IASI on Metop-A Operational Level 2 retrievals after five years in orbit
IASI on Metop-A Operational Level 2 retrievals after five years in orbit
IASI on Metop-A Operational Level 2 retrievals after five years in orbit
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>on</strong> <strong>Metop</strong>-A: Operati<strong>on</strong>al <strong>Level</strong> 2 <strong>retrievals</strong><br />
<strong>after</strong> <strong>five</strong> <strong>years</strong> <strong>in</strong> <strong>orbit</strong><br />
Thomas August n , Dieter Klaes, Peter Schlüssel, Tim Hultberg, Marc Crapeau,<br />
Arl<strong>in</strong>do Arriaga, Anne O’Carroll, Dorothée Coppens, Rose Munro, Xavier Calbet<br />
EUMETSAT, Eumetsat Allee 1, D-64295 Darmstadt, Germany<br />
article <strong>in</strong>fo<br />
Available <strong>on</strong>l<strong>in</strong>e 9 March 2012<br />
Keywords:<br />
Hyperspectral <strong>in</strong>frared sound<strong>in</strong>g<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
<strong>Metop</strong><br />
Operati<strong>on</strong>al <strong>retrievals</strong><br />
1. Introducti<strong>on</strong><br />
1.1. The EUMETSAT Polar System<br />
abstract<br />
The EUMETSAT Polar System (EPS) is EUMETSAT’s<br />
c<strong>on</strong>tributi<strong>on</strong> to the Initial Jo<strong>in</strong>t Polar System (IJPS). The<br />
IJPS provides observati<strong>on</strong>s for operati<strong>on</strong>al meteorology<br />
and climate m<strong>on</strong>itor<strong>in</strong>g from both the mid morn<strong>in</strong>g <strong>orbit</strong>,<br />
under the resp<strong>on</strong>sibility of EUMETSAT, and the <strong>after</strong>no<strong>on</strong><br />
<strong>orbit</strong>, under the resp<strong>on</strong>sibility of NOAA.<br />
The <strong>Metop</strong> satellite series is the space comp<strong>on</strong>ent of<br />
the EUMETSAT Polar System (EPS). <strong>Metop</strong>-A, the first <strong>in</strong> a<br />
series of three spacecraft, was launched <strong>in</strong> October 2006.<br />
S<strong>in</strong>ce May 2007 it has provided data c<strong>on</strong>t<strong>in</strong>uously from<br />
the eight meteorological <strong>in</strong>struments <strong>on</strong> board. The three<br />
n<br />
Corresp<strong>on</strong>d<strong>in</strong>g author. Tel.: þ49 6151 807 5650;<br />
fax: þ49 6151 807 8380.<br />
E-mail address: thomas.august@eumetsat.<strong>in</strong>t (T. August).<br />
C<strong>on</strong>tents lists available at SciVerse ScienceDirect<br />
Journal of Quantitative Spectroscopy &<br />
Radiative Transfer<br />
0022-4073/$ - see fr<strong>on</strong>t matter & 2012 Elsevier Ltd. All rights reserved.<br />
doi:10.1016/j.jqsrt.2012.02.028<br />
Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
journal homepage: www.elsevier.com/locate/jqsrt<br />
Geophysical parameters from the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>in</strong>strument <strong>on</strong> <strong>Metop</strong>-A are essential products<br />
provided from EUMETSAT’s Central Facility <strong>in</strong> near real time. They <strong>in</strong>clude vertical<br />
profiles of temperature and humidity, related cloud <strong>in</strong>formati<strong>on</strong>, surface emissivity and<br />
temperature, and atmospheric compositi<strong>on</strong> parameters (CO, oz<strong>on</strong>e and several other<br />
trace gases). As compared to previous operati<strong>on</strong>al processor versi<strong>on</strong>s, the latest<br />
processor versi<strong>on</strong> 5 delivers significant improvements <strong>in</strong> retrieval performance for<br />
most major products. These <strong>in</strong>clude improvements to cloud properties products, cloud<br />
detecti<strong>on</strong> (with a positive impact <strong>on</strong> the knowledge of the sea surface temperature,<br />
SST), the temperature profile (especially <strong>in</strong> the mid and upper troposphere), and oz<strong>on</strong>e<br />
and carb<strong>on</strong> m<strong>on</strong>oxide total columns.<br />
This paper provides a comprehensive summary of the process<strong>in</strong>g algorithms, the<br />
latest scientific developments, and the related validati<strong>on</strong> studies and activities. It<br />
c<strong>on</strong>cludes with a discussi<strong>on</strong> of the future outlook.<br />
& 2012 Elsevier Ltd. All rights reserved.<br />
<strong>Metop</strong> satellites (with launches of <strong>Metop</strong>-B and -C<br />
planned <strong>in</strong> 2012 and 2017, respectively) will provide a<br />
c<strong>on</strong>t<strong>in</strong>uous service from the mid-morn<strong>in</strong>g <strong>orbit</strong> (9:30<br />
Local Solar Time equator cross<strong>in</strong>g time, descend<strong>in</strong>g node)<br />
for at least 15 <strong>years</strong>. More details <strong>on</strong> the EPS/<strong>Metop</strong><br />
system can be found <strong>in</strong> [1].<br />
The most <strong>in</strong>novative and <strong>on</strong>e of the key <strong>in</strong>struments <strong>on</strong><br />
<strong>Metop</strong> is the Infrared Atmospheric Sound<strong>in</strong>g Interferometer<br />
(<str<strong>on</strong>g>IASI</str<strong>on</strong>g>). Three <strong>in</strong>struments were developed for <strong>Metop</strong><br />
by CNES (Centre Nati<strong>on</strong>al d’Etudes Spatiales) <strong>in</strong> cooperati<strong>on</strong><br />
with EUMETSAT. They are built to provide temperature,<br />
moisture with unprecedented accuracy and<br />
resoluti<strong>on</strong> and additi<strong>on</strong>ally to provide <strong>in</strong>formati<strong>on</strong> for the<br />
m<strong>on</strong>itor<strong>in</strong>g of atmospheric trace gases. <str<strong>on</strong>g>IASI</str<strong>on</strong>g> is a Michels<strong>on</strong><br />
<strong>in</strong>terferometer measur<strong>in</strong>g <strong>in</strong> the <strong>in</strong>frared. It measures 8461<br />
spectral samples between 3.62 and 15.5 mm with a resoluti<strong>on</strong><br />
of 0.5 cm 1 <strong>after</strong> apodisati<strong>on</strong>. <str<strong>on</strong>g>IASI</str<strong>on</strong>g> scans across-track<br />
<strong>in</strong> 30 successive elementary fields of view (EFOV), each<br />
composed of 4 <strong>in</strong>stantaneous fields of view (IFOV). The<br />
EFOVs span a 748.331 range, symmetric with respect to
the Nadir, <strong>in</strong> steps of 3.331. The swath width <strong>on</strong> ground is<br />
approximately 2200 km, which provides global Earth coverage<br />
twice per day. The IFOV is a disc of 12 km diameter<br />
at sub-satellite po<strong>in</strong>t. The dimensi<strong>on</strong>s <strong>in</strong>crease to 39 km<br />
and 20 km at the swath edge <strong>in</strong> the across- and al<strong>on</strong>g-track<br />
directi<strong>on</strong>s, respectively. More details about the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>in</strong>strument<br />
are available <strong>in</strong> [2,3].<br />
Operati<strong>on</strong>al products from EPS/<strong>Metop</strong> are generated <strong>in</strong><br />
the EPS Core Ground Segment, located at EUMETSAT headquarters<br />
<strong>in</strong> Darmstadt, Germany, and also <strong>in</strong> eight decentralised<br />
Satellite Applicati<strong>on</strong> Facilities (SAF), hosted by<br />
EUMETSAT Member States. More details <strong>on</strong> the EPS products<br />
and applicati<strong>on</strong>s are discussed <strong>in</strong> [1] and, more<br />
especially, <str<strong>on</strong>g>IASI</str<strong>on</strong>g> applicati<strong>on</strong>s are discussed by Hilt<strong>on</strong> et al. [4].<br />
1.2. The operati<strong>on</strong>al <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>Level</strong> 2 products<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1341<br />
Development activities for the centrally processed <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
L2 products [5] took place at EUMETSAT and were supported<br />
by science expert groups and additi<strong>on</strong>ally through<br />
dedicated <strong>in</strong>ternal and external studies. For <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>in</strong> particular,<br />
as a highly <strong>in</strong>novative <strong>in</strong>strument, c<strong>on</strong>siderable<br />
scientific support was required. For this purpose and to<br />
foster scientific development for <str<strong>on</strong>g>IASI</str<strong>on</strong>g> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> Sound<strong>in</strong>g<br />
Science Work<strong>in</strong>g Group (ISSWG) was established. A Science<br />
Plan [6] emerged which was used as a guidel<strong>in</strong>e for the<br />
product process<strong>in</strong>g development and for the group’s activities.<br />
There was a str<strong>on</strong>g heritage from the Atmospheric<br />
Infrared Sounder (AIRS) embarked <strong>on</strong> NASA’s Aqua satellite<br />
<strong>in</strong> 2002, which paved the way for the use of hyperspectral<br />
data <strong>in</strong> meteorology and climatology [7]. The<str<strong>on</strong>g>IASI</str<strong>on</strong>g><strong>Level</strong>2<br />
process<strong>in</strong>g development targeted the generati<strong>on</strong> of temperature<br />
and humidity profile <strong>in</strong>formati<strong>on</strong>, the associated<br />
surface <strong>in</strong>formati<strong>on</strong> and the retrieval of some trace gas<br />
species: CO, O 3,CH 4,N 2OandCO 2 from the beg<strong>in</strong>n<strong>in</strong>g of<br />
<strong>Metop</strong> operati<strong>on</strong>s. The vertical temperature and watervapour<br />
profiles are currently represented and distributed<br />
<strong>on</strong> a 90-level grid extend<strong>in</strong>g between 0.005 and 1050 hPa.<br />
The <strong>in</strong>dependence of the retrieval algorithms from Numerical<br />
Weather Predicti<strong>on</strong> (NWP) was a strict requirement<br />
from the beg<strong>in</strong>n<strong>in</strong>g. A large number of scientific studies<br />
both <strong>in</strong>ternal and external to the ISSWG, supported the<br />
evoluti<strong>on</strong> of the science plan and the processor development,<br />
thereby enabl<strong>in</strong>g the operati<strong>on</strong>al producti<strong>on</strong> of <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
L2 products by the end of the <strong>in</strong>strument’s commissi<strong>on</strong><strong>in</strong>g.<br />
As with the other EPS products, the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 processor has<br />
been subject to post-launch developments to extend the<br />
scope of and improve further the products. These are<br />
referred to as ‘‘Day2’’ activities.<br />
The versi<strong>on</strong> 5 (v5) of the processor was released <strong>on</strong> the<br />
operati<strong>on</strong>al process<strong>in</strong>g cha<strong>in</strong> <strong>on</strong> 14 September 2010. In<br />
the follow<strong>in</strong>g we provide a full descripti<strong>on</strong> of this new<br />
processor versi<strong>on</strong>, the associated validati<strong>on</strong> results, the<br />
improvements with respect to versi<strong>on</strong> 4, and the latest<br />
developments for each of the retrieved parameters <strong>after</strong> 5<br />
<strong>years</strong> <strong>in</strong> operati<strong>on</strong>. Each parameter is covered <strong>in</strong> <strong>on</strong>e or<br />
more dedicated algorithm descripti<strong>on</strong> documents and<br />
validati<strong>on</strong> reports, which are cited <strong>in</strong> the text. They are<br />
accessible <strong>on</strong>-l<strong>in</strong>e <strong>in</strong> the ‘‘Data & Products secti<strong>on</strong>’’ at<br />
www.eumetsat.<strong>in</strong>t/Home/Ma<strong>in</strong>/DataProducts/Resources/<br />
<strong>in</strong>dex.htm (last accessed 17/01/2012) or alternatively <strong>on</strong><br />
request with the reference number.<br />
2. The <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>Level</strong> 2 PPF versi<strong>on</strong> 5<br />
The <strong>Level</strong> 2 <str<strong>on</strong>g>IASI</str<strong>on</strong>g> Product Process<strong>in</strong>g Facility (PPF) has a<br />
modular structure, represented <strong>in</strong> Fig. 1. The <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements<br />
and various auxiliary data (e.g. measurements from<br />
other EPS <strong>in</strong>struments, NWP forecasts etc.) are first collocated<br />
<strong>in</strong> a data pre-process<strong>in</strong>g step, which also <strong>in</strong>cludes the<br />
c<strong>on</strong>figurati<strong>on</strong> of the retrieval algorithms with coefficients<br />
and thresholds adapted to the time and locati<strong>on</strong> of the<br />
acquisiti<strong>on</strong>s. The cloud<strong>in</strong>ess with<strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> pixel is assessed<br />
<strong>in</strong> a sec<strong>on</strong>d step us<strong>in</strong>g a number of cloud detecti<strong>on</strong> methods.<br />
In the presence of a cloud, the cloud properties – fracti<strong>on</strong>al<br />
coverage, height and phase – are determ<strong>in</strong>ed. The retrieval<br />
steps follow with a series of statistical methods to estimate<br />
the temperature and humidity profiles, the surface emissivity<br />
as well as the sea and land surface temperature (SST and<br />
LST, respectively), and some trace gas species. A physical<br />
retrieval us<strong>in</strong>g the optimal estimati<strong>on</strong> method (OEM) is<br />
subsequently performed to ref<strong>in</strong>e the temperature and<br />
oz<strong>on</strong>e profiles. These successive process<strong>in</strong>g steps are<br />
detailed <strong>in</strong> the follow<strong>in</strong>g secti<strong>on</strong>s.<br />
2.1. Pre-process<strong>in</strong>g<br />
Besides the calibrated and apodised spectra stored <strong>in</strong><br />
the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L1c products, the L2 PPF <strong>in</strong>gests two types of<br />
auxiliary data.<br />
Fig. 1. Modular representati<strong>on</strong> of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>Level</strong> 2 operati<strong>on</strong>al process<strong>in</strong>g cha<strong>in</strong> (versi<strong>on</strong> 5), operati<strong>on</strong>al s<strong>in</strong>ce 14/09/2010.
