Vertical profile of peroxyacetyl nitrate (PAN) from MIPAS-STR measurements over Brazil in February 2005 and the role of PAN in the UT tropical NOy partitioning

We report on the retrieval of PAN (CH 3 C(O)OONO 2 ) in the upper tropical from limb measurements by the remote-sensor MIPAS-STR on board the Russian high altitude research aircraft M55-Geophysica. The measurements were performed close to Arac¸atuba, Brazil, on 17 February 2005. The retrieval was made in the spectral 5 range 775–820 cm − 1 where PAN exhibits its strongest feature but also more than 10 species interfere. Especially trace gases such as CH 3 CCl 3 , CFC-113, CFC-11, and CFC-22, emitting also in spectrally broad not-resolved branches, make the processing of PAN prone to errors. Therefore, the selection of appropriate spectral windows, the separate retrieval of several interfering species and the careful handling of the water 10 vapour proﬁle are part of the study presented. The retrieved proﬁle of PAN has a maximum of about 0.14 ppbv at 10 km altitude, slightly larger than the lowest reported values ( < 0.1 ppbv) and much lower than the highest (0.65 ppbv). Besides the NO y constituents measured by MIPAS-STR (HNO 3 , ClONO 2 , PAN), 15 the situ instruments aboard the Geophysica provide simultaneous measurements of NO, NO 2 , and the sum NO y . Comparing the sum of in-situ and remotely derived NO + NO 2 + HNO 3 + ClONO 2 + PAN with total NO y a deﬁcit of 30–40% (0.2–0.3 ppbv) in the troposphere remains unexplained whereas the values ﬁt well in the stratosphere.

Interactive Discussion reaction of OH with hydrocarbons. After intermediate reactions involving acetaldehyde, the acetyl radical and molecular oxygen, the peroxyacetyl radical (CH 3 CO 3 ) is formed which further reacts with NO 2 to PAN (Singh, 1987). The lifetime of PAN in the lower troposphere is in the order of hours and is dominated by thermolysis. In the upper troposphere, the lifetime, dominated by photolysis, is of the 5 order of months, or even longer in dark Arctic regions (Talukdar et al., 1995;Kirchener et al., 1999). Median PAN/NO y ratios of more than 0.6 at altitudes from 4 km to 8 km have been observed in long-range transported Asian pollution plumes (Roberts et al., 2004). For details on the formation and distribution of PAN see Warneck (1999) and Finlayson-Pitts and Pitts (2000). 10 Although PAN concentrations as high as 0.65 ppbv (up to 8 km) have been observed (Roberts et al., 2004), its typically low concentrations (<0.1 ppbv) (Tanimoto et al., 1999) make it difficult to measure. Various in situ techniques have been used to determine the volume mixing ratios of PAN in the atmosphere. These are Fourier transform infrared spectroscopy (FTIR) (Stephens et al., 1956;Hanst et al., 1982), 15 gas chromatography with electron capture detection (GC/ECD) (Lovelock, 1961;Müller and Rudolph, 1989), gas chromatography with luminol-chemiluminescence detection (GC/LCD) (Gaffney et al., 1998), proton transfer reaction mass spectrometry (PTR-MS) (Hansel et al., 1995) coupled with a selected ion flow drift tube (SIFDT) method (Hansel and Wisthaler, 2000) and gas chromatography/negative ion chemical ioniza-20 tion mass spectrometry (GC/NICI MS) (Tanimoto et al., 2001).
Remote sensing in the infrared provides an alternative and independent method for the measurement of PAN. From occultation measurements of ACE-FTS onboard SCISAT-1 it was possible to retrieve PAN in a young biomass burning plume (Coheur et al., 2007). Recently, Remedios et al. (2007a) have shown the clear pres- 25 ence of the signatures of PAN in the emission spectra obtained by the balloon born MIPAS. Global upper tropospheric PAN distributions were derived from MIPAS/Envisat spectra by Glatthor et al. (2007).
In this paper we report the retrieval of PAN from measurements of MIPAS-STR  (MIPAS-STRatospheric aircraft, Piesch et al., 1996) an instrument operated on board the high-altitude aircraft Geophysica. The work was initiated by the observation that large differences exist in the upper troposphere (above 10 km) between MIPAS-STR measurements of HNO 3 and coincident in situ measurements of NO y -NO by SIOUX (StratospherIc Observation Unit for nitrogen oXides, Schmitt, 2003) also aboard the 5 Geophysica.
In the following we give a short description of the MIPAS-STR instrument and its measurement strategy, an overview of the flight of 17 February 2005 and compare HNO 3 data from MIPAS-STR with coincident in-situ measurements of (NO y -NO-NO 2 ) to obtain an upper limit PAN profile (Sect. 2). Further we give a simulation on the 10 feasibility of detecting PAN from the MIPAS-STR observations (Sect. 3), relevant general details on the data processing (Sect. 4) and finally the retrieval of PAN from the measured spectra, including the error estimation (Sect. 5). The last section gives a summary and a discussion of the results. 15

