IASI nitrous oxide ( N 2 O ) retrievals : validation and application to transport studies at daily time scales

The aim of this paper is to present a method to retrieve nitrous oxide (N2O) vertical profiles from the Infrared Atmospheric Sounding Interferometer (IASI) onboard the MetOp platform. We retrieved N2O profiles using IASI clear sky radiances in 2 spectral bands: B1 and B2 centered at ∼1280 cm−1 and ∼2220 cm−1, respectively. Both retrievals in B1 and B2 (hereafter referred to as N2O_B1 and N2O_B2, respectively) are sensitive to the mid-to-upper troposphere with a maximum of sensitivity at around 309 hPa. The degrees of freedom for N2O_B1 and N2O_B2 are 1.38 and 0.93, respectively. 5 We validated the retrievals using the High-performance Instrumented Airborne Platform for Environmental Research Pole-toPole Observations (HIPPO). The comparisons between HIPPO and the two retrieved datasets show relatively low standard deviation errors around 1.5% (∼4.8 ppbv) and 1.0% (∼3.2 ppbv) for N2O_B1 and N2O_B2, respectively. However, the impact of H2O contamination on N2O_B1 due to its strong absorption bands in B1 significantly degrades the quality of the retrievals in tropical regions. We analysed the scientific consistency of the retrievals at 309 hPa with a focus on the long-range transport of 10 N2O especially during the Asian summer monsoon. Over the mid-latitude regions, both variations of N2O_B1 and N2O_B2 at 309 hPa are influenced by the stratospheric N2O-depleted air because of the relative coarse shape of the averaging kernel. The analysis of N2O_B2 using results from backtrajectories exhibits the capacity of these retrievals to capture long-range transport of air masses from Asia to northern Africa via the summer monsoon anticyclone on a daily basis. Thus, N2O_B1 and N2O_B2 offer an unprecedented possibility to study global upper tropospheric N2O on a daily basis. 15


Introduction
Nitrous oxide (N 2 O) is a long-lived greenhouse gas with a lifetime of about 120 years which is essentially produced in the terrestrial and oceanic surfaces by the microbial processes of nitrification and denitrification (Butterbach-Bahl et al., 2013).
In terms of radiative forcing, N 2 O is the third anthropogenic greenhouse gas after methane (CH 4 ) and carbon dioxide (CO 2 ) (Ciais et al., 2014).Its main sink is the photolysis in the stratosphere but it is also destroyed by reacting with the excited atomic oxygen O( 1 D).This reaction is the main source of the nitrogen oxides, which are the main responsible of the destruction of the stratospheric ozone.N 2 O is therefore becoming the main ozone depleting substance emitted in the 21 st century (Ravishankara  et al., 2009).The natural and anthropogenic N 2 O emissions are about 60% and 40%, respectively (Syakila and Kroeze, 2011;Bouwman et al., 2013).The anthropogenic N 2 O emissions are dominated by agricultural sources which represent more than 66% of these emissions.An increase of the N 2 O volume mixing ratio (vmr) with a mean rate of 0.75 ppbv.yr−1 since the late 1970s has been observed (Ciais et al., 2014).This positive trend is driven by anthropogenic emissions because of the increasing use of nitrogen fertilizers to meet the growing demand of food production, especially in Asia.Moreover, according to the Intergovernment Panel on Climate Change (IPCC), this trend is likely to continue until 2100.Monitoring N 2 O emissions and its atmospheric concentration are therefore becoming major issues in the framework of anthropogenic pollution mitigation.
Nowadays, surface measurements of N 2 O provide the longer time series of N 2 O measurements and are used to characterize the trends and the sources of tropospheric N 2 O.Such measurements are performed by several organizations or institutes such as the National Oceanic and Atmospheric Administration/Earth Systems Research Laboratory/Global Monitoring Division (NOAA/ESRL/GMD) or in the framework of joint projects such as the Advanced Global Atmospheric Gases Experiment (AGAGE) (Ganesan et al., 2015) and the Network for the Detection of Atmospheric Composition Change (NDACC) (http://www.ndsc.ncep.noaa.gov).Despite their reliability and the long-term records of surface measurements, their limited geographical coverage makes them difficult to use in order to assess N 2 O tropospheric variations at global scale.In addition to surface measurements, there are also some aircraft campaigns like the High-performance Instrumented Airborne Platform for Environmental Research Pole-to-Pole Observations (HIPPO) (Wofsy, 2011;Wofsy et al., 2012) over the Pacific Ocean.N 2 O is also measured in some passenger aircraft based measurements including the Comprehensive Observation Network for TRace gases by AIrlLner (CONTRAIL) (Sawa et al., 2015) and the Civil Aircraft for the Regular Investigation of the atmosphere Based on an Instrument Container (CARIBIC) (Assonov et al., 2013).
Since satellite measurements of stratospheric N 2 O began in the 1970s, tropospheric N 2 O retrievals using satellite measurements are relatively recent.Clerbaux et al. (2009) exhibit the N 2 O signature from the infrared measurements of the Infrared Atmospheric Sounding Interferometer (IASI) showing some promising results in view of using these measurements to retrieve N 2 O tropospheric profiles.Ricaud et al. (2009a) analysed the equatorial maximum of N 2 O during March-May using the operational total columns of N 2 O retrieved from IASI measurements using artificial neural networks.These operational N 2 O total column products also show seasonal cycles and annual trends consistent with the retrieved N 2 O from the ground-based Fourier Transform Spectrometer (FTS) observations at the Izaña Atmospheric Observatory (IZO, Spain) (García et al., 2014(García et al., , 2016)).First results of N 2 O total columns retrievals using a partially scanned IASI interferogram with an accuracy of ±13 ppbv (∼4%) are described in Grieco et al. (2013).Retrievals of N 2 O tropospheric profiles have been performed using the Atmospheric Infrared Sounder (AIRS) and the results showed interannual trends consistent with surface measurements (Xiong et al., 2014).N 2 O profiles retrieved from the Greenhouse Gas Observing Satellite (GOSAT) measurements have been used to study the transport of Asian summertime high N 2 O emissions to the Mediterranean upper troposphere (Kangah et al., 2017).
In this paper, we describe the IASI instrument and the Radiative Transfer for Tiros Operational Vertical sounder (RTTOV) used as forward model in our retrieval system in sections 2 and 3, respectively.We present the retrieval strategy and the validation of the results using HIPPO airborne in situ measurements in sections 5 and 6, respectively.In section 7, we analyse the scientific consistency of the retrievals focusing on the long-range transport of N 2 O during the Asian summer Monsoon using backtrajectories from the Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT) model (Stein et al., 2015).Conclusions are presented in section 8. MetOp-B are in the same orbital plane and have an orbit phasing of about 49 min.IASI is a Michelson interferometer that measures infrared spectrum in the spectral range from 645 to 2760 cm −1 (15.5 to 3.62 µm) (Clerbaux et al., 2009).Although its apodized spectral resolution is about 0.5 cm −1 , IASI provides each spectrum with a sampling of 0.25 cm −1 giving a total of 8461 channels.The large spectral domain of IASI contains absorption bands of several atmospheric constituents (Hilton et al., 2012) among which the major absorbers are water vapour (H 2 O), ozone (O 3 ), CO 2 , N 2 O, CH 4 and carbone monoxide (CO).IASI observes the Earth with a swath of about 2200 km (1100 km on each side) and its instantaneous field of view is composed of four circular pixels of 12 km diameter footprint on the ground at nadir.The operational IASI H 2 O, temperature and O 3 products are retrieved simultaneously using an optimal estimation method (Pougatchev et al., 2009;Rodgers et al., 2000) whereas total columns of the other molecules are retrieved using artificial neural networks (Turquety et al., 2004).In this work, we used the IASI level 1c spectra (calibrated and apodized spectra) to perform our retrievals.

