Many efforts are made in research and operations to ensure accurate monitoring of the atmospheric chemical state. This is necessary in order to be able to take, whenever necessary, the right steps to rectify unhealthy behaviours or to take measures to protect the environment and health of populations. To do so, chemistry transport models (CTMs) that predict the evolution of the atmospheric chemical state with quality and precision are required, with a specific focus on both the efficiency of codes and methods themselves and their integration with real observations of the atmospheric state.

At present the observing system of the atmospheric chemical composition is mainly based on in situ measurements and observations from polar-orbiting satellites. The in situ measurements can concern surface observations, carried out through ground-based measurement stations, or airborne observations, acquired through instruments on board aircraft, balloons or drones. Although these kinds of observations have proved to be extremely helpful for atmospheric sciences, they are spatially sparse.

Measurements from polar-orbiting satellites, on the other hand, cover much more land surface on a regular basis. Among the most efficiently used tools belonging to this category, the Infrared Atmospheric Sounding Interferometer (IASI), one of the main payloads of the polar-orbiting Meteorological Operational Satellite (MetOp) series, developed by ESA and operated by the EUMETSAT agency (), is definitely worth mentioning. IASI provides accurate measurements of the meteorological and chemical state of Earth's atmosphere. The instrument acquires spectra of atmospheric emission between 645 and 2760 cm-1, with a spectral apodised resolution of 0.5 and 0.25 cm-1 spectral sampling. Consequently, it measures at 8461 wavelengths (or channels). Thanks to its fine spectral resolution, signal-to-noise ratio and wide spectral range, it is a precious resource for detecting trace gases like ozone, methane and carbon monoxide, as well as clouds, aerosols and greenhouse gases ().

However, satellite instruments in polar orbit are able to provide repeated measurements at the same point on Earth only a few times during the same day. Having IASI-like measurements coming from instruments on board a geostationary (GEO) platform has the potential to provide significant advantages in the area of atmospheric chemistry monitoring. These instruments would acquire much wider views of Earth with a much higher temporal acquisition frequency.

Geostationary motions require the satellite to cover an orbit much further from Earth's surface than satellites in polar orbit. For years, this has been at the expense of both the spatial and the spectral resolution of the acquired data. However, the enormous technological developments of recent decades have allowed the development of geo-instruments that produce high-resolution spectral information close to that obtainable from a polar instrument.

For this reason, only very recently have IASI-like satellite instruments for the study of chemistry monitoring been appearing on board geostationary platforms. The Chinese Geostationary Interferometric Infrared Sounder (GIIRS) is the first hyperspectral infrared sounder on board a geostationary satellite, namely the FengYun-4 series launched in 2016 (FY-4A) and 2021 (FY-4B) (). With its hyperspectral coverage (i.e. from 680 to 1130 cm-1 and from 1650 to 2250 cm-1) with a spectral resolution of 0.625 cm-1, it is able to provide important data for monitoring CO () or, for instance, the NH3 cycle (). GIIRS, however, is focused on the monitoring of a limited area over Asia.

The European agency EUMETSAT, on the other hand, has also envisaged putting an infrared sounder on board the geostationary Meteosat Third Generation – Sounding (MTG-S) platforms scheduled to be launched as part of the MTG satellite series (). This instrument, developed by Thales Alenia Space, is the Meteosat Third Generation Infrared Sounder (MTG-IRS) indeed (). This will provide data of the full Earth disc with acquisition every 30 min over Europe. Although initially designed for meteorological observations, MTG-IRS has ample potential to be exploited in atmospheric chemistry monitoring. It will acquire spectra in a long-wavelength infrared (LWIR) (679.70–1210.44 cm-1) and a medium-wavelength infrared (MWIR) (1600.00–2250.20 cm-1) spectral band with 0.603 and 0.604 cm-1 spectral sampling, respectively (https://user.eumetsat.int/resources/user-guides/mtg-irs-level-1-data-guide, last access: 2 January 2024).

