The OMPS instrument was successfully launched aboard the SNPP satellite on 28 October 2011 (Zhu et al., 2025a). The satellite operates in a sun-synchronous polar orbit at an average altitude of 833 km with a 13:30 local time ascending node, commencing routine scientific observations in early 2012 (Kramarova et al., 2022). The OMPS suite integrates three distinct sensors: the Nadir Mapper (NM), Nadir Profiler (NP), and Limb Profiler (LP) (Flynn et al., 2014). Among them, OMPS/LP is centrally aimed at retrieving the vertical distribution of ozone in the Earth's middle atmosphere with high precision, employing a limb observation mode to sound the atmosphere by imaging the edge of the Earth's atmosphere. During a limb observation, the sensor's line-of-sight passes tangentially through the atmosphere, and the point along this path with the lowest altitude is termed the tangent point (TP). The vertical distance from this point to the Earth's geoid is referred to as the tangent height (TH). The fundamental geometry of this observation mode is depicted in Fig. 1a.

The spectral coverage of OMPS/LP ranges from 290 to 1000 nm, with spectral resolution varying with wavelength from 1.5 nm at short wavelengths to 40 nm at the longer-wavelengths (Kramarova et al., 2014). Equipped with a charge-coupled device (CCD), the instrument can simultaneously observe scattered solar radiation across the full spectral range at altitudes from 0 to 100 km. Each detector pixel possesses an instantaneous vertical field of view of approximately 1.5 km. This configuration enables a high-precision vertical sampling of 1 km at the tangent point (Jaross et al., 2014). To expand cross-track coverage, OMPS/LP is configured with three observation slits horizontally spaced by 4.25° (approximately 250 km), whose observational geometry and field-of-view characteristics are illustrated in Fig. 1b. Each slit has a 1.85° vertical field of view (FOV), corresponding to a 110 km vertical observation range at the TP. This study focuses on measurement data from the central slit, which is aligned with the satellite's ground track. The SNPP satellite completes 14 orbits daily, with OMPS/LP performing approximately 160–180 measurements per orbit (at a latitudinal sampling interval of ∼ 1°). LP can achieve global coverage every 3–4 d (Kramarova et al., 2024).

Radiometric errors and sensor pointing errors are the two main error sources affecting limb-scattering ozone retrieval accuracy (Kramarova et al., 2024). The OMPS/LP L1G v2.6 dataset incorporates essential corrections to address these issues.

Pointing (altitude registration) corrections are applied to mitigate tangent height offsets caused by instrument alignment and thermal effects. Multi-point corrections include static, intra-orbit, and time-dependent adjustments following Moy et al. (2017).

Stray light correction is performed using an updated point spread function (PSF) based on pre-launch measurements (Jaross et al., 2014). In version 2.6, the PSF tail intensity in UV and VIS/NIR bands is increased by ∼ 12 % to improve high-altitude stray light estimation (Kramarova et al., 2024). An additional factor of 1.5 is applied in VIS/NIR wavelengths to correct for in-band scattering. While thermally induced wavelength shifts have negligible impact on height-normalized radiances in ozone retrieval, we note that residual wavelength-dependent errors could affect cross-section matching in regions of strong ozone absorption.

These calibration steps are critical for ensuring the radiometric and geometric accuracy of the radiances used in our retrieval. Further details can be found in the cited references.

The retrieval method for vertical ozone concentration distributions based on OMPS/LP measurements in this study draws on the technical framework developed by Zhu et al. (2021), who derived ozone number density profiles using SCIAMACHY limb scattering measurements in the Chappuis–Wulf band. It shares similar methodological principles with the approaches proposed by Roth et al. (2007) and Degenstein et al. (2009), all of which employ retrieval vectors positively correlated with ozone concentrations for calculations.

