude bin (as in Hurst et al., 2011) from 1-30 km is shown in Fig. 4. The tight grouping of RS92 launches about the zero line is distinguishable, with considerably more spread near the top of the profiles measured with the other radiosonde types. Most individual radiosondes show less variable pressure offsets in the stratosphere, with the RS92s converging to zero. The iMet-P radiosondes exhibit a peculiar S-shape pressure offset peak around 5 km that is not understood (we can find no artifact or geophysical cause).
At 26 km (an altitude 69 % of profiles reach, also chosen because p ≈ 20 hPa at 26 km), the iMet and RS80 radiosondes exhibit the most variable pressure offsets, with mean offsets of −0.93 hPa and −1.01 hPa, respectively (see Table 4 for offset averages and percentiles). In Fig. 5, we see the radiosonde-measured pressure is consistently lower than p GPS for many of the radiosonde types, and nearly a third of all launches have an offset of ±1.0 hPa at 26 km. The least variability is exhibited by the RS92s with only a −0.12 hPa average offset and just two outlier profiles beyond ±1.0 hPa at 26 km. Figure 6 shows pressure offsets at various altitudes as a function of the pressure offset at the burst altitude. The variance within the figure at different altitudes implies that the pressure offsets are not constant throughout most of the profile, and that a constant pressure correction cannot be applied to the entire profile. Only when the balloon reaches the stratosphere and around 15-20 km is a strong relationship evident. The tropospheric offsets appear much less constant than the stratospheric offsets, likely from variability in the GPS altitude and pressure sensor causing significant noise in the pressure offset below 10 km. In the troposphere an offset of several hPa only represents a few percent of the total atmospheric pressure. At 20 km, 96 % of all launches have less than ±5 % error. As a result, the true magnitude of the pressure offset and need for reprocessing cannot be determined until the balloon has reached the stratosphere (see Fig. A1 in Appendix A for altitude differences with pressure offset).

O 3MR offsets
Pressure offsets of only a few tenths of 1 hPa are the equivalent of 5-10 % errors in the total atmospheric pressure at the balloon burst altitude near 30 km. This pressure offset error results in an error in the calculated O 3MR of the same magnitude (Fig. 7). We define the O 3MR offsets as [O 3MR(p) -O 3MR(GPS) ]/O 3MR(GPS) . Figure 7 demonstrates how a nearly constant stratospheric pressure offset results in an O 3MR offset that grows in magnitude with altitude, with many profiles beyond ±10 % error in the stratosphere. At such magnitudes, this error becomes a significant component of the overall error budget associated with O 3 profile data from ozonesondes, and is beyond the intrinsic uncertainty of the O 3 measurements.  Fig. 4. Median pressure offset (p-p GPS ) for every 1 km altitude bin from 1-30 km for each radiosonde type (grey). Average offsets (black solid line) for each grouping of radiosondes are shown along with 10th and 90th percentiles (black dashes). A red dashed line marks the zero line for reference. Table 3. Radiosonde types with number of launches, quoted pressure uncertainties/accuracies from the manufacturer, and dates of available launches. We note the various iMet series have had no appreciable changes to the pressure sensors, but are split in these analyses for convenience and ease of interpretation. at 30 km is common; over a quarter of all launches that reach this altitude have O 3MR errors > ±10 %. Two distinct offset regimes are detected in the RS92s in Figs. 4 and 7, separable mainly by the launch sites Beltsville (one summer of data) and RHB near Galveston Bay (single campaign; see Figs. A2 and A3 in Appendix A for pressure and O 3MR offsets by launch site). The Beltsville pressure offsets lie slightly to the left of the zero line, and the RHB offsets straddle the zero line. Similar offset groupings are also observed in the campaign-based launches from Porterville, CA (iMet, only one iMet-S), Las Tablas, Panama (RS80) and the set of iMet-P sondes launched in the course of 10 months at Idabel and Houston. This suggests that particular "batches" of radiosondes, regardless of manufacturer/type, may have offsets that generally behave in similar manners. As a result, we caution against drawing conclusions about radiosonde types (particularly iMet-P radiosondes in this study) from offsets appearing in only one set or batch of sondes.

Column ozone measurements
Because the pressure offset affects both the apparent altitude and magnitude of O 3MR , it is also of interest to compute the influence on total column amount of O 3 . Each ozonesonde that reached 26 km was integrated to obtain a column O 3 amount in Dobson Units (1 DU = 2.69 × 10 16 molecules cm −2 ) from both the original pressure profile and the recalculated p GPS profile. As expected, considerable differences in the column integrated to the sonde burst altitude appear closely related to the pressure offset magnitude  in the stratosphere (Fig. 8a) -the radiosonde types that displayed the largest pressure and O 3MR offsets also present the largest sonde column offsets. The iMet, iMet-P and RS80 sonde-only column O 3 is consistently ∼ 10 DU higher than the O 3 column computed using p GPS (Table 4; average column O 3 difference from all sondes is +6.5 DU).