Since it will cover a part of the infrared portion of the electromagnetic spectrum, MTG-IRS will be able to provide both daytime and nighttime data. This could potentially complement the information issued from UV–VIS sensors currently used for monitoring from GEO platforms ().

This research work is among the preliminary studies being carried out to prepare for the arrival of this powerful tool. Our main objective has been to assess the contribution of the data that MTG-IRS will soon provide for the characterisation of the atmospheric chemical composition over Europe through the assimilation of these data into a CTM. Such evaluation has been carried out by simulating MTG-IRS Level 1 (L1) data first and then performing their assimilation into MOCAGE (abbreviation from the French, Modèle de Chimie Atmosphérique de Grande Echelle).

Since this is a preliminary study, the focus is put on a single species which is among those that the instrument will be able to detect and which, on the other hand, plays an important role in the unfolding of many atmospheric processes, namely ozone.

The latter is a trace gas and a secondary species (i.e. not emitted) resulting from photochemical reactions. It is mainly found in the stratosphere (almost 90 %), resulting from photodissociation of oxygen. The stratospheric ozone layer is of paramount importance in shielding harmful ultraviolet solar radiation, and it makes life on Earth possible. On the other hand, a smaller percentage of ozone is found in the troposphere. It comes in small part from the stratosphere itself and in larger amounts from reactions in the lower layers involving primary compounds of natural or anthropogenic origin. Tropospheric ozone can cause issues to human health and the ecosystem, as well as to agriculture and material goods, due to its high oxidising power.

In the course of this paper, the characteristics of MTG-IRS (Sect. 2), those of the MOCAGE CTM (Sect. 3), and the data assimilation methods chosen and employed (Sect. 4) will be described in more detail. In Sect. 5, an extensive space will then be devoted to the necessary description of the construction of an observing system simulation experiment (OSSE). The choice to focus on the ozone species and to simulate and assimilate L1 data will be better justified. In Sect. 6 the results of the research will be presented, while conclusions and perspectives will be provided in the last section.

The information about MTG-IRS reported in this subsection is based on EUMETSAT, Thales Alenia Space and .

MTG-IRS is a Fourier transform spectrometer, built by Thales Alenia Space, which will be launched on board the geostationary MTG-S mid-2025. Once the instrument is operational, it will provide data of the full Earth disc with a 4 km spatial sampling at nadir.

The MTG-IRS scanning sequence will divide the Earth disc into four Local Area Coverage (LAC) zones, as in Fig. . Each LAC will be covered in 15 min through the acquisition of successive stares, called “dwells”, of about 10 s each. Each dwell will consist of 160 × 160 pixels. The total number of Earth dwells for the whole disc will be 280, while the number per LAC is listed in Fig. . LAC4 will be covered every 30 min, while the other LACs will be acquired in between.

The sounder will cover 1960 channels spread on two bands in the thermal infrared. In order to identify the spectral location of the two bands, they have been highlighted (in green) on an IASI spectrum in radiance units. Band 1 in the LWIR will be bounded in the range of 679.70–1210.44 cm-1, with spectral sampling of ∼ 0.603 cm-1, for a total of 881 channels. Band 2, in the MWIR, in the range of 1600.00–2250.20 cm-1 and with spectral sampling of ∼ 0.604 cm-1, will cover 1079 channels. The difference in the spectral sampling is due to the on-ground maximum optical depth (δmax), which is slightly different for the two bands: 0.829038 cm-1 for Band 1 (LWIR) and 0.828245 cm-1 for Band 2 (MWIR). The central wavenumber wn is computed as follows: 1wn=N2δmax, with 1127 ≤N≤ 2007 in the LWIR and 2650 ≤N≤ 3728 in the MWIR (Bertrand Théodore (EUMETSAT), personal communication, June 2021).

The instrument noise, provided by EUMETSAT in terms of the noise equivalent of differential temperature (NEΔT) at 280 K, is depicted in Fig.  (dash-dotted line). This noise can be converted into the corresponding scene temperature using the following formula: 2NEΔT(Tb)=NEΔT(280K)(∂Bν/∂T)(280K)(∂Bν/∂T)(Tb). Values of the noise in radiance units are in shown in Fig.  using a solid line.