RTTOV
RTTOV is a fast model of transmittances of the atmospheric gases that are generated from a database of accurate line-byline (LBL) transmittances (Saunders et al., 1999).The database of accurate transmittances is generated from a set of diverse atmospheric profiles and then a linear regression is computed linking the optical depths of the vertical layers and a set of atmospheric profile-dependent predictors.The regression coefficients are actually given for different Instrument Spectral Response Functions (ISRF) including the ones of IASI.For our retrieval system, we used RTTOV version 11.2 together with the regression coefficients v9 based on the model LBLRTM (LBL Radiative Transfer Model) (Hocking et al., 2015).In this version, the predictors depend on the trace gases profiles including H 2 O, O 3 , CO 2 , N 2 O, CH 4 and CO.It takes less than 25 ms to compute 183 IASI channels together with weighting functions using an input of atmospheric profiles on 54 vertical levels and surface emissivities.Comparing with accurate LBL models, the biases of RTTOV simulations for IASI Brightness Temperature (BT) over sea in clear sky conditions are within ±1 K in the spectral range 645 to 2000 cm −1 and within ±1.6 K in the N 2 O/CO 2 ν3 region between 2200 and 2300 cm −1 (Matricardi, 2009).

N 2 O absorption bands
Previous studies from Clerbaux et al. (2009) have highlighted three absorption bands of N 2 O in the IASI spectral range centered at ∼1280 cm −1 , ∼2220 cm −1 and ∼2550 cm −1 .Figure 1 shows a N 2 O weighting function matrix (called hereafter Jacobian matrix) calculated in units of brightness temperature (BT) using a N 2 O profile derived from the Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) reference atmosphere (V3) daytime mid-latitude climatology.This matrix represents the sensitivity of the calculated BT to a unit change in the N 2 O volume mixing ratio (vmr).The spectral signature of N 2 O appears in the three spectral regions with significant differences of intensity.The most intense absorption band (called hereafter B2) is between 2190 and 2240 cm −1 and shows sensitivity to N 2 O from the lowermost troposphere to 100 hPa with a maximum of sensitivity between 500 and 200 hPa.The absorption band located between 1250 and 1310 cm −1 (called hereafter B1) is less intense than B2 and is sensitive to N 2 O between 800 and 100 hPa.The third band (called hereafter B3) located between 2500 and 2600 cm −1 is much less intense than B1 and B2 and is sensitive to N 2 O from 900 to 300 hPa.To illustrate the sensitivity of these three bands to N 2 O and to the other atmospheric and surface parameters, a sensitivity study has been performed using the MIPAS climatology for N 2 O, CO 2 and O 3 profiles and a set of atmospheric and surface parameters representative of a given atmospheric state on 13 June 2011 at 11.8 • N and 142.9 • W. This study consists in calculating of the variation of the BT (called hereafter ∆BT ) over the IASI spectral range for a given variation of the major atmospheric and surface parameters consistent with their actual accuracy.Figure 2  is mainly impacted by temperature, H 2 O, CH 4 and surface temperature.The signal corresponding to 10% change of H 2 O is more than twice greater than the signal corresponding to a change of N 2 O by 4% in most spectral domains of B1.The signal corresponding to a change of CH 4 by 2% is half the size to the signal of N 2 O.A total of 126 channels is selected in B1.The signal of N 2 O is twice larger than the NEDT for all selected channels in B1 whilst CH 4 , H 2 O and temperature are critical parameters for the N 2 O retrieval using the 126 selected channels in B1.In B2, we selected a total of 103 channels where the signal of N 2 O is more than twice greater than the signals of the other parameters except for atmospheric temperature and NEDT.The NEDT level of magnitude is similar to the signal of N 2 O while the |∆BT | signal corresponding to the temperature variation is slightly greater than that of N 2 O.The radiometric noise and the atmospheric temperature are therefore the critical parameters for the N 2 O retrieval in B2.In B3, we selected no channels because the radiometric noise is too large compared to the signal of N 2 O.In summary, the absorption band of N 2 O in B2 is sufficiently isolated from the absorption band of the other gases but presents the same level of magnitude as the IASI radiometric noise whereas in B1 the signal of N 2 O is more than twice greater than the noise but is impacted by the absorption bands of CH 4 and H 2 O.

