Using the satellite measured solar irradiance for the DOAS SO₂ retrieval can effectively reduce instrument-related errors. However, due to the degradation and non-uniformity of the diffuser plate of OMS, the OMS L1 irradiance measurements experienced increasing errors after launch, especially in the shortwave UV region. For example, after one year on orbit, the intensity of OMS irradiance at the shorter wavelength of 317 nm has decreased by about 8.83 %, while at the longer wavelength of 331 nm, it has decreased by about 6.07 %. Therefore, in this study the Total and Spectral Solar Irradiance Sensor-1 (TSIS-1) Hybrid Solar Reference Spectrum (HSRS) hybrid solar reference spectrum (Coddington et al., 2021) was used for OMS SO₂ retrievals instead of OMS daily measured solar irradiance. The TSIS-1 HSRS hybrid solar reference spectrum was developed by normalizing high spectral resolution solar datasets to the absolute irradiance scale of the TSIS-1 Spectral Irradiance Monitor (SIM) and the CubeSat Compact SIM (CSIM). The high spectral resolution solar data are sourced from the Air Force Geophysical Laboratory (AFGL) ultraviolet solar irradiance balloon observations, the ground-based Quality Assurance of Spectral Ultraviolet Measurements in Europe Fourier Transform Spectrometer (QASUMEFTS) solar irradiance observations, the Kitt Peak National Observatory (KPNO) solar transmittance atlas, and the semi-empirical Solar Pseudo Transmittance Spectrum (SPTS) atlas. The TSIS-1 HSRS spans 202–2730 nm at 0.01 to ∼ 0.001 nm spectral resolution with uncertainties of 0.3 % between 460 and 2365 nm and 1.3 % at wavelengths outside that range (Coddington et al., 2021). The TSIS-1 HSRS hybrid solar reference spectrum was convolved with the OMS-N ISRF to match the spectral characteristics of OMS radiance.

Before the DOAS fitting retrieval, spectral soft calibration was applied to the FY-3F/OMS radiance data. This procedure corrects wavelength shifts caused by instrument drift, temperature fluctuations, radiation effects, and nonlinearities by using known absorption features. It ensures the spectral accuracy and consistency of the OMS radiances, which is essential for accurate SO₂ retrieval and for reducing the impact of calibration-related errors.

The spectral soft calibration is performed using the peaks and valleys of the TSIS-1 HSRS hybrid solar reference spectrum (van Geffen and van Oss, 2003; Coddington et al., 2021). The detailed process includes: (1) selecting the high-resolution hybrid solar reference spectrum (Coddington et al., 2021), and convolving the reference spectrum with the slit function of the FY-3F/OMS instrument; (2) fitting the ratio of the solar reference spectrum to the observed radiance spectrum with a low-order polynomial to enhance the observed radiance spectrum; (3) for each observed radiance spectrum in the SO₂ fitting window, Gaussian peak-finding is performed to match the peak position of a set of Fraunhofer lines with their corresponding wavelengths. A least-squares method is used to fit the spectral line wavelengths and the peak position number data with a third-order polynomial, generating a spectral calibration equation for each observed pixel. Then the spectral soft calibration of the FY-3F/OMS radiance is realized by using the above spectral calibration equation.

The reflected radiance detected by the satellite instrument contains information of trace gases integrated along the slant observation path. From the reflected radiance and above DOAS spectral fitting, the SO₂ SCD can be derived; however, the SCD is not suitable for the application of satellite-derived SO₂ in monitoring global climate change and air pollution, as it represents the SO₂ column along the slant path, which is influenced by the observation geometry and atmospheric conditions. The SCD can be converted to VCD using AMF = SCD/VCD (Palmer et al., 2001), which represents the relative length of the mean slant path at a certain wavelength for photons interacting with a certain absorber in the atmosphere relative to the vertical path (Lorente et al., 2017).

The AMF can be expressed as a function of these parameters: 2AMF=f(SZA,VZA,RAA,AS,HS,S(z),O3,λ,clouds,aerosols) Where AS is the surface reflectance, HS is the terrain height, S(z) represents the SO₂ vertical profile shape, O₃ is total ozone column. In practice, the AMF is typically computed as the weighted average of altitude-dependent Box-AMFs (equivalent to scattering weights) across all layers, with the weights determined by the SO₂ distribution in each layer (Eq. 3) (Chen et al., 2009; Wagner et al., 2007; Palmer et al., 2001; Boersma et al., 2004): 3AMF=∑iBox-AMFi×ci⋅Δhi∑jcj⋅Δhj Where ci represents the SO₂ number density in the ith layer, and Δhi denotes the thickness of that layer. Box-AMFs quantify the contribution of each atmospheric layer to the total AMF and allows for flexible updates of AMF with new SO₂ profiles, eliminating the procedure of rebuilding the AMF lookup table (LUT).

