Consistency and quality assessment of the Metop-A / IASI and Metop-B / IASI operational trace gas products ( O 3 , CO , N 2 O , CH 4 and CO 2 ) in the Subtropical North Atlantic

Introduction Conclusions References

Abstract.This paper presents the tools and methodology for performing a routine comprehensive monitoring of consistency and quality of IASI (Infrared Atmospheric Sounding Interferometer) trace gas Level 2 (L2) products (O 3 , CO, N 2 O, CH 4 , and CO 2 ) generated at EUMETSAT (European Organisation for the Exploitation of Meteorological Satellites) using ground-based observations at the Izaña Atmospheric Observatory (IZO, Tenerife).As a demonstration the period 2010-2014 was analysed, covering the version 5 of the IASI L2 processor.Firstly, we assess the consistency between the total column (TC) observations from the IASI sensors on board the EUMETSAT Metop-A and Metop-B meteorological satellites (IASI-A and IASI-B respectively) in the subtropical North Atlantic region during the first 2 years of IASI-B operations (2012IASI-B operations ( -2014)).By analysing different timescales, we probe the daily and annual consistency of the variability observed by IASI-A and IASI-B and thereby assess the suitability of IASI-B for continuation of the IASI-A time series.The continuous intercomparison of both IASI sensors also offers important diagnostics for identifying inconsistencies between the data records and for documenting their temporal stability.Once the consistency of IASI sensors is documented we estimate the overall accuracy of all the IASI trace gas TC products by comparing to coincident ground-based Fourier transform infrared spectrometer (FTS) measurements performed at IZO from 2010 to 2014.The IASI L2 products reproduce the ground-based FTS observations well at the longest temporal scales, i.e. annual cycles and long-term trends for all the trace gases considered (Pearson correlation coefficient, R, larger than 0.95 and 0.75 for long-term trends and annual cycles respectively) with the exception of CO 2 .For CO 2 acceptable agreement is only achieved for long-term trends (R ∼ 0.70).The differences observed between IASI and FTS observations can be in part attributed to the different vertical sensitivities of the two remote sensing instruments and also to the degree of maturity of the IASI products: O 3 and CO are pre-operational, while N 2 O, CH 4 , and CO 2 are, for the period covered by this study, aspirational products only and are not considered mature.Regarding shorter timescales (single or daily measurements), only the O 3 product seems to show good sensitivity to actual atmospheric variations (R ∼ 0.80), while the CO product is only moderately sensitive (R ∼ 0.50).For the remainder of the trace gases, further improvements would be required to capture the day-to-day real atmospheric variability.

