The upper troposphere lower stratosphere (UTLS) region plays a critical role in regulating the Earth's climate. Ozone (O3) and water vapour (H2O), which are among the most effective greenhouse gases, exert their largest radiative effects in the upper troposphere and around the tropopause . Changes in the UTLS composition therefore have a direct impact on surface temperature . One of the key processes shaping the distribution of these gases is the Brewer-Dobson circulation (BDC) – the large-scale circulation system that spans the middle atmosphere between roughly 10 and 80 km altitude. Climate models consistently predict that the BDC will accelerate in response to global climate change , with significant implications for the distribution of trace gases in the UTLS . This includes increased exchange across the tropopause, potentially injecting O3-rich stratospheric air into the upper troposphere .

To better understand these processes, amongst others, the Changing Atmosphere Infra-Red Tomography Explorer (CAIRT) was proposed to the European Space Agency (ESA) as a candidate for Earth Explorer 11 (EE11), the ESA's Earth observation flagship missions. CAIRT was down-selected for Phase 0 pre-definition studies between 2021 and 2023 and then subject to Phase A feasibility studies until mid-2025. Although CAIRT was not selected for implementation in the EE11 competition, the mission concept was assessed as scientifically and technically mature, and could still be implemented under another ESA programme or through international collaboration.

CAIRT is designed to employ a limb-imaging Fourier Transform Spectrometer (FTS) to observe the atmosphere from 5 to 115 km of altitude in the thermal infrared spectral domain, providing observations of temperature, O3, and H2O, among numerous other trace gases, with unprecedented spatial resolution relying on a tomographic retrieval approach. If implemented, CAIRT would improve our understanding of how changes in the BDC influence the distribution of O3 and H2O in the UTLS by observing small-scale structures and gradients of these gases across the tropopause.

Moreover, CAIRT UTLS H2O data would be highly relevant for numerical weather prediction (NWP) systems. Many NWP models exhibit a cold temperature bias in the tropopause region caused by long wave radiative cooling associated with a moist bias in this region . The origin of this moist bias has been linked to excessive numerical diffusion in model transport schemes around the tropopause . Recent updates to the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecast System (IFS) have reintroduced water vapour data assimilation profiles – particularly those from the Microwave Limb Sounder (MLS) – showing significant improvement in temperature forecasts .

The present study has two goals. First, it aims to assess whether CAIRT measurement requirements are adequate to support the CAIRT scientific objectives targeting UTLS ozone and water vapour, with emphasis on providing scientific and societal added value via improved constraints on these radiative forcing gases. Second, it aims to compare CAIRT's capabilities with those of the existing MLS instrument, which serves as one of the main references for O3 and H2O observations in the UTLS . This assessment is performed using an Observing System Simulation Experiment (OSSE).

OSSEs are typically designed to investigate the potential impacts of prospective observing systems using data assimilation (DA) techniques . They are set up using several numerical experiments: the nature run (NR), the control run (CR), and at least one assimilation run (AR). The nature run corresponds to a model simulation defining the true state of the atmosphere. It is used to simulate observations with the instrument of interest and, if needed, with other existing observing systems used in data assimilation. The control run, being either a free model run or a data assimilation run using only simulated observations of existing instruments, provides the baseline atmospheric state and serves to evaluate the impact of the new instrument. The assimilation run includes the new observations in addition to those assimilated in the CR, if any. The comparison between AR-NR and CR-NR differences quantifies the added value of the new instrument. Additional assimilation runs can also be included to test alternative instrument configurations. To avoid the identical-twin problem, which can lead to over-optimistic results, assimilation and control runs should employ a different model than the nature run. This ensures that improvements in the AR relative to the CR are primarily due to the assimilation of new observations. Although OSSEs have traditionally been used to evaluate meteorological satellite systems, several studies have also applied this technique to instruments measuring atmospheric composition .

This paper is organized as follows. The concept of CAIRT and its capabilities to observe the UTLS composition are briefly introduced in Sect. . The nature run is presented in Sect. . The simulated observations of CAIRT and MLS are described in Sects.  and , respectively, followed by a discussion comparing the performance of both instruments in Sect. . The control and assimilation runs are presented in Sect. . Results are provided in Sect. , followed by the conclusions.

This section provides a concise overview of the CAIRT mission concept; further details are available in and in the CAIRT Report for Mission Selection . CAIRT aims to observe temperature and trace gases from infrared emission spectra of the Earth's atmosphere measured from a satellite in limb-viewing geometry. The instrument concept is based on infrared imaging, allowing for simultaneous vertical coverage (approximately 5–115 km altitude) and horizontal across-track coverage (a swath width of about 300–500 km, depending on design trade-offs). Each pixel of the imaging detector provides an interferogram within a finite acquisition time, which can be processed into a line-of-sight limb emission radiance spectrum (Level-1 data).

