Validation analysis of deriving acetonitrile (CH3CN) profiles by observations of SMILES from the International Space Station, in the stratosphere and lower mesosphere

Acetonitrile (CH3CN) is one of the volatile organic compounds (VOC) and a potential tracer of biomass burning. We evaluated the capability of using observations derived from the Superconducting Submillimeter-Wave Limb-Emission Sounder (SMILES) on the International Space Station (ISS) to measure CH3CN profiles. The error in a CH3CN vertical profile from the Level-2 research (L2r) product version 3.0.0 was estimated by both theoretical error analysis and compared with other instrumental measurements. We estimated the systematic and random errors to be ∼ 5.8 ppt (7.8 %) and 25 ppt (60 %) for 5 a single observation at 15.7 hPa, respectively, in the Tropics, where the CH3CN measurements are enhanced. The major source of systematic error was a pressure broadening, and its contribution to the total systematic error was approximately 60 % in the middle stratosphere (15.7–4.8 hPa). The random error decreased to less than 40 % after averaging 10 profiles in the pressure range of 28.8–1.6 hPa. The total error due to uncertainties in other molecular spectroscopic parameters was comparable (2.8 ppt) to those of CH3CN spectroscopic parameters. We compared the SMILES CH3CN profiles with those of 10 the Microwave Limb Sounder (MLS) on the Aura satellite (version 4.2). The SMILES CH3CN values were consistent with those from MLS within the standard deviation (1σ) of the MLS observations. The difference between the SMILES and MLS CH3CN profiles increased with altitude and was within 20–35 ppt (20–260 %) at 15.7–1.6 hPa. We observed discrepancies of 5–10 ppt (10–30 %) between the SMILES CH3CN profiles observed by different spectrometers, so we do not recommend merging SMILES CH3CN profiles derived from the different spectrometers. We found that SMILES CH3CN VMR in the 15 upper stratosphere has a seasonal maximum in February, which is consistent with the fact that biomass burning events are highest from December–March.

HEMT: High electron mobility transistor   The SMILES Level 2 research (L2r) product version 3.0.0 (v3.0.0) was used in this study. The CH 3 CN VMR profile was 55 retrieved from the measurement spectra data of the Level-1b (L1b) version 008. Major improvements of the v3.0.0 from the previous version 2.1.5 were the AOS response function and a priori temperature profile. The details can be found in the JEM/SMILES L2r data product guideline (see http://smiles.nict.go.jp/pub/data/index.html). The optimal estimation method (OEM) was used for the retrieval processing. The OEM leads to the maximum a posteriori solution, which minimizes the value of χ 2 described below.
where F(x, b) is the forward model depending on x state vector and on the known model parameters b, S −1 y the measurement covariance matrix, x a the a priori state of x, and S a the a priori covariance matrix. Detailed retrieval algorithm of L2r product can be found in Baron et al. (2011) and Sato et al. (2012).
Quality of the retrieval processing was quantified by the chi-square statics, or goodness of the fit (Eq. 1), and the measurement response (m) defined as, wherex is the solution of the retrieval, A is the averaging kernel, D is the contribution function, and K is the weighting function.
m, A and D were derived using K (Urban et al., 2004). Details on m are explained by Sato et al. (2014). The χ 2 of CH 3 CN for 75 v3.0.0 had a range of 0.4-0.6. In cases where the measurement response was low, information was retrieved from the a priori state. Here, the data selection thresholds of χ 2 and measurement response were set to be χ 2 < 0.6 and m > 0.80, respectively.  ters in a forward model (Sato et al., 2012;Kasai et al., 2013;Sagawa et al., 2013). We used a typical CH 3 CN profile derived using observations from the Tropics, where BB (a major source of CH 3 CN) frequently occurs. The total error (E total ) is given by where E n is the error due to spectral noise, E s is the smoothing error, and E p is the model parameter error. The error due to 90 the spectral calibration was ignored in this study, because the L1b data was updated in this version, and the error due to the spectral calibration was not significant in previous SMILES error analyses (e.g. Sato et al., 2012).
Error E n and E s was calculated by where and where 100 Here, S n and S s are the error covariance matrices for measurement noise and the errors from S a , respectively. U is the unit matrix.
The model parameter error E p includes errors caused by uncertainties in the parameters used in both the forward and inversion calculations. Error sources for the model parameters are summarized in Table 3. Error related to each of the individual model parameters was calculated using the perturbation method following Sato et al. (2012). The total error E p for all of the 105 parameters was calculated using the root sum square of the individual errors. Figure 4 shows the estimated systematic errors.The left panel (a) shows the uncertainties in the AOS response function ("AOS"), the antenna beam pattern ("Antenna"), the spectral line strength ("Strength"), the air pressure broadening coefficient ("γ"), its temperature dependence ("n"), and their root sum square ("Total"). The largest error source (∼2 ppt (5 %) was from the air pressure broadening coefficient ("γ") across the entire pressure range, followed by line intensity ("Strength") and 110 temperature dependence of air pressure broadening coefficient("n") ( 1.5 ppt). The error from spectroscopic parameters was more significant than that from instrumental functions.  . 2). The spectral shape of O 3 and H 37 Cl should therefore influence the retrieval of the CH 3 CN VMR profiles. To estimate the influence from the other spectral lines, error due to the spectroscopic parameters γ and its temperature dependence n of the O 3 and H 37 CL lines were also calculated. γ and 115 temperature dependence of γ were perturbed for each species, and are expressed as "O 3 γ", "O 3 n", "H 37 Clγ" and "H 37 Cln".
As shown in Fig. 4 (b), "H 37 Clγ" is the largest error source, whose maximum absolute difference was 1.1 ppt. Error analyses completed for O 3 and ClO demonstrated that error caused by other molecular spectral lines was negligible as they have high, isolated line strengths (Sato et al., 2012;Sagawa et al., 2013;Kasai et al., 2013). In the case of CH 3 CN retrieval, however, the total error caused by uncertainties in other molecular spectroscopic parameters was comparable to the error caused by CH 3 CN 120 spectroscopic parameters. The errors due to H 37 Cl was larger than that from O 3 at each pressure level.
The measurement noise and smoothing error from a single scan are shown in the Fig. 5 (a). These errors are considered random error for a CH 3 CN profile. SMILES CH 3 CN total error consists of both the systematic and random error. Figure 5 (b) shows the total systematic error, the random and total error averaged by the number of profile (N = 1, 10 and 100). The random error was larger than the systematic error from a single scan. However, the random error averaged by 100 profiles was 125 comparable to the systematic error, except for the highest systematic error, which was found at a pressure level of about 1 hPa.

