Development of the Retrieval Algorithm and Capability

Study of Ozone for JEM/SMILES

(JEM/SMILES

Chikako Takahashi
Contents

Abstract 1

I Introduction 4

1 Status for Atmospheric Ozone 4

2 Atmospheric Measurements Techniques 11

3 JEM/SMILES 14

4 Retrieval Techniques for JEM/SMILES 22

5 Aim of This Thesis 27

II Operational Retrieval Algorithms 34

6 Introduction 34

7 Framework of DPS-L2 35
7.1 Ground Data Processing System and Data Flow . . . . . . . . 35
7.2 Structure of the DPS-L2 . . . . . . . . . . . . . . . . . . . . . 38
7.3 Requirements for the ORC . . . . . . . . . . . . . . . . . . . . 40

8 Basic Retrieval Algorithm 41

8.1 State Vector x . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

I
9 Optimized Forward Model 43
9.1 Overview of Forward Model . . . . . . . . . . . . . . . . . . . 43
9.2 Radiative Transfer Calculation . . . . . . . . . . . . . . . . . . 45
9.2.1 Ray Tracing . . . . . . . . . . . . . . . . . . . . . . . . 45
9.2.2 Absorption Coefficient . . . . . . . . . . . . . . . . . . 46
9.2.3 Radiative Transfer . . . . . . . . . . . . . . . . . . . . 47
9.2.4 Doppler Shift . . . . . . . . . . . . . . . . . . . . . . . 48
9.3 Accurate Instrument Model for the SMILES . . . . . . . . . . 50
9.3.1 FOV Convolution . . . . . . . . . . . . . . . . . . . . . 50
9.3.2 Sideband Ratio . . . . . . . . . . . . . . . . . . . . . . 53
9.3.3 Frequency Response . . . . . . . . . . . . . . . . . . . 54
9.4 Fast Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . 55
9.4.1 Line Selection . . . . . . . . . . . . . . . . . . . . . . . 55
9.4.2 Frequency Selection . . . . . . . . . . . . . . . . . . . . 55

10 Results 57
10.1 Validation of ORC . . . . . . . . . . . . . . . . . . . . . . . . 57
10.2 Error Budget . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
10.3 Algorithm Performance and Hardware System . . . . . . . . . 65

11 Summary 66

III Capability for Ozone retrieval 73

12 Introduction 73

13 O3 Measurements with the JEM/SMILES 74

II
14 Retrieval Results 77
14.1 Setup Conditions . . . . . . . . . . . . . . . . . . . . . . . . . 77
14.2 Random Error and Vertical Resolution . . . . . . . . . . . . . 77

15 Error Analysis 81

16 Diurnal Variation of O3 Abundance in the Upper Strato-

sphere 88

17 Retrieval Capability in the Lower Stratosphere and Upper

Troposphere 90
17.1 Two-Bands Simultaneous Retrieval . . . . . . . . . . . . . . . 90
17.2 Effects from Cirrus Clouds . . . . . . . . . . . . . . . . . . . . 93

18 Conclusion 95

Summary 104

Acknowledgments 107

Appendix: Ozone retrieval by the assured data 109

III
List of Tables
1 Target species of SMILES . . . . . . . . . . . . . . . . . . . . 16
2 JEM/SMILES Specifications . . . . . . . . . . . . . . . . . . 21
3 JEM/SMILES Data Sets . . . . . . . . . . . . . . . . . . . . . 37
4 Operation Environment of DPS-L2 . . . . . . . . . . . . . . . 39
5 JEM/SMILES design specifications . . . . . . . . . . . . . . . 78

IV
List of Figures
1 Atmospher ozone destribution by Scientific Assessment of Ozone
Depletion 2010 (WMO, 2010) . . . . . . . . . . . . . . . . . . 8
2 Global total ozone changes by Scientific Assessment of Ozone
Depletion 2010 (WMO, 2010) . . . . . . . . . . . . . . . . . . 9
3 1980 baseline-adjusted multi-model trend estimates of annu-
ally averaged total column ozone (DU; left) and Cly at 50
hPa (ppb; right) for the tropics (25◦ S–25◦ N, upper row) and
midlatitudes (middle row: 35◦ N–60◦ N, lower row: 35◦ S–60◦ S)
(thick dark gray line) with 95% confidence and 95% prediction
intervals appearing as light- and dark-gray shaded regions, re-
spectively, about the trend (note the different vertical scale
among the panels) (WMO, 2010). . . . . . . . . . . . . . . . . 10
4 The ISS and SMILES (NASDA and CRL, 2002). . . . . . . . . 14
5 The SMILES payload (NASDA and CRL, 2002). . . . . . . . . 15
6 The limb observation by SMILES. . . . . . . . . . . . . . . . . 17
7 The world map of orbit and measurement positions of SMILES
(NASDA and CRL, 2002). . . . . . . . . . . . . . . . . . . . . 17
8 Block diagram of SMILES (NASDA and CRL, 2002). . . . . . 20
9 Overview of DPS-L2 . . . . . . . . . . . . . . . . . . . . . . . 39

V
10 Simultaneous retrieval of ozone and HOCl. The red line indi-
cates the retrieval precision in ozone retrieval, the green line in-
dicates the incremental error in ozone retrieval without HOCl,
and the blue line indicates the incremental error in ozone re-
trieval with HOCl. In all these cases, the true profiles of HOCl
are 100% (Right) and 200% (Left) greater than the a priori
profile of HOCl. Here, the incremental error is defined as the
difference between the true profile and the retrieved profile. . . 44
11 Effect of wind. The red line indicates the retrieval precision of
ozone. The other lines indicate the error due to the difference
between the reference profile and the true profile of wind.(The
wind velocities for the pink, blue, and green profiles are 50, 10,
5 m/s, respectively. The definition of the incremental error is
the same as that in Figure 10 . . . . . . . . . . . . . . . . . . 49
12 Error due to inclination of antenna scan axis. The red line
indicates the retrieval precision of ozone. The other lines indi-
cate the error due to the inclination of the antenna scan axis.
The inclinations of the antenna scan axis in the aqua blue,
pink, dark blue, and green profiles are 15◦ , 10◦ , 5◦ , and 1◦ , re-
spectively. The definition of the incremental error is the same
as that in Figure 10. . . . . . . . . . . . . . . . . . . . . . . . 52
13 Error due to line selection. The red line is the retrieval pre-
cision of ozone, and the green line indicates the incremental
error in retrieved ozone due to line selection. The definition
of the incremental error is the same as that in Figure 10. . . . 56

VI
14 Residual error due to frequency selection. Top: spectrum of
band A obtained by using equal frequency step of 0.1 MHz,
middle: spectrum of band A obtained by using unequal fre-
quency steps produced by our algorithm, and bottom: residual
error. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
15 Residual error of absorption coefficients at 0 km (red), 10
km (green), 20 km (blue), 30 km (pink), 40 km (aqua), and
50 km (yellow). The residual error is defined by (SMOCO-
ORC)/SMOCO. . . . . . . . . . . . . . . . . . . . . . . . . . . 60
16 Residual error of brightness temperature. Legends are the
same as those used in Figure 15 . . . . . . . . . . . . . . . . . 61
17 Retrieval precision of target species that can be retrieved from
single-scan data. . . . . . . . . . . . . . . . . . . . . . . . . . . 64
18 Measured spectra by SMILES. There are three bands; band
A (top), B (middle) and C (bottom) at approximately 10 km
(red/solid line), 20 km (green/dashed line), 30 km (blue/dotted
line), and 40 km (pink/fine dotted line). . . . . . . . . . . . . 76
19 The solid red line and dashed green line indicate the random
error for the system noise of 500 K and 5000 K, respectively. . 79
20 Averaging kernels of O3 for bands A (left), B (middle) and
C (right). A thick red line is an integration of the averaging
kernel for each altitude, which indicates amount of information
at that altitude. Averaging kernel altitudes and corresponding
vertical resolutions are indicated on the right side of each plot. 80

VII
21 The standard mid-latitudes atmosphere profiles have been used
in this study. The solid red line is the ozone profile, the dot-
ted blue line is the atmospheric temperature profile and the
dashed green line is H2 O profile. . . . . . . . . . . . . . . . . . 82
22 Estimations for the influence of priori profiles in O3 retrieval
(in this case, a priori profiles are same as initial profiles). The
solid red line is the random error of O3 . The other lines are
additional errors between the true profiles of O3 and the re-
trieved profiles of O3 that are the final results of the iteration
process in the cases where the differences between the a priori
profiles and true profiles are ±5%, ±10%, and ±50%. (top:
mid-latitudes, bottom: tropics). . . . . . . . . . . . . . . . . . 84
23 Information content of O3 at 625.37 GHz (red/solid line) and
oxygen at 118 GHz (green/dashed line). . . . . . . . . . . . . 85
24 Estimation of the influence of uncertainty of initial and a pri-
ori profiles of atmospheric temperature (in this case, we re-
trieve atmospheric temperature and O3 profile simultaneously
and the priori profiles of atmospheric temperature are same
as initial profiles). Lines are additional errors between the
true profiles of O3 and the retrieved profiles of O3 that are
the final results of the iteration process in the cases where the
uncertainties of atmospheric temperature are 1 K (red/solid
line), 2 K (green/dashed line), 5 K (blue/dotted line) and 10
K (pink/fine-dotted line). . . . . . . . . . . . . . . . . . . . . 87

VIII
25 Diurnal variability of the O3 measurements. The left side of
the figure shows the ozone profiles and the right figure shows
the random error in the daytime (red/solid line) and in the
nighttime (green/dashed line), respectively. . . . . . . . . . . . 89
26 Results of synthetic retrievals using different bands. The left
side of the figure shows the profiles used in each simulation
and the right side of the figure shows the random error in
the case of the mid-latitudes (top) and the tropics (bottom).
The lines on the left side of the figure are the profiles of O3
(red/solid line), temperature (green/dashed line), and H2 O
(blue/dotted line). The lines on the right side of the figure
are the random error of O3 using band A only (red/solid line),
using bands A and B (green/dashed line), and using bands A
and C (blue/dotted line). . . . . . . . . . . . . . . . . . . . . . 91
27 Relationship between the ice water path (IWP) and the bright-
ness temperature depression due to cirrus clouds. The tangent
altitudes are 4 km (solid line), 8 km (dashed line), and 12 km
(dotted line) and the cloud layer is assumed to be uniformly
between 12 km and 13 km. . . . . . . . . . . . . . . . . . . . 92
28 Effects from the cirrus clouds whose IWP are 1.0 (red/solid
line) and 0.5 (green/dashed line) g/m2 with the random error
of the bands A and C (blue/dotted line) and A only (pink/fine-
dotted line) in the case of the mid-latitudes (top) and the
tropics (bottom). . . . . . . . . . . . . . . . . . . . . . . . . . 94

IX
29 The measured spectra on 12 October 2009 at several tangent
altitudes with log scale (Kikuchi et al., 2010). (a) is Band A,
(b) is Band B, and (c) is Band C. . . . . . . . . . . . . . . . 110
30 Brightness temperature and residual errors of the measured
brightness temperature measured on 12 October 2009 at 20
km, 30 km, 40 km and 50 km altitudes. . . . . . . . . . . . . 111
31 the measurements of ozone diurnal variation by SMILES. Green
and blue points with error bars show ozone mixing ratios (ppb)
at 61 km altitude of ascending and descending, respectively. . 113
32 Latitude-height sections of the zonal mean ozone on 12 Octo-
ber 2009. Left side is SMILES and right side is AURA/MLS.
In this case, data for daytime and nighttime are averaged
(Kikuchi et al., 2010) . . . . . . . . . . . . . . . . . . . . . . . 114
33 Statistics estimation of comparison SMILES and AURA/MLS
for ozone retrieval profile, using coincident SMILES and AURA/MLS
ozone profiles at the 55◦ N - 65◦ N latitude. Left: the mean
profiles for SMILES (blue) and AURA/MLS (red), Middle:
the differences between the SMILES and AURA/MLS profiles
in mixing ratio, Right: the differences in percentage (Suzuki
et al., 2010). . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

X
34 Statistics estimation of comparison SMILES and AURA/MLS
for HCl retrieval profile using coincident SMILES and AURA/MLS
ozone profiles in the 55◦ N - 65◦ N latitude region. Left: the
mean profiles for SMILES (blue) and AURA/MLS (red), Mid-
dle: the differences between the SMILES and AURA/MLS
profiles in mixing ratio, Right: the differences in percentage
(Suzuki et al., 2010). . . . . . . . . . . . . . . . . . . . . . . . 116
35 Example of HNO3 coincidence compared with ENVISAT/MIPAS
(MIPAS-IMK ver.40) at 67.1◦ N, 101.5◦ on Oct. 12, 2009. Two
SMILES profiles (blue) are compared with 1 MIPAS profiles
(red) (Suzuki et al., 2010). . . . . . . . . . . . . . . . . . . . . 117

XI
Abstract
The Superconducting Submillimeter-Wave Limb-Emission Sounder (SMILES)
have been launched and aboard the Japanese Experiment Module (JEM) of
the International Space Station (ISS) in September 2009. It had measured
the submillimeter wave limb emission from atmospheric minor species mainly
in the stratosphere. It is the first to use 4-K mechanical refrigerator to cool
superconducting devices and a major feature of the SMILES is very sensitive
measurements by low system noise temperature. It achieved approximately
400 K. The SMILES requires the high accuracy and fast for the retrieval al-
gorithm, because the SMILES has very high sensitivity thanks to supercon-
ducting receiver and retrieves the density distributions of the target species
from calibrated spectra in near-real-time.

The retrieval algorithm is based on the optimal estimation method ap-

plied for atmospheric soundings by C. Rodgers. The retrieval process consists
of two parts: the forward model, which computes radiative transfer, and the
inverse model, which deduces atmospheric states. The forward model must
provide the most accurate basis for results because SMILES has very high
sensitivity. Therefore the forward model algorithm for an operational sys-
tem has to be accurate. It should be fast because of limitation of computing
resources. Hence, the algorithm is improved (1) by designing accurate in-
strument functions such as the instrumental field of view (FOV), sideband
rejection ratio of sideband separator, and spectral responses of acousto-optic
spectrometer (AOS) and (2) by optimizing radiative transfer calculation. The

1
accuracy of this algorithm has been achieved better than 1%, and the process-
ing time for single-scan spectra and the processing speed has been achieved
faster than 1 min with 8 parallel processing using a 3.16-GHz Quad-Core
Intel Xeon processor. In Chapter II show the details of these algorithms.

In Chapter III, we estimate the capability of O3 retrieving from mea-

surements by SMILES. Since SMILES has very high sensitivity, the ozone
retrieval is expected to be highly precision. We examine the retrieval pre-
cision and the accuracy of O3 vertical profiles based on the launch-ready
retrieval algorithm developed for the SMILES. The best retrieval precision
with single scan spectra is 0.2% at an altitude of 30 km with 3 km vertical
resolution in the mid-latitude. The retrieval precision is better than 5% in
the altitude region from 15 to 70 km in the nighttime and from 15 to 55 km
in the daytime. By averaging 10 profiles, the retrieval precision is improved
to 1% at 70 km altitude in the nighttime and to 5% in the daytime. It is
expected to reveal the diurnal variation of O3 vertical profiles with high pre-
cision in the upper stratosphere. Moreover, the retrieval capability in the
lower stratosphere is estimated. By retrieving spectral data using two re-
ceiver bands (624.32 GHz - 626.32 GHz and 649.12 GHz - 651.32 GHz), the
retrieval precision at 13 km altitude will be expected to be approximately
3% under clear sky conditions, which is 10 times better than one retrieved
from a single band. We have clarified the retrieval capabilities for the diurnal
variation in the upper stratosphere, the effects from the model parameters
such as the a priori profiles, uncertainty of the atmospheric temperature and
the cirrus clouds, and the effectiveness for the simultaneous retrieval of the

2
lower and upper sidebands in the lower stratosphere. Finally, ozone retrieval
with measured data are shown in appendix.

3
Part I
Introduction
1 Status for Atmospheric Ozone
Ozone is one of the key constituents of the Earth’s atmosphere. Most ozone
resides in the lower stratosphere in what is commonly known as the ozone
layer and extends up to about 50 km altitude (see Figure 1). The remaining
ozone, about 10%, is found in the troposphere. The ozone layer plays a major
role in shielding us from solar ultraviolet (UV) rays, radiation budget and
atmospheric chemistry. Increases in ozone occur near the surface as a result
of pollution from human activities (WMO, 2010), recentry.

Stratospheric ozone is naturally formed in chemical reactions involving

UV rays and oxygen molecules. The processes of ozone production and de-
struction are as follow:

O2 + hν → 2O (1)

O2 + O + M → O3 + M. (2)

Here, M is a catalytic molecule, such as a nitrogen molecule an oxygen

molecule. UV rays (hν) breaks apart one oxygen molecule (O2 ) to produce
two oxygen atoms (2 O) and each atom combines with an oxygen molecule to
produce an ozone molecule (O3 ). The production and destruction of strato-
spheric ozone is balanced in these reactions. This is not cause of ozone
depletion. Reactions of the cause of ozone depletion are refered to the next.
Stratospheric ozone is destroyed by reactions involving reactive halogen gases,

4
which are produced in the chemical conversion of halogen source gases. The
processes of ozone destruction are as follow:

X + O3 → XO + O2 (3)

XO + O → X + O2 (4)

=⇒ O + O3 → 2O2 . (5)

Here, X is a reactive halogen gas, such as chlorine (Cl) or bromine atoms (Br).
Global ozone has decreased because the amounts of reactive gases containing
chlorine and bromine have increased in the stratosphere.

