Retrieval of temperature , H 2 O , O 3 , HNO 3 , CH 4 , N 2 O , ClONO 2 and ClO from MIPAS reduced resolution nominal mode limb emission measurements

Retrieval of temperature, H2O, O3, HNO3, CH4, N2O, ClONO2 and ClO from MIPAS reduced resolution nominal mode limb emission measurements T. von Clarmann, M. Höpfner, S. Kellmann, A. Linden, S. Chauhan, B. Funke, U. Grabowski, N. Glatthor, M. Kiefer, T. Schieferdecker, G. P. Stiller, and S. Versick Forschungszentrum Karlsruhe, Inst. für Meteorologie und Klimaforschung, Karlsruhe, Germany Instituto de Astrof́ısica de Andalucı́a, CSIC, Spain Received: 18 December 2008 – Accepted: 7 January 2009 – Published: 30 January 2009 Correspondence to: T. von Clarmann (thomas.clarmann@imk.fzk.de) Published by Copernicus Publications on behalf of the European Geosciences Union.


Introduction
In March 2004 the MIPAS (Michelson Interferometer for Passive Atmospheric Sounding) (Fischer et al., 2008) on ESA's Envisat research satellite ceased to operate in its full spectral resolution measurement mode.Instead, MIPAS has been operated at reduced spectral but improved spatial resolution since then.While a data product containing vertical profiles of temperature and six trace species was generated by an operational processor (Ridolfi et al., 2000;Raspollini et al., 2006) under ESA responsibility and routinely distributed for the full spectral resolution data, only a limited official reduced spectral resolution data set is available by now (Ceccherini et al., 2008).Instead, research groups which had complemented the operational data product by a scientific data product already when MIPAS was operational at full spectral resolution, have attempted to fill the gap and to provide vertical profiles of temperature and several trace species.One of the data processors used to generate such a scientific data product is the one built and operated in a joint effort by the Institut für Meteorologie und Klimaforschung (IMK) at Forschungszentrum Karlsruhe and the Instituto de Astrofísica de Andalucía (IAA).Recently, the retrieval of temperature, O 3 , HNO 3 , N 2 O and H 2 O from reduced spectral resolution so-called "upper troposphere -lower stratosphere #1 (UTLS-1) measurement mode" spectra has been published by Chauhan et al. (2009).In this paper, retrievals of temperature, H 2 O, O 3 , CH 4 , N 2 O, HNO 3 , ClONO 2 , and ClO from spectra recorded in the "nominal reduced resolution measurement mode", the measurement mode in which T. von Clarmann et al.: MIPAS reduced resolution measurements MIPAS is operated most of its measurement time, are presented and discussed in terms of error analysis and vertical as well as horizontal resolution, and the temporal development of these atmospheric state variables is presented.

Measurements
MIPAS performs limb emission measurements from a sunsynchroneous polar orbit at about 800 km altitude.Envisat's equator passing time is about 10 a.m.local time on the southbound leg of the orbit.In the so-called reduced resolution nominal measurement mode, one limb sequence is recorded every 410 km.This horizontal sampling leads to about 100 limb sequences per orbit, depending on time needed for calibration measurements.At polar regions, the nominal tangent altitude spacing is 1.5 km from 6-21 km, 2.0 km from 21-31 km, 3.0 km from 31-46 km and 4 km up to the uppermost tangent altitude which is 70 km (cf.http://www.atm.ox.ac.uk/group/mipas/rrmodes.html).Towards lower latitudes, the entire scan pattern is moved upward by up to 3 km.Since this pattern is denser than the vertical extent of the instantaneous field of view of MIPAS at the tangent point, the atmosphere is oversampled in the vertical dimension.The spectral coverage of MIPAS is 4.15 to 14.6 µm (685-2410 cm −1 ) at 0.121 cm −1 spectral resolution (in terms of full width at half maximum, after application of the Norton Beer "strong" (Norton and Beer, 1976) apodization, corresponding to a maximum optical path difference of 8.2 cm).Data presented in this paper cover the period 27 January 2005 to 8 April 2008, however not at complete temporal coverage.These data are based on 139 824 limb scans recorded during 1626 Envisat orbits.

Retrieval
The retrieval strategy applied to the reduced resolution measurements is similar as the one applied to the MIPAS full spectral resolution measurements and has been extensively discussed by von Clarmann et al. (2003).In summary, it is based on constrained inverse modelling of limb radiances.

