Effects of atmospheric light scattering on spectroscopic observations of greenhouse gases from space: Validation of PPDF-based CO2 retrievals from GOSAT

Effects of atmospheric light scattering on spectroscopic

observations of greenhouse gases from space:

Validation of PPDF-based CO

retrievals

from GOSAT

Sergey Oshchepkov,

Andrey Bril,

Tatsuya Yokota,

Isamu Morino,

Yukio Yoshida,

Tsuneo Matsunaga,

Dmitry Belikov,

Debra Wunch,

Paul Wennberg,

Geoffrey Toon,

Christopher O

’

Dell,

André Butz,

Sandrine Guerlet,

Austin Cogan,

Hartmut Boesch,

Nawo Eguchi,

Nicholas Deutscher,

9,10

David Griffith,

Ronald Macatangay,

Justus Notholt,

Ralf Sussmann,

Markus Rettinger,

Vanessa Sherlock,

John Robinson,

Esko Kyrö,

Pauli Heikkinen,

Dietrich G. Feist,

Tomoo Nagahama,

Nikolay Kadygrov,

Shamil Maksyutov,

Osamu Uchino,

and Hiroshi Watanabe

Received 21 January 2012; revised 2 April 2012; accepted 13 May 2012; published 23 June 2012.

[

]

This report describes a validation study of Greenhouse gases Observing Satellite

(GOSAT) data processing using ground-based measurements of the Total Carbon Column

Observing Network (TCCON) as reference data for column-averaged dry air mole fractions

of atmospheric carbon dioxide (X

CO2

). We applied the photon path length probability

density function method to validate X

CO2

retrievals from GOSAT data obtained during

22 months starting from June 2009. This method permitted direct evaluation of optical path

modifications due to atmospheric light scattering that would have a negligible impact on

ground-based TCCON measurements but could significantly affect gas retrievals when

observing reflected sunlight from space. Our results reveal effects of optical path

lengthening over Northern Hemispheric stations, essentially from May

–

September of each

year, and of optical path shortening for sun-glint observations in tropical regions. These

effects are supported by seasonal trends in aerosol optical depth derived from an offline

three-dimensional aerosol transport model and by cirrus optical depth derived from

space-based measurements of the Cloud-Aerosol Lidar with Orthogonal Polarization

(CALIOP) instrument. Removal of observations that were highly contaminated by aerosol

and cloud from the GOSAT data set resulted in acceptable agreement in the seasonal

variability of X

CO2

over each station as compared with TCCON measurements. Statistical

comparisons between GOSAT and TCCON coincident measurements of CO

column

abundance show a correlation coefficient of 0.85, standard deviation of 1.80 ppm,

and a sub-ppm negative bias of

0.43 ppm for all TCCON stations. Global distributions

of monthly mean retrieved X

CO2

with a spatial resolution of 2.5

latitude

2.5

longitude

show agreement within

2.5 ppm with those predicted by the atmospheric tracer transport

model.

National Institute for Environmental Studies, Tsukuba, Japan.

Division of Geological and Planetary Sciences, California Institute of

Technology, Pasadena, California, USA.

Jet Propulsion Laboratory, California Institute of Technology,

Pasadena, California, USA.

Department of Atmospheric Science, Colorado State University, Fort

Collins, Colorado, USA.

IMK-ASF, Karlsruhe Institute of Technology, Karlsruhe, Germany.

Netherlands Institute for Space Research, Utrecht, Netherlands.

Earth Observation Science, Space Research Centre, University of

Leicester, Leicester, UK.

Research Institute for Applied Mechanics, Kyushu University,

Kyushu, Japan.

School of Chemistry and Centre for Atmospheric Chemistry,

University of Wollongong, Wollongong, New South Wales, Australia.

Institute of Environmental Physics, University of Bremen, Bremen,

Germany.

IMK-IFU, Karlsruhe Institute of Technology, Garmisch-

Partenkirchen, Germany.

National Institute of Water and Atmospheric Research, Wellington,

New Zealand.

FMI-Arctic Research Center, Sodankylä, Finland.

Max Planck Institute for Biogeochemistry, Jena, Germany.

Solar-Terrestrial Environment Laboratory, Nagoya University,

Nagoya, Japan.

Laboratoire des Sciences du Climat et de l

’

Environnement, CEA,

С

NRS, UVSQ, IPSL, Gif-sur-Yvette, France.

Corresponding author: S. Oshchepkov, National Institute for

Environmental Studies, 16-2 Onogawa, Tsukuba, Ibaraki 305-8506,

Japan. (sergey.oshchepkov@nies.go.jp)

0148-0227/12/2012JD017505

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 117, D12305,

doi:10.1029/2012JD017505, 2012

D12305

1of18

Citation:

Oshchepkov, S., et al. (2012), Effects of atmospheric light scattering on spectroscopic observations of greenhouse

gases from space: Validation of PPDF-based CO

retrievals from GOSAT,

J. Geophys. Res.

117

, D12305,

doi:10.1029/2012JD017505.

1. Introduction

[

] Atmospheric carbon dioxide (CO

) is acknowledged as

being the dominant anthropogenic greenhouse gas and an

important factor in global climate change. Over the past

century, the CO

concentration has significantly increased

from about 280 ppm to 380 ppm (

mol/mol) due to the

burning of fossil fuels as well as changes in land use and

forestry associated with expanding human activities

[

Intergovernmental Panel on Climate Change

, 2007].

Recent advances in carbon cycle science have heightened

the need for accurate space-based global observations of

for use in quantifying surface fluxes of greenhouse

gases [

Chevallier et al.

, 2007].

[

] The Greenhouse Gases Observing Satellite (GOSAT)

was launched on 23 January 2009 to monitor the global

distributions of greenhouse gases (CO

and CH

) from

space. The satellite has a Sun-synchronous orbit at an alti-

tude of 666 km and a 3-day recurrence with the descending

node around 13:00 local time. Over the 3-day orbital repeat

cycle, GOSAT measures several tens of thousands of single

soundings that cover the globe. The most useful observa-

tions for retrieving gas concentrations are limited to areas

under clear-sky conditions; only about 10% of the total

daytime satellite soundings are available for gas retrievals

throughout the entire atmosphere. At the same time, the

number of remaining data points collected by GOSAT far

surpasses the current number of ground monitoring stations

(approximately 200) registered in the World Data Centre for

Greenhouse Gases (WDCGG) [

World Data Centre for

Greenhouse Gases

, 2011]. GOSAT thus helps to fill gaps

in the ground-based observation network and has shown

promise for improved CO

surface flux inverse modeling

[

Chevallier et al.

, 2009], even though the precision of indi-

vidual soundings is less than that of observations by ground

monitoring stations. Generally, a precision of

1.5 ppm or

better for monthly mean column-averaged dry air mole

fraction of CO

CO2

) on a regional scale with sub-ppm

biases is required to improve estimates of surface CO

fluxes

from those based on in situ measurements [e.g.,

Rayner and

’

Brien

, 2001;

Chevallier et al.

, 2009;

Kadygrov et al.

2009].

[

] Although space-based observations of greenhouse

gases are usually processed under clear-sky conditions,

atmospheric light scattering, such as from high-altitude

subvisible cirrus or aerosols, is always present and can

introduce large biases in retrieved gas amounts [

’

Brien and

Rayner

, 2002;

Dufour and Bréon

, 2003;

Mao and Kawa

2004;

Aben et al.