Methodology
We used an optimal estimation method based on the Levenberg-Marquardt iterative algorithm (Rodgers et al., 2000) to retrieve N 2 O profiles over 13 fixed pressure levels from IASI clear sky radiances in the bands B1 and B2.Hereafter, the retrievals in B1 and B2 are referred to N 2 O_B1 and N 2 O_B2, respectively.In the retrieval algorithm, the i+1 th retrieval vector is expressed as: where X a is an a priori vector with an error covariance matrix S a .Y is the observed radiances with an error covariance matrix S y .F ( Xi ) and K i are the calculated forward spectrum and the Jacobian matrix at the iteration i, respectively.γ is the Levenberg-Marquardt parameter (Rodgers et al., 2000).The vertical sensitivity of the retrieval can be characterised using the averaging kernel matrix (A) defined as: N 2 O_B1 profiles are retrieved simultaneously with the vmr profiles of H 2 O and CH 4 whilst N 2 O_B2 profiles are retrieved simultaneously with the vmr profiles of H 2 O, CO and CO 2 .The air temperature profiles and the surface parameters (temperature and emissivity) are also retrieved simultaneously with the N 2 O profiles for N 2 O_B1 and N 2 O_B2.
The a priori error covariance matrix S a is calculated as follows: where σ 2 a is an a priori variance error fixed for each parameter of the state vector and P i the pressure level at the level i.For the retrievals, we used a fixed N 2 O a priori profile derived from the MIPAS V3 reference atmosphere daytime midlatitude climatology.Since this climatology is given for the year 2001, we adjusted it for the year 2011 by applying the averaged increase rate of 0.75 ppbv.yr−1 consistently with Ricaud et al. (2009b).We fixed σ a for the N 2 O profile to 4% consistently with Grieco et al. (2013).
The a priori states of H 2 O, temperature and surface temperature were taken from the IASI level 2 operational products (August et al., 2012).A validation using radiosonde data gave a standard error (std) of ∼2 K for the surface temperature, of about 10% for the relative humidity and between 0.6 and 1.5 K for the temperature profile (Pougatchev et al., 2009).Thus, we took for N 2 O_B1 and N 2 O_B2, σ a values of 1 K and 2 K for the temperature profile and the surface temperature, respectively.
A σ a value of 10% for the H 2 O profile was used for N 2 O_B2.Clerbaux et al. (2009) show the presence of a relatively strong absorption band of the deuterium hydrogen oxide (HDO) also called semiheavy water in the band B1.However, this chemical species is not taken into account as a variable parameter in RTTOV.Therefore, after sensitivity studies, we fixed the σ a value for the H 2 O profile to 30% for N 2 O_B1.In a similar approach to the N 2 O a priori profile, we took the CO 2 a priori from the MIPAS reference atmosphere v3 daytime mid-latitude climatology and applied an annual trend of 2.3 ppmv.yr−1 (Ciais et al., 2014).A σ a of 2% is used for the CO 2 a priori profile after a sensitivity study.
In addition, CH 4 and CO a priori profiles were taken from the Monitoring Atmospheric Composition and Climate (MACC) project reanalysis (Inness et al., 2013).σ a was fixed to 10% for CO after a sensitivity study and consistently with the CO validation reports (http://www.gmes-atmosphere.eu/services/aqac/global_verification/validation_reports/).For CH 4 , σ a was fixed to 2% which is approximately the std error on the IASI retrieved CH 4 profiles (Xiong et al., 2013).The land surface a priori emissivity is derived from a global atlas of land surface emissivity based on inputs from the Moderate Resolution Imaging Spectroradiometer (MODIS) operational product (Borbas and Ruston, 2010;Seemann et al., 2008).Over sea surface, we used the version 6 of the Infrared Surface Emissivity Model (ISEM) (Sherlock and Saunders, 1999) as an a priori surface emissivity.σ a is fixed to 10% for the surface emissivity since this parameter is also used as a sink parameter.An observation error diagonal covariance matrix S y was used for the retrievals in both bands with the IASI radiometric noise as the diagonal elements of the matrix.