Representation of the atmospheric state
The atmosphere is represented on a discrete altitude grid (1km gridwidth from 4-44 km altitude, 2-km gridwidth from 44-70 km, 5-km gridwidth from 70-120 km, with some additional levels for temperature and H 2 O).This grid is a compromise between accuracy and efficiency and has been developed during numerous test retrievals.Depending on the retrieved variable, the formal retrieval quantity is either temperature, volume mixing ratio (vmr) or log (vmr).In case of log(vmr) retrievals particular care is needed when the data are used in a statistical context.Arithmetic averaging will bias the mean values high, if the precision is close to or worse than 100%.Particularly problematic are averages where the expectation value is zero but the true measurements scatter around zero due to measurement noise.In a linear retrieval these noisy values would in tendency cancel out to zero, while arithmetic averaging of results obtained by logretrievals would, by suppression of negative vmr values but maintaining the positive ones, cause a pronounced bias.Averaging of log-retrievals in the log-domain does not fully remedy the problem because it does not properly account for the natural variability in the sample.The decision for or against the log(vmr) retrieval is made for each species separately under consideration of the gas-specific advantages and drawbacks.Unless explicitly mentioned, the retrievals discussed in this paper are vmr retrievals in the linear domain.
Usually, during analysis of one limb scan the atmosphere is assumed to be horizontally homogeneous with respect to mixing ratios of constituents but horizontally inhomogeneous with respect to temperature.For the latter, a linear horizontal variation was allowed in a range of ±400 km around the nominal geolocation of the limb scan.As already found in the context of the most recent full spectral resolution measurement mode retrievals, the consideration of horizontal temperature gradients generally improves the accuracy of the retrieval, and in many cases it helps to reduce the number of convergence failures.Particularly near the polar vortex boundary many retrievals did not converge when horizontal temperature gradients were ignored.
For reasons extensively discussed in von Clarmann et al. (2003), besides the target quantity, also an additive offset calibration correction per limb scan and per spectral analysis window and a profile of atmospheric background continuum absorption coefficients per spectral analysis window were jointly fitted.While the IMK-IAA MIPAS data processor supports consideration of non-local (non-LTE) thermodynamic equilibrium emissions (Funke et al., 2001(Funke et al., , 2005)), this is not taken into account for retrieval of species discussed in this paper; instead, the spectral analysis windows were chosen such that non-LTE radiation is best possibly excluded (cf.Sect.3.4).For the gases discussed in this paper, the same strategy had also been applied to the IMK-IAA full spectral resolution nominal mode retrievals.

Sequence of operations
The first step of the retrieval chain is determination of the spectral shift of the measurements.Secondly, temperatures and altitude pointing information (i.e. the elevation of the line of sight of the instrument) are jointly retrieved.The following sequence of operations is H 2 O, O 3 , HNO 3 , CH 4 along with N 2 O, then ClONO 2 , and ClO.As a general rule, results of preceding steps are used as input for the subsequent retrieval steps, i.e. the H 2 O retrieval uses retrieved temperatures and pointing information, and the subsequent O 3 retrieval uses retrieved H 2 O abundances, etc.This approach implies that major contributors to the infrared spectrum are analyzed first, while analysis of gases with tiny spectral features is postponed until the mixing ratios of the major contributors have been retrieved.

Regularization
Contrary to the operational MIPAS level-2 processor operated by ESA for analysis of the original full spectral resolution measurements (Ridolfi et al., 2000;Raspollini et al., 2006), which performs maximum likelihood retrievals on a vertical grid defined by the MIPAS tangent altitudes, the IMK-IAA processor performs regularized retrievals on a finer altitude grid (1 km gridwidth in the troposphere up to the middle stratosphere; for details, see Sect.3.1).Thus, stable solutions can only be obtained by regularization.While other MIPAS processors (Burgess et al., 2004(Burgess et al., , 2006;;Hoffmann et al., 2008) regularize by the maximum a posteriori (also known as "optimal estimation") method (Rodgers, 2000), the IMK-IAA processor uses a smoothing constraint, which operates by weighted minimization of the squared first order finite differences of adjacent profile values, using a Tikhonov (1963) formalism.The intent of this choice is to make the resulting profiles less dependent on the a priori profiles.Details are reported under the respective species.Adjustment of the -in many cases altitude-dependent -strength of the regularization, during a series of test retrievals was driven by the aim to achieve the best possible vertical resolution in the upper troposphere and lower stratosphere at still stable retrievals without unphysical oscillations of the resulting profiles.All trace gases are regularized against an altitude-constant a priori profile.Its actual value is meaningless, since the smoothing regularization applied here affects the shape of the profile only but not the actual values.The altitude-constant choice of the a priori profile avoids artefacts when the altitude of a stratospheric peak or a sharp bend near the tropopause is located at a different altitude in the a priori data compared to the true atmosphere.This approach implies that the resulting profile is, apart from deviations due to measurement, parameter or model errors, a smoothed version of the true profile without bias or further distortion.
The only variables regularized by maximum a posteriori approach are the elevation of the lines of sight (von Clarmann et al., 2003) and the horizontal temperature gradients.The latter (but not the temperatures themselves!) are constrained with a zero order Tikhonov-type regularization towards ECMWF data, which is equivalent to a maximum a posteriori formalism (Rodgers, 2000) with a diagonal ad hoc a priori covariance matrix.The latter was set such that for high altitudes, where ECMWF analysis data are thought to be less reliable, the information on the horizontal temperature gradients is extracted mainly from the measurements, while for lower altitudes, where a single limb scan contains little information on the horizontal temperature distribution but where ECMWF analyses are more reliable, the retrieval is strongly constrained towards the a priori.

Microwindows
Microwindows are narrow spectral intervals suited for retrieval of temperature or species abundances profiles.Their optimal boundaries are found by minimization of the estimated total error (von Clarmann and Echle, 1998).Starting from a set of automatically generated microwindows (Echle et al., 2000), the microwindow selection used here was modified in order to better remove impact of non-LTE which might have been underestimated in the initial microwindow selection, reduce the impact of further interfering species not yet considered then, and avoid spectral lines whose spectroscopic data prove inconsistent with those of the majority of lines.Details are reported under the respective species.The microwindow selection is listed in Tables 1a-1c.