Since the discovery of the Antarctic ozone hole in the mid 1980s, numerous
observations have been performes and understanding for the mechanisms of
ozone depletion have been advanced. According to the Ozone Depletion 2010
(WMO, 2010), in the early 1990s the depletion of global total ozone reached
a maximum of about 5% below the 1964–1980 average. The depletion has
lessened since then and now averages about 3.5% for 2006–2009. The total
global accumulation of ozone-depleting substances has slowed and begun to
decrease. The Ozone Depletion 2010 (WMO, 2010) refers a prediction that
the ozone will be recovered by 2050, but there is still some variability in each
climate model.

Meanwhile, a relationship between ozone depletion and climate change

is also becoming important issue recentry. The ozone depletion itself is not
the principal cause of climate change, but changes in ozone and climate are
directly linked. Ozone and the ozone-depleting substances are greenhouse

5
gases along with carbon dioxide (CO2 ). Both of stratospheric and tropo-
spheric ozone absorb infrared radiation but these ozone affect the Earth’s
surface oppositely. Stratospheric ozone depletion causes a negative radia-
tive forcing (i.e. it leads to a cooling of Earth’s surface) and in contrast
increases in tropospheric ozone cause a positive radiative forcing (i.e. it
leads to a warming of Earth’s surface). The mechanisms of the link of the
ozone changes and climate changes remain to be clarified, because various
factors are mixed complexly. The incleasing the green gases warm the tro-
pospheric temperature but it cool the statrospheric and above temperature.
The cooling of the lower strasphere cause the polar stratospheric cloud and it
lead to the ozone depletion but the cooling of the upper tropoephere provide
incleasing ozone contrastly. Furthermore transport processes of atmosphere
are affected by climate changes and changes of transport processes also affect
the ozone layer.

Finally, issues of the upper troposphere and lower stratosphere (UT/LS)

are mentioned here. UT/LS has an important role in the dynamical pro-
cesses. In UT/LS, the time scale of ozone change on the dynamical processes
is as comparable as that of the photochemical time scale. According to
(Holton et al., 1995), the planetary wave induced torque mainly drives the
upward motion at equatorial latitudes and the downward motion at mid and
high latitudes in the lower stratosphere. At equatorial latitudes, a modu-
lation of the meridional circulation affects the tropopause height and tem-
perature, etc., and all these factors are closely connected with each other.
However our knowledge is still insufficiency and comprehensive understand-

6
ing of these mechanism is needed.

The understanding for the mechanisms of ozone depletion have been ad-
vanced but there are still unsolved issues, such as the link with the climate
change mentioned above. Figure 3 shows 1980 baseline-adjusted multi-model
trend estimates of annually averaged total column ozone and Cly at 50 hPa
for the tropics (25◦ S–25◦ N, upper row) and midlatitudes (middle row: 35◦ N–
60◦ N, lower row: 35◦ S–60◦ S) with 95% confidence and 95% prediction in-
tervals appearing as light- and dark-gray shaded regions, respectively, about
the trend (note the different vertical scale among the panels) (WMO, 2010).
There are large difference in these models. Therefore, there are further
needs of systematic understandings of ozone chemistry and transport pro-
cesses. Since there have been various measurements, every ozone-depleting
substances have not been measured thoroughly. Measurements for some mi-
nor morecules like BrO are stil insufficient. To accelerate the understanding
of these issues, accurate and global measurements of three-dimensional and
simultaneous measurements of ozone and the ozone-depleting substancess
are required. The accurate measurement can reveal not only averaged abun-
dance of minor morecules but it can reveal time variation for abundance of
measured morecules finer than the previous measurements. For the accurate
measurement, developments of both sensor techniques and data processing
algorithms are indispensable. The sensor techniques and the data processing
algorithms are refered later.

7
Figure 1: Atmospher ozone destribution by Scientific Assessment of Ozone
Depletion 2010 (WMO, 2010)

8
Figure 2: Global total ozone changes by Scientific Assessment of Ozone De-
pletion 2010 (WMO, 2010)
9
Figure 3: 1980 baseline-adjusted multi-model trend estimates of annually
averaged total column ozone (DU; left) and Cly at 50 hPa (ppb; right) for
the tropics (25◦ S–25◦ N, upper row) and midlatitudes (middle row: 35◦ N–
60◦ N, lower row: 35◦ S–60◦ S) (thick dark gray line) with 95% confidence and
95% prediction intervals appearing as light- and dark-gray shaded regions,
respectively, about the trend (note the different vertical scale among the
panels) (WMO, 2010). 10
2 Atmospheric Measurements Techniques
Since the Antarctic ozone hole have been discoverd in the mid 1980s, nu-
merous observations have been conducted and these have helped to revel the
mechanism of ozone depletion. Atmospheric ozone have been measured from
various platforms, such as the ground, balloons, aircrafts, and satellites.

Total column ozone have been measured from the ground with Dobson
spectrophotometers since 1920’s. The Dobson spectrometer measures the
total ozone by measuring the relative intensity of two ultraviolet wavelength
range. It is a standard method for these measurement because of its high
sensitive measurements, but it is strongly affected by aerosols and pollutants
in the atmosphere. The ground measurement with Umkehr method can
mesure the ozone vertical distribution, but its vertical resolution is not so
fine. An Ozone laser radar is another technique to measure the atmospheric
ozone using laser light and it can meaasure the ozone vertical distributions
in the upper stratosphere. It can only work in night time. Othter technique
to measure the vertical distributions of ozone is millimeter wave radiometer,
which can measure it in the mesosphere in noon and night. Fourier transform
infrared spectroscopy (FTIR) is also one of the technique to measure the
vertical distributions of ozone and it can measuren not only ozone but also
other species like HCl and HNO3 simultaneously. The ground measurement
has advantage to know long-term variability at a certain location but it is
difficult to encompass the entire globe finely, but the vertical resolutions are
not so fine compared with baloons, aircrafts, ozonesonde and satelite.

11
 Satellites are useful platforms for measureing atmospheric ozone globally
and for measureing the vertical distribution with fine resolution up to high-
altitude. To measure the vertical distribution with fine resolution, baloons,
aircrafts and ozonesonde are useful, but these can also not be cover the entire
globe and these can not measure the high altitude region (generally below
30 km). There have been global, long-term and high-quality O3 measure-
ments by a solar occultation technique with the Stratospheric Aerosol and
Gas Experiment (SAGE) series from 1979 to 2005 (Rusch et al., 1997), the
Halogen Occultation Experiment (HALOE) (Russell et al., 1993) from 1991
to 2005 and the Polar Ozone and Aerosol Measurement (POAM) (Bevilac-
qua et al., 1995) from 1993 to 2005. Also the SCISAT/ACE (Atmospheric
Chemistry Experiment) mission (Bernath et al., 2005), launched in 2003,
uses the solar occultation technique. The Michelson Interferometer for Pas-
sive Atmospheric Sounding (MIPAS) launched in 2002 (Fischer and Oelhaf,
1996; Fischer et al., 2008), the Earth Observing System (EOS) Microwave
Limb Sounder (MLS)(Read et al., 2006; Froidevaux et al., 2008) instrument
launched in 2004, and the Submillimeter Radiometer (SMR) onboard the
Odin satellite (Murtagh et al., 2002) launched in February 2001 measure the
atmospheric limb emission in infrared, millimeter-wave and submillimeter-
wave ranges, respectively. Those emission measurements have the advantage
of independence of the solar angle. Furthermore, there are various other
types of global ozone measurements. The SCanning Imaging Absorption
spectroMeter for Atmospheric CHartographY (SCIAMACHY) (Bovensman
et al., 1999) launched in 2002 onboard ENVISAT provides nadir, limb and
solar/lunar occultation measurements. Global Ozone Monitoring by Occul-

12
tation of Stars (GOMOS) also onboard ENVISAT performs the stellar occul-
tation measurements, and the OSIRIS instrument onboard the Odin satellite
measures the scattered solar radiation in the visible spectral range.

Submillimeter-wave radiometry techniques are powerful tools for the re-

mote measurements of the earth’s middle atmosphere. These techniques
allow the simultaneous measurements of a variety of important species for
studying atmospheric chemistry and dynamics. The Microwave Limb Sounder
(MLS) aboard the Upper Atmosphere Research Satellite (UARS) launched in
1991 has demonstrated that the millimeter-wave limb sounding from satellites
is suitable technique for measuring atmospheric constituents and tempera-
ture on a global scale (Waters et al., 1999). The first submillimeter limb
measurement from space was conducted by the Sub-Millimeter Radiometer
(SMR) on board the Odin satellite launched in February 2001 Murtagh et al.
(2002). The follow-on MLS aboard the Aura satellite launched in July 2004
measures millimeter and submillimeter bands comprehensively (Read et al.,
2006). These technologies have contributed to improving our understanding
of stratospheric chemistry.

For future missions, measurements with grobal and high sensitivity are
desired in order to study the dynamical features of the atmosphere. These
studies require advanced submillimeter measurement technologies, including
superconductive receivers with high sensitivity.

13
 Figure 4: The ISS and SMILES (NASDA and CRL, 2002).

3 JEM/SMILES
The Superconducting Submillimeter-Wave Limb-Emission Sounder (SMILES)
instrument has been put in operation as a part of the Japanese Experi-
ment Module (JEM) on board the International Space Station (ISS) since
September 2009 (Figure 5). It is a collaborative project between the Japan
Aerospace Exploration Agency (JAXA) and the National Institute of Infor-
mation and Communications Technology (NICT). It measures the submillimeter-
wave limb emission from atmospheric minor species mainly in the strato-
sphere. Figure 6 show the image of the limb observation by SMILES.

14
Figure 5: The SMILES payload (NASDA and CRL, 2002).

15
 Table 1: Target species of SMILES
Type Band A Band B Band C
Species retrieved from O3 O3 O3
single-scan data H37 Cl H35 Cl ClO
18
OOO 18 OOO HNO3
HNO3 O17 OO 18 OOO
17
CH3 CN OOO
HOCl
O17 OO
Species retrieved from BrO HO2 BrO
multi-scan data (noisy products) HO2

A major feature of JEM/SMILES is the ability to receive high sensi-

tive measurements with a low system-noise temperature using 4–K cooled
superconductor-insulator-superconductor (SIS) mixers (Figure ??. The tar-
get species are O3 , ClO, HCl, HNO3 , HOCl, CH3 CN, HO2 , BrO, and O3
isotopes (See Table1). The SMILES mission focuses on the details of halogen
chemistry related to ozone destruction. SMILES takes 53 s for a complete
vertical scan and there are 68 individual limb rays in a single scan in the
nominal altitude coverage from 10 to 60 km. Since the orbital period of the
ISS is approximately 93 min, approximately 105 scans per orbit give 1630
scans per day. The spatial coverage is on a near global basis, or the nominal
latitude coverage is 38◦ S – 65◦ N by inclining the antenna beam to 45◦ left
from the direction of orbital motion. It is expected that it will be possible
to make measurements within the elongated polar vortex in the Northern
Hemisphere. The world map of orbit and measurement positions is shown in
Figure 7.

Figure 8 shows the block diagram of the SMILES instrument (NASDA

16
 Atmosphere

SMILES
Earth

Figure 6: The limb observation by SMILES.

Figure 7: The world map of orbit and measurement positions of SMILES

(NASDA and CRL, 2002).

17
and CRL, 2002). The main components of SMILES are a submillimeter an-
tenna, a submillimeter receiver, and a radio spectrometer. The submillimeter
antenna is an elliptical mirror with dimensions of 40 cm invertical direction
and 20 cm inhorizontal because of the trade-off between the spatial resolu-
tion and the envelope for JEM payloads (Manabe et al., 2008). The vertical
beam size is 0.09 degree (half-power beamwidth) and the angular accuracy of
the antenna driving mechanism must be better than 0.01 degree. The angu-
lar accuracy affects an accuracy of measured altitude determination, which
is important factor for retrieval processing. The submillimeter receiver is a
key instrument of SMILES experiment and its noise temperature is less than
400 K mentioned above. SMILES has the two SIS mixers that receive the
atmospheric signal in the upper sideband (USB) and lower sideband (LSB)
separately (USB: 649.12–650.32 GHz, LSB: 624.32–626.32 GHz). The atmo-
spheric emissions are separated by a quasi-optical sideband filter, based on
a modiffed Martin-Puplett interferometer (Manabe et al., 2003). A suppres-
sion over the other sideband is more than 15 dB but it cannot be ignored in
the retrieval processing. The radio spectrometer of SMILES is the acousto-
optical spectrometer (AOS) (Ozeki et al., 2000). It analyzes the atmospheric
signal and detects its power spectrum. AOS will have two analyzing units
with a resolution of 1.8 MHz (full width at half maximum:FWHM) and each
power spectrum is obtained with a CCD array of 1728 channels. Table 2
shows the main specifications of SMILES. Further details are given in the
SMILES mission plan (NASDA and CRL, 2002) and recent status reports
are presented by (Kikuchi et al., 2010) and (Ochiai et al., 2010).

18
 SMILES achieved that the system noise temperature is less than 400 K.
In terms of the noise level of the brightness temperature, it is approximately
0.7 K for a single scan. It is better than previous millimeter-submillimeter
limb measurements from satellites such as Aura/MLS (Read et al., 2006),
whose system noise temperature is approximately 10,000 K at a 600 GHz
frequency range, and Odin/SMR (Murtagh et al., 2002), whose system noise
temperature is approximately 3,000 K.

19
20
Figure 8: Block diagram of SMILES (NASDA and CRL, 2002).
 Table 2: JEM/SMILES Specifications
System Parameter Description
Frequency bands Band A (624.3–625.5 GHz)
Band B (625.1–626.3 GHz)
Band C (649.1–650.3 GHz)
System noise temperature Less than 700 K
Instrumental height resolution 3.5–4.1 km (nominal)
Frequency resolution 1.8 MHz (FWHM)
Channel separation 0.8 MHz channel
Integration time 0.5 s for each observation point
Altitude range 10–60 km (nominal)
Global coverage 38◦ S–65◦ N (nominal)
Processing time 53 s/scan

21
4 Retrieval Techniques for JEM/SMILES
The retrieval algorithm for SMILES is based on the standard Optimal Es-
timation Method (OEM) applied for atmospheric sounding, which is a well-
established method used in the field of atmospheric science and has been
reported in standard literatures (Rodgers, 1976, 1990, 2000). This section
gives a brief outline of the basic retrieval algorithm, and terms and notations
used in the subsequent descriptions are defined in Rodgers (2000).

The atmospheric state is represented by a state vector x of length n. In

the SMILES, a state vector represents the vertical profiles of target species
concentrations and atmospheric temperature along with other selected pa-
rameters described in Chapter II. A measurement vector y of length m de-
notes the calibrated brightness temperature observed by the SMILES, and
the forward model F is formulated to describe the brightness temperature
according to our knowledge of the physics (including the instrument). A
predicted measurement vector ŷ based on the atmospheric state represented
by x and the forward model parameters represented by b is described by

ŷ = F(x, b). (6)

The retrieval algorithm is based on the OEM applied for atmospheric sound-
ing (Rodgers, 2000). The most probable solution can be derived from sta-
tistical combination of a priori knowledge of x and the information of the
measurement. A priori knowledge is represented by the expected state xa
and its covariance matrix Sa . The most probable state is the one which

22
minimizes the χ2 statistics given by

χ2 = {y − F(x, b)}T · S−1

y · {y − F(x, b)}

+ (x − xa )T · S−1
a · (x − xa ), (7)

where Sy is the covariance matrix of y.

The maximum a posteriori estimate can be derived from statistical com-

bination of a priori knowledge of a state vector x and the information about
the measurement. We use a modification of the Gauss-Newton method called
the Levenberg-Marquardt method (Levenberg, 1944; Marquardt, 1963). The
retrieved state vector xi+1 at the iterative step i + 1 is calculated as

−1 −1 −1
xi+1 = xi + (KT
xi · Sy · Kxi + Sa + γD)

−1 −1
·{KT
xi · Sy · [y − F(xi , b)] + Sa · (xi − xa )}. (8)

The matrix Kxi is a differential weighting function (or the Jacobian) for each
of the retrieval parameters evaluated at xi . xa normally corresponds to the
initial guess x0 , D is a scaling matrix that is usually assumed to be S−1
a ,

and γ is a Levenberg-Marquardt parameter, which is initially set to 0.0001

in this study. For noisy products such as vertical profiles of mesospheric BrO
and HO2 , whose atmospheric densities are small and emissions are weak, the
products should be averaged to obtain the information. We use an algorithm
of averaging for the noisy products following that of the Aura/MLS (Livesey
et al., 2006) (Chapter III).