Data quality control
To assess the quality of the retrieved N 2 O profiles, we used quality parameters derived from the optimal estimation theory (Rodgers et al., 2000).Our retrieval process consists in the minimization of the cost function χ 2 defined as: where dim( X) and dim(Y ) are the dimensions of the state vector and of the radiances (number of channels), respectively.χ 2 allows to evaluate the quality of the retrieval by combining the calculated residuals relative to the observations error covariance matrix and the difference between the estimated and the a priori profiles relative to the a priori error covariance matrix.In our case, we performed simultaneous retrievals for both N 2 O_B1 and N 2 O_B2.Therefore, the χ 2 derived from the optimal estimation theory is a quality control parameter for the whole retrieved state vectors which include N 2 O profiles and the other interfering parameters.In addition to χ 2 , we computed another variable to assess the quality of the retrieved tropospheric N 2 O profile which is our target species.Thus, we calculate the difference between the a priori and the retrieved N 2 O relative to the N 2 O a priori errors σ a .This variable called χ 2 N2O is defined as: where Xj and X aj are the retrieved and the a priori N 2 O at the pressure level P j , respectively.β aj is the diagonal element of the a priori error precision matrix (the inverse of the a priori error covariance matrix) at the pressure level P j and n p is the number of levels used for the calculation.
An upper limit for the χ 2 parameter is generally used to select good quality pixels.For instance, an upper limit of 3 on a χ 2 calculated in the radiances space was used to select good quality pixels for CH 4 retrievals from IASI measurements (Xiong et al., 2013).Following the same methodology, we applied an upper limit on χ 2 N2O to select good quality pixels.After performing sensitivity studies for both N 2 O_B1 and N 2 O_B2, we rejected all the data with a χ 2 or a χ 2 N2O greater than or equal to 4.
Moreover, to evaluate the impact of the other retrieved parameters on N 2 O_B1 and N 2 O_B2, we calculated the Contamination Factor (called hereafter CF ) defined as follows: Here, CF (i) is the contamination of the parameter c on the retrieved N 2 O at the level i (x i ).∆c j is the uncertainty on the parameter c at the level j.We fixed ∆c j to the a priori error σ a for each parameter.Then for the parameter c, we defined CF tot (c) as the sum of the CF over the 13 retrieval levels.CF indicates the influence of the uncertainties in the knowledge of the co-retrieved parameters on the variability of the target species N 2 O retrievals.Here, the uncertainties on the co-retrieved parameters have been fixed to the a priori uncertainties.Thus, CF does not take into account the effects due to the spatial and temporal variations of these uncertainties.But CF estimates, a priori, how critical is the characterisation of each co-retrieved parameter for the quality of the N 2 O retrievals.As consequence, a posteriori sensitivity studies should be performed on each critical parameter to determine which co-retrieved parameters uncertainties have the most significant impact on the quality of the N 2 O retrievals.

Validation
In this section, we analyse the performance of our retrieval system by comparing the results with the in-situ measurements from the five HIPPO airborne campaigns (Figure 3): HIPPO 1 (January 2009), HIPPO 2 (October-November 2009), HIPPO 3 (March-April 2010), HIPPO 4 (June-July 2011) and HIPPO 5 (August-September 2011).For this purpose, we processed 26850 N 2 O_B1 and N 2 O_B2 profiles along the flight paths from the five HIPPO campaigns.Using a similar method as explained in Kangah et al. (2017), we used for these comparisons the measurements from the Harvard/Aerodyne Quantum Cascade Laser Spectrometer (QCLS), one of the airborne instruments of HIPPO, and the retrieved profiles selected within a collocation temporal and spatial window of ±200 km and ±12h, respectively.Our aim is to characterise the retrieval errors as well as the ability of the retrieval system to capture N 2 O tropospheric variations.