Diagnostics
IMK-IAA MIPAS results are characterized by error estimates, as well as vertical and horizontal averaging kernels.The latter two are used to estimate the spatial resolution of the retrievals.

Error analysis
The retrieval error is reported in terms of noise and parameter errors.The former is the mapping of measurement noise onto the retrieval, the latter is the propagation of uncertain parameters (for example the abundance of interfering species) onto the result.The mapping of each of these primary errors onto the retrieved quantity is estimated by application of Gaussian error propagation.The total error in terms of variances is the sum of the error variances of the noise error and all parameter errors.Precision is the total random error including measurement noise and those parameter errors which vary randomly in the time domain.Instrument line shape errors and uncertainties of spectroscopic data are systematic and thus contribute to the total error but not to the precision.Noise is estimated for each individual profile, while the full error analysis is performed for representative profiles only.We assume the following primary errors: 1% for gain calibration uncertainty, 3% uncertainty of the instrument line shape in terms of linear loss of modulation efficiency toward the maximum optical path difference, 150 m elevation pointing uncertainty, 2 K temperature uncertainty (conservatively estimated in order to also account for effects due to suboptimal horizontal representation of the temperature field).Uncertainties of abundances of interfering species, whenever available from a preceding retrieval were accounted for following their dedicated error estimates.When no preceding retrievals were available, and climatological profiles had to be used, uncertainty estimates by Remedios et al. (2007) were used.Noise equivalent spectral radiance typically is 11-19 nW/(cm 2 sr cm −1 ) in MIPAS band A (695-970 cm −1 ), relevant to the retrieval of temperature, H 2 O,  3125-745.5000 808.1875-808.7500 763.5625-764.3125 748.9375-749.8125 825.1250-825.4375 766.8750-767.1875 765.8750-766.5625 827.3750-827.8125 776.1875-776.5000 780.4375-780.6250 1224.7500-1225.2500 777.6875-777.9375 798.1250-798.5000 1239.1250-1239.4375 781.0000-781.7500 810.8125-811.0625 1244.0000-1244.4375 782.5000-782.8750 812.2500-812.5625HNO 3 , ClONO 2 and ClO; 8.5 nW/(cm 2 sr cm −1 ) in band AB (1020-1170 cm −1 ), relevant to the retrieval of O 3 and 6 nW/(cm 2 sr cm −1 ) in MIPAS band B (1215-1500 cm −1 ), relevant to the retrieval of H 2 O, CH 4 and N 2 O, all values after apodization.The following uncertainties in spectroscopic data were assumed (based on estimates by J.-M.Flaud, personal communication, 2008): the uncertainties of line intensities and broadening coefficients, respectively are: for CO 2 3 to 5% (at the band head, depending on the vibrational band, and increasing towards higher rotational quantum numbers) and 6%; for H 2 O 3-15% and 20%; for O 3 2-30% (the latter value for some hot bands) and 12%; for HNO 3 8-15% and 25%; for CH 4 2-8% and 25%; for N 2 O 4-6% (at the band head, depending on the vibrational band, and increasing towards higher rotational quantum numbers) and 7%; and for ClO 15%.The uncertainty of ClONO 2 absorption crosssections is estimated at 5%.The smoothing error (Rodgers, 2000) is not included; we report the spatial resolution instead.The reason for this choice is that the smoothing error depends on the actual variability of the respective state variable, on which for some species no robust knowledge is available.a Not all spectral gridpoints are used at each altitude.

Vertical averaging kernels
The vertical averaging kernel is the derivative of the retrieved profile with respect to the true profile (Rodgers, 2000).Any deviation from unity of the averaging kernel matrix characterizes the influence of the a priori information used to constrain the retrieval.When a retrieval grid is used which is finer than the altitude sampling of the measurement and no significant altitude information is contained in the spectral line shape, the largest values to be expected for an ideal retrieval where regularization compensates for vertical oversampling only are approximately the quotient of retrieval gridwidth over measurement gridwidth.We use the halfwidth of the row of the averaging kernel matrix to characterize the vertical resolution of the retrieval at the respective altitude.Due to regularization, the vertical resolution is usually wider than the tangent altitude spacing, and certainly wider than the grid on which the retrieval is performed (Table 2).

Horizontal averaging kernels
Most atmospheric state variables are assumed to vary only with altitude, while during the analysis of one limb scan no horizontal variation of the atmospheric state has been considered.The only exception is temperature, where a linear horizontal variation is allowed to avoid convergence problems at the boundary of the polar vortex.The assumption of local horizontal homogeneity of vmr causes horizontal smoothing, which is characterized by the horizontal averaging kernel (von Clarmann et al., 2008).This method has been extended such that a gain function with regularization term could be used and the retrieval grid could be defined independent of the tangent altitude grid.Consideration of regularization in the altitude domain complicates the assessment of horizontal smoothing, because, contrary to the simpler case of unconstrained maximum likelihood retrievals, there is interaction of horizontal and vertical smoothing.We characterize the horizontal smoothing in a 2-dimensional grid of i altitudes and j latitudes and by the i×j horizontal averaging kernel matrix A hor .The k-th row of A hor represents the weights of all latitudes1 1 to j contributing to the result assigned to the altitude k and the nominal latitude of the 1-dimensional profile retrieval.Each element a hor;k,l is calculated from the full two-dimensional i×(i×j ) averaging kernel matrix A 2D by summation over those elements a 2D;k,m of A 2D which represent the lth latitude.The horizontal averaging kernels give a measure of the horizontal smearing characteristics of the MI-PAS retrievals.For more quantitative applications like construction of observation operators for data assimilation applications, the use of the full two-dimensional averaging kernels is recommended.Unless limited by horizontal sampling, the horizontal resolution of the MIPAS retrievals can be estimated by the full width at half maximum of the row of the horizontal averaging kernel matrix (Table 3).The information displacement can be estimated as the horizontal distance of the median of the horizontal averaging kernel and the nominal geolocation of the limb scan.In case of MIPAS, the latter is the geolocation of the tangent point of the measurement at the middle tangent altitude of the limb scan.We use a sign convention that positive displacement is displacement towards the satellite (Table 4).