Characteristics of retrieval results in the standard OEM approach are

shown as follows. A contribution matrix Gy , which is the sensitivity of the

23
retrieved state to the measurement, and an averaging kernel matrix A, which
is the sensitivity of the retrieved state to the true state, are defined as follows:

−1 −1 −1 −1
Gy = (KT
x · Sy · Kx + Sa ) · KT
x · Sy ; (9)

A = Gy · K x . (10)

In the non-linear case, these matrices are calculated by using the results of
the final iteration process. The errors due to the retrieval process are shown
as follows: A retrieval covariance matrix S is defined by

−1 −1 −1
S = (KT
x Sy Kx + Sa ) . (11)

The retrieval precision of the product is defined as the square root of the
diagonal elements of S. The retrieval bias due to incorrect knowledge of the
forward model parameter is given by the error covariance matrix Sf ,

Sf = Gy Kb Sb KT T
b Gy , (12)

where Kb and Sb are a weighting function matrix and a covariance matrix

of b, respectively. In the SMILES, some of these parameters have signifi-
cant errors compared to S. Therefore, it is important to develop an accurate
forward model. In the equation (10) and (11), we ignore γD term of (8)
in accordance with the definition of Rodgers (2000). In the nonlinear case,
these matrices are calculated by using the results of the final iteration.

The retrieval process consists of two main parts; the forward model and
the inverse model. The inverse model deduces the distributions of geophys-
ical parameters from a given set of calibrated brightness temperatures and

24
brightness temperatures simulated using the forward model. The forward
model simulates the brightness temperatures measured by the SMILES for
a given atmospheric state. The output is a convolution of the brightness
temperature calculated with atmospheric radiative transfer with the instru-
mental field of view (FOV) and the spectral response of the AOS.

The brightness temperatures detected by the SMILES antenna is derived

by radiative transfer model, which is defined as a transmission process of
emissions from atmospheric molecules through the Earth atmosphere. Here,
the rays of the emissions are curved by refraction. The emissions from mi-
nor components are calculated by line-by-line but the emissions from major
components like water vapor are calculated by the model. In the line-by-line
calculation, spectroscopic parameters are important for the accuracy of the
retrieval process. The convolution with the FOV uses a measured antenna
pattern by the ground and includes effects from tilt of SMILES. Since the
antenna pattern of SMILES is ellipse, the pattern is changed depending on
tilt of SMILES. Furthermore signal from the other sideband of the sideband
filter and responce functions of AOS are also considerd in the forward moldel.

The measurement noise of SMILES is very small as I mentioned previously

and it is mostly less than 1 K in a single scan. It means that the forward
model must provide the most accurate basis for results. The forward model
algorithm has to be more accurate than ever before, to make the best of
SMILES performance. It also should be required to be fast because of limited
computing resources. Therefore to develop the accurate and fast forward

25
model algorithm is very important issue in the SMILES observation and this
study is one of a main part of this thesis.

26
5 Aim of This Thesis
The various types of atmospheric ozone observation have been performed
since 1970’s. The trend of the ozone depletion is monitored by worldwide co-
operation as previously described. However, the mechanism of the ozone de-
pletion and the prediction of the ozone recovery are still uncertain. SMILES
has been planed to improve our understanding of an expected slow recovery
of ozone from the depletion, climate change and coupled processes of them
by high sensitive measurements of distributions of ozone and ozone-depleting
substance.

To maximize the performance of SMILES, an accurate retrieval algorithm

must be indispensable because accuracy of the data is determined by both
sensitivity of the sensor and accuracy of the retrieval algorithm. Therefore
accurate retrieval algorithm has been developed to introduce high-quality
vertical profiles of ozone and ozone depleting substance. Furthermore for
a requirement of near-real time processing fast retrieval algorithm is also
required. The development the retrieval algorithm to balance competing
requirements for accuracy and high processing speed is one of a purpose of
this thesis. The other purpose of this thesis is estimation for a capability for
ozone high-precision retrieval on SMILES observation based on the algorithm
and differences from previous measurements are introduced.

The development the fast and accurate retrieval algorithmth for SMILES
is described in Chapter II. This Chapter describes the detailed algorithm of
the forward model optimized for the SMILES and the validation and perfor-

27
mance. The study for the capability of ozone observation using SMILES is
described in Chapter III. This Chapter refers to the retrieval errors caused
by model parameter uncertainties such as a priori profile and uncertainty of
the atmospheric temperature, identify the capability for measuring ozone di-
urnal variation in the upper stratosphere, and present a further improvement
of the retrieval algorithm in the lower stratosphere and its random error.

28
References
Bernath PF, McElroy CT, Abrams MC, Boone CD, Butler M, Camy-Peyret
C, Carleer M, Clerbaux C, Coheur PF, Colin R, DeCola P, DeMazie M,
Drummond JR, Dufour D, Evans WF, Fast H, Fussen D, Gilbert K, Jen-
nings DE, Llewellyn EJ, Lowe RP, Mahieu E, McConnell JC, McHugh M,
McLeod SD, Michaud R, Midwinter C, Nassar R, Nichitiu F, Nowlan C,
Rinsland CP, Rochon YJ, Rowlands N, Semeniuk K, Simon P, Skelton
R, Sloan JJ, Soucy MA, Strong K, Tremblay P, Turnbull D, Walker K,
Walkty I, Wardle DA, Wehrle V, Zander R, Zou J Atmospheric Chemistry
Experiment (ACE):Mission overview. Geophys. Res. Lett., 32, L15S01,
doi:10.1029/2005GL022386.

Bevilacqua RM, Hoppel K, Hornstein J, Lucke R, Shettle E, Ainsworth T,

Debrestian D, Fromm M, Lumpe J, Krigman S, Glaccum W, Olivero J,
Clancy RT, Rusch D, Randall C, Dalaudier F, Deniel C, Chassefiere E,
Brogniez C, Lenoble J, First results from POAM II: The Dissipation of
the 1993 Antarctic Ozone Hole Geophys. Res. Lett.;1995;21;909-912.

Bovensman H, Burrows JP, Buchwitz M, Frerick J, Noël S, Rozanov VV,

Chance KV and Goede APH, SCIAMACHY: Mission objectives and mea-
surement modes. J. Atmos.Sci;1999;56(2);127-150

Fischer H and Oelhaf H, Remote sensing of vertical profiles of atmo-

spheric trace constituents with MIPAS limb-emission spectrometers. Appl.
Opt;1996;35(16);2787796

Fischer H, Birk M, Blom C, Carli B, Carlotti M, Clarmann T, Delbouille

29
 L, Dudhia A, Ehhalt D, Endemann M, Flaud JM, Gessner R, Kleinert A,
Koopman R, Langen J, Lopez-Puertas M, Mosner P, Nett H, Oelhaf H,
Perron G, Remedios J, Ridolfi M, Stiller G, Zander R, MIPAS: An instru-
ment for atmospheric chemistry and climate research, ESA Publications
Division, Atmospheric Chemistry and Physics;2008;8;8;2188

Froidevaux L, Jiang YB, Lambert A, Livesey NJ, Read WG, Waters JW,
Browell EV, Hair JW, Avery MA, McGee TJ, Twigg LW, Sumnicht GK,
Jucks KW, Margitan JJ, Sen B, Stachnik RA, Toon GC, Bernath PF,
Boone CD, Walker KA, Filipiak MJ, Harwood RS, Fuller RA, Manney
GL, Schwartz MJ, Daffer WH, Drouin BJ, Cofield RE, Cuddy DT, Jarnot
RF, Knosp BW, Perun VS, Snyder WV, Stek PC, Thurstans RP, Wag-
ner PA, Validation of Aura Microwave Limb Sounder stratospheric ozone
measurements. JGR;2008;113;D15S20, doi:10.1029/2007JD008771.

Holton JR, Haynes PH, McIntyre ME, Duglass AR, Rood RB, and Pfster L,
Stratosphere-troposphere exchange, Rev. Geophys.;1995;33;403-439.

Kikuchi K,Nishibori T,Ochiai S,Ozeki H, Irimajiri Y,Kasai Y, Koike

M,Manabe T, Mizukoshi K, Murayama Y,Nagahama T,Sano T,
Sato Ryota,Seta M,Takahashi C. Takayanagi M, Masuko H,Inatani
J,Suzuki M, and Shiotani M, Overview and Early Results of the Su-
perconducting Submillimeter-Wave Limb-Emission Sounder (SMILES).
JGR;2010;15;D23306;12.

Livesey NJ, Snyder WV, Read WG, Wagner PA, Retrieval algorithms for
the EOS Microwave limb sounder (MLS). IEEE-MTT, 44(5), 1144–1155,
2006. doi 10.1109/TGRS.2006.872327

30
Manabe T, Inatani J, Murk A, Wylde RJ, Seta M, and Martin DH,
A New Configuration of Polarization-Rotating Dual-Beam Interferom-
eter for Space Use. IEEE Transactions on Microwave Theory and
Techniques;2003;vol.51(6):1696–1704

Manabe T, Fukami T, Nishibori T, Mizukoshi K, and Ochiai S, Measurement

and evaluation of submillimeter-wave antenna quasioptical feed system by
a phase-retrieval method in the 640-GHz band. IEICE Trans. Commun.;
2008;E91-B(6):1760-1766

Murtagh D, Frisk U, Merino F, Ridal M, Jonsson A, Stegman J, et al., An

overview of the previous termOdinnext term atmospheric mission. Can J
Phys;2002;80;30919.

NASDA/CRL, JEM/SMILES mission plan, Version 2.1.[Online]. Available:

http://smiles.tksc.nasda.go.jp/document/indexe.html; 2002

Ochiai S, Kikuchi K, Nishibori T, Manabe T, Ozeki H, Mizukoshi K, Ohtsubo

F, Tsubosaka K, Irimajiri Y, Sato R, and Shiotani M, Performance of
JEM/SMILES in orbit. 21th Int’l Symp. Space Terahertz Technol., S8.1,
(Oxford, UK), March 2010.

Ozeki H, Kasai Y, Ochiai S, Tsujimaru S, Inatani J, Masuko H, Takahashi C,

Mazuray L, and Rosen C, Submillimeter-wave spectroscopic performance
of JEM/SMILES. SPIE;2000;4152;255–262.

Read WG, Shippony Z, Schwartz MJ, Livesey NJ, and Snyder WV, The
Clear-sky Unpolarized Forward Model for the EOS Aura Microwave Limb
Sounder (MLS). IEEE Trans. Geosci. and remote sens;2006;44(5).

31
Rusch DW, Bevilacqua RM, Randall CE, Lumpe JD, Hoppel KW, Fromm
MD, Debrestian DJ, Olivero JJ, Hornstein JS, Guo F, Shettle EP, Vali-
dation of POAM II Ozone Measurements with Coincident MLS, HALOE,
and SAGE II Observations. JGR;1997;102(23);615–627

Russell JM, Gordley LL, Park JH, Drayson SR, Hesketh DH, Cicerone RJ,
Tuck AF, Frederick JE, Harries JE, and Crutzen P, The Halogen Occulta-
tion Experiment. JGR; 1993;98;D6;10;777–797.

Rodgers CD, Retrieval of atmospheric temperature and composition from

remote measurements of thermal radiation. Rev Geophys and Space
Phys;1976;14(4);609-624.

Rodgers CD, Characterization and error analysis of profiles retrieved from

remote sounding measurements. JGR;1990;95(5);5587-5595.

Rodgers CD, Inverse methods for atmospheric sounding: theory and practice,
Series on atmospheric, oceanic and planetary physics, vol.2. Singapole,
World Scentific; 2000

Waters JW, Read WG,Froidevaux L, Jarnot RF, Cofield RE, Flower DA,
Lau GK, Pickett HM, Santee ML, Wu DL, Boyles MA, Burke J. R, Lay R.
R, Loo M. S, Livesey N. J, Lungu TA, Manney GL, Nakamura LL, Perun
VS, Ridenoure BP, Shippony Z, Siegel PH, Thurstans RP, Harwood RS,
Pumphrey HC, and Filipiak MJ. The UARS and EOS Microwave Limb
Sounder (MLS) experiments. J.Atmos.Sci.;1999;56;194-218.

WMO (World Meteorological Organization), Scientific Assessment of Ozone

32
Depletion: 2010, Global Ozone Research and Monitoring Project Report
No. 52, 516 pp., Geneva, Switzerland, 2011.

33
Part II
Operational Retrieval
Algorithms
6 Introduction
As a part of the ground system for the SMILES, a level 2 data processing
system (DPS-L2) has been developed. The retrieval algorithm of the DPS-
L2 is based on the standard Optimal Estimation Method (OEM), which is a
well-established method used in the field of atmospheric science and has been
reported in standard literatures (Rodgers, 1976, 1990, 2000). The DPS-L2
converts the calibrated measurements of brightness temperatures, as referred
Level 1B data, which is a processed SMILES data containing calibrated limb
emission spectra, into distributions of geophysical parameters along the mea-
surement track of the instrument. The main retrieved geophysical parameters
are the vertical profiles of target species and temperature and tangent alti-
tudes. The retrieval process consists of two main parts; the forward model
and the inverse model. The forward model simulates the brightness temper-
atures observed by the SMILES for a given atmospheric state. The output
is a convolution of the brightness temperature calculated with atmospheric
radiative transfer with the instrumental field of view (FOV) and the spectral
response. Since the forward model must provide the most accurate basis for
results and be implemented under limited computing resources, the forward
model algorithm for the operational code has to be accurate and fast. The
inverse model deduces the distributions of geophysical parameters from a

34
given set of calibrated brightness temperatures and brightness temperatures
simulated using the forward model.
This Chapter describes the development of the DPS-L2 along with the
details on its algorithm and the algorithm performance. Section 7 presents
the framework of the DPS-L2, and Section 8 presents the fundamental re-
trieval theory. Section 9 gives the detailed algorithm of the forward model
optimized for the SMILES. Finally, the validation and performance of this
algorithm are reported in Section 10.

7 Framework of DPS-L2
7.1 Ground Data Processing System and Data Flow

The SMILES ground data processing system consists of an Experiment Op-

erating System (EOS) and Data Processing System (DPS). The EOS receives
telemetry and sends commands. The DPS is divided into two main units; a
data processing system for level 0 and level 1 (DPS-L0/L1) and the DPS-L2.
Downlinked raw data from the SMILES will be received by the DPS-
L0/L1 at the User Operation Area (UOA) in the Tsukuba Space Center
(TKSC). The DPS-L0/L1 processes the raw data consisting of house keep-
ing (HK) data and science data into level 1B data immediately. The level
1B data are periodically transfered via non-network media to a data server
connected to the internet. The DPS-L2 located at the Institute of Space and
Astronautical Science (ISAS) automatically receives the level 1B data from
the data server through the internet and processes the level 1B data into
level 2 data. The level 2 data are converted into the HDF-EOS format and
are distributed to users accompanied with the ancillary data on the SMILES

35
status through a web server. A list of the SMILES data sets is presented in
Table 3.

36
 Table 3: JEM/SMILES Data Sets
Data Type Description
Raw Unprocessed mission data at binary packets
Level 0 Reconstructed, unprocessed mission data at binary packets
Level 1b Calibrated instrument radiances and related data
Level 2 Derived geophysical variables at the same resolution and
location as the Level 1 source data
Level 3 Variables mapped on uniform space-time grid scales, usu-
ally with some completeness and consistency

37
7.2 Structure of the DPS-L2

Figure 9 shows an overview of the DPS-L2. As the DPS-L2 is designed to

be an automatic operating system, a control function of the DPS-L2 trans-
fers the level 1B data and invokes the data processing automatically. Data
handling and visualization functions are to support the analysis of the level
2 data and to carry out data checks in real time. These functions are im-
plemented on the Gfdnavi (geophysical fluid data navigator) (Horinouchi et
al., 2007). Gfdnavi is a suite of software that facilitates databasing, analysis,
data publication, and visualization of geophysical fluid data. The princi-
pal function of the DPS-L2 is data processing. In the following sections,
we discuss the algorithms of the operational retrieval code (ORC) and their
performance.
The DPS-L2 is designed to use a highly portable software and works
on a general Linux operating system. The operational environments of the
DPS-L2 are summarized in Table 4.

38
 Figure 9: Overview of DPS-L2

Table 4: Operation Environment of DPS-L2

System Software
Operating System Linux (kernel: up to 2.6)
Language gcc (or Intel C++), Ruby, GNU Octave, java
Database MySQL
Web Application GFDnavi, ruby on rails

39
7.3 Requirements for the ORC

There are three significant requirements for developing the ORC, ie. the ac-
curacy, processing speed and portability. The accuracy of the ORC should be
sufficiently high to ensure high sensitivity of the SMILES, and is determined
by estimation of errors expected in the retrieval. The estimated errors are
calculated by the inverse model assuming several parameters such as SMILES
radiometric error, FOV characteristics, spectroscopic parameters, and other
possible inevitable errors in the forward model. We decided the accuracy
of the forward model should be better than 1 % of radiometer noise. Data
processing should be carried out in near-real-time (a single scan takes 53 s).
Hence, a fast algorithm is required in order to simplify the operations of the
system.
For portability, the ORC is written by using two programming languages:
C programming language and GNU Octave. GNU Octave is a high-level lan-
guage, primarily intended for numerical computations (it is mostly compat-
ible with MATLAB). It can be linked with ATLAS (Whaley et al., 2000) to
accelerate the processing speed of the system. The forward model is written
in C programming language because of its high processing speed and its flex-
ibility to implement considerably complex numerical integration. However,
GNU Octave is the most suitable language for the inverse model because the
main parts of this model are matrix calculations.