The MIPAS-STR instrument
MIPAS-STR is a cryogenic Fourier transform emission sounder operating in the middle infrared (Fischer and Oelhaf, 1996;Keim et al., 2004). The emission method allows limb and upward viewing, yielding about 2 km vertical resolution below the flight level (up to 20 km). Reduced vertical information above the flight level is obtained by upward 20 measurements with several elevation angles. The final results are 2-dimensional distributions of the trace gases along the flight track in an altitude range covering the lowest stratosphere and the upper troposphere.
The first deployment of MIPAS-STR was made during the Antarctic campaign APE-GAIA in 1999(Höpfner et al., 2000. The performance of the instrument has been 25 considerably improved in recent years. The pointing of the limb measurements has been operated at fixed tangent heights between 6 km and the flight altitude with a spacing of 1 km. Considering the instrumental field of view of 0.44 degrees (FWHM) over-sampling by a factor 2-3 was applied at the lower tangent heights. In addition upward measurements at elevation angles of 0, 1, 3 and 10 degrees, as well as zenith and cold blackbody (210 K) measurements were 5 performed. Two-sided interferograms were obtained with a maximum optical path difference L of 14.4 cm, resulting in an unapodised spectral resolution (1/2 L) of 0.035 cm −1 . For a flight altitude of 19 km the complete sequence, including calibration, takes 200 s. This results in a horizontal resolution in flight direction of about 36 km.
The data shown in this paper are obtained from channel 1, which covers the 10 wavenumber range of 770-970 cm −1 .

Flight scenario
The flight track of the Geophysica with the location of the tangent points of MIPAS-STR limb sequences is given in Fig. 1. From Araçatuba (21.2 • S, 50.4 • W) the flight was conducted northbound and returned south on a straight track from 14 • S to 23 • S. 15 Optically thick clouds were observed in the northern part of this leg which prevented trace gas retrieval for that region. However, cloudless condition were found in the southern part, just before the descent. The red rectangle in Fig. 1 indicates the tangent points of six cloud free limb sequences measured between 13:05 and 13:20 UTC. These six limb sequences cover the track of the aircraft on the descent, which started 20 at the southernmost point of the path. The flight thus gives an excellent opportunity to compare the MIPAS-STR profile with in situ data measured during descent.
where square brackets indicate concentrations, k(T) denotes the temperaturedependent rate coefficient of the reaction of O 3 with NO, and J NO 2 is the NO 2 photolysis frequency. The NO concentrations are taken from SIOUX measurements, the O 3 concentrations from FOZAN (Fast OZone ANalyzer, Ulanovsky et al., 2001), and the tem-5 peratures from a high-precision TDC (thermodynamic complex) sensor (Rosemount sensor customized at CAO, Central Aerological Observatory, Dolgoprudny, Russia). All three in situ instruments are aboard the Geophysica. The J NO 2 values are calculated with the radiative transfer model of Ruggaber et al. (1994). The vmr profile of ClONO 2 , also included in NO y , was retrieved from the MIPAS-STR measurements, but 10 due to its very low vmr (see Fig. 12) neglected in the comparison. The altitude of the cold point tropopause (see Fig. 11) is about 18 km. Below this altitude, NO y -NO-NO 2 is always higher than HNO 3 by up to 0.32 ppbv. In the following we investigate how much of this difference can be attributed to PAN. 15 A well suited band for mid-IR PAN analysis is located between 775 and 820 cm −1 (Glatthor et al., 2007;Remedios et al., 2007a). To indicate the contribution of different atmospheric trace species in this spectral region we show simulations performed with KOPRA ( Sitnikov et al., 2007) and FISH (Fast In situ Stratospheric Hygrometer, Zöger et al., 1999) aboard the Geophysica (see Fig. 5).