Temperature and line of sight
The retrieval strategy for temperature and elevation of the line of sight (LOS) from CO 2 emissions follows the one applied to MIPAS full resolution spectra (von Clarmann et al., 2003) except for the following upgrades: The retrieval grid has been modified (1-km gridwidth from 4-50 km altitude, 2-km gridwidth from 50-70 km, 2.5-km gridwidth from 70-80 km, 5.0 km gridwidth from 80-110 km, and a 10 km step from 110 to 120 km).Also parameter settings relevant to numerical integration over wavenumber, instrument field of view etc. within the radiative transfer forward model have been re-adjusted in order to obtain an optimal trade-off between accuracy and efficiency.All microwindows used are located in the shortwave wing of the 15 µm CO 2 band system (Table 1a).Some gridpoints within the microwindows which were particularly sensitive to interfering species were discarded, and the selection of microwindows is altitudedependent, i.e. a particular microwindow is not always used for all tangent altitudes.Compared to the original temperature retrievals published by von Clarmann et al. (2003), microwindows in the 15 µm band system below 700 cm −1 , where the spectrum is fairly saturated, and the microwindow in the CO 2 laser band, where non-LTE is a particular issue, have been discarded in favour of further microwindows in the 730-810 cm −1 spectral range.The current microwindow selection is not a modification of the old one but has been newly constructed from scratch for the UTLS-1 reduced resolution mode retrievals by Chauhan et al. (2009) and further modified for reduced resolution nominal mode measurements.
ECMWF analysis and MIPAS engineering information is used as a priori information for temperature and line of sight elevation, respectively.Along with the temperature profile, a profile of latitudinal horizontal temperature gradients is retrieved, such that a temperature field is retrieved which varies linearly in the horizontal.This linear variation is confined to ±400 km around the nominal geolocation of the limb scan under analysis.For the horizontal temperature gradients, the ECMWF analysis is used as a priori information, and for regularization the maximum a posteriori approach (Rodgers, 2000) has been chosen.
Further, the climatological CO 2 profiles are trendcorrected.CO 2 ground values are calculated as and the vertical profile is scaled by the ratio of the actual ground value and the reference value of the year 2004.The original full-resolution data were validated by Wang et al. (2005).The MIPAS UTLS-1 retrievals have been compared to Microwave Limb Sounder (MLS) data by Chauhan et al. (2009); the retrieval setup of the UTLS-1 and the re-duced resolution nominal mode retrievals differ only in a few details (see below), and the differences of the datasets are the wider altitude coverage of the nominal measurements and the finer horizontal sampling of the UTLS-1 measurements.The dominating error source below 30 km altitude is the gain calibration error, while at higher altitudes instrument line shape (ILS) uncertainties are the largest error component.Further important error sources are uncertainties in the assumed abundances of N 2 O 5 and O 3 as well as measurement noise.Spectroscopic data account for an additional estimated uncertainty of 0.1-0.6K. Compared to MIPAS full spectral resolution temperature retrievals, the measurement noise is slightly reduced, while the other error components are of comparable size.
Figure 1 shows the vertical averaging kernels of this particular retrieval.The altitude resolution is between 2 and 3.5 km between 10 and 45 km altitude (Table 2), which is slightly better than that of MIPAS full spectral resolution measurements, for which the altitude resolution of temperature was estimated at 4 km in this altitude range.
The horizontal smoothing varies between about 128 km (at 10 km altitude, Table 3) and 499 km at 13 km altitude, with values around 300 km throughout wide parts of the According to the sampling theorem, all periodic atmospheric structures of wavelengths shorter than 2×410 km are prone to aliasing, unless the limited horizontal resolution of the measurement filters out these short-wavenumber components.For the actual horizontal resolution of the MIPAS retrievals, however, the horizontal smearing is not sufficient to act as anti-aliasing filter.The risk of aliasing has been reported for full-resolution MIPAS retrievals by von Clarmann et al. (2008).For reduced resolution nominal mode measurements, however, this risk is not as large as with the measurements taken in the reduced resolution UTLS-1 measurement mode (Chauhan et al., 2009), where the horizontal smoothing width is at some altitudes smaller than the horizontal sampling by a factor of about three.The aliasing problem is of general nature and does affect retrievals of all state variables of the atmosphere whose horizontal averaging kernels are narrow enough.It is not limited to temperature retrievals.At 10 km the information displacement of the reduced resolution nominal mode measurements is larger than the horizontal smoothing width (Table 4), which means that the measurement contains little information about the temperature at the nominal geolocation of the measurement but on geolocations closer to the sub-satellite point.
Figure 2  band, where non-LTE of CO 2 turned out to be a problem have been discarded (Table 1a).Instead, some microwindows have been added in the 1360-1375 cm −1 spectral region.In addition, the microwindow boundaries have been adjusted to the spectral resolution and sampling of the reduced resolution measurements.Measurements in MIPAS band C (1570-1750 cm −1 ) covering the 6.3 µm-band of H 2 O have finally been discarded because of problems with non-LTE and interference by NO 2 .Discard of these transitions unfortunately leads to reduced sensitivity of the H 2 O retrieval above 50 km, which is why these transitions were considered in some previous data versions.H 2 O spectroscopic data are taken from the most recent version of the dedicated MIPAS spectroscopy database (Flaud et al., 2006).As for all other gas retrievals, horizontal temperature gradients are taken into account.
While for most other trace gases the vmr is retrieved directly, for H 2 O we retrieve log(vmr).This allows for an altitude-, latitude-and season-independent a priori profile   and regularization strength although the variability of water vapour is extremely large particularly in the upper troposphere.On the negative side, the averaging kernels refer to log(vmr) and thus are not as easy to interpret, and averages of H 2 O values may be biased high, particularly if in a dry atmospheric region the spread is caused primarily by the propagation of measurement errors rather than natural variability.Further, the H 2 O profile is sampled on a 10-km grid only above 80 km.Beyond these changes in the H 2 O retrieval setup, certainly all changes in the setup of the temperature and line of sight retrieval also affect the subsequent H 2 O retrieval and thus have to be considered also in the H 2 O data version management.This version-interdependence does not only apply to H 2 O but to all retrievals of species which use results of preceding retrievals.
The total retrieval error, evaluated for the limb scan recorded at 61 • N, 124 • W on 27 January 2005, varies between 0.3 and 1.1 ppmv (Table 6) and is dominated by uncer-tainties of spectroscopic parameters which contribute to the systematic error, while the precision is much better, except for the stratopause region where the error budget is dominated by measurement noise.
The altitude resolution as inferred from the rows of the vertical averaging kernels (Fig. 3) varies from about 2.3-3.5 km at the lower stratosphere and about 3-4 km at 30 km, and coarser above (Table 2).Particularly at lower altitudes, the altitude resolution of the reduced spectral resolution nominal measurements is better than that of the full resolution measurements, which varies between 3 and 4 km at altitudes between 10 and 40 km for V3o H2O 13, but coarser than the MIPAS reduced resolution UTLS-1 mode results presented by Chauhan et al. (2009).
The horizontal smoothing varies between about 100 km and 500 km, depending on altitude (Table 3) and is generally better than that of the latest full spectral resolution retrievals (Milz et al., 2009).At most altitudes the horizontal resolution is limited by sampling rather than smoothing, i.e. the atmosphere is undersampled in the horizontal.At all altitudes except the lowest, the information displacement (Table 4) is dominated by the displacement of the actual tangent points of the limb scan with respect to the nominal geolocation of the limb scan.As discussed in von Clarmann et al. (2008), opacity shifts the source region of information towards the satellite at lower altitudes, while a displacement in the opposite direction at higher altitudes is explained by better sampling of the atmosphere behind the tangent point.The erratic negative value at 10 km does not describe the source of information well because the horizontal averaging kernel is quite an irregular distribution with considerable negative values near the nominal geolocation.Thus, it is not easily characterized by its median.The centroid of information, e.g., is +199 km at this altitude.
The MIPAS full spectral resolution retrievals, versions up to V3o H2O 13 (Milz et al., 2009) have been validated against numerous independent measurements and the reduced resolution UTLS-1 retrievals were compared to MLS by Chauhan et al. (2009).However, the V3o H2O 13 full  spectral resolution microwindows still include transitions in the 6.3 µm region which have been discarded for the reduced spectral resolution analysis.This suggests that a potential bias due to non-local thermodynamic equilibrium emission should be reduced in the reduced resolution nominal mode data set presented here.
Figure 4 shows the latitudinal water vapour distribution at 20 km altitude from February 2005 to April 2008.Most prominent features are dehydration events in Southern polar spring, humid midlatitude winters and dry tropical summers, in particular summer 2006, at this altitude.

