A Global Climatology of Tropopause Folds in CAMS and MERRA‐2 Reanalyses

Tropopause folds are the main mechanism underlying stratosphere‐to‐troposphere transport and influence tropospheric composition and weather systems by triggering convection. Here, we present the global climatology of tropopause folds in Copernicus Atmosphere Monitoring Service (CAMS) and Modern‐Era Retrospective analysis for Research and Applications, version 2 (MERRA‐2) reanalysis of atmospheric composition products for the time period from 2003 to 2018. We applied a 3‐D labeling algorithm in CAMS and MERRA‐2 reanalysis data to detect tropopause folding events. In constructing their climatologies, we show that the bulk of the folds are vertically shallow and are mainly found at the subtropical zones in the vicinity of the jet streams, while deeper folds also occur over the storm tracks, consistent with previous studies. The spatiotemporal characteristics of fold climatology are captured in a similar manner in CAMS and MERRA‐2, with MERRA‐2 capturing slightly higher frequencies during all seasons. In quantitative terms, there is a good agreement between CAMS and MERRA‐2 fold frequencies with spatiotemporal R2 values of ∼0.9 for DJF, MAM, and JJA, and 0.75 for SON. The two reanalysis products are in close agreement regarding the intra‐ and interannual variability in fold frequency, with temporal correlation scores higher than 0.7 over the subtropical bands where the majority of folds are found. The agreement between the two reanalyses is lower in the Southern Hemisphere compared to the Northern Hemisphere. Thus, the global climatology of tropopause folds in both CAMS and MERRA‐2 reanalyses are similar to those of previous studies.

The environmental significance of tropopause folds can be largely explained by two phenomena: (a) they are the main STT mechanism influencing tropospheric ozone budget, and (b) they are linked with surface weather systems, potentially leading to severe weather events. Tropospheric ozone is critical for the composition of the troposphere, climate, and air quality, as it regulates the oxidation capacity of the troposphere ; in terms of climate change, it is the third most important anthropogenic greenhouse gas (Myhre et al., 2013), and near the Earth's surface excessive ozone exposure is harmful to human health and the ecosystems (Monks et al., 2015). Although the main source of ozone in the troposphere is photochemical production, the downward transport of ozone from the stratosphere is also a significant contributor (Archibald et al., 2021;Stohl et al., 2003), especially for regions where the meteorological conditions favor subsidence and the formation of tropopause folds, such as during the summertime in the eastern Mediterranean and Middle East (Akritidis et al., 2016;Zanis et al., 2014), Afghanistan (Ojha et al., 2017;Tyrlis et al., 2014), and during springtime in the western United States (Langford et al., 2009;Lin et al., 2012Lin et al., , 2015. Tropopause folds are largely responsible for mediating the STT of ozone, making them a key factor influencing tropospheric ozone levels and variability. When deep folds reach into the lower troposphere, ozone concentrations may be significantly increased in both high (Cristofanelli et al., 2010;Lefohn et al., 2011Lefohn et al., , 2014 and low (Akritidis et al., 2010;Gerasopoulos et al., 2006) altitude sites, which occasionally results in violations of air quality regulations (Kaldunski et al., 2017;Langford et al, 2009Langford et al, , 2015Yates et al., 2013;Zhang et al., 2014). In a global study applying a Lagrangian methodology on ERA-Interim reanalysis, it was shown that near-surface ozone concentrations along the west coast of North America and around the Tibetan Plateau are likely to be markedly influenced by deep folds; particularly pronounced positive trends in the net downward mass flux were revealed for the period 1979-2011 over North America (Škerlak et al., 2014). Recently, Akritidis et al. (2019) stressed the role of tropopause folding in STT processes under a changing climate, suggesting that tropopause folds will be associated with both future increases and interannual variability in ozone STT.
Since folding events are characterized by positive PV anomalies, they affect tropospheric dynamics and may influence surface weather systems. The stratospheric reservoir that descends into the troposphere can impose positive PV advection and therefore induce or enhance the cyclonic circulation in the lower troposphere (Hoskins et al., 1985;Uccellini, 1990;Wernli et al., 2002). Furthermore, tropopause folds are known to promote or suppress convective storms, affecting the location, intensity, and morphology of the convection (Antonescu et al., 2013;Cooper et al., 2005;Homeyer et al., 2011;Waugh & Funatsu, 2003). In addition, tropopause folds can give rise to extreme surface winds through downward transport of momentum (Browning & Reynolds, 1994;Škerlak et al., 2015).
