Validation of 10-year SAO OMI Ozone Profile (PROFOZ) Product Using Ozonesonde Observations

68 We validate the Ozone Monitoring Instrument (OMI) ozone-profile (PROFOZ) product from 69 October 2004 through December 2014 retrieved by the Smithsonian Astrophysical Observatory 70 (SAO) algorithm against ozonesonde observations. We also evaluate the effects of OMI Row 71 anomaly (RA) on the retrieval by dividing the data set into before and after the occurrence of 72 serious OMI RA, i.e., pre-RA (2004-2008) and post-RA (2009-2014). The retrieval shows good 73 agreement with ozonesondes in the tropics and mid-latitudes and for pressure < ~50 hPa in the 74 high latitudes. It demonstrates clear improvement over the a priori down to the lower troposphere 75 in the tropics and down to an average of ~550 (300) hPa at middle (high latitudes). In the tropics 76 and mid-latitudes, the profile mean biases (MBs) are less than 6%, and the standard deviations 77 (SDs) range from 5-10% for pressure < ~50 hPa to less than 18% (27%) in the tropics (mid- 78 latitudes) for pressure > ~50 hPa after applying OMI averaging kernels to ozonesonde data. The 79 MBs of the stratospheric ozone column (SOC) are within 2% with SDs of < 5% and the MBs of 80 the tropospheric ozone column (TOC) are within 6% with SDs of 15%. In the high latitudes, the 81 profile MBs are within 10% with SDs of 5-15% for pressure < ~50 hPa, but increase to 30% with 82 SDs as great as 40% for pressure > ~50 hPa. The SOC MBs increase up to 3% with SDs as great 83 as 6% and the TOC SDs increase up to 30%. The comparison generally degrades at larger solar- 84 zenith angles (SZA) due to weaker signals and additional sources of error, leading to worse 85 performance at high latitudes and during the mid-latitude winter. Agreement also degrades with 86 increasing cloudiness for pressure > ~100 hPa and varies with cross-track position, especially with 87 large MBs and (60° S-30° S), and southern high latitudes (90° S-60° S) to the latitudinal variation of the retrieval performance. We investigated the seasonal variations of the comparisons mainly at northern mid-latitudes where ozone retrieval shows distinct seasonality and there are adequate coincidence pairs. To investigate the RA impacts on OMI retrievals, we contrasted the comparison before (2004-2008, i.e., pre-RA) and after (2009-2014, i.e., post-RA). Although we OMI data based on cloud fraction, cross-track position, and SZA, we conduct the comparison winter and MBs of -3% and best correlation in summer. The TOC comparison also shows noticeable station-to-station variability in similar latitude ranges with much larger MBs and/or SDs at the two Indian stations and larger MBs at several Japanese stations before they switched from KC 699 ozonesondes to ECC ozonesondes. This demonstrates the impacts of ozonesonde uncertainties 700 due to sonde types, manufacturers, sensor solution and operations. Without applying OMI AKs, 701 the TOC correlation with ozonesondes typically becomes worse at higher latitudes, ranging from 702 0.83 in the tropics to 0.5-0.6 at high latitudes. The linear regression slope is within 0.6-0.8, 703 typically smaller at higher latitudes, reflecting the smaller retrieval sensitivity down to the 704 troposphere at higher latitudes mainly resulting from larger SZA. The convolution of AKs Studies of Emissions and Atmospheric Composition, Clouds and Climate


Introduction 97
The Dutch-Finnish built Ozone Monitoring Instrument (OMI) on board the NASA Aura satellite 98 has been making useful measurements of trace gases including ozone and aerosols since October 99 2004. There are various retrieval algorithms to retrieve ozone profile and/or total ozone from OMI 100 data (Bak et al., 2015). The ozone profile retrieval algorithm initially developed at the Smithsonian 101 Astrophysical Observatory (Liu et al., 2005) for Global Ozone Monitoring Experiment (GOME) 102 data was adapted to OMI data (Liu et al., 2010b). Total ozone column (OC), Stratospheric Ozone 103 Column (SOC) and Tropospheric Ozone Column (TOC) can be directly derived from the retrieved 104 ozone profile with retrieval errors in in the range of a few Dobson Units (DU) (Liu et al., 2006b;105 Liu et al., 2010a). This algorithm has been put into production in the OMI Science Investigator-106 led Processing System (SIPS), processing the entire OMI data record with approximately one-107 month delay. The ozone profile product titled PROFOZ is publicly available at the Aura Validation 108 Data Center (AVDC) (http://avdc.gsfc.nasa.gov/index.php?site=2045907950). This long-term 109 ozone profile product, with high spatial resolution and daily global coverage, constitutes a useful 110 dataset to study the spatial and temporal distribution of ozone. 111 To effectively use the retrieval dataset, it is necessary to evaluate and understand its retrieval 112 quality and long-term performance. Although validation of the ozone profile product (mostly 113 earlier versions) has been partially performed against aircraft, ozonesonde, and Microwave Limb 114 Sounder (MLS) data, these evaluations are limited to certain time periods and/or spatial region 115 and/or to only portion of the product (e.g., total ozone columns (OC) or TOC only) (Pittman et al., 116 2009;Liu et al., 2010a;Liu et al., 2010b;Sellitto et al., 2011;Wang et al., 2011;Bak et al., 2013a;117 Lal et al., 2013;Ziemke et al., 2014;Hayashida et al., 2015). Additionally, the quality of ozone 118 profile retrievals is very sensitive to the signal to noise ratio (SNR) of the radiance measurements 119 as well as their radiometric calibration, which may degrade over time as shown in GOME and 120 GOME-2 retrievals (Liu et al., 2007;Cai et al., 2012). Although OMI's optical degradation is 121 remarkably small to within 1-2% over the years, the SNR and the number of good spectral pixels 122 (not flagged as bad/hot pixels) have been gradually decreasing over the years due to the expected 123 CCD degradation (Claas, 2014). Furthermore, the occurrence of RA, which affects level 1b data 124 at all wavelengths for particular viewing directions or cross-track positions and likely due to 125 blocking objects in the optical path, started in June 2007 affecting a few positions. This effect 126 abruptly worsened in January 2009 affecting ~1/3 of the cross-track positions (Kroon et al., 2011). 127 The impacts of RA not only evolve with time but also vary over the duration of an orbit. Analysis 128 indicates that radiances in the UV1 channels (shorter than ~310 nm) used in our retrievals might 129 have been affected at all positions (Personal communication with S. Marchenko) and are not 130 adequately flagged for RA. Therefore, we need to evaluate the impacts of instrument degradation 131 and especially row anomaly on the temporal performance of our ozone profile product. Currently, 132 we are planning an update of the ozone profile algorithm to maintain the long-term consistency of 133 the product. The update will include empirical correction of systematic errors caused by the 134 instrument degradation and row anomaly as a function of time. Such correction also requires us to 135 evaluate the long-term retrieval quality of our product. 