O 3
Ozone retrievals based on MIPAS full resolution spectra have been extensively studied by Glatthor et al. (2006) and validated by Steck et al. (2007).Since then, the following changes in retrieval strategy have been made: The microwindow selection (Table 1a) has been newly performed from scratch, under consideration of the reduced spectral resolution and modified measurement scenario.While the spectral regions where suitable lines are found remain the same, different ozone transitions have been chosen.As for the MIPAS full spectral resolution retrievals, O 3 spectroscopic data are taken from the most recent version of the dedicated MIPAS spectroscopy database (Flaud et al., 2006).Except for the eight highest altitudes, the strength of the regularization has been chosen independent of altitude.
Also for ozone version V4o O3 202, the total error budget, evaluated for the limb scan recorded at 61 • N, 124 • W on 27 January 2005, is dominated by uncertainties of spectroscopic parameters, which account for a relative error of more than 10% at most altitudes.The precision varies be-   tween about 50 and 275 ppbv and is limited by gain calibration uncertainties (Table 7).
The altitude resolution as inferred from the rows of the vertical averaging kernels (Fig. 5) is about 2.5-3.5 km (Table 2), and is slightly better than that of the latest full spectral resolution retrievals (3.5-5 km between 10 and 40 km altitude for version V3o O3 9).The horizontal smoothing varies between about 250 km and 400 km and is smaller than the horizontal sampling width of 410 km (Table 3).The horizontal information displacement (Table 4) follows roughly the tangent point displacement, with the usual positive and negative deviations due to opacity and sampling.
Figure 6 shows the latitudinal ozone distribution at 20 km altitude from February 2005 to April 2008.The annual cycle