Adding a typical O 3 climatology (e.g., McPeters and Labow, 2012) above-balloon burst allows calculation of total O 3 column abundance for both the original and pressurecorrected ozonesonde profiles. In this case, offsets are reduced to within a few DU (Fig. 9a). Note that sonde and/or satellite-based climatologies have become standard, replacing a constant mixing ratio assumption (McPeters et al., 1997(McPeters et al., , 2007Thompson et al., 2003;McPeters and Labow, 2012;Morris et al., 2013). The constant mixing ratio assumption takes the O 3 mixing ratio at balloon burst and extrapolates that value to the top of the atmosphere to provide the above-burst residual column. Thus, O 3MR errors such as those observed here will lead to significant errors in the residual and total column O 3 if a constant mixing ratio method is used. The sonde-only O 3 column discrepancies brought about by the differences in the balloon burst altitudes between the original and corrected pressure profiles are reconciled with the satellite climatological add-on above-balloon burst and comparison of the total column O 3 . The amount of total column offset is reduced to a mean offset within 3.8 DU for every radiosonde type with the above-burst addition (Table 4), signifying that both the O 3MR error and altitude differences are contributing to total column discrepancies.
A common practice within the ozonesonde community is to cut off total column O 3 integration at 10 hPa (Thompson et al., 2003, rather than integrating the entire profile, and to apply a climatology such as that of McPeters and Labow (2012) to the remainder. This approach is employed and recommended for a variety of reasons including mitigation of increasing pump efficiency uncertainties with altitude in the stratosphere (Johnson et al., 2002) and the reduced accuracy of the O 3 measurements above 10 hPa (Komhyr et al., 1995b). The same technique was applied to the ozonesondes in this study to test if the sonde-only and total column O 3 offset is reduced due to elimination of increasing O 3MR errors routinely observed above 10 hPa.
The 10 hPa cut off considerably reduces the differences between the uncorrected and pressure offset corrected sondeonly columns for most radiosonde types. Exceptions are the iMet-P launches, which in our data set rarely reached 10 hPa due to use of a smaller balloon (portions of the Houston and Idabel launches), and the RS92 profiles, which had little O 3 column error to begin with. Since the iMet-P launches rarely reached 10 hPa, the entire balloon profile was integrated to the burst altitude, eliminating any effect this cut off would have had. With the exception of the iMet-P sondes, sonde-only column O 3 average differences are reduced from a maximum of 11.2 DU to within 3.5 DU (Table 4, Fig. 8b) with a 10 hPa cut off. Considering the total column O 3 with the 10 hPa cut off and subsequent McPeters and Labow (2012) climatological add-on, the agreement between the uncorrected and corrected pressure O 3 columns is further improved and most differences are essentially noise within the uncertainty of the total column integration from the ozonesonde. All radiosonde types agree to an average offset within −1.7 to +0.8 DU, with the poorest agreement from the iMet radiosonde 90th percentile of +3.5 DU (Table 4, Fig. 9b). Figure 10 shows analysis of an individual profile to understand better the improved agreement in total column O 3 after the pressure correction is implemented. It appears the standard 10 hPa cut off may provide a serendipitous solution to reconciling the differences between p and p GPS total and sonde-only column O 3 . The compensating effects of the pressure offset are viewed in terms of O 3MR , p O3 , and integrated sonde-only column with p and p GPS . Because 10 hPa is above the p O3 maximum in the stratosphere, the discrepancies on either side of the O 3 peak routinely compensate for one another when sonde integration is truncated (i.e., the column differences below the O 3 peak are negative (positive) while above the peak they are positive (negative)). Integrating to the burst altitude for those sondes that reach above 10 hPa results in poorer agreement with altitude -the further above 10 hPa the sonde reaches before burst, the greater the column error becomes. Thus it appears that the 10 hPa recommended limit for using the O 3 profile data results in a fortuitous minimization of the column errors caused by the pressure offsets and therefore our analysis argues in favor of the application of an O 3 climatology such as that used by McPeters and Labow (2012) above-balloon burst, with a cut off at 10 hPa if necessary.