40
8 Basic Retrieval Algorithm
The retrieval algorithm is based on the OEM applied for atmospheric sound-
ing (Rodgers, 1976, 1990, 2000). The details are in Introduction (Chapter
I).

8.1 State Vector x

The components of the state vector x , which is mentioned in Introduction

(Chapter I), for the DPS-L2 are as follows:

1. the vertical profiles of the target species concentration in volume mixing

ratio;

2. the vertical profiles of atmospheric temperature;

3. the vertical profiles of atmospheric continuum absorption;

4. the offset of the instrument pointing.

The offset of the instrument pointing is a scalar quantity and the other
components are vector quantity gridded on the geometrical altitude. The
uncertainty of the atmospheric continuum absorption is mainly introduced
due to the uncertainty in H2 O distributions. The uncertainties of the instru-
ment pointing are caused by the time synchronous error between the clocks
in the SMILES and the ISS, an error in a reference pressure profile, etc.
The standard approach in microwave-limb soundings is to derive the in-
formation on the atmospheric temperature and the instrument pointing from
oxygen emission lines (Wehr et al., 1998; Livesey et al., 2006). However, since

41
the SMILES does not observe the oxygen emission lines, these two compo-
nents can be mainly derived from the ozone emission line at 625.37 GHz
(NASDA and CRL, 2002; Verdes et al., 2002). The atmospheric temper-
ature can be retrieved with ozone concentration at a time thanks to the
different frequency dependence between weighting functions with respect to
ozone and temperature. Most of the information comes from the brightness
temperature near the line center where the opacity is sufficiently high to con-
sider the brightness temperature to be independent of ozone concentration
but to depend on the atmospheric temperature.
The SMILES can measure the three bands but it measures the two bands
simultaneously. Combinations of the simultaneous measurement bands are
AB, AC, and BC. For the case of AB, band A and B are processed indepen-
dently in the launch ready algorithm. However, for the case of AC or BC,
the processing should be performed band A (or B) and C, in this order. The
reason is that the precision of the retrieved instrument pointing by using of
band A (or B) is better than by using band C, and the retrieval result is
passed to the retrieval processing of band C.
Our retrieval approach for each band consists of two processes. The
first process is the retrieval of the target species in the upper row of Table
1, atmospheric temperature, atmospheric continuum, and the offset of the
instrument pointing. These component are retrieved simultaneously. The
second process, which is performed after the first processes for the two bands,
is the retrieval of the target species in the lower row of Table 1, which are
noisy products. These products are retrieved for each product.

42
 We estimate availability of the simultaneous retrieval for the first process.
Figure 10 indicates the difference of the incremental error for retrieved ozone
between simultaneous (blue line) and sequential retrieval (green line) of ozone
and HOCl. The red line is the retrieval precision of ozone, which is the
square root of the diagonal elements of S. The simulation setting is that the
retrieval altitude step is set to 3 km in 4–70 km region, the measurement
tangent altitude step is set to 2 km in 0–80 km region, and the instrument
parameters are conformed to Table 5. This simulation is performed on band
A using all channels. Here we do not estimate the availability of the channel
selection to simplify the retrieval scheme.
The incremental error in ozone retrieval for the simultaneous retrieval is
approximately 0.1% in the stratosphere and is almost independent of the a
priori. However the incremental error in ozone retrieval for the sequential
retrieval depends on the assumptions of profiles of other molecules such as
HOCl. For the case that the true profile of HOCl is 100% and 200% greater
than the a priori, the incremental errors of ozone are approximately 1.0% and
2.0% in the stratosphere, respectively. Furthermore, the iteration number of
the sequential retrieval is larger than that of the simultaneous retrieval. Since
this behavior is same as other molecules in the upper row of Table 1, we adopt
the simultaneous retrieval for the first process.

9 Optimized Forward Model

9.1 Overview of Forward Model

The forward model calculates the brightness temperatures under the given
atmospheric state and also the Jacobians with respect to temperature, the

43
 true:a priori = 2:1
70
Retrieval error
without HOCl
60 with HOCl

50
Altitude [km]

40

30

20

10

0
0.001 0.01 0.1 1 10 100
Error Ratio [%]

true:a priori = 3:1

70
Retrieval error
without HOCl
60 with HOCl

50
Altitude [km]

40

30

20

10

0
0.001 0.01 0.1 1 10 100
Error Ratio [%]

Figure 10: Simultaneous retrieval of ozone and HOCl. The red line indicates
the retrieval precision in ozone retrieval, the green line indicates the incre-
mental error in ozone retrieval without HOCl, and the blue line indicates
the incremental error in ozone retrieval with HOCl. In all these cases, the
true profiles of HOCl are 100% (Right) and 200% (Left) greater than the a
priori profile of HOCl. Here, the incremental error is defined as the difference
between the true profile and the retrieved profile.

44
target species concentration, atmospheric continuum, which mainly comes
from H2 O, and the offset of the instrument pointing. The forward model
carries out the following calculations:

1. ray tracing for the evaluation of the atmospheric state along the limb
line-of-sight (LOS) with refraction

2. absorption coefficient calculation for a predefined frequency grid

3. radiative transfer calculation of the single-ray brightness temperature

4. Doppler shift calculation from the ISS velocity, the rotational velocity
of the earth, and wind velocity

5. signal convolution with the SMILES antenna FOV considering the in-
clination of the ISS

6. summation of both sideband signals according to sideband ratio

7. convolution of infinite resolution spectrum and the instrument fre-

quency response

The detailed algorithm is given in the following subsections.

9.2 Radiative Transfer Calculation

9.2.1 Ray Tracing

Refracted ray of limb measurement with arbitrary tangent height is calcu-

lated by ray tracing procedures with the following assumptions and condi-
tions. The 1-D radiative transfer calculations described later are defined
along the ray calculated here. The earth is assumed to be a sphere whose

45
radius is curvature radius at the tangent point. The atmospheric refractivity
at an altitude z [km] is approximated by

n(z) = 1 + 315 · 10−6 · exp(−z/7.35). (13)

The variations in the atmosphere are negligible for altitudes above 2 km in

the SMILES frequency region (NASDA and CRL, 2002). The step size used
in ray tracing is approximately 0.1 km. This step size and the step size used
in the path integration in the radiative transfer computation are independent
of each other (the step size of a path integration is approximately 1 km).

9.2.2 Absorption Coefficient

For a molecule i with its cross section given by αi , the absorption coefficient
k is given by
! xi p
k(ν, z) = αi (ν, z) + kcont (ν, z), (14)
i
kb T
where z is the altitude, ν is the frequency, xi is the volume mixing ratio of
molecule i, p is pressure, T is temperature, and kb is the Boltzmann constant.
The first term of the right-hand side of Eq. 14 can be calculated line-by-
line (LBL) using the JPL Spectral Line Catalog (Pickett et al., 1992), and
the second term of the right-hand side of Eq. 14, kcont , is the absorption
coefficient of the background continuum for water vapor and dry air (O2
and N2 ) calculated by using Liebe’s MPM (Liebe, 1989, 1993). The line
shape is described using the Voigt function, which is implemented using a
fast algorithm (Kuntz, 1997).
Errors associated with line parameters such as air-broadening parame-
ters and partition functions were estimated by Verdes et al. (Verdes et al.,

46
2005a,b). According to the papers, partition function errors do not intro-
duce significant error for most of the target species of the SMILES except
BrO and the uncertainties in air-broadening parameters lead to comparable
errors in the retrieval. These studies on the laboratory measurements of
air-broadening parameters have resulted in significant improvements in the
retrieval accuracy; therefore, we use the air-broadening parameters of main
target species that have already been measured by some research groups (Ya-
mada and Amano, 2005; Oh and Cohen, 1994; Yamada et al., 2003; Drouin,
2004; Goyette et al., 1998; Fischer et al., 2003).
Since the computing cost will be sufficiently reduced by improving al-
gorithm as shown below, k should be recalculated in each scan and each
iteration for the accuracy.

9.2.3 Radiative Transfer

Radiative transfer is defined as the process of transmission of the electro-

magnetic radiation through the atmosphere. The forward model assumes a
homogeneous distribution of geophysical parameters along the horizontal. A
single-ray radiance at a sensor position ss and at a frequency ν, Ip (ν, ss ), is
given by
" ss
−τ (ν,se ,ss )
Ip (ν, ss ) = I0 e + B(ν, T )k(ν, s)e−τ (ν,s,ss ) ds. (15)
se

The first term on the right-hand side of Eq. 15 corresponds to the radiation
entering the atmosphere along the direction of the LOS. τ denotes the optical
depth between two positions s1 and s2 along the LOS and B(ν, T ) is the
Planck function, which gives the radiation of a black body at temperature T

47
and frequency ν: " s2
τ (s1 , s2 ) = k(ν, s)ds; (16)
s1

2hν 3 1
B(ν, T ) = 2 hν . (17)
c e kb T − 1
Here, h is Planck’s constant and c is the speed of light. k(ν, s) is the absorp-
tion coefficient at position s along the LOS.

9.2.4 Doppler Shift

The Doppler shift must be taken into account in our forward model. The
factors affecting the Doppler shift are the LOS components of the ISS velocity,
the rotational velocity of the earth, and wind velocity. The velocity of the
ISS is approximately 8 km/s, and the rotational velocity of the earth at the
tangent point on the equator is approximately 460 m/s.
Next, we discuss the effect of wind on the Doppler shift. As the maxi-
mum velocity of wind in the stratosphere is approximately 100 m/s, there is a
Doppler shift of approximately 0.2 MHz. This value is not negligible as com-
pared to the frequency resolution of the acousto-optic spectrometer (AOS),
i.e, 1.8 MHz (FWHM). Figure 11 shows the effect of wind on ozone retrieval.
Ozone is the molecule that is the most susceptible to these types of effects.
The other molecules are not susceptible to these effects, and the incremental
errors caused by the Doppler shift are sufficiently small as compared to the
retrieval precision. This behavior is also exhibited by other model parameter
errors. The red line in Figure 11 indicates the precision of ozone retrieval,
and the other lines indicate the incremental error in ozone retrieval for wind
velocities equal to 50, 10, and 5 m/s. If the difference of the “true” velocity

48
 Ozone
70
Retrieval error
5 m/s
60 10 m/s
50 m/s
50
Altitude [km]

40

30

20

10

0
0.0001 0.001 0.01 0.1 1 10 100
Incremental Error [%]

Figure 11: Effect of wind. The red line indicates the retrieval precision of
ozone. The other lines indicate the error due to the difference between the
reference profile and the true profile of wind.(The wind velocities for the pink,
blue, and green profiles are 50, 10, 5 m/s, respectively. The definition of the
incremental error is the same as that in Figure 10

and “reference” velocity of wind is less than approximately 10 m/s, the in-
cremental error can be less than 0.005%. This value is sufficiently small as
compared to the retrieval precision of ozone. Therefore, we use re-analysis or
realtime-analysis data for the reference wind velocity instead of the retrieving
the wind speed.

49
9.3 Accurate Instrument Model for the SMILES

The instrument model is based on the knowledge of the SMILES instrument

according to the SMILES missionplan (NASDA and CRL, 2002) and the
recent experimental results of the flight model.

9.3.1 FOV Convolution

The FOV of the antenna can be taken into account in the following manner.
The brightness temperature convolved by the FOV at its central tangent
altitude z0 , TA (ν, z0 ), is obtained by
" zmax
TA (ν, z0 ) = P̄ (z, z0 ) · Tp (ν, z)dz, (18)
zmin

where Tp (ν, z) is a single-ray brightness temperature, P̄ (z, z0 ) is a nor-

malized antenna beam pattern convolved along the horizontal direction, z is
the tangent altitude, and zmin and zmax are lower and upper tangent altitudes
of field viewed by specified main beam when the antenna is pointing to the
tangent altitude of z0 .
It is not improbable that the SMILES antenna scan axis is tilted by 15◦
from the horizon, which is inclination of the ISS, from the normal inclination.
Since the antenna beam of the SMILES has a horizontally flattened elliptical
pattern, P̄ (z, z0 ) depends on the inclination of the antenna scan axis. Figure
12 shows the incremental error in retrieved ozone due to the inclination of the
antenna scan axis. If the antenna scan axis inclines by 15◦ , the incremental
error is approximately 1% in the stratosphere, and this value is approxi-
mately two times larger than the retrieval precision. Therefore, the effect of
the inclination of the antenna scan axis should be taken into consideration in

50
the derivation of P̄ (z, z0 ). The antenna beam pattern P (θ̃, ψ̃) is measured on
the ground with high accuracy, where θ̃ and ψ̃ are the polar and azimuthal
angles respectively with respect to the xyz-axis fixed to the antenna. The
y-axis of the antenna is the direction of the main beam. The x-axis of the
antenna is perpendicular to the plane formed the y-axis and a longer axis
of elliptically shaped main reflector, and is aligned with rotating axis of the
antenna that is elevation steerable. The z-axis of the antenna completes the
right-handed axes and is positive toward zenith. P (θ̃, ψ̃) is measured by a
phase-retrieval method (Manabe et al, 2008), which is considered to be ef-
fective for measuring and evaluating large-scale submillimeter-wave reflector
antennas.
The following equations define θ and φ. Those are the polar and azimuthal
angles which are exactly the same with θ̃ and ψ̃ when the x-axis of the antenna
becomes perpendicular to the vector toward the center of the earth.

θ̃ = cos−1 (−sinθ · cosψ · sinα + cosθ · cosα) , (19)

# $
−1 sinθ · sinψ
ψ̃ = tan , (20)
sinθ · cosψ · cosα + cosθ · sinα
where α is the angle of the x-axis of the antenna from the local horizontal
plane at the tangent point. P (θ̃, ψ̃) convolved by θ, P ∗ (θ), is
" ψ2
∗
P (θ, α) = P (θ̃(θ, ψ, α), ψ̃(θ, ψ, α)) · cosθ · dψ. (21)
ψ1

By using the pre-calculated P ∗ (θ, α) for every one degree of α, the incre-
mental error for ozone retrieval can be reduced to less than 0.001%. Finally,
P̄ (z, z0 ) is translated from P ∗ (θ) by ray tracing with refraction in every op-
eration calculation.

51
 O3-normal
70
Retrieval error
1o
60 5o
10o
50 15o
Altitude [km]

40

30

20

10

0
0.0001 0.001 0.01 0.1 1 10 100
Incremental Error [%]

Figure 12: Error due to inclination of antenna scan axis. The red line indi-
cates the retrieval precision of ozone. The other lines indicate the error due
to the inclination of the antenna scan axis. The inclinations of the antenna
scan axis in the aqua blue, pink, dark blue, and green profiles are 15◦ , 10◦ ,
5◦ , and 1◦ , respectively. The definition of the incremental error is the same
as that in Figure 10.

52
9.3.2 Sideband Ratio

To separate the two sidebands, the SMILES is equipped with a quasi-optical

sideband separator in the submillimeter range. The sideband separator of
the SMILES, which is a modified Martin-Puplett interferometer (Manabe et
al., 2003), can reject the image band signal more than 20 dB in the SMILES
measurement bands. Even this rejection ratio, the signal from the image band
rises up to approximately 2 K in the lower stratosphere. Since this value is
not negligible, the image signal should be calculated in our forward model by
b
using the measured response function. The transmission function, Ki,j (ν, T ),
from the antenna (j=ANT) or cold sky terminator (j=CST) to mixer-i in
upper sideband (b=USB) or lower sideband (b=LSB) can be expressed by (19)
where mbi,j (T ), ν0i,j
b b
(T ), αi,j (T ), are constants those have been experimentally
determined based on SMILES ground tests, and i=R or T, j=ANT or CST,
b=USB or LSB, T is the temperature of sideband separator.

b
Ki,j (ν, T ) = mbi,j (T )(ν − ν0i,j
b
(T ))2 + αi,j
b
(T ). (22)

The brightness temperature after the SIS mixer-i, Tmix(i) (ν, z0 ), is given
by (NASDA and CRL, 2002)
% LSB
&T % &
Ki,a (νLO − νif ) TA (νLO − νif , z0 )
Tmix(i) (ν, z0 ) = LSB ·
Ki,c (νLO − νif ) Tc (νLO − νif )
% USB &T % &
Ki,a (νLO + νif ) TA (νLO + νif , z0 )
+ USB · .
Ki,c (νLO + νif ) Tc (νLO + νif )

Here, νif is the intermediate frequency (IF), νLO is the local frequency,
which is 637.32 GHz, and TA and Tc are the brightness temperature from
ANT and CST ports, respectively.