Introduction
Continuous, consistent, and high-quality long-term monitoring of the composition of the atmosphere is fundamental for addressing the challenges of climate research.In this context space-based remote sensing observations are of particular importance, since they are unique in providing a global coverage.Among the current space-based remote sensing instruments, IASI (Infrared Atmospheric Sounding Interferometer, Blumstein et al., 2004) has special relevance since it combines high quality (very good signal-to-noise ratio and high spectral resolution), good horizontal resolution (12 km at nadir), global coverage, and long-term data availability.Its mission is guaranteed until 2022 through the meteorological satellites Metop, the space component of the EUMETSAT (European Organisation for the Exploitation of Meteorological Satellites) Polar System (EPS) programme: the first sensor (IASI-A) was launched in October 2006 on board Metop-A, the second (IASI-B) was launched in September 2012 on board Metop-B, and the third (IASI-C) is expected to be launched in October 2018 aboard Metop-C.A successor to IASI, IASI-NG (Crevoisier et al., 2014), with improved spectral resolution and radiometric performance is under development as part of the EPS-SG (Second Generation) programme and will continue the mission after Metop-C and extend the data record by 2 decades.All these features make the IASI missions very promising for monitoring atmospheric composition in the long term as a key instrument for the EU-METSAT Earth observation programme (e.g.Clerbaux et al., 2009;Crevoisier et al., 2009a, b;Herbin et al., 2009;Schneider and Hase, 2011;August et al., 2012;Kerzenmacher et al., 2012).However, for correct scientific use of these long-term observational records, an assessment of the consistency of the atmospheric observations from the IASI sensors currently in orbit, as well as a documentation of their quality is required.To date there has been no comprehensive consistency and validation study for all the trace gas products disseminated by EUMETSAT as IASI Level 2 (L2) products.Such activities have been mostly performed in the context of short campaigns or have been focused on specific atmospheric parameters (e.g.Keim et al., 2009;Viatte et al., 2011;Schneider and Hase, 2011;García et al., 2013).By such campaigns alone it is not possible to extensively evaluate the quality of the different IASI atmospheric products as well as the potential of IASI for long-term climate studies.In order to address these two critical tasks, high-quality ground-based reference data sets are needed.
While there are several techniques for measuring total column (TC) amounts of atmospheric trace gases such as water vapour or ozone that can be used as a validation reference (e.g.radiosondes, UV-VIS spectrometers), there is currently only one technique that routinely estimates all the atmospheric trace gases retrieved operationally from IASI measurements (ozone, O 3 , carbon monoxide, CO, nitrous oxide, N 2 O, methane, CH 4 , and carbon dioxide, CO 2 ): the ground-based Fourier transform spectrometers (FTSs).FTS instruments record very high-resolution infrared solar absorption spectra and use a similar measurement approach as IASI.By evaluating these solar spectra, the FTS systems can provide TC amounts and volume mixing ratio (VMR) profiles of many different atmospheric trace gases with high precision.Within the NDACC (Network for Detection of Atmospheric Composition Change, www.acom.ucar.edu/irwg/)such FTS experiments are operated at about 25 sites distributed worldwide.Since the 1990s, when these instruments began to be used for atmospheric composition monitoring, there have been continuous efforts to assure and even further improve the high quality of the FTS data products: monitoring the instrument line shape (e.g.Hase et al., 1999;Hase, 2012), monitoring and improving the accuracy of the solar trackers (e.g.Gisi et al., 2011), and developing and intercomparing sophisticated retrieval algorithms (e.g.Hase et al., 2004).The good quality of these long-term ground-based FTS data sets has been extensively documented by theoretical and empirical validation studies (e.g.Schneider et al., 2006Schneider et al., , 2008;;Sepúlveda et al., 2012;García et al., 2012b;Sepúlveda et al., 2014;Barthlott et al., 2015).
In this context this work intends to demonstrate monitoring capabilities for the IASI L2 atmospheric composition products, to help identifying potential areas for retrieval algorithm improvement, and to support the related ongoing and future development activities.For this purpose, it is firstly analysed the consistency between the IASI-A and IASI-B L2 atmospheric trace gas products provided by the IASI mission (TC amounts of O 3 , CO, N 2 O, CH 4 , and CO 2 ) in the subtropical North Atlantic region after the 2 first years in operation of IASI-B (2012IASI-B ( -2014)).Secondly, the documentation of the overall IASI quality is addressed by using the high-precision FTS observations that have been carried out at the Izaña Atmospheric Observatory (IZO) since 1999.Due to its strategic location, IZO is affected by background free troposphere air masses with very distinct history and regions of origin (Atlantic Ocean, Europe, North and Central Africa, etc; see Cuevas et al., 2013, and references therein), making IZO a unique place for documenting the quality of IASI atmospheric products for different scenarios.To address all these tasks, this paper is structured as follows: Sect. 2 presents the EUMETSAT Metop/IASI mission describing the IASI sensor as well as the IASI L2 atmospheric trace gas products routinely disseminated by EUMETSAT, while Sect. 3 introduces the ground-based FTS products, including the FTS activities at IZO and the retrieval strategies used for obtaining the different FTS products.Section 4 presents the strategy developed for comparing IASI-A, IASI-B, and FTS observations and Sects.5 and 6 address the consistency and intercomparison study at different timescales (single measurements, daily, annual, and long-term trends).Finally, Sect.7 summarises the main results and conclusions of this work.(August et al., 2012).To do so, IASI records thermal infrared emission spectra of the Earthatmosphere system in the 645-2760 cm −1 region (apodized spectral resolution of 0.5 cm −1 ) with a surface swath width of about 2200 km twice per day.2. IASI L2 V5 introduces significant improvements in the retrieval of the atmospheric trace gas products as well as cloud products and cloud detection in contrast to the previous version, V4.Now, under cloud-free conditions, the O 3 profiles are simultaneously retrieved, together with the humidity and temperature profiles and the surface temperature, from the IASI measured radiances using an optimal estimation method (Rodgers, 2000).This approach uses a global a priori with a single unique covariance matrix, computed from a collection of ECMWF (European Centre for Medium-Range Weather Forecast) analysis records, independent on seasonal or latitudinal variations.Therefore, all observed atmospheric variability comes from the measurements rather than the a priori information (August et al., 2012).
The TC amounts of the other molecules (CO, N 2 O, CH 4 , and CO 2 ) are retrieved using an inversion algorithm based on artificial neural networks (ANNs).The feasibility of retrieving those quantities with ANNs from IASI measurements was first studied prior to launch (Turquety et al., 2004) and this formed the basis for the IASI L2 processors until revision V4.The method was refined in V5 specifically for CO with an updated channel selection and the addition of new predictors adding important information about the surface characteristics and the viewing geometry (August et al., 2012).The ANNs were trained with simulated radiances using the RT-TOV model (Matricardi and Saunder, 1999;Saunders et al., 1999) and the atmospheric composition profiles from the MOZART model (Brasseur et al., 1998).N 2 O, CH 4 , and CO 2 benefit from the same improvements in the ANNs design introduced for CO in V5, but they have not been specifically optimised and validated.They are only distributed as aspirational products while scientific development is ongoing and research products from the wider community (Clerbaux et al., 2009;Crevoisier et al., 2013) can be made operational.
The cloud screening strategy is key for the space-based atmospheric trace gas retrievals.For IASI L2 products, the cloud detection relies on five different cloud detection tests, based on the combined information of the measured IASI L1C spectra and of other remote sensors flying with IASI (AVHRR, AMSU, and ATOVS) as well as of comparison to synthetic radiances.An IASI pixel is flagged cloudy if at least one of the cloud tests detects the presence of clouds.For more details, refer to August et al. (2012) and the Products User Guide (EUM/OPSEPS/MAN/04/0033, EUMETSAT).
3 Ground-based FTS programme at the Izaña atmospheric observatory and FTS products The IZO, run by the Izaña Atmospheric Research Centre (IARC, http://izana.aemet.es)belonging to the Spanish Meteorological Agency (AEMET), is a subtropical highmountain observatory on Tenerife (Canary Islands, 28.3 • N, 16.5 • W, and 2373 m a.s.l.; Fig. 1), and only 350 km away from the African continent.It is usually well above the level of a strong subtropical temperature inversion layer, which acts as a natural barrier for local pollution.This fact, together with the quasi-permanent subsidence regime typical of the subtropical region, makes the air surrounding the observatory representative of the background free troposphere (particularly at night-time) (Gómez-Peláez et al., 2013;Cuevas et al., 2015, and references therein).These conditions are only significantly modified by episodes of mineral desert dust exports during summertime, when Saharan dust long-range transports within the so-called Saharan Air Layer (Prospero et al., 2002) are quite common (e.g.García et al., 2012a, and references therein).During wintertime, the Saharan events rarely reach the IZO altitude, since they are confined below 2 km of altitude (Díaz et al., 2006;Alonso-Pérez et al., 2011, and references therein).
Within the IZO's atmospheric research activities, the ground-based FTS measurements started in 1999 and continue until the present with two Bruker IFS spectrometers (an IFS 120M from 1999 to 2005 and an IFS 120/5HR from 2005 until present).These activities have contributed to the international networks NDACC and TCCON (Total Car- bon Column Observing Network, www.tccon.caltech.edu)since 1999 and 2007 respectively.The FTS experiment at IZO records highly resolved infrared solar absorption spectra from 740 to 9000 cm −1 .However, in order to be consistent with the spectral range covered by IASI, for this study we only work with those measured in the middle infrared spectral region, i.e. between 740 and 4250 cm −1 (corresponding to the standard NDACC measurements).These solar spectra are acquired at an apodized spectral resolution of 0.005 cm −1 .Table 1 lists the main FTS characteristics and highlights the main differences and similarities between IASI and these ground-based FTS instruments.For further details about the FTS instrument at IZO refer to Schneider et al. (2005); Sepúlveda et al. (2012); García et al. (2012b, and references therein).
The high-resolution FTS solar absorption spectra allow an observation of the pressure broadening effect and thus the retrieval of trace gas VMR profiles.In this work the target gas VMR profiles are retrieved using the algorithm PROFFIT (PROFile FIT; Hase et al., 2004) which follows the formalism given by Rodgers (2000).Then, TC amounts are computed by integrating the retrieved VMR profiles from the FTS altitude (2373 m a.s.l. for IZO) to the top of the atmosphere.For all the target gases considered we use a nearly identical retrieval strategy, which is summarised in Table 3.The target gas VMR profiles are retrieved using specific microwindows, also taking into account the absorption signatures of those trace gases interfering with the target gas (see Table 3).For O 3 and N 2 O, the retrieval has also been refined by including a simultaneous temperature fit, which significantly reduces the error (Schneider et al., 2008;O. E. García et al., 2014).For this purpose, we add additional micro-windows containing well-isolated CO 2 lines.
Table 3. Description of the FTS retrieval setups used in this work.The profile retrieval is done by using a Tikhonov-Phillips slope constraint (TP1), while the scale retrieval corresponds to a scale to the a priori VMR profile.A summary of the error estimation is also provided: theoretical total systematic (SY) and statistical (ST) errors for the FTS total column products (median, 1σ × 10 −1 , being σ the standard deviation of the error distributions between 2010 and 2014).XN 2 O, XCH 4 , and XCO 2 correspond to the total column-averaged dry air mole fractions of N 2 O, CH 4 , and CO 2 respectively.In order to reduce the interference error due to water vapour (the main interfering gas), we firstly perform a prefit of H 2 O and in a second step simultaneously perform the target gas profile retrieval and a scale retrieval of all the interfering species considered.For all the target gases, the VMR profiles are retrieved on a logarithmic scale using an ad hoc Tikhonov-Phillips slope constraint (TP1 constraint), with the exception of CO 2 , which is scaled to the a priori profile on a linear scale.The a priori profiles are taken from WACCM (Whole Atmosphere Community Climate Modelversion 6, http://waccm.acd.ucar.edu)provided by NCAR (National Center for Atmospheric Research; J. Hannigan, private communication, 2014), averaged between 2008 and 2014 (period of IASI data).As a priori temperature we use the NCEP (National Centers for Environmental Prediction) 12:00 UTC daily temperature profiles.Note that only the a priori temperature profiles are updated daily.For all the target gases the a priori information is always kept constant; i.e. it does not vary on a daily or seasonal basis.Therefore, similarly to IASI, all the observed variability directly comes from the measured FTS spectra.Regarding spectroscopy, the spectroscopic line parameters are taken from HITRAN 2008database (Rothman et al., 2009) including 2009 and 2012 updates (www.cfa.harvard.edu/hitran)for all the gases except for CH 4 .For CH 4 we use a preliminary line list provided by D. Dubravica and F. Hase, obtained within a current project of the Deutsche Forschungsgemeinschaft, IUP Bremen, DLR Oberpfaffenhofen, and KIT, which has demonstrated lower spectroscopic residuals than the HITRAN 2008 linelist (Dubravica et al., 2013).
The FTS spectra are only recorded when the line of sight between the instrument and the Sun is cloud free.However, to avoid possible contamination of thin clouds, the FTS observations are, in a second step, filtered according to colocated global solar radiation observations taken at IZO in the framework of the Baseline Solar Radiation Network (BSRN, http://bsrn.awi.de).By using a cloud detection method on the coincident solar radiation measurements (based on Long and Ackerman, 2000, and adapted for IZO by R. D. García et al., 2014), the cloud-free periods in the FTS records are easily identified.Once the FTS retrievals are computed, they are filtered in a third step according to (i) the number of iterations at which the convergence is reached and (ii) the residues of the simulated-measured spectrum comparison.This final step ensures that unstable or imprecise FTS retrievals can be considered (which could likely be introduced by remaining thin clouds).
It is important to remark that the FTS products used here contain further refinements over the standard NDACC approaches (NDACC/Infrared Working Group, IRWG, www.acom.ucar.edu/irwg/)and, thus, they do not correspond to the FTS products publicly available at the NDACC archive.Refer to the references given in Table 3 for further details about the specific retrieval strategies used in this work.
Theoretically, the error of the different FTS products can be estimated by following the formalism detailed by Rodgers (2000) where three types of error can be distinguished: the smoothing error associated with the limited vertical sensitivity of the FTS instruments, the errors due to uncertainties in the input/model parameters (instrumental characteristics, spectroscopy data, etc), and the measurement noise.Using the error estimation as provided by PROFFIT and assuming the error sources and values listed in Appendix A, where the theoretical error estimation of the FTS products is detailed, the total statistical errors for the FTS TC products range from 0.2 to 0.6 %, while the systematic error is between 2 and 4 % (Table 3).
At IZO, other different high-quality measurement techniques for monitoring atmospheric trace gases are available (Cuevas et al., 2015).By using those data, a continuous em-pirical documentation of the quality and long-term consistency of our FTS products has been carried out since the FTS instrument was installed at IZO in 1999.The FTS precision obtained from these experimental studies for the trace gases considered here is also listed in Table 3, showing a rather good agreement between theoretical and experimental errors.