Despite the great importance of tropopause folds for tropospheric ozone and weather, the number of climatological studies on folds is rather limited. Ozonesonde (Beekmann et al., 1997;Van Haver et al., 1996) and radar (Antonescu et al., 2013) data were employed to detect foldings of the tropopause and to construct the climatology of the folds, but the analysis was restricted to specific sites. The first attempts to create global tropopause fold climatologies used meteorological analysis data and identified folds as regions of PV and Q-vector divergence maxima (Ebel et al., 1996;Elbern et al., 1998), while other studies were based on Lagrangian trajectories (James et al., 2003;Stohl, 2001). With increasing resolution in global atmospheric models, a new technique for fold detection was developed by Sprenger et al. (2003); this method is based on structural features and was used by the authors to construct a one-year global climatology of tropopause folding events. More recently, Škerlak et al. (2015) built on the ideas behind the 3-D labeling algorithm of Sprenger et al. (2003); the authors applied this algorithm to the ERA-Interim reanalysis data set for the time period 1979-2012 and were able to construct a more robust global climatology of folds. As well as climatologies of tropopause folds, there are also studies describing the climatologies of STT ozone flux based on reanalysis products (Jaeglé et al., 2017;Škerlak et al., 2014).
In recent years, several atmospheric composition reanalysis data sets were produced through the synergistic use of global atmospheric chemistry models, observations (ground-based, satellite, and aircraft), and assimilation techniques. The two most recent state-of-the-art atmospheric composition products are a) the Copernicus Atmosphere Monitoring Service (CAMS) reanalysis (CAMSRA; Inness et al., 2019) produced by the CAMS (http://atmosphere.copernicus.eu), which is operated by the European Center for Medium-Range Weather Forecasts (ECMWF, https://www.ecmwf.int/) on behalf of the European Commission, and b) the Modern-Era Retrospective analysis for Research and Applications, version 2 (MERRA-2) reanalysis (Gelaro et al., 2017) provided by the NASA Global Modeling and Assimilation Office (GMAO, https://gmao.gsfc. nasa.gov/). As state-of-the-art reanalysis data sets that provide both meteorological and atmospheric composition fields, CAMSRA and MERRA-2 might constitute a framework in which to study STT processes and their implications for tropospheric ozone and air quality. Several studies investigated STT processes using atmospheric composition reanalysis products, such as the Monitoring Atmospheric Composition and Climate (MACC) reanalysis (Knowland, Doherty, et al., 2017;Knowland et al., 2015) and the MERRA-2 reanalysis (Ott et al., 2016;Ryoo et al., 2017) over the last years. Nevertheless, to date, there is no report in the literature documenting a global climatology of tropopause fold occurrence in an atmospheric composition reanalysis product.
Here we present a 16-year (2003-2018) global climatology of tropopause folding events in CAMSRA and MERRA-2 reanalysis products. We used the latest version of the 3-D labeling and fold detection algorithm by Škerlak et al. (2015), with the aim of assessing the level of agreement both between the two reanalysis products as well as with previous studies. Section 2 provides information about the CAMSRA and MER-RA-2 reanalysis data and describes the 3-D labeling and fold detection algorithm. Section 3 presents the fold climatology in the two reanalysis products and the comparison between them, and finally, Section 4 summarizes the key conclusions.

CAMS Reanalysis
CAMSRA (Inness et al., 2019) is the latest reanalysis product of atmospheric composition produced by ECMWF, including 3-dimensional fields of meteorology, chemical species, and aerosols for the time period from 2003 onwards; CAMSRA is a follow-up of earlier reanalysis products: the MACC reanalysis (Inness et al., 2013) and the CAMS interim reanalysis (Flemming et al., 2017). CAMSRA is based on the ECMWF's Integrated Forecast System (IFS) CY42R1 release and the 4D-Var data assimilation system with two 12-h (09:00-21:00 UTC and 21:00-09:00 UTC) assimilation windows. It is based on the minimization of a penalty function that takes the deviations of the model's background fields from the observations to provide the optimal forecast during the assimilation window by adapting accordingly the initial conditions. The IFS incorporates meteorological observations, including satellite, in situ, PILOT (wind report from pilot balloon), radiosonde, dropsonde, and aircraft measurements. In addition, satellite retrievals of total column CO, tropospheric column NO 2 , aerosol optical depth, and total column, partial column and profile ozone retrievals are also assimilated in the IFS. More information on the satellite/instruments and the obtained satellite retrievals assimilated into CAMSRA are provided in Table 2 of Inness et al. (2019). The chemical mechanism used in the IFS is an extended version of the Carbon Bond 2005 (CB05) chemical mechanism (Flemming et al., 2015). The CAMSRA data have a spatial resolution of approximately 80 km (0.7° × 0.7° grid) with 60 hybrid sigma-pressure (model) levels up to 0.1 hPa, and a temporal resolution of 3 h.