In this study, OMS SO₂ AMFs were calculated using Box-AMFs derived from SCIATRAN and SO₂ vertical profiles from the Goddard Earth Observing System Composition Forecast (GEOS-CF) system (Fahrland et al., 2020) (Data source: https://portal.nccs.nasa.gov/datashare/gmao/geos-cf, last access: 10 March 2026). The GEOS-CF system combines the GEOS weather analysis and forecasting system with the GEOS-Chem chemistry module (Keller et al., 2021) to provide detailed chemical analyses of a wide range of air pollutants, including O₃, NO₂, SO₂, and PM_2.5. The GEOS-CF SO₂ vertical profiles, with a horizontal resolution of 0.25° × 0.25°, a temporal resolution of 1 h, and 72 vertical model layers extending up to 0.01 hPa, were temporally matched to the OMS overpass time and spatially interpolated to the center of each OMS pixel. The profiles were normalized and used as weighting functions in the AMF calculation.

To improve computational efficiency, a cloud-free Box-AMF LUT was built using SCIATRAN, considering six variables: SZA, VZA, RAA, AS, HS, and O₃ column. The detailed values for each variable are listed in Table 3. Although Box-AMFs are wavelength-dependent, the LUT in this study was calculated at 320 nm, approximately at the center of the OMS SO₂ retrieval window (312–326 nm). For each OMS pixel, Box-AMF values were obtained by multi-dimensional interpolation within the LUT according to the corresponding observational geometry (SZA, VZA, and RAA) and atmospheric and surface conditions (AS, HS, and O₃). The observational geometry parameters and surface height were directly obtained from the OMS L1 radiance product. The corresponding surface reflectance was interpolated from the OMI surface reflectance climatology (Kleipool et al., 2008). The total ozone column used for LUT interpolation was obtained from the corresponding OMS ozone product (Xu et al., 2025). Due to the current unavailability of synchronized and reliable OMS cloud and aerosol products, the effects of clouds and aerosol on SO₂ retrievals were not considered in this study. Consequently, the AMFs for all OMS pixels were calculated using the cloud-free Box-AMF LUT. Section 5.2 of this study provides a detailed description of the AMF error analysis.

The retrieval accuracy of SO₂ columns from FY-3F/OMS is affected by spectral and radiometric calibration errors, which are difficult to remove from OMS L1 radiance data and result in systematic biases in the retrieval results, such as along-track stripes and cross-track asymmetries (Boersma et al., 2004). The TSIS-1 HSRS hybrid solar reference spectrum used in OMS SO₂ retrievals does not contain instrument-related characteristics, meaning that certain calibration errors in the radiance data cannot be cancelled out during the DOAS fitting process. This may further contribute to systematic biases in the retrieved OMS SO₂ columns. In addition, due to the low SNR of measurements in the fitting window, the weak SO₂ absorption information included in the TOA reflected radiance, and the interference from strong O₃ absorption in the fitting window, the OMS SO₂ retrievals tend to be systematically overestimated or underestimated over the whole orbit. These systematic biases in the retrieval results of SO₂ columns are mixed with the absorption information of SO₂ and limit the applicability of the OMS SO₂ product. Therefore, it is necessary to apply a background offset correction to reduce these systematic biases in the retrieved SO₂ columns. The reference region method (Khokhar et al., 2005; Richter et al., 2006) was used to correct the background offsets, which are latitude-dependent and related to the cross-track position. This method selects retrieval values over a reference region (e.g., clean oceanic Pacific regions, assumed to be areas with no SO₂) as the background area. The SO₂ retrievals are then adjusted by subtracting the background offsets of the same latitude over the ocean. Yang proposed the sliding median correction method (Yang et al., 2007) and applied it to the OMI SO₂ product. This method performs averaged sampling within a sliding window centered on the pixel (selecting pixels with values less than 2 DU) to get background offsets for each row of pixels. With the sliding window method, the cross-track and along-track biases varying with time and location can be effectively eliminated.