1342<br />
Table 1<br />
Static atlases used <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF.<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
Data type Name Resoluti<strong>on</strong><br />
Land/sea mask – Quadtree a<br />
Surface GTOPO 30 [10] 30 <strong>in</strong>. 30 <strong>in</strong>.<br />
elevati<strong>on</strong><br />
( 1 km)<br />
Emissivity Global land surface<br />
0.51 0.51<br />
Atlas<br />
climatology<br />
Full spectrum<br />
for <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements [9] (via PCs)<br />
M<strong>on</strong>thly means<br />
a A quadtree is a tree data structure stor<strong>in</strong>g values <strong>on</strong>ly for homogeneous<br />
regi<strong>on</strong>s rather than for every pixel. The spatial resoluti<strong>on</strong> is thus<br />
vary<strong>in</strong>g and adapted here to the land/sea coastal <strong>in</strong>formati<strong>on</strong>. The<br />
dataset detailed descripti<strong>on</strong> and usage can be found <strong>in</strong> the EUMETSAT<br />
Technical Note ‘‘Descripti<strong>on</strong> of Land/sea and Coastl<strong>in</strong>e Data Bases for<br />
AAPP upgrade’’, EUM/EPS/SYS/TEN/00/018.<br />
The first type are the static data: the auxiliary datasets<br />
c<strong>on</strong>figur<strong>in</strong>g the various algorithms <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF (e.g.<br />
thresholds) and the static atlases (listed <strong>in</strong> Table 1) used<br />
to compute the land/sea fracti<strong>on</strong>al coverage, the surface<br />
elevati<strong>on</strong> and the land surface emissivity with<strong>in</strong> a given<br />
field of view. In previous releases, the land surface<br />
emissivity was derived from the Internati<strong>on</strong>al Geosphere/Biosphere<br />
Programme (IGBP) land surface properties<br />
database [8]. In the operati<strong>on</strong>al processor versi<strong>on</strong> 5<br />
this is replaced by the recent land surface emissivity<br />
m<strong>on</strong>thly atlas based <strong>on</strong> <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements developed<br />
by Zhou et al. [9]. The emissivity is def<strong>in</strong>ed <strong>on</strong> a coarser<br />
grid (0.51 0.51 l<strong>on</strong>gitude/latitude) but is provided at the<br />
full <str<strong>on</strong>g>IASI</str<strong>on</strong>g> spectral resoluti<strong>on</strong>.<br />
The sec<strong>on</strong>d type of auxiliary <strong>in</strong>formati<strong>on</strong> is data flow<strong>in</strong>g<br />
<strong>in</strong> near real time (NRT) to support the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
<strong>retrievals</strong>, such as NWP forecasts from the European<br />
Centre for Medium-Range Weather Forecasts (ECMWF)<br />
and data from other <strong>in</strong>struments <strong>on</strong>-board <strong>Metop</strong>:<br />
namely the Advanced Very High Resoluti<strong>on</strong> Radiometer<br />
(AVHRR), the Advanced Microwave Sound<strong>in</strong>g Unit<br />
(AMSU) and the (Microwave Humidity Sounder) MHS<br />
L1B products [1]. The validity of these variable data and<br />
of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L1C spectra is checked before process<strong>in</strong>g by<br />
<strong>in</strong>specti<strong>on</strong> of their quality flags when applicable, and by<br />
comparis<strong>on</strong> aga<strong>in</strong>st predef<strong>in</strong>ed and c<strong>on</strong>figurable validity<br />
bounds.<br />
2.1.1. Auxiliary data and measurements collocati<strong>on</strong><br />
When the spatial resoluti<strong>on</strong> of the auxiliary data is<br />
higher than the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> IFOV the <strong>in</strong>formati<strong>on</strong> is averaged<br />
with<strong>in</strong> the IFOV and weighted with the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> po<strong>in</strong>t spread<br />
functi<strong>on</strong> (PSF) to account for n<strong>on</strong>-homogeneities <strong>in</strong> the<br />
detector’s resp<strong>on</strong>se. This is the case with the digital<br />
elevati<strong>on</strong> model (DEM), the AVHRR cloud <strong>in</strong>formati<strong>on</strong><br />
and the land/sea mask.<br />
In c<strong>on</strong>trast, when the spatial resoluti<strong>on</strong> is of the order<br />
the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> footpr<strong>in</strong>t size (of 12 km at nadir), as with the<br />
numerical weather predicti<strong>on</strong> (NWP) forecast fields (this<br />
<strong>in</strong>formati<strong>on</strong> is used <strong>in</strong> the cloud detecti<strong>on</strong> <strong>on</strong>ly), the<br />
collocati<strong>on</strong> is performed us<strong>in</strong>g a nearest neighbour<br />
method or by bil<strong>in</strong>ear <strong>in</strong>terpolati<strong>on</strong> between the four<br />
nearest grid-po<strong>in</strong>ts. For the atmospheric forecast parameters<br />
provided <strong>on</strong> a vertical grid, such as the temperature,<br />
humidity and oz<strong>on</strong>e profiles, a vertical <strong>in</strong>terpolati<strong>on</strong><br />
is performed from the orig<strong>in</strong>al grid to the c<strong>on</strong>stant<br />
pressure grid used with<strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF prior to the<br />
horiz<strong>on</strong>tal <strong>in</strong>terpolati<strong>on</strong>. S<strong>in</strong>ce 08/11/2011, the 3-h forecasts<br />
(at 00, 03, 06, 09, 12, 15, 18, 21 UTC) from ECMWF<br />
are <strong>in</strong>gested. Older L2 products were processed with the<br />
6-h forecasts for the synoptic times 00, 06, 12 and 18 UTC.<br />
The forecast parameters are then <strong>in</strong>terpolated <strong>in</strong> time as a<br />
first order approximati<strong>on</strong> of the atmospheric state at the<br />
actual sens<strong>in</strong>g time.<br />
Collocati<strong>on</strong> of AMSU and MHS data has been already<br />
implemented <strong>in</strong> preparati<strong>on</strong> for the planned synergistic<br />
use of microwave measurements with the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>in</strong>frared<br />
(IR) spectra with<strong>in</strong> the cloud and atmospheric temperature<br />
and humidity profiles <strong>retrievals</strong>.<br />
2.1.2. Radiance noise filter<strong>in</strong>g<br />
The radiance noise filter<strong>in</strong>g is the last step of the data<br />
pre-process<strong>in</strong>g sequence and was <strong>in</strong>troduced with PPF v5.<br />
It c<strong>on</strong>sists of filter<strong>in</strong>g out part of the noise present <strong>in</strong> the<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> L1C spectra us<strong>in</strong>g pr<strong>in</strong>cipal comp<strong>on</strong>ent analysis (PCA)<br />
techniques. At <str<strong>on</strong>g>IASI</str<strong>on</strong>g> spectral resoluti<strong>on</strong>, the top of atmosphere<br />
(TOA) radiances sampled <strong>in</strong> the 8461 channels are<br />
spectrally correlated and do not represent as many<br />
<strong>in</strong>dependent <strong>in</strong>formati<strong>on</strong>. PCA theory states that these<br />
radiances can be represented by a small number of<br />
eigenvectors of their covariance matrix. A base of empirical<br />
orthog<strong>on</strong>al functi<strong>on</strong>s (EOF) was therefore computed at<br />
EUMETSAT from a large set of apodised <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L1C spectra<br />
corresp<strong>on</strong>d<strong>in</strong>g to a representative collecti<strong>on</strong> of atmospheric<br />
situati<strong>on</strong>s. We c<strong>on</strong>sider here the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements<br />
as the sum of the TOA radiances and the<br />
<strong>in</strong>strument noise. While the atmospheric signal subspace<br />
lies <strong>in</strong> the most significant eigenvectors, the random<br />
comp<strong>on</strong>ent (i.e. the noise) is equally distributed <strong>in</strong> all<br />
eigenvectors. Hence, discard<strong>in</strong>g the eigenvectors of higher<br />
rank allows the TOA radiances to be rec<strong>on</strong>structed with<br />
part of the noise removed [11]. To rec<strong>on</strong>struct and noisefilter<br />
the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> bands 1, 2 and 3, respectively, 90, 120 and<br />
80 eigenvectors are currently used <strong>in</strong> operati<strong>on</strong>s. The<br />
difference between the orig<strong>in</strong>al and the rec<strong>on</strong>structed<br />
radiance is called the residual. For rare atmospheric<br />
situati<strong>on</strong>s – e.g. excepti<strong>on</strong>al volcanic erupti<strong>on</strong>s or wild<br />
fires with unique plume compositi<strong>on</strong>s – it might happen<br />
that some spectral features are not well represented by<br />
the eigenvectors. These situati<strong>on</strong>s are usually of particular<br />
<strong>in</strong>terest for specific research studies <strong>on</strong> atmospheric<br />
compositi<strong>on</strong> which might require the orig<strong>in</strong>al spectra<br />
while the operati<strong>on</strong>al purposes of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 products<br />
are better served with low-noise level (i.e. filtered)<br />
radiances. The ability to represent the atmospheric signal<br />
with these lead<strong>in</strong>g eigenvectors is rout<strong>in</strong>ely m<strong>on</strong>itored.<br />
The collecti<strong>on</strong> of spectra support<strong>in</strong>g the EOF has been<br />
iteratively extended and the eigenvectors updated to<br />
<strong>in</strong>clude rare spectra, e.g. from the Kasatochi erupti<strong>on</strong><br />
(08/08/2008) [12,13] and from the Russian Fires (summer<br />
2010) [14]. The reader is referred to the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> pr<strong>in</strong>cipal<br />
comp<strong>on</strong>ent compressi<strong>on</strong> product generati<strong>on</strong> specificati<strong>on</strong><br />
(EUM/OPS-EPS/SPE/08/0199) and validati<strong>on</strong> report
(EUM/OPS-EPS/REP/10/0148), accessible <strong>on</strong>-l<strong>in</strong>e at www.<br />
eumetsat.<strong>in</strong>t <strong>in</strong> the secti<strong>on</strong> Data & Products, for further<br />
details and c<strong>on</strong>figurati<strong>on</strong> descripti<strong>on</strong>.<br />
2.2. Cloud detecti<strong>on</strong><br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1343<br />
Efficient cloud detecti<strong>on</strong> is essential for accurate L2<br />
product generati<strong>on</strong> because most of the subsequent<br />
retrieval functi<strong>on</strong>s are tailored for cloud-free atmospheres.<br />
Thus, undetected cloud c<strong>on</strong>tam<strong>in</strong>ati<strong>on</strong> would<br />
degrade the quality of the f<strong>in</strong>al <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 products. The<br />
cloud screen<strong>in</strong>g <strong>in</strong> previous <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF versi<strong>on</strong>s relied<br />
solely <strong>on</strong> the NWP cloud test (see Secti<strong>on</strong> 2.3.1 below). In<br />
the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF v5 two additi<strong>on</strong>al cloud tests were added.<br />
They are described <strong>in</strong> Secti<strong>on</strong>s 2.3.2 and 2.3.3. A more<br />
recent detecti<strong>on</strong> technique based <strong>on</strong> artificial neural networks<br />
(ANN) is discussed <strong>in</strong> Secti<strong>on</strong> 2.4.1. It was <strong>in</strong>troduced<br />
<strong>in</strong> October 2011 for m<strong>on</strong>itor<strong>in</strong>g purposes <strong>on</strong>ly and<br />
will be used rout<strong>in</strong>ely start<strong>in</strong>g <strong>in</strong> 2012.<br />
2.2.1. NWP cloud test<br />
The NWP cloud test c<strong>on</strong>sists of compar<strong>in</strong>g the measured<br />
radiances <strong>in</strong> a selecti<strong>on</strong> of channels from IR atmospheric<br />
micro-w<strong>in</strong>dows with synthetic clear-sky TOA radiances calculatedforthesamescene.Theaprioriknowledgeofthe<br />
atmospheric state and of the surface temperature and w<strong>in</strong>ds<br />
comes from ECMWF forecasts. The surface emissivity is either<br />
computed with an analytical model for ocean surfaces <strong>after</strong><br />
Masuda [15] and Watts [16] or, for c<strong>on</strong>t<strong>in</strong>ental surfaces,<br />
retrieved from the static atlas described <strong>in</strong> Secti<strong>on</strong> 2.2. The<br />
synthetic clear-sky radiances are then computed with a fast<br />
radiative transfer model (FRTM). The FRTM was RT<str<strong>on</strong>g>IASI</str<strong>on</strong>g>-4 for<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF versi<strong>on</strong> 5.0 (and previous versi<strong>on</strong>s) but was<br />
replaced, with the release of versi<strong>on</strong> 5.2 <strong>in</strong> October 2011,<br />
with the state-of-the-art RTTOV-10. Both versi<strong>on</strong>s result from<br />
development of the FRTM implemented by Saunders et al.<br />
[17,18]. The reader is referred to the NWP-SAF product<br />
reports (www.research.metoffice.gov.uk/research/<strong>in</strong>terproj/<br />
nwpsaf/rtm/<strong>in</strong>dex.html) and to the ECMWF Technical Memo<br />
425 (2003) for a specific descripti<strong>on</strong> of the RTMs used <strong>in</strong><br />
operati<strong>on</strong>s.<br />
In this cloud test, if the observed and the synthetic<br />
radiances differ by more than a c<strong>on</strong>figurable threshold, then<br />
the scene is declared cloudy. In the current operati<strong>on</strong>al<br />
processor, this threshold is set to 1 K <strong>in</strong> brightness temperature.<br />
The threshold was determ<strong>in</strong>ed statistically with<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> observati<strong>on</strong>s. It accounts for the errors <strong>in</strong> the prior state,<br />
especially <strong>in</strong> the surface parameters. Studies are currently<br />
<strong>on</strong>go<strong>in</strong>g to calibrate this test. They <strong>in</strong>dicate that such a<br />
c<strong>on</strong>figurati<strong>on</strong> is more appropriate for sea than for land<br />
surfaces, where uncerta<strong>in</strong>ties <strong>in</strong> the surface temperature<br />
and emissivity sometimes translate <strong>in</strong>to errors much larger<br />
than 1 K <strong>in</strong> the computati<strong>on</strong> of synthetic brightness temperatures.<br />
As a c<strong>on</strong>sequence, it is believed that the residual<br />
cloud c<strong>on</strong>tam<strong>in</strong>ati<strong>on</strong> is higher for land surfaces than for<br />
oceans. The choice of the IR micro-w<strong>in</strong>dow channels is also<br />
c<strong>on</strong>figurable. The channels 751 (832.5 cm 1 ) and 1023<br />