, 2007;

Oshchepkov et al.

, 2008, 2009,

2011;

Reuter et al.

, 2010]. The primary source of these

biases is the uncertainty in the modification of the light path

through the atmosphere [

Oshchepkov et al.

, 2011].

[

] Several algorithms have been developed for proces-

sing spectroscopic observations of greenhouse gases from

space. Most of the algorithms that have been developed

include the numerical solution of the radiative transfer

equation when modeling measured radiance spectra [

Bösch

et al.

, 2006;

Connor et al.

, 2008;

Butz et al.

, 2009;

Reuter

et al.

, 2010;

’

Dell et al.

, 2012;

Yoshida et al.

, 2011].

These are often referred to as full physics algorithms. The

effects of atmospheric light scattering are taken into account

using modeled vertical profiles of aerosol and cloud optical

depth. Whether the vertical profiles of a set aerosol and cloud

optical characteristics are assumed [

Yoshida et al.

, 2011] or

retrieved [

’

Dell et al.

, 2012] simultaneously with gas

amounts is a basic distinguishing feature of different versions

of full physics algorithms. It is important to note that aerosol

and cloud optical characteristics are very smooth spectral

functions within the gas absorption bands and only optical

path modification is essential for radiative transfer spectral

calculations. Any other impacts of atmospheric light scat-

tering on measured radiance spectra, such as those related to

near-ground aerosols, can be taken into account by spectral

polynomials of low orders when applying differential optical

absorption spectroscopy (DOAS) [

Buchwitz et al.

, 2000;

Frankenberg et al.

, 2005].

[

] Another physical background to consider in regard to

atmospheric light scattering is light path modification, which

can be examined using the photon path length statistical

characteristics and the equivalence theorem [

Bennartz and

Preusker

, 2006]. This formalism has been used to develop

the photon path length probability density function (PPDF)

method [

Bril et al.

,2007;

Oshchepkov et al.

, 2008, 2009] to

rapidly process spectroscopic observations of greenhouse

gases from space. The PPDF-based method is constructed to

allow reduction to DOAS when neglecting light path mod-

ifications. It can also account for any a priori data on aerosol

and cloud optical characteristics available in full physics

algorithms [

Oshchepkov et al.

,2009].

[

] These algorithms are currently under development using

the experience gained from processing of actual GOSAT data.

It is important to evaluate algorithms for satellite data pro-

cessing, such as by validating their CO

retrievals using

ground-based high-resolution Fourier Transform Spectrome-

ter (FTS) observations from the Total Carbon Column

Observing Network (TCCON) [

Wunch et al.

,2011a].

Morino

et al.

[2011] compared data products retrieved operationally

with the version 01.XX algorithm of the National Institute for

Environmental Studies (NIES) with reference TCCON data,

and found a large X

CO2

negative bias of

9ppm.

Butz et al.

[2011] validated about 1 year of GOSAT data processing

with the

“

RemoTeC

”

algorithm developed at the Netherlands

Institute for Space Research (SRON) and the Karlsruhe Insti-

tute of Technology (KIT). They implemented the empirical

correction of positive bias in surface pressure retrieval that

permitted correction of the X

CO2

negative bias.

Wunch et al.

[2011b] evaluated version B2.8 and partially evaluated ver-

sion B2.9 of the algorithm that is operationally used to process

the GOSAT radiance spectra in the National Aeronautics and

Space Administration (NASA) Atmospheric CO

Observa-

tions from Space (ACOS) project. In their study, the global

negative bias seen in version B2.8 was adjusted according to

the discrepancy between TCCON and space-based retrievals

OSHCHEPKOV ET AL.: VALIDATION OF PPDF-BASED CO

RETRIEVALS

D12305

2of18

of X

CO2

over Southern Hemisphere stations where the X

CO2

gradients were assumed to be small [

Wunch et al.

, 2011b].

The version B2.9 results had an empirical correction applied

to the oxygen A-band spectroscopy, similar to the

Butz et al.

[2011] correction, and required no further adjustments using

the TCCON data.

[

] This paper validates PPDF-based GOSAT data pro-

cessing by comparing the retrievals of column-averaged dry

air mole fraction of CO

seasonally with FTS ground-based

measurements at each TCCON station. We also compare

global distributions of the retrievals by atmospheric transport

modeling and analyze seasonal and spatial variability in light

path modification over each TCCON site. We consider this

latter issue to be important for the validation because atmo-

spheric light scattering can influence satellite data, whereas

it has little influence on ground-based FTS measurements

that record direct solar light transmitted through the atmo-

sphere. In most cases, light path modification offers a

physical explanation of the discrepancy between satellite-

and ground-based CO

retrievals.

[

] The paper is organized as follows. Section 2 briefly

outlines the basic specifications of the GOSAT instrument

and the PPDF-based CO

retrieval method. In section 3, we

describe the reference data sources for the CO

column-

averaged dry air mole fractions used to validate the retrievals

(TCCON and atmospheric transport model). Section 4

defines coincidence criteria for GOSAT and TCCON data

selection. Section 5 presents results of the validation study in

terms of CO

seasonal variability over each TCCON site

(section 5.1), a pairwise TCCON

–

GOSAT X

CO2

statistical

comparison (section 5.2), and validation of the retrieved

CO2

global distribution (section 5.3) using atmospheric

transport modeling. A summary and concluding remarks are

given in section 6.

2. GOSAT Instruments and Data Processing

[

] In this section, we briefly outline the basic specifica-

tions of the GOSAT instruments related to this study (e.g.,

spectral bands, observation modes, cloud screening) and

focus on the GOSAT data processing by the PPDF-based

method that retrieves both CO

column abundance (X

CO2

)

and optical path modification.

2.1. GOSAT Specifications

[

] The GOSAT project has been promoted jointly by the

Japan Aerospace Exploration Agency (JAXA); NIES, Japan;

and the Ministry of the Environment (MOE), Japan

[

Hamazaki et al.

, 2005;

Yokota et al.

, 2009;

Kuze et al.

2009]. GOSAT was successfully launched on 23 January

2009 to monitor global CO

and CH

amounts. The satellite

is in a Sun-synchronous orbit with an equator crossing time

of about 13:00 local time and an inclination angle of 98

GOSAT flies at an altitude of approximately 666 km, com-

pletes an orbit in about 100 min, and operates on a global

basis with a 3-day repeat cycle.

[

] GOSAT is equipped with two instruments, the Ther-

mal And Near-infrared Sensor for carbon Observation

–

Fourier Transform Spectrometer (TANSO-FTS) and the

TANSO Cloud and Aerosol Imager (TANSO-CAI), which

have been described in detail by

Kuze et al.

[2009].

[

] TANSO-CAI is an ancillary imager that observes the

state of the atmosphere and the surface during daytime. The

image data from CAI are used to determine the existence of

cloud over an extended area that includes the FTS

’

field of

view as described by

Ishida and Nakajima

[2009]. Radiance

in the visible and near-infrared bands (380, 674, 870, and

1600 nm) would provide additional basic information on

aerosol and cloud properties, but these retrieval algorithms

remain under development. Currently, TANSO-CAI data are

used for cloud prescreening (section 4) and for the cloud

spatial coherence test [

Yoshida et al.

, 2011].

