Total column ozone comparisons
A time series of satellite and reanalysis data as well as individual measurements at the EMA in Quito (see “Methods” section) is shown in Fig. 1. The main features observable in the time series are consistent across all datasets, mainly in terms of the shape of the profiles and the time of year when the TCO peaks (mid-September, Fig. S1) . However, there are differences in the magnitude of the TCO with respect to ozonesondes and between datasets, which have been quantified and are discussed below. For reference, the spatial and temporal resolution of the spatial measurements, as well as the number of points compared to the ozonesondes, are summarized in Table 1.
The TROPOMI/S5P TCO overestimates the sounding measurements with an average positive bias of +8.8 DU, which corresponds to an average difference of 3.7% (Fig. 2a,b). Even if the spatial resolution of TROPOMI/S5P is far superior to GOME-2/MetOP-B, the latter works in a similar way since its average bias (Fig. 2c,d) compared to the soundings is +7.7 DU (difference of 3.2%). Similar amplitude shifts in measurements from these sensors have been documented at other tropical stations.17. On the other hand, the difference between OMI/Aura and the soundings (+ 2.7 DU or 1.2% mean difference; Fig. 2e,f), is one third of that encountered for TROPOMI/5SP and GOME-2/MetOP -B. Meanwhile, the TCO of OMPS/Suomi NPP turned out to be the best of all the products (Fig. 2g,h) because the average bias against the polls is -0.6 DU (-0.2 % mean difference). In all cases, the slope of the linear regression is 0.7 and the correlation coefficient (R2) is 0.6. From an observational point of view, the TCO of OMPS/Suomi NPP and OMI/Aura outperforms TROPOMI/S5P and GOME-2/MetOP-B in the tropical Andes. Previous work shows that OMI/Aura and OMPS/Suomi agree within 2% of measurements from the majority of other tropical stations1.
As presented, the product with the finest spatial resolution (TROPOMI/S5P) does not give the best comparison. This is counterintuitive as a finer resolution would seem more likely to resolve the significant elevational gradient at the study site. However, an important aspect to consider is the structure and magnitude of ozone in the troposphere, which has been identified as a factor that causes biases in TCO recoveries.18. Satellite algorithms use “a priori” information on ozone profiles to recover TCO from backscattered UV radiation measurements19. Due to the high longitudinal variability of TrCO, an improvement of the TROPOMI/S5P algorithm incorporates the climatology of Ziemke et al.20 allow a better adjustment in the troposphere11. However, ozone climatologies use site profiles mainly at sea level in the Atlantic and Pacific basins, while high altitude profiles in the tropical Andes deviate from these locations mainly in the troposphere. Thus, previous research demonstrates significant differences in profile structure and magnitude of TrCO (lower) from Galapagos (Ecuador) and Natal (Brazil) stations, while stratospheric ozone is similar and consistent with Microwave Limb Sounder measurements (MLS/Aura)12.13. Therefore, we compared the average differences of TCO, TrCO and stratospheric column ozone (SCO) between Ziemke’s climatology and the EMA observations compendium in Quito (Table S1). While the SCO is practically the same (+ 1.3 DU or 0.6% difference), the TCO from climatology shows an average bias of + 9.3 DU (3.8% average difference), which comes from an overestimation of ozone in the troposphere (+ 8 DU or 42.5%). This result is similar to the comparison of individual TROPOMI/S5P measurements with ozone sensors, discussed previously. Although further research is needed and more data are needed, our results highlight the need to incorporate profiles in the Andean tropics into satellite climatologies, in particular to better represent TrCO in the region, which would have a globally beneficial to TCO.
With respect to reanalysis products, MERRA-2 gives a closer comparison (Fig. 2i,j) to observations than CAMS (Fig. 2k,l), as the average bias is +4.5 DU versus +8 DU (less than 2% against mean difference of 3.3%). Data assimilation in MERRA-2 incorporates TCO from IMO/Aura and stratospheric mixing ratios from MLS/Aura21. Meanwhile, CAMS adds to the old ozone sources the TCO of GOME-2/MetOP-B22. From an observational perspective, MERRA-2 outperforms CAMS in estimating TCO, but further research should be conducted to better identify the specific differences between the two models and the data sources they use.