The first step in the retrieval process involves normalizing the limb radiance at selected wavelengths. This operation entails normalizing the limb radiance at each wavelength to a reference TH, which effectively eliminates interference from the solar Fraunhofer structure, weakens the impact of surface reflection, and simultaneously achieves instrument self-calibration (Jia et al., 2015). 1Inorλ,H=Iλ,H/Iλ,Href where, H denotes the TH, and λ represents the wavelength. Iλ,Href and Inorλ,H refer to the radiance at the reference TH and the normalized radiance, respectively. The reference TH is an upper altitude where ozone sensitivity is low; in this study, it is selected as 40.5 km (i.e., the reference TH above the maximum retrieval altitude). Although radiance normalization cannot completely eliminate the influence of surface reflection or correct spectral errors such as wavelength shifts (which affect the calculation of ozone absorption cross-sections), it significantly reduces the requirements for both absolute radiometric calibration accuracy and modeling accuracy (Flittner et al., 2000). To mitigate the effect of aerosol scattering, the Chappuis triplet vector (CTV) method proposed by Degenstein et al. (2009) and Flittner et al. (2000) is employed for wavelength pairing. In the Chappuis-Wulf band, the CTV is defined as the difference between the logarithmic average of normalized radiances at two weakly ozone-absorbing wavelengths and the logarithm of the normalized radiance at a wavelength near the ozone absorption peak, thereby isolating the ozone absorption signal from common background scattering effects (e.g., aerosol scattering). It is expressed as: 2yj=lnInorλref1,j⋅Inorλref2,jInorλp,j where, j denotes the index of the TH measured by the instrument, and yj represents the retrieval vector after wavelength pairing at the jth of TH. λref1, λref2, and λp correspond to the weakly ozone-absorbing reference wavelengths and the strongly absorbing peak wavelength, respectively. In this study, the peak wavelength λp=606.3 nm corresponds to the visible channel configuration employed in the NASA OMPS/LP v2.6 operational algorithm (Kramarova et al., 2024), thereby ensuring consistency with established OMPS retrieval products. The weakly absorbing reference wavelengths λref1=512 nm and λref2=675.5 nm were optimized according to the selection criteria proposed by Zhu et al. (2021) for limb scattering ozone retrievals within the Chappuis-Wulf band, which take into account the specific spectral response and noise characteristics of OMPS/LP. Unlike the NASA algorithm, which uses spectral averages over multiple wavelengths for its visible triplet (510, 606, 675 nm; Kramarova and DeLand, 2023), the proposed method adopts individual discrete wavelength channels.

The CTV is designed to be positively correlated with ozone concentration (Degenstein et al., 2009). As expected, the CTV values and the retrieved ozone profiles show consistent vertical and latitudinal variations, with peak altitudes decreasing from the tropics to high latitudes. In this study, CTV values near zero above 35 km exhibit insufficient sensitivity to ozone, and values above 40 km become negative; therefore, the retrieval is restricted to altitudes below 40 km.

Aerosols, as suspended particles capable of absorbing and scattering light, have sources including both natural and anthropogenic factors. Stratospheric aerosols mainly originate from SO₂, HCl released by volcanic eruptions, naturally generated OCS, and pollutants such as SO₂ from industrial emissions (Li et al., 2023b). The presence of aerosols enhances the intensity of atmospheric scattered light, with the effect being stronger at longer (red) wavelengths than at shorter (blue) wavelengths due to wavelength-dependent scattering. Based on the SCIATRAN model (radiative TRANsfer model for SCIAMACHY), this study conducted simulation experiments on aerosol extinction coefficients to explore their impacts on radiance and CTV.

Figure 2 shows the effects of aerosol extinction profiles perturbed with different scaling factors on radiance and CTV. We perturbed the aerosol extinction profiles with factors from 0.1 to 10 and we found that the radiance profile is positively correlated with the aerosol extinction coefficient, while CTV decreases as the extinction coefficient increases, with the impact mainly concentrated below 30 km. For instance, when the aerosol extinction profile doubles, the radiance value at 21 km increases by 27 %, while CTV decreases by only 5 %. It indicates that wavelength pairing can weaken the aerosol scattering effect but cannot completely eliminate it. In addition, the study found that both radiance errors and CTV errors increase with an increase in SZA.