MERRA-2 Reanalysis
MERRA-2 is the latest atmospheric reanalysis product (Gelaro et al., 2017) provided by the NASA GMAO, covering the time period from 1980 onwards. It represents an update of the modeling and data assimilation system of the original MERRA data set (Rienecker et al., 2011). MERRA-2 was produced using the Goddard Earth Observing System, Version 5 (GEOS-5) atmospheric model (Molod et al., 2015) and the Gridpoint Statistical Interpolation (GSI) assimilation system (Kleist et al., 2009). Specifically, GSI applies a 3D-VAR algorithm (W. Wu et al., 2002) with 6 h windows, producing the analyses through a process of incremental analysis update (Bloom et al., 1996). GEOS-5 integrates a radiatively coupled version of the Goddard Chemistry Aerosol Radiation and Transport (GOCART) model (Chin et al., 2002) to simulate aerosol components. Along with meteorological observations (see Table 1 of Gelaro et al. (2017) for further details), the model also assimilates ozone partial and total column, stratospheric ozone profiles, and aerosol optical depth at 550 nm. More information on the observational system used in MERRA-2 for ozone and aerosol optical depth assimilation are provided in Table 1 of Wargan et al. (2017) and Table 2 of Randles et al. (2017), respectively. The horizontal resolution is 0.5° × 0.625° latitude by longitude, with 72 hybrid sigma-pressure model layers up to 0.01 hPa.

Detection of Tropopause Folds
We adopted the latest version of the 3-D labeling algorithm of Škerlak et al. (2015) to identify tropopause folds in CAMSRA and MERRA-2 data, similarly to Akritidis et al. (2016Akritidis et al. ( , 2019. The initial inputs for the algorithm are the 3-D fields of PV, potential temperature, and specific humidity and the 2-D fields of surface pressure. Subsequently, the algorithm constructs the 3-D fields of pressure and determines the pressure level of the dynamical tropopause (Holton et al., 1995;Sprenger et al., 2003;Stohl et al., 2003) using the lower of the isosurfaces of PV at 2 PVU and potential temperature at 380 K. A vertical profile is then taken for each grid point and a fold is assigned when multiple crossings of the tropopause are identified. Yet, there are specific situations where air with PV > 2 PVU is either not connected to the stratosphere (stratospheric cut-offs) or is not of stratospheric origin (diabatic PV anomalies or surface-bound PV anomalies), which should not be considered as stratospheric. To this end, the algorithm performs a 3-D labeling (from 1 to 5) of air masses as follows: tropospheric = 1, stratospheric = 2, stratospheric cut-off or diabatically produced PV anomaly = 3, tropospheric cut-off = 4 and surface-bound PV anomaly = 5. A more detailed description of the criteria used for the 3-D labeling can be found in Škerlak et al. (2015). Therefore, a fold is detected when a 2→1→2→1 or 3 transition is found on a vertical profile from the upper to the lower model level, and the algorithm produces as outputs a binary variable (0:no fold, 1:fold) for every grid point and time step. In addition, the upper (p u ), middle (p m ), and lower (p l ) pressure levels of the tropopause crossings are identified along with the difference Δp = p m -p u , which reveals the vertical extent of the fold. According to the Δp values, the detected folding events are divided into the three following categories: -shallow folds, 50 ≤ Δp < 200 hPa -medium folds, 200 ≤ Δp < 350 hPa -deep folds, Δp ≥ 350 hPa As PV and potential temperature are not available in model levels for the CAMSRA product, the "pot_vort_ hybrid" NCL (NCAR Command Language) function (https://www.ncl.ucar.edu/Document/Functions/ Contributed/pot_vort_hybrid.shtml) is applied to calculate them, using as input the 3-D fields of pressure, u-v wind components, and temperature for every time step. For consistency, PV and potential temperature in MERRA-2 are obtained with the same method as in CAMSRA. The 3-D labeling algorithm is implemented in CAMSRA and MERRA-2 data for the period from 2003 to 2018 with a time interval of 3 h. The average of the binary output variable (0:no fold, 1:fold) for a grid point over a time period (month, season, year, etc.) multiplied by 100 reflects the percentage of time steps that exhibit a fold over the examined period and is hereafter called "tropopause fold frequency." A snapshot of PV pressure-longitude cross section at ∼38°N for CAMSRA and MERRA-2 at 00:00 UTC on October 9, 2003 is presented in Figure 1, depicting a folding of the dynamical tropopause in both reanalysis products down to ∼600 hPa, accompanied by the sloping isentropes that denote downward air transport into the troposphere. The vertical dashed black line that crosses the dynamical tropopause at three points depicts the detected fold both in CAMSRA ( Figure 1a) and MERRA-2 ( Figure 1b).

