A process-oriented evaluation of CAMS reanalysis ozone during tropopause folds over Europe for the period 2003–2018

. Tropopause folds are the key process underlying stratosphere-to-troposphere transport (STT) of ozone, and thus they affect tropospheric ozone levels and variability. In the present study we perform a process-oriented evaluation of Copernicus Atmosphere Monitoring Service (CAMS) reanalysis (CAMSRA) O 3 during folding events over Europe and for the time period from 2003 to 2018. A 3-D labeling algorithm is applied to detect tropopause folds in CAMSRA, while ozonesonde data from WOUDC (World Ozone and Ultraviolet Radiation Data Centre) and aircraft measurements from IAGOS (In-service Aircraft for a Global Observing System) are used for CAMSRA O 3 evaluation. The proﬁles of observed and CAMSRA O 3 concentrations indicate that CAMSRA reproduces the observed O 3 increases in the troposphere during the examined folding events. Never-theless, at most of the examined sites, CAMSRA overestimates the observed O 3 concentrations, mostly at the upper portion of the observed increases, with a median fractional gross error (FGE) among the examined sites > 0 . 2 above 400 hPa. The use of a control run without data assimilation reveals that the aforementioned overestimation of CAMSRA O 3 arises from the data assimilation implementation. Overall, although data assimilation assists CAMSRA O 3 to follow the observed O 3 enhancements in the troposphere during the STT events, it intro-duces biases in the upper troposphere resulting in no clear quantitative improvement compared to the control run without data assimilation. Less biased assimilated O 3 products, with ﬁner vertical resolution in the troposphere, in addition to higher IFS (Integrated Forecasting System) vertical resolution, are expected to provide a better representation of O 3 variability during tropopause folds.

magnitude can vary widely across chemistry climate models (Young et al., 2018;Griffiths et al., 2021;Szopa et al., 2021). The net stratospheric influx results from STE processes, comprised of stratosphere-to-troposphere transport (STT) and troposphere-to-stratosphere transport (TST) . The latter constitutes an important pathway through which very short-lived substances (VSLS) emitted at the surface can be transported to the lower stratosphere influencing ozone (Levine et al., 2007;Aschmann et al., 2009;Liang et al., 2014). The main mechanism for STT is tropopause folding , which results in the downward transport of stratospheric ozone-rich air into the troposphere, a process known as stratospheric intrusion (Danielsen and Mohnen, 1977). Therefore, tropopause folding events affect tropospheric composition and in particular tropospheric ozone levels (Beekmann et al., 1997;Ott et al., 2016;Cooper et al., 2005;Tarasick et al., 2019;Zhao et al., 2021a), especially in regions that are known to be hot spots of fold activity (Zanis et al., 2014;Akritidis et al., 2016;Ojha et al., 2017). Occasionally, during deep and intense folding events, stratospheric air is transported down to the lower troposphere or even to the planetary boundary layer leading to changes in tropospheric and surface ozone concentrations (Langford et al., 2009;Lin et al., 2015;Knowland et al., 2017). Model projections suggested that under a changing climate tropopause folds will be associated with both future increases and interannual variability in ozone STT (Akritidis et al., 2019).
The spatial and temporal characteristics of tropopause folds occurrence around the globe have been the subject of study in recent years, suggesting the jet stream location, intensity, and seasonality as their main drivers (Elbern et al., 1998;Sprenger et al., 2003;Škerlak et al., 2014;Akritidis et al., 2021). The springtime western United States region is a hot spot of deep folding events with well-known implications for tropospheric ozone and air quality (Langford et al., 2009;Lin et al., 2012Lin et al., , 2015Knowland et al., 2017). Recently, Luo et al. (2019) explored the seasonal features of tropopause folds over the Tibetan Plateau where folds occur frequently , while other studies investigated the effect of tropopause folds on lower-tropospheric ozone levels and air quality in China (Lu et al., 2019;Zhao et al., 2021b, a). Regarding the broader European region, the summertime Eastern Mediterranean is a well-known hot spot of fold activity Akritidis et al., 2016), resulting from the interaction of the subtropical jet stream and the South Asian monsoon , while further north folds occur in the vicinity of cyclones (Reutter et al., 2015;Antonescu et al., 2013;Knowland et al., 2015). During the past 2 decades several studies have explored the impact of stratospheric intrusions on tropospheric ozone levels and variability over Europe (Stohl et al., 2000;Cristofanelli et al., 2006;Trickl et al., 2020), as well as the quality of their forecast Trickl et al., 2010;Akritidis et al., 2018).
Nowadays, a comprehensive framework to study the contribution of stratospheric intrusions to tropospheric ozone are atmospheric composition reanalysis products that provide global meteorological and ozone data in relatively high spatial and temporal resolution. Yet, before estimating the impact of STT events on tropospheric ozone, a processoriented evaluation of the reanalysis product during such events is deemed essential. The latest reanalysis of atmospheric composition produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) is the Copernicus Atmosphere Monitoring Service (CAMS) reanalysis (CAMSRA) (Inness et al., 2019a). Within the framework of the CAMS service element CAMS_84, Akritidis et al. (2018)  In the present study we perform a process-oriented evaluation of CAMSRA O 3 during STT events, selected from the CAMSRA tropopause folds database by Akritidis et al. (2021), for the European region and over the time period 2003-2018. Compared with other regions worldwide, the European region exhibits relatively high observational data availability for the examined period. In addition, the role of IFS chemical data assimilation in O 3 STT is explored. Section 2 describes WOUDC (World Ozone and Ultraviolet Radiation Data Centre) and IAGOS (In-service Aircraft for a Global Observing System) O 3 data used for the evaluation, ECMWF IFS system and CAMSRA data, the 3-D labeling algorithm applied for tropopause fold detection, and the methodological approach for the selection of STT events. Section 3 presents the main evaluation results, and finally Sect. 4 summarizes the key findings of the study.