Temporal decomposition
The consistency and quality assessment of IASI-A and IASI-B products is addressed at different timescales: single measurements, daily, annual, and long-term trends.This temporal decomposition provides an added value for validating trace gases with a rather small variability, such as N 2 O or CO 2 .For such gases the uncertainty is often larger than the day-today concentration variations and thus a validation at longer temporal scales is more meaningful than a validation limited to a comparison of individual measurements.Moreover, this analysis allows us to quickly detect instrumental issues or inconsistencies.For this purpose, we follow the procedure proposed by Sepúlveda et al. (2014) (and references therein), explained in detail in the following.Firstly, for analysing the time series on different timescales the measured TC time series of each target gas ([TC] gas ) is fitted to a time series model, which considers a mean [TC] gas value and [TC] gas variations on two different timescales (see Eq. 1): a linear trend and intra-annual variations.
where t is measured in years, f o is a baseline constant, and f trend the linear trend in change per year.The annual cycle is modelled in terms of a Fourier series where a i and b i are the parameters of the Fourier series to be determined and ω i = 2π i/T with T = 365.25 days.Once the model fit is computed, the seasonal variations are obtained by subtracting the fitted linear trends from the measured time series.The averaged annual cycle is then computed by averaging these de-trended time series on a monthly basis.It represents the de-trended multi-annual seasonal cycle of the target gas.In addition to the seasonal timescale we look on measurement-to-measurement and long-term timescales.For the separation into these two timescales we use the aforementioned time series model.The measurement-to-measurement timescale signal is calculated as the difference between the measured time series and the modelled time series (whereby all fitted timescales are considered: mean value, linear trend, and seasonal cycle).The so-calculated de-trended and deseasonalised time series represents the very short-term variations, corresponding to the variations among individual observations.Finally, in order to calculate the long-term timescale signal (annual means) we reconstructed a time series that only considers the fit results obtained for the mean [TC] gas and the seasonal cycle.Then, by subtracting it from the measured time series, we get a de-seasonalised time series, for which we then calculate the annual mean values.Note that for IASI-A and IASI-B consistency study, the bias between both IASI sensors is also calculated.To do so, we directly compare the measured TCs of all the trace gases and compute the median difference of this difference time series.This temporal decomposition has been done on a logarithmic scale, i.e. our measured time series correspond to the logarithm of the measured TCs of all the trace gases.This approach has two clear advantages in the subsequent IASI-FTS comparison: (i) the [TC] gas variations on this scale can be interpreted as variations relative to the reference mean values ( ln[TC] gas ≈ [TC] gas /[TC] gas ) and we thereby directly compare the anomalies observed by both remote sensing instruments, and (ii) the relative differences between IASI and FTS observations can directly be computed as the subtraction of the corresponding variability on the different timescales (note that the temporal decomposition produces values very close to zero, thereby computing the standard relative differences provides very extreme values in some cases).