53
9.3.3 Frequency Response

The SMILES carries two units of AOSs (Ozeki et al., 2000). Each AOS has
1728 channels and covers a bandwidth of approximately 1.2 GHz. Their
frequency resolution is approximately 1.8 MHz (FWHM), and the channel
separation is typically 0.8 MHz. The central frequency of each channel is well
related to channel number by 3rd-order polynomial. The frequency of AOSs
are calibrated with a comb generator during every scan. The coefficients of
the polynomial are supplied by L1B processing.
According to (NASDA and CRL, 2002) and (Ozeki et al., 2000), the
brightness temperatures after the mixer i, Tmix(k ) (ν, z0 ), are convolved with
the channel response function of the AOS l, HAOS(l) (ν − νj ), for each channel
and the brightness temperature of the jth channel, TAOS(l) (νj , z0 ), is denoted
by ' νmax
νmin
HAOS(l) (ν − νj ) · Tmix(i) (ν, z0 )dν
TAOS(l) (νj , z0 ) = ' νmax . (23)
νmin
HAOS(l) (ν − νj )dν
Here, νj is the central frequency of the jth channel. The channel response
function of an AOS is represented by the superposition of some Gaussian
profiles, and the channel response function for a channel j is given by
Ni
! # $
Ai,j (ν − νj − xci,j )2
HAOS(l) (ν − νj ) = ( · exp −2 2
, (24)
i−1 wi,j π/2 wi,j

where Ai,j , wi,j , and xci,j are ith parameters of the Gaussian fit. Ai,j , wi,j ,
and xci,j depend on each channel of the AOS. The forward model takes in
the dependence on the channel based on the measurements.

54
9.4 Fast Algorithm
9.4.1 Line Selection

Since the absorption coefficient is calculated by LBL, except the background

continuum for H2 O and dry air, the computing cost of the absorption coef-
ficient is proportional to the number of absorbing lines. Therefore, it is nec-
essary to select the lines to be calculated as minimum in number as possible
to reduce computing cost without sacrificing the accuracy of the calculation.
In our estimation, lines that affect more than 5.0e–4 K on the target band
should be calculated to satisfy the accuracy. Figure 13 shows the effect of line
selection on ozone retrieval. The incremental error due to this line selection
(green line) is less than 0.01% in the stratosphere. This value is sufficiently
small as compared to the retrieval precision (red line).
The total number of selected lines is approximately 2000 to 3000 per
band. Approximately 80% of these lines are outside the bands, and contri-
bution of these lines to the band is not significant because the intensity is
relatively low and it does not have steep spectral feature. Therefore, the ab-
sorption coefficient of the lines outside the bands can be calculated by using
coarse and equally spaced frequency grid (approximately 10 MHz) by spline
interpolation. This results in a speed increase of approximately 10.

9.4.2 Frequency Selection

The computing cost of the forward model is proportional to the number of

frequency grid points. The interval of the frequency grid around a line center
should be suitable for the Doppler width of the line in order to achieve the
required accuracy. (Since the Doppler width is approximately 0.4 MHz, the

55
 Ozone
70
Error Ratio
60 Incremental Error
50
Altitude [km]

40
30
20
10
0
0.001 0.01 0.1 1 10 100
Error Ratio [%]
Figure 13: Error due to line selection. The red line is the retrieval precision
of ozone, and the green line indicates the incremental error in retrieved ozone
due to line selection. The definition of the incremental error is the same as
that in Figure 10.

56
interval of the frequency grid should be less than approximately 0.1 MHz.)
On the other hand, the interval can be wide for line wings because the signal
level of the line wings is low, and the change is smooth. By optimizing the
frequency grid, we can develop a fast algorithm. However, it is difficult to
automatically optimize unequally spaced frequency grid.
In our method, the frequency grid is optimized by a sequential process as
follows. First, the spectra are calculated by using an initial frequency grid
defined by the line centers of a target band and interpolated by a cubic spline
function. Next, by comparing the above spectra with the spectra calculated
by using the 0.1 MHz equally spaced frequency grid, the maximum difference
frequency point is determined. This frequency point is added to the initial
frequency grid, and the spectra are calculated by using the new frequency
grid. This process is repeated until the maximum difference frequency point
is less than 0.001 K.
By using this method, we have found that approximately 900 unequally
spaced frequency grid of each band represents a high-accuracy spectrum com-
parable to the 0.1 MHz equally spaced frequency grid (in this case, we require
12000 points in a band). Using this method, the calculation speed increases
by more than a factor of 10. The residual difference of the two frequency
sets is less than 0.001 K in all tangent altitude ranges (Figure 14).

10 Results
10.1 Validation of ORC

To validate the ORC, we have compared the various output of the ORC with
those of the SMOCO (SMILES observation retrieval code) (NASDA and

57
 40.00(km)
Equal frequency step (0.1 MHz)
1000

100

10
Brightness Temp. (K)

0.1

0.01

0.001

0.0001
624.4 624.6 624.8 625.0 625.2 625.4
Frequency (GHz)

Unequal frequency step

1000

100

10
Brightness Temp. (K)

0.1

0.01

0.001

0.0001
624.4 624.6 624.8 625.0 625.2 625.4
Frequency (GHz)

Residual error
Brightness Temp. (K)

0.002

0.000

−0.002
624.4 624.6 624.8 625.0 625.2 625.4
Frequency (GHz)

Figure 14: Residual error due to frequency selection. Top: spectrum of band
A obtained by using equal frequency step of 0.1 MHz, middle: spectrum of
band A obtained by using unequal frequency steps produced by our algo-
rithm, and bottom: residual error.
58
CRL, 2002; Kasai et al., 2008). The SMOCO is a simulator that includes
the radiative transfer calculation for simulating atmospheric sounding in the
millimeter and submillimeter wavelength range. The SMOCO has been com-
pared with several models such as ARTS (Bühler et al., 2005), MOLIERE
(Ueban et al., 2004), and MAES (Ochiai et al., 2000), and its accuracy has
been verified (Melsheimer et al., 2005). Figure 15 shows the error ratio of the
absorption coefficient for the ozone line at 625.37 GHz under the same at-
mospheric conditions. The error ratio is less than 0.03%. This difference can
be attributed to the difference in the line shape model of the Voigt profile.
There are no fundamental problems in the absorption coefficient calculation.
Next, we have compared the monochromatic brightness temperature of the
limb emission from the ozone line obtained using the ORC and SMOCO. In
this case, we have ignored the effects of the instruments such as the antenna
and AOSs. Figure 16 shows the error of the brightness temperature obtained
using the ORC and the SMOCO. The error in both cases is less than 0.2%
(less than 0.05 K error in absolute value). The largest difference is appeared
at 10 km and this altitude range corresponds to the tropopause. Around the
tropopause, the gradient of absorption with respect to altitude is so steep
that small difference of models tends to be amplified. We concluded that this
level of difference is acceptable in the retrieval for the stratosphere, which is
our target and that there are no fundamental problems encountered in the
radiative transfer calculation.

59
 0.005
0km
10km
20km
0 30km
40km
50km
-0.005
Error [%]

-0.01

-0.015

-0.02

-0.025
624.2 624.4 624.6 624.8 625 625.2 625.4 625.6
Frequency [GHz]

Figure 15: Residual error of absorption coefficients at 0 km (red), 10 km

(green), 20 km (blue), 30 km (pink), 40 km (aqua), and 50 km (yellow). The
residual error is defined by (SMOCO-ORC)/SMOCO.

60
 0.04

0.02

-0.02
Error [%]

-0.04

-0.06

-0.08
0 km
-0.1 10 km
20 km
30 km
-0.12 40 km
50 km
-0.14
624.2 624.4 624.6 624.8 625 625.2 625.4 625.6
Frequency [GHz]

Figure 16: Residual error of brightness temperature. Legends are the same
as those used in Figure 15

61
10.2 Error Budget

Here, we focus on the retrieval errors in the ozone profile in order to discuss
the error budgets, because ozone is the representative species of the SMILES
mission. The main error budgets are

• Smoothing error

• Retrieval noise

• Forward model parameter errors

– In sufficient information on the profiles of nonretrieved parameters

– Approximations of the instrument functions

– Incorrect input parameters

– Approximations of the fast algorithm.

The retrieval noise can be evaluated from the covariance of measurements,

Sy . In case of radiometer whose spectral channels are almost uncorrelated
each other, Sy is a diagonal matrix and each diagonal element, σy , is calcu-
lated by
Tsys + TA
σy = √ , (25)
Bτ
where TA is the measurement brightness temperature, Tsys is the system
noise (500 K), B is the noise band width (2.5 MHz), and τ is the integration
time (0.5 s). σy is one order smaller than that of existing satellite sensors,
and the error of ozone retrieval mainly due to retrieval noise is less than
0.5% at 30 km (Figure 19). This estimation is based on the standard mid-
latitude profile, and the standard deviation of a priori profile is assumed to

62
be 100% for all species. Because of the high sensitivity, the forward model
parameter errors cannot be considered to be negligible as compared to the
retrieval noise. In some cases, the forward model parameters are the most
significant errors. Since most of the forward model parameter errors are
systematic errors, they affect the accuracy of the retrieved profiles. We have
studied the technique to reduce these errors and have developed an accurate
algorithm on the basis of these studies. As we have showed in Section 8.1,
this effect can be reduced to retrieve all species whose signal is larger than a
few kelvins (species in the upper row of Table 1). The instrument functions
have been designed to model the measurements, which have been carried out
meticulously for each module of the SMILES on the ground by the hardware
team of the SMILES. In the determination of the attitude of the ISS for the
antenna beam pattern, the error due to FOV convolution is minimized and
assumed to be negligible. Basic studies on the incorrect input parameters
such as spectroscopic parameters have been carried out by the University of
Bremen. However, further studies have to be carried out after the launch.
Although these calculations increase the computing cost since importance
is given to accuracy instead of the computing cost, the computing cost can
be reduced by using the fast algorithm given in Section 9.4. The speed of
calculation using this algorithm is nearly ten times higher than that of the
normal algorithms such as the line selection algorithm and frequency selection
algorithm without degradation of the accuracy. The accuracy of the spectra
of the stratospheric minor species is up to 0.01 K (1% of the system noise).

63
 70
O3
HCl
CH3CN
60 HOCl
HNO3
asym18-O3
50 sym17-O3
asym17-O3
Altitude [km]

ClO
40

30

20

10

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Error ratio [-]

Figure 17: Retrieval precision of target species that can be retrieved from
single-scan data.

64
10.3 Algorithm Performance and Hardware System

The algorithm performance has been tested using a 3.16–GHz Quad-Core

Intel Xeon processor. The test was performed using the single-scan data of
band A, which requires the longest processing time among the three SMILES
bands. In band A, seven molecules (Table 1), temperature, and the offset
of the tangent altitude are retrieved simultaneously, while in bands B and
C, only four molecules (Table 1) are retrieved simultaneously. Hence, the
processing time of bands B and C is approximately 30% lesser than that
of band A. We have used the Intel C++ compiler and the 64-bit linux OS
because the Intel C++ compiler is 2.5 times faster than gcc, and the 64-bit
linux OS is 1.5 times faster than a 32 bit OS. The total processing time of
the forward and inverse models without iteration is 56 s per core. To assume
four iterations, the total processing time for a single scan, which has two
bands, is performed 1 min per core with 8 parallel processing.
According to this estimation, we have designed the hardware system con-
figuration suitable for the DPS-L2. Assuming three times redundancy and
approximately four times iterations, the runtime processing is performed us-
ing five servers that have two Quad-Core processors compatible with the
3.16-GHz Quad-Core Intel Xeon on the safe side. Further, two or three small
servers are required to process the level 2 data, to manage the database, and
to distribute the level 2 data along with the information of the SMILES to
users. The hardware system configuration is very compact, and it is conve-
nient to operate and maintain the system.

65
11 Summary
We have developed a launch ready retrieval algorithm. This algorithm is
based on a modification of the Gauss-Newton method called the Levenberg-
Marquardt method so that it can support nonlinear cases also. The most
important objective of this study is how to realize a highly accurate forward
model. As shown in Section 10, this retrieval algorithm can keep the forward
model parameter errors at a low level compared to the retrieval precision.
Furthermore, the algorithm is sufficiently fast to process the single-scan data
within 1 min with 8 parallel processing. Using this algorithm, we expect
to obtain data with high accuracy compared to the data obtained using the
existing sensors in orbit.

66
References
Baron P, Merino F, and Murtagh D, Simultaneous retrievals of temperature
and volume mixing ratio constituents from non-oxygen Odin submillimeter
bands. Opt.;2001;40(33);6102-・110.

arts Bühler SA, Eriksson P, Kuhn T, Engeln A, and Verdes C, ARTS, the
atmospheric radiative transfer simulator. JQSRT;2005;91(1);65–93.

Drouin BJ, Temperature dependent pressure induced lineshape of the HCl J

= 1 ← 0 rotational transition in nitrogen and oxygen. JQSRT;2004;83(3–
4);321–331.

Fischer J, Gamache RR, Goldman A, Rothman LS, and Perrin A, Total

internal partition sums for molecular species in the 2000 edition of the
HITRAN database. JQSRT;2003;82(1–4);401-・12.

Goyette MT, Cohen EA, and Lucia CF, Pressure broadening of HNO3 by
N2 and O2: an intercomparison of results in the millimeter wave region.
JQSRT;1998;60(1);77–84.

Horinouchi T, Nishizawa S, Watanabe C, Morikawa Y, Koshiro T, Ishiwatari

M, Hayashi Y, and Shiotani M, Development of Gfdnavi: a new desk-
top/server tool for geophysical fluid database, analysis, and visualization.
Proceedings of Data Engineering Workshop 2007;D2-8;8 (in Japanese)

Irimajiri Y, Noguchi T, Shi SC, Manabe T, Ochiai S, and Masuko H, A

650-GHz band SIS receiver for balloon-borne limb-emission sounder. Int.
J. Infr. Millimeter Waves;2000;21(4);519–526.

67
Irimajiri Y, Manabe T, Ochiai S, Masuko H, Yamagami T, Saito Y, Izutsu N,
Kawasaki T, Namiki M, and Murata I, BSMILES–A Balloon-Borne Super-
conducting Sub-Millimeter-Wave Limb-Emission Sounder for statrospheric
measurements. IEEE Trans. on Geoscience and Remote Sensing;2006;3(1).

Kasai Y, Urban J, Takahashi C, Hoshino S, Takahashi K, Inatani J, Shiotani

M, and Masuko H, Stratospheric ozone isotope enrichment studied by sub-
millimeter wave heterodyne radiometry: the observation capabilities of
SMILES. IEEE Trans. on Geoscience and Remote Sensing;2006;3(1).

Kasai Y, Ochiai S, Super-Conductive Submillimeter-Wave Limb Emission

Sounder Onboard International Space Station: Algorithm Development of
for the Data Processing. Jounal of The National Institute of Information
and Communications Technology;2008;55(1),97–108

Kuntz M, A new implementation of the Humlicek algorithm for the calcula-

tion of the Voigt profile function. JQSRT;1997;57(6);819–824.

Levenberg K, A method for the solution of certain problems in least squares.

Q. J. Appl. Math.;1944;2;164–168.

Liebe HJ, MPM–an atmospheric millimeter-wave propagation model. Int. J.

Inf. Millim. Waves;1989;10(6):631–650.

Liebe HJ, Hufford G, and Cotton M, Propagation modeling of moist air and
suspended water/ice articles at frequencies below 1000 GHz. AGARD 52nd
Specialist Meeting of the Electromagnetic Wave Propagation Panel, Palma
De Mallorca, Spain, 1993.

68
Livesey NJ, Snyder WV, Read WG, Wagner PA, Retrieval algorithms for
the EOS Microwave limb sounder (MLS). IEEE-MTT, 44(5), 1144–1155,
2006. doi 10.1109/TGRS.2006.872327

Manabe T, Inatani J, Murk A, Wylde RJ, Seta M, and Martin DH,

A New Configuration of Polarization-Rotating Dual-Beam Interferom-
eter for Space Use. IEEE Transactions on Microwave Theory and
Techniques;2003;vol.51(6):1696–1704

Manabe T, Fukami T, Nishibori T, Mizukoshi K, and Ochiai S, Measurement

and evaluation of submillimeter-wave antenna quasioptical feed system by
a phase-retrieval method in the 640-GHz band. IEICE Trans. Commun.;
2008;E91-B(6):1760-1766

Marquardt DW, An algorithm for the least-squares estimation of nonlinear

parameters SIAM J. Appl. Math.;1963;11(2);431–441.

Melsheimer C, Verdes C, Bühler S, Emde C, Eriksson P, Ichizawa S,

John V, Kasai Y, Kopp G, Koulev N, Kuhn T, Lemke O, Ochiai S,
Schreier F, Sreerekha T, Takahashi C, Tsujimaru S, and Urban J. In-
tercomparison of general purpose clear-sky atmospheric radiative trans-
fer models for the millimeter and sub-millimeter spectral range. Radio
Sci.;2005;40;R51007,doi:10.1029/2004RS003110.

NASDA/CRL, JEM/SMILES mission plan, Version 2.1. [Online]. Available:

http://smiles.tksc.nasda.go.jp/document/indexe.html; 2002.

Ochiai S, Irimajiri Y, and Masuko H, 270 GHz SIS radiometer for strato-
spheric ClO observation. SPIE;2000;4152;372–379.

69
Ochiai S, Tsujimaru S, Irimajiri Y, and Manabe T, Stratospheric ozone and
clo measurement using balloon-borne submillimeter limb sounder. IEEE
Trans. on Geoscience and Remote Sensing;2004;8.

Oh J and Cohen EA, Pressure broadening of ClO by N2 and O2 near 204

and 649 GHz and new frequency measurements between 632 and 725 GHz.
JQSRT;1994;54;151–156.