Observational data
To evaluate CAMSRA O 3 during folding events, ozonesonde measurements were obtained from the WOUDC network (WMO/GAW Ozone Monitoring Community, 2022) (last access: 7 February 2022) for nine European sites, namely Lerwick, United Kingdom (UK) (LER); Uccle, Belgium (UCC); Hohenpeissenberg, Germany (HOH); Payerne, Switzerland (PAY); Legionowo, Poland (LEG); Madrid, Spain (MAD); De Bilt, the Netherlands (DBI); Lindenberg, Germany (LIN); and Prague, Czech Republic (PRA). At all sites the measurements are carried out with electrochemical concentration cell (ECC) ozonesondes (Komhyr, 1969), except at Hohenpeissenberg where the Brewer Mast ozonesonde (Brewer and Milford, 1960) is used. Both ozonesonde types are based on the same measurement principle of ozone electrochemical detection in potassium iodine. The major differences between ECC and Brewer Mast ozonesondes are that the latter uses only one reaction chamber and a silver anode instead of a platinum anode, requiring an external electrical potential in contrast to the ECC (Beekmann et al., 1994). The precision of ECC ozonesondes in the troposphere (below 200 hPa) is between −7 % and +17 %, as reported by Komhyr et al. (1995), while for the Brewer Mast ozonesondes the same order of precision was found by Steinbrecht et al. (1998). The ozonesonde observations are compared against CAM-SRA O 3 concentrations of the nearest grid point and time step.
Aircraft ozone measurements from the IAGOS (In-service Aircraft for a Global Observing System) program were also used (http://iagos-data.fr/, last access: 21 April 2020). Within the framework of IAGOS, instruments are carried on commercial airlines, measuring ozone, carbon monoxide, and water vapor along with meteorological parameters and cloud particles. Details of the IAGOS project can be found in Petzold et al. (2015), with the technical details of the instrumentation, operations, and validation presented in . Five IAGOS airports were selected for the evaluation, based on their data temporal coverage, namely Paris, France (PAR); Düsseldorf, Germany (DUS); Frankfurt, Germany (FRA); Munich, Germany (MUN); and Vienna, Austria (VIE). The IAGOS O 3 data have an accuracy of ±2 ppb, a precision of ±2 %, and a detection limit of 2 ppb (Blot et al., 2021). Landing and takeoff O 3 profiles are compared against CAMSRA O 3 profiles. It should be noted that the IAGOS profiles are not strictly vertical. To this end, and in order to perform a more realistic evaluation of CAMSRA O 3 according to the flight position (longitude, latitude, pressure), the respective CAMSRA grid points are extracted at the nearest time to that of the takeoff or landing. The selection of both ozonesonde sites and IAGOS airports was based in the availability of at least 500 profile observations throughout the 2003-2018 period. This objective criterion ensures a sufficient number of both observational sites and folding events to be selected for the analysis. It is noteworthy to mention that both ozonesondes and IAGOS profiles are not assimilated and hence they constitute completely independent validation data. The location of the examined WOUDC ozonesonde sites and IAGOS airports are depicted in Fig. 1.