For OMS SO₂ retrievals, based on the above background offset correction methods, we developed an improved iterative sliding correction scheme to avoid seam problems due to discontinuous integration times within the same orbit. The details of the background offset correction used for OMS SO₂ retrievals are as follows.

Firstly, based on the integration time of FY-3F/OMS L1 data, the orbital data is divided into several data blocks corresponding to different integration times. For each data block, the mean vector (V0) at each cross-track position is estimated using all valid pixels within the block (i.e., pixels with normal L1 data, as indicated by the L1 QA quality flag). Each scan line within the data block is processed by subtracting V0 from the SO₂ retrievals of each scan line to obtain the initially corrected SO₂ columns.

Ideally, the retrieved SO₂ values from satellite observations in clean regions should be close to zero. The scatter of SO₂ columns in clean regions can reflect the reliability and stability of the satellite data and the retrieval algorithm (Krotkov et al., 2008). In this study, two clean oceanic regions, where SO₂ emissions are extremely low and assumed to be zero, were selected as the case studies to compare and evaluate the precision of FY-3F/OMS SO₂ retrievals. One is the area of latitude from 5° S to 5° N and longitude from 30 to 20° W (Region 1 in Fig. 5), and the other is the area of latitude from 5° S to 5° N and longitude from 135 to 125° W (Region 2 in Fig. 5). Due to the geolocation differences between OMS and TROPOMI orbits, the orbital pixels of OMS and TROPOMI DOAS and TROPOMI COBRA SO₂ product over the two clean oceanic regions were resampled to 0.15° × 0.15° equal latitude-longitude grid for comparison. As suggested in the TROPOMI README file, only pixels with QA > 0.5 were used for the TROPOMI DOAS SO₂ product in this section. For the cases of clean oceanic regions, TROPOMI COBRA L3 grid PBL SO₂ products were used instead of the COBRA 7 km product.

As shown in Figs. 6, 7, 8, and 9, the SO₂ columns retrieved from both OMS and TROPOMI over Region 1 and Region 2 are generally low and close to zero, and they approximately follow a normal distribution centered around 0, with most values concentrated between -2 and 2 DU. However, there are small differences between OMS and TROPOMI SO₂ over clean oceanic regions, which become larger as the pixels approach the edge of the orbit, where retrieval uncertainties are larger. For Region 1, the standard deviations of OMS, TROPOMI DOAS and TROPOMI COBRA SO₂ columns are 0.1537, 0.2493 and 0.1156 DU on 23 August 2024, and 0.1672, 0.3934 and 0.1865 DU on 15 November 2024, respectively. For Region 2, the corresponding standard deviations are 0.1539, 0.2636 and 0.1035 DU on 23 August 2024, and 0.1284, 0.3595 and 0.1615 DU on 15 November 2024, respectively. Compared with TROPOMI DOAS SO₂ results, both TROPOMI COBRA and OMS DOAS SO₂ results show lower standard deviations and are closer to zero over Regions 1 and 2. The differences between OMS and TROPOMI SO₂ columns may be primarily related to differences in local overpass times, observation angles, L1 and L2 processing algorithms.

Overall, comparisons from different dates and different clean oceanic regions indicate that FY-3F/OMS SO₂ retrievals have a reliable precision of approximately 0.15 DU over low-emission regions, and the data quality is relatively stable over time. It is worth noting that the retrieval errors for both OMS and TROPOMI are relatively large at the edges of the orbit, which is likely due to fewer valid pixels in these regions, resulting in larger standard deviations; thus, these pixels should be used with caution.

The massive amounts of SO₂ released over a short period during volcanic eruptions, as well as their long-distance transport and the subsequent formation of sulfate aerosols, not only influence the global radiative energy balance (Mccormick et al., 1995) but also pose risks to aviation in the tropopause or stratosphere (Miller and Casadevall, 2000). Through the comparison in volcanic regions, the capability of OMS SO₂ retrievals at large columns can be evaluated. In this section, we took the eruptions of Sundhnúkur volcano (Region 3 in Fig. 5) and Nyamuragira volcano (Region 4 in Fig. 5) as case studies for comparing FY-3F/OMS SO₂ with TROPOMI DOAS and TROPOMI COBRA 7 km SO₂ results. To present the SO₂ maps of OMS and TROPOMI DOAS more clearly, the resampling scheme in Sect. 4.1 was not used in Sect. 4.2. For TROPOMI DOAS, we used all pixels from the TROPOMI DOAS SO₂ product instead of applying QA > 0.5 to filter the high-quality pixels when comparing with the OMS SO₂ results. The reason is that after applying the QA > 0.5 filter, a lot of high SO₂ pixels in TROPOMI DOAS over volcanic regions would be missing (not shown here), making it difficult to compare with OMS SO₂ results. In this section, TROPOMI COBRA L3 grid 7 km SO₂ products were used instead of PBL products.

