The operational CAMS global forecasting system uses fully integrated chemistry in the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecasting System (IFS). The IFS meteorology drives atmospheric composition changes, and the IFS simulates atmospheric chemistry at a resolution of about 40 km . CAMS uses the IFS data assimilation system to assimilate observations of atmospheric composition and includes weather–chemistry feedbacks and references therein. For ozone the CAMS near-real-time system only assimilates satellite retrievals. These include total column ozone retrievals from the Ozone Monitoring Instrument (OMI) and the Global Ozone Monitoring Experiment-2 (GOME-2) on Metop-A and Metop-B, profile data from the Microwave Limb Sounder (MLS) and partial columns from Solar Backscatter Ultra-Violet (SBUV/2) and from the Ozone Mapping and Profiler Suite (OMPS). Details of the ECMWF's 4-D data assimilation system for aerosol, greenhouse gases and reactive gases can be found in .

In addition to chemistry, IFS also includes greenhouse gases and aerosols . IFS applies the Carbon Bond 2005 (CB05) chemical mechanism, which describes tropospheric chemistry with 55 species and 126 reactions . Stratospheric ozone chemistry in IFS is parameterized by the Cariolle scheme . Chemical tendencies of stratospheric and tropospheric ozone are merged at an empirical interface of the diagnosed tropopause height in IFS . In this paper we use IFS day 1 forecasts of ozone, geopotential, u and v wind components, specific humidity, relative humidity and PV. In order to assess the impact of chemical data assimilation on ozone representation during an STT event, an additional IFS control run without data assimilation (free running ozone) is used for intercomparison. Moreover, to evaluate the forecast performance of CAMS global forecast system during the STT event the IFS day 2 to day 5 forecasts of ozone are also used.

The CAMS regional forecasting service is operated by Météo-France and provides daily 4-day forecasts of the main air pollutants and pollens, from seven state-of-the-art regional atmospheric chemistry models (http://atmosphere.copernicus.eu/documentation-regional-systems, last access: 19 October 2018) and from the median ensemble calculated from the seven model forecasts. The 96 h forecasts are available with an hourly resolution and a spatial resolution of 0.1∘ from the surface up to 5 km. Currently the CAMS regional ensemble (RegEns) consists of the following regional models: CHIMERE from INERIS (National Institute for Industrial Environment and Risks) , EMEP from MET-Norway , EURAD-IM from the University of Cologne , LOTOS-EUROS from KNMI (Royal Netherlands Meteorological Institute) and TNO (Netherlands Organisation for Applied Scientific Research) , MATCH from SMHI (Swedish Meteorological and Hydrological Institute) , MOCAGE from Météo-France and SILAM from FMI (Finnish Meteorological Institute) . All regional model data are produced on a horizontal domain of 25∘ W–45∘ E and 30–70∘ N, covering a large European domain. The RegEns members have been documented and evaluated during the MACC projects . The ozone results from RegEns and RegEns members presented here correspond to day 1 forecasts. The meteorological conditions in every model are driven by the operational ECMWF meteorological forecasts, which are at 10 km horizontal resolution during the period of study. The anthropogenic emissions are issued from the TNO MACC-III emission inventory over Europe for the year 2011, which is an updated version of the TNO MACC-II inventory . All models use the concentrations of gas and aerosol species from the global CAMS system as lateral boundary conditions, which makes the regional model outputs consistent with the global model output. The differences between the seven models thus come from the different representation of the chemistry and aerosols, of the physical and dynamical processes and of the natural emissions inside the domain. Table  presents the CAMS models and simulations used in the present study.

The observational data used in this paper include images by the Meteosat Second Generation (MSG) (Geo-Stationary) Satellite (NERC Satellite Receiving Station, Dundee University, Scotland, http://www.sat.dundee.ac.uk/, last access: 17 March 2017). MSG carries the Spinning Enhanced Visible and InfraRed Imager (SEVIRI) instrument, which has the capacity to observe the Earth in 12 spectral channels. Here, we present images from the mid-infrared/water vapour channel (5.35–7.15 µm) for 12:00 Z on 6 January 2017 and 12:00 Z on 7 January 2017. Radiosonde data in the form of skew-T log-P diagrams (taken from the Wyoming University, Department of Atmospheric Science, http://weather.uwyo.edu/upperair/sounding.html last access: 27 April 2018) are used from four European stations:

Norderney (10113), Germany, 53.71∘ N–7.15∘ E (12:00 Z on 3 January 2017 and 12:00 Z on 4 January 2017);

In early January 2017, severe winter weather struck several European regions, namely the Baltic Sea, northern Germany, Italy, the Balkan Peninsula and Turkey, with flooding, extreme cold and snow . The international news media reported that at least 61 people died because of the extremely cold weather conditions in central, eastern and southern Europe . The prevailing synoptic conditions associated with these weather events are depicted in Fig. , which presents the temporal evolution (every 12 h) of IFS geopotential height, wind speed and wind direction at 300 hPa during the time period 3–9 January 2017. An upper level ridge gradually formed over the eastern Atlantic and western Europe in conjunction with a deep upper level trough over eastern and central Europe. Additionally, the jet stream was found on the western side of the upper level trough, with wind speeds occasionally exceeding 65 ms-1 (12:00 Z on 4 January 2017 and 00:00 Z on 5 January 2017). This synoptic situation resulted in the advection of very cold arctic air masses towards eastern, central and southern Europe and favoured the formation of tropopause folds along the path of the jet stream. In its later stage (00:00 Z on 8 January 2017 and after), the southernmost part of the system detached from the main stream, forming a cut-off low over the Balkans. The IFS temperatures at 850 hPa, averaged from 00:00 Z on 7 January to 21:00 Z on 10 January 2017, were below -14 ∘C in most of the Balkans, reaching values below -18 ∘C in the western Balkans (not shown). To stress the exceptional intensity of the cold intrusion, it is noted that the monthly mean climatological temperatures for January at 850 hPa, derived from ERA-Interim reanalyses for the 1981–2010 period, are not lower than -4 ∘C in the Balkan region (not shown).

The horizontal thermal advection at 850 hPa was calculated at 3 h intervals, using the IFS data and employing second-order centred finite differences for the estimation of the horizontal derivatives. Cold advection at 850 hPa occurred in large parts of central, eastern and southern Europe in early January. Strong negative values of the horizontal thermal advection (<-1.5 K hr-1) were exhibited continuously in large parts of Italy (5–7 January), the northern Balkans and central Europe (4–9 January), the western Balkans along the Adriatic coast (5–11 January) and northern Greece, southern FYROM (former Yugoslav Republic of Macedonia) and southwest Bulgaria (6–9 January). The latter maximum in cold advection resulted in a record period of 7 (5) consecutive days with frost (maximum daily temperature below 0 ∘C) from 6 to 12 (7 to 11) January at Thessaloniki (northern Greece), which is located a few metres above sea level.

To further explore the meteorological conditions and to investigate the case of stratospheric intrusions into the troposphere during the examined period, several stratospheric tracers are analysed from both IFS and observations. The water vapour satellite images at 12:00 Z on 6 and 7 January 2017, presented in Fig. a and b, respectively, display a “hook-shaped” streamer of dry air (dark shades) extending from north-eastern Europe to the central Mediterranean. This is a typical pattern encountered during STT events . The fields of IFS specific humidity at 500 hPa on the same days (Fig. c and d) resemble the observed satellite images. These depict a hook-shaped region of air with low specific humidity, affirming that the presence of dry air in the troposphere is captured well by the IFS global model. The respective PV isosurfaces of 1.5 pvu (Fig. c and d) overlap the band of dry air in the troposphere, while high ozone concentrations, up to 130 ppb, are also found over this dry streamer (Fig. e and f). Altogether, Fig.  indicates that this dry air with relatively high PV values and high ozone concentrations is of stratospheric origin.