136 To understand retrieval quality and the resulting spatial and temporal performance of our OMI 137 product, we evaluate our data from October 2004 through December 2014 against available 138 ozonesonde and MLS observations, respectively, in two papers. This paper evaluates our ozone 139 product including both ozone profiles and stratospheric and tropospheric ozone columns using 140 ozonesonde observations with a focus on retrieval quality in the troposphere. More than 27,000 141 ozonesonde profiles from both regular ozonesonde stations and field campaigns are used in this 142 study to provide a comprehensive and global assessment of the long-term quality of our OMI ozone 143 product. This paper is followed by the validation against collocated MLS data with a focus on the 144 retrieval quality in the stratosphere (Huang et al., 2016), also submitted to this special issue). 145 This paper is organized as follows: Section 2 describes OMI retrievals and ozonesonde data. The 146 validation methodology is introduced in Section 3. Section 4 presents results, analysis and 147 discussions regarding the OMI and ozonesonde comparisons. Section 5 summarizes and concludes 148 this study. 149

OMI and OMI Ozone Profile Retrievals 151
OMI is a Dutch-Finnish built nadir-viewing pushbroom UV/visible instrument aboard the NASA 152 Earth Observing System (EOS) Aura satellite that was launched into a sun-synchronous orbit in 153 July 2004. It measures backscattered radiances in three channels covering the 270-500 nm 154 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 (Levelt et al., 2006). Measurements across the track are binned to 60 156 positions for UV2 and visible channels, 30 positions for the UV1 channels due to the weaker 157 signals. This results in daily global coverage with a nadir spatial resolution of 13 km × 24 km 158 (along × across track) for UV2 and visible channels, and 13 km × 48 km for the UV1 channel. 159 The SAO OMI ozone profile algorithm was adapted from the GOME ozone profile algorithm (Liu 160 et al., 2005) to OMI and was initially described in detail in Liu et al. (2010b). Profiles of partial 161 ozone columns are retrieved at 24 layers, ~2.5 km for each layer, from the surface to ~60 km using 162 OMI radiance spectra in the spectral region 270-330 nm with the optimal estimation technique. In 163 addition to the OC, SOC and TOC can be directly derived from the retrieved ozone profile with 164 the use of tropopause from the daily National Center for Environmental Protection (NCEP) 165 reanalysis data. The retrievals are constrained with month-and latitude-dependent climatological 166 a priori profiles derived from 15-year ozonesonde and SAGE/MLS data (McPeters et al., 2007) 167 with considerations of OMI random-noise errors. OMI radiances are pre-calibrated based on two 168 days of average radiance differences in the tropics between OMI observations and simulations 169 with zonal mean MLS data for pressure less than 215 hPa and climatological ozone profile for 170 pressure greater than 215 hPa. This "soft calibration" varies with wavelength and cross-track 171 positions but does not depend on space and time. 172 The updated algorithm of our SAO OMI ozone product was briefly described in Kim et al. (2013). 173 The radiative transfer calculations have been improved through the convolution of simulated 174 radiance spectra at high resolutions rather than effective cross sections, which is done by 175 interpolation from calculation at selected wavelengths assisted by weighting function. In addition, 176 four spatial pixels along the track are coadded to speed up production processes at a nadir spatial 177 resolution of 52 km × 48 km. Meanwhile, minimum measurement errors of 0.4% and 0.2% are 178 imposed in the spectral ranges 270-300 nm and 300-330 nm, respectively, to stabilize the 179 retrievals. The use of floor errors typically reduces the Degree of Freedom for Signals (DFS) and 180 increases retrieval errors. Compared to the initial retrievals, the average total, stratospheric, and 181 tropospheric DFS decrease by 0.49, 0.27, and 0.22, respectively, and the mean retrieval errors in 182 OC, SOC, and TOC increase by 0.6, 0.5, and 1.2 DU, respectively. The corresponding changes to 183 the retrievals are generally within retrieval uncertainties except for a systematic increase in 184 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 15 February 2017 c Author(s) 2017. CC-BY 3.0 License. tropospheric ozone at SZA larger than ~75, where the TOC increases to ~12 DU. Validation 185 against ozonesonde data indicates that this TOC increase at large SZA makes the retrieval worse. 186 Therefore retrieved tropospheric ozone at such large SZA should not be used, but the retrieved 187 total ozone still shows good quality (Bak et al., 2015). 188 For current products, retrievals contain ~5.5-7.4 DFS, with 4.6-7.3 in the stratosphere and 0-1.2 in 189 the troposphere. Vertical resolution varies generally from 7-11 km in the stratosphere to 10-14 190 km in the troposphere, when there is adequate retrieval sensitivity to the tropospheric ozone. 191 Retrieval random-noise errors (i.e., precisions) typically range from 0.6-2.5 % in the middle 192 stratosphere to approximately 12% in the lower stratosphere and troposphere. The solution errors, 193 dominated by smoothing errors, vary generally from 1-7% in the middle stratosphere to 7-38% in 194 the troposphere. The solution errors in the integrated OC, SOC, and TOC are typically in the few 195 DU range. Errors caused by the forward model and forward model parameter assumptions are 196 generally much smaller than the smoothing error (Liu et al., 2005). The main sources of these 197 errors include systematic errors in temperature and cloud-top pressure. Systematic measurement 198 errors are the most difficult to estimate, mostly due to lack of full understanding of the OMI 199 instrument calibration. 200 Certain cross track positions in OMI data have been affected by RA since June 2007 (Kroon et al., 201 2011). Loose thermal insulating material in front of the instrument's entrance slit is believed to 202 block and scatter light, causing measurement error. The anomaly affects radiance measurements 203 at all wavelengths for specific cross-track viewing directions that are imaged to CCD rows. 204 Initially, the anomaly only affected a few rows. But since January 2009, the anomaly has spread 205 to other rows and shifted with time. The RA also shows slight differences among different spectral 206 channels, and varies during the duration of an orbit. Pixels affected by the RA are flagged in the 207 level 1b data. The science team suggested that they are not be used in research. For data before 208 2009, the RA flagging is not applied in the processing. Pixels seriously affected by RA will 209 typically show enhanced fitting residuals. The algorithm was updated to use RA flagging in the 210 UV1 channel and was used to process the data starting from 2009. If a pixel is flagged as a row 211 anomaly then it is subsequently not retrieved to speed up the processing except that the cross-track 212 position 24 is still retrieved due to reasonably good fitting. It should be noted that the retrieval 213 quality of those non-flagged pixels may still be affected by the RA, because of the different RA 214 flagging in the UV1 and UV2, the lack of RA flagging before 2009and inadequacy of the RA 215 flagging. 