The NASA OMI aerosol retrieval algorithms for cloud-free conditions (OMAERUV; Torres et al., 2007, 2013, 2018), and for above-cloud aerosols (OMACA; Torres et al., 2012; Jethva et al., 2018) have been combined into a single algorithm (TropOMAER) and applied to TROPOMI observations. TropOMAER uses observations at two near-UV wavelengths to calculate the UVAI and to retrieve total column aerosol optical depth (AOD) and single-scattering albedo (SSA). Although detailed documentation on the heritage algorithm is available in the published literature, a brief description is presented here for completeness.

In this work, we use the UV–Vis (UV–visible) band 3 of the TROPOMI level 1b product (Kleipool et al., 2018). TROPOMI version 1 reflectances for band 3 are within 5 %–10 % compared with OMI and OMPS (Rozemeijer and Kleipool, 2019a, b). It is expected that the upcoming version 2 of the TROPOMI level 1b product will solve inconsistencies of the radiometric calibration detected in the UV and UV–Vis spectrometers using in-flight measurements, and it will include degradation correction for the affected bands (Ludewig et al., 2020).

For this application, we use TROPOMI correction coefficients at 354 and 388 nm derived using an ice-reflectance-based vicarious approach that has been used to evaluate the calibration of UV–Vis sensors (Jaross and Warner, 2008).

TROPOMI-measured reflectances over Antarctica on 28 and 29 November 2017 were compared to radiative transfer model results. We calculate the ratio of each observed across-track ground pixel's reflectance at a specified wavelength to that of the modeled value for the same viewing conditions to obtain an error for that measurement. The model used is exactly the same as was used in the generation of OMI Collection 3 level 1b data (Dobber et al., 2008). The static corrections applied to TROPOMI reflectances elsewhere on the globe were derived by first averaging over all measurement errors at a given across-track position, then further smoothing with a five-pixel boxcar in the across-track direction. Corrections range from -4 % to +2 % in the across-track direction for the two wavelengths. We plan to repeat the calibration adjustments and to reprocess when an improved version 2 of the level 1b product is released by the ESA.

Two consecutive orbit views by OMI and TROPOMI of the smoke plume from the Pacific Northwest fires on 18 August 2018 are shown in Fig. 2. OMI's depiction of this event appears in Fig. 2a, whereas Fig. 2b illustrates the same aerosol feature as reported by the TropOMAER algorithm. Both products cover a similar range of UVAI values from a slightly negative background to values as high as 10. OMI's coarse spatial resolution, however, is in stark contrast to TROPOMI's fine resolution that allows for the mapping of the smoke plume UVAI signal with an unprecedented level of detail. Missing data in OMI's depiction in Fig. 2a is associated with the row anomaly that has reduced the sensor's observing capability by nearly 50 % since about 2008 (Torres et al., 2018; Schenkeveld et al., 2017). Figure 2c shows the operational TROPOMI ESA/KNMI UVAI product for the same event. The main difference between the NASA (Fig. 2b) and ESA/KNMI (Fig. 2c) UVAI products is the background values that, while near zero for the NASA product, reach values a low as -2 for the KNMI product. The large background difference between the two products is likely the combined effect of calibration uncertainties in the operational ESA/KNMI product and algorithmic differences in the treatment of clouds in the calculated component of the UVAI definition. In the KNMI UVAI calculation, clouds are modeled as opaque reflectors at the ground (Herman et al., 1997), whereas in the NASA UVAI, clouds are explicitly modeled as polydispersions of liquid water droplets using Mie theory (Torres et al., 2018). A comparative analysis of OMAERUV and TropOMAER UVAI is presented in Sect. 3.3.

We separately evaluate the effect of instrumental and algorithmic improvements in the TropOMAER retrieval algorithm by direct comparison of the satellite product to ground-based globally distributed (over land) level 2 version 3 measurements of AOD (Giles et al., 2019) by the Aerosol Robotic Network (AERONET; Holben et al., 1998).

Measurements of AOD at 380 nm are available at most AERONET sites, allowing for a direct comparison to OMI and TROPOMI 388 nm retrievals. No attempt was made to account for the small AERONET–TROPOMI wavelength difference. AERONET AOD measurements at the 12 sites listed in Table 1 over a 2-year period (May 2018 through May 2020) were used in the analysis. These locations were chosen based on the availability of 380 nm AOD measurements and on the representativity of environments where most common aerosol types (carbonaceous, desert dust, and sulfate-based) are observed.