Climatology of Tropopause Folds Frequency
We first examined the global spatial distribution of shallow, medium, and deep tropopause fold frequency (%) in CAMSRA and MERRA-2 for DJF (December, January, and February), MAM (March, April, and May), JJA (June, July, and August), and SON (September, October, and November). Figure 2 presents the mean seasonal shallow tropopause fold frequency along with the horizontal wind speed (≥20 m/s) at 250 hPa for CAMSRA (left panel) and MERRA-2 (right panel). During DJF and MAM, and based on the seasonal behavior of the subtropical jet stream, the vast majority of shallow folds in both CAMSRA (Figure 2a and 2c) and MERRA-2 (Figure 2b and 2d) are found in the NH mainly in the vicinity of high wind speed regions. In both reanalyses, the highest shallow fold frequencies during DJF and MAM are seen over eastern Asia and further west over Northern India, respectively, with regional values exceeding 15%. During JJA, and as the subtropical jet stream strengthens in the SH, a band of high fold frequencies up to 15% is found over the southern Indian Ocean and Australia, while in the NH a hotspot of shallow fold activity is identified over the eastern Mediterranean and the Middle East as a result of the complex interaction between the subtropical jet stream and the south Asian Monsoon . Overall, the geographical distribution of shallow fold frequency in the two reanalysis products is similar, reproducing the spatiotemporal features globally. Nevertheless, shallow folds occur somewhat more frequently in MERRA-2 compared to CAMSRA during all seasons.
AKRITIDIS ET AL.  The mean seasonal frequency of medium folds for CAMSRA and MERRA-2 is presented in Figure 3, revealing that medium folds are rarer than shallow ones as they occur with frequencies of approximately one order of magnitude lower. More specifically, medium folds mainly co-occur with shallow folds in the proximity of high wind speed regions and over the eastern Mediterranean and the Middle East, yet with lower frequencies that are up to 1%. In addition, medium folds are also found along the storm tracks in both the NH and SH, namely the North Atlantic (mainly appearing during DJF and SON), the North Pacific (during DJF, MAM, and SON), and the Southern Ocean (during MAM, JJA, and SON). Although there is a qualitative agreement between the frequencies of medium folds captured by CAMSRA and MERRA-2, in quantitative terms MERRA-2 exhibits up to ∼0.5% higher frequencies over almost all aforementioned regions, which on average corresponds to approximately one extra medium fold event per month. AKRITIDIS ET AL.     The mean annual cycle of zonal mean tropopause fold (folds with Δp ≥ 50 hPa namely shallow, medium, and deep) frequency is presented in Figure 5 for CAMSRA ( Figure 5a) and MERRA-2 (Figure 5b). In the NH, both reanalyses capture the highest fold frequencies during the period between January and April and within the latitudinal band of 20°-40°N where the subtropical jet stream occurs, with frequencies in MERRA-2 ∼1% higher compared to those in CAMSRA. From June to August and as the NH subtropical jet stream weakens and shifts further north, the highest zonal mean fold frequencies are seen at ∼40°N, which is due to the high summertime fold frequencies over the eastern Mediterranean and Middle East. In the SH, AKRITIDIS ET AL.  and following the strengthening of the subtropical jet stream during the austral winter, folds occur more often during the period from May to October at ∼−30°N, a feature that is more pronounced in MERRA-2.