Comparison with Aura/MLS
In this section, we compare SMILES CH 3 CN observations with Aura/MLS observations and discuss the validity of SMILES observations. Figure 6 shows (  VMRs observations retrieved at SMILES pressure grids for AOS1 (blue) and AOS2 (green). The relative difference between AOS1 and AOS2 is approximately 12 ppt (30 %) with the maximum at 15.7 hPa, indicating that the difference between the two AOSs is due to sensitivity differences. The difference between the two decreases down to less than 10 % at an upper altitudes 135 than 4.8 hPa.
We also investigated seasonal variation of SMILES CH 3 CN observations for each altitude grid as shown in Fig. 7. This figure shows daily scatter plots and daily averages for AOS1 (red shaded) and AOS2 (blue shaded) observations. The red circles and bars represent the daily mean values and 1 σ standards deviation, when more than one hundred observation points were obtained in one day. Like in Fig. 1, at lower altitudes (28 km to 36 km) the difference between the two AOSs observations  CH 3 CN levels peaked in February, and can be seen from approximately 40 km to 52 km where the difference between the two AOSs can be negligible.

Comparison with Aura/MLS v4.2 data 145
We investigated the difference of CH 3 CN VMRs between SMILES and MLS observations. We set the data quality thresholds and the coincidence selection criteria for the SMILES and MLS observations, as summarized in Table 4. The MLS data quality criteria was based on the MLS v4.2 Level-2 data quality and description document.
The geolocation and measurement time criteria were determined as follows; the distance of measurement location within 300 km;   difference in the measurement time within 6 hour.
We investigated the diurnal variation of SMILES CH 3 CN observations at several altitudes (32 km, 40 km, and 48 km) for AOS1 and AOS2 individual observational periods, and confirmed that there is no diurnal variation for stratospheric CH 3 CN observations. Figure 8 shows the distribution of coincident points satisfying these criteria at 8.6 hPa.The interpolation of VMRs was done 155 using a linear interpolation with respect to the logarithm pressure levels. There are on average 10 coincident points in each bin at this pressure level and the total coincident data number was 17910.
For the comparison between SMILES and MLS observations, the mean absolute difference, ∆ abs , and relative difference, ∆ rel , at the pressure levels, p, between coincident CH 3 CN profiles of the two observations were calculated as follows, where N (p) is the number of coincidences at p, x s (p) and x m (p) are the VMRs at p for SMILES and MLS observations, and the reference (x p ) is x p = 1 2 (x s (p) + x m (p)).  spectral regions (Waters et al., 2006). MLS Level-2 CH 3 CN profiles were observed in 640 GHz spectral regions. Although the pressure range of a retrieved MLS CH 3 CN is 147 to 0.001 hPa, the pressure range of CH 3 CN version 4.2.0 is 46-1.0 hPa (Livesey et al., 2006).

Result of comparisons
175 Figure 9 shows the vertical profile, the absolute differences and the relative differences between SMILES AOS1/AOS2 and MLS CH 3 CN observations. The left panel in Fig. 9  T sys was much smaller than that of MLS, indicating that SMILES has an advantage in the upper stratosphere. SMILES was also able to observe CH 3 CN VMR in the upper stratosphere with a much lower uncertainty of ∼20 ppt although the uncertainty of MLS CH 3 CN VMR was approximately 100 ppt in the altitude. The differences of the CH 3 CN VMR observed by two AOSs was sufficiently small in comparison with the difference between SMILES and MLS observations. Theoretical systematic error (blue broken lines in the middle panel) derived in Sect. 3 was less than the differences between SMILES and MLS observations, 185 except at 8.9 hPa. We also investigated latitudinal and seasonal variation between the two observation methods. Figure 10 shows the seasonal variation of SMILES and MLS CH 3 CN observations, and the absolute differences for each pressure level at coincident points, as a function of latitude. The left column represents SMILES CH 3 CN VMR in units of ppt which were separated into two AOSs observations. The middle column represents MLS CH 3 CN VMR, and the right column represents the absolute differences 190 between SMILES and MLS observations. At lower altitudes of 15.7 hPa and 8.6 hPa, SMILES observations were overestimated