Summary and recommendations
A total of 731 radiosondes were compared to quantify errors in radiosonde pressure sensor measurements relative to pressure calculated from GPS measurements and to assess the impact these pressure offsets have on O 3MR and column O 3 measurements. The pressure offset was shown to detrimentally affect O 3 measurements, particularly in the stratosphere, where errors in O 3MR frequently exceed the laboratory uncertainty of the ozonesonde measurements of ±5 % in the lower stratosphere (Komhyr et al., 1995b). The performance of Vaisala RS92 radiosondes was superior to RS80s and three series of iMet radiosondes, and was characterized by offsets of only ±0.1-0.2 hPa at balloon burst, translating to O 3MR errors generally within ±1-2 % at 26 km. The RS80 and iMet-P radiosondes had the greatest 26 km average offsets of −1.01 and −1.95 hPa, respectively, translating to average O 3MR errors of −4.75 and −8.75 %.
The differences between the radiosonde-measured and GPS-calculated pressures also introduced an altitude shift in the profile that must be considered for satellite validation studies and column O 3 integration. The ozonesondeonly column exhibited a robust relationship with 26 km pressure offsets; sonde column differences between p and p GPScorrected profiles often exceeded +10 DU, or ∼ 3 % of the total column when offsets were beyond −1.0 hPa at 26 km. These column differences were reduced with the application Atmos. Meas. Tech., 7, 65-79 of the above-balloon burst O 3 climatology of McPeters and Labow (2012). When an integration cut off of 10 hPa was applied the agreement in total column O 3 between p and p GPS profiles improved to within a few DU. The improved agreement between the uncorrected and corrected total O 3 columns using a standard profile climatology and the 10 hPa cut off argues for adopting this technique for column abundance estimates, especially with ozonesondes launched without GPS technology. Note that in the absence of GPS verification of the pressure profiles and O 3MR , this cut off technique only improves the resulting calculated column abundance and does not improve the accuracy of the O 3 profile shape or O 3MR profile magnitude at the top of the profile.
The ozonesonde community is currently in the process of homogenizing data (Deshler, 2012), seeking the highest accuracy trends and measurements, particularly at altitudes where satellite validation plays a vital role, from a global data set spanning dozens of stations and up to 40 yr of measurements. The homogenization process will take into account sources of discrepancies and biases between different ozonesonde manufacturers, potassium iodide sensing solution strengths, and pump efficiency corrections. The pressure offset introduces an additional source of error (often significant) that is independent of the ozonesonde partial pressure measurement, and an error that is not constant from one flight to the next, either with altitude or within a specific radiosonde manufacturer/type. It is anticipated that the analyses here will contribute to pressure corrections required as part of the ozonesonde data reprocessing. 38 Figure A1. Altitude differences between pressure and GPS altitude with GPS altitude based 1 on magnitude of pressure offset. Pressure offsets from -3 hPa to 3 hPa in increments of .25 2 hPa were plotted. This calculation assumes a scale height of ~7000 m ( ̅ = 240 K, g = g0 = 3 9.80665 ms -2 ), and was calculated from pGPS = 1013 to 5 hPa. 4 5 6 Fig. A1. Altitude differences between pressure and GPS altitude with GPS altitude based on magnitude of pressure offset. Pressure offsets from −3 hPa to 3 hPa in increments of 0.25 hPa were plotted. This calculation assumes a scale height of ∼ 7000 m (T v = 240 K, g = g 0 = 9.80665 m s −2 ), and was calculated from p GPS = 1013 to 5 hPa.
The results of this study suggest the following recommendations regarding the pressure offset: 1. Whenever possible, ozonesondes should always be launched with a GPS-enabled radiosonde to ensure an accurate O 3MR magnitude and profile shape.
2. Pressure-dependent variables should be recalculated using p GPS , especially when pressure offsets exceed ±1.0 hPa or ±5 % of the total atmospheric pressure at 26 km/20 hPa.
3. An above-burst climatology such as that used by McPeters and Labow (2012) should be applied using a 10 hPa cut off (if applicable), particularly with ozonesondes launched prior to the GPS era for column abundance observations.

Appendix A
The effect the pressure offset has on the difference between radiosonde-reported geopotential altitude and GPS altitude is presented in Fig. A1. Using standard gravity, g o , it is seen how a pressure offset of ±1.0 hPa (frequently observed in this study) can lead to an altitude discrepancy of ±1.0-1.5 km, having implications for column O 3 and a shifting of the O 3 profile shape.  analysis work for this problem began with Elizabeth Thompson at Valparaiso. Special thanks to hosts in Japan during Gary Morris' Fulbright: Jun Hirokawa and Fumio Hasebe (Hokkaido University, Sapporo, Japan) and Hajime Akimoto (Frontier Research Center for Global Change, Yokohama, Japan). Thanks also to Barry Lefer at University of Houston (Houston, TX) and Bob Heinemann at the Oklahoma State University Kiamichi Forestry Research Station (Idabel, OK) and to the many students who have been involved in the ozonesonde launches from the various sites over the years. Access to Beltsville data was facilitated by Cassie