Given the nonlinear nature of the retrieval problem, this study employs an iterative method for solution and selects the multiplicative algebraic reconstruction technique (MART) to perform ozone profile retrieval. As an improved algorithm of nonlinear relaxation techniques, MART has a main advantage in that it can utilize multiple sets of measurement vectors to realize the retrieval of atmospheric state parameters at any altitude (Roth et al., 2007). During the iteration process, the update of atmospheric states at each altitude depends on a multiplicative factor, which is obtained by weighted averaging the ratios of all valid observation vectors to simulation vectors. The general formula of the MART algorithm is as follows: 3xin+1=xin∑jyjobsyjmodWji=xinαi where, xindenotes the ozone number density at atmospheric height i during the nth iteration; yjobs and yjmod represent the observation vector and simulation vector processed via Eqs. (1) and (2), respectively, where ymod is generated by the radiative transfer model based on the ozone profile xn obtained from the nth iteration; Wji is the line-of-sight weight factor, indicating the importance of the jth TH or line of sight to the ozone retrieved at altitude i. At each altitude, ∑jWji=1. The value of Wji in this study follows the setting in Zhu et al. (2021). αi is the multiplicative update factors at atmospheric height i.

The sensitivity of retrieval to the true state and the contribution of prior information can be formally described through the averaging kernel and measurement response (Rodgers, 2000; von Clarmann et al., 2020). However, unlike Optimal Estimation (OE) approaches, the MART algorithm used in this study does not produce formal averaging kernels.

A critical step in the OMPS/LP ozone profile retrieval is to define the lower boundary by determining the cloud top height. This is accomplished using a cloud detection method, modified from Chen et al. (2016), which leverages the spectral contrast in radiance between red and near-infrared bands. The method quantifies this contrast by computing the change in the vertical radiance gradient between two selected wavelengths. The underlying premise of the gradient-based cloud detection algorithm is that clouds generate a significantly larger radiance gradient compared to aerosols. This gradient is quantitatively defined as the rate of change of radiance with respect to TH: 4Gλ,H=∂lnIλ,H/∂H As shown in Fig. 3a, the variation characteristics of radiance gradient with wavelength in the 500–900 nm provide a basis for determining cloud top height. In cloud-free conditions, the radiance intensity varies slightly with wavelength; whereas in the presence of clouds, the wavelength dependence of radiance is significantly stronger than that of aerosols. Based on this, in this study, the cloud top height is determined by calculating the spectral gradient difference, with the formula as follows: 5lnRH=Gλs,H-Gλl,H where, λs and λl denote the short wavelength and long wavelength, respectively. In this study, λs is set to 674 nm, and λl is set to 868 nm. The positive cloud detection threshold for LP data is 1.5, which is also applicable to the detection of polar mesospheric clouds (PMCs). Taking the SNPP satellite orbit 51220 on 15 September 2021 as an example (Fig. 3b), the characteristics of lnR profiles differ significantly between two cloudy events and one cloud-free event: the maximum value of lnR in the cloud-free event is below 1.5, while that in the cloudy events is above this threshold, and the TH corresponding to the maximum lnR value is the cloud top height. Retrievals are not performed below the cloud top height, while the multiplicative update factor above the cloud top is propagated downward into the cloudy region.

In this study, the SCIATRAN v2.2 toolbox (Rozanov et al., 2017) is employed as forward model to calculate the simulated radiances required for ozone concentration retrieval. The observed and simulated radiances are processed through normalization and wavelength pairing to form retrieval vectors, which serve as inputs to the MART algorithm to drive the iterative update of ozone profiles.