The continuity of the OMI and TROPOMI records of aerosol properties is analyzed in this section. Monthly average values of AOD and AAOD for the May 2018 to May 2020 2-year period are calculated for three regions: the eastern United States (EUS) between 25–45∘ N and 60–90∘ W, southern Africa (SAF) bounded by 5–25∘ S and 15–35∘ E, and the Saharan desert (SAH) zone between 15–30∘ N and 30∘ E–10∘ W. The EUS region is representative of areas predominantly associated with non-absorbing aerosols and clouds. The SAF region is known as an important source area of carbonaceous aerosol–cloud mixtures, whereas the SAH region is the source area of the desert dust part, the most abundant aerosol type.

Figure 5 shows the 2-year AOD record produced by the OMAERUV (blue) and TropOMAER (red) algorithms for the three regions. TropOMAER-generated AOD values are consistently higher by about 0.2 than the OMAERUV record for the SAF and SAH regions where the absorbing aerosol load is typically large most of the year. The EUS region shows significantly smaller OMI–TROPOMI differences in monthly mean values. The comparison was also done using a TropOMAER version of the algorithm that uses the heritage cloud screening approach, yielding similar results.

Figure 6 depicts the 2-year record in terms of AAOD. Differences as large as 0.03 in the SAH region during the 2018 spring–summer months are significantly lower in the 2019 record. Overall, the AAOD time series over the three regions show closer agreement between the two sensors, suggesting a partial cancellation of retrieval errors in SSA and AOD when combined in the AAOD parameter.

Figure 7 shows global 3-month (June, July, August 2018) average maps of AAOD from TROPOMI (top) and OMI (bottom) observations. Seasonally occurring features, such as the Saharan desert dust signal over northern Africa and the smoke plumes associated with biomass burning over Namibia, Angola, and Congo, are clearly picked by both sensors with comparable AAOD values. Other continental aerosol features, such as a dust and smoke signal over the western US and smoke plumes from wildfires in the Northwest Pacific moving eastward across Canada, are detected at similar AAOD values by the two sensors, albeit with a higher level of detail in the TROPOMI product. Similar aerosol signals are also picked up by the two sensors over Saudi Arabia, northwest India, Pakistan, and western China. Perhaps the most striking continental difference in the seasonal map in Fig. 7 is the much larger OMI background AAOD in South America, possibly linked to the difficulty of removing sub-pixel cloud effects at OMI's resolution.

Surprisingly, OMI only shows a very scattered signal of the North Atlantic Saharan dust plume between northern Africa and the plume's leading edge north of Venezuela over the Caribbean, whereas the TROPOMI product shows an almost continuous North Atlantic plume. In spite of the geographically sparse nature of the OMI AAOD data, there is high consistency in the retrieved values by the two sensors. A similar but less severe difference is also observed over the South Atlantic, where the OMI-retrieved carbonaceous aerosol plume is more disperse than what is shown in the TROPOMI map. The combined effect of prevailing sub-pixel cloud contamination and OMI's row anomaly explains the spatially scattered OMI retrievals over the oceans.

Clearly, the full TROPOMI coverage at much higher spatial resolution than OMI and the high-resolution VIIRS cloud mask contribute to significantly improving the near-UV aerosol product.

The OMI and TROPOMI gridded 2018 monthly data used to produce the seasonal average maps discussed above are also displayed in Fig. 8 as density AAOD (left) and UVAI (right) plots. Although small offsets in UVAI (∼ 0.2) and AAOD (∼ 0.02) between the sensors are apparent, a high degree of correlation between the observations by the two instruments is clearly observed.