Ozeki H, Kasai Y, Ochiai S, Tsujimaru S, Inatani J, Masuko H, Takahashi C,

Mazuray L, and Rosen C, Submillimeter-wave spectroscopic performance
of JEM/SMILES. SPIE;2000;4152;255–262.

Pickett HM, Poynter RL, and Cohen EA, Submillimeter, millimeter, and
microwave spectral line catalogue. Technical Report JPL Publication 80–
23, Rev.3, JPL, 1992.

Read WG, Shippony Z, Schwartz MJ, Livesey NJ, and Snyder WV, The
clear-sky unpolarized forward model for the EOS Aura Microwave Limb
Sounder (MLS). IEEE Trans. Geoscience and Remote sensing;2006;44(5).

Rodgers CD, Retrieval of atmospheric temperature and composition from

remote measurements of thermal radiation. Rev. Geophys. and Space
Phys.;1976;14(4);609–624.

Rodgers CD, Characterization and error analysis of profiles retrieved from

remote sounding measurements. JGR;1990;95(5);5587–5595.

Rodgers CD, Inverse methods for atmospheric sounding: theory and practice,
Series on atmospheric, oceanic and planetary physics, vol.2. Singapole,
World Scentific; 2000

70
Urban J, Baron P, Lautié N, Dassas K, Schneider N, Ricaud P, and La Noë J,
Moliere (v5): A versatile forward- and inversion model for the millimeter
and sub-millimeter wavelength range. JQSRT;2004;83(3–4);529–554.

Verdes C, Bühler SA, Engeln A, Kuhn T, Kuenzi K, Eriks-

son P, and Sinnhuber BM, Pointing and Temperature Re-
trieval from Millimeter/Sub-Millimeter Limb Soundings.
JGR;2002;107(D16);4299;doi:10.1029/2001JD000777.

Verdes CL, Bühler SA, Perrinb A, Flaudb JM, Demaisonc J, Wlodarczakc

G, Colmontc JM, Cazzolid G, and Puzzarinid C, A sensitivity study on
spectroscopic parameter accuracies for a mm/sub-mm limb sounder instru-
ment. J. Mol. Spectrosc;2005;229(2);266-・75.

Verdes C, Engeln A, and Bühler SA, Partition function data and impact on
retrieval quality for a mm/sub-mm limb sounder. JQSRT;2005;90(2);217-・
38.

Yamada M, Kobayashi M, Habara H, Amano T, and Drouin BJ,

Submillimeter-wave measurements of the pressure broadening of BrO.
JQSRT;2003;82(1–4);391–399.

Yamada M and Amano T, Pressure broadening measurements of

submillimeter-wave lines of O3. JQSRT;2005;95(2);221-230.

Waters JW, Read WG, Froidevaux L, Jarnot RF, Cofield RE, Flower DA,
Lau GK, Pickett HM, Santee ML, Wu DL, Boyles MA, Burke JR, Lay
RR, Loo MS, Livesey NJ, Lungu TA, Manney GL, Nakamura LL, Perun
VS, Ridenoure BP, Shippony Z, Siegel PH, Thurstans RP, Harwood RS,

71
 Pumphrey HC, and Filipiak MJ, The UARS and EOS Microwave Limb
Sounder (MLS) experiments. J. Atmos. Sci.;1999;56;194–218.

Wehr T, Bühler SA, Englen A, Kunzi K, and Langen J, Retrieval of spectro-

scopic temperatures from space borne microwave limb sounding measure-
ments. J. Geophys. Res.;1998;103;25,997–26,006.

Whaley RC, Petitet A, and Dongarra JJ, Automated empirical optimizations

of software and the ATLAS project. Parallel Computing;2000;27(1–2);3–
35.

72
Part III
Capability for Ozone retrieval
12 Introduction
In this Chapter, we refer to theoretical evaluations for the random error and
the systematic error of O3 vertical profiles based on the launch-ready retrieval
algorithm developed for the operational data processing system of SMILES
(Takahashi et al., 2009). Since the O3 emission line at 625.37 GHz is the most
intense emission line within the SMILES measurement bands and it affects
the retrieval of the other target species measured together, it is important to
evaluate the O3 retrieval so that other species can be retrieved accurately.
The altitude region where SMILES can detect O3 has been estimated
between 20 km and 60 km (NASDA and CRL, 2002) but the altitude re-
gions below 20 km and above 60 km have not been adequately assessed to
determine the quality of O3 retrievals possible in these regions. Therefore,
we will determine the measurable altitude region more strictly here. It is
difficult to retrieve the O3 profiles in the low- and high-altitude regions. In
the low-altitude region, the uncertainty of the continuum emission by water
vapour and scattering by clouds reduces the accuracy of the retrieved O3
profiles and in the high-altitude region the random error becomes worse with
decreasing abundance of O3 . It is important to determine the O3 distribu-
tion in these low- and high-altitude regions to improve understanding of the
ozone chemistry.
The launch-ready retrieval algorithm is based on the optimal estima-
tion method (OEM), which is discussed in Section 8. Section 13 provides

73
an overview of SMILES, including the SMILES measurement technique and
Section 14 provides O3 retrieval results. Section 15 refers to the retrieval
errors caused by model parameter uncertainties such as a priori profile and
uncertainty of the atmospheric temperature. In Section 16, we identify the
capability for measuring O3 diurnal variation in the upper stratosphere. Fi-
nally, we present a further improvement of the retrieval algorithm in the
lower stratosphere and its random error in Section 17.

13 O3 Measurements with the JEM/SMILES

SMILES measures the submillimeter-wave limb emission from the atmo-
spheric minor species, such as O3 , ClO, HCl, HNO3 , HOCl, CH3 CN, HO2 ,
BrO and O3 isotopes. The nominal altitude coverage is 10–60 km while the
altitude coverage of tangent altitude by the antenna scanning is typically be-
tween -10 km and 100 km. The most unique characteristics of the SMILES
measurements is their high-sensitivity to atmospheric limb emission in the
submillimeter-wave range thanks to the 4-K cooled SIS mixers. Overall sys-
tem noise temperature of SMILES was designed to be less than 700 K and
after the launch it was found to be less than 400 K (Kikuchi et al., 2010). It is
better than previous millimeter-submillimeter limb measurements from satel-
lites such as Aura/MLS (Read et al., 2006), whose system noise temperature
is approximately 10,000 K at a 600 GHz frequency range, and Odin/SMR
(Murtagh et al., 2002), whose system noise temperature is approximately
3,000 K. The relationship between the system noise temperature Tsys (ν) and

74
the noise level of the brightness temperature σ(ν, zT ) is

Tsys (ν) + TA (ν, zT )

σ(ν, zT ) = √ , (26)
Bτ
where TA (ν, zT ) is the measurement brightness temperature at a frequency ν
and a tangent altitude zT , B is the noise bandwidth (1.5 to 2.0 MHz), and τ
is the integration time (0.5 s). The noise level of the brightness temperature
of SMILES for a single scan is approximately 0.7 K.
SMILES measures three frequency bands: band A (624.32–625.52 GHz),
band B (625.12–626.32 GHz) and band C (649.12–650.32 GHz). Only two
of them are measured simultaneously. Figure 18 shows measured spectra in
the mid-latitudes atmosphere at approximately 10 (red line), 30 (green line)
and 50 km (blue line) tangent altitude for each band. The brightest emission
line in the SMILES measurement bands is the O3 emission line at 625.37
GHz in bands A and B. The spectra at 30 km tangent altitude indicate that
SMILES is a suitable instrument to measure the target species in the middle
stratosphere. The spectra at 10 km tangent altitude are almost saturated by
water vapor continuum whereas at 50 km tangent altitude emission lines of
the most of target species (except O3 and HCl) are less than the noise level
of SMILES (approximately 1 K). Therefore, we need further improvements
for the retrieval algorithm in these altitude regions. Detailed estimations for
O3 retrieval in these altitude regions are in Section 16 and 17.

75
 Band A
250
10 km O3
20 km
30 km
40 km
200

Brightness Temperature [K] 150

HCl

100

50 CH3CN O3v1v3
HNO3 HNO3
HOCl BrO 18
OOO O3v2
HOCl HNO3
0
624.2 624.4 624.6 624.8 625 625.2 625.4 625.6
Frequency [GHz]

Band B
250
10 km
20 km
30 km O3
40 km
200
Brightness Temperature [K]

HCl

150

100

50
18
O17OO
OOO
HO2 O17OO
0
625 625.2 625.4 625.6 625.8 626 626.2 626.4
Frequency [GHz]

Band C
250
10 km
20 km
30 km
40 km
200
Brightness Temperature [K]

150

100

50 BrO
18 ClO 17 17
OOO 17 OOO OOO
OOO HO2

0
649 649.2 649.4 649.6 649.8 650 650.2 650.4
Frequency [GHz]

Figure 18: Measured spectra by SMILES. There are three bands; band A
(top), B (middle) and C (bottom) at approximately 10 km (red/solid line),
20 km (green/dashed line), 30 km (blue/dotted line), and 40 km (pink/fine
dotted line). 76
14 Retrieval Results
14.1 Setup Conditions

The retrieval algorithm is based on the OEM applied for atmospheric sound-
ing (Rodgers, 1976, 1990, 2000). The details are in Introduction (Chapter
I). The species included in the simulation are O3 , HCl, ClO, HOCl, HNO3 ,
CH3 CN, HO2 , BrO and H2 O. Spectroscopic line parameters for those species
are taken from the JPL catalog (Pickett et al., 1992) and H2 O background
continua is calculated by Liebe’s model (Liebe, 1989). The retrieval pa-
rameters are the vertical profiles of the mixing ratio for these species, the
atmospheric temperature profile and a pointing offset. The profiles used in
this simulation are from the US standard atmosphere. Standard deviations
of the a priori profiles for the species, temperature and pointing offset are
100% of the a priori profiles, 5 K in the entire altitude range, and 5 km (as
estimated for the uncertainty of the ISS attitude (NASDA and CRL, 2002)),
respectively.
The measured tangent altitude region used in the retrieval is 0–80 km
and the altitude grid step is approximately 2 km width. The vertical grid
step is set to 3 km in the 4–70 km region for all retrieval parameters. Other
instrument parameters are listed in Table 5. Standard deviations of the mea-
surement brightness temperature (the square root of the diagonal elements
of Sy ) are defined by the equation 26.

14.2 Random Error and Vertical Resolution

We estimate the random error of the retrieved O3 profiles under clear sky
conditions by using single-scan spectral data of band A, because band A pro-

77
 Table 5: JEM/SMILES design specifications
System Parameter Description
Frequency bands Band A (624.3–625.5 GHz)
Band B (625.1–626.3 GHz)
Band C (649.1–650.3 GHz)
System noise temperature Less than 700 K
Instrumental vertical resolution 3.5–4.1 km (nominal)
Frequency resolution 1.8 MHz (FWHM)
Channel separation 0.8 MHz channel
Integration time 0.5 s for each observation point
Altitude range 10–60 km (nominal)
Global coverage 38◦ S–65◦ N (nominal)
Antenna scan period 53 s/scan
Horizontal sampling interval 360 km (in orbital direction)
Orbit period 90 min.

vides the highest sensitivity for the stratospheric O3 . The red line in Figure
19 is the random error of O3 , which is the square root of diagonal elements
of S, in the case of Tsys = 500 K (typical for SMILES observations). The
green line is the random error of O3 in the case of Tsys = 5000 K (comparable
with previous submillimeter or millimeter radiometers). In fact, according to
Livesey et al. (2007) the random error of O3 measured by the Aura/MLS is
estimated 2% at approximately 25 km and 5% in the altitude range from ap-
proximately 20 to 40 km. The best random error of O3 by using the SMILES
measurement is approximately 0.4% at 28 km altitude. The random error is
better than 10% in the altitude region from 13 to 70 km, 5% in the altitude
region from 15 to 55 km and 2% in the altitude region from 16 to 48 km.
The random error of O3 is approximately one order of magnitude better than
that from previous submillimeter or millimeter radiometers. Here, we only
discuss the random error, but systematic errors caused by the uncertainty

78
 70
500 K
5000 K
60

50
Altitude [km]

40

30

20

10

0
0.001 0.01 0.1 1
Error ratio [%]

Figure 19: The solid red line and dashed green line indicate the random error
for the system noise of 500 K and 5000 K, respectively.

of the model parameters are also important in the discussion of the product
quality. These errors will be estimated in Section 15.
Next, we refer to the vertical resolution of the retrieved O3 profile. The
averaging kernels of O3 for the bands A, B and C are shown in Figure 20.
The full width at half maximum (FWHM) of the averaging kernel for each
altitude indicates the vertical resolution of the retrieved O3 profiles. Since
the retrieval altitude grid step is 3 km in the retrieval, the minimum altitude
resolution is 3 km in this case. The vertical resolution of 3 km is achieved
from 10 to 70 km in the case of the bands A and B and 10 to 40 km in the
case of band C.

79
 $CPF# $CPF$
#NVKVWFG=MO?

#NVKVWFG=MO?

# #
$CPF%
#NVKVWFG=MO?

Figure 20: Averaging kernels of O3 for bands A (left), B (middle) and C

(right). A thick red line is an integration of the averaging kernel for each
altitude, which indicates amount of information at that altitude. Averaging
kernel altitudes and corresponding vertical resolutions are indicated on the
right side of each plot.

80
 In this calculation, we assume that the profiles correspond to the standard
mid-latitude atmosphere (see Figure 21). The thick red line, which is an area
of ai (ith row of A) and is calculated by aT
i u using a vector with unit elements

u, indicates the amount of information in the measurement (Rodgers, 2000).

Bands A and B contain more information for the O3 measurement than band
C in the upper stratosphere because the center of the line in these two bands
is located there. Band C contains more information in the low stratosphere
than bands A or B, because the tail of the O3 emission line is detected in
band C. To retrieve the O3 vertical profile at low stratosphere with high
precision, band C plays a significant role (Section 17).

15 Error Analysis
Other possible error sources for O3 retrieval using SMILES are:

• Pointing error,

• Bias error from the a priori profile,

• Uncertainty of the atmospheric temperature,

• Forward model parameter errors.

An error of pointing is one of the most significant issues for atmospheric

remote sensing measurements using limb-viewing instruments. We retrieve
the offset of the pointing to reduce the effect from the pointing error and
the details of pointing retrieval for SMILES are described in Takahashi et al.
(2009).

81
 [K]
0 50 100 150 200 250 300
100
Ozone
90 Temperature
H2O
80

70
Altitude [km]

60

50

40

30

20

10

0
1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01
VMR

Figure 21: The standard mid-latitudes atmosphere profiles have been used in
this study. The solid red line is the ozone profile, the dotted blue line is the
atmospheric temperature profile and the dashed green line is H2 O profile.

82
 The bias errors from the a priori assessment of the O3 profiles are es-
timated in Figure 22. It shows additional errors caused by the differences
between the true and the a priori O3 profiles in the case of the mid-latitudes
and the tropics. Here, the true profiles are the profiles that are used for
the calculations of the simulated measurement vector (y). Systematic errors
are derived by comparing these true profiles with profiles retrieved by the
software when a bias is applied to one of the input parameters (in this case
the O3 a priori profile). The additional error, σδxi , is defined by
)
σδxi = σi2 + (δxi )2 − σi . (27)

the σi is the square root of the ith diagonal component of S and δxi is ith
component of δx (δx = xtrue − xret ), where xtrue and xret are the true and
the retrieved profiles, respectively. The error ratio is represented by σδxi /xa .
Excluding the case of the –50% difference (i.e. xtrue is half of xa ), the
additional errors are less than 3% in the stratosphere for both mid-latitudes
and the tropics. Since we prepared a priori profiles of O3 for every 10–degree
latitude grids, every month and day and night, the influence of the a priori
profiles should be insignificant for the O3 retrieval.
Next, we show effects from the uncertainty of the atmospheric tempera-
ture. The standard approach in the previous millimeter and submillimeter
limb sounders (Waters et al., 1999; Read et al., 2006) is to derive the informa-
tion on the atmospheric temperature from the oxygen emission lines (Livesey
et al., 2006; Wehr et al., 1998). Since there is no oxygen line in the SMILES
measurement bands, the atmospheric temperature can be derived from the
O3 emission line at 625.37 GHz (NASDA and CRL, 2002; Verdes et al., 2002;
Baron et al., 2001). Figure 23 shows a comparison of the information con-

83
 Error at the mid-latitude
70 Random error
5%
60 -5%
10%
-10%
50 50%
-50%
Altitude[km]

40

30

20

10

0
0.001 0.01 0.1 1 10 100
Error [%]

Error at the tropics

70 Random error
5%
60 -5%
10%
-10%
50 50%
-50%
Altitude[km]

40

30

20

10

0
0.001 0.01 0.1 1 10 100
Error [%]

Figure 22: Estimations for the influence of priori profiles in O3 retrieval (in
this case, a priori profiles are same as initial profiles). The solid red line is
the random error of O3 . The other lines are additional errors between the
true profiles of O3 and the retrieved profiles of O3 that are the final results of
the iteration process in the cases where the differences between the a priori
profiles and true profiles are ±5%, ±10%, and ±50%. (top: mid-latitudes,
bottom: tropics). 84
 70
Ozone
Oxygen
60

50
Altitude [km]

40

30

20

10

0
9 10 11 12 13 14 15 16 17
Information Content [bits]

Figure 23: Information content of O3 at 625.37 GHz (red/solid line) and

oxygen at 118 GHz (green/dashed line).

tent (Rodgers, 1998) of the atmospheric temperature for the O3 emission

line at 625.37 GHz and for the oxygen emission line at 118 GHz under the
same conditions as SMILES. It indicates that the atmospheric temperature
can be retrieved by using the O3 emission line, although it is not as precise
as from the oxygen emission line especially in the upper stratosphere. The
atmospheric temperature and O3 vertical profiles are unknown quantities to
be retrieved. Both of them can be retrieved from the same measurement
because of the different frequency dependence of their weighting functions.
Figure 24 shows the effect of the atmospheric temperature uncertainty
upon retrieved O3 profiles. Each line is σδxi corresponding to the difference

85
between the true and the a priori profiles of atmospheric temperature, which
are 1 K (red line), 2 K (green line), 5 K (blue line), and 10 K (pink line),
respectively. If the difference between the true and a priori profiles is below 10
K, the additional error is below 1% in the stratosphere. Following this result,
we will use the reanalysis and reprocessing data produced by Global Modeling
and Assimilation Office (GMAO) of Goddard Space Flight Center (GSFC)
for atmospheric temperature. By applying GMAO data, the error caused
by the atmospheric temperature will be less than 1% in the stratosphere
(Rienecker et al., 2007).
The forward model parameter errors in the launch-ready algorithm have
been estimated in Takahashi et al. (2009). It shows that each forward model
parameter error from optimizing the radiative transfer calculation and design-
ing the accurate instrument functions (such as the instrumental field of view,
the sideband rejection ratio of sideband separator, and the spectral responses
of acousto-optic spectrometers) is below 0.01%. Errors associated with line
parameters such as air-broadening parameters and partition functions were
estimated by Verdes et al. (2005a). According to the paper, partition func-
tion errors do not introduce significant errors for most of the target species
of SMILES except BrO. It is known that there are still some forward model
parameter errors that have not been estimated such as spectroscopic parame-
ters and cirrus clouds. To estimate and improve the accuracy of the retrieval,
we have been estimating these errors using the real observation data.