The Sundhnúkur volcano (Region 3 in Fig. 5) is the first case study for comparing FY-3F/OMS SO₂ with TROPOMI DOAS and TROPOMI COBRA 7 km SO₂ results over a volcanic eruption. On 22 August 2024, Sundhnúkur volcano within the Reykjanes volcanic system began erupting and continued to emit SO₂ for approximately 14 d (https://volcano.si.edu/, last access: 10 March 2026). The eruption created a fissure approximately 3.9 km long, with lava and smoke reaching a height of about 1 km. The region of latitude from 55 to 65° N and longitude from 30 to 5° W was selected as the study area for the Sundhnúkur volcano. There are two OMS orbits (20240823_1036 and 20240823_1217) (Format: YYYYMMDD_HHMM, UTC time of the first scan line) overpassing the Sundhnúkur volcano region on 23 August 2024. For the OMS orbit 20240823_1217, most of the pixels covering the volcanic region are near the edge of the orbit where the measurement noise tends to be higher. For TROPOMI DOAS, although two orbits (20240823T125304 and 20240823T111134) (Format: YYYYMMDDTHHMMSS, observation start UTC time) passed over the Sundhnúkur volcano area, only orbit 20240823T125304 is presented because most of its pixels over the Sundhnúkur volcano region were near the nadir of the orbit where the data quality is higher. Compared with OMS orbit 20240823_1036, OMS orbit 20240823_1217 has a local overpass time closer to that of TROPOMI orbit 20240823T125304 in the volcanic region, making the SO₂ results of OMS orbit 20240823_1217 more consistent with those of TROPOMI orbit 20240823T125304.

As shown in Fig. 10, both OMS and TROPOMI successfully captured the high SO₂ distribution around the Sundhnúkur volcano on 23 August 2024. The spatial distributions of OMS, TROPOMI DOAS and TROPOMI COBRA SO₂ over Sundhnúkur volcano are similar, but differ at the edge of the SO₂ plume. The correlation between OMS and TROPOMI DOAS reaches ∼ 0.78 over the Sundhnúkur volcano on 23 August 2024, while the correlation between OMS and TROPOMI COBRA reaches ∼ 0.70. However, when SO₂ values exceed 50 DU, OMS SO₂ retrievals are significantly lower than those of TROPOMI DOAS over the Sundhnúkur volcano region on 23 August 2024. Moreover, the relative biases between OMS and TROPOMI DOAS increase with increasing SO₂ columns. This may be attributed to the reason that the OMS SO₂ retrieval uses the 312–326 nm fitting window, where SO₂ has strong absorption and is prone to saturation in the case of high SO₂ concentrations, leading to an underestimation of SO₂ columns. In order to mitigate the risk of saturation, TROPOMI DOAS uses two additional fitting windows (325–335 and 360–390 nm) (S5P-BIRA-L2-ATBD-400E) for volcanic eruption cases. In addition, the different overpass time of OMS and TROPOMI, along with varying volcanic eruption strength and meteorological conditions, may also be major contributors to the differences in SO₂ columns of OMS and TROPOMI DOAS. The TROPOMI COBRA SO₂ results retrieved from the 310.5–326 nm window are also lower than those from TROPOMI DOAS, but are more consistent with those from OMS.