HNO 3
The reduced resolution nominal mode HNO 3 retrievals version V4o HNO3 201 are based on full spectral resolution retrievals as published by Mengistu Tsidu et al. (2005), who already used altitude dependent regularization strength (Steck and von Clarmann, 2001;Steck, 2002).Since then, the regularization strength, while still altitude-dependent, has been modified and the microwindows have been adjusted to the spectral sampling of the reduced resolution measurements (Table 1b).However, the same spectral transitions are used as Mengistu Tsidu et al. (2005) used for full spectral resolution measurements.Thus, the lessons learned from the validation of the full spectral resolution HNO 3 retrievals by Wang et al. (2007) should be applicable also to the V4o HNO3 201 data.HNO 3 spectroscopic data are taken from the dedicated MIPAS spectroscopy database (Flaud et al., 2006).The HNO 3 total retrieval error, evaluated for the limb scan recorded at 61 • N, 124 • W on 27 January 2005, varies between 4.4 and 17.2% and is noise-dominated below 15 km and above 35 km.In the middle stratosphere the precision is excellent (2.4 to 4.9%) and spectroscopic uncertainties are the major contributor to the total error (Table 8).
The altitude resolution as inferred from the rows of the vertical averaging kernels (Fig. 7) is better than 3.5 km in the lower stratosphere and degrades with altitude towards 6 km at 40 km altitude (Table 2).It is better than that of the latest full spectral resolution retrievals, V3o HNO3 9 at altitudes below 25 km (on average 4 km) but similar above.The horizontal smoothing width varies between about 250 to 500 km  (Table 3) and the horizontal information displacement behaves as expected (Table 4).
Figure 8 shows the latitudinal HNO 3 distribution at 20 km altitude from February 2005 to April 2008.The annual cycle is clearly visible.Denitrification is visible in all Southern polar winters and in northern polar winter 2007/2008.

CH 4 and N 2 O
As for the full spectral resolution retrievals, also V4o CH4 201 and V4o N2O 201 versions of CH 4 and N 2 O are retrieved jointly in order to minimize mutual error propagation.In the boundary region of the polar vortex severe convergence problems were encountered before horizontal temperature gradients were considered, and convergence is significantly helped by consideration of these.As for the MIPAS full spectral resolution retrievals (Glatthor et al., 2005), microwindows in the 1220-1305 cm −1 spectral regions are used (Table 1b).For some transitions, only the boundaries of the original full spectral resolution microwindows were adjusted, others were exchanged.The goal of these microwindow changes was to reduce the known high bias of CH 4 but this was only partly successful.N 2 O spectroscopic data are taken from the HITRAN2K database, while for CH 4 the HITRAN update of 2002 was used (Rothman et al., 2003).
The estimated total retrieval error of CH 4 , evaluated for the limb scan recorded at 61 • N, 124 • W on 27 January 2005, varies between 13 and 34%, and uncertainties in spectroscopic data are the dominating error source (Table 9).The estimated precision is in the order of 5 to 8%, and degrading to 14% towards the stratopause.For N 2 O the total retrieval error varies between 8 and 16% in major parts of the    45.0 (2.8) 14.0 (0.9) 91.9 (5.6) 12.0 (0.7) 31.0 (1.9) 97.7 (6.0) stratosphere and reaches 30.5% at the stratopause.For N 2 O the dominating error source at altitudes up to 35 km is uncertainties in spectroscopic data, while above measurement noise is the leading source of error (Table 10).
Both the full and the reduced spectral resolution CH 4 and N 2 O retrievals with the IMK-IAA processor are known to be biased high, particularly in the lowermost stratosphere (Glatthor et al., 2005).This bias, which, however, is within the error margins associated with uncertainties in spectroscopc data, is smaller in the reduced resolution data compared to the full spectral resolution retrievals.
The altitude resolution as inferred from the rows of the vertical averaging kernels (Figs. 9 and 11) is around 2-3 km in the altitude region between 15 and 35 km for both species.At the tropopause region and in the upper stratosphere it is 4 km or even worse (Table 2).At these altitudes the altitude resolution of full spectral resolution retrievals V3o N2O 11 and V3o CH4 11 is 4 km and 6 km, respectively.The horizontal smoothing varies between about 140 and 470 km for CH 4 and between 190 and 540 km for N 2 O (Table 3).For CH 4 , the horizontal displacement behaves roughly as expected, while for N 2 O there is an erratic value (421 km) at 30 km altitude (Table 4).In this case, also the centroid of information is quite large (246 km).The horizontal averaging kernel is bimodal, with a broad maximum at +450 km displacement and secondary maximum around zero displacement.While the  latter are associated with N 2 O emissions near the tangent point, the largely displaced information is associated with tropospheric tangent altitudes.
Figure 10 shows the Southern polar (averaged over 60.0    subsides into the stratosphere during each of the three Southern winters.Then, after the vortex breakdown, CH 4 -rich air is transported to polar latitudes in the lower stratosphere.Also in the northern polar winters subsidence occurs in each polar winters, as seen in the N 2 O temporal development (Fig. 12).Rapid enhancements, are attributed to final warmings (March/April 2005-2007) or a midwinter warming (January 2006, Manney et al. 2008).At altitudes above 40 km, subsided air rich in N 2 O produced in the mesosphere (Funke et al., 2008) might also have contributed to N 2 O enhancements.