Figure  presents the evolution (12 h interval) of ozone concentrations exceeding 50 ppb, geopotential height and PV isosurfaces of 1.5 pvu from IFS at 500 hPa for the time period 4–8 January 2017, to examine ozone enhanced in the middle troposphere owing to STT in relation to the predominant synoptic–dynamic conditions. At 12:00 Z on 4 January 2017 a streamer of high ozone concentrations with values up to about 100 ppb is found over the Baltic Sea and northern Germany, near the ridge exit and trough entrance, where convergence and descending motions prevail, and in the vicinity of the jet stream (Fig. ). During the next 24 h as the system moves further south, the streamer of high ozone concentrations crosses central Europe following the path of the jet stream. At 00:00 Z and 12:00 Z on 6 January 2017 ozone concentrations exceeding 130 ppb, linked with high PV values (>1.5 pvu), are found over the central Mediterranean, highlighting the vertical transport of ozone from the stratosphere down to the middle troposphere. During the next 48 h, the high ozone streamer moves further eastward, affecting the island of Crete (7 and 8 January 2017), and gradually dissipates.

In order to explore the capability of the regional models to reproduce the enhanced ozone seen in the mid-troposphere due to STT, the fields of RegEns ozone exceeding 50 ppb at 5000 m are shown in Fig.  for the same dates as in Fig. . Visual inspection of Fig.  indicates that the RegEns compares well with IFS as it synchronously captures the spatial distribution of ozone concentrations. In more detail, the hook-shaped patterns of high ozone are seen well in the CAMS regional product, with ozone mixing ratios exceeding 90 ppb at 12:00 Z on 6 January 2017 over the central Mediterranean. Although the spatio-temporal features of ozone in the RegEns agree well with that of the IFS, in quantitative terms, there are discrepancies between the regional and the global product. This is likely due to the fact that (a) the RegEns is presented at a 5000 m level (the uppermost level available) and the IFS at 500 hPa, (b) different resolution and advection schemes are used in global and regional models and (c) pressure and temperature values from the US Standard Atmosphere were used for units' conversion in RegEns. Considering the ERA-Interim temperatures during the period of interest for the units conversion may result in even lower RegEns ozone concentrations of up to ∼7 % in the regions exhibiting the lower temperatures (not shown). Overall, the agreement between the CAMS global and regional products highlights the critical role that the IFS boundary conditions and meteorological drivers play in the regional models for forecasting an STT event and the induced downward transport of ozone.

Four sites are selected (Norderney, Germany; Muenchen, Germany; Trapani, Italy; Heraklion, Greece), located within the system transit path with available radiosonde observations, in order to study the vertical structure of the STT event and the subsequent transport of stratospheric ozone into the troposphere. To better depict the impact of STT on tropospheric ozone, two dates for analysis are selected for each site: one prior and one during the STT occurrence.

Starting from Norderney (see location in Fig. a), the skew-T log-P diagrams for 12:00 Z on 3 January 2017 and 12:00 Z on 4 January 2017 are presented in Fig. b and c, respectively. As can be seen from the comparison between the two figures, a distinct decrease of humidity (departure of dew-point curve (left) and temperature curve (right)) is found at 12:00 Z on 4 January 2017 between 250 and 400 hPa, while the tropopause drops to approximately 400 hPa. Furthermore, the vertical profile of the IFS ozone mixing ratio over Norderney during the examined dates (Fig. d) indicates a remarkable increase of ozone down to 400 hPa, verifying the aforementioned observed folding of the tropopause. The vertical profiles of the observed and IFS relative humidity (Fig. d) show a sharp decrease at 400 hPa, revealing that the intrusion of dry stratospheric air in the troposphere is captured well by the IFS. A comprehensive view of the induced stratospheric intrusion over Norderney is provided through the longitude–pressure cross section at 53.6∘ N, showing ozone, PV (2 pvu isosurface) and wind speed at 12:00 Z on 4 January 2017 (Fig. e). An impressive downward penetration of ozone-rich and PV-rich (>2 pvu) air down to approximately 600 hPa is found in the free troposphere and over the greater Norderney longitude band. The 2 pvu PV isosurface (dynamical tropopause; e.g. ) illustrates the tropopause folding on the right side of the jet stream (black contours) and down to 450 hPa at 5∘ E. The stratospheric origin of ozone in the upper troposphere over Norderney is also supported by the IFS ozone and specific humidity time series at 400 hPa, revealing a significant anti-correlation at the 95 % confidence level (Fig. f). The respective diagrams for Muenchen are presented in Fig.  for 00:00 Z on 4 January 2017 and 12:00 Z on 5 January 2017. Similarly, an intrusion of dry air is observed in the upper and middle troposphere (down to 550 hPa) at 12:00 Z on 5 January 2017 (Fig. b and c), which, along with the sharp increase/decrease of IFS ozone/relative humidity above 550 hPa (Fig. d), partially seen in RegEns ozone vertical profiles, indicates the downward transport of dry stratospheric air into the troposphere. The longitude–pressure cross section over Muenchen at 12:00 Z on 5 January 2017 (Fig. e) depicts the folding of the tropopause (2 pvu isosurface) in the vicinity of the jet stream and the associated vertical transport of ozone-rich air down to 600 hPa. In support of the above, the distinct increase of IFS ozone at 400 hPa is combined with a sharp decrease of IFS specific humidity (significant anti-correlation at the 95 % confidence level) (Fig. f).