86
 Errors of ozone retrieval
70 Random error
1K
60 2K
5K
10K
50
Altitude [km]

40

30

20

10

0
0.001 0.01 0.1 1 10 100
Error [%]

Figure 24: Estimation of the influence of uncertainty of initial and a priori

profiles of atmospheric temperature (in this case, we retrieve atmospheric
temperature and O3 profile simultaneously and the priori profiles of atmo-
spheric temperature are same as initial profiles). Lines are additional errors
between the true profiles of O3 and the retrieved profiles of O3 that are the
final results of the iteration process in the cases where the uncertainties of
atmospheric temperature are 1 K (red/solid line), 2 K (green/dashed line),
5 K (blue/dotted line) and 10 K (pink/fine-dotted line).

87
16 Diurnal Variation of O3 Abundance in the
Upper Stratosphere
Ozone has a strong diurnal variation in the upper stratosphere and the meso-
sphere (Allen et al., 1984). Since the ISS is not in a sun-synchronous orbit and
its orbital plane rotates every 60 days, SMILES can produce one day datasets
that can describe the diurnal variation using 60 days SMILES observation
data. In this section we estimate the capability of SMILES observations for
the diurnal variation of ozone by investigations of the ozone random error at
two extreme times, which are the midnight and the noon.
The solar occultation sensors (ACE, SAGEs, etc.) can measure in sunrise
and sunset conditions. The scattering instruments can measure the diurnal
variation only in the daytime but most of these instruments are on a sun-
synchronous orbit, so their observations are limited within fixed local time.
SMILES has an advantage in measuring diurnal variation but it has not been
previously discussed (NASDA and CRL, 2002). Here, we study the retrieval
capability of the diurnal variation with SMILES.
Figure 25 shows the random error from single-scan spectra of band A
in the daytime (red/solid line) and in the nighttime (green/dashed line).
Ozone a priori profiles in this estimation have been applied with the annually
averaged and 10◦ degree zonal mean (35–44◦ N) (Froidevaux et al., 2008;
Livesey et al., 2007) of the Aura/MLS version 2.2 data for day and night,
respectively. At 70 km altitude ozone can be measured with 5% random error
in the nighttime and, because of a low abundance, with 20% random error
in the daytime.
SMILES is a powerful instrument to investigate O3 chemistry in the upper

88
Figure 25: Diurnal variability of the O3 measurements. The left side of the
figure shows the ozone profiles and the right figure shows the random error
in the daytime (red/solid line) and in the nighttime (green/dashed line),
respectively.

89
stratosphere and the mesosphere. To improve the random error, we apply
the averaging algorithm (Livesey et al., 2006) in the high-altitude region.
After averaging ten spectra, the random error at 70 km altitude is improved
to 5% in the daytime and is 1% in the nighttime.

17 Retrieval Capability in the Lower Strato-

sphere and Upper Troposphere
In our early estimation (NASDA and CRL, 2002), it was stated that SMILES
can measure O3 profiles above 20 km (i.e. SMILES cannot measure the lower
stratosphere). The results of the early estimation were based on the retrieval
using a single band independently. In this section we study the retrieval capa-
bility of ozone below 20 km using two bands combinations because SMILES
mounts two Acousto-Optical Spectrometer and measures the two bands si-
multaneously. Also we estimate effects of cirrus clouds.

17.1 Two-Bands Simultaneous Retrieval

SMILES measures two bands simultaneously as mentioned in Section 13. We

consider a possibility of improving the random error for O3 by the simultane-
ous retrieval of two bands (Figure 26) in the cases of the mid-latitudes (top)
and the tropics (bottom). Since SMILES has poor coverage in the polar
region, we focus on the mid-latitudes and the tropics. We use the profiles
of the standard 45◦ N and 0◦ N atmosphere for mid-latitude and the tropics
estimation, respectively. The lines on the left of Figure 26 are the profiles of
O3 (red/solid line), temperature (green/dashed line), and H2 O (blue/dotted
line).

90
Figure 26: Results of synthetic retrievals using different bands. The left side
of the figure shows the profiles used in each simulation and the right side of
the figure shows the random error in the case of the mid-latitudes (top) and
the tropics (bottom). The lines on the left side of the figure are the profiles of
O3 (red/solid line), temperature (green/dashed line), and H2 O (blue/dotted
line). The lines on the right side of the figure are the random error of O3
using band A only (red/solid line), using bands A and B (green/dashed line),
and using bands A and C (blue/dotted line).
91
 Tangent height: 4km 8km 12km

15
∆Tb (K)

10

0
0 0.5 1 1.5 2 2.5 3
IWP (g/m 2)

Figure 27: Relationship between the ice water path (IWP) and the brightness
temperature depression due to cirrus clouds. The tangent altitudes are 4 km
(solid line), 8 km (dashed line), and 12 km (dotted line) and the cloud layer
is assumed to be uniformly between 12 km and 13 km.

The right side of Figure 26 shows the random errors of O3 by using (1)
band A only (red line), (2) both bands A and B (green line), and (3) both
bands A and C (blue line). By using bands A and C simultaneously, the
random error is less than 5% above 13 km in the mid-latitudes and above
15 km in the tropics. In contrast, the random error at 13 km in the mid-
latitudes is approximately 50% using only band A. The random error at 15
km in the tropics is approximately 40% using only band A and 30% using
both bands A and B. It indicates that the simultaneous retrieval of bands A
and C improves the random error especially in the lower stratosphere under
the clear sky conditions.

92
17.2 Effects from Cirrus Clouds

The O3 profile cannot be retrieved if there are thick clouds such as cumu-
lonimbi in the lines of sight of SMILES, but we have not yet discussed the
possibility of the O3 retrieval under the thin clouds like cirrus clouds exis-
tence. Here, we discuss the possibility of retrieval of the O3 profile under the
condition of cirrus clouds existence. It is important to estimate the cirrus
clouds because they (including invisible thin clouds) exist globally with high
possibility below the lower stratosphere.
Figure 27 shows the relationship between the ice water path (IWP) and
the brightness temperature depression because of cirrus clouds estimated
with the atmospheric radiative transfer simulator (Bühler et al., 2005). A
uniform cloud layer between 12 and 13 km is assumed. The brightness tem-
perature depressions are 1 K in the case of IWPs = 0.5 g/m2 and 3 K in the
case of IWPs = 1.0 g/m2 .
We estimate the error, δx, due to the cirrus clouds whose IWPs are equal
to 0.5 g/m2 and 1.0 g/m2 in the mid-latitudes and the tropics (See Figure
28). The error ratio in Figure 28 is defined δx/xa . δx increases rapidly
below 13 km in the case of IWPs = 0.5 g/m2 and below 16 km in the case of
IWPs = 1.0 g/m2 in the mid-latitudes. It means that the O3 profile in the
mid-latitudes cannot be retrieved below 13 km (i.e. around the tropopause
altitude) if there are clouds with IWP greater than 0.5 g/m2 (uniformly
distributed between 12 km and 13 km).
In the tropics, the clouds whose IWP is less than 1.0 g/m2 have little effect
on the O3 retrieval. The retrieval errors increase less than approximately 10%
above 16 km. SMILES has a capability to measure O3 distributions above

93
 30
0.5 g/m22 (1 K)
1.0 g/m (3 K)
random error (A+C)
random error (A)
25
Altitude [km]

20

15

10
0 2 4 6 8 10
Error ratio [%]
30
0.5 g/m22 (1 K)
1.0 g/m (3 K)
random error (A+C)
random error (A)
25
Altitude [km]

20

15

10
0 2 4 6 8 10
Error ratio [%]

Figure 28: Effects from the cirrus clouds whose IWP are 1.0 (red/solid line)
and 0.5 (green/dashed line) g/m2 with the random error of the bands A and
C (blue/dotted line) and A only (pink/fine-dotted line) in the case of the
mid-latitudes (top) and the tropics (bottom).

94
16 km that can be retrieved with an error below 10% in the tropics, even in
a presence of thin clouds between 12 and 13 km.

18 Conclusion
We have estimated O3 random error with its altitude range by the SMILES
observations. In the early estimations in NASDA and CRL (2002), it was
estimated that the random error of O3 vertical profiles was better than 5%
from 20 to 45 km and 10% from 18 to 53 km but these errors includes the
random error and some systematic errors together. In this study, we reveal
and classify the random error and other systematic errors. Furthermore
we estimated the retrieval capability of the diurnal variation and the lower
stratosphere and upper troposphere for the first time in this study.
The best random error is found to be about 0.4% at 28 km and better
than 5% between 15 and 70 km in the nighttime and between 15 and 55 km
in the daytime. This high precision is mainly due to the low system-noise
temperature of SMILES, which is less than 400 K. In rough estimation, the
random error of SMILES is also one order of magnitude better than similar
previous sensors. The systematic errors are estimated in section 15. We
mentioned the systematic errors for the bias error of the a priori profile and
uncertainty of the atmospheric temperature. The first one is almost same as
the random error at 28 km (the minimum point of the random error), and
the second one is 0.2% at 28km. Therefore, the total error, which is the sum
of the random error and the systematic errors, at 28 km is approximately
1%.
Above 55 km O3 profiles can be retrieved with the random error better

95
than 5% in the nighttime. In the daytime, the random error gets worse
because the O3 abundance in this altitude region decreases. Since the high-
altitude region is important for investigations of the O3 diurnal variation
in the upper stratosphere, we will adopt the averaging algorithm of MLS
(Livesey et al., 2006), which is unsusceptible to the bias of the a priori profile
to improve the random error. After averaging ten spectra, the random error
can be improved from 20% (single scan) to 5% (averaging ten times). We
are planning to use the averaging scheme after sufficient validation of the
single-scan retrieval.
Furthermore, using simultaneous retrievals of bands A and C, random
errors above 15 km in the tropics and above 13 km in the mid-latitudes are
expected to be about 5% under clear sky conditions. The SMILES observa-
tions are expected to provide information on O3 vertical distribution from the
lower to upper stratosphere. In previous studies the effect of clouds below 20
km has been found to be non-negligible, however, no quantitative tests were
performed. Here, we have found that in a presence of thin clouds between
12 and 13 km (with IWP less than 1.0 g/m2 ) O3 can be retrieved only above
16 km for both mid-latitudes and tropics.
There are some important issues to resolve in the low-altitude region
such as uncertainty of H2 O abundance and the continuum model. In the
SMILES measurement bands, there is no H2 O line but there is the continuum
component. In the launch-ready algorithm, the continuum components are
fitted by the MPM-89 model (Liebe, 1989) and the components that can not
be removed are fitted by a slope function. It works reasonably well but we
continue to research strategies to improve data quality.

96
 Finally, we mention some important issues to resolve, such as pointing
error and spectroscopic parameters errors. In our algorithm, the offset of
the pointing is retrieved (the details are in (Takahashi et al., 2009)) and
this algorithm works properly in real data processing. For the spectroscopic
parameters, detailed study is now performed with real data.

97
References
Allen M, Lunine JI, Yung YL, The vertical distribution of ozone in the meso-
sphere and lower thermosphere. J. Geophys. Res;1984;89;48414872.

Baron P, Merino F, Murtagh D, Simultaneous retrievals of temperature and

volume mixing ratio constituents from non-oxygen Odin submillimeter
bands. Opt;2001;40(33);61026110.

Bernath PF, McElroy CT, Abrams MC, Boone CD, Butler M, Camy-Peyret
C, Carleer M, Clerbaux C, Coheur PF, Colin R, DeCola P, DeMazie M,
Drummond JR, Dufour D, Evans WF, Fast H, Fussen D, Gilbert K, Jen-
nings DE, Llewellyn EJ, Lowe RP, Mahieu E, McConnell JC, McHugh M,
McLeod SD, Michaud R, Midwinter C, Nassar R, Nichitiu F, Nowlan C,
Rinsland CP, Rochon YJ, Rowlands N, Semeniuk K, Simon P, Skelton
R, Sloan JJ, Soucy MA, Strong K, Tremblay P, Turnbull D, Walker K,
Walkty I, Wardle DA, Wehrle V, Zander R, Zou J Atmospheric Chemistry
Experiment (ACE):Mission overview. Geophys. Res. Lett., 32, L15S01,
doi:10.1029/2005GL022386.

Bovensman H, Burrows JP, Buchwitz M, Frerick J, Noël S, Rozanov VV,

Chance KV and Goede APH, SCIAMACHY: Mission objectives and mea-
surement modes. J. Atmos.Sci;1999;56(2);127-150

Bevilacqua RM, Hoppel K, Hornstein J, Lucke R, Shettle E, Ainsworth T,

Debrestian D, Fromm M, Lumpe J, Krigman S, Glaccum W, Olivero J,
Clancy RT, Rusch D, Randall C, Dalaudier F, Deniel C, Chassefiere E,

98
 Brogniez C, Lenoble J, First results from POAM II: The Dissipation of
the 1993 Antarctic Ozone Hole Geophys. Res. Lett.;1995;21;909-912.

Bühler SA, Eriksson P, Kuhn T, Engeln A, and Verdes C. ARTS, The atmo-
spheric radiative transfer simulator. JQSRT;2005;91(1);65-93.

European Space Agency: Envisat, MIPAS An instrument for atmospheric

chemistry and climate research, ESA Publications Division, ESTEC,
P.O.Box 299,2200 AG NOOrdwijk, The Netherlands, SP-1229

Fischer H and Oelhaf H, Remote sensing of vertical profiles of atmo-

spheric trace constituents with MIPAS limb-emission spectrometers. Appl.
Opt;1996;35(16);2787796

Froidevaux L, Jiang YB, Lambert A, Livesey NJ, Read WG, Waters JW,
Browell EV, Hair JW, Avery MA, McGee TJ, Twigg LW, Sumnicht GK,
Jucks KW, Margitan JJ, Sen B, Stachnik RA, Toon GC, Bernath PF,
Boone CD, Walker KA, Filipiak MJ, Harwood RS, Fuller RA, Manney
GL, Schwartz MJ, Daffer WH, Drouin BJ, Cofield RE, Cuddy DT, Jarnot
RF, Knosp BW, Perun VS, Snyder WV, Stek PC, Thurstans RP, Wag-
ner PA, Validation of Aura Microwave Limb Sounder stratospheric ozone
measurements. JGR;2008;113;D15S20, doi:10.1029/2007JD008771.

Harris N, Hudson R, and Phillips C, Assessment of Trends in the

Vertical Distribution of Ozone. SPARC Rep;1;289;World Meteorol.
Org.;Geneva;Switzerland;1998

Kikuchi K,Nishibori T,Ochiai S,Ozeki H, Irimajiri Y,Kasai Y, Koike

M,Manabe T, Mizukoshi K, Murayama Y,Nagahama T,Sano T,

99
 Sato Ryota,Seta M,Takahashi C. Takayanagi M, Masuko H,Inatani
J,Suzuki M, and Shiotani M, Overview and Early Results of the Su-
perconducting Submillimeter-Wave Limb-Emission Sounder (SMILES).
JGR;2010;15;D23306;12.

Levenberg K, A method for the solution of certain problems in least squares

Quart.Appl.Math;1944;2;164-168.