CAMS reanalysis
CAMSRA is the latest reanalysis dataset produced by ECMWF, including three-dimensional fields of meteorological, chemical, and aerosol species for the period from 2003 onwards. It comes as a follow-up of the previous successful reanalysis products, the Monitoring Atmospheric Composition and Climate (MACC) reanalysis (MACCRA) (Inness et al., 2013) and the CAMS interim reanalysis (CAM-SIRA) (Flemming et al., 2017). CAMSRA is based on the ECMWF's IFS CY42R1 cycle and the 4D-VAR data assimilation system (Inness et al., 2019a). In more detail, 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 12 h assimilation windows (from 09:00 to 21:00 and 21:00 to 09:00 UTC) by modifying the initial conditions accordingly. To this end, satellite retrievals of total column CO, tropospheric column NO 2 , aerosol optical depth, and total column, partial column and profile ozone retrievals are assimilated in the IFS system. More details on the satellite retrievals (product, satellite, period) assimilated in CAMSRA can be found in Table 1 of the CAMSRA evaluation study by Wagner et al. (2021). In addition, meteorological observations, including satellite, PILOT, in situ, radiosonde, dropsonde, and aircraft measurements are also incorporated in IFS. The chemical mechanism used in the IFS is an extended version of the Carbon Bond 2005 (CB05) tropospheric chemical mechanism  and stratospheric ozone chemistry is parameterized by a "Cariolle scheme" (Cariolle and Déqué, 1986;Cariolle and Teyssèdre, 2007). The emissions consist of the MACCity (MACC and CityZEN EU projects) anthropogenic emissions (Granier et al., 2011), the GFAS (Global Fire Assimilation System) fire emissions (Kaiser et al., 2012), and the MEGAN2.1 (Model of Emissions of Gases and Aerosols from Nature) biogenic emissions (Guenther et al., 2006). The CAMSRA data have a spatial resolution of approximately 80 km (0.7 • × 0.7 • grid) with 60 hybrid sigma-pressure (model) levels (13 levels between approximately 400 and 100 hPa) in the vertical (top level at 0.1 hPa) and a temporal resolution of 3 h. The quality of the CAMSRA O 3 field is documented in Wagner et al. (2021) and comprehensive validation reports that can be found on the CAMS website https: //atmosphere.copernicus.eu/eqa-reports-global-services (last access: 10 May 2022) (e.g., Errera et al., 2021).
To investigate the role of chemical data assimilation in tropospheric ozone representation during folding events, a control simulation of IFS without the use of chemical data assimilation (CAMSRA no DA) is also used. As it would have been computationally too expensive to produce a control analysis experiment that was identical to CAMSRA but did not actively assimilate observations of reactive gases, a forecast run was carried out that applied the same settings (model code, resolution, emissions) as used in CAMSRA. The control run was carried out as a sequence of 24 h. The meteorological initial conditions were taken from CAMSRA, but the initial conditions for the atmospheric composition species, including ozone, from the previous forecast. It thus allows us to detect the impact of the assimilation of ozone data, for example, by comparing its ozone fields with CAM-SRA. This approach was also followed by Akritidis et al. (2018) in their evaluation of CAMS forecasting systems during a deep STT event over Europe, indicating an overall improvement of IFS performance due to the chemical data assimilation implementation.
Apart from O 3 , a stratospheric ozone tracer (O 3s ) is also used from CAMSRA providing a diagnostic of O 3 STT. In principle, O 3s in IFS is defined identically with O 3 in the stratosphere; however, in CAMSRA O 3s is equal to the modeled (Cariolle scheme) O 3 tracer and not the assimilationresulted O 3 . In the troposphere O 3s is subject to transport and chemical destruction just like O 3 . The tropopause in CAMSRA is calculated based on the temperature lapse rate, switching the chemistry scheme from CB05 (troposphere) to Cariolle (stratosphere) accordingly. It should be noted that O 3s is used here only as a qualitative diagnostic of ozone STT to support evidence of stratospheric ozone downward transport during the folding events.