Collocation criteria
IASI and ground-based FTS instruments are sensing areas of different size and the acquisition times are generally not exactly simultaneous.In addition, both instruments have different vertical sensitivities.All these features have to be taken into account in the definition of the comparison strategy that, on the one hand, ensures the representativeness of the reference FTS data and, on the other hand, accounts for the spatial and temporal variability of the trace gases considered.The collocation criteria selected are the result of a compromise between the spatial and temporal variability of each trace gas and the uncertainties and spatial range covered by the FTS observations and therefore can vary from gas to gas.Appendix B describes in detail the methodology followed to define the optimal coincidence criteria adopted for each trace gas, summarised in Table 4.In summary, we consider all the IASI observations within the box ±1 • latitude/longitude centred at the IZO location (Fig. 1) and pair those IASI and FTS observations taken within ±1 h for O 3 and CO, and we pair daily median observations corresponding to the same day for the rest of trace gases.As previously mentioned, we consider all the TC observations disseminated in V5: September 2010-September 2014 for IASI-A and December 2012-September 2014 for IASI-B.From this data set we only work with the best quality IASI L2 V5 measurements: over sea, cloud free, and with the highest level of quality and completeness of the IASI retrieval as indicated in the Products User Guide (EUM/OPSEPS/MAN/04/0033, EUMETSAT).
For the consistency study between both IASI sensors (IASI-A and IASI-B) the 2 • square has been divided in boxes of ±0.25 • .Moreover, we distinguish between daily morning (10:00-11:00 UTC) and daily evening (22:00-23:00 UTC) TC observation overpasses in order to carefully analyse a possible bias for each overpass.Then, we compare TC amounts from each overpass and sensor for the same day.Note that for this study the IASI observations are paired without forcing temporal coincidence with the FTS data.

Vertical sensitivity
The vertical sensitivity of a remote sensing spectrometer depends, among other things, on the geometry of observations, the target gas considered as well as the specific characteristics of each instrument (e.g. the signal to noise ratio, spectral resolution).Hence, the responses of IASI and FTS to real atmospheric variability are significantly differ-ent.This fact can be observed in Fig. 2, where the rows of the IASI and FTS averaging kernels (A) are displayed for O 3 and CH 4 .The IASI A are not operationally disseminated in V5; therefore, to illustrate its vertical sensitivity, we have taken the O 3 A from the EUMETSAT/IASI L2 version 6 products (EPS Product Validation Report: IASI L2 PPF v6, EUM/TSS/REP/14/776443 v4C, EUMETSAT), while the IASI CH 4 A have been obtained in the framework of the European project MUSICA (Schneider et al., 2013).Note that the rows of A describe the altitude regions that mainly contribute to the retrieved VMR profile and therefore these kernels can be used to identify the independent layers without significant overlap with other layers.Indeed, the trace of A (so-called the degrees of freedom for signal, DOFS) is a measure of the number of independent layers retrieved from the remote sensing measurements.Also, Fig. 2 includes the vertical profiles of the cumulative DOFS, calculated from the top of the atmosphere to surface for IASI and inversely for FTS as well as the a priori VMR used for the FTS retrievals in order to compare with the vertical sensitivity of the two remote sensing instruments.
Figure 2 illustrates two relevant facts that have to be taken into account when comparing IASI and FTS observations.First, IASI has a lower number of DOFS than FTS and the maximum sensitivity within these detected layers is located at different altitudes.Therefore, the TC amounts observed by both instruments could differently reflect the atmospheric composition variability.For example, when retrieving CH 4 , IASI only detects one CH 4 layer (DOFS ∼ 1), located in the upper troposphere/tropopause region (∼ 12-14 km), while the FTS system detects two independent CH 4 partial columns (DOFS ∼ 2.5), corresponding to the troposphere and the stratosphere.Similar conclusions might be derived for N 2 O and CO (see, for example, Fig. 2 of Kerzenmacher et al., 2012).For O 3 , the difference between the vertical sensitivities is not so significant and IASI is expected to be sensitive to the maximum O 3 concentrations in the Chapman layer and to the tropopause/upper troposphere regions (DOFS ∼ 2.5).The expected IASI DOFS and the obtained FTS DOFS for all the trace gases are also listed in Table 4.
The second important fact is that IASI has a weak sensitivity in the lower troposphere for the trace gases considered in this study, leading to the variability of the partial columns missed by FTS below 2373 m a.s.l.(IZO altitude) not being crucial for the IASI-FTS comparison.

Consistency between IASI-A and IASI-B observations
In order to probe the continuity provided by IASI-B as well as the consistency of each individual IASI sensor, it is indispensable to first analyse the temporal stability of their observations.For this purpose, we examine possible drifts and discontinuities in the times series of the differences between the de-seasonalised variability from IASI-A and IASI-B averaged on a weekly basis.The drift is defined as the linear trend in the differences, while the change points (changes in the weekly median of the difference time series) are analysed by using a robust rank order change-point test (Lanzante, 1996; O. E. García et al., 2014).By using these tools, we observe that the time series of the differences between the observations from the morning and evening overpasses for each IASI sensor as well as between the observations from the two sensors for each overpass are homogenous for all the trace gases considered (i.e.no change points were detected at 95 % confidence level).Moreover, all the difference time series reveal no significant drifts at 95 % confidence level.Figure 3 shows the time series of O 3 TC as observed by IASI-A and IASI-B and the corresponding differences.This temporal stability study is complemented by analysing whether the distributions of the IASI-A and IASI-B observations could be statistically considered equivalent.To do so, we have used the Friedman non-parametric test, which detects differences in the distributions of related variables by checking the null hypothesis that multiple dependent samples come from the same statistical population (Sawilowsky and Fahoome, 2005).By applying this test on the observed short-term variability time series from the two IASI sensors for each overpass, which can be considered as four related samples for each trace gas, we only observe significant differences for the O 3 distributions between evening and morning overpasses (at 95 % confidence level).Nonetheless, these discrepancies disappear when comparing the observations for the same overpass; therefore, the IASI sensors distinguish the O 3 intra-day concentration variations.Indeed, for this trace gas the agreement between both sensors for the same overpass is significantly better than between the two overpasses for each sensor, as observed in Fig. 4, which displays the scatter plots between the O 3 de-trended and de- seasonalised variability as observed by the two IASI sensors for each overpass.
All of these findings suggest that, on the one hand, the observations from each sensor are consistent with themselves and, on the other hand, both sensors similarly reproduce the atmospheric composition variations.The statistics for the IASI-A and IASI-B intercomparison are summarised in Fig. 5 (Pearson correlation coefficient and standard deviation of the de-trended and de-seasonalised differences, and median bias).In summary, we observe that both IASI sensors similarly reproduce the annual cycle of all the trace gases considered (R > 0.95 except for CO 2 , for which we observe a poorer agreement, R ∼ 0.70-0.85),while for the very short-term concentration variations we find a large correlation for O 3 (∼ 0.80) and moderate for the rest of trace gases (R ∼ 0.30-0.60).The scatter (1σ ) of the differences among sensors and overpasses, on a measurement-to-measurement basis, is less than 10 % for CO, ∼ 2 % for O 3 and between 1 and 2 % for the rest of the trace gases, while it decreases when comparing annual cycles: 2 % for CO, less than 0.5 % for the rest of trace gases.Regarding the biases between the IASI sensors (IASI-A-IASI-B), the values of ∼ −3 % for  CO and between 0.4 and 0.6 % for N 2 O and CH 4 are remarkable (bottom panel in Fig. 5).For O 3 and CO 2 the bias is lower than 0.2 % in absolute value.