A period of 12 h later (00:00 Z on 6 January 2017), and as the system moved further south, a dramatic decrease of humidity is observed in the middle troposphere and down to approximately 550 hPa over Trapani (Fig. b and c), with specific and relative humidity at 500 hPa dropping from 0.75 gkg-1 and 58 % (00:00 Z on 5 January) to 0.01 gkg-1 and 2 % (00:00 Z on 6 January 2017) respectively. The IFS specific humidity values at 500 hPa for the same dates are 0.49 and 0.025 gkg-1, respectively. On top of that, the vertical profiles of the observed and IFS relative humidity (Fig. d) indicate that the sharp decrease of humidity is reproduced well by the CAMS global model. The IFS system captures the dynamical features of the stratospheric intrusion as it is depicted in the vertical profiles of ozone, showing increased concentrations at 00:00 Z on 6 January 2017 down to 600 hPa, which is also seen in CAMS RegEns (Fig. d). The intense tropopause folding over Trapani is illustrated in Fig. e, with the dynamical tropopause dropping down to 550 hPa and ozone-rich air penetrating down to 800 hPa. Again, a significant anti-correlation at the 95 % confidence level is found between the IFS ozone and specific humidity time series at 400 hPa, indicating that the ozone increase results from the downward transport of ozone from the stratosphere (Fig. f). The three-dimensional field of IFS ozone concentrations exceeding 80 ppb at 00:00 Z on 6 January 2017 is presented in Fig. a, depicting the stratospheric ozone intrusion into the troposphere and over the broader Trapani region. The three-dimensional IFS ozone concentration isosurface of 100 ppb (Fig. b) resembles the folding of the tropopause along a north-east-oriented conceivable axis, which coincides with the high wind speed flow in the upper troposphere (Fig. ). Later on and over Heraklion (see location in Fig. a), the skew-T log-P diagrams for 12:00 Z on 5 January 2017 and 00:00 Z on 8 January 2017 (Fig. b and c) and the respective vertical profiles of IFS ozone and relative humidity (Fig. d) reveal the presence of dry ozone-rich air in the upper and middle troposphere (down to 500 hPa). The increase in IFS ozone time series at 400 hPa is synchronized with the decrease of IFS specific humidity (significant anti-correlation at the 95 % confidence level), indicating that dry stratospheric air rich in ozone is transported into the troposphere over Heraklion (Fig. f). A more illustrative representation of the development and evolution of the examined STT event is provided in the three-dimensional animation (from 12:00 Z on 3 January 2017 to 21:00 Z on 8 January 2017 with 3 h interval) of IFS ozone concentrations exceeding 80 ppb in the Supplement.

In order to evaluate the forecasting capability of both IFS and RegEns regarding the downward transport of ozone during the examined STT event, we compare CAMS forecasts with profile observations from ozonesondes (WOUDC) and aircraft measurements (IAGOS). Two sites located across the passage of the examined system with available observational data during the examined period were selected: (a) Prague (ozonesondes) and (b) Frankfurt (aircraft measurements). The model error is quantified using the fractional gross error (FGE), which ranges between 0 and 2 and behaves symmetrically with respect to under- and overestimation: FGE=2N∑iNMi-OiMi+Oi, where Mi represents the model value for level i, Oi is the corresponding observed value and N is the number of sample values.