MTG-IRS spectra will be distributed in the form of principal component (PC) scores. Its 1960 channel acquisitions will be compressed into 300 PC scores that preserve the vast majority of the information content for near-real-time applications. As this study aims to carry out a preliminary analysis, however, this step will be not mimicked. As better explained in Sect. , we chose to work with raw radiances in the ozone band.

The Modèle de Chimie Atmosphérique de Grande Echelle (MOCAGE) is a three-dimensional chemistry transport model that has been developed at the Centre National de Recherches Météorologiques (CNRM) since 2000 (e.g. ). It has been exploited in the last 2 decades for a wide range of operational and research applications. For instance, it has been used for several studies aiming to evaluate the climate change impact on atmospheric chemistry (e.g. ), as well as the trace gases transport throughout the troposphere (). Many efforts have been made to employ MOCAGE to investigate the exchanges taking place between the troposphere and stratosphere using data assimilation (e.g. ) or also to extend the representation of aerosols within the model simulations through aerosol optical depth (AOD) assimilation (e.g. ). The model is also a valuable resource for air quality monitoring and forecasting on the French PREV'AIR platform () and over Europe within the Monitoring Atmospheric Composition and Climate (MACC) project ().

At present, MOCAGE provides two geographical configurations, using a two-way nested-grid capacity (Fig. a):

GLOB11, a global scale with a 1° latitude × 1° longitude horizontal resolution;

For the vertical levels, MOCAGE uses σ–pressure vertical coordinates . Hence the model has a non-uniform vertical resolution: 47 vertical altitude–pressure levels from the surface up to 5 hPa. The levels are denser near the surface, with a resolution of about 40 m in the lower troposphere and 800 m in the lower stratosphere. A 60-hybrid-level version is also used in research mode, and it is also the one that has been exploited for the present work. This consists of the 47 levels computed as just described, plus 13 additional levels going up to 0.1 hPa. The resolution in the upper stratosphere is around 2 km. Please note that the MOCAGE vertical levels are numbered in descending order from the ground: the level closest to the ground is number 60, while the highest in the atmosphere is level 1. A schematic representation of the MOCAGE vertical configuration of the 60 hybrid pressure levels is provided in Fig. b, while an equivalence for the vertical level number and pressure can be found in Appendix A.

The model is able to simulate the chemistry in the lower stratosphere and troposphere, taking into account photochemical processes and the transport of longer-lived species in detail. It uses different chemical schemes in order to reproduce the atmospheric chemical composition: the Reactive Processes Ruling the Ozone Budget in the Stratosphere (REPROBUS) is used for the stratosphere (), while for the tropospheric representation the Regional Atmospheric Chemistry Mechanism (RACM) is exploited (). Through the combined use of the two aforementioned schemes (called RACMOBUS), MOCAGE is able to simulate 118 gaseous species, 434 chemical reactions, primary aerosols and secondary inorganic aerosols.

MOCAGE CTM runs in an off-line mode. Depending on the application, it can be coupled with a general circulation climate model (for climate studies) or with numerical weather prediction (NWP) models (e.g. for near-real-time applications). The core of the chemical reactions used in MOCAGE is also exploited on-line in the Integrated Forecast System (IFS) . In this study, MOCAGE has been used off-line and the meteorological forcing comes from Météo-France's operational global NWP model ARPEGE (Action de Recherche Petite Echelle Grande Echelle) , in a configuration with the two domains GLOB11 and MACC01.

The data concerning chemical emissions are provided to MOCAGE as external data sets. For the present operational configuration, MEGAN–MACC () and the Global Emissions Inventory Activity (GEIA) are used for biogenic emissions. MACCity (), RCP60 (), CAMS-REG-AP () and GEIA are those provided to cover information about anthropogenic emissions. For the representation of the biomass burning, data from the Copernicus Atmosphere Monitoring Service (CAMS) Global Fire Assimilation System (GFAS), from ECMWF, are exploited ().