By acquiring a new image every ∼ 7.5 s, arrays of limb line-of-sight spectral radiance measurements would be acquired approximately every 50 km along the satellite track, enabling tomographic retrieval. Spectra from several pixels are typically binned to achieve sampling of about 1 km in the vertical and finer than 50 km in the across-track direction around the tangent point. Furthermore, to facilitate potential synergies with nadir instruments, CAIRT tangent points are designed to be co-registered with MetOp second generation (SG) nadir instruments, particularly with the Infrared Atmospheric Sounding Interferometer New Generation (IASI-NG), allowing both satellites to sample approximately the same air masses simultaneously.

The CAIRT spectral range is specified in the long- to mid-wave infrared regions (718–2200 cm⁻¹) with a spectral resolution better than 0.25 cm⁻¹ (before apodisation), enabling the retrieval of temperature and numerous trace gases like O3, H2O, CH4, nitrogen and sulfur species, CFCs, SF6, and others seeFig. 4.11 for an example of simulated CAIRT spectra. In the UTLS, CAIRT O3 observational requirements are specified with a vertical resolution better than 2 km, a horizontal (across-track, ACT) resolution better than 100 km, and a total uncertainty better than 30 ppbv in the upper troposphere and better than 150 ppbv between the lower stratosphere and 30 km altitude (see Table ). For H2O, the specified spatial resolution is similar, with a total uncertainty better than 10 %. These requirements are designed to ensure sensitivity to stratosphere–troposphere exchange processes, such as stratospheric intrusions. This is illustrated in Fig. a for ozone. The figure shows ozone at 7 km during a stratospheric intrusion which spans from Scandinavia to Algeria. Average ozone at 7 km is around 50 ppbv and exceeds 100 ppbv in the stratospheric intrusion. An uncertainty of better than 30 ppbv is thus reasonable if one wants to observe such kinds of intrusions. Similarly, if horizontal gradients of water vapour at 12 km like those illustrated in Fig. e are to be observed, an uncertainty of better (thus smaller) than 10 % seems reasonable.

The Level-1 limb radiance spectra corresponding to multiple across-track spatial samples are averaged to reduce noise before Level-2 retrievals see Sects.  and , andSect. 7.2.3. For O3 and H2O in the UTLS, the spectra are binned across-track to get a spatial resolution of 100 km. Considering a default swath of 400 km, CAIRT would thus retrieve L2 profiles along four orbit tracks spaced out by 100 km for O3 and H2O in the UTLS (this can change for other species or altitude region).

CAIRT builds upon the strong space heritage of infrared limb emission measurements established by the Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) instrument onboard Envisat, which operated between 2002 and 2012 . MIPAS was a FTS scanning instrument with several co-aligned detectors, covering different spectral ranges, providing one retrieved profile approximately every 400 km. Given its 50 km ALT resolution and its four tracks, CAIRT would retrieve ∼ 32 times more profiles than MIPAS, and ∼ 12 times more profiles than MLS which provides one profile every 167 km along the orbit track (see Sect. ). If implemented, CAIRT would be the first infrared limb imager in space, building on the heritage of the Gimballed Limb Observer for Radiance Imaging of the Atmosphere (GLORIA), which has flown on balloons and aircraft since 2011 . Experience with the airborne demonstrator instrument GLORIA provides confidence that the estimated performance of the CAIRT concept are realistic.

In this study, a CAMS control simulation is used as the nature run providing the synthetic truth state for ozone, water vapour and cloud. It is based on the Integrated Forecast System for COMPOsition (IFS-COMPO) cycle 47r3. The meteorological upgrade implemented in cycle 47r3 is described in , while the chemical scheme is summarized in and references therein. This cycle became operational in October 2021 but had been running as an experimental control simulation since March of that year.

Unlike the operational CAMS analyses, the control simulation does not assimilate either chemical or meteorological observations. However, its meteorology – including the specific humidity – is reinitialized every 24 h from the operational CAMS analyses at 00:00 UTC. The simulation employs a reduced Gaussian grid at TL511 horizontal resolution (approximately 40 km) with 137 vertical levels, providing a vertical resolution of about 400 m at 50 hPa, 300 m in the UTLS, and getting finer toward the surface.