Error Characterisation
The total retrieval errors can be divided into four components: a smoothing error, a forward model error, a model parameter error and a retrieval noise.We used a simultaneous retrieval strategy to include all the parameters which influence RTTOV in each band and we removed RTTOV systematic biases consistently with Matricardi (2009).Therefore, the forward model and the model parameter errors can be, as a first approximation, considered as negligible compared to the smoothing error and the retrieval noise.The covariance matrix of the smoothing error (S s ) is defined as: where A is the N 2 O averaging kernels matrix; I is the identity matrix and S e is the covariance matrix of the real ensemble of states consistently with Rodgers et al. (2000).For our retrieval algorithm, we use a simple "ad hoc" matrix (see Eq. 3) as a priori covariance matrix (S a ) to constrain the retrieval system.Since this matrix may or may not be representative of the variability of a real ensemble of N 2 O profiles, we took S e as the covariance matrix of HIPPO profiles.
The retrieval noise covariance matrix (S n ) is defined as: where G is the gain matrix which represents the change in the vmr profile for a unit change in the observation Y .The theoretical covariance matrix of the total errors (S tot ) is therefore defined as: The theoretical covariance matrix of the total errors is then compared with an empirical total errors covariance matrix calculated using the HIPPO measurements and the retrievals along the HIPPO campaigns flight paths (namely the covariance matrix of the difference between HIPPO profiles and IASI retrieved profiles).Figure 4 shows the standard deviation errors (std errors) corresponding to all these covariance matrices (square roots of the diagonal elements of the covariance matrix) and averaged over the set of retrievals for N 2 O_B1 and N 2 O_B2.The empirical std error which we consider as our reference standard deviation of the total errors (σ tot ) is about 1.5% (∼4.8 ppbv) for N 2 O_B1 and about 1.0% (∼3.2 ppbv) for N 2 O_B2 in the troposphere.For N 2 O_B2, the theoretical σ tot is consistent with the empirical σ tot but, for N 2 O_B1, the theoretical σ tot is about 0.5% less than the empirical σ tot .This means that our hypothesis of two sources of errors to characterise the total error is correct for N 2 O_B2 but is not enough for N 2 O_B1 for which other sources of error should be considered (forward model errors and/or model parameter errors).Concerning the forward model errors, we removed the biases on RTTOV IASI clear sky radiances consistently with Matricardi (2009) both in the band B1 and B2.Therefore the difference between the theoretical and the empirical std errors for N 2 O_B1 is certainly due to the existence of other sources of variation of the radiances in the band B1 which are not correctly taken into account in our retrieval system.The HDO absorption which is the only significant absorption band not included in the predictor parameters of RTTOV could be responsible of at least part of these unexplained variations.To summarise, we can consider that the std errors on N 2 O_B1 and N 2 O_B2 are on averaged about 1.5% (∼4.8 ppbv) and 1.0% (∼3.2 ppbv), respectively.However, for the users, the retrieved profiles will be given with the empirical S tot together with the theoretical S tot associated with each retrieval.

Sensitivity in the observation and retrieval spaces
Figure 5 shows the averaged observed and calculated (using a priori and retrievals) radiances together with the averaged calculated residuals for both B1 and B2.In B1, the mean residual is reduced from -0.8% (using the a priori) to 0.01% (using the retrievals) whereas in B2, the mean residual is reduced from -0.5% (using the a priori) to 0.01% (using the retrievals).The differences between the a priori residuals in B1 and B2 are due to the existence of more interfering parameters in B1 than in B2.
Therefore, some differences between N 2 O_B1 and N 2 O_B2 due to the contamination of CH 4 and H 2 O are expected.Figures 6 and 7 show the mean N 2 O normalized (Deeter et al., 2007) averaging kernels matrix together with the altitude of the kernels maximum and the mean CF from CH 4 , temperature, surface temperature and H 2 O for N 2 O_B1 and N 2 O_B2, respectively.
Considering the averaging kernels, the maximum of sensitivity is located at the retrieval level 309 hPa for both N 2 O_B1 and N 2 O_B2.In addition, the averaging kernels corresponding to this level peak at 309 hPa.Therefore, retrieved vmrs at this level are the most reliable for both N 2 O_B1 and N 2 O_B2.For N 2 O_B2, all the averaging kernels peak at the levels 309 hPa.This means that the retrieved N 2 O vmr profiles are mainly sensitive to the real N 2 O vmr at this level.This result is consistent with previous studies from Kangah et al. (2017) and Xiong et al. (2014).The degree of freedom (DOF), which represents the number of independent vertical pieces of information of the retrieved profile and is computed as the trace of the averaging kernels matrix, is on average equal to 1.38 and 0.93 for N 2 O_B1 and N 2 O_B2, respectively.The DOF for N 2 O_B1 is greater than that of N 2 O_B2 because the SNR is higher in B1 than in B2.Thus, more channels with better SNR are selected in B1 than in B2.Although the retrieved N 2 O is impacted by temperature in the two bands, we have in B1 an additional significant impact of CH 4 and H 2 O.In conclusion, we expect more contamination on N 2 O_B1 than on N 2 O_B2.