The assimilation system used within MOCAGE was originally defined in the ASSET (Assimilation of Envisat) project (). It was jointly developed by CERFACS (Centre Européen de Recherche et de Formation Avancée en Calcul Scientifique) and Météo-France and has been further refined over the years. It has already been exploited for many studies on the assimilation of chemical data (e.g. ) and also on aerosol assimilation (e.g. ), on the exchanges between the troposphere and stratosphere (e.g. ), and in many other fields.

For this work we use the three-dimensional variational (3D-Var) method with an hourly assimilation window. The aim of the method is to look for the best representation of the atmospheric state or, in other words, the best compromise between the background model state and the observations. This is done by minimising the following cost function: 3J(x)=12(x-xb)TB-1(x-xb)+12[y-H(x)]TR-1[y-H(x)], where y is the vector of the observations, while xb and x are the a priori background and the model state vector, respectively. In the present case xb is obtained from a forecast from the previous assimilation. The state that minimises the cost function J(x) will then be defined as xa, i.e. the analysis state. H is the observation operator, which is usually a radiative transfer model (RTM) whose function is to transform a model state to a vector comparable to the observed radiances (or vice versa). In this work this function is covered by Radiative Transfer for TOVS (RTTOV) version 12 () in clear-sky conditions (the scattering effect of clouds and aerosols is not taken into account). Last but not least, the covariance matrices of the background and observation errors (B and R, respectively) are two essential components in the equation, since they allow each term to be given its proper weight.

In the context of this work, it is necessary to have good quality and reliable simulated spectral radiances in order to assess the impact of MTG-IRS assimilation into the CTM MOCAGE. Since MTG-IRS was not yet flying at the time the research was carried out, the strategy we adopted, and the one most commonly adopted in the atmospheric sciences in similar scenarios, was to perform an observing system simulation experiment (OSSE) (e.g. ). Such experiments imply a series of steps and validations to ensure that the observations are reliable and truthful and to put them to the test in a data assimilation system.

An OSSE basically consists of simulating synthetic observations from an atmospheric model state representing reality, assimilating them within another model state representing the model itself, and finally evaluating their impact on analyses and forecasts.

The reference reality is usually referred to as the nature run (NR). This consists of an atmospheric state that must realistically reproduce the true state of the atmosphere. Throughout the experiment, it serves as the state from which observations are simulated and as the reference against which the final assimilations are verified. It is usually produced using a good-quality model in a free-running set-up or without providing any information from real observations. In the present study, the approach is elaborated upon further and introduced in Sect. .

The NR reality is used to feed an observation simulator, which is usually an RTM through which the sought-after synthetic observations will be produced. Some specific instrumental proprieties have to be specified (such as optics, observational geometry, spatial and temporal resolution). The observations obtained through the simulator are, finally, perturbed with an instrumental error, to be properly assessed.

Another fundamental step in the development of an OSSE is the creation of the so-called control run (CR), which is a run of a model simulating the reality. The differences between the NR and CR should be those existing between the reality itself and the output of a good-quality model trying to reproduce it.

The synthetic observations will then be assimilated into the CR. This final run, referred to as the assimilation run (AR), is realised with the same configuration as the CR, by assimilating the synthetic observations created from the NR.

Finally, the impact of assimilation is assessed by comparing the results of the assimilation to the reference reality (i.e. the NR) and to the run without assimilation (i.e. the CR).

For an OSSE to be considered robust, it is very important that the CR is consistent with the NR but different enough to avoid the so-called “identical-twin” problem (e.g. ). This can be avoided by using different models to create the two scenarios or by sufficiently differentiating the inputs and the configurations in the same model so that the errors are properly represented and the outputs are consistent but divergent at the same time. In any case, the spatio-temporal variability in the NR must be properly evaluated against the CR before assimilation is carried out ().