Comparison between IASI and ground-based FTS observations
This section presents the IASI-FTS comparison at different timescales: single measurements, daily, annual, and longterm trends.An example of this strategy is displayed in Fig. 6 for O 3 , while the summary of the intercomparison results for all the trace gases is shown in Fig. 7 (Pearson correlation coefficient and standard deviation of the differences between IASI and FTS products).The consistency between both IASI sensors has been documented in the previous section.Therefore, here we only focus on IASI-A since it has the longest time series of measurements (September 2010-September 2014 for V5) as well as to ensure a homogeneous sampling during the whole period analysed.
As observed in Fig. 7, IASI reproduces the ground-based FTS observations well at the longest temporal scales, i.e. annual cycles and long-term trends.For the latter the correlation is larger than 0.95 for all the trace gases with the exception of CO 2 (R ∼ 0.70), while on an annual basis, the correlation is larger than 0.95 for O 3 and CO and between 0.75 and 0.85 for CH 4 and N 2 O.The discrepancies found for the annual cycles (amplitude and phase) can be explained by the different sensitivity of IASI and ground-based FTS instruments.As observed in Figs.6c and 8, for O 3 and CO the IASI and FTS annual cycles are completely in phase, but the peak-to-peak amplitude is slightly different.For O 3 the largest differences are observed during spring-early summer due to the missing sensitivity of IASI in the troposphere and in part in the tropopause region (recall Fig. 2), while for CO we observe that IASI tends to overestimate the variability observed by the FTS.For CH 4 and N 2 O the results are very similar: we observe that IASI products follow the annual shift of the tropopause altitude, where the maximum IASI sensitivity is located, and while FTS also reflects that annual shift, it is also sensitive to the tropospheric and stratospheric CH 4 and N 2 O variations.As a consequence, the annual cycles observed by both remote sensors are slightly out of phase.For CO 2 the annual cycles are not correlated either in phase or in amplitude.This is likely due to the algorithm used in  (d, e) time series of the de-trended+de-seasonalised variability (in %), and the difference between coincident de-trended+de-seasonalised variability from IASI-A and FTS (in %) respectively; and (f) scatter plot of the de-trended + de-seasonalised variability.For (c) and (f) the Pearson correlation coefficient is included in the legends and for (e) the standard deviation of the differences.The number of coincident measurements is 2338.the IASI L2 processor which has not been specifically optimised for CO 2 , since the middle infrared FTS CO 2 products have successfully proven their reliability to monitor the CO 2 concentrations (Barthlott et al., 2015).In particular, the IASI CO 2 retrieval is solely based on the IASI measurements unlike in Crevoisier et al. (2009a), where collocated microwave measurements are exploited together with IASI data to disentangle the temperature and the CO2 signals in the thermal infrared spectra.
For the shortest-term variations we find poorer agreements, although the correlation is significantly larger for O 3 (∼ 0.80), but not for the rest of trace gases (R ∼ 0.10-0.30).When comparing daily values for CO the agreement improves (R ∼ 0.50), suggesting that the IASI sensor could moderately capture the day-to-day concentration variations but not the intra-day variability.The scatter (1σ ) of the differences between IASI and FTS observations, which can be used as a conservative estimate of the IASI uncertainties, is less than ∼ 10 % for CO, ∼ 3 % for O 3 , and between 1 and 2 % for the rest of the trace gases.All of them are within the target uncertainties of the IASI mission (recall Table 2) and seem to be good enough to capture the day-to-day variations for O 3 and CO when comparing to those observed by FTS records at IZO (values also included in Fig. 7 as a reference).The uncertainties reported here agree well with previous validation studies of IASI L2 V5 operational products using different measurement platforms.For example, errors below 5 % have been reported for O 3 (e.g.Viatte et al., 2011;August et al., 2012), between 10 and 15 % for CO (e.g.August et al., 2012), and ∼ 2 % for N 2 O and CH 4 (García et al., 2013).Note that IASI-FTS comparison also confirms the results observed for the consistency study of IASI-A and IASI-B sensors.The correlations and the scatter of the differences observed for both analysis are very similar both at short-term and intra-annual timescales (recall Figs. 6 and 8) with the exception of CO 2 .For this trace gas the consistency study reveals a moderate agreement between both IASI sensors (correlations between 0.6 and 0.8 for the annual cycles), but we do not document any agreement for the IASI-FTS comparison (correlation less than 0.2).This is likely due to the degree of maturity of the IASI CO 2 products, as aforementioned.Therefore, the continuous intercomparison of both IASI sensors could successfully be used as a quality control for identifying inconsistencies or instrumental issues in lack of reference ground-based observations.
The differences between the IASI and FTS short-term variability have been analysed as a function of the different parameters, such as the relative horizontal distance between IASI footprints and FTS location, or of the different viewing geometry of the two remote sensing instruments, without identifying significant patterns.As example, the study of viewing geometry is displayed in Fig. 9, where we observe that the differences are uncorrelated to the viewing geometry (IASI-A air mass (Liou, 1980) and difference between IASI-A and FTS AMS).Only the differences for O 3 are displayed as example.However, the same behaviour has been observed for all the trace gases considered.
In addition, the strategic location of IZO allows us to address the IASI-FTS comparison for different atmospheric conditions.Figure 9 also shows the IASI-FTS differences as a function of the aerosol optical depth (AOD) records.This aerosol parameter characterises the total extinction in the line of sight due to atmospheric aerosols and, thus, it can be used as a tracer of Saharan desert air masses in the subtropical North Atlantic region (e.g.García et al., 2012a).To identify these Saharan events, we have considered the AOD observations performed at the Santa Cruz de Tenerife (SCO) station from AERONET network (Holben et al., 1998).SCO, managed by AEMET, is also located on Tenerife, but at sea level (≈ 50 m of altitude) and very close to the sea.Therefore, the SCO AOD records can better quantify the mineral dust outbreaks occurring both in winter (below 2 km of altitude) and in summer (between 2 and 6 km of altitude).Note that although the winter events rarely reach IZO, they will affect the IASI observations including the surrounding ocean.As observed in Fig. 9, during summer, the differences between IASI and FTS observations seem to be affected by the range of the aerosol load, increasing (absolute value) as the corresponding AOD values increase.This pattern is likely due to the different type of observations: while the FTS measurements are performed from the ground in the direct solar path, the IASI sensors record thermal emission of the Earth-atmosphere from the space and could be more affected by aerosol signatures (thermal emission and scattering processes) (e.g.Vandenbussche et al., 2013;Peyridieu et al., 2013, and references therein).Indeed, when comparing the observations from the two IASI sensors (also displayed in Fig. 9) this difference disappears, which is expected given the very good spectral and radiometric consistency of the two instruments.However, in winter, the IASI-FTS differences seem to be independent on AOD.This fact is likely due to the limited sensitivity of IASI to the boundary layer because of decreasing thermal contrast between the surface and the atmosphere when approaching the surface.In addition to the Saharan conditions, we have also analysed the IASI-FTS comparison under polluted air masses likely coming from North America or Europe (Cuevas et al., 2013, and references therein) by using the intra-day CO concentration variations from the GAW (Global Atmospheric Watch) in situ observations recorded at IZO as a tracer (see Appendix B for details about GAW programme at IZO).No significant patterns were observed (data not shown), but further analysis and longer time series are needed to extract more robust conclusions.
Until now, the IASI-FTS comparison has been addressed in terms of relative variability, but a comparison of absolute TC amounts also provides us useful information.Therefore, to roughly estimate possible biases between IASI-A and FTS observations, the partial column amounts below IZO altitude, computed from the WACCM climatological data, have been added to the FTS observations.By using those data, we find that the IASI observations are consistently lower than FTS observations for all the trace gases, with a median bias (IASI-FTS) of ∼ −6 % for O 3 and CH 4 (−6.4 and −5.9 % respectively) and ∼ −12 % for N 2 O and CO 2 (−12.4 and −12.1 % respectively), except for CO.For CO we observe the contrary behaviour; i.e. the IASI sensor overestimates the FTS observations by ∼ 15 % (15.2 %).These discrepancies could be partly attributed to systematic IASI and FTS error caused, for example, by the lower IASI sensitivity to the lower troposphere, uncertainties in the IASI ANNs training procedure or in the spectroscopic line parameters.As previously mentioned in Sect.3, the errors in the spectroscopic line parameters could explain between 2 and 4 % of the FTS bias, according to our error estimation (see Appendix B).Indeed, experimental intercomparisons between FTS and Brewer observations carried out at IZO in the last years found that FTS systematically overestimates the Brewer O 3 TC amounts by ∼ 4 % (Schneider et al., 2008;Viatte et al., 2011;García et al., 2012b), which may be due to inconsistencies in the ultraviolet and infrared spectroscopic parameters.This implies that the IASI O 3 ob-servations should have less bias with respect to the actual O 3 concentrations (less than 2 %).However, for CO, the bias obtained is likely introduced by the WACCM estimation since previous studies comparing to other space-based CO products, like the MOPITT sensor (August et al., 2012), or to dedicated IASI CO retrievals, like the FORLI-CO algorithm (Kerzenmacher et al., 2012), found biases lower than 7 % for regions with low background CO concentrations such as the Atlantic Ocean.