Fold detection in CAMS reanalysis
Tropopause folds are identified in CAMSRA using the latest version of the 3-D labeling and fold detection algorithm by Škerlak et al. (2015), initially developed by Sprenger et al. (2003). Here, we adopted the 3-D labeling algorithm to detect folds in CAMSRA, using as inputs the fields of potential vorticity (PV), potential temperature, specific humidity, and surface pressure. The 3-D fields of pressure are constructed and the pressure level of the dynamical tropopause (Holton et al., 1995;Stohl et al., 2003) is determined using the lower of the isosurfaces of PV at 2 PVU and potential temperature at 380 K. Subsequently, the vertical profile for each grid point is examined, and a fold is assigned when multiple crossings of the tropopause are identified. Still, there are specific cases 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 3-D labeling algorithm, using physical and geometrical criteria, labels the air masses as follows: tro- pospheric (label = 1), stratospheric (label = 2), stratospheric cut-off or diabatically produced PV anomaly (label = 3), tropospheric cut-off (label = 4), surface-bound PV anomaly (label = 5). The diabatically produced PV anomalies merged with the stratosphere are distinguished using a specific humidity threshold of 0.1 g kg −1 . Further details on the criteria used for the 3-D labeling can be found in Škerlak et al. (2015). Therefore, a fold is identified when a 2 → 1 → 2 → 1 or 3 transition is detected on a vertical profile (from top to bottom), with the algorithm outputting 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 p = p m − p u , which depicts the vertical extent of the fold. The spatial distribution of CAM-SRA monthly mean tropopause folds (with p ≥ 50 hPa) frequency over Europe for the period 2003-2018 is presented in Fig. S1 of the Supplement.

Selection of STT events
To perform the process-oriented evaluation of CAMSRA O 3 , the STT events are selected for each WOUDC ozonesonde site by applying the following methodology.
a. For every ozonesonde profile, the time and location of release are extracted.
b. For the CAMSRA grid cell, including the ozonesonde site location, and for the CAMSRA 3 h time steps before and after the time of ozonesonde release, the presence of a tropopause fold with p ≥ 50 hPa is explored (e.g., if the ozonesonde release was 14:00 UTC, we search for folds at the 12:00 and 15:00 UTC time steps of the respective grid cell).
c. If a fold is found, the respective ozonesonde profile is classified in "STT events", while it otherwise is classified in "rest of events".
d. The ozonesonde data are vertically interpolated (linear) with a step of 25 hPa, and only the profiles exhibiting a data completeness ≥ 75 % from 900 to 300 hPa are kept in the STT events and rest of events records. Merging the STT events record with the rest of events record provides the climatology of ozonesonde profile.
The same approach is followed (steps a, c, and d) for the IAGOS data with one difference in step b. Since the aircraft measurement profiles during takeoff and landing are not strictly vertical, a tracking of the aircraft position is performed and the respective CAMSRA grid cells that include the aircraft route are extracted. Subsequently, the presence of a tropopause fold with p ≥ 50 hPa is explored if it is found in at least one of the extracted grid cells. A schematic representation of the applied methodology for the STT events selection is illustrated in Fig. 2. For direct comparison with observations, CAMSRA O 3 concentrations are also vertically interpolated (linear) with a step of 25 hPa.