The Nyamuragira volcano (Region 4 in Fig. 5) is the second case study for comparing FY-3F/OMS SO₂ with TROPOMI DOAS and TROPOMI COBRA 7 km SO₂ results over a volcanic eruption. Nyamuragira is Africa's most active volcano and a high-potassium basaltic shield volcano located in the eastern part of the Democratic Republic of the Congo, approximately 25 km north of Lake Kivu and 13 km north-northwest of the Nyiragongo volcano (https://volcano.si.edu/; Global Volcanism Program, 2024b). Based on the Volcanic Explosivity Index (VEI) classification (Newhall and Self, 1982) and eruptive history reports of Nyamuragira from the Global Volcanism Program (GVP; Global Volcanism Program, 2024a), the magnitude of Nyamuragira's eruptions can generally be classified as small to moderate. According to the GVP weekly reports, Nyamuragira had continuing eruptive activities in November 2024. The spatial distribution maps (Fig. 11) show that OMS, TROPOMI DOAS, and TROPOMI COBRA results clearly detected the high-concentration SO₂ plume from the Nyamuragira eruption, although the shape of the SO₂ plume differs due to differences in overpass time, observation angles and retrieval strategies. Compared with OMS and TROPOMI DOAS, TROPOMI COBRA shows lower SO₂ columns over the Nyamuragira volcano. Although OMS and TROPOMI DOAS exhibit more comparable SO₂ levels, the quantitative agreement between them is weaker in scatter plot comparisons (not shown). This is primarily because Nyamuragira volcano is located near the equator, unlike the high-latitude Sundhnúkur volcano, making it difficult to identify OMS and TROPOMI orbits with closely matched observation times. Differences in overpass times lead to shifts in plume position, which hampers direct pixel-to-pixel quantitative comparisons over the Nyamuragira region.

Overall, from the above comparisons between FY-3F/OMS SO₂ and TROPOMI DOAS and TROPOMI COBRA 7 km SO₂ results, we can see that FY-3F/OMS has the capability to monitor volcanic activities, and with high spatial resolution of 7 km × 7 km and a local overpass time different from TROPOMI, FY-3F/OMS can contribute to a more effective satellite SO₂ product for the continuous monitoring of global volcanic activity.

Compared to the monitoring of high SO₂ emissions from natural sources such as volcanoes, the monitoring of anthropogenic SO₂ emissions from satellite observations is more challenging. The atmospheric SO₂ from anthropogenic emissions is generally much lower compared to that of volcanic eruptions, and is primarily concentrated near the surface. The sensitivity of satellite measurements in the UV band near the surface is relatively low because solar UV radiation is partially absorbed and scattered by atmospheric components such as air, aerosols, and clouds during its transmission. As a result, the weakened UV radiation reaching the boundary layer reduces the sensitivity of satellite instruments to PBL SO₂ and makes it harder to distinguish SO₂ signals from background noise, especially at large solar zenith and satellite viewing angles.

Based on the SO₂ emission sources observed by TROPOMI in 2018 (Fioletov et al., 2020), we selected three representative regions – the Persian Gulf (oil and gas exploration, Region 5 in Fig. 5), Norilsk (smelters, Region 6 in Fig. 5), and Eastern India (power plants, Region 7 in Fig. 5) – to compare three SO₂ column products: OMS SO₂, TROPOMI DOAS SO₂, and TROPOMI COBRA PBL SO₂. These comparisons aim at evaluating the capability of OMS in monitoring SO₂ emissions from anthropogenic sources. Similar to the cases of volcanic eruptions, we used all pixels from the TROPOMI DOAS SO₂ product instead of applying QA > 0.5 to filter the high-quality pixels because the QA > 0.5 filter would remove many high SO₂ pixels over anthropogenic emission regions, especially over snow/ice surfaces such as Norilsk. Meanwhile, comparison results for the Persian Gulf and Eastern India regions during the study period show negligible differences whether the QA > 0.5 filter is applied or not (not shown here). In the case of anthropogenic emissions, TROPOMI COBRA L3 grid PBL SO₂ products were used for comparison instead of the COBRA 7 km SO₂ products.