The radiative transfer solution in the forward model is based on the discrete ordinate method applied to a spherical atmosphere with a pseudo-spherical approximation for multiple scattering. The solution incorporates the effects of multiple scattering and refraction while explicitly omitting polarization. Radiance calculations in the model are focused solely on ozone, an absorbing gas, with the ozone absorption cross-sections taken from Bogumil et al. (2000). The pressure and temperature profiles used in this study were obtained from the Global Modeling and Assimilation Office (GMAO) interpolated dataset. These meteorological data are incorporated in the OMPS/LP L1G dataset provided by NASA (.2025a). In addition, the model sets the stratospheric background aerosol type as LOWTRAN (Kneizys, 1988), the boundary layer humidity as 80 %, and the boundary layer aerosol type as marine. The retrieved ozone profiles are reported on the same vertical grid as the OMPS/LP L1G input data, which has a fixed spacing of 1 km in tangent height. However, this sampling interval does not imply an equivalent effective vertical resolution. The true vertical resolution is generally coarser than 1 km, particularly in regions of lower measurement sensitivity. The prior profiles are from SCIATRAN's built-in database. These profiles are provided by McLinden climatology (User's Guide for the software package SCIATRAN, 2018) and include monthly and latitude-dependent vertical distributions of volume mixing ratios for O₃, NO₂, BrO, and OClO, as well as pressure and temperature in the 0 to 100 km altitude range.

In the field of error analysis on limb-scattering ozone retrieval, there is a wealth of academic studies. Zhu et al. (2022) used numerical perturbation to conduct formal error analysis on the retrieval method of the weighted multiplicative algebraic reconstruction technique, accurately quantifying ozone retrieval errors at different altitudes. Arosio et al. (2022) systematically evaluated random errors and systematic errors for stratospheric ozone profile retrieval based on OE algorithms. These research results provide important references for the error analysis of MART retrieval algorithm, and the error estimation results of this study are consistent with those in Arosio et al. (2022) and Zhu et al. (2022).

To ensure data quality, radiance measurements contaminated by clouds were systematically excluded during retrieval. Furthermore, the effective surface albedo was determined directly from OMPS/LP radiance measurements at 675 nm (referred to as scene reflectance in the NASA product), and the corrected THs provided in the NASA L1G data were adopted. Accordingly, the retrieval error budget focuses on four primary sources: prior profiles, aerosols extinction profile, ozone absorption cross-sections, and measurement noise.

We assess the sensitivity of the retrieval to the prior profile through a perturbation-based approach. In this study, the prior sensitivity analysis matrix A0 is used to quantify the sensitivity of retrieval to the prior profile, and its expression is as follows: 6A0=∂x^∂x0 where x0 and x^ represent the initial ozone profile and the retrieved ozone profile, respectively. To calculate the column vectors of A0, the ozone concentration at a single altitude was perturbed in x0 by 5 % and analyzes the corresponding changes in x^. Specifically,∂x0 is the perturbation applied to the prior, and ∂x^ is the difference between the retrieval using the perturbed prior and the retrieval using the unperturbed prior. A0 is a dimensionless matrix, whose characteristics can intuitively reflect the impact of changes in the prior profile on the retrievals.

Figure 4a illustrates the distribution of the prior sensitivity analysis matrix (A0) column vectors across the 12–40 km altitude range, with each curve plotted at a vertical grid resolution of 1 km. A peak centered near the perturbation altitude indicates that the retrieval at that altitude retains sensitivity to the prior value at the same level. The width of the peak reflects the degree of vertical smoothing inherent in the retrieval. In the lower stratosphere (below 20 km), the response amplitudes are strong, indicating that strong prior dependence due to reduced measurement information content under weak limb-scattering signals. This is consistent with the reduced information content of limb measurements in the upper troposphere and lower stratosphere, where the retrieval is primarily constrained by the prior.

While Fig. 4a illustrates the pattern of prior influence, it does not quantify the actual retrieval error that would result from an inaccurate prior. To assess this, the entire prior profile was uniformly scaled by +5 % at all altitudes, and the relative difference between the perturbed retrieval and the standard retrieval was computed. Figure 4b shows the relative error induced by a +5 % perturbation of the prior profile. Below 20 km, the retrieval shows sensitivity to the a priori, with relative errors -5 % in tropical regions. This indicates that a small increase in the prior profile leads to a noticeable underestimation of retrieved ozone concentrations in the tropical lower stratosphere, reflecting the high sensitivity of the retrieval to prior information in this region where measurement information content is low. With increasing altitude, the magnitude of the error progressively decreases. Above 25 km, the error approaches 0 % across all latitudes, At high latitudes, the error magnitude remains relatively small at all altitudes, indicating weaker prior dependence compared to tropical and mid-latitude regions.