[

] TANSO-FTS is the key instrument for observing CO

and CH

amounts in the atmosphere. It observes solar light

reflected from the Earth

’

s surface in the Short Wavelength

Infra-Red (SWIR) region and thermal radiation from the

Earth

’

s surface and atmosphere in the Thermal Infra-Red

(TIR) region. TANSO-FTS measures raw interferograms

that are then converted to radiance spectra (Level 1B; L1B

data). This instrument has three narrow bands in the SWIR

region (0.76, 1.6, and 2.0

m as the center; also referred to

as TANSO-FTS bands 1, 2, and 3, respectively) and one

wide band in the TIR region (5.56

–

14.3

m; TANSO-FTS

band 4) at a high spectral resolution (interval) of about

0.2 cm

. The full width at half-maximum of the instrument

line shape function is about 0.27 cm

, which clearly iden-

tifies the rovibrational absorption lines of CO

and CH

the observed spectra [

Kuze et al.

, 2009;

Yoshida et al.

2011]. TANSO-FTS band 4 data were not used in this study.

[

] For the SWIR bands, the incident light is divided by a

polarization beam splitter. It is then simultaneously recorded

as two orthogonal polarization components, designated

“

”

and

“

”

whose polarization axes are aligned roughly along

and perpendicular to the direction of spacecraft motion,

respectively. The TANSO-FTS collects interferograms within

an instantaneous, circular field of view with a 15.8-mrad

diameter, yielding footprints that are approximately 10.5 km

in diameter for nadir observations.

[

] The TANSO-FTS has a pointing mechanism that

allows it to conduct off-nadir observations within pointing

mirror driving angles of

in the cross-track direction

and

in the along-track direction. Over ocean, the

TANSO-FTS also observes the glint spot, the point of

specular reflection at the water surface. The TANSO-FTS

pointing mechanism includes several point cross-track scan

modes, a sun-glint tracking mode, and a target mode.

Between 4 April 2009 and 31 July 2010 the 5-point cross-

track scan mode was utilized for routine science observa-

tions. This mode has been changed to the three-point cross-

track scan mode for all subsequent routine science observa-

tions from 1 August 2010. It yields footprints separated by

263 km cross-track and

283 km along-track [

Watanabe

et al.

, 2010]. In this mode, each footprint is sampled three

times.

2.2. PPDF-Based Method

[

] In this section, we apply the version of the PPDF-

based method that incorporates PPDF retrievals as a pre-

screening step to identify GOSAT soundings that are not

distinctly affected by atmospheric light scattering. These

clear sky soundings are recognized by low values of PPDF

parameters from band 1 (the threshold is defined below)

when the optical path modification is negligible in bands 2

OSHCHEPKOV ET AL.: VALIDATION OF PPDF-BASED CO

RETRIEVALS

D12305

3of18

and 3 [

Oshchepkov et al.

, 2011]. Then, we retrieve gas

amounts from radiance spectra in bands 2 and 3 with PPDF

parameters equal to zero, representing negligible optical

path modification in these bands. This is shown to be

equivalent to applying DOAS [

Oshchepkov et al.

, 2008].

We refer to this version as the NIES PPDF-D. Figure 1

presents a simplified flowchart of the data processing.

[

] Details of the PPDF-based method have been pub-

lished elsewhere [

Bril et al.

, 2007;

Oshchepkov et al.

, 2008,

2009, 2011]. Below, we outline the key features of the

PPDF-D inversion scheme and radiative transfer modeling.

[

] The GOSAT radiance spectra

∗

(

: wave number,

L1B data [

Kuze et al.

, 2009]) were processed with respect to

retrievals of both PPDF parameters and gas amounts

according to the maximum a posteriori rule:

^

arg min

∗

′

ðÞ

:

[

] In equation (1),

∗

is the measurement vector

(ln

∗

′

(

) is the vector of the radiative transfer forward

model (

′

~

ðÞ

) at a set of spectral

channels

is a spectral polynomial designed to

remove the low-frequency portion of the processed spectra

within each GOSAT band in the retrieval process (e.g.,

portions associated with near-ground aerosols or surface

albedo), solar irradiance

and the effective transmittance

̃

ðÞ

spectra are convolved with the GOSAT instrumental

line-shape function within each channel

is a retrieval

state vector with a priori

, and the squared norms

‖

are weighted by covariance matrices of

measurement errors of

∗

(

) and of a priori assumptions

(

). PPDF parameters are spectrally identical (wave-

length independent) over each GOSAT spectral band [

Bril

et al.

, 2007;

Oshchepkov et al.

, 2008]. The effective trans-

mittance and Jacobians are analytically expressed through

either gas profiles or PPDF parameters allowing for rapid

radiative transfer spectral calculations in the data processing.

[

] Under negligible light path modification, this method

reduces to DOAS-based retrievals with a simplified expres-

sion for the effective transmittance at an altitude

equation (1):

~

ðÞ

½¼

exp

cos

;

where

ðÞ

is the optical depth of gaseous

atmosphere (in the vertical direction);

and

are the solar

zenith angle and ray incident angle to the satellite, respec-

tively;

(

) is the vertical profile of the gas absorption

coefficient; and

is the top altitude of the absorbing

atmosphere [

Oshchepkov et al.

, 2008]. Atmospheric light

scattering could modify the optical path

cos

through modification of the path length depending on aero-

sol and cloud optical properties, such as the aerosol and

cloud optical depth, single scattering albedo, and scattering

phase function within each individual atmospheric layer

[

] Two key PPDF parameters,

and

, are mainly

responsible for the optical path modifications in the PPDF

Figure 1.

Simplified flowchart showing the basic steps for the PPDF-D level-2 data processing. PPDF

parameters are retrieved in the oxygen A-band 1 (0.76

m). GOSAT soundings with large light path mod-

ification are filtered out from further processing. The remaining GOSAT scans are used in the retrieval of

amounts at gas bands 2 (1.61

m) and 3 (2.0

m) with zero PPDF parameters.

OSHCHEPKOV ET AL.: VALIDATION OF PPDF-BASED CO

RETRIEVALS

D12305

4of18

model. They are retrieved in the oxygen A-band (12,950

–

13,190 cm

) and have the following physical interpretations:

[

] 1. Parameter

is the relative cloud/aerosol layer

reflectivity, i.e., the ratio of photons scattered by the cloud/

aerosol to the total number of photons within the field of

view of the satellite detector (it is clear by definition that

increases as the surface albedo decreases);

[

] 2. Parameter

is the scaled first moment of the PPDF

under the cloud or within the aerosol layer, i.e., this param-

eter describes the path length modification due to multiple

scattering/reflection between the ground surface and cloud/

aerosol particles (

increases with increasing surface albedo

because of the higher probability that contributing photons

will survive after multiple interactions with the surface).

[

] The admissible level of these parameters in the oxygen

A-band within which the optical path modification due to

aerosol and clouds is negligible in the gas absorption bands

was set empirically at 0.04. These values are roughly appro-

priate for Rayleigh light scattering in the oxygen A-band of

absorption at 0.76

Oshchepkov et al.

, 2011].

[

] The retrieved state vector in the CO

absorption bands

at 1.61

m(6368

–

6192 cm

) and at 2.0

m(4815

–

4885 cm

) included a vertical profile of the CO

mixing ratio

^

ðÞ

distributed over pre-defined pressure levels, and a

scaling factor for water vapor the

prior

profile. The retrieved

profile was converted to X

CO2

at a given pressure incre-

ment using the following expression:

^

dry

^

ðÞ

where

dry

is the dry air number concentration.