In IFS-COMPO 47r3, ozone is represented using a detailed chemistry scheme in the troposphere and a linearized parameterization in the stratosphere . For water vapour, we use the specific humidity field provided by IFS-COMPO. The control simulation was selected because its output is archived every 3 h, compared to every 6 h for the operational analyses. In this configuration, ozone is not constrained by any observations, whereas water vapour is indirectly constrained through the daily reinitialization of specific humidity from the CAMS analyses, within which this variable is assimilated in the troposphere.

Ozone from the CAMS control simulation is routinely evaluated against a wide range of observational datasets by the CAMS Evaluation and Quality Control (EQC) service see, e.g.,and https://atmosphere.copernicus.eu/eqa-reports-global-services, last access 13 November 2025. Following the CAMS EQC methodology, ozone from the CAMS control simulation is interpolated at the profile geolocation of ozonesondes and a set of limb-sounding instruments: the Atmospheric Chemistry Experiment Fourier Transform Spectrometer (ACE-FTS) version 5.3 , MLS version 5.0 , the Ozone Mapping and Profiler Suite Limb Profiler (OMPS-LP) onboard the S-NPP satellite, version 2.6 , and the Stratosphere Aerosol and Gas Experiment III onboard ISS (SAGE-III/ISS) version 6.0 . Statistical differences between CAMS interpolated profiles and these observations are displayed in Fig.  in five latitude bands for the period October–December 2021 (see Fig. ). The rationale for selecting this period is provided in the following paragraphs. The statistical indicators shown are the normalized mean bias (NMB) between CAMS and the observations, the associated normalized standard deviation (NSD), and the correlation coefficient (CORR) between CAMS and the observations. NMB and NSD use the mean values of CAMS for normalization.

In the stratosphere (above 100 hPa), except over the southern polar region, the agreement between the CAMS control simulation and both ozonesonde and satellite observations is generally good. The mean bias remains within ±10 %, the standard deviation of the difference is below 10 %, and the correlation typically exceeds 0.8. In the troposphere, the agreement is less good: comparisons with ozonesondes reveal positive biases between +10 % to +20 % in the tropics and biases within ±10 % at mid-latitudes. The standard deviation increases notably in the tropics, although the correlation remains between 0.5 and 0.8. While ozone from the operational CAMS analysis exhibits better agreement due to the assimilation of multiple satellite ozone retrievals, the quality of the control simulation is considered sufficient for the purposes of this study.

For specific humidity, major upgrades to the moist-physics parameterizations and the assimilation of satellite observations were introduced in IFS-COMPO cycle 47r3 , reducing the stratospheric temperature bias by up to 50 %. Using the same evaluation methodology as for ozone, specific humidity from the CAMS control simulation was compared with measurements from two limb-sounding instruments (see Fig. ). Because the stratospheric water vapour content increased by approximately 10 % following the Hunga eruption in mid-January 2022 , and since this source of water vapour is not represented in CAMS, the comparison was restricted to the October-December 2021 period.

In the stratosphere (above 100 hPa), the mean bias in specific humidity is positive, between 0 % and +10 %, with a standard deviation below 10 %. In the troposphere, both the mean bias and standard deviation increase with pressure, reaching approximately +30 % and 50 %, respectively, at 250 hPa. The correlation remains relatively high above 50 hPa and below 200 hPa, but decreases to around 0.5 between these pressure levels, except in the tropics near 50 hPa where it approaches zero. Nevertheless, as for ozone, the quality of the control simulation is deemed adequate for use in this study.

The CAMS control simulation is archived in the ECMWF MARS system, and the corresponding MARS parameters required for data retrieval are provided in the Code and data availability section at the end of this paper.

Simulating CAIRT Level-2 (L2) profiles from the nature run involves several steps, illustrated in Fig. . First, the satellite orbit must be simulated, i.e. the satellite position and velocity vectors are determined at a frequency consistent with the along-track sampling. Second, given the satellite position and velocity and the CAIRT viewing direction, the CAIRT retrieval grid is computed, providing the geolocation of the points observed by CAIRT. Third, the nature run is interpolated at the observed points, yielding the NR@L2 dataset.

The performance characteristics of the L2 products, namely the averaging kernels and observational uncertainties, must also be simulated (step 4). Subsequently (step 5), the simulated L2 profiles are generated by convolving NR@L2 with the L2 performance characteristics. To account for the potential saturation of Level-1 (L1) spectra caused by the presence of clouds, cloud information is incorporated at this stage.