The uncertainty of stratospheric aerosol extinction coefficients must be considered in the error budget for ozone profile retrievals. Previous studies have shown that retrieved aerosol extinction profiles have an average upper error bound of approximately 30 % in the lower stratosphere (Arosio et al., 2022). To assess the impact of this uncertainty on our ozone retrieval, we perturbed the climatological aerosol extinction profile by uniformly scaling it by +30 % at all altitudes and repeated the retrieval. The relative difference between the perturbed retrieval and the standard retrieval was then computed. Figure 5a depicts the resulting ozone retrieval errors as a function of latitude and altitude. It is evident that retrieval errors induced by variations in aerosol extinction coefficients are predominantly distributed within the 15–25 km altitude range, with a magnitude of approximately 5 %, and errors in the southern high latitudes are more pronounced. Within the 25–30 km range, the error is around 2 %, while above 30 km, it is less than 1 %. Below 15 km, errors vary with latitude, mostly falling within the range of ± 2 %.

The temperature dependence of ozone absorption cross-sections has the potential to introduce errors in the retrieved profiles. To assess this effect quantitatively, we followed the approach which applied a uniform +2 % perturbations to the ozone absorption cross-sections at all temperatures used in the forward model. This perturbation magnitude represents a typical conservative estimate of cross-section uncertainty in the Chappuis band (Arosio et al., 2022). The retrieval was then repeated using the perturbed cross-sections, and the relative difference with respect to the standard retrieval was computed.

As shown in Fig. 5b, the resulting retrieval error exhibits a distinct vertical and latitudinal structure. In the tropics, the largest negative deviations (-3 % to -5 %) appear below 20 km, indicating that retrievals in the tropical lower stratosphere are most sensitive to uncertainties in ozone absorption cross-sections. In the Southern Hemisphere (SH) mid-to-high latitudes, prominent negative deviations (-3 % to -4 %) are found below 26 km. In the Arctic region, the negative bias below 20 km is relatively smaller, at approximately -2 %. Across all latitudinal bands, the error stabilizes near -2 % above 25 km. These results confirm that uncertainties in ozone absorption cross-sections introduce systematic biases in lower stratospheric ozone retrievals, especially in the tropics and SH mid-to-high latitudes. Specifically, a positive perturbation in the cross-sections leads to an underestimation of ozone concentrations, as observed in the negative biases in Fig. 5b.

To quantify the impact of random measurement noise on retrieval precision, a Monte Carlo simulation was performed using data from OMPS/LP orbit 51220. Sixteen representative latitudes spanning from 80° S to 80° N were selected. For each latitude, Gaussian random noise with a standard deviation of 1 % was added independently at each tangent height to the retrieval vector yj (Eq. 2). This process was repeated 100 times, generating 100 independent noisy realizations per latitude. A full MART retrieval was conducted for each realization, producing an ensemble of 100 retrieved ozone profiles for each latitude. The random uncertainty due to measurement noise was quantified as the standard deviation of the 100 retrieved profiles at each altitude, expressed as a percentage of the average of the unperturbed profiles.

Figure 6 shows the latitudinal and altitudinal distribution of the resulting random uncertainty. Uncertainty remains low (<10 %) at most latitudes within the 20–33 km mid-stratosphere, reflecting robust and stable retrieval performance. Above 30 km, especially at high latitudes, uncertainty increases sharply to above 20 %, which is mainly attributed to weaker signals in the visible spectral range. In the tropics below 20 km, a region of elevated uncertainty (>15 %) is identified, likely associated with low ozone abundances, strong atmospheric variability, or reduced information content from the measurements.

The OMPS/LP v2.6 ozone profile retrieval algorithm developed by the NASA team is built on wavelength pairing and an optimal estimation with prior constraints (Kramarova and DeLand, 2023). This algorithm operates on combined UV-Vis measurement data from 12.5 km (or cloud top) to 57.5 km, producing a single ozone profile from each retrieval. The measurement vectors are obtained via doublet and triplet methods, with specific parameters detailed in Table 1. During algorithm implementation, the retrieved surface albedo, cloud top height, and corrected tangent height are incorporated. In the forward model, aerosol extinction coefficients retrieved from OMPS/LP measured data are used. For validation, only altitudes above the detected cloud top height are included in the comparison.