(900.25 cm 1 ) are currently used <strong>in</strong> the PPF v5. On average,<br />
approximately 75% of the IFOVs tested with this method are<br />
declared cloudy, with regi<strong>on</strong>al variati<strong>on</strong>s.<br />
2.2.2. AVHRR cloud fracti<strong>on</strong><br />
The imager AVHRR, fly<strong>in</strong>g with <str<strong>on</strong>g>IASI</str<strong>on</strong>g>, provides measurements<br />
<strong>in</strong> both visible and <strong>in</strong>frared bands, from which the<br />
presence of a cloud is evaluated for each pixel. It has a<br />
resoluti<strong>on</strong> of 1 km at Nadir. At this resoluti<strong>on</strong>, the pixels<br />
are classified as either clear or cloudy <strong>in</strong> a b<strong>in</strong>ary fashi<strong>on</strong>.<br />
The determ<strong>in</strong>ati<strong>on</strong> of the cloud mask is performed <strong>in</strong> the<br />
AVHRR L1 process<strong>in</strong>g cha<strong>in</strong>. The detecti<strong>on</strong> algorithms<br />
<strong>in</strong>volve various tests based <strong>on</strong> <strong>in</strong>ter-channel brightness<br />
differences <strong>in</strong> the <strong>in</strong>frared, reflectances <strong>in</strong> the visible and<br />
near-<strong>in</strong>frared (NIR), and some NWP forecast data. They<br />
are more exhaustively listed and described <strong>in</strong> Secti<strong>on</strong><br />
5.4.4 of the ‘‘EPS Ground Segment AVHRR L1 Product<br />
Generati<strong>on</strong> Specificati<strong>on</strong>’’, EUM/EPS/SYS/SPE/990004<br />
(available <strong>on</strong>-l<strong>in</strong>e at: www.eumetsat.<strong>in</strong>t).<br />
The <strong>in</strong>tegrated AVHRR cloud fracti<strong>on</strong> (CFR) <strong>in</strong> this<br />
c<strong>on</strong>text is computed as the proporti<strong>on</strong> of AVHRR cloudy<br />
pixels with<strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> footpr<strong>in</strong>t, weighted by the <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
po<strong>in</strong>t spread functi<strong>on</strong>. This <strong>in</strong>tegrated AVHRR CFR has<br />
been rout<strong>in</strong>ely embedded <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L1C products s<strong>in</strong>ce<br />
18 May 2010 follow<strong>in</strong>g <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L1 Day-2 developments. A<br />
scene is declared cloudy if the amount of AVHRR cloudy<br />
pixels with<strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> IFOV exceeds a c<strong>on</strong>figurable threshold,<br />
set to 2% <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF versi<strong>on</strong> 5. On average,<br />
approximately 78% of the IFOVs tested with this method<br />
are declared cloudy. This AVHRR cloud test is systematically<br />
used <strong>in</strong> c<strong>on</strong>juncti<strong>on</strong> with the NWP cloud test and<br />
both methods agree <strong>in</strong> 90% of the cases. The use of<br />
different cloud tests <strong>in</strong> comb<strong>in</strong>ati<strong>on</strong> is specifically<br />
addressed <strong>in</strong> Secti<strong>on</strong> 2.2.5.<br />
2.2.3. Optical thickness test<br />
In this test, the atmospheric optical thickness is evaluated<br />
us<strong>in</strong>g the pr<strong>in</strong>cipal comp<strong>on</strong>ents of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> spectra<br />
as predictors <strong>in</strong> a l<strong>in</strong>ear regressi<strong>on</strong>. This retrieval bel<strong>on</strong>gs<br />
to an algorithms suite developed <strong>in</strong> [19] for retrieval of<br />
atmospheric and surface parameters <strong>in</strong> cloud-free as well<br />
as <strong>in</strong> partially cloudy c<strong>on</strong>diti<strong>on</strong>s. The result<strong>in</strong>g cloud<br />
optical thickness triggers the choice of regressi<strong>on</strong> coefficients<br />
tailored for the atmospheric and surface parameters<br />
retrieval under either clear or partially cloudy<br />
c<strong>on</strong>diti<strong>on</strong>s. This test was also <strong>in</strong>troduced with the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
PPF v5 and, with the current c<strong>on</strong>figurati<strong>on</strong>, is the least<br />
c<strong>on</strong>servative of the three operati<strong>on</strong>al tests with a global<br />
cloud detecti<strong>on</strong> rate between 40% and 45%.<br />
2.2.4. Stand-al<strong>on</strong>e ANN cloud test<br />
This test is a novel cloud detecti<strong>on</strong> method based <strong>on</strong><br />
ANNs which was developed <strong>in</strong> the frame of an external<br />
study. The pr<strong>in</strong>ciple and performance of this method are<br />
summarised <strong>in</strong> this secti<strong>on</strong>. The full descripti<strong>on</strong> of the<br />
methodology is accessible <strong>in</strong> the EUMETSAT study reports<br />
by Brockmann et al. (‘‘Technical Report for the study <strong>on</strong><br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g>/AVHRR Visual Scenes Analysis and Cloud Detecti<strong>on</strong>—<br />
IAVISA, Issue 3.0, Revisi<strong>on</strong> 4’’, 21 November 2008, 85 pp.<br />
and ‘‘<str<strong>on</strong>g>IASI</str<strong>on</strong>g>/AVHRR Cloud detecti<strong>on</strong> and Characterisati<strong>on</strong>,<br />
v1.1’’, 21 July 2010, 100 pp.).<br />
The ANNs described here are multi-layer perceptr<strong>on</strong>s<br />
(MLP) with two or three hidden layers. They can be seen<br />
as n<strong>on</strong>-l<strong>in</strong>ear regressi<strong>on</strong> operators between a set of<br />
predictors, stored <strong>in</strong> an <strong>in</strong>put layer, and a predictand, <strong>in</strong>
1344<br />
this case the required cloud<strong>in</strong>ess estimate. The <strong>in</strong>puts are<br />
composed of <str<strong>on</strong>g>IASI</str<strong>on</strong>g> radiances and AVHRR <strong>in</strong>formati<strong>on</strong>. The<br />
<strong>in</strong>put <str<strong>on</strong>g>IASI</str<strong>on</strong>g> radiances are sampled at selected wavelengths<br />
comm<strong>on</strong>ly used <strong>in</strong> cloud detecti<strong>on</strong> with IR data [20–22];<br />
they are listed <strong>in</strong> Table 2. The AVHRR <strong>in</strong>formati<strong>on</strong> c<strong>on</strong>sists<br />
of the means and standard deviati<strong>on</strong>s of the collocated<br />
radiance and reflectance clusters, which comprise the<br />
AVHRR radiance analyses present <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L1C products.<br />
The <strong>in</strong>puts are propagated through the networks,<br />
where successive l<strong>in</strong>ear comb<strong>in</strong>ati<strong>on</strong>s and c<strong>on</strong>voluti<strong>on</strong>s<br />
apply. The l<strong>in</strong>ear comb<strong>in</strong>ati<strong>on</strong>s <strong>in</strong>volve different sets of<br />
weights and offsets. The c<strong>on</strong>voluti<strong>on</strong> (activati<strong>on</strong>) functi<strong>on</strong><br />
is <strong>in</strong> this case the logistic functi<strong>on</strong> f(x)¼1/(1þe x), which<br />
gives their n<strong>on</strong>-l<strong>in</strong>ear properties to the nets. The weights<br />
and offsets used to c<strong>on</strong>figure the networks are iteratively<br />
adjusted dur<strong>in</strong>g a so-called tra<strong>in</strong><strong>in</strong>g phase, where pairs of<br />
<strong>in</strong>put vectors (<str<strong>on</strong>g>IASI</str<strong>on</strong>g> and AVHRR observati<strong>on</strong>s) and their<br />
associated outputs (IFOV cloud<strong>in</strong>ess) are successively<br />
presented to the network. The tra<strong>in</strong><strong>in</strong>g c<strong>on</strong>sists of<br />
Table 2<br />
List of <str<strong>on</strong>g>IASI</str<strong>on</strong>g> channels used <strong>in</strong> the ANN cloud detecti<strong>on</strong>.<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
Channels Wavenumber (cm 1 ) Channels Wavenumber (cm 1 )<br />
142 680.250 2733 1328.00<br />
546 781.250 3391 1492.50<br />
751 832.500 3770 1587.25<br />
754 833.250 3790 1592.25<br />
1023 900.500 6390 2242.25<br />
1059 909.500 6512 2272.75<br />
1282 965.250 7391 2492.50<br />
2019 1149.50 7547 2531.50<br />
2616 1298.75 8232 2702.75<br />
m<strong>in</strong>imis<strong>in</strong>g the errors between the retrieved output and<br />
the teach<strong>in</strong>g target (here the cloud<strong>in</strong>ess). This is achieved<br />
by simple back-propagati<strong>on</strong> of the error gradient [23].<br />
To c<strong>on</strong>stitute the teach<strong>in</strong>g patterns, the cloud<strong>in</strong>ess <strong>in</strong><br />
24923 <str<strong>on</strong>g>IASI</str<strong>on</strong>g> IFOVs was assessed by visual <strong>in</strong>specti<strong>on</strong> of the<br />
collocated AVHRR images and classified <strong>in</strong>to four qualitative<br />
categories: clear-sky, low cloud fracti<strong>on</strong>, high cloud<br />
fracti<strong>on</strong> and full cloud coverage. Together with the corresp<strong>on</strong>d<strong>in</strong>g<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> and AVHRR observati<strong>on</strong>s, they will be<br />
referred to as the IAVISA database. These IFOVs were<br />
randomly selected to provide a full Earth coverage (see<br />
Fig. 2) and an equal representati<strong>on</strong> of day/night, land/sea,<br />
seas<strong>on</strong>al and geographical/climatological c<strong>on</strong>figurati<strong>on</strong>s.<br />
Dedicated artificial neural networks were tra<strong>in</strong>ed for each<br />
of the day/night and land/sea comb<strong>in</strong>ati<strong>on</strong>s. The IAVISA<br />
database was split <strong>in</strong>to a tra<strong>in</strong><strong>in</strong>g subset and a verificati<strong>on</strong><br />
subset (not used for tra<strong>in</strong><strong>in</strong>g), to evaluate the generalisati<strong>on</strong><br />
skill of the networks (i.e. ensure that they are not<br />
too specific to the tra<strong>in</strong><strong>in</strong>g data).<br />
The performances of the NWP and AVHRR cloud tests<br />
(see Secti<strong>on</strong>s 2.2.1 and 2.2.2) with these visually classified<br />
scenes were assessed and compared to the performances<br />
of the ANN cloud test. They are summarised <strong>in</strong> the<br />
c<strong>on</strong>t<strong>in</strong>gency Table 3. The global ability to correctly classify<br />
clear-sky and the capacity to screen out the cloud c<strong>on</strong>tam<strong>in</strong>ated<br />
IFOVs are <strong>in</strong>creased by approximately 25% with<br />
the ANN method <strong>in</strong> comparis<strong>on</strong> to the NWP test, with a<br />
success rate exceed<strong>in</strong>g 90%. The respective scores have<br />
also been studied <strong>in</strong>dividually for different surface types<br />
and day/night c<strong>on</strong>figurati<strong>on</strong>s. A more exhaustive performance<br />
analysis can be found <strong>in</strong> the EUMETSAT validati<strong>on</strong><br />
report EUM/MET/TEN/10/0343. The ANN test is usually<br />
more str<strong>in</strong>gent than the other tests, with better success<br />
rates for the identificati<strong>on</strong> of cloud-free IFOVs.<br />
Fig. 2. Distributi<strong>on</strong> of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> IFOVs <strong>in</strong> the IAVISA database used for tra<strong>in</strong><strong>in</strong>g the ANN cloud detecti<strong>on</strong>.
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1345<br />
2.2.5. Cloud tests comb<strong>in</strong>ati<strong>on</strong> and impact <strong>on</strong> <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
product quality<br />
Currently, all operati<strong>on</strong>al tests (NWP, AVHRR and<br />
optical thickness) are systematically applied. An IFOV is<br />
flagged cloudy if at least <strong>on</strong>e of these tests detects the<br />
presence of a cloud. While this c<strong>on</strong>servative approach<br />
<strong>in</strong>creases the c<strong>on</strong>fidence that the rema<strong>in</strong><strong>in</strong>g clear IFOVs<br />
Table 3<br />
C<strong>on</strong>t<strong>in</strong>gency table of the true (classified by visual <strong>in</strong>specti<strong>on</strong>) vs. retrieved<br />
cloud<strong>in</strong>ess with the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> NWP and ANN cloud tests. The numbers <strong>in</strong>dicate<br />
the size of the subsamples classified by the respective tests.<br />
NWP test ANN test<br />
Clear Cloudy Clear Cloudy<br />
Visual classificati<strong>on</strong> Clear 4803 2456 6708 551<br />
Cloudy 4074 13,590 1363 16,301<br />
are <strong>in</strong>deed cloud-free it has the disadvantage of discard<strong>in</strong>g<br />
many pixels with false cloud detecti<strong>on</strong> from the<br />
subsequent atmospheric parameters <strong>retrievals</strong>. Approximately<br />
10–15% of the IFOVs are declared cloud-free with<br />
the current NWP and AVHRR tests but <strong>on</strong>ly about 5–8%<br />
would rema<strong>in</strong> if the n<strong>on</strong>-l<strong>in</strong>ear cloud test were simply<br />
used <strong>in</strong> additi<strong>on</strong> to the other two.<br />
Fig. 3 shows the agreement/disagreement rate between<br />
three cloud detecti<strong>on</strong> methods, computed for the period<br />
19–24 March 2010. It illustrates that simply c<strong>on</strong>sider<strong>in</strong>g<br />
the <strong>in</strong>tersecti<strong>on</strong> of the cloud tests repeatedly reject IFOVs<br />
<strong>in</strong> regi<strong>on</strong>s with a specific climatology or soil type. For<br />
<strong>in</strong>stance sea ice, c<strong>on</strong>t<strong>in</strong>ental snow cover and Polar Regi<strong>on</strong>s<br />
are typical areas where the cloud detecti<strong>on</strong>, and especially<br />
with the AVHRR and NWP tests, is less reliable. The<br />
reflectance properties of such surfaces are similar to those<br />
of clouds <strong>in</strong> the AVHRR visible channels and often c<strong>on</strong>fuse<br />
the albedo test. Visual <strong>in</strong>specti<strong>on</strong> c<strong>on</strong>firmed that this test<br />
Fig. 3. Average agreement rates between the ANN cloud detecti<strong>on</strong>s and the NWP tests (left), and AVHRR tests (right) dur<strong>in</strong>g the period 19–24 March<br />
2010. In red: the methods agree, <strong>in</strong> green: cloud tests disagree with clear-sky as per ANN test <strong>on</strong>ly, <strong>in</strong> blue: cloud tests disagree with cloud-free as per<br />
NWP test (top) or AVHRR (bottom) <strong>on</strong>ly. (For <strong>in</strong>terpretati<strong>on</strong> of the references to colour <strong>in</strong> this figure legend, the reader is referred to the web versi<strong>on</strong> of<br />
this article.)