[

] The stretch factor was included in the desired state

vector in all bands for adjusting the position of the wave

number grids to the observational spectra. Temperature and

surface pressure were predefined and derived from Grid Point

Value (GPV) data provided by the Japan Meteorological

Agency (JMA). Constraints on the retrieved CO

profiles

were unified: an altitude-independent 385-ppm a priori pro-

file with a 30-ppm a priori standard deviation for all pressure

levels and GOSAT soundings. For the assumed diagonal

matrix

these constraints correspond to

–

8 ppm uncer-

tainty in X

CO2

depending on the surface pressure.

[

] We apply a unified retrieval scheme to processing the

GOSAT observations over land and for both regular and sun-

glint observation modes over ocean using the

“

”

polariza-

tion state (section 2.1). A post-processing and instrumental

quality assessment eliminated retrieved results for which the

discrepancy between the forward model and the observed

spectrum was significant (chi-square

> 5, where

defined within the curly braces of equation (1)), signal-to-

noise ratio (SNR) <75 for each spectral band, and degrees of

freedom for the signal (DFS)

≤

Rodgers

, 2000] to ensure

the low impact of a priori assumptions on a posteriori X

CO2

estimations.

3. Ground-Based and Modeled X

CO2

Data

[

] This section briefly describes the sources of reference

data for validating GOSAT X

CO2

retrievals. We focused on

validation using TCCON ground-based FTS measurements.

A NIES global atmospheric tracer transport model is used as

an ancillary for three reasons: for selection of GOSAT obser-

vations when applying the confidence criteria (section 4), to

fill gaps in seasonal variations of X

CO2

when TCCON data

were not available (section 5.1), and as a part of the validation

of X

CO2

retrievals on a global scale (section 5.3).

3.1. TCCON FTS Data

[

] The Total Carbon Column Observing Network was

established in 2004 with a primary focus on measuring pre-

cise and accurate columns of CO

,CH

, and other atmo-

spheric constituents. It is a ground-based network of high-

resolution FTSs recording direct solar absorption spectra in

the near-infrared spectral region. The scientific goals of the

network are to improve our understanding of the carbon

cycle, to provide a transfer standard between satellite mea-

surements and ground-based in situ measurements, and to

compile a primary validation data set for retrievals of X

CO2

and X

CH4

from space-based instruments, such as the Scan-

ning Imaging Absorption Spectrometer for Atmospheric

Cartography (SCIAMACHY), Tropospheric Emission

Spectrometer (TES), Atmospheric Infrared Sounder (AIRS),

Orbiting Carbon Observatory-2 (OCO-2) instrument, and

TANSO-FTS [

Wunch et al.

, 2010, 2011a;

Morino et al.

2011].

[

] TCCON data processing uses the GFIT nonlinear least

squares spectral fitting algorithm developed at NASA

’

sJet

Propulsion Laboratory [

Toon

,1992;

Wunch et al.

, 2011a]. The

retrieved TCCON X

CO2

data are corrected for air mass-

dependent artifacts [

Wunch et al.

, 2011a;

Deutscher et al.

2010]. Aircraft profiles measured over most of these stations

were used to determine an empirical scaling factor to place the

TCCON data on the World Meteorological Organization

(WMO) standard reference scale. On average, the uncertainty

in X

CO2

from the ground-based FTS measurement was esti-

mated to be 0.8 ppm by comparing the TCCON retrievals with

aircraft measurements [

Wunch et al.

, 2010;

Deutscher et al.

2010;

Messerschmidt et al.

, 2010]. Below, we use the uncer-

tainties from individual TCCON and GOSAT soundings as

weights when averaging X

CO2

data according to TCCON and

GOSAT coincidence criteria (section 4).

[

] Figure 2 shows the locations of the TCCON stations

from which the data analyzed in this study were obtained

(yellow stars). Although the Moshiri site (red cross) is not

included in the TCCON, the FTS at this station is operating

under the TCCON observational protocol (IFS120HR). The

spatial coordinates of these stations are listed Table 1.

3.2. NIES Transport Model

[

] Simulated X

CO2

values for GOSAT and TCCON

observations were prepared with the NIES global atmo-

spheric tracer transport model (NIES TM). NIES TM is an

offline model driven by the Japanese 25-year Reanalysis

(JRA-25) and JMA Climate Data Assimilation System

(JCDAS) data covering more than 30 years from 1 January

1979 [

Onogi et al.

, 2007]. The present version (NIES-08.1i)

of NIES TM uses a flexible hybrid sigma-isentropic (

)

vertical coordinate, which combines both terrain-following

and isentropic levels switched smoothly near the tropopause.

Vertical transport in the stratosphere is controlled by the cli-

matological heating rate derived from JRA-25/JCDAS

OSHCHEPKOV ET AL.: VALIDATION OF PPDF-BASED CO

RETRIEVALS

D12305

5of18

reanalysis, which was adjusted to fit to the observed mean

age of the air in the stratosphere.

[

] The model uses a flux-form advection algorithm with

a second-order van Leer scheme and deep convection fol-

lowing

Tiedtke

[1989], with penetrative updraft mass fluxes

adjusted to the JRA-25/JCDAS convective precipitation rate,

a turbulent diffusion calculation as described by

Maksyutov

et al.

[2008], and a bulk boundary layer with a 3-hourly

planetary boundary layer height taken from the ERA-Interim

reanalysis. Further details, except for the recently added

hybrid sigma-isentropic coordinate option and implementa-

tion of JCDAS meteorological data, have been described

elsewhere [

Belikov et al.

, 2011, 2012].

[

] NIES TM has been evaluated against GLOBALVIEW-

and WDCGG observations [

Belikov et al.

, 2011], through

transport model intercomparison studies by

Niwa et al.

[2011]

based on CO

data from the Comprehensive Observation

Network for Trace Gases by Airliner (CONTRAIL) aircraft

Figure 2.

Global locations of 12 ground-based FTS stations (yellow stars and red cross). Stars show 11

operational TCCON sites whose ground-based FTS data were used in this study. The site of future ground-

based FTS measurements at Ascension Island (red star) was used to demonstrate the large impact of opti-

cal path modifications on GOSAT data processing due to atmospheric light scattering over the ocean.

Although the Moshiri site (red cross) does not belong to TCCON, the FTS at this station is operating under

the TCCON observational protocol.

Table 1.