216 To screen out OMI profiles for validation, we only use OMI ozone profiles meeting the following 217 criteria based on three filtering parameters: 1) nearly clear-sky scenes with effective cloud fraction 218 less than 0.3; 2) cross track positions between 4 and 27 due to relatively worse quality and much 219 larger footprint size for these greater off-nadir positions; 3) SZA should be less than 75° due to 220 very limited retrieval sensitivity to tropospheric ozone and the aforementioned positive biases. We 221 will use all OMI pixels of each filtering parameter when evaluating retrieval quality as a function 222 of that specific parameter. The fitting quality of each retrieval is shown in the fitting RMS (root 223 mean square of the fitting residuals relative to the assumed measurement errors). The mean fitting 224 RMS including both UV1 and UV2 channels has been increasing with time as shown in Figure 1. 225 This is primarily due to the increase of fitting residuals in UV1 caused by the instrument 226 degradation and RA since the fitting residuals of UV2 only slightly increase with time. As 227 aforementioned, the retrieval information of stratospheric and tropospheric ozone mainly comes 228 from UV1 and UV2, respectively. Consequently, retrievals in the troposphere, the focus of this 229 paper, are less impacted by the increasing fitting RMS. However, to apply consistent filtering in 230 validation against both ozonesonde in this study and MLS data in the companion paper (Huang et 231 al., 2016, submitted to the same special issue), we set the RMS threshold based on the overall 232 fitting RMS and select retrievals with fitting RMS smaller than the sum of monthly mean RMS 233 and its 2σ (i.e., Standard Deviations (SDs) of fitting RMS). 234

Ozonesondes 235
The balloon-borne ozonesonde is a well-established technique to observe the ozone profile from 236 the surface to ~35 km with vertical resolution of ~100-150 m and approximately 3-5% precision 237 and 5-10% accuracy (Komhyr, 1986;Komhyr et al., 1995;Johnson, 2002;Smit et al., 2007;238 Deshler et al., 2008). Ozonesonde data have been widely used in the studies of stratospheric ozone, 239 climate change, tropospheric ozone and air quality, as well as the validation of satellite 240 observations (Kivi et al., 2007;Wang et al., 2011;Huang et al., 2015;Thompson et al., 2015). 241 However, the accuracy of ozonesonde observations depends on data processing technique, sensor 242 solution, and instrument type and other factors. Consequently, station-to-station biases may occur 243 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 Manuscript under review for journal Atmos. Meas. Tech. in ozonesonde measurements and could be as great as 10% (Thompson et al., 2007c;Worden et 244 al., 2007). Ozonesondes (SHADOZ) (Thompson et al., 2007a;Thompson et al., 2007b), as well as archives 255 of recent field campaigns including DISCOVER-AQ (Deriving Information on Surface Conditions 256 from Column and Vertically Resolved Observations Relevant to Air Quality, http://discover-257 aq.larc.nasa.gov/) (Thompson et al., 2015) and SEACR 4 S (Studies of Emissions and Atmospheric 258 Composition, Clouds and Climate Coupling by Regional Surveys, 259 https://espo.nasa.gov/home/seac4rs) (Toon et al., 2016). Almost all of the ozonesonde data in this 260 study were obtained from electrochemical concentration cell (ECC) ozonesondes, which is based 261 on the oxidation reaction of ozone with potassium iodide (KI) in solution. The exceptions are 262 Hohenpeissenberg station in Germany that uses Brewer-Mast (BM) ozonesondes, the New Delhi, 263 Poona, and Trivandrum stations that use Indian ozonesondes, and four Japanese stations (i.e., 264 Sapporo, Tsukuba, Naha and Syowa) that switched from KC ozonesondes to ECC ozonesondes 265 during late 2008 and early 2010. These types of ozonesondes have been reported to have larger 266 uncertainties than ECC ozonesondes (WMO, 1998;Liu et al., 2013;Hassler et al., 2014). 267 To avoid using anomalous profiles, we screen out ozonesondes that burst at pressure exceeding 268 200 hPa, ozone profiles with gaps greater than 3 km, more than 80 DU TOC or less than 100 DU 269 SOC. In the SOC comparison, we also filter measurements that do not reach 12 hPa. Some 270 ozonesonde data used in this paper (e.g. WOUDC data) are provided with a correction factor (CF) 271 derived by normalizing the integrated ozone column (appended with ozone climatology above 272 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 Manuscript under review for journal Atmos. Meas. Tech. In this paper, we determine our coincident criteria based on the balance between finding most 283 coincident OMI/ozonesonde pairs to minimize differences due to spatiotemporal samplings and 284 finding a sufficient number of pairs for statistical analysis. For each screened ozonesonde profile, 285 we first select all filtered OMI data within ±1° latitude, ±3° longitude and ± 6 hours and then find 286 the nearest OMI retrieval within 100 km from the ozonesonde station to perform the validation on 287 the individual profile basis. 288 Ozonesondes have much finer vertical resolution than OMI retrievals. To account for the different 289 resolutions, ozonesonde profiles are first integrated into the corresponding OMI vertical grids and 290 then degraded to the OMI vertical resolution by using the OMI retrieval Averaging Kernels (AKs) 291 and a priori ozone profile based on the following equation: 292 293 where x is the ozonesonde profile integrated into the OMI grid, ̂ is the retrieved ozone profile if 294 the ozonesonde is observed by OMI, , A is the OMI AK matrix, and is the OMI a priori ozone 295 profile. We refer to this retrieval as "convolved ozonesonde profile", which is a reconstruction of 296 ozonesonde profile with OMI retrieval vertical resolution and sensitivity. Missing ozone profiles 297 above ozonesonde burst altitude are filled with OMI retrievals. The convolution process essentially 298 removes OMI smoothing errors and the impacts of a priori from the comparison so that 299 OMI/ozonesonde differences are mainly due to OMI/ozonesonde measurement precision, 300 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 15 February 2017 c Author(s) 2017. CC-BY 3.0 License. spatiotemporal sampling differences and other errors. However, in the regions and altitudes where 301 OMI has low retrieval sensitivity, the comparisons can show good agreement because both the 302 retrieval and convolved ozonesonde approach the a priori profile. To overcome the limitation of 303 such a comparison, we also compare with unconvolved ozonesonde profiles since it indicates how 304 well the retrievals can represent the actual ozonesonde observations (i.e., smoothing errors are 305 included as part of retrieval errors). In addition, we also compare OMI a priori and 306 convolved/unconvolved ozonesonde profiles to indicate the retrieval improvement over the a 307 priori. 