Spectral simulations for PAN
The sensitivity of the MIPAS-STR observation on PAN is demonstrated by plotting simulated difference spectra (with -without PAN) for various tangent heights between 15 8 and 18.6 km (see Fig. 4).
In small spectral regions, the information on PAN is reduced due to saturation caused by interfering trace gases. This is the case around the CO 2 Q-branch (792 cm −1 ) and at the position of strong CO 2 and H 2 O lines. Apart from these regions the radiance abates rather quickly with increasing tangent height. At 13, 14, and 15 km, the maximum 20 radiance is only 50, 25, and 12.5 nW/(cm 2 sr cm −1 ), respectively, comparable with the spectral noise (15 nW/(cm 2 sr cm −1 )) in the single MIPAS-STR spectra. However, with the high resolution spectra the broadly emitting PAN can be retrieved by multi-line retrieval from much lower radiances. In the present work 1171 independent spectral points were used to obtain a PAN profile. Level-1 processing of the MIPAS-STR data provides the input data for the subsequent profile retrieval. Basically, it converts raw interferograms of the atmospheric measurements stored during the flight into radiometrically calibrated atmospheric spectra for 5 each tangent height or elevation angle. The spectral gain and offset of the instrument were obtained from the zenith and cold blackbody measurements of each individual sequence. The zenith spectra were corrected for the contained atmospheric features.
Level-1 processing also provides the auxiliary data which are derived from the stored housekeeping information as well as from the line of sight calibration and the field of 10 view measurements made before and after the flight. The auxiliary data include information on the corrected flight altitudes, elevation-and azimuth angles, and relevant instrument parameters.

Level-2 processing
Vertical profiles of the atmospheric parameters (vmr of gases, temperature, pressure 15 and absorption/emission of aerosols) are retrieved by use of the atmospheric radiative transfer model KOPRA and its inversion algorithm KOPRAFIT. The profiles are iteratively changed to minimise the residuum between measured spectra and forward calculated spectra of a complete sequence. Regularisation of the profile shape against an a priori profile is necessary for each retrieved atmospheric parameter because the 20 chosen retrieval grid (0.5 km) is finer than the achievable vertical resolution.
In KOPRAFIT the Tikhonov-Philips regularisation method (Tikhonov, 1963;Phillips, 2003) was adopted: where i denotes the iteration index; x the vector with the unknowns; x a the a priori values; y the measurement vector; S y the measurement covariance matrix of y; f the forward model; K the spectral derivatives matrix; γ the regularisation parameter and L the first derivative regularisation operator. The regularisation strength is chosen as small as possible, just to avoid oscillations 5 in the resulting profile.
The achieved vertical resolution of the retrieved profile is the FWHM (full width at half maximum) of the columns of the averaging kernel matrix, given by: 4.3 PAN retrieval method 10 Here we describe the strategy used for the retrieval of PAN. To minimise the error contribution from spectral noise, we have averaged all spectra of the same tangent height/elevation angle within the six southernmost limb sequences (see Fig. 1), which reduces the noise from 15 to 6 nW/(cm 2 sr cm −1 ). Furthermore, we have used all spectral points between 775 and 820 cm −1 , with the exception of the region 790-794 cm −1 . 15 We excluded this interval to avoid any error on the retrieval from line-mixing of the CO 2 Q-branch located there. A summary of atmospheric parameters (12 species and temperature) that have been considered in the retrieval scheme is given in Table 1. Among those parameters, five species (CH 3 CCl 3 , CFC-113, CFC-22, CFC-11, and ClONO 2 ) have been determined 20 in steps previous to the PAN retrieval and are kept constant. The remaining profiles are fitted simultaneously with PAN.
ClONO 2 is fixed to the profile derived from the nearby ν 4 Q-branch in the interval 779.5-781 cm −1 . CFC-11 has been determined on the basis of the major band in the interval 838-856 cm −1 and CFC-22 has been obtained from its signature at 828.7- Interactive Discussion scaling factor corrects the profiles for the annual decrease. The vmr profiles of all five pre-determined species are plotted in Fig. 12. As a priori vmr profile for PAN (PAN a priori ) a mid-latitude profile of the MOZART model is used (see Fig. 6). Beside trace gases and temperature we determine a continuum extinction profile for aerosols and a tangent height constant radiation offset for minor 5 calibration errors.
In the retrieval we have considered spectra of tangent heights from 8 km upwards, because the lower spectra are contaminated by clouds.