1346<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
usually overestimates the presence of clouds. The NWP test<br />
requires an accurate descripti<strong>on</strong> of the atmospheric temperature<br />
and humidity, which the forecast fields are less<br />
able to provide at high latitudes, as well as an accurate<br />
knowledge of the surface temperature and emissivity. At a<br />
surface temperature of 270 K, the cloud detecti<strong>on</strong> threshold<br />
of 1 K described <strong>in</strong> Secti<strong>on</strong> 2.2.1 corresp<strong>on</strong>ds to an error<br />
smaller than 2% <strong>in</strong> the surface emissivity at the wavelengths<br />
c<strong>on</strong>sidered for this test. Measurements of natural<br />
materials [24] however show that the natural variability of<br />
the surface emissivity for water, ice and snow covers is as<br />
high as 5% <strong>in</strong> this spectral regi<strong>on</strong> and depend <strong>on</strong> the ice and<br />
snow type, density and moisture. This variability is not<br />
necessarily captured by the static emissivity database.<br />
Furthermore, the land/sea classificati<strong>on</strong> is so far still a<br />
static process with the c<strong>on</strong>sequence that the IFOVs covered<br />
by sea-ice are still treated as open oceans. The current<br />
developments <strong>in</strong> this area <strong>in</strong>volve the synergistic use of the<br />
microwave measurements from the AMSU compani<strong>on</strong><br />
<strong>in</strong>strument <strong>on</strong>-board <strong>Metop</strong> to dist<strong>in</strong>guish pla<strong>in</strong> water<br />
and sea ice.<br />
Similarly, clouds are systematically <strong>in</strong>correctly reported<br />
<strong>in</strong> the Ganges and Indus crop land valleys by the NWP test.<br />
C<strong>on</strong>versely, the AVHRR and NWP tests statistically report<br />
more cloud-free IFOVs <strong>in</strong> the <strong>in</strong>ter-tropical belt over the<br />
oceans than the ANN test. Visual <strong>in</strong>specti<strong>on</strong>s, as illustrated<br />
by Fig. 4, c<strong>on</strong>firm the actual presence of clouds reported by<br />
the ANN classificati<strong>on</strong> when this method disagrees with<br />
the other two. However, the n<strong>on</strong>-l<strong>in</strong>ear test was not<br />
tra<strong>in</strong>ed to detect dust clouds, which the NWP and especially<br />
the AVHRR tests usually capture more efficiently.<br />
This can be seen <strong>in</strong> the sub-Saharan belt where the AVHRR<br />
images c<strong>on</strong>firm a large dust storm event which was not<br />
detected by the ANN test but is correctly reported by the<br />
AVHRR <strong>in</strong>tegrated cloud <strong>in</strong>formati<strong>on</strong>.<br />
The identificati<strong>on</strong> of clear-sky is important as the retrieval<br />
algorithms def<strong>in</strong>ed <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 science plan are<br />
nom<strong>in</strong>ally tailored for cloud-free radiances. The impact of<br />
cloud c<strong>on</strong>tam<strong>in</strong>ati<strong>on</strong>, and therefore the ability to tolerate it,<br />
does however vary with the atmospheric parameter <strong>in</strong><br />
questi<strong>on</strong>, the retrieval method and the required product<br />
accuracy def<strong>in</strong>ed <strong>in</strong> the EPS End User Requirements Document<br />
(reference EUM/EPS/MIS/REQ/93/001 available <strong>on</strong>-l<strong>in</strong>e<br />
at: www.eumetsat.<strong>in</strong>t). Clerbaux et al. [25], for <strong>in</strong>stance,<br />
retrieve CO c<strong>on</strong>centrati<strong>on</strong>s from <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements with<br />
potential IFOV cloud coverage of up to 25% us<strong>in</strong>g a retrieval<br />
scheme designed for clear radiances. While this is c<strong>on</strong>sistent<br />
with the expected accuracy of 10% for the CO total column,<br />
such high cloud fracti<strong>on</strong>s are not compatible with the<br />
accuracy requirement of 0.4 K for clear-sky sea surface<br />
temperature <strong>retrievals</strong>. In subsequent secti<strong>on</strong>s we illustrate<br />
the impact of the three cloud tests discussed <strong>in</strong> this secti<strong>on</strong><br />
<strong>on</strong> the SST product yield and quality.<br />
The reference SST product, from the Advanced Al<strong>on</strong>g<br />
Track Scann<strong>in</strong>g Radiometer (AATSR) <strong>in</strong>strument <strong>on</strong>-board<br />
ENVISAT [26], the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 SST retrieval and the <strong>in</strong>tercomparis<strong>on</strong><br />
protocol are described <strong>in</strong> Secti<strong>on</strong> 2.5.3. The<br />
<strong>in</strong>ter-comparis<strong>on</strong>s are performed dur<strong>in</strong>g the period 19–24<br />
March 2010 where <str<strong>on</strong>g>IASI</str<strong>on</strong>g> and AATSR acquisiti<strong>on</strong>s over the<br />
same area were taken with<strong>in</strong> 20 m<strong>in</strong> of <strong>on</strong>e another.<br />
Fig. 5a–c shows maps of the SST departures (AATSR–<str<strong>on</strong>g>IASI</str<strong>on</strong>g>)<br />
<strong>after</strong> cloud filter<strong>in</strong>g with the AVHRR, the NWP and the<br />
ANN test respectively. Fig. 5e–g shows the corresp<strong>on</strong>d<strong>in</strong>g<br />
distributi<strong>on</strong>s of the AATSR–<str<strong>on</strong>g>IASI</str<strong>on</strong>g> SST differences. All three<br />
distributi<strong>on</strong>s have a ma<strong>in</strong> Gaussian mode with an asymmetric<br />
tail c<strong>on</strong>ta<strong>in</strong><strong>in</strong>g less than 10% of the sample where<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> SST are colder by up to 2 K than the AATSR SST. This<br />
tail is the result of undetected cloud and dust c<strong>on</strong>tam<strong>in</strong>ati<strong>on</strong>.<br />
While the overall statistics are similar for all three<br />
cloud-tests (positive bias and standard deviati<strong>on</strong> of<br />
approximately 0.47 K) the central mode is narrower with<br />
the ANN than with the NWP and the AVHRR cloud<br />
filter<strong>in</strong>g (a standard deviati<strong>on</strong> of 0.25 versus 0.30 K,<br />
respectively). Furthermore, the visual <strong>in</strong>specti<strong>on</strong> of<br />
AVHRR images and MODIS aerosol optical depth products<br />
[27] (as shown <strong>in</strong> Fig. 6 and available at: http://disc.sci.<br />
gsfc.nasa.gov/giovanni/overview/<strong>in</strong>dex.html, last accessed<br />
19/01/2012) c<strong>on</strong>firmed the presence of dust aerosols off<br />
the West African coast, <strong>in</strong> the Arabian Sea and off the East<br />
Asian coast. The dust patterns correlate with the larger<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> and AATSR SST discrepancies. The yield with the ANN<br />
test is also 30% smaller. Outside these three dust c<strong>on</strong>tam<strong>in</strong>ated<br />
areas, identified us<strong>in</strong>g the MODIS aerosol<br />
product, the AATSR–<str<strong>on</strong>g>IASI</str<strong>on</strong>g> SST departure <strong>after</strong> the ANN<br />
cloud filter<strong>in</strong>g is nearly Gaussian, with a bias and standard<br />
deviati<strong>on</strong> of þ0.25 and 0.22 K, respectively. This c<strong>on</strong>firms<br />
Fig. 4. Cloud detecti<strong>on</strong>s <strong>in</strong> <str<strong>on</strong>g>IASI</str<strong>on</strong>g> IFOVs over central Pacific Ocean <strong>in</strong> March 2010 (black: clear, white: cloudy). From left to right: ANN cloud test, ‘NWP<br />
cloud test’, AVHRR CFR test, over an AVHRR image composite (channels 3 and 4). The clouds show up <strong>in</strong> p<strong>in</strong>k/red and the ocean surface <strong>in</strong> cyan. (For<br />
<strong>in</strong>terpretati<strong>on</strong> of the references to colour <strong>in</strong> this figure legend, the reader is referred to the web versi<strong>on</strong> of this article.)
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1347<br />
Fig. 5. Impact of the cloud tests <strong>on</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 SST yield and quality as compared to AATSR L2P SST.
1348<br />
the superior skill of the ANN cloud test for clear-sky<br />
identificati<strong>on</strong> (see Fig. 5d and h) over oceans.<br />
Work is still <strong>on</strong>go<strong>in</strong>g <strong>in</strong> the area of dust detecti<strong>on</strong> and<br />
<strong>on</strong> the comb<strong>in</strong>ed use of the various cloud detecti<strong>on</strong><br />
schemes with the objective of br<strong>in</strong>g<strong>in</strong>g the ANN tests <strong>in</strong>to<br />
operati<strong>on</strong>s <strong>in</strong> 2012. The ultimate c<strong>on</strong>figurati<strong>on</strong> will maximise<br />
the quantity and the quality of the f<strong>in</strong>al <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
<strong>retrievals</strong> available.<br />
2.3. Cloud characterisati<strong>on</strong><br />
If a cloud is detected <strong>in</strong> an <str<strong>on</strong>g>IASI</str<strong>on</strong>g> field of view, its<br />
characterisati<strong>on</strong> <strong>in</strong> terms of cloud microphysics, fracti<strong>on</strong>al<br />
coverage and cloud top pressure (CTP) is performed. In<br />
previous versi<strong>on</strong>s of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF, the cloud fracti<strong>on</strong> and<br />
cloud top height determ<strong>in</strong>ed with AVHRR were comb<strong>in</strong>ed<br />
with the <strong>retrievals</strong> made from <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements with the<br />
CO2-slic<strong>in</strong>g method (see Secti<strong>on</strong> 2.3.2 below). Because of<br />
their different characteristics this was modified and <strong>on</strong>ly<br />
the CO2-slic<strong>in</strong>g <strong>retrievals</strong> are reta<strong>in</strong>ed <strong>in</strong> the L2 products<br />
from versi<strong>on</strong> 5.0. Assessment of cloud<strong>in</strong>ess with<strong>in</strong> each<br />
AVHRR pixel is <strong>in</strong>deed b<strong>in</strong>ary: either the pixel is clear or it<br />
is cloudy, regardless of the fracti<strong>on</strong>al cover with<strong>in</strong> the pixel<br />
and of the cloud optical depth, whereas the CO2-slic<strong>in</strong>g<br />
returns the effective cloud amount (ECA). The latter takes<br />
<strong>in</strong>to account the cloud transparency and is more c<strong>on</strong>sistent<br />
with the radiative transfer <strong>in</strong> the atmosphere.<br />
2.3.1. Cloud phase retrieval<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
Fig. 6. MODIS aerosol optical depth at 550 nm for the period 19–24 March 2010, http://disc.sci.gsfc.nasa.gov/giovanni/overview/<strong>in</strong>dex.html [27].<br />
2.3.1.1. Methodology and algorithm descripti<strong>on</strong>. The <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
PPF c<strong>on</strong>ta<strong>in</strong>s a cloud-phase-determ<strong>in</strong>ati<strong>on</strong> method that is<br />
based <strong>on</strong> a published algorithm described <strong>in</strong> the MODIS<br />
ATBD <strong>on</strong> ‘‘Cloud top properties and cloud phase, v5’’ [81],<br />
which is <strong>in</strong> turn based <strong>on</strong> the algorithm described <strong>in</strong> [28].<br />
We describe here modificati<strong>on</strong>s to and validati<strong>on</strong> of the<br />
algorithm. It was further compared aga<strong>in</strong>st a method<br />
described <strong>in</strong> [29].<br />
The methodology is based <strong>on</strong> the different spectral<br />
emissivity of water and ice clouds <strong>in</strong> the spectral regi<strong>on</strong><br />
between 8 and 12 mm. In case of clouds, the slope of the<br />
emissi<strong>on</strong> changes, so that the difference DT1¼Tb(8 mm)<br />
Tb(11 mm) <strong>in</strong> comparis<strong>on</strong> to a sec<strong>on</strong>d difference DT2¼<br />
Tb(11 mm) Tb(12 mm) varies depend<strong>in</strong>g <strong>on</strong> the cloud<br />
phase. The corresp<strong>on</strong>d<strong>in</strong>g test until <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF v4 reads:<br />
If DT1 DT2Zt1 then the cloud phase is ice,<br />
else if t14DT1 DT2Zt2 then the cloud phase<br />
is mixed,<br />
else the cloud phase is liquid. The thresholds t1 and t2 as<br />
well as the exact wavelengths with<strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> spectrum<br />
are c<strong>on</strong>figurable and can be optimised <strong>on</strong> the basis of<br />
<strong>in</strong>dependent knowledge.<br />
The algorithm implemented <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF V5.0<br />
c<strong>on</strong>sists of that described above, plus an additi<strong>on</strong>al test. A<br />
third threshold can be set to identify ice clouds based <strong>on</strong><br />
the fact that super-cooled water cannot exist at temperatures<br />
below t3¼ 40 1C; i.e. at temperatures lower than<br />
t3 the cloud phase must be ice. This threshold must be set<br />
us<strong>in</strong>g a w<strong>in</strong>dow channel near 11 mm, which is close to the<br />
cloud-top temperature <strong>in</strong> case of opaque clouds. In the<br />
case of semi-transparent clouds this test will <strong>on</strong>ly predict<br />
ice if the cloud-top temperature is much lower than t3,<br />
and therefore also provides correct results.<br />
2.3.1.2. An alternative method. Another cloud-phase test,<br />
published by Wei et al. [29], has been implemented as a<br />
prototype for comparis<strong>on</strong> with the algorithms described<br />
<strong>in</strong> Secti<strong>on</strong> 2.3.1.1. This method was optimised for the use<br />
with AIRS data. Ice clouds are detected<br />
if Tb(900 cm 1 )o238 K<br />
or Tb(1231 cm 1 ) Tb(900 cm 1 )40.5 K<br />
or Tb(1231 cm 1 ) Tb(900 cm 1 )4 0.5 K and<br />
Tb(900 cm 1)4258 K.