308 For consistent calculations of TOC and SOC from the OMI/ozonesonde data, the tropopause 309 pressure included in the OMI retrieval and ozonesonde burst pressure (required to be less than 12 310 hPa or ~30 km) are used as the proper boundaries. The TOC is integrated from the surface to the 311 tropopause and the SOC is integrated from the tropopause pressure to the ozonesonde burst 312 pressure. 313 The relative profile difference is calculated as (OMI-Sonde) / OMI a priori ×100% in the present 314 comparison with ozonesonde and with MLS in the companion paper. Choosing OMI a priori rather 315 than MLS/ozonesonde is to avoid unrealistic statistics skewed by extremely small values in the 316 reference data especially in the MLS retrievals of upper troposphere and lower stratosphere ozone 317 (Liu et al., 2010a). Unlike the profile comparison, ozonesonde/OMI SOC/TOC values are used in 318 the denominator in the computation of relative difference. To exclude remaining extreme outliers 319 in the comparison statistics, values that are exceeding 3σ from the mean differences are filtered. 320 After applying the OMI/ozone filtering and coincident criteria, approximately 10,500 ozonesonde 321 profiles are used in the validation. We performed the comparison for five latitude bands: northern 322 high latitudes (60° N-90° N), northern mid-latitudes (30° N-60° N), tropics (30° S-30° N), southern 323 mid-latitudes (60° S-30° S), and southern high latitudes (90° S-60° S) to understand the latitudinal 324 variation of the retrieval performance. We investigated the seasonal variations of the comparisons 325 mainly at northern mid-latitudes where ozone retrieval shows distinct seasonality and there are 326 adequate coincidence pairs. To investigate the RA impacts on OMI retrievals, we contrasted the 327 comparison before (2004 and after (2009-2014, i.e., post-RA). Although we 328 filter OMI data based on cloud fraction, cross-track position, and SZA, we conduct the comparison 329 as a function of these parameters using coincidences at all latitude bands to show how these 330 parameters affect the retrieval quality. In these evaluations, the filtering of OMI data based on 331 cloud fraction, cross-track position, and SZA are switched off, respectively. Approximately 15,000 332 additional ozonesonde profiles are used in this extended evaluation. To evaluate the long-term 333 performance of our ozone profile retrievals, we analyze the monthly mean biases (MBs) of the 334 OMI/ozonesonde differences as a function of time using coincidences in the 60° S-60° N region 335 and then derive a linear trends over the entire period as well as the pre-RA and post-RA periods. and SDs vary spatially with altitude and latitude. Vertically, the SD typically maximizes in the 342 upper troposphere and lower troposphere (UTLS) in all latitude bands due to significant ozone 343 variability and a priori uncertainty. Bak et al. (2013b) showed that the use of Tropopause-Based 344 (TB) ozone profile climatology with NCEP Global Forecast System (GFS) daily tropopause 345 pressure can significantly improve the a priori, and eventually reduce the retrieval uncertainty. 346 Consequently, the SDs of OMI/sonde differences in the UTLS at mid-and high-latitudes can be 347 reduced through reducing the retrieval uncertainties. Latitudinally, the agreement is better in the 348 tropics and becomes worse at higher latitudes. The patterns are generally similar in the northern 349 and southern hemispheres. The MBs between OMI and ozonesonde are within ~6% with AKs and 350 10% without AKs in the tropics and the middle latitudes. Large changes in the biases between with 351 and without AKs occur in the tropical troposphere where the bias differences reach 10%. The MBs 352 increase to 20-30% at high latitudes consistently with large oscillation from ~-20-30% at ~300 hPa 353 to +20% near the surface both with and without the application of AKs. At pressure < 50 hPa, the 354 SDs for comparisons with OMI AKs are typically 5-10% at all latitudes except for the 90° S-60° 355 S region. For pressure > 50 hPa, the SDs are within 18% and 27% in the tropics and middle-356 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 15 February 2017 c Author(s) 2017. CC-BY 3.0 License. latitudes, respectively, but increase to 40% at higher latitudes. The SDs for comparison without 357 applying OMI AKs, i.e., including OMI smoothing errors in the OMI/ozonesonde differences, 358 typically increase up to 5% for pressure < 50 hPa, but increase up to 15-20% for pressure > ~50hPa. 359 The smoothing errors derived from root square differences of the MBs with and without OMI AKs 360 are generally consistent with the retrieval estimate from the optimal estimation. 361 The improvements of OMI over the climatological (a priori) profiles can be reflected in the 362 reduction of MBs and SDs in the comparisons between ozonesondes and OMI retrievals, and 363 between ozonesondes and a priori. The retrieval improvements in the MBs are clearly shown in 364 the tropics and at ~ 100 hPa pressure in the middle latitudes. At high latitudes, the MBs and 365 corresponding oscillations in the troposphere are much larger than these in the a priori comparison, 366 suggesting that these large biases are mainly caused by other systematic measurements errors at 367 high latitudes (larger SZAs and thus weaker signals). As can be seen from the reduction of SDs, 368 OMI retrievals show clear improvements over the a priori at pressure < 300 hPa. For pressure > 369 300 hPa, the retrieval improvements vary with latitudes. There are consistent retrieval 370 improvements throughout the surface -300hPa layer in the tropics and only the 550 -300 hPa 371 layer at middle latitude, while there is no retrieval improvement over the a priori for > 300 hPa at 372 high latitudes. The failure to improve the retrieval over a priori in part of the troposphere at middle 373 and high latitudes is caused by several factors. They are the inherent reduction in retrieval 374 sensitivity to lower altitudes at larger SZAs as a result of reduced photon penetration into the 375 atmosphere, unrealized retrieval sensitivity arising from retrieval interferences with other 376 parameters (e.g., surface albedo) as discussed in Liu et al. (2010b) and the use of floor-noise of 377 0.2% that underestimate the actual OMI measurement SNR. In addition, the a priori ozone error 378 in the climatology is quite small since the SDs of the differences between the a priori and 379 ozonesonde without AKs are typically less than 20% in the lower troposphere for middle and high 380 latitudes, which also makes it more difficult to improve over the a priori comparison. 