Figure  displays the vertical profiles of observed and forecasted (IFS and RegEns) ozone concentrations over Prague at 11:00 Z (12:00 Z for CAMS models) on 2 January 2017 (prior the STT event) and 11:00 Z (12:00 Z for CAMS models) on 4 January 2017 (during the STT event). The intercomparison between the observed vertical profiles of ozone on the two dates indicates a distinct increase of ozone concentrations in the upper troposphere, probably related to the vertical transport of ozone from the stratosphere, reaching down to approximately 500 hPa. In support of the above findings, the respective vertical profiles of the observed and IFS relative humidity both show a distinct decrease at 500 hPa. Although the CAMS global model seems to underestimate (overestimate) ozone in (above) the free troposphere, the transition from the neutral condition to the STT event is captured well by the IFS day 1 forecast (Fig. ), with an FGE value of 0.13 (300–500 hPa) on 4 January 2017. Whilst data assimilation resulted in overestimating ozone near the tropopause compared with the control run (Fig. a), it is clearly beneficial in reproducing the increase of ozone in the upper troposphere during the STT event (Fig. b). Notably, the respective FGE value for the control run at 4 January 2017 is 0.29, revealing an improvement in model performance due to data assimilation. Ozone in the RegEns forecast is higher within the planetary boundary layer than in IFS, with a relatively small spread among the RegEns members (Fig. a). In the free troposphere, the range of regional variability increases; however the RegEns remains close to the global forecast. The RegEns is also able to reproduce the ozone enhancement, following the IFS forecast closely (Fig. b). Day 1 to day 5 forecasts of IFS ozone indicate that the observed ozone increase in upper troposphere during the STT event is satisfactorily forecasted up to 3 days in advance, with FGE values not higher than 0.22 (Fig. b).

Three ozone profiles from aircraft measurements (two take-offs and one landing) over the broader region of Frankfurt at 13:00 Z on 4 January 2017, 6:00 Z on 5 January 2017 and 13:00 Z on 5 January 2017 are compared with the respective IFS and RegEns ozone profiles in Fig. . At 13:00 Z (12:00 Z for CAMS models) on 4 January 2017, the profile of the IFS ozone day 1 forecast is found to be in very good agreement with the IAGOS data, both depicting the increase of ozone down to approximately 500 hPa (Fig. a). The FGE was 0.04 for the 300–500 hPa altitude range. The respective profiles 17 h later (06:00 Z on 5 January 2017) also reveal enhanced ozone concentrations in the upper troposphere, which are captured by the IFS day 1 forecast (Fig. b) (FGE = 0.19 at 300–500 hPa). Finally, at 13:00 Z (12:00 Z for CAMS models) on 5 January 2017, the IFS day 1 forecast is found to overestimate the observed high ozone concentrations between 250 and 350 hPa, while it qualitatively captures the observed high ozone pattern in the middle troposphere between 400 and 600 hPa (Fig. c) (FGE = 0.30 at 400–600 hPa). The advantageous role of data assimilation can be affirmed from the intercomparison with the IFS control run, which exhibits FGE values of 0.34, 0.30 and 0.12 for the three dates respectively. A better agreement with observations is found for IFS when implementing data assimilation at 13:00 Z on 4 January 2017 and 06:00 Z on 5 January 2017 (Fig. a and b), while at 13:00 Z on 5 January 2017 (Fig. c), although the control run performs better in terms of bias, the data assimilation seems to help in the direction of reproducing the observed ozone peak in the middle troposphere. Concerning the RegEns, due to its limited vertical profile, up to about 550 hPa, the evaluation of its forecast performance is restricted. Nevertheless, there is a clear signal of increased ozone in the uppermost vertical level during all three dates. Regarding the forecast performance of CAMS global model, a relatively good agreement with observations is seen up to forecast day 3 at 13:00 Z on 4 January 2017 and 06:00 Z on 5 January 2017 (FGE values not higher than 0.25) (Fig. a and b), while at 13:00 Z on 5 January 2017 the observed ozone peak in the middle troposphere is somehow captured up to forecast day 3 but overestimated. Figure  depicts the FGE values of IFS ozone in relation to the forecast day for the observational instances of Prague and Frankfurt. Overall, a satisfactory forecast performance is revealed up to 3 days in advance with FGE values not higher than 0.3. Forecast day 1 exhibits the best agreement with observations, while after forecast day 3, more discrepancies are found between the forecast and the observations (see also Figs.  and ).