Figure 7 shows a comparison between the retrievals of this study and those of OMPS/LP v2.6, involving approximately 770 000 profiles. Among them, Fig. 7a presents an example of number density for the annual average profile, and Fig. 7b shows the relative differences of the annual data. In this study, the relative difference is calculated as follows: 7Edif=2⋅O3Ret-O3RefO3Ret+O3Ref×100% where O3Ret denotes the ozone profile retrieved in this study, and O3Ref represents the reference ozone profile product.

As presented in Fig. 7b, the ozone concentrations from our retrieval systematically exceed those of the OMPS/LP v2.6 product at altitudes below 19 km and above 35 km. The positive deviation increases with decreasing altitude, reaching a maximum of approximately 10 %–24 % at the upper and lower retrieval boundaries. The ozone concentration is slightly lower between 20 and 28 km, with a deviation within 3 %. There is an inherent negative deviation of about -6 % around ∼ 33 km. Overall, the deviation between 17 and 36 km is confined within 5 %.

Figure 8 shows the mean relative differences between the retrievals of this study and OMPS/LP v2.6 in tropical regions and southern, northern mid-high latitudes. In the tropical regions within the 18–36 km altitude range, the deviation is within 5 %, showing good consistency. At northern mid-high latitudes, the difference between 12 and 32 km reaches 8 %, and the difference above 32 km is as high as 8 %–13 %, with similar deviations in the Arctic region. In the southern mid-high latitudes, most of the deviations above 18 km are less than 3 %, but there is a positive deviation of up to 19 % near 15 km. In the Antarctic region, this positive deviation reaches approximately 10 % around 18 km.

The two datasets differ significantly in the upper troposphere and lower stratosphere (UTLS) region, especially in the tropical region, mainly due to the extremely low ozone concentration at this altitude. The large positive deviation at the upper boundary of retrieval is caused by the decreased ability of the visible spectrum to retrieve ozone at high altitudes, while the NASA product uses combined ultraviolet and visible spectrum information for retrieval at this altitude. Although there are differences between the retrievals of this study and the OMPS/LP v2.6 product in terms of ozone absorption cross-sections, prior profiles, aerosol settings, retrieval algorithms, and spectra, the overall consistency is high.

The Earth Observing System-Microwave Limb Sounder (EOS-MLS) aboard the Aura satellite was successfully launched on 15 July 2004 (Waters et al., 2006). The satellite completes about 14 orbits daily, achieving global coverage between 82° S and 82° N. MLS provides vertical ozone profiles from the upper troposphere to the middle atmosphere using the 240 GHz frequency band by detecting naturally emitted microwave thermal radiation from the Earth's atmospheric limb measurements. Detailed descriptions can be found by Waters et al. (2006).

For validation purposes, this study employs the latest MLS L2 version 5.0 data product (Schwartz et al., 2020), with data filtering applied in accordance with the protocols recommended by Livesey et al. (2022). To ensure collocation quality from the dataset, stringent criteria were enforced: we retained only those data pairs where the geographical separation between the OMPS/LP and MLS footprints was within 1° in both latitude and longitude, and the observation time difference was less than 6 h. In cases where multiple MLS profiles corresponded to a single OMPS/LP measurement, their average was computed and used. For consistent comparison with NASA products and ozonesonde data, the MLS volume mixing ratio (VMR) and pressure were first transformed to number density and altitude, utilizing the MLS geopotential height and temperature. These converted profiles were subsequently interpolated onto the regular altitude grid of the OMPS retrievals using a spline method.

Figure 9 presents the average profiles and relative differences between the retrieval results of this study and MLS v5.0, involving approximately 93 000 profiles. The results show that compared with MLS v5.0, the ozone concentration retrieved in this study has a relatively large positive deviation of 5 %–20 % below 18 km; the positive deviation above 32 km increases with altitude; within the 18–35 km height range, the deviation between the two is confined within 5 %.