381 The right column of Figure 3 shows the comparisons between OMI retrievals and ozonesondes 382 convolved with OMI AKs in the pre-RA and post-RA periods, respectively. In the tropics and mid-383 latitudes, the pre-RA comparison is better than the post-RA comparison, with SDs smaller by up 384 to ~8% at most altitudes especially in the troposphere. The pre-RA comparison also shows smaller 385 biases near ~300 hPa at middle latitudes while the post-RA comparison exhibits negative biases 386 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 Manuscript under review for journal Atmos. Meas. Tech. As seen from the number of OMI/ozonesonde coincidences shown in Figure 3, the northern mid-396 latitudes and the tropics have sufficient coincidences to validate the retrievals as a function of 397 season. In the tropics, the retrieval comparison does exhibit little seasonality as expected (not 398 shown). Figure 4 shows the comparison similar to Figure 3(c) for each individual season at 399 northern middle latitudes. The comparison results are clearly season-dependent with best 400 agreement in the summer (except for the MBs) and the worst agreement in the winter. This 401 indicates the general best retrieval sensitivity to lower tropospheric ozone during the summer as a 402 result of small SZAs and stronger signals and worst retrieval sensitivity during the winter as a 403 result of large SZAs and weaker signals. The MBs for with and without AKs at 300 hPa vary from 404 ~12% in the winter to -10% in the summer. The overall MBs are the smallest during the spring, 405 within 6%; but the MBs at pressure < 50 hPa are the best during the summer. The maximum SDs 406 vary from 31% in the winter to 20% in the summer. Also, the retrieval in the summer shows the 407 most improvements over the a priori in the lower troposphere at all tropospheric layers except for 408 the bottom layer, while the retrievals during other seasons show the improvement over a priori 409 only above the lowermost two/three layers. The seasonal variation of retrieval quality is partially 410 caused by the seasonal variations of the retrieval sensitivity and ozone variability. Bak et al. 411 (2013b) showed that the use of TB ozone climatology with daily NCEP GFS tropopause pressure 412 can significantly reduce the seasonal dependence of the comparison with ozonesondes. In addition, 413 radiometric calibration errors such as those caused by stray light and RA also contribute to the 414 seasonal variation of retrieval quality. 415 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 15 February 2017 c Author(s) 2017. CC-BY 3.0 License.

Solar Zenith Angle Dependence 416
The SZA of low earth orbit (LEO) satellite observation varies latitudinally and seasonally; 417 therefore the SZA dependence of the retrieval can cause latitudinal and seasonal dependent 418 retrieval biases. SZA is one of the main drivers that affect retrieval sensitivity especially to 419 tropospheric ozone. At large SZA, the measured backscattered signal becomes weak due to weak 420 incoming signal and long path length; the retrieval sensitivity to the tropospheric ozone decreases 421 due to reduced photon penetration to the troposphere. In addition, measurements are subject to 422 relatively larger radiometric errors such as those from stray light and as a result of weaker signal, 423 and radiative transfer calculations can lose accuracy at larger SZA (Caudill et al., 1997). 424 Figure 5 gives the MBs and SDs of differences between OMI and ozonesondes (with OMI AKs) 425 in a function of SZAs. We can see that retrieval performance generally becomes worse at large 426 SZA. The SD typically increases with SZA especially at pressure > 300 hPa. At SZA larger than 427 75°, the SD at ~300 hPa increases to greater than ~45%. The variation of MBs with SZA is more 428 complicated. We see generally larger positive biases at larger SZA in the troposphere with > 20% 429 biases at SZA larger than 75°. The MBs near ~ 30 hPa becomes more negative at larger SZAs. 430 There is a strip of positive biases of ~10% that slightly decreases in pressure from ~50 hPa at low 431 SZA to ~10 hPa at large SZA; it might be due to some systematic radiometric biases that can affect 432 ozone at different altitudes varying with SZA. Because of the clear degradation of the retrieval 433 quality at large SZA, we set the SZA filtering threshold of 75° to filter OMI data. 434

Cloud Fraction Dependence 435
The presence of cloud affects retrieval sensitivity since clouds typically reduce sensitivity to ozone 436 below clouds and increase sensitivity to ozone above clouds. The accuracy of ozone retrievals is 437 sensitive to the uncertainties of cloud information and cloud treatment (Liu et al., 2010a;Antón 438 and Loyola, 2011;Bak et al., 2015). Our OMI ozone algorithm assumes clouds as Lambertian 439 surfaces with optical centroid cloud pressure, and partial clouds are modeled using independent 440 pixel approximation such that the overall radiance is the sum of clear and cloudy radiances 441 weighted by the effective cloud fraction. The cloud albedo is assumed to be 80% and is allowed 442 to vary (>80%) with the effective cloud fraction.  Figure 6 gives the influences of effective cloud fraction on the comparisons between OMI and 444 ozonesonde observations convolved with OMI AKs. The MBs and SDs do not change much with 445 cloud fraction for pressure < 100 hPa, and typically increase with the increase of cloud fraction for 446 pressure > 100 hPa. The MBs at pressure > 100 hPa, especially greater~300 hPa, increase to more 447 than 10% with cloud fraction greater than ~0.3. This indicates that the cloud fractions have small 448 impacts on the stratospheric retrievals but large impacts on the tropospheric retrievals as expected. 449 Some of the variation with cloud fraction such as negative biases near ~300 hPa at cloud fraction 450 of ~0.4 and the decreases of positive biases at ~ 50 hPa for cloud fraction greater than ~0.8 may 451 be partially related to the uncertainties of the cloud parameters. The chosen filtering threshold of 452 0.3 in cloud fraction is a tradeoff between validating OMI data with adequate retrieval sensitivity 453 to tropospheric ozone and finding adequate number of OMI/ozonesonde coincidences. 454

Cross-Track Position Dependence 455
The OMI swath is divided into 30 cross-track pixels at the UV1 spatial resolution of our product. 456 Each cross-track position is measured by a different part of the CCD detector, i.e., essentially a 457 different instrument. Radiometric calibration coefficients of the instrument are characterized 458 during pre-launch only at selected CCD column pixels and then interpolated to other columns, 459 causing variation in the radiometric calibration performance across the CCD detector. This in turn 460 causes cross-track dependent biases in the calibrated radiance (Liu et al., 2010b), which therefore 461 causes stripping in almost all the OMI data products if no de-striping procedure is applied. Our 462 retrieval algorithm has included a first-order empirical correction independent of space and time 463 to remove the cross-track variability (Liu et al., 2010b). However, residual dependence on cross-464 track position remains and the radiometric calibration at different position can degrade differently 465 with time (e.g., the RA impact). In addition, the viewing zenith angle ranges from ~0° to ~70° and 466 the footprint area increases by approximately an order of magnitude from nadir to the first/last 467 position. So the varying viewing zenith angle causes the variation of retrieval sensitivities and 468 atmospheric variabilities within varying footprint areas may also cause additional cross-track 469 dependence in the retrieval performance.