Conclusions
This paper documents, for the first time, the uncertainty and the long-term consistency of all EUMETSAT/IASI trace gas products at the same time and using a unique measurement technique as reference, the ground-based FTS experiment.
Firstly, we show that the EUMETSAT/IASI trace gas observations, from both Metop-A/IASI and Metop-B/IASI, are consistent; i.e. neither drifts in time nor were biases found.Therefore, the observations from both remote sensing instruments could be merged to obtain a unique IASI database.Secondly, we focus on the IASI versus ground-based FTS measurement comparison.IASI adequately captures the dayto-day TC variation, the annual cycle, and the long-term trend for its operational products O 3 and CO, as compared to ground-based FTS measurements.Likewise, for N 2 O and CH 4 (trace gases with a rather small day-to-day variability), IASI observations can successfully be used to describe their seasonality and interannual trend.However, for CO 2 an acceptable agreement is only achieved at the long-term scale.For the latter three gases, disseminated as aspirational products, improvements in the EUMETSAT retrieval algorithms are currently being carried out to include mature algorithms from the wider scientific community.FORLI-CO and FORLI-O 3 are also being included in the operational IASI L2 suite (Clerbaux et al., 2009).The same methodology can be applied to the upgraded products to support their monitoring in the long term and in view of the expectation of a continued data record from IASI on Metop-C.
This consistency and quality assessment has been carried out using the ground-based FTS located at the IZO and, thus, it is valid for the subtropical North Atlantic region under free troposphere conditions.Although this quality documentation can be used as a benchmark for studies that apply EU-METSAT/IASI trace gas products in climate research, further comparison studies covering other regions might be desirable in order to analyse the possible impact of latitude or other environments, such as urban-industrial or biomass burning areas, on the IASI products accuracy.
Finally, this paper highlights the potential of ground-based FTS experiments once again as an indispensable reference for validating the current space-based observations as well as those anticipated from the next generation of satellite sensors.Theoretically, the error of the different FTS products can be estimated by following the formalism detailed by Rodgers (2000), where the difference between the retrieved state, x, and the real state, x, can be written as a linear combination of the a priori state, x a , the real and estimated model parameters, b and b respectively and the measurement noise :

Atmos
where G represents the gain matrix, K b a sensitivity matrix to model parameters, I the identify matrix, and A the averaging kernel matrix.A relates the real variability to the measured variability of the considered atmospheric state and, thus, represents the way in which the remote sensing system smoothes the real vertical profiles (Rodgers, 2000).Therefore, Eq. (A1) defines three types of error: the first term is the smoothing error associated with the limited vertical sensitivity of the FTS instruments, the second one represents the errors due to uncertainties in the input/model parameters (instrumental characteristics, spectroscopy data), and the third one corresponds to the measurement noise).
The theoretical error estimation strongly depends on the assumed uncertainties.In our case, we consider the error sources and values listed in Table A1 for the input parameters, which are the leading error sources affecting the different FTS products, identified from our experience and the literature (Schneider and Hase, 2008;Sepúlveda et al., 2014, and references therein), while the smoothing error is calculated as (A − I)S a (A − I) T , where S a matrix is the is the covariance matrix of the target gas.Strictly, to estimate the smoothing error contribution, the covariance matrix of a real ensemble of atmospheric states must be known (Rodgers, 2000).However, due to lack of real observations of the vertical profiles of all the trace gas considered at IZO, the S a for each target gas is assumed and calculated from the WACCM-V6 model estimates.WACCM is a global chemistry model of well-recognised prestigious that has widely demonstrated its ability to provide reliable estimations of the vertical profiles of trace gases and their expected concentration variations (Pan and Brasseur, 2006;SPARC Report, 2010;Smith et al., 2011;Brakebusch et al., 2013).Therefore, here the S a is calculated considering the variance of the corresponding gas concentrations at each altitude from the WACCM-V6 climatological data and a Gaussian distribution of strength 5 km for the inter-layer correlation.Note that the total error values are calculated as the root sum squares of all the error sources considered, where the contribution of each error source has been split into statistical and systematic contributions.The exceptions are the spectroscopic parameters and the measurement noise, which are considered as purely systematic and statistical respectively.This error estimation has been applied to the IZO FTS observations between 2010 and 2014 (period studied in the current work).
Table A1.Error sources used for the theoretical error estimation for all the FTS products (chann.: channeling; eff.: efficiency; err.: error; int.: intensity; ν-scale: spectral position; S: intensity; γ : pressure broadening parameter).The second column gives the assumed error value and the third column the partitioning of this error between statistical (ST) and systematic (SY) contributions (Sepúlveda et al., 2014).The FTS total errors (statistical and systematic) depend on the observing geometry at which FTS observations are carried out.As illustrated in Fig. A1, the larger theoretical errors are found at high solar zenith angles (SZAs), mainly due to the fact that the FTS observations are more sensitive to possible misalignments of the solar tracker at these SZAs.Therefore, these data (SZA > 75 • ) are excluded from the study to avoid unrealistic FTS retrievals in the FTS-IASI intercomparison, which represent between 1 % for CO and 8 % for N 2 O, CH 4 , and CO 2 .Considering the filtered FTS observations, the total statistical errors (medians and ±1σ ) are 0.40 ± 0.03 % for O 3 , 0.50 ± 0.01 % for CO, 0.20 ± 0.03 % for N 2 O, 0.30 ± 0.03 % for CH 4 , and 0.60 ± 0.14 % for CO 2 .The major error sources for the tropospheric gases (N 2 O, CH 4 , CO 2 , and CO) are the baseline uncertainties and the measurement noise, while the uncertainties in the FTS's instrumental line shape (described by the modulation efficiency and the phase error) dominate the statistical errors for the stratospheric gas O 3 .For all the target gases, the systematic error budget is lead by the spectroscopic errors, with median values and ±1σ of 2.00 ± 0.01 % for O 3 , 2.10 ± 0.04 % for CO, 2.10 ± 0.01 % for N 2 O, 2.35 ± 0.01 % for CH 4 , and 3.50 ± 0.04 % for CO 2 .