The Number of Available GOSAT Single Scans (

), Percentage of Remaining Scans After PPDF Screening (

), Number of

GOSAT Scans Coincident With TCCON Data (

), Number of Average Points (

) Meeting the Coincidence Criteria, and Characteristics

of Statistical Relationships (5)

–

(8) Between Ground-Based FTS and GOSAT X

CO2

for Each TCCON Site

TCCON Site: Global Location

(%)

Bias (ppm)

(ppm)

Sodankyla: 67.4

N, 26.6

386

253

1.34

0.13

1.69

2.29

0.90

0.84

Bialystok: 53.2

N, 23.1

1179

652

1.01

0.05

0.07

1.25

0.95

0.91

Bremen: 53.1

N, 8.85

824

237

0.89

0.09

1.87

2.74

0.77

0.62

Orleans: 48.0

N, 2.11

1183

432

0.83

0.07

0.23

1.62

0.88

0.79

Garmisch: 47.5

N, 11.1

1502

1140

0.89

0.05

0.43

1.41

0.90

0.84

Park Falls: 45.9

N, 90.3

2777

2289

1.19

0.05

0.34

1.66

0.94

0.91

Moshiri: 44.4

N, 142.3

773

226

1.29

0.17

0.38

2.84

0.82

0.96

Lamont: 36.6

N, 97.5

5299

5271

1.01

0.05

0.56

1.20

0.92

0.90

Tsukuba: 36.0

N, 140.2

906

279

1.21

0.12

0.04

1.50

0.91

0.97

All northern TCCON sites

14829

1079

412

1.02

0.02

0.47

1.65

0.90

0.86

Ascension Is.: 7.9

S, 14.3

2603

Darwin: 12.4

S, 130.9

2114

871

1.76

1.36

Wollongong: 34.4

S, 150.9

2579

1717

0.86

1.44

Lauder: 45.0

S, 169.7

103

1.66

2.71

All southern TCCON sites

4796

2646

130

0.38

2.05

All TCCON sites

19625

1345

542

0.98

0.02

0.43

1.80

0.85

0.78

Both data sets were averaged weekly within 15

-latitude

-longitude grid boxes centered at each TCCON station.

OSHCHEPKOV ET AL.: VALIDATION OF PPDF-BASED CO

RETRIEVALS

D12305

6of18

measurements, and by

Patra et al.

[2011] using CH

and SF

observations. NIES TM is able to reproduce the observed

seasonal patterns of CO

in near-surface layers and in the free

troposphere. Implementation of the hybrid sigma-isentropic

vertical coordinate allows for simulation of seasonal mean

vertical profiles and of the vertical propagation of seasonal

variations in tracers in the troposphere and lower stratosphere

that agree well with aircraft and balloon observations. In

general, the performance of NIES TM is consistent with

models participating in the TransCom intercomparison,

although small biases in simulated interhemispherical gra-

dients of SF

[

Patra et al.

, 2011] and CO

[

Niwa et al.

,2011]

were found.

[

] The model was run at a horizontal resolution of

2.5

and 32 vertical levels from the surface up to the

level of 3 hPa. The CO

simulation for the period of 2000

–

2011 (we used 2000

–

2001 as a spin-up period) utilized the

initial CO

global distribution derived from

GLOBALVIEW-

[2010] observations [

Belikov et al.

, 2012]. We use pre-

scribed fluxes from the CONTRAIL model intercomparison

[

Niwa et al.

,2011],asfollows:

[

] 1. Interannually varying anthropogenic emissions are

derived from the Emission Database for Global Atmospheric

Research (EDGAR-1998) with annual mean distribution

[

Olivier and Berdowski

, 2001]. The emission totals are

scaled using the growth rate of the top 20 country-specific

fossil fuel consumptions from the Carbon Dioxide Informa-

tion Analysis Center [

Boden et al.

, 2009];

[

] 2. Distribution of all natural (non-fossil) sources/sinks

over land and ocean is represented by the inversion flux,

derived by inverse modeling with 12 TransCom3 models

[

Gurney et al.

, 2004] and observational data from GLOBAL-

VIEW-CO

at 87 sites in the period of 1999

–

2001 [

Miyazaki

et al.

, 2008].

[

] The 2007 fluxes were used from 2008 onwards.

4. GOSAT, TCCON, and NIES TM

Data Selection

[

] We analyzed data from 22 months of GOSAT oper-

ation from June 2009 to March 2011. The analyzed obser-

vations fell into the category of cloud-free scenarios detected

by TANSO-CAI [

Kuze et al.

, 2009] when applying the cloud

flag test developed by

Ishida and Nakajima

[2009]. After

eliminating cloudy observations, we had approximately

700,000 GOSAT single scans over both land and ocean. As

Nakajima et al.

[2008] noted, TANSO-CAI often fails to

detect optically thin cirrus clouds because it has no thermal

infrared channel that is sensitive to the presence of clouds in

the upper troposphere. On the basis of actual data proces-

sing,

Yoshida et al.

[2011] also found that the TANSO-CAI

cloud flag test tended to categorize subpixel-sized clouds as

clear pixels over the ocean. Thus, the GOSAT observations

in this data set may contain contamination by light scattering

from thin cirrus clouds. We consider the effects of atmo-

spheric light scattering by examining the retrieved optical

path modification in the GOSAT validation study. The sea-

sonal variability of PPDF parameters is supported with data

from the Cloud Aerosol Lidar with Orthogonal Polarization

(CALIOP) instrument onboard the Cloud-Aerosol Lidar and

Infrared Pathfinder Satellite Observations (CALIPSO) plat-

form [

Winker et al.

, 2007].

[

] The number of ground-based FTS and GOSAT

coincident observations within spatially small pixels around

TCCON stations is rather limited [

Morino et al.

, 2011]. Our

TCCON

–

GOSAT coincidence criteria for validating the

PPDF-based retrievals included weekly mean GOSAT data

within 15

-latitude

-longitude grid boxes centered at

each TCCON station. For these grid boxes, we excluded

GOSAT observations for which the global atmospheric

tracer transport model (mod X

CO2

) showed >1 ppm differ-

ence from modeled X

CO2

at the TCCON site (mod X

CO2

TCCON

modX

CO2

modX

TCCON

CO2

≤

ppm

:

[

] We chose this value to limit spatial and temporal

variability in X

CO2

to within the a posteriori standard devi-

ation of X

CO2

derived by PPDF-D retrievals (

0.8

–

2 ppm).

A similar approach to extending the sample size was made

Wunch et al.

[2011b] with a larger grid box size (30

latitude

longitude) and longer period of 10 days,

which included at least three GOSAT repeat cycles. They

used potential temperature constraints (

2 K) at 700 hPa to

minimize the variability in X

CO2

, which is dynamic in origin

[

Keppel-Aleks et al.

, 2011], when defining coincidence cri-

teria in the Northern Hemisphere and used TCCON mea-

surements in the Southern Hemisphere when defining the

global X

CO2

bias. We limited the ground-based FTS data

1 h of the GOSAT overpass time. The same temporal

limitation was used for sampling the NIES TM simulated

concentration. The modeled X

CO2

data were taken from the

spatial grid cell (with horizontal a resolution of 2.5

2.5

)

containing the TCCON site. In this study, we use initial data

when comparing X

CO2

from GOSAT, TCCON, and NIES

TM without considering the averaging kernel effect.

5. Validation of PPDF-Based Retrievals

5.1. Seasonal Variations

5.1.1. Northern Hemisphere Stations

[

] Figures 3 (left) and 4 (top) display 22-month seasonal

variations in X

CO2

retrievals (blue symbols) in comparison

with those derived by ground-based FTS measurements

(green symbols). Six TCCON stations in the Northern

Hemisphere (from north to south: Sodankyla, Bialystok,

Garmisch, Park Falls, Lamont, and Tsukuba) were chosen

(Figure 3, top to bottom, and Figure 4, top). As noted in

section 4, all X

CO2

data were weekly means. For temporal

correspondence, the ground-based FTS data were limited

1 h of the GOSAT overpass time. The GOSAT mean

values

(

) and their variance

)

were computed for noncorrelated noise of

single scans

within the coincidence criterion, where

and

are

dimensional vector-columns of unity and X

CO2

, respec-

tively, and

is the

diagonal covariance matrix of

TCCON data were processed in the same way. Bars in

Figure 3 display the standard deviations (

) for both

GOSAT and TCCON X

CO2

. In Figures 3 and 4, we also plot

the X

CO2

time series according to NIES TM (red crosses) for

the GOSAT soundings that were available for PPDF-based

data processing after the quality assessment (section 2.2).