Comparison of Partial Ozone Columns 483
We investigate and validate OMI partial ozone columns, including SOCs, TOCs, and surface-550 484 hPa and surface-750 hPa ozone columns in this section. We define the lowermost one and two 485 layer as surface-750 hPa and surface-550 hPa in this paper, respectively, for conveniences. 486 Similarly, we also analyze the validation results of SOCs and TOCs during pre-RA and post-RA, 487 respectively, to test the impacts of RA on OMI partial ozone columns. In addition, we validate 488 ozone columns from the surface to ~550 hPa (bottom two layers) and ~ 750 hPa (bottom one layer) 489 against ozonesonde observations in the tropics and mid-latitude summer where there is better 490 retrieval sensitivity to these quantities. 491

Comparison of Stratospheric Ozone Columns (SOCs) 492
The left column of Figure 8  range. The SDs are typically larger than the comparisons with MLS data (Liu et al., 2010a) due to 500 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 Manuscript under review for journal Atmos. Meas. Tech.

Comparison of Partial Ozone Columns in the Troposphere 510
The left column of Figure 9 shows the comparison of OMI and ozonesonde (with and without OMI 511 AKs) TOCs for each of the five latitude bands during 2004-2014. Without applying OMI AKs, the 512 MBs are within 1-3%except for 9% at northern high latitudes; The SDs are within 20% in the 513 tropics and mid-latitudes and increase to ~30-40% at high-latitudes. The correlation coefficient 514 ranges from 0.83 in the tropics to ~0.7 at middle latitudes, and 0.5-0.6 at high-latitudes. The linear 515 regression slopes are in the range 0.6-0.8 typically smaller at high latitudes due to reduced retrieval 516 sensitivity to the lower troposphere. After applying the OMI AKs to ozonesonde data to remove 517 smoothing errors, we see significant improvement in the comparison statistics except for MBs, 518 which are within 6% at all latitudes. The SDs are reduced to within 15%in the tropics and middle 519 latitudes and ~30% (5.5-8.1 DU) at high latitudes; the correlation improves by 0.04-0.12 and the 520 slope significantly increases by 0.12-0.23 to the range 0.8-1.0 at different latitude bands due to 521 accounting for inadequate retrieval sensitivity to the lower and middle troposphere. 522 The middle and right columns of Figure 9 show comparisons during pre-RA and post-RA, 523 respectively. The comparison between OMI and ozonesondes with OMI AKs TOCs during the 524 pre-RA period is significantly better than these during the post-RA period in the tropics and mid-525 latitudes with SDs smaller by 3.4-5.5% and greater correlation. The MBs during the post-RA 526 period is smaller by ~2 DU at mid-latitudes, but larger by ~1 DU in the tropics. However, the post-527 RA comparison is similar to the pre-RA comparison at northern high-latitudes and is even better 528 at southern high latitudes probably due to the aforementioned ozonesonde issues. 529 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 Manuscript under review for journal Atmos. Meas. Tech.  shows good agreement with ozonesondes at these stations with overall MBs ≤ 3 DU and SDs less 533 than 5.1 DU. The comparison is also good even in the high latitude regions partially because the 534 Summit and Neymayer stations only have ozonesonde launches during local summer. Seasonal 535 dependent biases are clearly seen at Payerne, and bias trends can be seen at several stations with 536 positive trends at Summit and Neumayer and a negative trend at Naha. In the pre-RA and post-RA 537 periods, the MBs are typically within 2 DU and the SDs are typically smaller during the pre-RA 538 period except for Naha. The better comparison (both mean bias and standard deviation) during the 539 post-RA period at Naha is likely due to the switch to ECC ozonesondes beginning on November 540 13, 2008 from KC ozonesonde that have greater uncertainty (WMO, 1998). 541 Figure 2 also shows the MBs and SDs of the TOC differences between OMI and ozonesonde 542 convolved with OMI AKs at each station/location where there are at least 10 coincident 543 OMI/ozonesonde pairs. OMI data generally exhibit good agreement with ozonesondes at most of 544 the stations, with MBs of ≤ 3 DU and SDs of ≤ 6 DU. In the tropics (30° S-30° N), very large SDs 545 (>11 DU) occur at the two Indian stations (New Delhi, and Trivandrum). The large bias of >6 DU 546 at New Delhi is likely associated with the large uncertainties of the Indian ozonesonde data. Hilo 547 has large biases of ~4.5 DU with 3.2 and 6.2 DU for pre-RA and post-RA, respectively. Java also 548 has a large bias of ~5 DU but shows no much difference between pre-RA and post-RA. Consistent 549 ~2% and ~5% underestimates of OC by ozonesondes compared to OMI total ozone are found in 550 Hilo and Java, respectively (Thompson et al., 2012). These OC underestimates may partly explain 551 the large TOC biases in Hilo and Java. However, the reason for underestimates of ozonesonde-552 derived OC is unknown. In the middle latitudes, noticeably large SDs and/or biases occur at a few 553 stations such as Churchill, Sable Islands, Hohenpeissenberg, and Parah. Three Japanese stations, 554 Sapporo, Tateno, and Naha, exhibit relatively large biases of 2-3 DU and even larger biases before 555 switching from KC to ECC sondes. Almost half of the 11 northern high latitude stations (60° N-556 90° N) and two of the 6 southern high-latitude stations have large SDs/biases. In addition to 557 retrieval biases from the OMI data, some of the large biases or SDs might be partially related to 558 ozonesonde type with different biases and uncertainties due to different types (e.g., Indian sonde 559 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 15 February 2017 c Author(s) 2017. CC-BY 3.0 License. stations, Brewer-Mast ozonesonde at Hohenpeissenberg, three KC sonde stations), manufacturers 560 (e.g., SP vs. ENSCI for ECC sonde), sensor solution or related to individual sonde operations, 561 which was shown in the validation of GOME ozone profile retrievals (Liu et al., 2006a). 562 Figure 11 shows the comparison for each season at northern mid-latitudes. Consistent with profile 563 comparison, the TOC comparison is season-dependent. When applying OMI AKs, the mean bias 564 varies from 3 DU in winter to -1.5 DU in summer. The SDs are within 6.8 DU with the smallest 565 value during fall due to less ozone variability. The regression slopes are very close, within 0.04 566 around 0.67. The retrieval sensitivity is smallest during the summer as seen from the greatest 567 correlation and slope and relatively small standard deviation, and is the worst during the winter. 