Otherwise, liquid phase is assumed. No provisi<strong>on</strong> is<br />
made for mixed-phase clouds.<br />
2.3.1.3. Reference data set and validati<strong>on</strong> results. For the<br />
validati<strong>on</strong> of cloud-phase determ<strong>in</strong>ati<strong>on</strong> and for the<br />
def<strong>in</strong>iti<strong>on</strong> of new thresholds (tra<strong>in</strong><strong>in</strong>g), a set of 672<br />
globally distributed co-located <str<strong>on</strong>g>IASI</str<strong>on</strong>g> and AVHRR data has<br />
been compiled. The data set c<strong>on</strong>sists of day-time scenes<br />
<strong>on</strong>ly, allow<strong>in</strong>g proper identificati<strong>on</strong> of ice clouds <strong>in</strong> the<br />
multi-spectral AVHRR imagery, us<strong>in</strong>g measurements of<br />
the reflected solar radiati<strong>on</strong> <strong>in</strong> the visible and near<strong>in</strong>frared<br />
spectral regi<strong>on</strong>s. In particular the use of the<br />
1.6 mm channel allows proper identificati<strong>on</strong> of ice clouds<br />
due to their low reflectivity <strong>in</strong> this channel as compared<br />
to the short-wave channels. The visual <strong>in</strong>specti<strong>on</strong> of RGB<br />
pseudo-colour images from AVHRR channels 1, 2, and 3a<br />
allows the discrim<strong>in</strong>ati<strong>on</strong> of ice and liquid phases. Mixed<br />
phases are not however easily identified.<br />
The data set composed of 672 AVHRR scenes and<br />
associated <str<strong>on</strong>g>IASI</str<strong>on</strong>g> samples has been collected and visually<br />
<strong>in</strong>terpreted with respect to cloud phase. It <strong>in</strong>cludes 371<br />
samples with ice clouds, 228 samples with liquid phase,<br />
and 73 samples with mixed phase. A separati<strong>on</strong> <strong>in</strong>to two<br />
subsets allows for tra<strong>in</strong><strong>in</strong>g and validati<strong>on</strong>.<br />
Us<strong>in</strong>g the orig<strong>in</strong>al coefficients implemented <strong>in</strong> <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
PPF V4.3.2, 79.0% of the cloud phases are determ<strong>in</strong>ed<br />
correctly. The method <strong>in</strong> [29] provides 78.6% correctly<br />
determ<strong>in</strong>ed cloud phases. Subsequently, the algorithm<br />
has been amended as described <strong>in</strong> Secti<strong>on</strong> 2.3.1.1. The<br />
spectral sample positi<strong>on</strong>s as well as the thresholds have<br />
been modified to enhance the performance. The f<strong>in</strong>al<br />
versi<strong>on</strong> of the tuned algorithm detects 84.5% of the cloud<br />
phases correctly. Am<strong>on</strong>g the ice samples 97.3% are correctly<br />
determ<strong>in</strong>ed, while of the liquid samples 84.6% are<br />
determ<strong>in</strong>ed correctly. The algorithm does not have any<br />
skill <strong>in</strong> detect<strong>in</strong>g mixed-phase clouds, <strong>on</strong>ly 5.5% are<br />
correctly identified, the majority of mixed-phase clouds<br />
are reproduced as ice clouds. The reas<strong>on</strong> for this is the<br />
uncerta<strong>in</strong> visual determ<strong>in</strong>ati<strong>on</strong> of mixed-phase clouds. In<br />
practice, however, mixed phase clouds will always be<br />
detected as either ice or liquid clouds depend<strong>in</strong>g <strong>on</strong> the<br />
relative proporti<strong>on</strong>s.<br />
The algorithm described <strong>in</strong> Secti<strong>on</strong> 2.3.1.1 requires the<br />
follow<strong>in</strong>g brightness temperatures and is c<strong>on</strong>figured with<br />
the follow<strong>in</strong>g thresholds:<br />
DT1 ¼ Tbð1209:75cm 1 Þ2Tbð900:25cm 1 Þ<br />
DT2 ¼ Tbð900:25cm 1 Þ2Tbð829:00cm 1 Þ<br />
t1 ¼ 0:9K<br />
t2 ¼ 1:15K<br />
t3 ¼ 233:15K:<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1349<br />
2.3.2. The CO 2-slic<strong>in</strong>g method<br />
The cloud top pressure (CTP), and subsequently the<br />
cloud top temperature (CTT) obta<strong>in</strong>ed us<strong>in</strong>g a reference<br />
temperature profile, together with the cloud fracti<strong>on</strong><br />
(CFR) are additi<strong>on</strong>al parameters provided <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> level<br />
2 cloud products. They are currently retrieved with the<br />
CO2-slic<strong>in</strong>g method <strong>in</strong> the operati<strong>on</strong>al processor. Validati<strong>on</strong><br />
results are presented and discussed here.<br />
The CO2-slic<strong>in</strong>g method established by Menzel et al.<br />
[30] and Smith and Frey [31] for high resoluti<strong>on</strong> <strong>in</strong>terferometer<br />
sounders was adapted to <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>after</strong> Arriaga, 2007<br />
(EUMETSAT Technical Report: ‘‘CO2 Slic<strong>in</strong>g Algorithm for<br />
the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 Product Process<strong>in</strong>g Facility’’, EUM/MET/REP/07/<br />
0305). It was implemented <strong>in</strong> the first versi<strong>on</strong> of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
PPF to rout<strong>in</strong>ely retrieve the cloud top height and equivalent<br />
cloud amount with<strong>in</strong> a <str<strong>on</strong>g>IASI</str<strong>on</strong>g> field of view. In this<br />
implementati<strong>on</strong>, the observati<strong>on</strong>s <strong>in</strong> CO 2-channels sensitive<br />
to different pressure levels are associated with their<br />
clear-sky synthetic radiances counterparts. The forward<br />
computati<strong>on</strong> is performed with RTTOV us<strong>in</strong>g as <strong>in</strong>put the<br />
atmospheric profile forecast from ECMWF. A reference<br />
channel (currently at 796.75 cm 1 ) and 41 channels<br />
selected <strong>in</strong> the CO 2-band between 707.5 and 756 cm 1 ,<br />
c<strong>on</strong>figure the cloud-top retrieval. The equivalent cloud<br />
amount is estimated us<strong>in</strong>g a w<strong>in</strong>dow channel<br />
(900.50 cm 1 ) and the retrieved CTP. When a low cloud<br />
is suspected, the collocated forecast model temperature<br />
profile is <strong>in</strong>spected for a potential <strong>in</strong>versi<strong>on</strong>. If a temperature<br />
<strong>in</strong>versi<strong>on</strong> is found, the cloud top pressure is placed<br />
below the base of the <strong>in</strong>versi<strong>on</strong>. Retrievals are attempted<br />
<strong>on</strong>ly if the c<strong>on</strong>trast between observed and simulated clear<br />
radiances is larger than the <strong>in</strong>strument noise. In additi<strong>on</strong>,<br />
the algorithm also returns a quality <strong>in</strong>dicator flag and<br />
<strong>retrievals</strong> with ECA lower than 10% are discarded as they<br />
are usually associated with too large uncerta<strong>in</strong>ties.<br />
2.3.2.1. Validati<strong>on</strong> of the cloud fracti<strong>on</strong>. The use of a simple<br />
digital camera equipped with a fish-eye lens <strong>in</strong> support of<br />
cloud detecti<strong>on</strong> and m<strong>on</strong>itor<strong>in</strong>g has developed <strong>in</strong> recent<br />
<strong>years</strong> [32]. The validati<strong>on</strong> of the retrieved cloud fracti<strong>on</strong> as<br />
compared to such whole sky imagery was performed <strong>in</strong><br />
2008 dur<strong>in</strong>g the dedicated EPS validati<strong>on</strong> campaign which<br />
took place from June to September 2007 at the meteorological<br />
stati<strong>on</strong> of L<strong>in</strong>denberg, Germany. A Daylight VIS/NIR<br />
Whole Sky Imager (WSI) manufactured at the University of<br />
California San Diego has been operat<strong>in</strong>g <strong>in</strong> L<strong>in</strong>denberg<br />
s<strong>in</strong>ce 2003. The <strong>in</strong>strument acquires images of the upper<br />
hemisphere with a view<strong>in</strong>g angle of 1801 <strong>in</strong> different<br />
spectral regi<strong>on</strong>s at time steps of 5–10 m<strong>in</strong> between<br />
sunrise and sunset. Calibrated and corrected basic<br />
image data are stored. Automatic post process<strong>in</strong>g<br />
algorithms provide cloud cover and cloud distributi<strong>on</strong><br />
of both optically th<strong>in</strong> and thick clouds. More details <strong>on</strong><br />
the <strong>in</strong>strument and <strong>on</strong> the results of measurements<br />
<strong>in</strong>clud<strong>in</strong>g comparis<strong>on</strong>s of WSI data with c<strong>on</strong>venti<strong>on</strong>al<br />
cloud observati<strong>on</strong>s are given by Feister and Shields [33].<br />
Ground-based measurements with<strong>in</strong> 10 m<strong>in</strong> of the<br />
overpass time were matched to the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> IFOVs with<strong>in</strong> 11<br />
of L<strong>in</strong>denberg. The associated cloud fracti<strong>on</strong>s are shown <strong>in</strong><br />
Fig. 7, where <strong>on</strong>ly scenes with <strong>on</strong>e cloud layer are shown.<br />
Horiz<strong>on</strong>tal error bars show the variability of the cloud<br />
fracti<strong>on</strong> with<strong>in</strong> 10 m<strong>in</strong> of the overpass time while the<br />
vertical error bars show the spatial variability (with<strong>in</strong><br />
11 of the L<strong>in</strong>denberg site) as seen by <str<strong>on</strong>g>IASI</str<strong>on</strong>g> with the<br />
CO2-slic<strong>in</strong>g algorithm. The area encompassed by the WSI<br />
does not exactly co<strong>in</strong>cide <strong>in</strong> size and locati<strong>on</strong> with the<br />
collocated <str<strong>on</strong>g>IASI</str<strong>on</strong>g> foot-pr<strong>in</strong>ts, as illustrated by Fig. 8, which
1350<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
Fig. 7. Comparis<strong>on</strong> of cloud fracti<strong>on</strong> as measured from the L<strong>in</strong>denberg<br />
Whole Sky Imager (WSI) and the <strong>on</strong>e from the CO 2 slic<strong>in</strong>g algorithm <strong>in</strong><br />
the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF.<br />
Fig. 8. C<strong>on</strong>vective cells near L<strong>in</strong>dcolourenberg (red square) at overpass<br />
20070620190941. The cloud cover from surface observati<strong>on</strong>s is 87%. The<br />
retrieved cloud fracti<strong>on</strong> with<strong>in</strong> <str<strong>on</strong>g>IASI</str<strong>on</strong>g> IFOVs (ellipses) is overlayed <strong>on</strong> the<br />
AVHRR 10.8 mm image. (For <strong>in</strong>terpretati<strong>on</strong> of the references to colour <strong>in</strong><br />
this figure legend, the reader is referred to the web versi<strong>on</strong> of this<br />
article.)<br />
expla<strong>in</strong>s the size of the errors bars and prevents a full<br />
quantitative validati<strong>on</strong>. The <strong>in</strong>ter-comparis<strong>on</strong> shows a<br />
good general agreement between the two products and<br />
visual <strong>in</strong>specti<strong>on</strong>s of the retrieved cloud coverage with<br />
AVHRR images c<strong>on</strong>firmed the quality of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 ECA<br />
(see an example <strong>in</strong> Fig. 9). The retrieved equivalent cloud<br />
amount was recently validated aga<strong>in</strong>st the visual cloud<strong>in</strong>ess<br />
classificati<strong>on</strong> <strong>in</strong> the IAVISA database (presented <strong>in</strong><br />
Secti<strong>on</strong> 2.2.4). It c<strong>on</strong>firmed the general good quality of the<br />
CO 2-slic<strong>in</strong>g results but also dem<strong>on</strong>strated the limitati<strong>on</strong>s<br />
of the method for low-c<strong>on</strong>trast scenes, i.e. when the cloud<br />
fracti<strong>on</strong> is smaller than 20% and/or when the cloud top<br />
temperature is too close to the surface temperature, as<br />
Fig. 9. Retrieved cloud fracti<strong>on</strong> with<strong>in</strong> <str<strong>on</strong>g>IASI</str<strong>on</strong>g> IFOVs over a cold fr<strong>on</strong>t <strong>in</strong> the<br />
North-West Atlantic imaged at 0.6 mm with AVHRR.<br />
happens with low-level clouds. For these c<strong>on</strong>figurati<strong>on</strong>s, a<br />
complementary method has been developed for the <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
L2 PPF with the ‘w 2 method’ (see Secti<strong>on</strong> 2.3.3 below).<br />
Furthermore, the current implementati<strong>on</strong> of the CO 2slic<strong>in</strong>g<br />
method cannot cope with multi-layer clouds<br />
with<strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> field of view and the current developments<br />
aim to use the AVHRR cloud <strong>in</strong>formati<strong>on</strong> based <strong>on</strong><br />
the cluster analysis for these situati<strong>on</strong>s.<br />
2.3.2.2. Validati<strong>on</strong> of the cloud top pressure with L<strong>in</strong>denberg<br />
cloud radar. The retrieved cloud top pressure was also<br />
validated <strong>in</strong> 2008 aga<strong>in</strong>st ground-based cloud-radar<br />
operated at the meteorological stati<strong>on</strong> of L<strong>in</strong>denberg,<br />
Germany, dur<strong>in</strong>g an atmospheric sound<strong>in</strong>g campaign<br />
from June to September 2007 <strong>in</strong> support of the<br />
validati<strong>on</strong> of <strong>Metop</strong> products. It is acknowledged that<br />
this validati<strong>on</strong> work is so far limited to this specific site.<br />
However, the campaign <strong>in</strong>volved a large variety of cloud<br />
types reported by an expert observer (Ac, Cb, Cc, Cb calvus,<br />
Cg, Cs, Cu, Ci, Sc, St), al<strong>on</strong>e or <strong>in</strong> comb<strong>in</strong>ati<strong>on</strong>, and a large<br />
range of cloud elevati<strong>on</strong>s between 950 and 200 hPa. The<br />
full analyses and validati<strong>on</strong> results of the retrieved cloud<br />
fracti<strong>on</strong> and cloud top pressure are available <strong>in</strong> the<br />
validati<strong>on</strong> report by Arriaga et al. ‘‘EPS <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF<br />
Validati<strong>on</strong> Report: Cloud Top Pressure and Effective<br />
Cloud Amount’’ (EUM/MET/REP/09/0698).<br />
Cloud radars measure profiles of the <strong>in</strong>tensity of<br />
particle-backscattered signals and their Doppler shift to<br />
derive <strong>in</strong>formati<strong>on</strong> about particle size, c<strong>on</strong>centrati<strong>on</strong> and<br />
moti<strong>on</strong>. The cloud radar used <strong>in</strong> this study is a vertically<br />
po<strong>in</strong>t<strong>in</strong>g Doppler radar measur<strong>in</strong>g at 35.5 GHz, manufactured<br />
by Metek GmbH, Elmshorn, Germany, and is operated<br />
c<strong>on</strong>t<strong>in</strong>uously. The measurements are 10-s averaged<br />
cloud reflectivity records, sampled every m<strong>in</strong>ute. Local<br />
atmospheric sound<strong>in</strong>gs support the c<strong>on</strong>versi<strong>on</strong> of the<br />
cloud top (base) height to pressure levels. The use of<br />
the cloud radar at L<strong>in</strong>denberg and the limitati<strong>on</strong>s of the<br />
ground-based cloud detecti<strong>on</strong> and characterisati<strong>on</strong> are<br />
discussed by Hennemuth et al. [34]. In particular, some<br />
optically relevant clouds with too small particles, such as<br />
cumulus humilis and high cirrus, may fall below the radar<br />
detecti<strong>on</strong> threshold.