Finally, we quantified the differences of all products compared to OMPS/Suomi NPP as the dataset that best compares to in situ observations. As shown in Fig. S2, the average differences encountered for TROPOMI/S5P, GOME-2/MetOP-B and OMI/Aura are respectively 3.8%, 3.5% and 1.4%. Concerning the reanalysis, the average differences encountered for MERRA-2 and CAMS are respectively 1.1% and 3%. Therefore, OMI/Aura and MERRA-2 agree best with OMPS/Suomi NPP.
Ground-level ozone comparisons
A time series of TrCO for datasets that have daily observations is shown in Fig. 3, although at appropriate pressure levels for comparison. For example, TROPOMI/S5P is available from the surface (760 hPa) down to 270 hPa, which corresponds to about 10 km altitude. However, from high-resolution profiles, the tropical Andean troposphere is located at 96 hPa (17 ± 0.7 km) (see “Methods” section, Table S2). In addition, previous work demonstrates that the tropopause level identified with the chemical definition (“Methods” section) mainly coincides with the coldest point of the temperature profile.13. This indicates that, unlike the Pacific and Atlantic sites23.24, ozone in the Andean tropics is generally well mixed up to the level of the tropopause. While 270 hPa in TROPOMI/S5P is suitable for determining TrCO at mid-latitudes, the rest of the tropospheric column is missing at the study site. We have not come across a specific explanation in the literature as to why the off-the-shelf TrCO product is nominally set at 270 hPa for the entire globe.2.11. We believe this might be somewhat misleading, especially for the end user in the Andean tropics. Similarly, TrCO of GOME-2/MetOP-B is available up to 200 hPa, albeit as monthly averages (Fig. S3). In contrast, reanalysis products are available as mass mixing ratios at pressure level intervals from the surface throughout the atmospheric column. Thus, comparisons capturing the entire Andean tropical troposphere can be made by integrating data down to 100 hPa.
Comparing the TrCO of TROPOMI/S5P with ozone sensors (integrated up to 270 hPa) gives a positive bias of +3.6 DU or 32.5% mean difference (Fig. 4a,b). Previous research shows that tropospheric ozone in the tropical Andes is low because altitude deducts 5-7 AU from TrCO, while boundary layer ozone is also low13. However, TROPOMI/S5P measures generally higher values even if the observations correspond to a fraction of the column. In recent research that assessed the quality of TROPOMI/S5P TrCO against ozonesondes in the tropics, biases were also encountered. For example, the differences at several stations at sea level were found to be +4 DU (up to 22% greater) when data were averaged over 2 years, while the overall positive bias was +2, 3 DU (or 11%) when data was smoothed over longitudes2. The cause of this positive bias has not been fully understood and has been partially attributed to possible systematic differences in the timing of measurements, provided TrCO follows a diurnal pattern. We also report a positive bias in the Andean tropics which also requires further investigation. In part, this overestimation probably stems from satellite climatologies overestimating ozone in the Andean troposphere, as discussed in the previous section. However, further comparisons are needed with more data in the future to better understand the nature and persistence of these differences.
MERRA-2 TrCO compared to soundings down to 100 hPa gives the best comparison in the troposphere (Fig. 4c,d). The bias compared to ozone probes is +1.5 DU (difference of 11.5%), which we consider to be low given the difficulty in correctly capturing the TrCO on this very complex site where no validation is possible. has been done before. In the case of CAMS, the integration was also performed up to 100 hPa. The average bias (Fig. 4e,f) is +3.5 DU (23%), which doubles that of MERRA-2, but the correlation coefficient for the linear regression is higher (R2= 0.8 versus 0.6). On the other hand, the TrCO of GOME-2/MetOP-B is only available to the end user in the form of monthly averages. Thus, this product was only qualitatively compared to EMA in a time series (Fig. S3), but there are insufficient data to draw quantitative conclusions.
Finally, from an end-user perspective, the TrCO MERRA-2 is currently best suited for the Andean tropics. First, because data integration can be done down to a pressure level that captures the entire Andean tropospheric column. Second, because the differences from observations are the least.