The present research took place in the very current context of preparing for the launch of the new MTG Infrared Sounder (MTG-IRS), which will fly aboard the Meteosat Third Generation – Sounding (MTG-S) satellites in the next few years. As already discussed in detail within the paper, such an instrument will have considerable potential for application in the sphere of atmospheric chemistry monitoring and forecasting. It will cover two bands in the LWIR and MWIR, with remarkable spectral sampling (about 0.603 and 0.604 cm-1, respectively), combined with a high frequency of spectra acquisition (every 30 min over Europe) of the entire Earth disc from a geostationary platform.

The purpose of this work was, therefore, to assess the contribution of the assimilation of radiances that MTG-IRS will acquire and what this could bring to the characterisation of the atmospheric chemical composition (focus on the ozone) over Europe when assimilated into a chemistry transport model (CTM) such as MOCAGE.

In order to fulfil these purposes, realistic data of the instrument had to be simulated. The method chosen was the observing system simulation experiment (OSSE) approach. A detailed study was carried out in order to build a robust OSSE framework over the European domain. For both the nature run (NR – i.e. the reference reality) and the control run (CR), MOCAGE was chosen. This provides a global (GLOB11) and regional (MACC01) domain configuration. To have the regional domain, i.e. the one of interest for the project, the global has to run too, since MOCAGE is a nested-grid chemistry transport model. The research configuration with 60 vertical levels, up to 0.1 hPa, has been exploited. The time period chosen for the evaluation of the simulations ranged between 1 June till 31 August 2019.

Planning for the construction of a realistic NR to be different enough from the CR but consistent with it required significant research effort. For both runs, MOCAGE, forced by the same meteorological input from the operational ARPEGE, has been used. Therefore, a strategy had to be chosen to differentiate the runs. First, different emission configurations, referring to different years, have been set for the NR and CR frameworks. Additionally, another, more unconventional, choice was made in order to differentiate the reference reality and the run of control: L1c radiances from the IASI instrument were assimilated within the global MOCAGE configuration in the NR framework. The MACC01 domain was then used as the NR, which was forced at the edges by the global domain thus obtained but within which no assimilation was performed. In this case, the ability of MOCAGE to benefit from MTG-IRS observations was studied.

A study of inter-comparison has been performed between the NR and CR (which we remind readers are the simulations on the regional MACC01 MOCAGE domain). The noticeable differences that emerged have been estimated enough to avoid the identical-twin problem. The Ox total columns averaged over the time frame considered for the evaluation (from 1 June to 31 August 2019) showed a CR that always produces lower values than the NR, with a relative difference between 10.5 % and 13.5 %. The spatial distribution of the values was consistent for the two runs, as were the errors.

From the NR, MTG-IRS observations have been simulated, using RTTOV v12, for the 195 contiguous wavelengths bounded at 982.464 and 1099.467 cm-1, i.e. the range that had been confirmed by a priori sensitivity studies (not shown) to be the most sensitive to the ozone species and, thus, the most suitable for the purposes of this work.

Only clear-sky observations have been simulated (cloud filtering applied through information issued from the meteorological forcing). A horizontal thinning of the observation was also applied in order to make the ensuing assimilation reliable and efficient in terms of computational time (1 pixel simulated per box of 0.4°). These actions have produced a good observation density and frequency in the time period of interest.

Perfect observations thus created, have been perturbed with the addition of the MTG-IRS instrument noise provided by EUMETSAT.

The synthetic observations have been assimilated within the CR using a 3D-Var method with an hourly assimilation window. Background errors have been derived as a percentage of the ozone profile (2 % over the entire column). Observation errors, on the other hand, have been obtained by means of the so-called Desroziers diagnostics.

Statistics of residuals and innovations have been computed and verified and have been found to be correct, i.e. with analysis closer to observations than the background. Furthermore, higher values have been found at the edges of the domain than in the inner area, consistent with the coupling information coming from the lateral boundary conditions.

The contribution of MTG-IRS synthetic radiance assimilation has been assessed.