The Persian Gulf was selected due to its high anthropogenic emissions, including SO₂ and NO₂, primarily from oil and gas extraction and refining industries (Mardani et al., 2025; Krotkov et al., 2016). This region has a higher probability of clear-sky days and is located in the low-latitude zone, which means relatively high SNR for satellite observations. As shown in Fig. 12, OMS effectively detects high SO₂ values over the Persian Gulf while maintaining a lower background noise. The variability of enhanced SO₂ columns from OMS are generally consistent with those from TROPOMI DOAS and TROPOMI COBRA. Differences are mainly manifested in magnitude and local-scale variability, which may be attributed to differences in satellite overpass time, retrieval algorithms, AMF, and spatial resolution. Based on the spatiotemporal distributions shown in Fig. 12, three representative days with relatively large and small differences (23 August and 4 and 8 November 2024) were selected for scatter plot comparisons among the three datasets (OMS SO₂, TROPOMI DOAS SO₂, and TROPOMI COBRA PBL SO₂). Scatter plots (Fig. 13) comparing OMS SO₂ with TROPOMI DOAS and TROPOMI COBRA SO₂ over the Persian Gulf show that: (1) Both TROPOMI DOAS and TROPOMI COBRA SO₂ products exhibit clear positive correlations with OMS, with correlation coefficients (R) ranging from ∼ 0.63 to 0.91. (2) The level of agreement among the three datasets varies with time. On 4 and 8 November, OMS shows good consistency with both TROPOMI products, with regression slopes close to unity (DOAS: slopes ≈ 1.00 and 0.98, R = 0.82 and 0.91; COBRA: slopes ≈ 0.93 and 0.96, R = 0.82 and 0.83), indicating good agreement under anthropogenic SO₂ conditions. However, on 23 August OMS retrieves higher SO₂ columns over the Persian Gulf emission hotspot region than both TROPOMI DOAS and COBRA, with regression slopes of ∼ 1.61 and 2.14, respectively. This discrepancy may be primarily attributed to the relatively low AMFs calculated using GEOS-CF profiles and OMI surface reflectance climatology over these high-SO₂ regions. As shown in Fig. 14, although the corresponding OMS SCD results on 23 August is relatively low, the low AMFs over region with high SCDs lead to enhanced VCD (VCD = SCD/AMF), resulting in higher OMS SO₂ VCDs over these high-SO₂ regions on 23 August. (3) For pixels with low SO₂ columns, the correlations among the three datasets are relatively weak, likely due to larger retrieval uncertainties associated with random noise and reduced sensitivity at low SO₂ conditions. It should be noted that Fig. 13 does not present results for consecutive days. This is mainly due to the unavailability of GEOS-CF data for some days, the presence of a large number of pixels located near the satellite swath edges, and/or extensive cloud coverage indicated by the Terra/MODIS true color images, which limit the reliability of the comparison among the three SO₂ datasets. Similar limitations also exist in the India case studies presented later.

Norilsk in northern Russia, located within the Arctic Circle, is one of the world's biggest sources of anthropogenic SO₂ emissions due to its massive nickel and metal smelting industry (Bauduin et al., 2014). With winter lasting 6 to 9 months and snow covering the ground for most of the year, the region's low temperatures and atmospheric stability hinder pollutant dispersion, leading to persistently high SO₂ concentrations over Norilsk. Large SO₂ emissions cause severe air pollution and acid rain, which make Norilsk one of the most polluted cities in the world. On 16 May 2024, the orbits of OMS and TROPOMI with close local overpass times over Norilsk were selected to minimize the influence of emission variability and meteorological differences on the comparisons of SO₂ columns. As shown in Fig. 15, both OMS and TROPOMI were able to detect the high SO₂ plumes over the Norilsk region, with similar spatial patterns and transport features. The SO₂ columns from OMS and TROPOMI DOAS show strong correlations, with correlation coefficients ranging from 0.80 to 0.85 over the Norilsk region. However, OMS SO₂ VCD retrievals are lower than those from TROPOMI DOAS, with the average relative biases (|OMS - TROPOMI| / TROPOMI) of the data from the Norilsk region in OMS orbit 20240516_0334 being approximately 38 % (excluding SO₂ columns smaller than 1 DU), and for OMS orbit 20240516_0516 being approximately 33 % (applying the same filtering criteria). These differences may be related to differences in AMF strategy, retrieval algorithm, and spatial resolution. In contrast, the TROPOMI COBRA PBL L3 products exhibits large data gaps due to quality filtering over the Norilsk region, limiting its applicability over Norilsk for this case. In addition, the Terra/MODIS true color image from the same day shows the presence of snow cover over the Norilsk region, highlighting OMS's capability to capture anthropogenic SO₂ emissions in high-latitude regions within the Arctic Circle.