Dur<strong>in</strong>g the validati<strong>on</strong> campaign, a total of 161 overpasses<br />
with valid <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements and cloud product<br />
<strong>retrievals</strong> were available with<strong>in</strong> 50 km of the L<strong>in</strong>denberg<br />
meteorological stati<strong>on</strong>. For the reas<strong>on</strong>s cited above, no<br />
radar cloud measurements were available <strong>in</strong> 64 cases. 15<br />
cases were further excluded because <str<strong>on</strong>g>IASI</str<strong>on</strong>g> and the groundbased<br />
radar were sens<strong>in</strong>g different cloud structures <strong>in</strong><br />
different layers, due to large scale cloud heterogeneities<br />
(full details <strong>on</strong> the cloud heterogeneity evaluati<strong>on</strong> and<br />
handl<strong>in</strong>g <strong>in</strong> the validati<strong>on</strong> report by Arriaga, 2009). A total<br />
of 82 match-ups were reta<strong>in</strong>ed for validati<strong>on</strong> of the CTP<br />
and are presented <strong>in</strong> Fig. 10. The correlati<strong>on</strong> is high with a<br />
global bias of 29.4 hPa (<str<strong>on</strong>g>IASI</str<strong>on</strong>g> products higher) and standard<br />
deviati<strong>on</strong> of 49.2 hPa. These statistics vary with the<br />
cloud elevati<strong>on</strong> and the details are presented <strong>in</strong> Table 4.<br />
2.3.2.3. Inter-comparis<strong>on</strong> with CALIOP cloud top pressure.<br />
More recently, a validati<strong>on</strong> of the cloud height assignment<br />
from <str<strong>on</strong>g>IASI</str<strong>on</strong>g> dedicated to the Polar Regi<strong>on</strong>s was <strong>in</strong>itiated <strong>in</strong><br />
order to support the development of the EPS Day-2 Polar<br />
w<strong>in</strong>d product. The <str<strong>on</strong>g>IASI</str<strong>on</strong>g> products were <strong>in</strong>ter-compared<br />
to the <strong>retrievals</strong> from the Cloud–Aerosol Lidar with<br />
Fig. 10. <str<strong>on</strong>g>IASI</str<strong>on</strong>g> retrieved vs. ground-based radar measured cloud-top<br />
pressure (hPa) over L<strong>in</strong>denberg dur<strong>in</strong>g the EPS Validati<strong>on</strong> Campaign<br />
from June to September 2007.<br />
Table 4<br />
Bias and standard deviati<strong>on</strong> of the retrieved cloud top pressure (hPa).<br />
CTP (hPa) Retrieved cloud top pressure<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1351<br />
Radar Nb cases Bias Std<br />
Above 400 37 þ57.1 46.6<br />
400–700 26 þ19.1 37.4<br />
Below 700 19 10.5 32.7<br />
Orthog<strong>on</strong>al Polarisati<strong>on</strong> (CALIOP) <strong>in</strong>strument. CALIOP<br />
was launched <strong>on</strong>-board the Cloud–Aerosol Lidar and<br />
Infrared Pathf<strong>in</strong>der Satellite Observati<strong>on</strong>s (CALIPSO)<br />
satellite [35] <strong>in</strong> April 2006 as part of the NASA A-Tra<strong>in</strong>.<br />
It is the first space-borne Lidar to provide l<strong>on</strong>g-term<br />
atmospheric measurements, with unique cloud profil<strong>in</strong>g<br />
capabilities. CALIOP provides high resoluti<strong>on</strong> vertical<br />
profiles of clouds and aerosols with a resoluti<strong>on</strong> of 30 m<br />
from ground to 8 km and of 60 m between 8 and 20 km<br />
[36]. The cloud products v3 at 1-km horiz<strong>on</strong>tal resoluti<strong>on</strong><br />
[37] are used <strong>in</strong> this study. They are available at NASA’s<br />
<strong>on</strong>-l<strong>in</strong>e Earth Science Discovery Tool http://reverb.echo.<br />
nasa.gov/ (last accessed 19/01/2012). Simultaneous<br />
observati<strong>on</strong>s by <str<strong>on</strong>g>IASI</str<strong>on</strong>g> and CALIOP <strong>on</strong>ly occur at Polar<br />
latitudes, due to the <strong>orbit</strong>s of their respective spacecrafts:<br />
09h30 descend<strong>in</strong>g node for <strong>Metop</strong> and 13h30 ascend<strong>in</strong>g<br />
node for CALIPSO. Collocati<strong>on</strong>s <strong>in</strong> the period September to<br />
December 2010 with a maximum time difference of<br />
10 m<strong>in</strong> and a maximum distance of 10 km between the<br />
CALIOP pixel and the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> IFOV centre are selected.<br />
Additi<strong>on</strong>ally, <strong>on</strong>ly homogeneous s<strong>in</strong>gle layer clouds, as<br />
def<strong>in</strong>ed by CALIOP, are reta<strong>in</strong>ed for comparis<strong>on</strong>; i.e.<br />
where the CTPs measured by CALIOP do not vary by<br />
more than 50 hPa standard deviati<strong>on</strong> with<strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> field<br />
of view. A total of 4568 match-ups are c<strong>on</strong>sidered which<br />
are displayed <strong>in</strong> the scatter plot <strong>in</strong> Fig. 11. The colour<br />
cod<strong>in</strong>g refers to the retrieved <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 ECA. The few outliers<br />
seen <strong>in</strong> the graph, with <str<strong>on</strong>g>IASI</str<strong>on</strong>g> CTP around 100 hPa, are<br />
actually artefacts due to an algorithm <strong>in</strong>itialisati<strong>on</strong> issue,<br />
which does not affect the other CTP <strong>retrievals</strong>. It was<br />
corrected with revisi<strong>on</strong> 5.2 which became operati<strong>on</strong>al <strong>in</strong><br />
October 2011. The discrepancies between the two satellite<br />
cloud height assignments are larger for the IFOVs with an<br />
ECAs of 25% and below. For ECAs larger than 30% the<br />
correlati<strong>on</strong> is high (r 0.9). Two modes are observed,<br />
spann<strong>in</strong>g the lower and higher halves of the troposphere.<br />
Between 1000 and 550 hPa CTPs from <str<strong>on</strong>g>IASI</str<strong>on</strong>g> show a small<br />
bias of 15 hPa (<str<strong>on</strong>g>IASI</str<strong>on</strong>g> lower) and a standard deviati<strong>on</strong> of<br />
approximately 90 hPa. The dispersi<strong>on</strong> is smaller (standard<br />
deviati<strong>on</strong> of around 60 hPa), with however a larger bias<br />
(110 hPa, <str<strong>on</strong>g>IASI</str<strong>on</strong>g> higher) for clouds located above 550 hPa.<br />
One should recall that the CO 2-slic<strong>in</strong>g with IR measurements<br />
characterises the effective radiative height of a<br />
cloud while the Lidar locates a physical cloud boundary.<br />
The two may differ depend<strong>in</strong>g <strong>on</strong> the cloud opacity and<br />
this accounts for the biases observed between the two<br />
cloud height assignments. Similar offsets were recently<br />
reported by Kim et al. [38] compar<strong>in</strong>g the cloud top heights<br />
derived from CALIOP and those retrieved <strong>in</strong> the IR from<br />
MODIS with the CO 2-slic<strong>in</strong>g method. Inaccuracies <strong>in</strong> the<br />
forecast atmospheric profiles support<strong>in</strong>g the CO2-slic<strong>in</strong>g<br />
retrieval are also source of errors [39].<br />
2.3.3. The w 2 -method<br />
This method c<strong>on</strong>sists of modell<strong>in</strong>g the differences<br />
between measured and simulated clear radiances with a<br />
m<strong>on</strong>o-layer cloud for channels selected <strong>in</strong> CO2 l<strong>in</strong>es. For<br />
implementati<strong>on</strong> and theoretical performances see<br />
EUMETSAT technical note ‘‘Assessment of the chi-square<br />
method for cloud top pressure and equivalent cloud<br />
amount <strong>retrievals</strong> with measurements from <str<strong>on</strong>g>IASI</str<strong>on</strong>g>’’ (EUM/
1352<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> top cloud pressure [hPa]<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
100<br />
200<br />
300<br />
400<br />
500<br />
600<br />
700<br />
800<br />
900<br />
1000<br />
1000<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> vs CALIOP TCP depend<strong>in</strong>g <strong>on</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> CO2−slic<strong>in</strong>g ECA<br />
(scatter_caliop_iasi_eca, stdev
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1353<br />
– the sk<strong>in</strong> surface temperature (for ocean and c<strong>on</strong>t<strong>in</strong>ental<br />
surfaces)<br />
– the surface emissivity at twelve c<strong>on</strong>figurable channels<br />
– a coarse O3, profile <strong>in</strong> thick partial columns (0–6 km,<br />
0–12 km, 0–16 km)<br />
– the total column of CO, O3, CH4, N2O and CO2.<br />
A brief overview of the retrieval algorithms is given<br />
here while validati<strong>on</strong> methodology and results of the<br />
respective parameters is summarised <strong>in</strong> Secti<strong>on</strong>s 2.5–2.7.<br />
2.4.1. The statistical <strong>retrievals</strong><br />
The retrieval of the full state vector listed above is <strong>on</strong>ly<br />
attempted for cloud-free pixels. In case of a partial cloud<br />
c<strong>on</strong>tam<strong>in</strong>ati<strong>on</strong>, the geophysical retrieval is achieved with<br />
EOF regressi<strong>on</strong> and limited to temperature and humidity.<br />
The reader is referred to the work by Zhou [19] for the<br />
complete descripti<strong>on</strong> of the algorithm <strong>in</strong> cloudy c<strong>on</strong>diti<strong>on</strong>s<br />
and its performance. In clear-sky, a collecti<strong>on</strong> of<br />
statistical methods is applied first, which c<strong>on</strong>sists of EOF<br />
l<strong>in</strong>ear and ANN n<strong>on</strong>-l<strong>in</strong>ear regressi<strong>on</strong>s. In the EOF<br />
method, the pr<strong>in</strong>cipal comp<strong>on</strong>ent scores of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
radiances are computed and are <strong>in</strong>put to the l<strong>in</strong>ear<br />
regressi<strong>on</strong> for determ<strong>in</strong>ati<strong>on</strong> of the sea surface temperature<br />
(SST) [42], the oz<strong>on</strong>e, temperature and humidity<br />
profiles [19] as well as the land surface temperature and<br />
emissivity [9].<br />
Artificial neural networks are used to retrieve the total<br />
column amount of CO, CH4, N2O and CO2. The algorithm<br />
was adapted from the work of Hadji-Lazaro et al. [43] and<br />
Turquety et al. [44] to account for the <strong>in</strong>strument view<strong>in</strong>g<br />
angle, variable surface emissivity and <strong>in</strong>terfer<strong>in</strong>g atmospheric<br />
species. The ANNs <strong>in</strong>gest radiances <strong>in</strong> selected and<br />
c<strong>on</strong>figurable absorpti<strong>on</strong> and basel<strong>in</strong>e channels as well as a<br />
coarse atmospheric temperature profile, the surface pressure<br />
and the satellite zenith angle. Improvements were<br />
sought also <strong>in</strong> the preparati<strong>on</strong> of normalised and centred<br />
<strong>in</strong>puts to optimise the learn<strong>in</strong>g skills and speed [23]. The<br />
descripti<strong>on</strong> of the updated algorithm which became<br />
operati<strong>on</strong>al with the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF versi<strong>on</strong> 5 <strong>on</strong> 14/09/<br />
2010 is available <strong>on</strong>-l<strong>in</strong>e at: www.eumetsat.<strong>in</strong>t (August T.,<br />
2010, ‘‘An Improved Artificial Neural Network CO Retrieval<br />
for <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 Processor’’, EUM/MET/TEN/09/0232).<br />
The geophysical parameters retrieved with the statistical<br />
methods may c<strong>on</strong>stitute the f<strong>in</strong>al L2 product or serve<br />
as a first guess for the optimal estimati<strong>on</strong> retrieval<br />
(Secti<strong>on</strong> 2.4.2). The choice is c<strong>on</strong>figurable <strong>in</strong>dividually<br />
for each parameter and is also a functi<strong>on</strong> of the quality<br />
of the respective statistic and iterative <strong>retrievals</strong>.<br />
2.4.2. The retrieval us<strong>in</strong>g optimal estimati<strong>on</strong><br />
The OEM is the f<strong>in</strong>al retrieval module <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
process<strong>in</strong>g cha<strong>in</strong> and implements the standard optimal<br />
estimati<strong>on</strong> <strong>after</strong> Rodgers [45]. The core algorithm <strong>in</strong> the<br />
PPF versi<strong>on</strong> 5 is similar to the algorithm <strong>in</strong> former<br />
revisi<strong>on</strong>s [5]. The c<strong>on</strong>figurati<strong>on</strong> and the state-vector<br />
compositi<strong>on</strong> have changed <strong>in</strong> the PPF versi<strong>on</strong> 5.0; they<br />
are presented here.<br />
2.4.2.1. Atmospheric state vector and cost functi<strong>on</strong><br />
m<strong>in</strong>imisati<strong>on</strong>. In additi<strong>on</strong> to the surface temperature (Ts)<br />
and to the temperature (T) and humidity (q) profiles<br />
which were simultaneously retrieved already <strong>in</strong> the<br />
previous versi<strong>on</strong>s, the oz<strong>on</strong>e profile (O3) has become an<br />
active state vector parameter and is now simultaneously<br />
retrieved with<strong>in</strong> the OEM. The basis of the retrieval is the<br />
m<strong>in</strong>imisati<strong>on</strong> of a cost functi<strong>on</strong> of the classical form:<br />
j ¼ðx xaÞ T UB 1 Uðx xaÞþðFðxÞ yÞ T UR 1 UðFðxÞ yÞ ð2Þ<br />
The first term of the sum represents a c<strong>on</strong>stra<strong>in</strong>t <strong>on</strong> the<br />
atmospheric state vector to be retrieved, x, <strong>in</strong> terms of an<br />
a priori vector xa and its covariance matrix B. The sec<strong>on</strong>d<br />
term describes the c<strong>on</strong>stra<strong>in</strong>t generated by weight<strong>in</strong>g the<br />
departure of the measurements (y) from simulated<br />
radiances computed with the forward model F(x), tak<strong>in</strong>g<br />
the retrieved state vector x as <strong>in</strong>put, by the <strong>in</strong>strument<br />
noise covariance matrix R. The state vector x is modified<br />
<strong>in</strong> successive iterati<strong>on</strong>s <strong>in</strong> order to m<strong>in</strong>imise J us<strong>in</strong>g the<br />
Levenberg–Marquardt method. The m<strong>in</strong>imisati<strong>on</strong> is performed<br />
<strong>in</strong> brightness temperature space and the c<strong>on</strong>vergence<br />
criteri<strong>on</strong> is a threshold <strong>on</strong> the norm of the gradient<br />
of the cost functi<strong>on</strong>. For operati<strong>on</strong>al performance and<br />
timel<strong>in</strong>ess reas<strong>on</strong>s, a maximum of 5 iterati<strong>on</strong>s is allowed<br />
and the state vector x is accepted and stored <strong>in</strong>to the f<strong>in</strong>al<br />
L2 product if the residual F(x) y is smaller than a<br />
c<strong>on</strong>figurable threshold.<br />
With<strong>in</strong> the cost functi<strong>on</strong>, the vertical profiles to be<br />
retrieved used to be expressed <strong>on</strong> a pressure grid <strong>in</strong> the<br />
previous PPF versi<strong>on</strong>s. T, q and O3 are now represented by<br />
their pr<strong>in</strong>cipal comp<strong>on</strong>ents <strong>in</strong> the EOF space of the atmospheric<br />
profiles. In this approach 28, 18 and 9 PCs were<br />
reta<strong>in</strong>ed to represent the temperature (<strong>in</strong> K), the humidity<br />
and oz<strong>on</strong>e (both <strong>in</strong> ln(ppmv)) profiles, respectively. More<br />
details <strong>on</strong> the algorithm descripti<strong>on</strong> can be c<strong>on</strong>sulted <strong>on</strong>l<strong>in</strong>e<br />
at www.eumetsat.<strong>in</strong>t <strong>in</strong> the ‘‘EPS <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 Product<br />
Generati<strong>on</strong> Specificati<strong>on</strong>’’ (EUM/OPS-EPS/SPE/08/0199).<br />
2.4.2.2. Radiative transfer model, channel selecti<strong>on</strong>, bias<br />
correcti<strong>on</strong> and observati<strong>on</strong> error covariance matrix. The<br />
functi<strong>on</strong> F <strong>in</strong> (2) is the radiative transfer model RTTOV-<br />
10 (presented <strong>in</strong> Secti<strong>on</strong> 2.2.1), <strong>in</strong> use s<strong>in</strong>ce October 2011<br />
with the <strong>in</strong>troducti<strong>on</strong> of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF v5.2 (RT<str<strong>on</strong>g>IASI</str<strong>on</strong>g>-4<br />
was used <strong>in</strong> prior versi<strong>on</strong>s). The channel selecti<strong>on</strong> was<br />
modified and 316 channels from the list established by<br />
Collard [46] are used <strong>in</strong> the PPF v5 whereas the <strong>retrievals</strong><br />
were based <strong>on</strong> 222 channels <strong>in</strong> previous releases.<br />
In the sec<strong>on</strong>d term of the cost functi<strong>on</strong> (2), the<br />
measurements are fitted with simulated radiances. It is<br />
therefore essential to account beforehand for any systematic<br />
differences between the measurements and the<br />
forward computati<strong>on</strong>s, which may orig<strong>in</strong>ate <strong>in</strong> <strong>in</strong>strument<br />
errors or <strong>in</strong> <strong>in</strong>accuracies <strong>in</strong> the radiative transfer<br />
model. Global biases have been calculated from a set of<br />
pairs of radiance vectors, <strong>on</strong>e observed by <str<strong>on</strong>g>IASI</str<strong>on</strong>g> (referred to<br />
as OBS), the other calculated from some measure of truth<br />
(called CALC here<strong>after</strong>). The (OBS CALC) statistics were<br />
computed <strong>on</strong> clear sky ocean FOVs, us<strong>in</strong>g collocated<br />
ECMWF analysis profiles for the atmospheric state vector<br />
and RTTOV as a forward model. The systematic departures<br />
evaluated this way are used to tune the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements<br />
before <strong>in</strong>gesti<strong>on</strong> <strong>in</strong> this retrieval module.