Retrieval accuracy
To assess the skills of the retrieval process, we applied the IASI N 2 O averaging kernels to the HIPPO profiles using the following equation (Rodgers et al., 2000): where x a is the IASI a priori profile, x the HIPPO profile, x the result of the averaging kernels application (called hereafter convolved HIPPO), I the identity matrix and A the IASI N 2 O averaging kernels matrix.
Figures 8 and 9 show the results from the comparisons between HIPPO measurements and N 2 O_B1 and N 2 O_B2 averaged within the spatial and temporal window around the HIPPO measurements, respectively.N 2 O_B1 and HIPPO measurements are moderately correlated (the Pearson linear correlation coefficient R=0.42) with a low bias and standard deviation (called hereafter std) error of -1.6 ppbv (∼0.5%) and 3.5 ppbv (∼1.0%), respectively.However, the quality of the retrievals depends on the latitude band.The consistency between N 2 O_B1 and HIPPO increases at mid-latitudes (e.g.R=0.63 for northern hemisphere mid-latitudes).We can also notice that there is a very low mean bias (-0.1 ppbv) in the northern hemisphere high-latitude regions.) and CF max tot (temperature), respectively.Then, we evaluated the mean bias (N 2 O_B1 − HIPPO) using these filtered N 2 O_B1.In order to have enough collocated IASI-HIPPO pixels, N 2 O_B1 around the HIPPO measurements are not averaged for this sensitivity study (see Figure 10).We observe that the lower CF max tot (H 2 O), the better the bias, whereas there is no significant improvement of the bias when we used CF max tot (CH 4 ) and CF max tot (temperature) to filter N 2 O_B1.Thus, when CF max tot (H 2 O) decreases from 10 to 4, the absolute value of the mean bias decreases from 2.5 to 1.0 ppbv.Therefore, we can consider that the degradation of the quality of N 2 O_B1 over the tropics is mainly due to the contamination of H 2 O.Although CF tot (temperature) and CF tot (CH 4 ) are greater than CF tot (H 2 O), H 2 O is actually the most critical parameter in the band B1 to retrieve N 2 O in our retrieval system.H 2 O has a high variability, especially over the tropical regions where maxima of H 2 O vmrs are observed.This variability is more difficult to retrieve than the variability of temperature and CH 4 .Thus, August et al. (2012) show that the std error on the IASI retrieved temperature at 800 hPa varies from 1 K (northern hemisphere sea) to 1.5 K (tropical land) whereas the std error on the IASI retrieved H 2 O at 800 hPa varies from 1500 ppmv (northern hemisphere sea) to 3500 ppmv (tropical land).Furthermore, we evaluated the linear correlation R and the std error using the different values of CF max tot (H 2 O) (see Figure 11).When CF max tot (H 2 O) decreases from 10 to 4, R increases from 0.17 to 0.57 and the std error decreases from 3.5 to 3.0 ppbv.CF tot (H 2 O) should therefore be considered carefully when analysing N 2 O_B1 over tropical regions.
N 2 O_B2 is moderately correlated with HIPPO measurements (R=0.6) with a std error of 3.2 ppbv and a very low mean bias of 0.3 ppbv.This moderate correlation is also observed when considering only data from the northern hemisphere mid-latitudes.
In the northern hemisphere tropical regions, the bias is slightly higher (-1.0 ppbv) and the correlation coefficient decreases to 0.31.The worst correlation coefficient is found for the southern hemisphere mid-latitudes (R=0.11).The very small slope (∼0.16) indicates that N 2 O_B2 does not capture optimally the N 2 O spatial and temporal variations in this region, although we observe a relatively low mean bias (1.6 ppbv) in this region.In tropical regions the correlation coefficient between N 2 O_B2 and HIPPO measurements becomes very high (0.71 and 0.92 in the northern and southern hemispheres, respectively) compared to the other regions.However, the large slope from the linear regression (2.51 and 3.32 in the northern and southern hemispheres, respectively) indicates that N 2 O_B2 tends to overestimate the spatial and temporal N 2 O vmr gradients in this region.
In summary, N 2 O_B1 and N 2 O_B2 are of sufficient quality to be used to analyse N 2 O variations in the mid and high latitude regions.N 2 O_B2 can even be used to analyse N 2 O transport processes between tropical regions and higher latitude regions whereas, for N 2 O_B1, we have to analyse vmrs in the tropics taking care to reject retrievals with high CF tot (H 2 O).The scientific users should fix CF max tot (H 2 O) to filter N 2 O_B1 according to their need in terms of accuracy and/or spatial and temporal sampling.The statistics presented on Figure 11 can be used for that purpose.is also observed over the eastern Mediterranean as a result of these transport processes.N 2 O_B1 also exhibits maxima over the eastern China (∼332 ppbv) which is the result of the vertical transport to the upper troposphere of the high summertime N 2 O emissions over this region.The high emissions and vertical transport are confirmed by the occurrence of relatively high convective precipitations over the eastern China region the days before (24-26 July, not shown).
To highlight these long-range transport processes, we used the HYSPLIT model (Stein et al., 2015) to perform a 4-day backtrajectory ensemble from a central point located in the upper troposphere (∼306 hPa) of the northern Africa (25 • N, 32 • E) on 28 July 2011 at 12h00 UTC.Then, N 2 O_B2 along the path of the air masses represented by the mean trajectory is evaluated.
Figure 15 shows the results of the backtrajectories with air masses transported from western Asia to northern Africa.Furthermore, the mean trajectory is located on a vertical range 420-316 hPa which is in the domain of vertical sensitivity of N 2 O_B2 considering the shape of the averaging kernel for the level 309 hPa (Figure 7).To study N 2 O_B2 along the mean trajectory, we calculated a Hovmöller diagram using the latitudinal range of the backtrajectory ensemble from 21 to 31 • N.This diagram calculated on a daily basis for the longitude range from 30 to 80 • E and superimposed with zonal winds from the ERA-Interim reanalysis is presented in Figure 16.As expected, the mean daily trajectory represented by the black stars is located in an easterly wind region (delimited by the blue contours).Moreover, this diagram exhibits the transport of high N 2 O_B2 maximum by the easterly wind fluxes associated to the Asian monsoon anticyclone.N 2 O_B2 corresponding to the mean trajectory and averaged over the latitude range of the Hovmöller diagram are within the range 330.5-331.5 ppbv.We averaged a maximum of 3 basic (4 • × 4 • ) pixels over the latitude range of the Hovmöller diagram.Since the std error for a single N 2 O_B2 pixel is about 2.8 ppbv (see Section 6.3), we can approximate the error on the retrieved N 2 O_B2 over the mean trajectory by 2.8/ √ 3 ≈ 1.6 ppbv.Therefore, we can conclude that N 2 O_B2 allows to follow the upper tropospheric N 2 O transport processes between tropical and mid-latitude regions at nearly daily time scales.