India's SO₂ emissions are mainly from coal fired power plants, transportation, and agricultural activities, which are growing rapidly, increasing by more than 100 % from 2005 to 2015 (Krotkov et al., 2016; Kuttippurath et al., 2022). Eastern India, in particular, is a major hotspot due to the dense distribution of coal-based power plants and industrial facilities. As shown in Fig. 16, OMS successfully captured strong SO₂ plumes over Eastern India on 4, 5, 8, and 9 November 2024, with spatial patterns that are generally consistent with those retrieved by TROPOMI DOAS and TROPOMI COBRA. Figure 16 also includes Terra/MODIS true color images for the same days, which indicate relatively low cloud coverage over Eastern India, thereby minimizing the influence of clouds on the intercomparison of SO₂ retrievals. Despite the overall consistency in spatial distribution, OMS SO₂ retrievals over regions with strong SO₂ emissions are higher than those from TROPOMI DOAS and TROPOMI COBRA on 5, 8, and 9 November 2024. This overestimation is likely related to uncertainties in AMF calculations associated with the use of GEOS-CF SO₂ vertical profiles, OMI surface reflectance climatology, and aerosol loading over Eastern India. To further improve the accuracy of OMS SO₂ retrievals over India, future work will focus on incorporating more representative local SO₂ vertical profiles and more accurate surface reflectance databases.

Overall, the comparisons over the Persian Gulf, Norilsk, and Eastern India demonstrate that OMS SO₂ retrievals show good spatial consistency and strong correlations with TROPOMI DOAS and TROPOMI COBRA SO₂ products; OMS is capable of effectively distinguishing intense anthropogenic SO₂ signals from background noises. Owing to differences in local overpass times, OMS can provide effective complementary data to fill the monitoring gaps in TROPOMI SO₂ products.

The main error sources affecting OMS SO₂ SCD primarily include instrument-related noise and algorithm settings in spectral fitting (Table 4). These error sources are difficult to separate, making it challenging to quantitatively estimate how each individual error source contributes to the SCD retrieval uncertainty. Although OMS L1 provides specified performance requirements before launch, on-orbit measurements indicate that the OMS L1 errors deviate from the planned specifications for certain parameters. At the time of writing, the updated OMS L1 errors have not yet been obtained. Therefore, in this study, we do not report the contribution of each individual error source to the final SCD uncertainty. Instead, we use the DOAS Spectral Fitting Error (SFE) to derive the overall SCD uncertainty under the current OMS L1 instrument status and DOAS settings.

AMF uncertainty is a primary source of the uncertainty of OMS SO₂ VCD retrieval. In this study, OMS SO₂ AMFs were calculated using SO₂ vertical profiles from GEOS-CF (Keller et al., 2021) and Box-AMFs derived from the SCIATRAN model (Chen et al., 2009; Wagner et al., 2007; Palmer et al., 2001; Boersma et al., 2004; see Sect. 3.4 for more information). The uncertainty in OMS SO₂ AMF is mainly associated with SZA, VZA, RAA, AS, HS, O₃ column, the shape of SO₂ vertical profile (S(z)), and wavelength (λ) dependence (Lee et al., 2009). It should be noted that the clouds and aerosols were not considered in the OMS SO₂ AMF calculation. Therefore, AMF uncertainties associated with clouds and aerosols were not included in the present AMF error analysis.

In this section, sensitivity tests of OMS SO₂ AMF were conducted using the SCIATRAN model by computing Box-AMFs under different forward-model configurations (Fig. 19). And then the contribution of each input parameter to the AMF uncertainty was quantified by perturbing it according to its estimated uncertainty while keeping other parameters fixed. Finally, an overall estimate of uncertainty in OMS SO₂ AMF was derived. For the perturbation analysis, a set of typical atmospheric and surface conditions was adopted as the default configuration, with SZA = 32.9°, VZA = 0°, RAA = 0°, AS = 0.05, HS = 0 km, wavelength = 320 nm, O₃ column = 275 DU, clear-sky conditions, and with the assumption of surface reflectance as isotropic Lambertian equivalent reflector (LER).

The contributions of different parameters to the AMF uncertainty were evaluated across five scenarios: clean, boundary-layer pollution, boundary-layer pollution over snow and ice, volcanic degassing, and volcanic eruption. To represent these scenarios, four assumed SO₂ vertical profiles were constructed (Fig. 20), corresponding to clean conditions (SO₂ column = 0.11 DU), anthropogenic SO₂ emissions (SO₂ column = 5 DU), volcanic degassing with plume heights around 2 km (SO₂ column = 16.5 DU), and volcanic eruptions with plume heights around 6 km (SO₂ column = 120 DU).