568 With OMI AKs applied to ozonesonde profiles, the MBs only slightly change (varying from 3.5 569 DU to -1.3 DU), but the SDs are significantly reduced to within 5.2 DU, the slopes significantly 570 increase by ~0.2 to 0.8-1.0, and the correlation improves significantly during the winter and spring. 571 Figure 12 compares the surface~550 hPa and surface~750 hPa ozone columns with ozonesonde 572 data in the middle latitudes during summer and the tropics. Compared to the TOC comparisons in 573 Figure 9 and Figure 11, the comparisons of these lower tropospheric ozone columns exhibit smaller 574 regression slopes and correlations that are a result of reduced retrieval sensitivity. In the tropics, 575 the slopes decrease from 0.78 in TOC to 0.65 in the surface~550 hPa ozone column and ~0.50 in 576 the surface~750 hPa column, with corresponding correlation from 0.83 to 0.74 in the surface-~550 577 hPa column, and 0.66 in the surface-~750 hPa column. This indicates that the retrievals in the 578 surface~550 hPa/750 hPa can capture ~65%/50% of the actual ozone change from the a priori. 579 During the middle latitude summer, the slope decreases from 0.71 in the TOC comparisons to 0.42 580 in the surface-~550 hPa comparisons and 0.32 in the surface-~750 hPa comparisons, with 581 corresponding correlation coefficients from 0.74 to 0.5 and 0.46. Thus, the retrievals in the 582 surface~550 hPa and ~750 hPa only capture ~40%/30% of the actual ozone change from the a 583 priori. The MBs are generally small within 0.5 DU (5%) with SDs of ~3.6 DU (20-28%) in the 584 surface~550 hPa ozone column and ~2.5 DU (25-36%) in the surface~750 hPa ozone column. 585 After applying OMI AKs to account for inadequate retrieval sensitivity and removing smoothing 586 errors, the slope significantly increases to approach 1 (as expected). SDs are reduced to ~10% in 587 the middle latitudes and ~15% in the tropics. 588

Evaluation of Long-term Performance 589
Previous evaluation indicated systematic differences between pre-RA and post-RA periods and 590 generally worse performance during the post-RA periods. To further illustrate the long-term 591 stability of our ozone profile product and understand the quality of OMI radiometric calibration as 592 a function of time, we analyze monthly MBs of OMI/ozonesonde differences with OMI retrieval 593 AKs in ozone profiles, SOCs, and TOCs. Due to the lack of OMI observations during some months 594 at high-latitudes, we focus the evaluation by using coincidence pairs in 60° S-60° N. Monthly MBs 595 are calculated only if there are more than 5 OMI-ozonesonde pairs in a given month. Linear 596 regression trend is on the MBs for the entire period (2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014) and/or for the pre-RA and post-597 RA periods, respectively. The trend is considered statistically significant if its P value is less than 598 0.05. 599 The linear trends of monthly mean ozone biases for each OMI layer between 60° S-60° N are 600 plotted in Figure 13  indicates the need to remove signal dependent errors and the calibration inconsistency across the 630 track. To maintain the spatial consistency and long-term stability of our ozone profile product, we 631 need to further improve OMI's radiometric calibration especially during the post-RA period. 632 Preferably, the calibration improvement should be done in the level 0-1b processing. If this option 633 is not possible, we can perform soft calibration similar to Liu et al. (2010b) but derive the 634 correction as a function of time and latitude/SZA. In addition, it should be noted that the trend 635 calculation might be affected by factors such as the availability of correction factors with 636 ozonesondes (Morris et al., 2013), station-to-station variability and the uneven spatiotemporal 637 distribution of the ozonesondes, which can introduce considerable sampling biases (Liu et al., 638 2009;Saunois et al., 2012). 639

Summary and Conclusion 640
We conducted a comprehensive evaluation of the quality of OMI ozone profile (PROFOZ) 641 products produced by the SAO algorithm, including their spatial consistency and long-term 642 performance using coincident global ozonesonde observations during the decade 2004-2014. To 643 better understand retrieval errors and sensitivity, we compared the retrieved ozone profiles and a 644 priori profile at individual layers with ozonesondes before and after being degraded to the OMI 645 vertical resolution with OMI retrieval average kernels (AKs). We also compared the integrated 646 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 Manuscript under review for journal Atmos. Meas. Tech. In addition, we quantified the dependence of retrieval performance on seasonality and several key 652 parameters including solar zenith angle (SZA), cloud fraction, and cross-track position. Finally, 653 we analyzed the monthly mean variation of the mean biases (MBs) to examine the long-term 654 stability of the PROFOZ product. 655 The comparison between OMI and ozonesonde profiles varies in altitude, with maximum standard 656 deviations (SDs) in the Upper Troposphere and Lower Stratosphere (UTLS) due to significant 657 ozone variability, and varies with latitude similarly in the northern and southern hemispheres. 658 There is good agreement throughout the atmosphere in the tropics and mid-latitudes. With the 659 application of OMI AKs to ozonesonde data, the MBs are within 6%, and the SDs increase from 660 5-10% for pressure < ~50 hPa to within 18%(27%) in the tropics/mid-latitudes for pressure > ~50 661 hPa. In the high latitudes, the retrievals agree well with ozonesondes only for pressure < ~50 hPa 662 with MBs of < 10% and SDs of 5-15% for pressure < ~ 50 hPa, but with MBs reaching 30% and 663 SDs reaching 40% for pressure > ~50 hPa. The comparison is seasonally dependent. At northern 664 mid-latitudes, the agreement is generally best (except for MBs) in the summer, with the best 665 retrieval sensitivity and the smallest SDs as great as 20%, and the worst in the winter with the 666 worst retrieval sensitivity and the largest SDs reaching 31%. The MBs near 300 hPa vary from 667 12% in the winter to -10% in the summer. The post-RA comparison is generally worse in the 668 tropics and mid-latitudes than the pre-RA comparison, with SDs larger by up to 8% in the 669 troposphere and 2% in the stratosphere, and with larger MBs around ~300 hPa in the mid-latitudes. 