Determination of the H 2 O a priori profile
Although H 2 O is simultaneously fitted with PAN, an impact of the applied a priori profile 10 for water vapour on the PAN result has been observed. The use of a climatological H 2 O a priori profile resulted in instabilities in the PAN vertical distribution. This was caused by the incorrect vertical position of the hygropause mapped into the resulting water vapour profile through the Tikhonow-Phillips regularisation constraint. To solve this problem we adopted a 2-step approach. In the first step we use a zero a priori H 2 O 15 profile and a relatively strong regularisation. This leads to a H 2 O profile (H 2 O first ) with reasonable position of the hygropause but relatively low vertical resolution. Its values are found to be higher than the in situ data between 10 and 12 km.
In the next step with weakened constraint, H 2 O first is used as the a priori to get the next H 2 O profile. This profile has been used as the "selected" H 2 O a priori in the PAN 20 retrieval. As shown in Fig. 5 the fitted H 2 O vmr profile is very similar to the selected a priori profile above 11 km but larger at lower altitudes. Both the fitted and the selected a priori profile tend to have some instability around 12-13 km. Such kind of feature is also present in the in situ data observed by the instruments FLASH and FISH (Fig. 5  The averaging kernel matrix for PAN MIPAS−STR is used to determine the sensitivity of the retrieval at different altitudes (see Fig. 7). The columns of the matrix are the answers of the retrieval to a delta function in the associated altitude. The diagonal 15 structures in the altitude range of the limb sequences between 8 and 18.6 km is clearly visible in Fig. 7. Below this range no measurements are available. The vertical resolution, determined as FWHM of each column of the averaging kernel matrix is given in Fig. 8. Above the flight level of 19 km the diagonal structure broadens strongly showing that there the vertical information is strongly reduced compared to the limb-range 20 where a vertical resolution of 2-2.5 km has been achieved (see Fig. 8).

Residual spectra
We investigate the quality in the spectral domain of the PAN retrieval described in Sect. 4.3 (RUN fit ) in comparison with that resulting from two further approaches (see Table 2). For the test "RUN limit " we fixed the PAN profile to PAN limit (see Fig. 2) and ACPD 8,2008 Tropical vertical profile of peroxyacetyl nitrate Interactive Discussion retrieved all other parameters like described in Sect. 4.3. The test "RUN zero " has been handled similarly but all PAN vmrs are fixed to zero. For all three runs, the residual spectra are shown in the lower panels of Fig. 9 for two selected tangent heights, 11 km (left part) and 13 km (right part). The top panels show the corresponding measured spectra. The rms (root mean square) of 5 the residuum is considerately lower [14.8 nW/(cm 2 sr cm −1 )] for the run RUN fit , than for RUN limit [32.1 nW/(cm 2 sr cm −1 )] and RUN zero [27.3 nW/(cm 2 sr cm −1 )]. The rms of RUN fit is higher than the spectral noise (6 nW/(cm 2 sr cm −1 )), because the residuum still contains residuals of lines, especially for low tangent heights. The broadband structure similar to the PAN contribution (see Figs. 3 and 4), present in the residua of RUN limit and RUN zero , is, however, removed in RUN fit .