Observations over both land and ocean were included in the

comparison if the GOSAT scans satisfied the coincidence

criteria in equation (4).

OSHCHEPKOV ET AL.: VALIDATION OF PPDF-BASED CO

RETRIEVALS

D12305

7of18

[

] For each TCCON station, Figures 3 (right) and 4

(middle) show time series of PPDF parameters (

blue and

orange symbols) and water vapor column abundance

H2O

, green histograms). The X

H2O

data were calculated

from a priori water vapor profiles provided by the JMA.

[

] Most retrieved values of

were within the threshold

of 0.04 (except over Tsukuba), and no remarkable seasonal

trend was found for the shortening of the light path (blue

symbols in Figure 3 (right) are within the filled area). We

explain this by the very limited number of glint observations

in the Northern Hemisphere around these stations, where

dark surfaces could cause significant shortening of the

optical path (section 5.1.2). At the same time, the effects of

light path lengthening are clearly discernible in the increased

values of

from May

–

September each year (orange symbols

in Figure 3, right). The similar seasonal trends in water

vapor column abundance shown by green histograms in

Figure 3 (right) provide supporting circumstantial evidence

for the path length modifications due to light scattering by

aerosols or clouds.

[

] Removal of contaminated observations (

and

> 0.04) from the GOSAT data set results in acceptable

agreement between the seasonal variability of CO

at each

station and the TCCON measurements (Figure 3, left).

Overall, PPDF-D X

CO2

retrievals reproduce the temporal

and spatial patterns observed in the TCCON measurements

and simulated by NIES TM well. The disagreement between

GOSAT and ground-based FTS X

CO2

is mostly within the

error bars of satellite-based X

CO2

data. A clear and

Figure 3.

(left) Time series of X

CO2

and (right) PPDF parameters and average water vapor column abun-

dance X

H2O

obtained from meteorological data of JMA over four TCCON stations (the station name is

given in the left panels) in the Northern Hemisphere within 15

- latitude

-longitude grid boxes cen-

tered at each TCCON station. For clearer display within the same axis scale, X

H2O

(mole fractions) was

multiplied by a factor of 30. The number of available GOSAT single scans for each TCCON station is

listed in Table 1 (

). The light blue shaded areas in the right panels represent the 0.04 threshold beyond

which the retrievals were eliminated.

OSHCHEPKOV ET AL.: VALIDATION OF PPDF-BASED CO

RETRIEVALS

D12305

8of18

pronounced seasonal cycle of CO

column abundance can

be seen, with minimums in late summer and maxima in

spring. The largest amplitude of the seasonal trend was

found at Park Falls, as previously reported in the GOSAT

validation study by

Butz et al.

[2011]. This large amplitude

may be attributable to the large forest area surrounding Park

Falls, which is known to be a strong sink of CO

in summer,

as well as to increased fossil fuel consumption in the winter

season.

[

] Retrievals over Sodankyla demonstrate the limits of

GOSAT SWIR observations at high latitudes, even on a

seasonal time scale (Figure 3, top), due to large solar zenith

angles, cloudy conditions, and ice-covered surfaces.

[

] We present X

CO2

validation results over Lamont and

Tsukuba in Figure 4. These stations are located at almost the

same latitude of

N but the surrounding areas have dif-

ferent effects of atmospheric light scattering, as well as dif-

ferent surface properties. Figure 4 (bottom) shows the

seasonal variability in aerosol optical depth (AOD - red

histogram) at the 1.6-

m wavelength calculated from an

offline three-dimensional aerosol transport model, the

Spectral Radiation-Transport Model for Aerosol Species

(SPRINTARS) [

Takemura et al.

, 2000]. This panel also

shows the product of the cirrus fraction by cirrus optical

depth at the 532-nm wavelength (

- blue stars) derived from

observations by CALIOP, which is the primary instrument

carried by CALIPSO [

Winker et al.

, 2007]. Here and in the

following sections, the SPRINTARS (AOD) and CALIOP

(

) data have a spatial resolution of 2.8

-latitude

2.8

longitude and 5

-latitude

-longitude grid boxes.

Whereas the seasonal trends and magnitude of cirrus optical

prosperities are similar for both stations (blue stars in

Figure 4, bottom), the values of AOD (red histograms) are

significantly larger over the Tsukuba site. This could be

attributable to light scattering by windborne soil and mineral

dust aerosols, which are transported to Japan from the Gobi

and Takla Makan deserts in China due to sandstorms

[

Shimizu et al.

, 2004]. Global distributions of the aerosol

optical depth from SPRINTARS in Figure 5 (top) support

this idea; the dark green area (indicating large AOD) covers

Japan in July (Figure 5, left). Tsukuba is near Kasumigaura

Lake and 57 km west of the Pacific Ocean. A large amount

of elevated aerosols and subvisible cirrus over the dark

surface around this station lead to the shortening of the

Figure 4.

Validation of GOSAT retrievals and characteristics of atmospheric light scattering (left) over

Lamont and (right) over Tsukuba. (top) Time series of X

CO2

from TCCON (green symbols), GOSAT

(blue symbols), and NIES TM (red crosses). (middle) Time series of PPDF parameters

(blue symbols),

(orange symbols), and average water vapor column abundance X

H2O

(green histogram). For clearer dis-

play within the same axis scale, X

H2O

[mole fractions] was multiplied by a factor of 10. (bottom) Time

series of aerosol optical depth (AOD) from the SPRINTARS model (red histogram) and from the product

of the cirrus fraction by cirrus optical depth (

, stars) derived from CALIOP data.

OSHCHEPKOV ET AL.: VALIDATION OF PPDF-BASED CO

RETRIEVALS

D12305

9of18

optical path during summer. Values of

are distinctly

higher over Tsukuba than over Lamont, where the light path

is dominantly lengthening at the seasonal maximums of the

aerosol and cirrus optical depth (Figure 4, left). Although the

FTS X

CO2

data at Tsukuba are rather noisy and limited to

around 1 year, GOSAT PPDF-D retrievals over both Tsu-

kuba and Lamont have similar seasonal trends and agree

with the TCCON data and NIES TM results (Figure 4, top).

5.1.2. Ascension Island

[

] Ascension Island is a future TCCON site [

Geibel

et al.

, 2010] (ground-based FTS observations were not

yet available at the time of this study). Here, we mostly

demonstrate the large impact of atmospheric light scat-

tering on GOSAT retrievals over Ascension Island when

observing reflected sunlight over the ocean surface. Indeed,

after the TANSO-CAI test and post-retrieval quality assess-

ment, only 27% of GOSAT soundings were unaffected by

light path modifications in the atmosphere (

scatter in

Figure 6 (middle), section 5.2). Most GOSAT soundings

over Ascension Island were related to the sun-glint observa-

tion mode. In the near-infrared region, the ocean surface is

known to be dark in all directions except at the sun glint

mode. For a dark surface, the detected backscattered light

does not form by the photons that reach the absorbing sur-

face; therefore, the light path tends to be shorter due to light

scattering by aerosol and clouds. The sun-glint observation

mode is not exempt from this rule because only a small

fraction of photons reach the detector in the glint direction

after being scattered by aerosol/cloud and then reflected by

the surface.