Because the number of CAIRT profiles is significantly larger than that of any existing limb-sounding instrument, direct assimilation into the BASCOE-EnKF system would lead to memory overflow. This is a well-known issue in data assimilation (DA), for which a common solution is the generation of superobservations (often referred to as superobbing in DA terminology), obtained by combining observations available at the same time within the same model grid cell.

All these steps are described in detail in the following subsections.

The computation of MLS simulated profiles differs from that of CAIRT profiles in that some of the required data are already available. This includes the geolocation of MLS profiles, the a priori profiles, the L2 error profiles, and the vertical averaging kernels. The MLS geolocation, a priori profiles, and L2 error profiles are obtained from MLS v5 daily observation files, while the MLS v5 vertical averaging kernels are provided for five latitudes and are independent of time and longitude.

The CAMS nature run is interpolated onto the MLS retrieval grid using BASCOE-MAO, as for CAIRT (see Sect. , and is then perturbed to produce the L2 profiles yl2 as follows: 6yl2=Ayn+(I-A)ya+ϵp Here, yn represents the nature run interpolated onto the MLS grid; ya denotes the existing MLS a priori profiles corresponding to the MLS geolocation; A is the MLS vertical averaging kernel matrix; and I is the identity matrix. Note that MLS along-track averaging kernels have not been taken into account. The perturbation profile ϵp is a Gaussian random profile multiplied by the MLS L2 error profile provided in the MLS data files.

Simulated profiles are only assimilated within the vertical range of validity specified in . This report also defines criteria for accepting or discarding profiles based on the variables “estimated precision,” “quality,” “convergence,” and “status” provided in the MLS files. These criteria were used to flag entire profiles or portions of profiles, discarding measurements contaminated by clouds, particularly for O3 (see discussion in the next section).

As previously mentioned, CAIRT will provide substantially higher sampling than MLS (or MIPAS) for O3 and H2O in the UTLS, delivering approximately 12 times more profiles. This is illustrated in Figs.  and e. Figure  compares various performance metrics for CAIRT and MLS: (a) random error, (b) systematic error, (c) vertical resolution, (d) horizontal resolution, and (e) number of available observations. For CAIRT performances discussed in Sect. , the displayed values correspond to January in the tropics. This provides a conservative view of the CAIRT performances compared to other latitudes (not shown). MLS performances are taken from the MLS v5 data quality document and are generally independent of latitude and season, except for the number of observations .

For O3, CAIRT and MLS exhibit comparable random errors, whereas for H2O, MLS shows slightly smaller (i.e., better) random errors (Fig. a). For O3, CAIRT systematic errors exceed 50 % below 7 km in the tropics; consequently, CAIRT O3 observations below this altitude have not been assimilated. The vertical resolutions of both instruments are comparable for O3, while MLS generally exhibits better vertical resolution in the stratosphere for H2O (Fig. c). For both species, CAIRT achieves superior horizontal resolution (Fig. d). The number of available observations for January is shown in Fig. e; the decrease with decreasing altitude is due to the presence of clouds.

Figure  compares CAMS ozone with the corresponding L2 simulated observations for CAIRT and MLS along an orbit on 15 December 2021. Note that CAIRT and MLS orbits differ in time and position; therefore, this comparison provides only a qualitative indication of their differences. In the stratosphere, CAIRT and MLS observations closely match the original CAMS data, whereas in the troposphere the agreement is weaker. In the tropics, a large portion of CAIRT data below 15 km is rejected due to clouds, whereas MLS is less affected. This difference arises from the measurement techniques, as microwave observations (MLS) are less sensitive to clouds than infrared (CAIRT). Conversely, CAIRT provides more valid observations than MLS at mid and high latitudes. Similar patterns are observed for H2O (see Fig. ).

Control and assimilation experiments were conducted with the BASCOE system , driven by ERA5 winds and temperatures. The simulations employed a horizontal grid of 91 latitudes by 144 longitudes (approximately 250 km horizontal resolution) and 51 vertical levels, with a vertical resolution of about 1.2 km at 50 hPa, 1 km in the UTLS, and decreasing toward the surface. This spatial resolution is considerably coarser than that of the nature run (see Sect. ).

In the stratosphere, BASCOE, similarly to CAMS, uses a linearized ozone chemistry scheme, although the two systems are constrained by different climatological data. In the troposphere, BASCOE applies the same linearized chemical formulation as in the stratosphere, whereas CAMS employs a comprehensive tropospheric chemistry scheme (see Sect. ).