1354<br />
The measurement error covariance matrix R <strong>in</strong> (2) was<br />
computed from these clear sky (OBS CALC) differences.<br />
Thus, besides the <strong>in</strong>strument noise (reduced with the<br />
noise filter<strong>in</strong>g, see Secti<strong>on</strong> 2.2.2 above), it also c<strong>on</strong>ta<strong>in</strong>s<br />
c<strong>on</strong>tributi<strong>on</strong>s from the model error. The full measurement<br />
error covariance matrix is now exploited. This also allows<br />
the correlati<strong>on</strong>s <strong>in</strong>troduced by the noise filter<strong>in</strong>g and the<br />
correlati<strong>on</strong>s due to model errors to be captured.<br />
2.4.2.3. Background a priori and covariance matrix, first<br />
guess. Until versi<strong>on</strong> 4.3, the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPFs were<br />
c<strong>on</strong>figured with a unique covariance matrix and n<strong>in</strong>e<br />
regi<strong>on</strong>al latitude-dependent a priori atmospheric state<br />
vectors which were based <strong>on</strong> the climatology established<br />
by Chevallier <strong>in</strong> 2001 from ECMWF analysis fields<br />
(<strong>in</strong> Chevallier F., 2001. ‘‘Sampled Database of 60 <strong>Level</strong>s<br />
Atmospheric Profiles from the ECMWF Analysis’’. Technical<br />
Report, ECMWF EUMETSAT SAF programme Research<br />
Report 4; ECMWF). In versi<strong>on</strong> 5.0, a global aprioriis used<br />
with a unique covariance matrix, computed from a<br />
collecti<strong>on</strong> of ECMWF analysis records cover<strong>in</strong>g all seas<strong>on</strong>s.<br />
The choice of a unique background rather than the coarse<br />
latitude stratificati<strong>on</strong> was made to avoid geographical<br />
disc<strong>on</strong>t<strong>in</strong>uities. With this c<strong>on</strong>figurati<strong>on</strong>, regi<strong>on</strong>al variati<strong>on</strong>s<br />
can also be expla<strong>in</strong>ed by the measurements rather than by<br />
the vary<strong>in</strong>g a priori.<br />
The atmospheric parameters to be retrieved are <strong>in</strong>itialised<br />
with the statistical <strong>retrievals</strong>, discussed <strong>in</strong> Secti<strong>on</strong><br />
2.4.1 above. The geophysical state vector has also a static<br />
part, i.e. <strong>on</strong>e that is not modified <strong>in</strong> successive iterati<strong>on</strong>s.<br />
For <strong>in</strong>stance, the profiles of CO2,CO,N2O and CH4 are set to<br />
the default trace gas profiles used <strong>in</strong> RTTOV-10 and kept<br />
fixed dur<strong>in</strong>g the iterati<strong>on</strong>s. Surface pressure and w<strong>in</strong>d<br />
vectors are obta<strong>in</strong>ed from ECMWF forecasts and the surface<br />
emissivity is either computed follow<strong>in</strong>g Masuda [15]<br />
over oceans or taken from a m<strong>on</strong>thly emissivity climatology<br />
database over land (see Secti<strong>on</strong> 2.1).<br />
2.5. The surface products<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
2.5.1. Land surface temperature<br />
The land surface temperature (LST) and land surface<br />
emissivity (LSE) enter<strong>in</strong>g the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 are currently taken<br />
from the statistical <strong>retrievals</strong> (Secti<strong>on</strong> 2.4.1). As part of the<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 validati<strong>on</strong> activities, the errors <strong>in</strong> the LST products<br />
were assessed by <strong>in</strong>ter-comparis<strong>on</strong> with the LST products<br />
generated at the Land Surface Analysis (LSA) SAF from the<br />
Sp<strong>in</strong>n<strong>in</strong>g Enhanced Visible and InfraRed Imager (SEVIRI)<br />
measurements acquired from the Meteosat Sec<strong>on</strong>d Generati<strong>on</strong><br />
(MSG) satellites, available <strong>in</strong> the EUMETSAT Earth<br />
Observati<strong>on</strong> data portal (https://eoportal.eumetsat.<strong>in</strong>t). This<br />
validati<strong>on</strong> study was therefore limited to the porti<strong>on</strong> of the<br />
Earth visible from the MSG geostati<strong>on</strong>ary <strong>orbit</strong>, namely<br />
ma<strong>in</strong>ly Africa, Europe and the Eastern part of South America.<br />
The short periodicity of the LSA LST products, repeated<br />
every 15 m<strong>in</strong>., however allowed very close temporal co<strong>in</strong>cidences<br />
between the two products which is essential as the<br />
surface sk<strong>in</strong> temperature diurnal variati<strong>on</strong>s can be as high<br />
as 6 1C/hforsomesoiltypes<strong>in</strong>deserts[47]. SEVIRIdata<br />
products from nom<strong>in</strong>al mode hav<strong>in</strong>g a spatial resoluti<strong>on</strong> of<br />
approximately 3 km at the sub-satellite po<strong>in</strong>t were used.<br />
The LSA-SAF product retrieved from SEVIRI measurements<br />
is based <strong>on</strong> a generalised split-w<strong>in</strong>dow algorithm<br />
us<strong>in</strong>g two adjacent channels—IR10.8 and IR12.0 mm. It<br />
was validated aga<strong>in</strong>st MODIS LST products and <strong>in</strong>-situ LST<br />
<strong>retrievals</strong> [48]. At night time, biases (standard deviati<strong>on</strong>)<br />
of around þ1.75 K (1.5 K) were found aga<strong>in</strong>st the former<br />
(LSA m<strong>in</strong>us MODIS) and of 1.7 K (2 K) aga<strong>in</strong>st the latter<br />
(LSA products be<strong>in</strong>g colder than <strong>in</strong>-situ <strong>retrievals</strong>).<br />
Each LSA LST comes with a quality flag <strong>in</strong>dicat<strong>in</strong>g the<br />
degree of c<strong>on</strong>fidence and the error associated with the<br />
retrieval. For this study, we <strong>on</strong>ly reta<strong>in</strong>ed the products with<br />
‘‘above nom<strong>in</strong>al’’ and ‘‘nom<strong>in</strong>al’’ quality for both the<br />
retrieved LST and surface emissivity. Departures (LSA SAF<br />
m<strong>in</strong>us <str<strong>on</strong>g>IASI</str<strong>on</strong>g>) were computed for each match-up where at<br />
least four good LSA LST <strong>retrievals</strong> were found with<strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
field of view. The match<strong>in</strong>g SEVIRI po<strong>in</strong>ts were averaged and<br />
used <strong>on</strong>ly if their standard deviati<strong>on</strong>s rema<strong>in</strong>ed lower than<br />
5 K to avoid highly heterogeneous scenes. The <strong>in</strong>ter-comparis<strong>on</strong>s<br />
were performed for day- and night-times separately<br />
to discrim<strong>in</strong>ate the effect of the different <strong>Metop</strong>/<br />
MSG–Sun–Surface geometry. At night-time the absence of<br />
solar illum<strong>in</strong>ati<strong>on</strong> allows a direct comparis<strong>on</strong> of the LST<br />
retrieved or modelled from different <strong>in</strong>struments. Dur<strong>in</strong>g<br />
daytime, the comparis<strong>on</strong> is affected by the different Sun–<br />
surface–<strong>in</strong>strument geometries, as a result of shadows due<br />
to orography or vegetati<strong>on</strong> for example. For a given place<br />
and time, the better the alignment between the <strong>in</strong>strument,<br />
the Sun and the scene, the smaller is the observed shadow<br />
fracti<strong>on</strong> and the warmer the sensed LST. This effect is visible<br />
<strong>in</strong> Fig. 13, where the LST–<str<strong>on</strong>g>IASI</str<strong>on</strong>g> mean difference varies by 2 K<br />
with scan angle <strong>in</strong> daytime and rema<strong>in</strong>s c<strong>on</strong>stant to better<br />
than 0.5 K at night time. The statistics are computed for the<br />
overall SEVIRI doma<strong>in</strong> with the exclusi<strong>on</strong> of the Sahara,<br />
where the PPF LST <strong>retrievals</strong> dur<strong>in</strong>g daytime <strong>in</strong> particular<br />
exhibit large variances. The African Sahara and the Arabian<br />
Pen<strong>in</strong>sula were then isolated and the statistics specifically<br />
repeated for these particular soil types.<br />
Fig. 14 illustrates the average and standard deviati<strong>on</strong> of<br />
(LSA–<str<strong>on</strong>g>IASI</str<strong>on</strong>g>) LST between the 19 and 24 March 2010 at night<br />
time. The distributi<strong>on</strong>s of the departures dur<strong>in</strong>g day and<br />
night times, displayed <strong>in</strong> Fig. 15, are essentially Gaussians<br />
(red shape), with the excepti<strong>on</strong> of few outliers. The correlati<strong>on</strong><br />
between the two products is of 0.98 over n<strong>on</strong>-desert<br />
places and of 0.7 for the Sahara. The best agreements are<br />
achieved at night, where the rms errors amount to about 2 K<br />
with usual surfaces. Larger differences were found <strong>in</strong><br />
elevated regi<strong>on</strong>s and for bare arid soils <strong>in</strong> the Sahara and<br />
the Arabic Pen<strong>in</strong>sula with a particular m<strong>in</strong>eralogy, especially<br />
<strong>in</strong> the Rub’ Al Khali subregi<strong>on</strong>. In additi<strong>on</strong>, global comparis<strong>on</strong>s<br />
aga<strong>in</strong>st the ECMWF LST analyses for the same period<br />
were performed. The results are summarised <strong>in</strong> Table 5. Itis<br />
referred to the dedicated validati<strong>on</strong> report (August, 2010,<br />
‘‘<str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 Surface Temperature: PPF v5 Validati<strong>on</strong> Results’’,<br />
EUM/MET/TEN/10/0188) for a more detailed analysis of the<br />
<strong>in</strong>ter-comparis<strong>on</strong> between the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 LST and LSA LST<br />
products as well as with the ECMWF analysis field.<br />
2.5.2. Land surface emissivity<br />
The land surface emissivity product is distributed <strong>in</strong> a<br />
pre-operati<strong>on</strong>al mode <strong>in</strong> 12 sampled channels. It has so<br />
far ma<strong>in</strong>ly been compared to the Global Infrared Land
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1355<br />
Fig. 13. LST (LSA–<str<strong>on</strong>g>IASI</str<strong>on</strong>g>) departures distributi<strong>on</strong> vs. <str<strong>on</strong>g>IASI</str<strong>on</strong>g> scan positi<strong>on</strong><br />
(Nadir at step 15) <strong>in</strong> the 19–24 March 2010 period at night (top) and day<br />
(bottom) time.<br />
Surface Emissivity database [49] (see EUMETSAT report,<br />
Hultberg, 2010, ‘‘Surface Emissivity with<strong>in</strong> <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF<br />
v5’’, EUM/OPS-EPS/TEN/10/0203). Validati<strong>on</strong> studies are<br />
<strong>on</strong>go<strong>in</strong>g to characterise it further. In future <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
processor versi<strong>on</strong>s, the OEM will <strong>in</strong>gest these <strong>retrievals</strong><br />
as an <strong>in</strong>put <strong>in</strong>stead of the emissivity m<strong>on</strong>thly climatology<br />
<strong>in</strong>troduced <strong>in</strong> Secti<strong>on</strong> 2.1. Retrieval of the land surface<br />
emissivity simultaneously with the temperature and<br />
humidity profiles is also planned, with a positive impact<br />
expected <strong>in</strong> the lower troposphere.<br />
2.5.3. Sea surface temperature<br />
The sea surface temperature selected for the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
products is currently based <strong>on</strong> the EOF retrieval, where<br />
PCs of the first two <str<strong>on</strong>g>IASI</str<strong>on</strong>g> bands are used for the l<strong>in</strong>ear<br />
regressi<strong>on</strong> (EOF <strong>in</strong> Secti<strong>on</strong> 2.4.1). The validati<strong>on</strong> of this<br />
product <strong>in</strong>volved <strong>in</strong>ter-comparis<strong>on</strong>s with other spaceborne<br />
sensors and to <strong>in</strong>-situ measurements from drift<strong>in</strong>g<br />
buoys. They are summarised here.<br />
2.5.3.1. Comparis<strong>on</strong> to AATSR SST products. In an <strong>in</strong>itial<br />
assessment phase, the SST <strong>retrievals</strong> from the Advanced<br />
Al<strong>on</strong>g-Track Scann<strong>in</strong>g Radiometer (AATSR) were used as<br />
reference products. The reader is referred to Llewellyn-<br />
J<strong>on</strong>es et al., 2001 ‘‘AATSR: Global-change and surfacetemperature<br />
measurements from ENVISAT’’ <strong>in</strong> the<br />
European Space Agency (ESA) Bulett<strong>in</strong> 105, pp. 10–21<br />
(available <strong>on</strong>l<strong>in</strong>e at: http://www.esa.<strong>in</strong>t/esapub/bullet<strong>in</strong>/<br />
bullet105/bul105_1.pdf, last accessed 09/02/2012) for the<br />
descripti<strong>on</strong> of the <strong>in</strong>strument and of the missi<strong>on</strong><br />
objectives. As with the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF, the AATSR processor<br />
is designed to retrieve the surface sk<strong>in</strong> temperature. The<br />
reference products <strong>in</strong> this study are the L2 ATS_NR_2P<br />
generated <strong>in</strong> the frame of the Group for High-Resoluti<strong>on</strong><br />
SST (GHRSST) [50] (www.ghrsst.org, last accessed 23/01/<br />
2012). They have a horiz<strong>on</strong>tal resoluti<strong>on</strong> of 1 km and were<br />
validated aga<strong>in</strong>st <strong>in</strong>-situ measurements (buoys and ships)<br />
and airborne radiometers [26]. The retrieval algorithm<br />
<strong>in</strong>cludes atmospheric correcti<strong>on</strong>s cover<strong>in</strong>g multi-spectral<br />
and dual-angle view capability. The L1 measurements<br />
underwent dedicated calibrati<strong>on</strong>s [51]. Ill<strong>in</strong>gworth et al.<br />
[52] evaluated the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> and AATSR radiance <strong>in</strong>tercalibrati<strong>on</strong><br />
at 11 and 12 mm and c<strong>on</strong>cluded that the<br />
brightness temperatures agreed with<strong>in</strong> 0.3 K, with even<br />
smaller differences of the order 0.1 K at 11 mm. The AATSR<br />
SST products are characterised by a slight warm bias of<br />
0.05 K at night (0.1 K for daytime) and by a typical<br />
standard deviati<strong>on</strong> of 0.25 K dur<strong>in</strong>g night-time (0.35 K<br />
dur<strong>in</strong>g day) [26,53].<br />
The <strong>in</strong>ter-comparis<strong>on</strong> was performed <strong>on</strong> the same 6day<br />
period (19–24 March 2010) used for the LST. As for<br />
the collocati<strong>on</strong> of AATSR pixels to <str<strong>on</strong>g>IASI</str<strong>on</strong>g> IFOVs, <strong>on</strong>ly the<br />
clear cases as identified <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> process<strong>in</strong>g cha<strong>in</strong> were<br />
reta<strong>in</strong>ed where at least 200 good AATSR pixels (accord<strong>in</strong>g<br />
to the L2 SST quality flags) could be found <strong>in</strong> a radius of<br />
15 km around the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> IFOV centre. Additi<strong>on</strong>ally, the<br />
match-ups were rejected if the standard deviati<strong>on</strong> of<br />
AATSR SST over the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> footpr<strong>in</strong>t exceeded 0.4 K <strong>in</strong> order<br />
to analyse homogeneous scenes <strong>on</strong>ly and limit the impact<br />
of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> po<strong>in</strong>t spread functi<strong>on</strong> <strong>in</strong> the <strong>in</strong>ter-comparis<strong>on</strong>s.<br />