[

] As a result, the effects of atmospheric light scattering

shorten the path length over the ocean, and the PPDF-D

retrievals underestimate X

CO2

[

Bril et al.

, 2012]. We dem-

onstrate this in Figure 6 (top) for retrieval of the CO

amount

for a set of GOSAT observations with

> 0.04 (brown

symbols). The negative bias due to shortening of the light

path could be as much as

8 ppm.

[

] Regarding the source of the high level of atmospheric

light scattering over Ascension Island, either heightened

amounts of high-altitude cirrus or aerosols could shorten the

light path. Figure 6 (bottom) displays seasonal variability in

the aerosol optical depth calculated by SPRINTARS

[

Takemura et al.

, 2000] (red histogram) and the product of

the cirrus fraction by the cirrus optical depth (

) derived

from the CALIOP observations [

Winker et al.

, 2007] (stars).

As Figure 6 (bottom) suggests, cirrus and aerosol impacts on

the optical path modifications have different seasonal vari-

ability. Whereas cirrus appears mostly between November

2009 and May 2010 (stars), which correlates seasonally

with water vapor amounts (green histogram in the middle

panel), the AOD has heightened values in July

–

August

2010 (red histogram), which is consistent with the peaks

Figure 5.

Global maps of (top) the total aerosol optical depth (AOD) from SPRINTARS and (bottom) aero-

sol size (in terms of the percentage of small particles) from MODIS for (left) July 2010 and (right) December

2010. The AOD data are monthly means, spatially averaged within 2.5

-latitude

2.5

-longitude grid boxes,

and values correspond to the color scales.

OSHCHEPKOV ET AL.: VALIDATION OF PPDF-BASED CO

RETRIEVALS

D12305

10 of 18

scatter (Figure 6, middle). Figure 5 displays global dis-

tributions of the monthly mean AOD (top) from SPRINTARS

and the percentage of small particles (bottom) from the

Moderate Resolution Imaging Spectroradiometer (MODIS)

on NASA

’

s Terra satellite (NASA Earth Observatory, http://

earthobservatory.nasa.gov) for July 2010 (left) and December

2010 (right). The percentage of small (fine) aerosol particles

was determined using the MODIS algorithm of

Remer et al.

[2005] with two aerosol modes that differ in particle size and

composition (effective particle radius of 0.1

m for fine mode

and >1

m for coarse mode). Aerosol plumes dominated by

large particles can be seen over Ascension Island in July

(green areas in the bottom left panel) when the AOD is sig-

nificant in northern Africa and tropical regions of the Atlantic

Ocean (green areas in the bottom left panel). This suggests that

the light path modification over this station may largely be due

to windborne mineral dust aerosols from sand storms in the

Sahara and Arabian Peninsula deserts.

[

] Blue symbols in Figure 6 (top) display X

CO2

retrievals

for the rest of the GOSAT soundings derived under the clearest

atmospheric conditions (

< 0.04). Although the retrievals

have a small persistent negative bias of about 1

–

2ppmin

comparison with NIES TM, the shape of the small seasonal

variations is generally in line with that of the model, at least

before July 2010.

[

] When ground-based measurements at this station

become available, they will be useful for testing space-based

observations of greenhouse gases where light scattering by

cirrus and windborne mineral dust aerosol in the sun-glint

mode could significantly shorten the light path.

5.1.3. Australian Stations

[

] Figure 7 displays the validation of NIES PPDF-D

GOSAT data processing for two TCCON stations in Aus-

tralia: Darwin (left) and Wollongong (right). As expected,

the CO

retrievals show no pronounced seasonal cycle in the

Southern Hemisphere, which is similar at both stations.

[

] The retrievals over the tropical Darwin station are

highly contaminated by atmospheric light scattering for sun-

glint observations. The summer/wet monsoonal season from

December

–

May is very humid and dominated by marine

winds [

Bouya et al.

, 2010]. The pronounced shortening of

the optical path shows good correlation with the seasonal

variability in water vapor column abundance (blue symbols

and green histogram in Figure 7, middle left), as well as with

seasonal trends in cirrus occurrence (

) and the heightened

values of the AOD (stars and red histogram in Figure 7,

bottom left). Most Australian deserts are located in the

central and northwestern parts of the Australian continent,

and winter is the dry season [

Bouya et al.

, 2010]. PPDF-D

retrievals over land (black symbols in the top left panel)

appear to be reasonable in comparison with ground-based

FTS X

CO2

observations (green symbols in the top left panel).

[

] Similar to Tsukuba, Wollongong was one of the

most difficult sites for identifying the tendency of optical

path modification (Figure 7, right). The total AOD over

Wollongong is lower than that over Darwin (red histograms

in Figure 7, bottom). However, Wollongong is located on

the outskirts of a small urban center and, as pointed out by

Wunch et al.

[2011b], the measurements at this site might

be affected by local pollution. The retrieved values of the

and

PPDF parameters show large variation and are fre-

quently compatible with each other (Figure 7, middle right).

The absence of a remarkable tendency in the light path

modification might explain why the GOSAT data processing

with simultaneous gas and aerosol/cloud retrievals over land

around this site showed relatively high GOSAT X

CO2

retrieval

scatter compared with other stations [

Morino et al.

, 2011;

Butz et al.

, 2011]. PPDF-D data processing over the ocean

around this site showed comparatively lower scatter and bias

from ground-based FTS X

CO2

(blue symbols in Figure 7, top

right). Even for these observations, PPDF-based screening

was found to be important, as demonstrated by the strong

scatter and negative bias of X

CO2

retrieved from observa-

tions for which

> 0.04 and thus shortening of the optical

path due to aerosols or clouds over the ocean was significant

(brown symbols).

5.2. Pairwise TCCON-GOSAT Statistical Comparison

[

] Table 1 and Figure 8 summarize the pairwise statis-

tical comparison between GOSAT and TCCON X

CO2

data

within the coincidence criterion (section 4). To statistically

Figure 6.

Validation of GOSAT retrievals and characteris-

tics of atmospheric light scattering over Ascension Island

(details are the same as in Figure 4). Brown symbols in the

top panel display PPDF-D retrievals from GOSAT observa-

tions beyond the threshold of

> 0.04 when shortening of

the optical path leads to significant underestimation of the

gas amount. For clearer display within the same axis scale,

H2O

(mole fractions) and

values were multiplied by

factors of 50 and 10, respectively.

OSHCHEPKOV ET AL.: VALIDATION OF PPDF-BASED CO

RETRIEVALS

D12305

11 of 18

compare GOSAT (

) and TCCON (

), mean values of

CO2

we computed the following:

bias

Bias

ðÞ

;

standard deviation

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

Bias

ðÞ

;

Pearson

’

s correlation coefficient

ðÞ

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ðÞ

;

coefficient of determination (goodness of fit)

ðÞ

;

where

denotes the least squares-adjusted points (expecta-

tion values of

) in the best linear fit between the GOSAT

and TCCON X

CO2

mean values according to

York et al.