Background offset correction is essential for the OMS SO₂ column retrievals which have obvious along-track stripes and cross-track asymmetries problems. However, it is hard to get the accurate background offset for SO₂ retrieval of each pixel. In this study, an iterative sliding correction scheme for background offset correction was applied to OMS SO₂ retrievals, with the purpose of forcing SO₂ values over clean or low SO₂ emission regions to zero. This approach helps addressing problems such as along-track stripes and cross-track asymmetry. However, the sliding window strategy assumes that pixels with values smaller than the threshold (2 DU) within the sliding window represent zero SO₂ emissions, which may lead to a loss of SO₂ information contained in the low-emission regions. This not only could limit the applicability of the OMS SO₂ product for monitoring anthropogenic emission sources, but also may lead to many negative retrievals of SO₂ column in clean regions. Furthermore, SO₂ retrievals at certain cross-track positions (assumed to be clean near-zero) may exceed the 2 DU threshold, requiring the threshold to be adjusted upwards to achieve a better effect after background offset correction. Therefore, although the current background offset correction strategy helps mitigate stripes, residual systematic errors may still remain after the correction, which could affect the accuracy of the OMS SO₂ column retrievals.

The quantification of the residual systematic errors is challenging, as it is difficult to fully separate random fitting errors from residual systematic components. Therefore, in this study, the residual systematic errors after the background offset correction were combined with the random fitting errors and treated as an effective SCD uncertainty, which was subsequently used in the total uncertainty estimation of the OMS SO₂ VCD.

In general, assuming that the error sources are independent of each other, the uncertainty of satellite-retrieved SO₂ VCD can be estimated using the following standard error propagation approach (Lee et al., 2009; Theys et al., 2017): 4σSO2=σSCDAMF2+σSCDbackAMF2+SCD-SCDback⋅σAMFAMF22 where σSCD is the random error from DOAS SO₂ SCD spectral fitting including instrument-related noise, σSCDback is the residual systematic error after background offset correction, and σAMF is the AMF uncertainty which includes two components: one is related to the atmospheric scattering weight and the other one is associated with the SO₂ profile shape.

In the standard error propagation method, σSCD is commonly approximated by the SFE. However, for the OMS SO₂ retrieval, the SFE is strongly affected by instrument-related systematic errors, as shown in Fig. 18. As a result, the SFE includes not only random fitting errors but also systematic components, and thus does not represent the random SCD uncertainty. If directly used as σSCD in the error propagation, it would lead to an overestimation of the total uncertainty of the OMS SO₂ VCD.

Therefore, the SFE was not adopted as σSCD in this study. Instead, the standard deviation of the corrected SCD over clean oceanic regions, approximately 0.2 DU in equatorial regions and about 0.4 DU at high latitudes, was used as an effective SCD uncertainty. This value reflects the combined effect of random fitting error and residual systematic errors after background offset correction and was propagated into the total uncertainty estimation. Accordingly, the total uncertainty of the OMS SO₂ VCD can be expressed as: 5σSO2=σSCD,resAMF2+SCDcorr⋅σAMFAMF22 where σSCD,res is the effective SCD uncertainty, and SCDcorr is the background-corrected SO₂ SCD.

The estimated total uncertainties of OMS SO₂ VCDs under four scenarios – clean, boundary layer pollution, volcanic degassing, and volcanic eruption – are summarized in Table 7. The results show that under clean conditions, the total uncertainty is dominated by measurement noise and fitting errors, ranging from 0.16–0.17 DU at low latitudes and 0.32–0.33 DU at high latitudes. The corresponding relative errors are ∼ 150 % at low latitudes and ∼ 295 % at high latitudes, reflecting the challenge of retrieving low SO₂ column. For boundary layer pollution scenarios, the enhanced SO₂ signal reduces the relative errors to 14 %–25 %, while the total uncertainty increases to 0.68–1.21 DU at low latitudes and 0.75–1.26 DU at high latitudes. For volcanic degassing, the relative errors range from 14 %–24 %, with total uncertainties ranging from 2.32 to 3.97 DU. Under the volcanic eruption scenario, the total uncertainty and relative error are approximately 60 DU and 50 %, respectively. It should be noted that these estimates still involve large uncertainties due to the large errors in radiance measurement and SO₂ vertical profile provided by GEOS-CF, particularly under volcanic eruption conditions. In addition, the uncertainties associated with clouds and aerosols were not included in the present error analysis, suggesting that the estimated uncertainties may be conservative.