Conclusions
We presented and validated an inversion algorithm to retrieve N 2 O profiles using IASI level 1c radiances.Consistently with previous studies, the N 2 O Jacobian exhibits three N 2 O absorption bands: B1 centered at ∼1280 cm −1 , B2 centered at ∼2220 cm −1 and B3 centered at ∼2550 cm −1 .We performed a sensitivity test in each band studying the radiometric noises and  et al., 2014;Kangah et al., 2017), both N 2 O_B1 and N 2 O_B2 are sensitive to mid-to-upper troposphere with a maximum of sensitivity in the upper troposphere (∼309 hPa).
We developed quality control parameters based on the standard χ 2 derived from the optimal estimation theory and on a reduced χ 2 parameter called χ 2 N2O .χ 2 gives a quality criteria for the whole state vector and χ 2 N2O gives a quality criteria for the N 2 O tropospheric profile.Besides these two parameters, another quality control parameter based on CF tot (H 2 O) was used to assess the impact of the H 2 O contamination on N 2 O_B1, especially in tropical regions.N 2 O_B1 and N 2 O_B2 at 309 hPa are validated using HIPPO airborne in situ measurements.From these comparisons, we calculated std errors around 1.5% and 1.0% for N 2 O_B1 and N 2 O_B2, respectively.Besides, we calculated relatively low biases (-1.6 ppbv for N 2 O_B1 and 0.3 ppbv for N 2 O_B2).Apart from an overestimation of gradients in tropical regions, N 2 O_B2 is of a good quality in all latitudinal bands.The quality of N 2 O_B1 is good except in tropical regions where H 2 O contamination characterised by high CF tot (H 2 O) degraded the quality of the retrievals.
We studied the scientific consistency of the retrieved N 2 O by focusing on transport processes.We showed that both N 2 O_B1 and N 2 O_B2 variations over the mid-latitudes regions are influenced by the N 2 O-depleted air from high latitudes and from the stratosphere.Using backtrajectory calculations, we also showed that the transport of high Asian N 2 O emissions from Asia to the Eastern Mediterranean basin by the summertime Asian monsoon anticyclone can be observed using N 2 O_B2 on a daily basis.N 2 O_B1 also offers good opportunities to study this N 2 O transport process but with limitations due to H 2 O contamination over the tropics.Thus, at this stage of our retrieval process, N 2 O retrieved in bands B1 and B2 offer an unprecedented possibility to study upper tropospheric N 2 O on a daily basis at global scale.This algorithm will be therefore applied to retrieve N 2 O profiles at a global scale using the 10 years of IASI measurements.
spaceborne instrument on board the platforms MetOp-A and MetOp-B.The MetOp (Meteorological Operational) mission consists of a series of three sun-synchronous Low Earth Orbits satellites developed jointly by the french space agency (CNES) and the EUropean organization for the exploitation of METeorological SATellites (EUMETSAT).The first satellite (MetOp-A) was launched in October 2006, the second (MetOp-B) in September 2012 and the third (MetOp-C) is expected to be launched in October 2018.MetOp-A and MetOp-B are operational at the present time.The mean MetOp altitude is ∼820 km and the satellite crosses the equator at ∼09:30 mean local solar time and have a repeat cycle of 29 days.MetOp-A and shows the absolute value of the ∆BT (|∆BT |) for variations of each major absorber (H 2 O, O 3 , CO 2 , N 2 O, CH 4 and CO) and for variations of temperature and surface temperature.The IASI radiometric noise expressed as the Noise Equivalent Delta Temperature (NEDT) is superimposed to the |∆BT | signals.In each band, channels were selected by optimizing the Signal to Noise Ratio (SNR) while reducing the spectral signature of the other parameters.B1 Atmos.Meas.Tech.Discuss., https://doi.org/10.5194/amt-2018-21Manuscript under review for journal Atmos.Meas.Tech.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.
Atmos.Meas.Tech.Discuss., https://doi.org/10.5194/amt-2018-21Manuscript under review for journal Atmos.Meas.Tech.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.Furthermore, N 2 O_B1 exhibits greater biases in tropical regions (-3.7 and -4.8 ppbv in the tropical northern and southern hemispheres, respectively) than in the other regions.Figure 6 suggest that the largest CF on N 2 O_B1 are from the temperature, CH 4 and H 2 O, respectively.To understand the degradation in the quality of N 2 O_B1 over the tropics, we examined the contamination of H 2 O, CH 4 and temperature (see Eq. 6).For that purpose, we filtered N 2 O_B1 over the northern hemisphere tropical regions by considering the pixels with CF tot (H 2 O), CF tot (CH 4 ) and CF tot (temperature) less than arbitrary maxima called CF max tot (H 2 O), CF max tot (CH 4 Atmos.Meas.Tech.Discuss., https://doi.org/10.5194/amt-2018-21Manuscript under review for journal Atmos.Meas.Tech.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.Atmos.Meas.Tech.Discuss., https://doi.org/10.5194/amt-2018-21Manuscript under review for journal Atmos.Meas.Tech.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.troposphere and redistributed westward by the easterly winds associated to the Asian summer monsoon anticyclone.Wind patterns on Figure 12 show the connection between the western Asia region and the eastern Mediterranean region.Thus, longrange transport processes between Asia and the eastern Mediterranean are expected.A maximum of N 2 O_B1 (∼331 ppbv) Atmos.Meas.Tech.Discuss., https://doi.org/10.5194/amt-2018-21Manuscript under review for journal Atmos.Meas.Tech.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.