Appendix B: Collocation criteria between IASI and ground-based FTS observations
To define the temporal collocation, we first estimate the intraday concentration variations of the target gases and, then, analyse whether the FTS system is good enough to detect this variability by comparing to the respective FTS uncertainties.To do so and to be independent from the FTS observations, we use the high-frequency and high-quality data routinely measured by different in situ analyzers and Brewer spectrometers at IZO.
Ground-level in situ atmospheric continuous measurements of CO 2 (since 1984), CH 4 (since 1984), CO (since 2008), and N 2 O (since 2007) have been routinely carried out at IZO as a contribution of AEMET to the WMO GAW programme (Cuevas et al., 2015).The high quality of these measurements has been externally assessed by ( 1 The intra-day concentration variations have been estimated through the intra-day variation coefficient (VC), cal- culated as the daily standard deviation divided by the daily mean of the corresponding observations, considering daytime Brewer measurements for O 3 and night-time GAW hourly mole fractions means for the rest of the gases (20:00-08:00 UTC).Note that the IZO in situ night-time data represent the background regional signal of the free troposphere well, while during daytime thermally driven up-slope flow from maritime boundary layer can reach the station and, thus, the in situ data are not well suited for comparing to remote sensing observations.By considering the available time series of these observations since the IASI data are operationally disseminated, i.e. 2008-2013 (see Fig. B1 for CH 4 and O 3 ), we have estimated the following typical intra-day VC (medians and ±1σ ): 0.63 ± 0.52 % for O 3 , 2.90 ± 2.65 % for CO, 0.08 ± 0.02 % for N 2 O, 0.29 ± 0.17 % for CH 4 , and 0.05 ± 0.07 % for CO 2 .When comparing experimental and theoretical FTS uncertainties (recall Table 3) to these values, we observe that for CO and O 3 the intra-day VC is larger or much larger than the FTS uncertainty and, therefore, the individual FTS measurements (or hourly medians) should be considered to ensure optimal temporal collocation.Likewise, for CH 4 , N 2 O, and CO 2 , daily medians of the FTS observations can be taken without losing information or affecting the validation results (FTS uncertainty is larger than or comparable to the typical intra-day VC).Note that the remaining concentration variations within the defined temporal window have to be considered when performing the IASI-FTS intercomparison.For CO the hourly VC, considering only nighttime observations, is estimated to be 0.98 ± 1.43 %, while for O 3 it is 0.31 ± 0.37 %.As for the temporal collocation, the spatial coincidence criteria have to take into account the spatial concentration variations of each trace gas and the maximal horizontal distance covered by the FTS observations.Since the FTS measurements are performed in the direct solar path, the horizontal projection of the air masses probed by the FTS can be easily calculated from the actual solar observing geometry and the effective altitude of the vertical column observed by the FTS.The latter has been defined as the altitude at which 95 % of the corresponding TC amount is observed and, thus, varies from gas to gas.These effective altitudes have been determined by using the WACCM-V6 climatological data and are ∼ 40 km for O 3 and ∼ 20 km for the rest of gases, resulting in a maximal horizontal distance of ∼ 150 km for O 3 and ∼ 80 km for the rest of gases (see Fig. B2).
Regarding the spatial concentration variations of each trace gas, we should consider that IZO is far away from the target gas sources/sinks and embedded in the free troposphere, thereby usually affected by long-range transports of aged and well-mixed air masses (e.g.Cuevas et al., 2013, and references therein).In addition, the latitudinal/longitudinal gradients of the trace gases considered here are rather smooth at oceanic subtropical latitudes.Indeed, the latitudinal relative difference of CO 2 between the Equator and 60 • N in 2012 was smaller than 2.5 % (WDCGG, 2014), leading to a mean CO 2 gradient smaller than 0.04 % per degree of latitude.For CH 4 and CO, the latitudinal relative difference between the means of the latitudinal bands 0-30 and 30-60 • N in 2012 was smaller than 3.8 and 40 % respectively (WDCGG, 2014).This implies mean CH 4 and CO gradients smaller than 0.13 and 1.3 % per degree of latitude respectively.In the previous three gradient estimations, we have also taken into account the seasonal cycles, which depend on latitude (i.e.we are not simply considering the annual mean latitudinal gradients).For N 2 O, the latitudinal relative difference between 20 and 40 • N is smaller than 0.32 % (Huang et al., 2008;Kort et al., 2011) and the seasonal cycle is insignificant as can be seen in WDCGG (2014).This implies a mean N 2 O gradient smaller than 0.016 % per degree of latitude.For O 3 a gradient of 0.92 % per degree of latitude could be expected at the IZO latitude (value obtained from the ozone observations in 2012 of the space-based Ozone Monitoring Instrument (OMI); Veefkind et al., 2006).Therefore, assuming constant latitudinal gradients within the box ±1 • latitude/longitude centred at IZO, the spatial concentration variations inside the box (defined in an equivalent way as the temporal intra-day VC) are expected to be 0.53 % for O 3 , 0.75 % for CO, 0.01 % for N 2 O, 0.08 % for CH 4 , and 0.023 % for CO 2 (where we have taken into account that the standard deviation, in a segment of length 2 • , of a linear function with slope Gr per degree, is equal to Gr / √ 3).These spatial VC are similar (for O 3 and CO) or much smaller (for the rest of trace gases) than the statistical uncertainties of the FTS (recall Table 3).Therefore, no significant concentration variations might be expected within the actual area probed by the FTS observations and, indeed, a slightly wider range than this can be applied for collocating IASI measurements without affecting the validation results.Thus, we define a validation box of ±1 • centred at IZO location (i.e. ± ∼ 110 km at IZO latitude) for all the trace gases.Previous studies at IZO latitudes found no significant impact of the spatial co-location criteria (50-100 km) on the differences between IASI and FTS TCs for N 2 O, CH 4 , or CO (Kerzenmacher et al., 2012;García et al., 2013).