[2004]. This technique effectively accounts for uncertainties

in both the

and

data sets when they vary from point to

point. The mean values

∑

ðÞ

∑

in equations

(5)

–

(8) are weighted as

⋅

;

through standard deviations

and

(

) of uncor-

related GOSAT (

) and TCCON (

) mean values. In

equation (9)

is the regression slope, which is known to be

an important measure when characterizing the bias correc-

tion. The slope values and their errors (1

) are listed in

Table 1. For each TCCON station, we also tabulated the

following:

[

]

–

the total number of GOSAT single scans

available after the quality assessment (section 2.2), application

of X

CO2

selection criterion (section 4), and PPDF screening

(section 2.2);

[

]

–

remaining percentage of GOSAT single scans

after PPDF-based screening;

[

]

–

the total number of GOSAT single scans

coincident with TCCON measurements, after quality assess-

ment and application of X

CO2

selection criterion; and

Figure 7.

Validation of GOSAT retrievals and characteristics of atmospheric light scattering over Australian

stations (left) Darwin and (right) Wollongong. Details are the same as in Figure 4. For clearer display within the

same axis scale, X

H2O

(mole fractions) was multiplied by a factor of 10.

OSHCHEPKOV ET AL.: VALIDATION OF PPDF-BASED CO

RETRIEVALS

D12305

12 of 18

[

]

–

the number of coincident GOSAT and TCCON

mean values of X

CO2

[

] Statistical characteristics are listed in Table 1 for each

TCCON site, as well as summarized separately for stations

in the Northern and Southern hemispheres. The last row in

Table 1 represents the data for all TCCON sites. Slope

correlation coefficients (7), and coefficients of determination

(8) are not summarized in Table 1 for three operational sta-

tions in the Southern Hemisphere (blank cells) due to the

low variability in X

CO2

, but both GOSAT and TCCON data

for these sites were included when estimating these quanti-

ties for all operational sites (last row in Table 1). Incorpo-

ration of data from the Southern Hemisphere into the total

data set had little impact on the slope of the linear regression,

but noticeably degraded the correlation coefficients and

standard deviations between GOSAT and TCCON obser-

vations. The best values of the slope were for Lamont and

Bialystok (1.01);

0.1 deviation of the slope from unity

were found for the Bremen and Garmisch sites. We believe

that the relatively high deviation of the slope from unity

| > 1) for Sodankyla, Moshiri, and Tsukuba was due to

insufficient numbers (<300) of coincident GOSAT and

TCCON observations for proper statistical analysis; this

could also explain the largest bias in X

CO2

over Moshiri site

and the lowest correlation coefficient over Bremen site. The

close correlation of GOSAT observations with TCCON

measurements over Tsukuba and Moshiri is not representa-

tive and mostly attributable to the large uncertainties in X

CO2

at these TCCON stations (green bars in Figure 4 for the

Tsukuba site).

[

] The clearest atmosphere within the available set of

GOSAT

observations was over Lamont (

= 76%) and

Sodankyla (

= 77%; Table 1). Note again that this

assessment concerns only PPDF-based screening (

< 0.04) of the initial GOSAT soundings that were filtered

by the CAI prescreening test (section 2.1) and passed the

GOSAT data and retrieval quality assessment (section 2.2).

The smallest number of remaining scans with negligible

light path modification was detected in the Southern Hemi-

sphere stations over Ascension Island (

= 27%) and

Lauder (

= 30%). We discussed possible reasons for the

high level of atmospheric light scattering over Ascension

Island in section 5.1.2. As noted by

Liley and Forgan

[2009], the aerosol optical depth data obtained continu-

ously over Lauder since 1999 are among the lowest

observed worldwide. The SPRINTARS model also showed

fairly low values of AOD during the 22 months of GOSAT

observations considered in this study. We have no explana-

tion for the large light path modifications around this station,

except for the presence of elevated subvisible cirrus. The

maximal value of

= 0.055 (the product of the cirrus frac-

tion by cirrus optical depth from CALIOP observations) was

detected in November

–

December 2009 and is noticeably

larger than that over Ascension Island (

= 0.017 in the same

period). This example shows that the total aerosol optical

depth is not always a sufficiently representative quantity to

characterize the possible impact of atmospheric light scat-

tering when retrieving gas amounts from space [

Oshchepkov

et al.

, 2011].

[

] For the coincident measurements over all stations,

GOSAT and TCCON data of CO

column abundance

showed a correlation coefficient of 0.85, standard deviation

of 1.80 ppm, and a sub-ppm negative bias of

0.43 ppm.

For the Northern Hemisphere sites, where pronounced sea-

sonal trends provide larger variability in X

CO2

, these statis-

tical characteristics were 0.90, 1.65 ppm, and

0.47 ppm,

respectively (Table 1).

5.3. Global Distribution of CO

[

] Figure 9 displays a comparison between the global

distributions of CO

monthly mean column abundance cal-

culated with NIES TM (right panels) and those retrieved

from GOSAT data processing (left panels represent the

retrievals as deviation from the modeled values). The mod-

eled data shown in Figure 9 were generated only for the

locations of those GOSAT soundings that passed through

the CAI prescreening test (section 2.1) and also passed the

GOSAT data and retrieval quality assessment (section 2.2).

This results in some gaps in the NIES X

CO2

monthly

averages. To produce these maps, we did not apply weighted

interpolation, such as by the kriging technique; we simply

calculated the spatial average of X

CO2

values within 2.5

latitude

2.5

-longitude grid boxes and excluded cases in

Figure 8.

Correlation diagram between GOSAT and

ground-based FTS measurements of X

CO2

. The GOSAT

data were selected within 15

-latitude

-longitude grid

boxes centered over each FTS station. The analyzed

ground-based FTS data were mean values measured within

1 h of the GOSAT overpass time, and both GOSAT and

TCCON data were averaged weekly. Red lines correspond

to the best fit for all sites (with the fitting equation given

in the inset) and the green line representing one-to-one cor-

respondence. The number of available GOSAT soundings

and characteristics of statistical relationships between

ground-based FTS and GOSAT X

CO2

for each TCCON sta-

tion are listed in Table 1.

OSHCHEPKOV ET AL.: VALIDATION OF PPDF-BASED CO

RETRIEVALS

D12305

13 of 18

which only one scan fell into the grid box as unrepresenta-

tive data.

[

] The period covers 10 months of GOSAT observations

in 2010 (every 3 months from February

–

November 2010).

These months were selected because they reflect significant

temporal and spatial X

CO2

variability. In particular, the X

CO2

latitudinal gradient changes sign between February and

August; the highest amount of CO

in the Northern

Hemisphere is in May; and a comparatively homogeneous

global distribution of CO

can be observed in November,

with considerable enhancement of CO

amount in the east-

ern parts of the USA and China (Figure 9).

[

] The global distribution of retrieved X

CO2

reasonably

reproduces the column abundance derived from NIES TM

(Figure 9, right). The displayed deviations

CO2

GOSAT

CO2

Model

between retrieved X

CO2

GOSAT

and modeled

Figure 9.

Global maps of the monthly mean X

CO2

(values (ppm) correspond to the color scales) (left)

from NIES TM and (right) from GOSAT data processing in terms of deviation

CO2

GOSAT

CO2

Model

from (top to bottom) February

–

November 2010. The data are averaged over 2.5

-latitude

2.5

-longitude

grid boxes.

OSHCHEPKOV ET AL.: VALIDATION OF PPDF-BASED CO

RETRIEVALS

D12305

14 of 18