Figure 10 shows the mean relative differences in five latitude zones, with shaded areas indicating the standard deviations. In terms of data sample size, there are approximately 18 000 profiles in the tropical region, about 20 000 in the northern high latitudes, and around 15 000 in the southern high latitudes. The analysis results reveal that within the 12–35 km height range, the zonal average relative differences in each latitude zone are basically confined within 10 %. Among them, the northern mid-high latitudes exhibit a stable negative deviation of 5 %–10 % at 20–35 km; the southern mid-latitudes have a constant positive deviation of 1 %–5 % at 18–36 km, which increases to 3 %–10 % in the polar regions. In the tropical regions the differences change sign with altitude: the retrieved ozone number density at 19–27 km is 5 %–10 % lower than that of MLS, while it is 3 %–10 % higher in the 28–36 km range. It is worth noting that below 19 km, the consistency between the retrieval results of this study and MLS data decreases significantly, with the relative difference in the tropical region even exceeding 30 %, although the absolute difference is relatively small (see Fig. 10a).

Figure 11 depicts the altitude-dependent relative differences between retrieved profiles and MLS zonal means in 1° latitude bins across the three selected periods. The full-year 2021 data shown in Fig. 11a indicate that within the 20–35 km altitude range, differences across all latitudes are basically confined within ± 10 %. The altitude-dependent behavior of the retrieval biases can be summarized as follows. In the tropical UTLS region, oscillating differences exceeding 30 % are observed, which may be attributed to several factors: the highly dynamic variability of ozone concentrations, the limited detection sensitivity at the lowest retrieval altitudes, and the influence of cloud filtering. Furthermore, the inherently low ozone abundance in this region exacerbates the discrepancies in relative values. A distinct negative bias in retrieved ozone values is evident at 20–23 km in the tropics, particularly pronounced during winter (Fig. 11c). Conversely, a positive bias is observed over Antarctica, possibly linked to inaccuracies in the retrieved effective surface albedo, particularly in polar regions with high seasonal variability. Above 35 km, the retrievals exhibit a positive bias in the tropics. This bias arises from the inherently limited sensitivity of the visible spectrum at these altitudes, where ozone absorption is weak and measurement signals become dominated by noise and stray light.

In summary, the comparative analysis of this study shows that the effectiveness of OMPS retrieval varies across different regions and altitudes: the accuracy in tropical regions is concentrated in the 20–35 km altitude range; in mid-latitude regions, good consistency is also observed below 15 km. Nevertheless, in some atmospheric regions and under different seasonal conditions, the relative deviation may still exceed 10 % compared with MLS data.

In February 2001, the OSIRIS instrument was launched aboard the Odin satellite into a nearly circular sun-synchronous orbit (Llewellyn et al., 2004). The orbit has an altitude of approximately 600 km, an orbital period of 96 min, an inclination of 97.8°, and covers an observational latitude range from 82° S to 82° N. The satellite's ascending node crosses the equator at approximately 18:00 local time. Detailed descriptions of the instrument can be found by Llewellyn et al. (2004). Degenstein et al. (2009) used the MART to retrieve ozone profiles in the altitude range from 10 km or cloud top to 60 km, with the retrieval algorithm integrating radiation information from the UV and VIS bands. In this study, the version of OSIRIS L2 v7.3 data (University of Saskatchewan, 2025) is used for verification.

Owing to the sparsity of coincident OSIRIS measurements, a relaxed collocation criterion was adopted. Data pairs were considered matched if the geographical distance between the instrument footprints was within 2° in latitude and 5° in longitude, and the time difference was within 24 h. When multiple OMPS/LP profiles corresponded to a single OSIRIS observation, their average was used. To unify the data format, the ozone concentration unit (mol m⁻³) of OSIRIS profiles is converted to number density (molecule cm⁻³).

Figure 12 shows the average profiles and relative differences between the retrieval results of this study and OSIRIS v7.3, involving approximately 44 000 profiles. The results indicate that compared with OSIRIS v7.3, the ozone concentration retrieved in this study has a relatively large positive deviation of 28 %–34 % below 18 km; above 35 km, the positive deviation increases with increasing altitude; in the 20–35 km altitude range, except near 23 km, most of the deviations between the two are confined within 5 %.