Level of Agreement Between CAMSRA and MERRA-2 Tropopause Fold Climatologies
We quantitatively assessed the agreement in spatiotemporal variability in tropopause fold climatologies in CAMSRA and MERRA-2. Due to the different horizontal resolutions in CAMSRA and MERRA-2, the MERRA-2 fold frequencies were bilinearly interpolated to the coarser CAMSRA grid. Figures 6a-6d present the scatter plots of tropopause fold (folds with Δp ≥ 50 hPa) frequency for all grid points (512 × 256) and for each season (16 seasonal means for each season and grid point) in CAMSRA and MERRA-2. Furthermore, Figure 6e shows the scatter plot of tropopause fold frequency for all grid points (512 × 256) and months (12 × 16 = 192 monthly means for each grid point). Also shown is the distribution of fold frequencies in the two reanalysis products in every case. The explained variance in Figures 6a-6d depicts the spatial and the interannual covariance between CAMSRA and MERRA-2 tropopause fold frequency for each hemisphere and season. In addition, the explained variance in Figure 6e includes the intra-annual covariance (from month to month within a year) between CAMSRA and MERRA-2 tropopause fold frequency.
As depicted for DJF (Figure 6a), there is a very good agreement between CAMSRA and MERRA-2 fold frequencies on a global scale (blue and orange points) with R 2 (proportion of MERRA-2 variance explained by CAMSRA) and mean absolute difference (MAD) values of 0.88% and 0.63%, respectively; the match is better in the NH (blues points) (R 2 = 0.91) than in the SH (orange points) (R 2 = 0.78). A similar behavior on a global scale is seen for MAM (Figure 6b), with R 2 = 0.88 and MAD = 0.59% and a better agreement in the NH (R 2 = 0.92) with regards to the SH (R 2 = 0.75). During JJA (Figure 6c), and as the fold activity shifts also in the SH, CAMSRA and MERRA-2 match each other closely, with R 2 = 0.88 and MAD = 0.58%; the closest agreement is found in the NH (R 2 = 0.93 compared to R 2 = 0.85 in the SH). The weakest agreement among all seasons is identified for SON (Figure 6d) with R 2 = 0.75 and MAD = 0.67%, but again with a better match in the NH (R 2 = 0.82 compared to R 2 = 0.7 in the SH). Overall, and when considering all monthly values over the examined period (Figure 6e), the above pattern remains the same with R 2 = 0.8 and MAD = 0.72%, while in the NH and SH the respective R 2 values are 0.86 and 0.69. Considering the embedded figures of fold frequency distributions, there is a clear preponderance of folds in the NH for the whole year, DJF, and MAM, which is reproduced by both reanalyses. During JJA, there are more instances of fold frequencies below ∼10% in the SH, while there are more above ∼15% in the NH, a feature in both CAMSRA and MER-RA-2. For SON, the distribution of fold frequencies in the two hemispheres is more similar than in the other seasons. Overall, the distributions reveal a good agreement between MERRA-2 and CAMSRA, with MER-RA-2 exhibiting slightly higher fold frequencies, an aspect which is also mentioned above in Section 3.1. Moreover, there is a better agreement between CAMSRA and MERRA-2 in the NH than in the SH for each season and when considering all months, with a higher R 2 and a lower MAD (except DJF). This may be related with the fact that more observations are available for assimilation in the NH compared to the SH.