670 But at high latitudes, the pre-RA comparison does not show persistent improvement over the post-671 RA comparison, with smaller biases and larger SDs at some altitudes, especially at southern high 672 latitudes. The retrieval improvement over a priori can be determined from the SD reduction of the 673 retrieval comparison from the a priori comparison. The retrievals demonstrate clear improvement 674 over the a priori down to the surface in the tropics, but only down to ~750 hPa during mid-latitude 675 summer, ~550 hPa during the other seasons of mid-latitudes and ~ 300 hPa at high latitudes. 676 Retrieval performance typically becomes worse at large SZA, especially at SZA larger than 75°, 677 where the MBs in the troposphere are >20% and the SDs near ~300 hPa are > 45%. The worse 678 performance at larger SZA is due to a combination of weaker signal and greater influence by 679 radiometric calibration errors such as due to stray light, and radiative transfer calculation errors. 680 The variation of SZA is likely responsible for the majority of the retrieval dependence on latitude 681 and season. The retrieval quality for pressure > ~100 hPa degrades with increasing cloudiness in 682 terms of MBs and SDs, with MBs greater than 10% at cloud fraction > 0.3. The retrieval 683 performance also varies with cross-track position, especially with large MBs and SDs at the 684 first/last extreme off-nadir positions (e.g., 1-3 and 28-30). The dependence is stronger during the 685 post-RA period. 686 The integrated SOCs and TOCs also exhibit good agreement with ozonesondes. With the 687 convolution of OMI AKs to ozonesonde data, the SOC MBs are within 2% with SDs within ~5.1% 688 in the tropics and mid-latitudes. These statistics do not change much even without the applications 689 of OMI AKs. The comparison becomes slightly worse at high latitudes, with MBs up to 3% and 690 SDs up to 6%. The pre-RA comparison is generally better with smaller SDs of 0.2-0.6% except 691 for southern high latitudes, although with slightly larger MBs. The TOC MBs and SDs with OMI 692 AKs are within 6%, with SDs of <~15% in the tropics and mid-latitudes but reach 30% at high 693 latitudes. The pre-RA TOC comparison is also better in the tropics and mid-latitudes with SDs 694 smaller by 3.4-5.5% but worse values at southern high latitudes. The TOC comparison at northern 695 mid-latitudes varies with season, with MBs of 11%. There are worse correlation during winter 696 and MBs of -3% and best correlation in summer. The TOC comparison also shows noticeable 697 station-to-station variability in similar latitude ranges with much larger MBs and/or SDs at the two 698 Indian stations and larger MBs at several Japanese stations before they switched from KC 699 ozonesondes to ECC ozonesondes. This demonstrates the impacts of ozonesonde uncertainties 700 due to sonde types, manufacturers, sensor solution and operations. Without applying OMI AKs, 701 the TOC correlation with ozonesondes typically becomes worse at higher latitudes, ranging from 702 0.83 in the tropics to 0.5-0.6 at high latitudes. The linear regression slope is within 0.6-0.8, 703 typically smaller at higher latitudes, reflecting the smaller retrieval sensitivity down to the 704 troposphere at higher latitudes mainly resulting from larger SZA. The convolution of AKs 705 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 15 February 2017 c Author(s) 2017. CC-BY 3.0 License. significantly improves the correlation and slope. The impact of retrieval sensitivity related to SZA 706 is also reflected in the seasonal dependence of the comparison at mid-latitudes. 707 The surface-~550/750 hPa ozone columns in the tropics during mid-latitude summer compare 708 quite well with ozonesonde data, with MBs of < 5% and SDs of 20-25%/28-36% without OMI 709 AKs. The correlation and slope decrease with decreasing altitude range due to reduced retrieval 710 sensitivity down to the lower troposphere. These columns capture ~65%/50% of the actual ozone 711 change in the tropics and ~40%/30% in the troposphere. Convolving ozonesonde data with OMI 712 AKs significantly increases the slope to ~1 and reduce the SDs to 10-15%. 713 The contrast of pre-RA and post-RA comparisons indicates generally worse post-RA performance 714 with larger SDs. Linear trend analysis of the OMI/ozonesonde monthly MBs further reveals 715 additional RA impact. The temporal performance over 60° S-60° N is generally stable with no 716 statistically significant trend during the pre-RA period, but displays a statistically significant trend 717 of 0.14-0.7%/year at individual layers for pressure < ~80 hPa, 0.7 DU/year in SOC and -0.33 718 DU/year in TOC during the post-RA period. Because of these artificial trends in our product, we 719 caution against using our product for ozone trend studies. 720 This validation study demonstrates generally good retrieval performance of our ozone profile 721 product especially in the tropics and mid-latitudes during the pre-RA period. However, the 722 spatiotemporal variation of retrieval performance suggests that OMI's radiometric calibration 723 should be improved, especially during the post-RA period, including the removal of signal-724 dependent errors, calibration inconsistency across the track and with time to maintain the long-725 term stability and spatial consistency of our ozone profile product. 726 727 Acknowledgements 728 This study was supported by the NASA Atmospheric Composition: Aura Science Team 729 (NNX14AF16G) and the Smithsonian Institution. The Dutch-Finnish OMI instrument is part of 730 the NASA EOS Aura satellite payload. The OMI Project is managed by NIVR and KNMI in the 731 Netherlands. We acknowledge the OMI International Science Team for producing OMI data. We 732 also acknowledge the ozonesonde providers and their funding agencies for making ozonesonde 733 Atmos. Meas. Tech. Discuss., doi:10.5194/amt-2017-15, 2017 Manuscript under review for journal Atmos. Meas. Tech. Discussion started: 15 February 2017 c Author(s) 2017. CC-BY 3.0 License.