ClONO 2
For ClONO 2 the same spectral analysis window as for the original full spectral resolution retrievals (Höpfner et al., 2004a(Höpfner et al., , 2007) ) was used, which includes the Q branch of the ClONO 2 ν 4 band.Absorption cross-sections by Birk and Wagner (2000) are used.Although O 3 retrievals are available from a previous retrieval step, O 3 is jointly fitted to avoid problems with misfitted lines at 779.6, 779,8, 780.15, 780.4, 780.6 and 780.8 cm −1 which might propagate errors to the ClONO 2 vmr.Compared to the original data version V2 ClONO2 1 (Höpfner et al., 2004b), later versions, e.g. the full spectral resolution data sets V3o ClONO2 10/11 validated by Höpfner et al. (2007) and the current reduced spectral resolution data set V4o ClONO2 201 uses a weaker regularization at higher altitudes, since the original retrieval was found to smear out ClONO 2 variability in the middle and  upper stratosphere.The analysis window has remained unchanged (Table 1b).
The estimated retrieval error, evaluated for the limb scan recorded at 61 • N, 124 • W on 27 January 2005, varies between 35.5 and 105.4 pptv and is dominated by measurement noise (Table 11).The altitude resolution as inferred from the rows of the vertical averaging kernels (Fig. 13) is 2.5-3.5 km in the lowermost stratosphere, and degrades towards about 9 km at 40 km altitude (Table 2).In contrast, the altitude resolution of the full spectral resolution retrievals, version V3o ClONO2 12 are never better than 3 km.The horizontal smoothing length of the reduced resolution nominal mode ClONO 2 measurements varies between 250 and 510 km for the test case under analysis (Table 3).The information displacement (Table 4) is largely dominated by the displacement of the actual tangent points.The low displacement at high altitudes reflects the effect of vertical regularization: The measurements on the uppermost tangent altitudes contain nearly no information on ClONO 2 , and the vertical averaging ker- nel peaks below the nominal altitude of 50 km (cf. the violet line in Fig. 13).This vertical information shift implies also a horizontal information shift, because lower tangent points are closer to the satellite than higher ones.Figure 14 shows a latitudinal distribution of ClONO 2 as a function of time.ClONO 2 reaches its maximum in northern polar spring.Since denitrification is stronger in the Southern polar vortex, and since ozone is largely depleted, leading to low ClO abundances, there is very little ClONO 2 in the core region of the vortex there.ClONO 2 is formed in the collar of polar vortices rather than the core region.As a consequence, at Southern polar latitudes the region of highest ClONO 2 mixing ratios is shifted from the pole towards midlatitudes.Wave activity moving the polar vortex off the pole masks this effect in the Northern Hemisphere.

ClO
For ClO the same spectral analysis windows were used as for the original full spectral resolution retrievals (Glatthor et    2004) (Table 1c), which includes lines in the P and Q branch region of the ClO 1-0 band.As for ClONO 2 , O 3 is jointly fitted in order to reduce related error propagation.The same regularization as for the original data version V2 ClO 1 and more recent full spectral resolution ClO data versions up to V2 ClO 11 was used.The total retrieval error, evaluated for a limb scan recorded at 57 For this case of unperturbed chemical conditions, i.e. without chlorine activation, the retrieval error exceeds the 100% limit at altitudes below 20 km and is never better than 34% above (Table 12).This means that under unperturbed conditions stratospheric ClO cannot be detected in the lower stratosphere from single limb scans, and the information in the upper stratosphere is limited.However, the significance of the results can be enhanced at the cost of spatial and temporal resolution by averaging of multiple limb scans.The altitude resolution as inferred from the rows of the vertical averaging kernels (Fig. 15) is about 3-7 km in the lower stratosphere (Table 2), and the horizontal resolution is in the order of 500 km (Table 3).The effect of horizontal information displacement at high altitudes (Table 4) is even more pronounced as for ClONO 2 but the same explanation applies.
Figure 16 is a latitudinal distribution of daytime ClO at 20 km as a function of time.Chlorine activation is clearly visible in Southern polar spring, and, to a lesser extent, also in northern polar late winters.Pronounced scatter and negative values at tropical latitudes reflect noisy data, since atmospheric background ClO is below the MIPAS detection limit.