The 2018 fire season in the western USA and Canadian British Columbia territory was one of the most active of the last few years. It is estimated that over 8500 fires were responsible for the burning of over 0.8 million ha, which is the largest burned area ever recorded according to the California Department of Forestry and Fire Protection (http://fire.ca.gov, last access: 2 December 2020) and the National Interagency Fire Center (http://nfic.gov, last access: 2 December 2020). From mid-July to August, intense fires in northern California, including the destructive Carr and Mendocino Complex fires, produced elevated smoke layers that drifted to the east and northeast. In 2018, the British Columbia (BC) province of Canada encountered its worst fire season on record, surpassing the 2017 record with more than 2000 wildfires and 1.55 million ha burned and accounting for about 60 % of the total burned area in Canada in 2018 (https://www2.gov.bc.ca/gov/content/safety/wildfire-status, last access: 2 December 2020). Figure 9 shows the spatial extent of the smoke plume generated by wildfires in Canadian BC and the northwestern USA on 18 August in terms of the 388 nm AOD and SSA products from both TROPOMI (top) and OMI (bottom) observations (the corresponding UVAI depiction is shown in Fig. 2). Observed gaps in the core of the plume are due to out-of-bounds retrieval conditions. The carbonaceous aerosol layers produced by the fires spread over a huge area, covering large regions of the US Midwest and central Canada. The height of the aerosol layer varies between 3 and 5 km according to CALIOP observations (not shown). Although OMI's coarse resolution and row-anomaly-related reduced spatial coverage are clearly observable, the retrieved AOD and SSA fields by the two sensors look remarkably similar. TROPOMI and OMI AOD retrievals reach values as high as 5.0 near the sources, generally consistent with AERONET ground-based observations that, on this day, reported AOD values as large as 1.5 (412 nm) at the Lake Erie site (41.8∘ N, 83.2∘ W) and values in excess of 3.0 at the Toronto station (43.8∘ N, 79.5∘ W). SSA values in the range 0.85–0.92 are retrieved by both sensors over the extended area. Minimum OMI-retrieved SSA (0.85) in the vicinity of a source area, however, is lower by about 0.02 than the corresponding TROPOMI measurement, consistent with the relative OMI–TROPOMI SSA differences reported in Fig. 4.

Figure 10 shows the spatial distribution of the September 2019 average TROPOMI UVAI, AOD, and AAOD over the region between the Equator and 40∘ S and between 35∘ W and 85∘ W. Monthly average AOD values of around 2.0 prevailed over the source areas. The smoke plumes were mobilized downwind towards southern Brazil, reaching highly populated areas where TROPOMI-measured monthly average AOD in the range 0.9 to 1.0 are reported.

Figure 11 shows the time series of monthly average OMI 388 nm AOD over the region for the last 15 years, along with the overlapping TROPOMI AOD observations over the last 2 years, illustrating the importance of the continuity of the long-term record. Although, as discussed earlier, there are small differences in the time series between the two sensors, these differences are not large enough to question the ability to differentiate years with large seasonal events from years with comparably reduced biomass burning activity. The seasonal carbonaceous aerosol concentration over the Amazon Basin, associated with intense agriculture-related biomass burning, has significantly decreased over the last 12 years since 2008. The OMI record shows a remarkable decrease since 2008 when near-record-high values were observed (Torres et al., 2010). After consecutive AOD September peaks larger than 2.0, in the 3-year 2005–2007 period, the monthly average AOD over the Amazon Basin was reduced to values of about 0.5. An isolated abrupt increase to larger than 2.0 was again observed in 2010. Since then, the September peak AOD value has remained much lower than 1, except for 2017 and 2019 when September average AOD larger than unity was observed. The 2019 peak AOD value (1.25) was also retrieved by TROPOMI observations. Although the overall regional average was slightly larger than in the previous year, it was about a third of the 2010 peak value. As a result of the prevailing regional atmospheric dynamics in 2019, carbonaceous aerosols generated by seasonal biomass burning over regions up north were transported towards the southeast, reaching large urban centers such as São Paulo and Curitiba, generating a lot of media attention (Hughes, 2019).

The 2019–2020 fire season in Australia resulted in 18.6 million burned hectares, most of them in the New South Wales and Victoria southeastern states (SBS News, 2020). It is estimated that tens of people died along with billions of animals that were exterminated, including species that were near extinction before the fire (Readfearn, 2020). The intense fire activity likely triggered a number of pyroCb clouds over a few days between 30 December 2019 and early January 2020, injecting large quantities of carbonaceous aerosols into the Southern Hemisphere UTLS (Ohneiser et al., 2020). In this section, we describe TROPOMI observations of these events in terms of UVAI and AOD retrievals. As observed in visible satellite imagery (not shown) most of the UTLS-injected carbonaceous aerosol material was initially above clouds. TROPOMI near-UV observations were used in conjunction with aerosol layer height from CALIOP observations as input to a modified version of the TROPOMI aerosol algorithm that handles stratospheric aerosol layers (TropOMAER-UTLS). The retrieved SSA over clear scenes was then used as input in the retrieval of AOD over cloudy pixels by the above-cloud aerosol module described in Sect. 2.1.3.