Figure 1 .
Figure 1.N2O Jacobian in brightness temperature calculated by RTTOV over IASI spectral range and for an atmospheric situation at 142.96 • W and 11.76 • N.

Figure 4 .
Figure 4. Top: Standard deviations of the smoothing errors (solid green line), the retrieval noise (solid red line), the theoretical total errors (solid blue line) and the empirical total errors (solid black line) on N2O_B1 averaged over a set of 26850 retrievals along the HIPPO campaigns flight paths.Bottom: same as top but for N2O_B2.

Figure 8 .
Figure 8. HIPPO N2O measurements vs N2O_B1 at 309 hPa averaged within a box of ± 200 km and a temporal window of ± 12h around the HIPPO measurements.The black and red lines represent the first bisector (y=x) and the linear regression line, respectively.The colorbar represents the different latitude bands of the HIPPO measurements.N is the number of collocated pixels.

Figure 9 .
Figure 9. Same as Figure 8 but for the N2O_B2.

Figure 11 .
Figure 11.From top to bottom: Pearson linear correlation coefficient (R), mean bias, std error and number of IASI retrieved pixels (N) for different values of CF max tot (H2O) corresponding to N2O_B1-HIPPO collocated pixels in the northern hemisphere tropical regions (0-30 • N) .
the signals from N 2 O, temperature, surface temperature and other major absorbers including CH 4 , H 2 O, CO, O 3 , CO 2 .By maximizing the SNR and minimizing the impact from the interfering parameters in each bands, 126 channels in B1 and 103 channels in B2 were selected to retrieve N 2 O profiles.We also deduced from this sensitivity study that, in addition to the impact of temperature and surface temperature, B1 is impacted by relatively strong absorption bands of CH 4 and H 2 O whereas B2 has relatively strong radiometric noises.A retrieval algorithm based on the Levenberg-Marquardt optimal estimation theory was used to retrieve N 2 O profiles using B1 and B2 (namely N 2 O_B1 and N 2 O_B2, respectively).N 2 O_B1 was retrieved simultane-Atmos.Meas.Tech.Discuss., https://doi.org/10.5194/amt-2018-21Manuscript under review for journal Atmos.Meas.Tech.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.ously with CH 4 , H 2 O, temperature, surface temperature and surface emissivity whereas N 2 O_B2 was retrieved simultaneously with H 2 O, temperature, CO, CO 2 , surface temperature and surface emissivity.Consistently with the previous studies (Xiong