In order to compare tropopause fold climatology in the two reanalysis products in more detail, we explored the interannual variability and calculated the Pearson correlation coefficients between the CAMSRA and MERRA-2 fold frequency for every grid point from both seasonal (16) and monthly (12 × 16 = 192) values of the period 2003-2018. Figure 7 presents the spatial distribution of these correlations, with all colors except black depicting significant correlation coefficients at the 95% significance level (following t-statistic) and the gray color denoting undefined values (cases where at least one of the series consists of zero frequencies). Significant positive correlations predominate over all seasons across the globe, and values exceed 0.7 over the subtropical zones where the main fold activity is identified; over the regions with lower fold activity, for example, the storm tracks, meanwhile, the correlation is lower but still significant and higher than 0.5 (Figures 7a-7d). This indicates that, overall, there is a good agreement between the year-to-year variability for the seasonal fold frequency between CAMSRA and MERRA-2. In addition to Figures 7a-7d, Figure 7e shows both inter-and intra-annual covariance between MERRA-2 and CAMSRA fold frequency for each grid point. The correlation coefficients of monthly fold frequency between CAMSRA and MERRA-2 show significant correlations higher than 0.8 over the subtropical regions; these values decrease with latitude and are higher over the NH with values exceeding 0.5 over the mid-latitudes (Figure 7e that, in both reanalysis products, the combined intra-and interannual variability of fold frequency is reproduced in a similar way, at least for the regions where the majority of tropopause fold events occur. Any discrepancies between the two climatologies are due to the different modeling and assimilation techniques applied for the production of CAMSRA and MERRA-2 reanalyses, as well as the differences between them in horizontal and vertical resolution. For example, for the western United States region which is a hotspot of medium and deep folds (see Figures 3 and 4) that are known to affect tropospheric ozone and air quality during MAM Lin et al., 2012Lin et al., , 2015, MERRA-2 captures more folds compared to CAMSRA. Nevertheless, when we consider fold events (Δp ≥ 200 hPa) over a specific grid point of this region that are detected in MERRA-2 and not in CAMSRA, a quite similar increase (decrease) of ozone (specific humidity) is found in the troposphere and near the folds for both reanalysis products, AKRITIDIS ET AL.  indicating that the STT processes are also reproduced by CAMSRA (see Figures S1-S3 in supporting information). Presumably, the coarser vertical and horizontal resolution in CAMSRA compared to MERRA-2 is responsible for the non-detection of these folds in CAMSRA by the applied algorithm.

Conclusions
We presented the climatology of tropopause folds in CAMS and MERRA-2 state-of-the-art reanalysis products for the time period 2003-2018 using a 3-D labeling algorithm for the detection of folding events. In particular, we examined the spatiotemporal characteristics of tropopause fold occurrence around the globe in both reanalyses and assessed the degree of agreement between the two products and with previous climatological studies. The most notable findings of the current study are summarized as follows: • The bulk of tropopause folds are vertically shallow (50 ≤ Δp < 200 hPa) and occur mainly in the vicinity of the NH/SH subtropical jet streams, while deeper folds, which are rarer, also develop over the storm track regions. During DJF and MAM, the majority of folds are seen in the NH, while during JJA and SON, most folds are observed in the SH. These features are seen in both CAMS and MERRA-2 reanalyses, which is in line with previous climatological studies such as the ERA-Interim-based approach of Škerlak et al. (2015). • The two reanalysis climatologies perform similarly well in capturing the global spatial distribution of tropopause fold frequency throughout the seasons, with MERRA-2 capturing an overall slightly higher frequency of occurrence. In quantitative terms, the comparison between CAMSRA and MERRA-2 indicates a good agreement for DJF, MAM, and JJA (R 2 = 0.88 and MAD ≈ 0.6%), while during SON the agreement is slightly weaker but remains strong (R 2 = 0.75 and MAD = 0.67%). • A lower agreement between CAMSRA and MERRA-2 is found for the SH compared to the NH, with R 2 values decreasing in all seasons. This probably reflects the lower amount of observational data available for assimilation in the SH. • Finally, the interannual variability in seasonal fold frequency is similar between the two reanalyses, with significantly positive correlations; over the subtropical bands where the majority of folds occur, these coefficients have values higher than ∼0.7.
In summary, this study indicates that both CAMS and MERRA-2 reanalysis products reproduce the key characteristics of tropopause fold occurrence reported by previous studies. Despite quantitative discrepancies, overall the two climatologies are highly similar. Tropopause folds are a key process that determine the influence of STT on tropospheric composition. Therefore, recording them and reporting their reliability in CAMS and MERRA-2 atmospheric composition reanalysis products is expected to be critical for future research studies.

Data Availability Statement
The CAMS Reanalysis data (Inness et al., 2019) were obtained from ECMWF's web repository at https:// apps.ecmwf.int/data-catalogues/cams-reanalysis/?class=mc&expver=eac4 through the corresponding We-bAPI and can be also accessed at the CAMS Atmosphere Data Store (ADS) https://ads.atmosphere.copernicus.eu/cdsapp#!/dataset/cams-global-reanalysis-eac4?tab=form. The MERRA-2 data (Gelaro et al., 2017) were obtained from the NASA Earthdata website (GMAO, 2015) and can be accessed at https://disc.gsfc. nasa.gov/datasets/M2I3NVASM_5.12.4/summary. This study contains modified Copernicus Atmosphere Monitoring Service Information (2020); neither the European Commission nor ECMWF is responsible for any use that may be made of the information it contains.