Error estimation
In this section we analyse the effects of various error sources on the retrieved PAN vertical profile. We distinguish in instrument-related error sources such as calibration and spectral noise and in retrieval-related like spectroscopy and the errors in the used 15 profiles. Here we consider temperature, water vapour, CCl 4 and the five interfering species (CH 3 CCl 3 , CFC-113, CFC-22, CFC-11, and ClONO 2 ) whose profiles have been kept constant during the PAN retrieval. Figure 10 presents the total error together with the individual errors described in the following paragraphs.
1. Temperature: A comparison of the retrieved vertical temperature profile from 20 MIPAS-STR with that of ECMWF and in situ observations by the Rosemount TDC is shown in Fig. 11. In general, good agreement is found between all profiles, providing us the confidence in the level-1 processing for the spectral band in which also PAN is retrieved. Since the MIPAS-STR temperature is still slightly lower in the comparison, especially in the lower part, the contribution from a 2 K  Fig. 5), are used to estimate the contribution of the H 2 O a priori profile on the PAN error budget. In both a priori test profiles the zigzag at 13 km is removed. Additionally, the a priori values in test 2 have been increased for altitudes below 11 km, adapting the FISH measurement. Test 1 only weakly influences PAN MIPAS−STR , 5 whereas test 2 leads to differences in the order of about 5%.
3. The five pre-determined species: An uncertainty of 5% in each of the vmr profiles (CH 3 CCl 3 , CFC-113, CFC-22, CFC-11, and ClONO 2 ), which have been determined in previous steps and kept constant during the PAN retrieval, is 10 assumed.

PAN cross sections:
To consider atmospheric temperatures lower than 250 K we linearly extrapolated the cross sections measured at 273 K and 250 K. For the error from the PAN cross section, we added the temperature dependent term 15 (T -250 K)×0.16% to the error of 3% given by Allen et al. (2005b) for 250 K. The first term, roughly 4 % for 25 K difference is the dominant term at temperatures close to 200 K.
7. The a priori profile of PAN: The influence of the chosen a priori profile on the 25 retrieved PAN has been investigated by using a zero profile instead of PAN a priori .  Figure 10 presents each individual error contribution together with the total error calculated from these by the root square sum of all individual errors for each altitude. The high relative errors are in altitudes with low vmr values (see Fig. 6). In the altitude range spanned by the tangent points from 9 km to 18 km, the total relative error is between 15 % and 20 %.
In the lower part (up to about 14 km), errors in the temperature and PAN cross section dominate, whereas above spectral noise and PAN cross sections are the major error sources. Error bars for the total error are given with the PAN MIPAS−STR profile in Fig. 6.

Discussion
This work was initiated by the comparison of the MIPAS-STR HNO 3 profile with the dif-10 ference profile NO y -NO, measured by the in-situ instrument SIOUX. The disagreement between the two profiles posed the question, which of the constituents of NO y have to be considered additionally. The profile of ClONO 2 was retrieved from the MIPAS-STR spectra, and NO 2 was calculated from O 3 and NO. However, the consideration of those two gases did not change the situation, as their vmrs are very small. So we 15 tried successfully to retrieve PAN vmrs from the MIPAS-STR spectra. But the retrieved PAN profile only accounts for a sixth to a half (depending on the height) of the deficit NO y -NO-NO 2 -HNO 3 -ClONO 2 . In Fig. 13 we show all derived profiles of the individual constituents of NO y , their sum and the measured NO y profile.
In Interactive Discussion not be measured or calculated individually by the instruments aboard the Geophysica.

Conclusions
We investigated the retrieval of the vertical profile (8-19 km) of PAN using MIPAS-STR tropic emission spectra obtained in February 2005. The largest peak in the retrieved PAN vmr profile is located at 10 km altitude with an amount of about 0.14 ppbv. Above 5 10 km PAN decreases with a second smaller maximum at 16 km (≈0.06 ppbv). The total relative error is estimated to be about 15-20% between 9 and 18 km. Our measurements took place in February, the end of the dry season when the biomass burning was almost finished. This may explain the low values compared to measurements by (Singh, 1987, 0.3 ppbv @ 3-11 km) and (Glatthor et al., 2007, 0.33 ppbv @ 8 km 10 and 0.23 ppbv @ 11 km) in the same region (tropic southern Atlantic) but in September/October. Introduction