An analysis of the total columns of ozone averaged over the whole period of study showed that the assimilation of MTG-IRS radiances into MOCAGE always has a positive impact compared to no assimilation. The evaluated total columns averages, obtained from the AR of MTG-IRS, deviated from the NR very little where synthetic observations are more frequently assimilated. Values were mostly close to ∼ 0 %, with just a few areas touching ∼ 3 %. This represents a significant improvement over the CR case, which, instead, deviated from the NR by up to ∼ 13 %. Standard deviation of the bias existing between the AR of MTG-IRS with respect to the NR had its smallest values in the area where MTG-IRS radiances have been assimilated (∼ 1.8 DU).

When evaluating tropospheric columns, slightly stronger variations compared to the case of the total columns have been estimated, with values closer to 3% than to zero. The error in the differences compared to the NR is lower than for total columns (minima ∼0.3 DU), due also to the lower concentrations of the tropospheric ozone field.

Compared to what was obtained for the CR, where vertical variations were much more evident, the impact of assimilation on the whole vertical column is considerable, with values of up to ∼ 25 %.

From the results obtained and examined, many perspectives emerge.

The first thing to do in the very short term is to improve the assessment of the assimilation. It has been stated that the impact of coupling between MOCAGE domains produces more variable values of innovations and, thus, of the ozone field at the southern edge of the treated MACC01 regional domain. To ensure that the averages over the domain are not impacted by these values, the impact of assimilation should also be evaluated on a smaller area that cuts off these edges.

It is official that principal components (PCs) will only be distributed for MTG-IRS. The present study, however, did not take this subject into account. A short study evaluating the different impact of PCs on the work illustrated so far should be carried out.

Another point that arose from the study on the preparation of the NR framework has been the possibility of performing a channel selection while maintaining good-quality assimilation results. This is also a commonly used procedure in NWP for many instruments, which has been explored at CNRM for MTG-IRS () and in the context of another future hyperspectral infrared instrument, IASI-NG (IASI New Generation) (). Since the work carried out here on MTG-IRS was a first analysis, we wanted to investigate the behaviour of all the consecutive wavelengths considered. Moreover, using the full spectral band was a fair way of comparing the impact of MTG-IRS. At a later stage, however, it will be of interest to carry out a selection of a smaller group of channels too.

If we consider assimilating MTG-IRS into the global domain of MOCAGE, the joint assimilation of MTG-IRS with GIIRS (a hyperspectral IR sounder on board Chinese geostationary satellites) could lead to interesting validation. At the same time, on board the MTG-S satellites, the ultraviolet–visible–near-infrared imaging spectrometer Sentinel-4 will also fly close to MTG-IRS. This is designed to monitor some key air quality trace gases and aerosols over Europe with a high spatial resolution and fast revisit time, in support of CAMS. Given the potential of the joint acquisition of the two instruments, their synergy in the characterisation of atmospheric pollution over Europe should definitely be tested.

The joint assimilation of MTG-IRS radiances with data acquired by other spectrometers working in a similar way, though on polar-orbiting platforms, would also be interesting to consider. First on the list are certainly IASI and IASI-NG (as soon as the latter's data are exploitable) on board European satellites. Other IR sounders of the US and Chinese polar systems (CrIS and HIRAS) could also be added.

Since MTG-IRS provides good vertical information in the stratosphere, upper troposphere–lower stratosphere (UTLS) and free troposphere, surface data could also bring additional details with respect to the lowermost atmospheric layers and then be assimilated as a complement.

Finally, to be addressed as a continuation of this work, there will certainly be the evaluation of the assimilation of radiances in the bands sensitive to other chemical species and aerosols. The OSSE framework created for this study was designed to provide, with minor modifications, the basics for extension to species other than ozone. Sulfur dioxide (SO2) could be investigated, since MTG-IRS has the potential to sense this species, with a sensitivity towards the end of the MWIR band (). A more specific case study, however, should be taken into account for this compound, since it is associated with specific and punctual natural events. Carbon monoxide (CO) is, however, going to be the first target, given the results encountered during the study performed to assess the sensitivity of MTG-IRS in its two versions.