Comparison of observed and CAMSRA climatological O 3 profiles
Before proceeding with the process-oriented evaluation of CAMSRA O 3 during the STT events, we present a comparison of CAMSRA O 3 profiles against observations (ozonesonde and aircraft measurements) during all the events (STT events + rest of events), to ensure that CAMSRA reproduces the climatological features of the observed O 3 profiles at the examined European sites. Figure  More specifically, CAMSRA exhibits higher O 3 concentrations throughout the troposphere at Hohenpeissenberg and Paris and to a lesser extent at Frankfurt and Munich, with the greatest overestimation seen in the upper troposphere at all sites. Similar O 3 overestimations in the free troposphere over Hohenpeissenberg are also reported for previous ECMWF atmospheric composition reanalysis products such as the MACC reanalysis (see Figs. 7 and 8 in the evaluation study by Katragkou et al., 2015). At Payerne and Düsseldorf, CAMSRA O 3 is only overestimated above 500 hPa. At the rest of the sites CAMSRA O 3 is quantitatively in very good agreement with observations. The differences seen in the comparison between the observed and CAMSRA O 3 concentrations among the examined sites are presumably related to regional differences and uncertainties in O 3 precursor emissions affecting modeled local net photochemical O 3 production rates, as well as the spatiotemporal representativeness of WOUDC vertical profiles and IAGOS aircraft takeoff and landing routes by the selected CAMSRA grid points and time steps.

Evaluation of CAMSRA O 3 during STT events
In Figs in the upper troposphere. As mentioned above, the O 3 overestimation in the upper troposphere is an already known issue in both CAMS near-real-time analysis and reanalysis products. This might be due to a bias in some of the assimilated data, the likely insufficient vertical resolution of O 3 data assimilated (total column and stratospheric profiles) in IFS to capture STT events, and the O 3 background error formulation in data assimilation. In particular, CAMSRA O 3 vertical profiles during both STT and rest of events exhibit a better agreement in the upper troposphere with observations during the years 2003 and 2004, indicating that the inclusion of the Aura data in the assimilation system from August 2004 and onward is likely to result in O 3 overestimation in the upper troposphere ( Fig. S2 and S3 in the Supplement). In addition, the individual dynamics and the different vertical location and geometrical characteristics of the selected STT events, especially for observational sites with a less extended number of events, may form somehow unique structures of CAMSRA and observed O 3 deviations. In particular, for sites exhibiting a very small number of STT events over the years (e.g., Legionowo), the O 3 vertical variability is not smoothed out, and CAMSRA is not found to be able to reproduce the high-resolution features    of O 3 increase, probably due to its coarser vertical resolution compared to ozonesonde measurements. For a quantitative comparison between CAMSRA and observations, in Fig. 6 we present the vertical profiles of fractional gross error (FGE) and modified normalized mean bias (MNMB) of CAMSRA O 3 for the WOUDC ozonesonde sites and IAGOS airports. The FGE is a normalized version of the mean error, while the MNMB is a normalized version of the mean bias. Both metrics are normalized by the mean of the observed and model (here CAMSRA) values, being dimensionless and relative, and they are thus suitable to use at different heights in the troposphere. FGE and MNMB are insensitive to outliers in the distribution and range between 0 to 2 and −2 to 2, respectively, behaving symmetrically with respect to underestimation and overestimation:

MNMB
where M i is the model (CAMSRA here) value for the ith STT event, O i is the corresponding observed value, and N is the number of STT events. As can be seen in Fig. 6a, the FGE is mostly increasing with height, with values > 0.3 found above 400 hPa at several sites. Indicatively, the median FGE value among the examined sites for each pressure level is > 0.2 above 400 hPa (Fig. 7). As expected, the respective profiles of MNMB in Fig. 6b indicate that the biases are mostly positive, confirming the aforementioned discussion. The median MNMB value among the examined sites for each pressure level ranges from approximately 0 to +0.1 below 400 hPa (Fig. 7), which is in agreement with the MNMB values of CAMSRA O 3 in the free troposphere reported by Inness et al. (2019a) and Wagner et al. (2021). Above 400 hPa, the respective median MNMB value ranges from +0.1 to +0.19 (Fig. 7).