TROPOMI-retrieved AOD was used to produce an estimate of resulting stratospheric aerosol mass (SAM). The SAM calculation procedure involves the separation of the stratospheric AOD component from the total AOD column measurement and the use of an extinction-to-mass conversion approximation described in Appendix B. This approach was previously applied to EPIC near-UV AOD retrievals to calculate the SAM associated with the 2017 British Columbia pyroCb events (Torres et al., 2020).

The identification of stratospheric aerosols is carried out by establishing a theoretical relationship between AOD and UVAI for a hypothetical aerosol layer at the tropopause for assumed values of ALH and AAE (see the discussion in Sect. 2.1). CALIOP-provided ALH information and an assumed AAE value of 4.8, similar to that in Torres et al. (2020), were used as input to TropOMAER-UTLS. AOD retrievals associated with UVAI values larger than those indicated by the AOD–UVAI relationship at the tropopause height are assumed to correspond to stratospheric aerosols. Figure 12 shows TROPOMI-observed UVAI (y axis) and retrieved AOD (x axis) for CALIOP-reported ALH on 31 December 2019. Data points in red indicate retrievals above the estimated tropopause height (12 km), while the blue points show retrievals at heights below that level. The altitudes of the retrievals in relation to the tropopause are determined based on a unique viewing-geometry-dependent UVAI–AOD relation for each pixel, which is difficult to visualize on a single plot. Therefore, a quadratic fit (black line) to all data, i.e., above and below the tropopause, was derived to illustrate, for visualization purposes, the separation of tropospheric and stratospheric aerosols.

Unlike during the 2017 British Columbia fire episodes, when a large fraction of the pyroCb-generated aerosol plume initially remained in the troposphere and some of it diabatically ascended to the stratosphere over the next few days (Torres et al., 2020), during the Australian 2020 pyro-convective fires most of the produced carbonaceous aerosols appear to have gone directly into the stratosphere. Figure 13 shows TROPOMI-retrieved UVAI and AOD fields (total column and stratospheric component) on 2 January 2020. Only small differences in the total column and above-tropopause AOD fields are observed, as most of the aerosol material was directly deposited in the stratosphere.

Stratospheric AOD values were converted to mass estimates using the procedure described in Torres et al. (2020) and also included as Appendix B of this paper. For mass estimation purposes, TropOMAER 388 nm AOD data were gridded to 0.25∘×0.25∘ lat.–long. resolution. Figure 14 shows calculated daily SAM values (in kilotons) from 31 December 2019 through 7 January 2020 resulting from aerosols above 12 km, the altitude used as a proxy for the tropopause height. Separate aerosol mass retrievals were carried out for cloud-free (blue bars) and cloudy scenes (green bars), with the daily total SAM given as the sum of these two components (orange bars). The observed daily monotonic increase from 119 kt on 31 December 2019 to 380 kt on 2 January 2020 is likely the result of distinct pyroCb events that seemingly injected most of the aerosol mass directly in the stratosphere. Following the 2 January maximum, SAM decreases over the following 3 d to a minimum of 87 kt on 5 January, likely due to the combined effect of dilution processes that spread the aerosol layer horizontally and thin it out to extremely low AOD values beyond the sensor's sensitivity to the total AOD column measurement and aerosol deposition bringing it down to lower than 12 km and therefore no longer included in the SAM calculation.

The sudden increase to 166 kt on 6 January is likely associated with another pyroCb event observed on 4 January that injected an additional 166 kt. Thus, the TROPOMI-based total SAM estimate is the sum of the two peaks on 2 and 6 January, yielding a total of 546 kt, which about twice as much as the 268 kt estimated SAM for the 2017 British Columbia pyroCb (Torres et al., 2020) using the same mass estimation technique. The uncertainty of the estimated SAM is ±40 %, which represents the combined effect of uncertainties on assumed AAE (±0.5) in the AOD retrieval and the uncertainty associated with the assumed aerosol density range of 0.79 to 1.53 g cm-3 (Reid et al., 2005).