Figure 13 further compares the mean relative differences of the five latitude zones, with shaded areas indicating the standard deviations. In terms of data sample size, there are approximately 17 000 profiles in the tropical region, about 10 000 in the northern high latitudes, and around 7000 in the southern high latitudes. The analysis shows that the northern mid-high latitudes have a significant negative deviation of 5 %–13 % at 21–25 km altitude, which is more prominent in polar regions; the difference in the southern mid-latitudes at 20–36 km altitude is less than 4 %, and the consistency in the Antarctic region at 23–35 km altitude is better than 2.5 %; most of the deviations in the tropical region at 26–36 km altitude are within 2 %. In addition, the differences in the region below 20 km are significant, with the relative difference in the tropical region exceeding 50 % and reaching 30 % at 13 km in the southern mid-latitude zone.

To robustly validate the retrieved ozone concentrations at altitudes below 30 km, this study employs a comparative analysis with ozonesonde measurements. The sonde data were obtained from the World Ozone and Ultraviolet Radiation Data Centre (WOUDC) and the Southern Hemisphere Additional Ozonesondes (SHADOZ) network (Thompson et al., 2007). Accounting for the sparse spatial distribution of ozonesonde stations, a relaxed collocation criterion was implemented: an OMPS/LP measurement was considered a match if it fell within ± 5° latitude and ± 10° longitude of a sonde station and occurred within ± 12 h of its launch. For each ozonesonde profile, all collocated OMPS/LP retrievals were averaged to form a single comparative data point.

In quantitative comparisons, to align the vertical resolution of ozonesonde data with that of OMPS data, a moving average filtering method is used for dimensionality reduction of ozonesonde data. The specific procedure begins with defining the window size of the moving average filter: 8N=ΔzlowΔzhigh where, Δzlow and Δzhigh represent the low vertical resolution of OMPS data and the high vertical resolution of ozonesonde data, respectively.

Filtering is applied to the original ozonesonde data xfine to obtain the dimensionality-reduced data xcoarse, which is expressed as: 9xcoarsezk=12N+1∑j=-NNxfinezk+j where zk denotes the altitude, and k represents the layer index.

In addition, another processing approach involves convolving ozonesonde measurements with the averaging kernels (AKs) retrieved from OMPS/LP v2.6 (see Arosio et al., 2018 for details). Taking the Alert station (82.5° N, 62.4° W) as an example, Fig. 14 presents the comparison results between the ozonesonde data and the collocated OMPS average profile on 15 September 2021. It is found that there are differences in the dimensionality-reduced ozonesonde data obtained by the two methods. The data curve processed by convolution with averaging kernels is smoother and its shape is closer to the OMPS product; while the data processed by moving average filter retains more original features with a sharper curve. After comprehensive consideration, the moving average filter is finally adopted for data processing in this study.

Figure 15 presents the annual average collocated profiles and their relative differences. The left side of the figure indicates the number of valid collocations at each altitude, with the total sample size amounting to approximately 1460. This study included data from 41 ozonesonde stations, involving over 1700 individual profiles. A significant positive bias of 10 %–15 % is observed in the retrievals compared to ozonesonde data below 18 km, while the deviation generally stays within 10 % across the 15–30 km altitude range.

Figure 16 presents the mean relative differences across five latitude zones. The number of available collocations in the tropical and Antarctic regions is approximately 200 and 75, respectively. Specifically, the tropical region shows good consistency above 20 km, with most relative differences within ± 4 % between 20 and 30 km. The southern mid-latitude region also exhibits high consistency, with positive biases generally less than 5 % above 20 km, but a relatively large positive bias of about 19 % near 15 km. In the Antarctic region, the bias is less than 2 % above 25 km, while significant differences occur below 25 km, with a positive bias as high as 21 % at 19 km. The northern mid-latitude zone has a bias of less than 5 % below 21 km, but a constant negative bias of 5 %–11 % between 21–28 km, and this negative bias is more pronounced in the Arctic region.