The role of chemical data assimilation
Here we investigate the role of chemical data assimilation in CAMSRA O 3 representation during the selected STT events.
To this end, we present CAMSRA no DA and observed O 3 profiles averaged during the selected STT events and the rest of events, at the WOUDC ozonesonde sites and IAGOS airports in Figs Fig. 10b reveal overall an underestimation of O 3 during the STT events. The median FGE and MNMB values depicted in Fig. 7 suggest an overall improvement of MNMB and FGE in CAMSRA due to chemical data assimilation between 500 and 400 hPa and a deterioration above 350 hPa reflecting the aforementioned CAMSRA O 3 overestimation in the upper troposphere.

Conclusions
A process-oriented evaluation of CAMS reanalysis O 3 during tropopause folding events over the period 2003-2018 is performed using WOUDC ozonesonde data and IAGOS aircraft measurements. The selected STT events were obtained from the CAMSRA tropopause folds database by Akritidis et al. (2021), which was constructed with the implementation of the 3-D labeling and fold detection algorithm by Škerlak et al. (2015). Moreover, the role of chemical data assimilation in O 3 representation during the examined STT events was investigated using a CAMS control simulation without chemical data assimilation. The most notable findings of the study are summarized as follows.
-CAMSRA reproduces the observed O 3 increases in the troposphere during the examined folding events, which as indicated by the respective O 3s profiles are of stratospheric origin.
-For some sites CAMSRA misses the observed return of O 3 concentrations back to normal tropospheric levels, resulting in an overestimation of O 3 in the upper troposphere, with FGE values at 350 hPa ranging from 0.13 to 0.38 (median of 0.28) at the observational sites.
-The use of chemical data assimilation in IFS is found to be beneficial for the representation of CAMSRA O 3 enhancements in the troposphere during the STT events. However, it leads to an overestimation of O 3 concentrations at the upper portion of O 3 increases.
-Overall, and in terms of O 3 bias and absolute bias, only a small improvement is found between 500 and 400 hPa due to chemical data assimilation implementation.
The present analysis indicates that CAMSRA satisfactorily reproduces the observed O 3 increases in the troposphere during the tropopause folding events. Although IFS chemical data assimilation helps CAMSRA O 3 to follow the observed    Author contributions. DA designed the study, performed the analysis, and wrote the manuscript. AP adopted the 3-D labeling algorithm and performed the fold identification calculations. JF and AI have contributed to the development of IFS and the production of CAMS reanalysis of atmospheric composition product. All authors contributed to reviewing and editing of the manuscript and interpretation of the results.

Competing interests.
At least one of the (co-)authors is a member of the editorial board of Atmospheric Chemistry and Physics.
The peer-review process was guided by an independent editor, and the authors also have no other competing interests to declare.

Disclaimer.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Acknowledgements. This research is co-financed by Greece and the European Union (European Social Fund-ESF) through the Operational Programme "Human Resources Development, Education and Lifelong Learning" in the context of the project "Reinforcement of Postdoctoral Researchers -2nd Cycle" (MIS-5033021), implemented by the State Scholarships Foundation (IKY). MOZAIC/-CARIBIC/IAGOS data were created with support from the European Commission, national agencies in Germany (BMBF), France (MESR), and the UK (NERC), and the IAGOS member institutions (https://www.iagos.org/iagos-partner/, last access: 10 May 2022 Review statement. This paper was edited by Andreas Petzold and reviewed by two anonymous referees.