Conclusions
We have presented the IMK-IAA retrievals of atmospheric state variables from MIPAS reduced resolution nominal mode measurements.The MIPAS reduced resolution nominal mode IMK-IAA retrievals are, along with related diagnostics, available to registered users via http://www.fzk.de/imk/asf/sat/envisat-data.For the species investigated in this paper, the reduction of spectral resolution seems to be rather an advantage than a drawback, because, as a consequence of denser vertical sampling of the measurements, the vertical resolution has become better.Since overlapping fields of view cause oversampling, also the retrieval error is reduced in many cases.The characteristics of data is roughly comparable to those of the reduced resolution UT/LS measurement mode retrievals (Chauhan et al., 2009), except for the horizontal smearing, which is larger for the "reduced resolution nominal mode" measurements due to the extended altitude range of the latter.In turn, due to the extension of measurements to tangent altitudes above 49 km up to 70 km, the nominal mode retrievals obviously contain more information on the upper part of the atmosphere.However, the horizontal sampling of the measurements is -similar as for the full spectral resolution measurements (von Clarmann et al., 2008) -significantly coarser than the horizontal smoothing, which implies the risk of aliasing if short scale periodic atmospheric structures are present.The inclusion of horizontal temperature gradients is important to prevent failure of convergence of many retrievals, particularly near the boundary of the polar vortex and to improve the data quality.

Fig. 2 .Fig. 2 .
Fig. 2. Temporal development of temperature as a function of altitude in the inner tropics (averaged from 10 • S to 10 • N.) Fig. 2. Temporal development of temperature as a function of altitude in the inner tropics (averaged from 10 • S to 10 • N).

Fig. 3 .
Fig. 3. Rows of the averaging kernel matrix for H 2 O.For details, see Fig. 1.

Fig. 4 .
Fig. 4. Temporal development of H2O as a function of latitude at 20 km altitude.White areas in midlatitudes are data gaps due to calibration measurements.White areas in high southern latitudes are data gaps due to polar stratospheric clouds (PSC).

Fig. 4 .
Fig. 4. Temporal development of H 2 O as a function of latitude at 20 km altitude.White areas in midlatitudes are data gaps due to calibration measurements.White areas in high southern latitudes are data gaps due to polar stratospheric clouds (PSC).

Fig. 6 .
Fig. 6.Temporal development of O 3 as a function of latitude at 20 km altitude.White areas in southern midlatitudes are due to data gaps caused by regular calibration measurements, while white areas at high southern latitudes are data gaps due to polar stratospheric clouds (PSC).

Fig. 8 .
Fig. 8. Temporal development of HNO3 as a function of latitude at 20 km altitude.White areas in southern midlatitudes are due to data gaps caused by regular calibration measurements, while white areas at high southern latitudes are data gaps due to polar stratospheric clouds (PSC).

Fig. 8 .
Fig. 8. Temporal development of HNO 3 as a function of latitude at 20 km altitude.White areas in southern midlatitudes are due to data gaps caused by regular calibration measurements, while white areas at high southern latitudes are data gaps due to polar stratospheric clouds (PSC).

Fig. 10 .
Fig. 10.Temporal development of CH4 as a function of altitude in the southern polar region (averaged from 60 • S to 90 • S). White areas at high altitudes represent insignificant negative values.

Fig. 10 .
Fig. 10.Temporal development of CH 4 as a function of altitude in the southern polar region (averaged from 60 • S to 90 • S). White areas at high altitudes represent insignificant negative values.

Fig. 11 .
Fig. 11.Rows of the averaging kernel matrix for N 2 O.For details, see Fig. 1.

Fig. 12 .Fig. 12 .
Fig. 12. Temporal development of N2O as a function of altitude in the northern polar region (averaged from 60 • N to 90 • N.) Fig. 12. Temporal development of N 2 O as a function of altitude in the northern polar region (averaged from 60 • N to 90 • N).

Fig. 14 .
Fig. 14.Temporal development of ClONO2 as a function of latitude at 20 km altitude.White areas in southern midlatitudes are due to data gaps caused by regular calibration measurements, while white areas at high southern latitudes are data gaps due to polar stratospheric clouds (PSC).

Fig. 14 .
Fig. 14.Temporal development of ClONO 2 as a function of latitude at 20 km altitude.White areas in southern midlatitudes are due to data gaps caused by regular calibration measurements, while white areas at high southern latitudes are data gaps due to polar stratospheric clouds (PSC).

Fig. 16 .
Fig. 16.Temporal development of daytime ClO as a function of latitude at 20 km altitude.White areas are data gaps caused by regular calibration measurements, due to polar stratospheric clouds (PSC), or at areas where during polar winter no daytime measurements are available.

Fig. 16 .
Fig. 16.Temporal development of daytime ClO as a function of latitude at 20 km altitude.White areas are data gaps caused by regular calibration measurements, due to polar stratospheric clouds (PSC), or at areas where during polar winter no daytime measurements are available.

Table 2 .
Vertical resolution for selected altitudes, in terms of full width at half maximum of the row of the profile averaging kernels.

Table 3 .
Horizontal resolution in terms of full width at half maximum of the row of the horizontal averaging kernels.

Table 4 .
Information displacement, calculated as the distance between the median of the row of the horizontal averaging kernel and the nominal geolocation of the MIPAS limb.A positive sign represents displacement towards the satellite.The column "tp" represents the displacement explained by measurement geometry, i.e. the displacement of the tangent point from the nominal geolocation of the limb scan, which is the geolocation of the tangent point of the middle tangent altitude of the limb scan.Rows of the averaging kernel matrix for temperature, evaluated for the limb scan recorded at 61 • N, 124 • W on 27 January 2005.Selected altitudes are highlighted by colours.

Table 5 .
Temperature retrieval error budget.

Table 6 .
H 2 O retrieval error budget.

Table 10 .
N 2 O retrieval error budget.