Liebe HJ, MPM-an atmospheric millimeter-wave propagation model. Int J

Inf Millim Waves;1989;10(6):631-50.

Livesey NJ, Snyder WV, Read WG, Wagner PA, Retrieval algorithms for
the EOS Microwave limb sounder (MLS). IEEE-MTT, 44(5), 1144–1155,
2006. doi 10.1109/TGRS.2006.872327

Livesey NJ,Read GW, Lambert A, Cofield ER, Cuddy TD, Froidevaux L,

Fuller AR, Jarnot FR, Jiang HJ, Jiang BY, Knosp WB, Kovalenko JL,
Pickett MH, Pumphrey CH, Santee LM, Schwartz MJ, Stek CP, Wagner
PA, Waters WJ, and Wu LD, Version 2.2 Level 2 data quality and descrip-
tion document JPL D-33509;May 22,2007.

Marquardt DW, An algorithm for the least-squares estimation of nonlinear

parameters SIAM J.Appl.Math;1963;11;431-441.

Murtagh D, Frisk U, Merino F, Ridal M, Jonsson A, Stegman J, et al., An

overview of the previous termOdinnext term atmospheric mission. Can J
Phys;2002;80;30919.

NASDA/CRL, JEM/SMILES mission plan, Version 2.1.[Online]. Available:

http://smiles.tksc.nasda.go.jp/document/indexe.html; 2002

100
Ochiai S, Kikuchi K, Nishibori T, Manabe T, Ozeki H, Mizukoshi K, Ohtsubo
F, Tsubosaka K, Irimajiri Y, Sato R, and Shiotani M, Performance of
JEM/SMILES in orbit. 21th Int’l Symp. Space Terahertz Technol., S8.1,
(Oxford, UK), March 2010.

Pickett HM, Poynter RL, Cohen EA, Submillimeter, millimeter, and mi-
crowave spectral line catalogue. Technical Report JPL Publication 80-
23,Rev.3,JPL, 1992.

Read WG, Shippony Z, Schwartz MJ, Livesey NJ, and Snyder WV, The
Clear-sky Unpolarized Forward Model for the EOS Aura Microwave Limb
Sounder (MLS). IEEE Trans. Geosci. and remote sens;2006;44(5).

Rienecker MM, Suarez MJ, Todling R, Bacmeister J, Takacs L, Liu HC, Gu

W, Sienkiewicz M, Koster RD, Gelaro R, Stajner I, and Nielsen JE, The
GEOS–5 data assimilation system:Documentation of versions 5.0.1, 5.1.0,
and 5.2.0, Tech. Rep. Ser. on Global Modeling and Data Assimilation.
NASA/TM;2007;104;606;27

Rodgers CD, Retrieval of atmospheric temperature and composition from

remote measurements of thermal radiation. Rev Geophys and Space
Phys;1976;14(4);609-624.

Rodgers CD, Characterization and error analysis of profiles retrieved from

remote sounding measurements. JGR;1990;95(5);5587-5595.

Rodgers CD, Information content and optimisation of hight spectral resolu-

tion remote measurement. ASR;1998;21;361–367.

101
Rodgers CD, Inverse methods for atmospheric sounding: theory and practice,
Series on atmospheric, oceanic and planetary physics, vol.2. Singapole,
World Scentific; 2000

Rusch DW, Bevilacqua RM, Randall CE, Lumpe JD, Hoppel KW, Fromm
MD, Debrestian DJ, Olivero JJ, Hornstein JS, Guo F, Shettle EP, Vali-
dation of POAM II Ozone Measurements with Coincident MLS, HALOE,
and SAGE II Observations. JGR;1997;102(23);615–627

Russell JM, Gordley LL, Park JH, Drayson SR, Hesketh DH, Cicerone RJ,
Tuck AF, Frederick JE, Harries JE, and Crutzen P, The Halogen Occulta-
tion Experiment. JGR; 1993;98;D6;10;777–797.

Takahashi C, Ochiai S, Suzuki M, Operational Retrieval Algorithms

for JEM/SMILES Level2 Data Processing System, JQSRT; 2009;
doi:10.1016/j.jqsrt.2009.06.005

Verdes C, Bühler SA, Engeln A, Kuhn T, Kuenzi K, Eriksson

P, and Sinnhuber BM, Pointing and Temperature Retrieval from
Millimeter/Sub-Millimeter Limb Soundings. JGR; 2002; 107(D16);
4299;doi:10.1029/2001JD000777.

Verdes C, Bühler SA, Perrinb A, Flaudb JM, Demaisonc J, Wlodarczakc

G, Colmontc JM, Cazzolid G, and Puzzarinid C, A sensitivity study on
spectroscopic parameter accuracies for a mm/sub-mm limb sounder instru-
ment. J. Mol. Spectrosc;2005a;229(2);266-75.

Verdes C, Engeln A, and Bühler SA, Partition function data and impact on

102
 retrieval quality for a mm/sub-mm limb sounder. JQSRT;2005b;90(2);217-
38.

Waters JW, Read WG,Froidevaux L, Jarnot RF, Cofield RE, Flower DA,
Lau GK, Pickett HM, Santee ML, Wu DL, Boyles MA, Burke J. R, Lay R.
R, Loo M. S, Livesey N. J, Lungu TA, Manney GL, Nakamura LL, Perun
VS, Ridenoure BP, Shippony Z, Siegel PH, Thurstans RP, Harwood RS,
Pumphrey HC, and Filipiak MJ. The UARS and EOS Microwave Limb
Sounder (MLS) experiments. J.Atmos.Sci.;1999;56;194-218.

Wehr T, Bühler SA, Englen A, Kunzi K, and Langen J, Retrieval of Spec-

troscopic Temperatures from Space Borne Mcrowave Limb Sounding Mea-
surements, J Geophys. Res., 103, 25997-26006, 1998.

WMO (World Meteorological Organization): Scientific assessment of ozone

depletion: 2006, WMO Global Ozone Res. and Monit. Project, Rep. 50,
World Meteorol. Org., Geneva, 2007.

Wu DL, Austin RT, Deng M, Durden SL, Heymsfield AJ, Jiang

JH, Lambert A, Li JL, Livesey NJ, McFarquhar GM, Pittman JV,
Stephens GL, Tanelli S, Vane DG, and Waliser DE, Comparisons
of global cloud ice from MLS, CloudSat, and correlative data sets.
JGR;2009;114;doi:10.1029/2008JD009946

103
Overall Summary
The purpose of this work is to develop the launch ready retrieval algorithm
with the highly accurate forward model and to clarify the capability of ozone
measurement by SMILES.

SMILES have been launched on 11 September 2009 by H-II Transporta-

tion Vehicle (HTV) and have begun atmospheric observations on 12 October.
The SMILES had achieved the system temperature was less than 400 K. The
system temperature was realized better than specific value of 700 K. The mea-
sured data by SMILES were retrieved and profiles of the target species were
produced as soon as the data had downloaded in the ground data processing
system without problems. The measured SMILES data have been able to be
retrieved by using the launch ready retrieval algorithm shown in Chapter II
without any significant problems. The processing speed achieved as we ex-
pected. Appendix shows the results of ozone retrieval by the measured data
using this algorithm. It shows that the SMILES data are consistent with
the MLS data. For the other standard products, the retrieval processing was
almost successful. It provided one of the evidence for the effectiveness of
this retrieval algorithm. Furthermore, it was confirmed that the ability of
processing was also sufficient.

In Chapter III, O3 random error with its altitude range by the SMILES
observations and systematic errors are estimated. The retrieval capability of
the diurnal variation and the lower stratosphere and upper troposphere are

104
also estimated for the first time in this study. It provide that the best random
error is found to be about 0.4% at 28 km and better than 5% between 15
and 70 km in the nighttime and between 15 and 55 km in the daytime.These
error is roughly one order of magnitude better than similar previous sensors.
The total error, which is the sum of the random error and the systematic
errors, at 28 km is approximately 1% . In the case of the upper stratosphere
in the daytime, the random error can be improved from 20% (single scan)
to 5% (averaging ten times) by averaging ten spectra. Meanwhile, in the
lower stratosphere using simultaneous retrievals of bands A and C, random
errors above 15 km in the tropics and above 13 km in the mid-latitudes are
expected to be about 5% under clear sky conditions. Here, we have found
that in a presence of thin clouds between 12 and 13 km (with IWP less than
1.0 g/m2 ) O3 can be retrieved only above 16 km for both mid-latitudes and
tropics. The SMILES observations are expected to provide information on
O3 vertical distribution from the lower to upper stratosphere.

SMILES stopped the observation because of the hardware problem in

April 2010 unfortunately and it was working about half a year. The study for
the retrieval algorithm is still continued and it is improved by using received
data of half a year. There are some important issues to resolve using real
data to improve data quality. For example, the uncertainty of H2 O is one of
the big issue. In the launch-ready algorithm, the continuum components are
fitted by the MPM-89 model (Liebe, 1989). It works reasonably well but we
continue to research strategies. Others are pointing error and spectroscopic
parameters errors. In our algorithm, the offset of the pointing is retrieved

105
and this algorithm works properly in real data processing. Some of the
spectroscopic parameters in the launch-ready algorithm are used measured
values in laboratories, but not all of them. It should be estimated with real
data whether the spectroscopic parameters should be measured.

106
Acknowledgments
I would like to express my deep gratitude to Professor Masato Shiotani of
the Research Institute for Sustainable Humanosphere (RISH), Kyoto Uni-
versity, for his enormous supports and insightful comments of this study. I
am deeply grateful to Dr. Makoto Suzuki of the Institute of Space and As-
tronautical Science / JAPAN Aerospace Exploration Agency (ISAS/JAXA),
whose opinions and information have helped me very much throughout the
production of this study.

Special thanks to Professor Akira Mizuno and Associate Professor Tomoo

Nagahama of the Solar-Terrestrial Environment Laboratory (STEL), Nagoya
University, who gave me invaluable comments and suggestions to complete
this thesis.

I would like to express my appreciation Dr. Masahiro Takayanagi, Mr.

Takuki Sano, Dr. Koji Imai , Dr. Naohiro Manago of ISAS/JAXA and Dr.
Chihiro Mitsuda and Mr. Hirotomo Taniguchi of Fujitsu FIP corporation,
and Associate Professor Hiroo Hayashi of RISH, Kyoto University, who are
the SMILES team members in ISAS/JAXA and study for the SMILES level
2 data processing and its algorithm together. They gave me valuable cooper-
ations to my study and invaluable discussion that make my research of great
achievement.

I also would like to thank the SMILES instrument team, Dr. Satoshi

107
Ochiai and Dr. Ken-ichi Kikuchi of National Institute of Information and
Communications Technology (NICT), Dr. Toshiyuki Nishibori of JAXA,
Professor Hiroyuki Ozeki of Toho University, Professor Takeshi Manabe of
Osaka Prefecture University and Professor Shinji Inatani of National Astro-
nomical Obsevatory of JAPAN, for their useful discussions and suggestions
about the instrument characteristics. Finally, I am grateful to my family for
their support and encouragement.

108
Appendix: Ozone retrieval by
the measured data
SMILES have been launched on 11 September 2009 by H-II Transportation
Vehicle (HTV) and it have begun atmospheric observations on 12 October
2009. SMILES have quit the atmospheric observation on 21 April 2010 be-
cause of the problem of the sub-millimeter local oscillator. SMILES have
measured and stored about half years data of atmospheric observations,
whose system noise temperature is better than 400 K (it better than the
specification value of 700 K). In this appendix, some recent results for the
ozone retrieval by using the SMILES measured data are shown.

The single scan spectra of the measured brightness temperature for band
A, B, and C are shown in Figure 29 (Kikuchi et al., 2010). Since the system
noise temperature is lower than specification value, weak emission lines like
BrO and HO2 can be seen in the single scan. Next, a sample of residual er-
rors of the measured brightness temperature are shown in Figure 30 to verify
the the retrieval algorithm. The residual errors of the measured brightness
temperature is useful to check the validity of the retrieval processing. The
residual errors in Figure 30 are almost less than 1 K for all tangent altitudes.
The results show the good agreements between the measurement and the re-
trieval processing in most regions except line center. The retrieval processing
is working as we expected but there are still remained the errors in the line
shapes.

109
Figure 29: The measured spectra on 12 October 2009 at several tangent
altitudes with log scale (Kikuchi et al., 2010). (a) is Band A, (b) is Band B,
and (c) is Band C.
110
Figure 30: Brightness temperature and residual errors of the measured
brightness temperature measured on 12 October 2009 at 20 km, 30 km, 40
km and 50 km altitudes.

111
 The capability for the measurements of ozone diurnal variation by SMILES
was discussed in Chapter III. Ozone mixing ratios processed by SMILES mea-
sured data are plotted with local times (see Figure 31). Figure 31 is ozone
mixing ratio at 61 km altitude whose altitude must be appeared the diurnal
variation (Garcia and Solomon, 1985). Ozone mixing ratios are decreasing in
daytime and the ozone diurnal variations are clearly appeared in the SMILES
measured data.

The first study for comparisons between SMILES and other sensors like
MLS will be shown discussed in Kikuchi et al. (2010). Figure 32 (Kikuchi et
al., 2010) shows the latitude-height sections of the zonal mean ozone on 12
October 2009 of SMILES (left) and AURA/MLS (right). This figure shows a
good agreement with SMILES and MLS in the global scale dimensions. The
comparison of the profiles of ozone concentration are in Figure 33 (Suzuki et
al., 2010). It shows statistics estimation of comparison between SMILES and
AURA/MLS for ozone retrieval profile using 284 coincidence data at the 55◦
N - 65◦ N latitude region. The difference between SMILES and AURA/MLS
are less than 5% from 20 km to 45 km.

Finally, the status of the other target species is mentioned. Almost other
target species are retrieved successful soon after receiving the measured data
as same as ozone. Figure 33 (Suzuki et al., 2010) shows Statistics estimation
of comparison between SMILES and AURA/MLS for HCl retrieval profile
using coincidence data in the 55◦ N - 65◦ N latitude region. The difference
between SMILES and AURA/MLS are less than 10% from 20 km to 45

112
Figure 31: the measurements of ozone diurnal variation by SMILES. Green
and blue points with error bars show ozone mixing ratios (ppb) at 61 km
altitude of ascending and descending, respectively.

113
Figure 32: Latitude-height sections of the zonal mean ozone on 12 October
2009. Left side is SMILES and right side is AURA/MLS. In this case, data
for daytime and nighttime are averaged (Kikuchi et al., 2010) .

114
Figure 33: Statistics estimation of comparison SMILES and AURA/MLS
for ozone retrieval profile, using coincident SMILES and AURA/MLS ozone
profiles at the 55◦ N - 65◦ N latitude. Left: the mean profiles for SMILES
(blue) and AURA/MLS (red), Middle: the differences between the SMILES
and AURA/MLS profiles in mixing ratio, Right: the differences in percentage
(Suzuki et al., 2010).

115
Figure 34: Statistics estimation of comparison SMILES and AURA/MLS
for HCl retrieval profile using coincident SMILES and AURA/MLS ozone
profiles in the 55◦ N - 65◦ N latitude region. Left: the mean profiles for
SMILES (blue) and AURA/MLS (red), Middle: the differences between the
SMILES and AURA/MLS profiles in mixing ratio, Right: the differences in
percentage (Suzuki et al., 2010).

km. The recent papers of Shiotani et al. (2011) and Suzuki et al. (2010)
shows the results of many more target species such as HNO3 (see Figure 35).
Some species especially whose spectra are mixable with baseline are difficult
to retrieve. Further estimation for the baseline components such as water
vapor needed for these species.

116
Figure 35: Example of HNO3 coincidence compared with ENVISAT/MIPAS
(MIPAS-IMK ver.40) at 67.1◦ N, 101.5◦ on Oct. 12, 2009. Two SMILES
profiles (blue) are compared with 1 MIPAS profiles (red) (Suzuki et al., 2010).

117
References
Garcia, R and Solomon, S, The effect of breaking gravity waves. JGR;1985;90,
3850-3852.

Kikuchi K, Nishibori T,Ochiai S,Ozeki H, Irimajiri Y,Kasai Y, Koike

M,Manabe T, Mizukoshi K, Murayama Y,Nagahama T,Sano T,
Sato Ryota,Seta M,Takahashi C. Takayanagi M, Masuko H,Inatani
J,Suzuki M, and Shiotani M, Overview and Early Results of the Su-
perconducting Submillimeter-Wave Limb-Emission Sounder (SMILES).
JGR;2010;15;D23306;12.

Shiotani M, Takayanagi M, Suzuki M, Sano T, SMILES Mission Team, Re-

cent results from the Superconducting submillimeter-wave limb-emission
sounder (SMILES) onboard ISS/JEM, SPIE;2011

Suzuki M, Mitsuda C, Takahashi C, Iwata T, Manago N, Sano T, Kikuchi

K, Ochiai S, Imai K, Nishimoto E, Naito Y, Hayashi H, and Sh-
iotani M, EARLY RESULTS FROM 4K-COOLED SUPERCONDUCT-
ING SUBMM WAVE LIMB EMISSION SOUNDER SMILES ONBOARD
ISS/JEM, International Archives of the Photogrammetry, Remote Sensing
and Spatial Information Science, Volume XXXVIII, Part 8, Kyoto Japan
2010

118