Figure 1 .
Figure 1.Site map indicating the location of the Izaña Atmospheric Observatory (IZO, marked by a black star) in the Canary Islands and the collocation box used for comparison (dashed lines), i.e. ±1 • latitude/longitude centred at IZO location.The grid lines divide the area into boxes of 0.25 • .Coloured filled circles correspond to IASI-A ozone total column observations on 31 January 2012.

Figure 2 .
Figure2.Row averaging kernels for O 3 and CH 4 as observed by IASI-A and FTS instruments, expressed on logarithm scale, for typical measurement conditions at IZO. US is upper stratosphere, MS is middle stratosphere, UTLS is upper troposphere and lower stratosphere, and T is troposphere.Also shown are the total degree of freedom for signal (DOFS), the vertical profile of the cumulative DOFS, as well as the a priori VMR profiles used for the FTS retrievals.

Figure 3 .
Figure 3.Time series of O 3 TC amounts (in 1 × 10 22 molec m −2 ) from the IASI-A and IASI-B overpasses (morning, M, and evening, E) (upper panel) and of the differences between the corresponding de-seasonalised variability (in %) for the morning and evening overpasses (middle and bottom panels).The solid lines represent the difference time series averaged on a weekly basis.The arrows distinguish the IASI-B commission mode (from 19 December 2012 to 18 June 2013) and the IASI-B operational mode (from 19 June 2013 onwards).

Figure 4 .
Figure 4. Scatter plots of the de-trended and de-seasonalised variability (in %) from the IASI-A and IASI-B overpasses (morning and evening) for O 3 .The legend shows the Pearson correlation coefficient, R, and the colour bar indicates the number of coincident data per bin.The dashed lines represent the diagonals (x = y).

Figure 5 .
Figure 5. Pearson correlation coefficient (upper panel) between the observations from IASI-A and IASI-B overpasses (morning, M, and evening, E) for all the trace gases considered at different timescales: single measurements (de-trended and de-seasonalised variability, Det + Des) and annual cycle (AC).Middle panel shows the same, but for the standard deviation (in %) of the corresponding differences, while the bottom panel displays the median bias (in %).The number of coincident observations is 675.

Figure 6 .
Figure 6.Summary of the IASI-A and FTS comparison for O 3 : (a, b) time series of the TC amounts (in 1 × 10 22 molec m −2 ) and the de-trended variability (in %) respectively; (c) averaged annual cycle;(d, e) time series of the de-trended+de-seasonalised variability (in %), and the difference between coincident de-trended+de-seasonalised variability from IASI-A and FTS (in %) respectively; and (f) scatter plot of the de-trended + de-seasonalised variability.For (c) and (f) the Pearson correlation coefficient is included in the legends and for (e) the standard deviation of the differences.The number of coincident measurements is 2338.

Figure 7 .
Figure 7. Pearson correlation coefficient (upper panel) between IASI-A and FTS observations for all the trace gases considered at different timescales: single measurements (detrended + de-seasonalised variability within ±1 h, Det + Des 1 h), daily (de-trended + de-seasonalised variability within the same day, Det + Des Day), annual, and long-term trend.The number of coincident data is 2338 for O 3 , 2003 for CO, and 425 for N 2 O, CH 4 , and CO 2 .Bottom panel shows the same, but for the standard deviation (in %) of the corresponding differences.The solid black lines represent the day-to-day variability calculated from the FTS observations at IZO between 2010 and 2014.

Figure 8 .
Figure 8. Multi-annual averaged annual cycle for CO, N 2 O, CH 4 , and CO 2 from IASI-A and FTS observations.

Figure 9 .
Figure 9. (a) Differences between the O 3 de-trended and deseasonalised variability time series from IASI-A and FTS (in %) as a function of the IASI-A air mass (AM) plotted as black squares with error bar (median and standard deviation per bins of 0.1 of AM), and simultaneous 2-D plot showing the difference between the IASI-A and FTS AMS.(b) Same as (a) but versus the aerosol optical depth (AOD) at 500 nm, recorded at AERONET SCO station during summer.Also, the differences between IASI-A and IASI-B are included.The black and white squares represent the median and standard deviation per bins of 0.025 of AOD for the IASI-A-FTS and IASI-A-IASI-B differences respectively.(c) Same as (b) but during winter.

Figure A1 .
Figure A1.Total statistical (ST) and systematic (SY) errors (in %) as a function of the solar zenith angle (SZA, in • ) for FTS O 3 and CH 4 measurements between 2010 and 2014.The black solid line represents the limit value of SZA = 75 • .Beyond this value the FTS observations are discarded.

Figure B1 .
Figure B1.Intra-day variation coefficient (VC) for O 3 total column and in situ CH 4 (in %]) as observed by a Brewer spectrometer and a GAW in situ GC-FID analyzer respectively between 2008 and 2013.The solid and dashed black lines represent the median and ±1σ of the reference intra-day VC respectively, and the dashed red lines represent the range of theoretical and experimental FTS errors.

Figure B2 .
Figure B2.Horizontal distance covered by the FTS observations (in km) versus the solar zenith angle (SZA; in • ), at which they are taken for all the target gases.The dashed lines represent the maximal horizontal distances.

Table 1 .
Technical specifications of the IASI and IZO ground-based FTS instrument.
The IASI remote sensing instruments are nadir-viewing atmospheric sounders based on FTSs and developed by CNES (Centre National d'Etudes Spatiales, www.cnes.fr) in cooperation with EUMETSAT.They are on board the EUMET-SAT Metop meteorological satellites, which operate in a polar, Sun-synchronous, low-Earth orbit since 2006 (Metop-A was launched in October 2006, Metop-B was launched in September 2012, and the Metop-C launch is scheduled for October 2018).The Metop-A and Metop-B currently operate in a co-planar orbit, 174 • out of phase.The IASI sensors were designed with the main goal of retrieving operational meteorological soundings (temperature and humidity) with high vertical resolution and accuracy for weather forecast use, as well as for monitoring atmospheric composition (O 3 , CO, N 2 O, CH 4 , and CO 2 ) at a global scale.Additionally, they provide land and sea surface temperature, surface emissivity, and cloud parameters

Table 1
L2 trace gas products: version 4 (V4) between June 2008 and September 2010, version 5 (V5) between September 2010 and September 2014, and version 6 (V6) from September 2014 onwards.Here we focus on the IASI V5 products, for which the longest IASI-A and coincident IASI-

Table 4 .
Clerbaux et al. (2009)l and spatial collocation criteria adopted for each trace gas.Also shown are the typical degree of freedom for signal (DOFS) for IASI and FTS products, and the number of coincident observations between IASI and FTS (N ).For IASI, the expected DOFS for O 3 and CO are taken from the Products User Guide (EUM/OPSEPS/MAN/04/0033, EUMET-SAT), while for CH 4 , N 2 O, and CO 2 those are obtained from Turquety et al. (2004) andClerbaux et al. (2009).The FTS DOFS are calculated from the corresponding retrievals between 2010 and 2014 (median ±1σ ).Note that daily means 24 h means.