For the experiments presented in this study, BASCOE does not account for any chemical processes involving water vapour. To prevent an unrealistic transport of tropospheric water vapour into the stratosphere, condensation is approximated by limiting the water vapour partial pressure to the saturation vapour pressure of water ice . Consequently, the BASCOE and CAMS configurations and simulations exhibit substantial differences in their treatment of both O3 and H2O.

Six BASCOE experiments were performed, as summarized in Table . The first is a control run (CR) using the BASCOE Chemistry Transport Model (CTM) without data assimilation. The five other experiments are all assimilation runs using the BASCOE Ensemble Kalman Filter (ENKF). The first assimilation run is constrained by the simulated MLS profiles (AR_MLS). The additional assimilation runs were conducted using different simulations of CAIRT superobservation profiles. One experiment assumes a 400 km swath (AR_CAIRT), which corresponds to the default CAIRT configuration adopted during the Phase A study, another experiment assumes a 300 km swath (AR_C300km), representing the threshold value for CAIRT swath. Since the simulated MLS profiles do not include any systematic error, an additional assimilation run was performed with CAIRT simulated superobservation profiles with a 400 km swath where the contribution of systematic errors in the L2 profiles was excluded. Finally, to assess the potential impact of clouds, a fifth assimilation run (AR_Cnocloud) was carried out using the simulated CAIRT superobservation profiles with a 400 km swath, excluding the contribution of systematic error and cloud filtering.

The control run was initialized from the nature run on 15 August 2021, whereas all assimilation runs were initialized on 15 September 2021 from the control run. All simulations are evaluated against the nature run for the period from October 2021 to February 2022, allowing for a 15 d spin-up period for each assimilation experiment.

CAIRT data help to improve the representation of ozone and water vapour, as illustrated in Fig. . The ozone distribution is shown at 7 km during a tropopause fold event above Europe, which is well represented by both the AR_CAIRT and AR_MLS assimilation runs, effectively correcting the bias present in the control run. One can also see the strong constraint provided by CAIRT observations, which substantially – though not completely – reduces the negative bias of the control run outside the tropopause fold region, and to a greater extent than MLS data. Similarly, CAIRT observations also provide a significant constraint, comparable to that of MLS, in the representation of water vapour at 12 km.

This study evaluated the added value of ozone and water vapour profiles from the proposed satellite limb instrument CAIRT, designed to observe the temperature and composition of the middle atmosphere. The evaluation was conducted through an Observing System Simulation Experiment (OSSE) using an IFS model simulation as the nature run and the BASCOE system to perform the control and assimilation runs. CAIRT observations were simulated through a comprehensive processing chain starting from the IFS model fields as the synthetic truth, and accounting for CAIRT's orbit, geographical coverage, cloud contamination, spatial and spectral resolution, error budget, and superobservation strategy. For comparison, MLS profiles of ozone and water vapour were also simulated from the same IFS synthetic truth and assimilated by BASCOE, to assess the relative added value of CAIRT against an existing limb sounder.

For ozone, the OSSE results show that CAIRT can provide valuable information down to the mid-troposphere. Compared with the nature run, the BASCOE control run exhibits a positive bias above the tropopause and a negative bias in the troposphere exceeding ±50 %. Assimilating either MLS or CAIRT observations reduces the bias to within ±2 % in the lower stratosphere.

In the troposphere, MLS assimilation significantly reduces the control run bias, but a notable residual remains (−5 % in the extra-tropics to more than −35 % in the tropical lower troposphere). CAIRT data provide a stronger constraint, reducing the bias below −5 % in the extra-tropics and below −15 % in the tropics. Additional runs indicate that the remaining bias in the tropical troposphere is due to the presence of cloud. Furthermore, reducing the CAIRT swath from 400 to 300 km has little impact, suggesting that a narrower swath could be implemented if required by technical constraints without significant loss of performance.

For water vapour, CAIRT and MLS provide useful information in the lower stratosphere, reducing the bias and standard deviation to un-significant values. In the upper troposphere, while the bias and standard deviation remains significant (e.g. larger biases than -30 % for both instruments in the tropics), the improvement against the control run is large in all statistical indicator. As for O3, reducing the swath to 300 km does not significantly affect CAIRT's ability to constrain the UTLS. Overall, these results indicate that CAIRT could deliver high-quality water vapour profiles suitable for constraining the UTLS region providing invaluable data to improve NWP systems.

This OSSE therefore demonstrates that CAIRT data are capable of meeting the mission's scientific objectives in the UTLS, particularly by improving the characterization of vertical gradients and small-scale structures of ozone and water vapour across the tropopause. Such capabilities directly support CAIRT's overarching goal of advancing our understanding of the coupling between atmospheric composition, circulation, and climate.