A total of approximately 60,000 match-ups were f<strong>in</strong>ally<br />
c<strong>on</strong>sidered <strong>in</strong> this study.<br />
The <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 SST has a global cold bias of 0.44 K and<br />
a standard deviati<strong>on</strong> of approximately 0.40 K. As seen<br />
<strong>in</strong> Fig. 5, the distributi<strong>on</strong>s of AATSR–<str<strong>on</strong>g>IASI</str<strong>on</strong>g> SST have a<br />
Gaussian ma<strong>in</strong> mode, c<strong>on</strong>ta<strong>in</strong><strong>in</strong>g at least 90% of the<br />
samples and peak<strong>in</strong>g at þ0.27 K. The standard deviati<strong>on</strong><br />
of the fitted Gaussian is also smaller and drops down to<br />
0.28 K at night-time, when AATSR products are expected<br />
to be the most accurate. A small fracti<strong>on</strong> (5–10%) of the<br />
<strong>retrievals</strong> lies outside this ma<strong>in</strong> mode. The associated<br />
departures are typically of 1–1.5 K (<str<strong>on</strong>g>IASI</str<strong>on</strong>g> colder) but can <strong>on</strong><br />
rare occasi<strong>on</strong>s be as high as a few Kelv<strong>in</strong>. This asymmetric<br />
tail is mostly attributed to undetected clouds and aerosols<br />
(dust clouds) c<strong>on</strong>tam<strong>in</strong>ati<strong>on</strong> as discussed <strong>in</strong> Secti<strong>on</strong> 2.2.5,
1356<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
Fig. 14. LSA–<str<strong>on</strong>g>IASI</str<strong>on</strong>g> LST mean (top) and standard deviati<strong>on</strong> (bottom) <strong>in</strong> the 19–24 March 2010 period.<br />
whichartificiallycoolstheretrievedSST.Thiseffectisshown<br />
<strong>in</strong> Fig. 16, where the differences (AATSR <str<strong>on</strong>g>IASI</str<strong>on</strong>g>) <strong>in</strong> SST <strong>in</strong> the<br />
Arabian Sea <strong>on</strong> 19/03/2010 correlate with higher dust load.<br />
The (AATSR–<str<strong>on</strong>g>IASI</str<strong>on</strong>g>) SST statistics computed outside the geographic<br />
areas with high aerosol optical depths, as identified<br />
with MODIS products, have a mean of approximately þ0.3 K<br />
and a standard deviati<strong>on</strong> of the same order. The current<br />
operati<strong>on</strong>al implementati<strong>on</strong> does not however <strong>in</strong>clude such<br />
an aerosol filter<strong>in</strong>g. These figures, although still theoretical as<br />
l<strong>on</strong>g as aerosols are not effectively detected, can be extrapolated<br />
to the sea areas which are climatologically clean of<br />
dust. Similar comparis<strong>on</strong>s were repeated with ECMWF<br />
analyses fields, which c<strong>on</strong>firmed the slight cold bias <strong>in</strong> <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
SST and the good precisi<strong>on</strong> assessed with AATSR. The<br />
departures from AATSR and ECMWF SST were assessed as a<br />
functi<strong>on</strong> of view<strong>in</strong>g geometry, which revealed a slight<br />
angular variati<strong>on</strong> of approximately 0.3 K amplitude <strong>in</strong> the<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> SST bias from Nadir to the swath edge.<br />
2.5.3.2. S<strong>in</strong>gle sensor error estimate and quality levels. To<br />
promote the use of comb<strong>in</strong>ed multiple SST datasets the SST<br />
community has agreed <strong>on</strong> a specificati<strong>on</strong> for all global<br />
satellite-derived level-2 SST products with<strong>in</strong> the GHRSST<br />
[50] (www.ghrsst.org, last accessed 23/01/2012). Highresoluti<strong>on</strong><br />
products are provided to the operati<strong>on</strong>al<br />
oceanographic, meteorological and climate community <strong>on</strong><br />
a daily basis <strong>in</strong> a comm<strong>on</strong> netCDF format. The GHRSST<br />
level-2 format is identified as ‘L2P’ and c<strong>on</strong>ta<strong>in</strong>s<br />
observati<strong>on</strong>al error estimates called S<strong>in</strong>gle Sensor Error<br />
Statistics (SSES). The good performances described <strong>in</strong> the<br />
previous secti<strong>on</strong> motivated the creati<strong>on</strong> of a dem<strong>on</strong>strati<strong>on</strong><br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> SST L2P product as a c<strong>on</strong>tributi<strong>on</strong> to the GHRSST, for<br />
which trial-dissem<strong>in</strong>ati<strong>on</strong> started <strong>on</strong> 24/03/2011. Six levels<br />
of quality are def<strong>in</strong>ed for the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2P SST from 0 to 5 with<br />
2 be<strong>in</strong>g the first usable quality and 5, the best quality.<br />
The <str<strong>on</strong>g>IASI</str<strong>on</strong>g> SST error has been further estimated by<br />
comparis<strong>on</strong> to other satellite SST products, namely from
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1357<br />
Fig. 15. LST (LSA–<str<strong>on</strong>g>IASI</str<strong>on</strong>g>) distributi<strong>on</strong>s at day (left) and night (right) times, for n<strong>on</strong>-arid areas (top row) and Sahara (bottom row) <strong>in</strong> the 19–24 March 2010<br />
period.<br />
Table 5<br />
Summary of the respective (LSA,ECMWF–<str<strong>on</strong>g>IASI</str<strong>on</strong>g>) LST departures for the 19–24 March 2010 period: bias, standard deviati<strong>on</strong> and correlati<strong>on</strong> coefficient. In<br />
parentheses, the statistics of the Gaussian fitt<strong>in</strong>g the ma<strong>in</strong> mode.<br />
the <strong>in</strong>struments AVHRR/<strong>Metop</strong> and AATSR, as well as with<br />
drift<strong>in</strong>g buoy <strong>in</strong>-situ measurements, <strong>in</strong> order to establish<br />
SSES for <str<strong>on</strong>g>IASI</str<strong>on</strong>g>. A summary of the validati<strong>on</strong> approach and<br />
of the results is presented here (they are more exhaustively<br />
discussed <strong>in</strong> EUMETSAT validati<strong>on</strong> report from<br />
O’Carroll A., 2010, ‘‘Validati<strong>on</strong> of <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2Pcore Sea Surface<br />
Temperature’’, EUM/MET/DOC/10/0472).<br />
Day Night<br />
Bias s r Bias s r<br />
LSA–<str<strong>on</strong>g>IASI</str<strong>on</strong>g> No Sahara 0.6 ( 0.7) K 2.4 (1.8) K 0.97 1.6 ( 1.2) K 1.6 (1.0) K 0.99<br />
Sahara 1.7 ( 1.3) K 4.6 (3.3) K 0.75 4.4 ( 2.9) K 5.6 (3.0) K 0.65<br />
ECMWF–<str<strong>on</strong>g>IASI</str<strong>on</strong>g> No mounta<strong>in</strong>s, no Poles, no Sahara 4.0 ( 3.2) K 3.6 (3.1) K 0.97 1.9 ( 0.9) K 2.9 (1.7) K 0.99<br />
Sahara 5.7 ( 5.4) K 5.7 (5.2) K 0.66 3.5 ( 2.3) K 5.0 (2.6) K 0.73<br />
Poles (Antarctica, Arctic) 2.9 ( 1.8) K 3.3 (2.5) K 0.92 3.6 ( 1.1) K 6.9 (2.0) K 0.83<br />
The <str<strong>on</strong>g>IASI</str<strong>on</strong>g> SSTs have been rout<strong>in</strong>ely compared to the<br />
EUMETSAT Ocean and Sea-Ice Satellite Applicati<strong>on</strong> Facility<br />
(OSI-SAF) AVHRR/<strong>in</strong> situ Matchup Dataset (MDB) (Le<br />
Borgne et al., ‘‘Operati<strong>on</strong>al SST retrieval from METOP/<br />
AVHRR validati<strong>on</strong> report’’, Ocean and Sea-Ice SAF CDOP<br />
report, Versi<strong>on</strong> 2.0, July 2008) to provide a multi-matchup<br />
dataset (MMD) of collocated <str<strong>on</strong>g>IASI</str<strong>on</strong>g>, AVHRR and drift<strong>in</strong>g
1358<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
Fig. 16. Clear-sky (AATSR–<str<strong>on</strong>g>IASI</str<strong>on</strong>g>) SST plotted over an AVHRR image (visible channel) of the Arabian Sea <strong>on</strong> 19 March 2010. Land and clouds appear <strong>in</strong><br />
saturated white, dust loads colourlight grey.<br />
buoy SSTs. This MMD was used to determ<strong>in</strong>e the biases<br />
and errors of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> SSTs. In order to enable direct<br />
comparis<strong>on</strong> with buoy sub-sk<strong>in</strong> SSTs, 0.17 K are added to<br />
the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> sk<strong>in</strong> SSTs <strong>after</strong> D<strong>on</strong>l<strong>on</strong> et al. [54], <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
provid<strong>in</strong>g the sk<strong>in</strong> surface temperature. Quality c<strong>on</strong>trol<br />
of the buoy data requires the buoy SSTs to be with<strong>in</strong> 5 K of<br />
climatology and a buoy blacklist supplied by the Ocean<br />
and Sea-Ice SAF (OSI-SAF) is used <strong>in</strong> additi<strong>on</strong>. As for the<br />
AVHRR products, the c<strong>on</strong>fidence flags are used to select<br />
good quality <strong>retrievals</strong> (QC flag 42) and match-ups are<br />
c<strong>on</strong>sidered if the AVHRR SST over an <str<strong>on</strong>g>IASI</str<strong>on</strong>g> footpr<strong>in</strong>t is less<br />
than 0.3 K. Additi<strong>on</strong>ally, al<strong>on</strong>g the GHRSST guidel<strong>in</strong>es, the<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> and <strong>in</strong> situ SST time difference must be with<strong>in</strong> 2 h;<br />
<strong>on</strong>ly drift<strong>in</strong>g buoys are used and <strong>on</strong>ly night-time <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
observati<strong>on</strong>s over sea are used <strong>in</strong> order to reduce diurnal<br />
variati<strong>on</strong>s.<br />
As the retrieval method does not provide an associated<br />
error estimate and a major error c<strong>on</strong>tributi<strong>on</strong> to SST<br />
<strong>retrievals</strong> is the amount of water vapour <strong>in</strong> the atmosphere<br />
[55], climatological SSTs are used to def<strong>in</strong>e <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
SST quality levels by stratificati<strong>on</strong> aga<strong>in</strong>st the <strong>in</strong>tegrated<br />
water vapour (IWV), <strong>in</strong> order to def<strong>in</strong>e the SSES thresholds<br />
and criteria. The <str<strong>on</strong>g>IASI</str<strong>on</strong>g> water vapour profiles c<strong>on</strong>ta<strong>in</strong>ed<br />
<strong>in</strong> the L2 product are taken as <strong>in</strong>put. AVHRR, buoys and<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> match-up databases are rout<strong>in</strong>ely c<strong>on</strong>structed and<br />
the characterisati<strong>on</strong> of the errors <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> SSTs, stratified<br />
aga<strong>in</strong>st the IWV is updated at EUMETSAT <strong>on</strong> a 6m<strong>on</strong>th<br />
basis to complete the <strong>in</strong>formati<strong>on</strong> <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> SST<br />
L2Pcore product. With the <strong>in</strong>troducti<strong>on</strong> of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF<br />
v5, the bias <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> SSTs compared to drift<strong>in</strong>g buoy<br />
SSTs, assessed between October 2010 and March 2011,<br />
gave a value of 0.3 K (<str<strong>on</strong>g>IASI</str<strong>on</strong>g> colder) with a standard<br />
deviati<strong>on</strong> of 0.3 K. Comparis<strong>on</strong>s aga<strong>in</strong>st AVHRR SSTs gave<br />
a <str<strong>on</strong>g>IASI</str<strong>on</strong>g> SST bias of 0.33 K (<str<strong>on</strong>g>IASI</str<strong>on</strong>g> colder) and standard<br />
deviati<strong>on</strong> of 0.28 K, over the same period. In additi<strong>on</strong>,<br />
given three different observati<strong>on</strong> sources, it is possible to<br />
estimate the overall standard deviati<strong>on</strong> of error of the<br />
observati<strong>on</strong> type [56], assum<strong>in</strong>g that the errors between<br />
the observati<strong>on</strong>s are uncorrelated. The global standard<br />
deviati<strong>on</strong> of errors us<strong>in</strong>g this method, over the period<br />
October 2010 to March 2011, are 0.26 K (<str<strong>on</strong>g>IASI</str<strong>on</strong>g>), 0.14 K<br />
(AVHRR), and 0.19 K (drift<strong>in</strong>g buoys).<br />
Future developments for this product will address the<br />
slight angular dependency and the <strong>in</strong>clusi<strong>on</strong> of the band 3<br />
(shorter wavelengths) <strong>in</strong> the retrieval at night time.<br />
Another aspect where improvements are anticipated is<br />
<strong>in</strong> the detecti<strong>on</strong> of dust layers with <str<strong>on</strong>g>IASI</str<strong>on</strong>g> for SST product<br />
quality flagg<strong>in</strong>g and possible correcti<strong>on</strong>.<br />
2.6. Temperature and humidity profiles<br />
2.6.1. Results of L<strong>in</strong>denberg validati<strong>on</strong> campaign<br />
To the first order, the validati<strong>on</strong> of remotely-sensed<br />
atmospheric profiles can c<strong>on</strong>sist of a level-to-level comparis<strong>on</strong><br />
with some <strong>in</strong>dependent representati<strong>on</strong> of the true<br />
atmospheric state. The reference data can be for <strong>in</strong>stance<br />
acquired <strong>in</strong>-situ by airborne, radio-s<strong>on</strong>de or drop-s<strong>on</strong>de<br />
systems. This is the approach followed <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> product<br />
assessment part of the ESA GlobVapour project, which<br />
aims at generat<strong>in</strong>g validated multi-annual global water<br />
vapour datasets with error estimates. (www.globvapour.<br />
<strong>in</strong>fo, last accessed 23/01/2012). One limitati<strong>on</strong> is that the<br />
retrieval and the reference system do not give a representati<strong>on</strong><br />
of the true state with the same vertical resoluti<strong>on</strong>.<br />
In fact, the OEM retrieval scheme dictates that the<br />
f<strong>in</strong>al retrieved state vector is a comb<strong>in</strong>ati<strong>on</strong> of the true<br />
state vector and of the a priori that c<strong>on</strong>stra<strong>in</strong>s Eq. (2).<br />
The a priori enters the retrieval for the porti<strong>on</strong>s of the
atmosphere where the measurements do not carry<br />
enough <strong>in</strong>formati<strong>on</strong> or are not given enough weight e.g.<br />
too large measurement noise <strong>in</strong> R <strong>in</strong> (2). Rodgers [45]<br />
showed that from (2) derives:<br />
^x ¼ Axv þðI AÞxa þe ð3Þ<br />
where ^x is the retrieved vector, enter<strong>in</strong>g the L2 product, x v<br />
is the true atmospheric state and xa the a priori profile. A<br />
is a matrix called the averag<strong>in</strong>g kernel, giv<strong>in</strong>g a measure<br />
of the vertical resoluti<strong>on</strong> and vertical sensitivity of a given<br />
parameter. I is the identity matrix and e is the retrieval<br />
error to be assessed. The trace of A gives a measure of the<br />
<strong>in</strong>dependent pieces of <strong>in</strong>formati<strong>on</strong> retrieved from the<br />
measurements, or degrees of freedom (DoF) for the<br />
retrieval [45]. They are smaller than the number of levels<br />
<strong>in</strong> the vertical grid and can vary with the atmospheric<br />
c<strong>on</strong>figurati<strong>on</strong>. For example, up to 14 and 10 DoFs <strong>in</strong> the<br />
temperature and water-vapour profiles retrieved from<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> for 2007 summer cases <strong>in</strong> L<strong>in</strong>denberg.<br />
Additi<strong>on</strong>ally, for n<strong>on</strong>-uniform and dynamic atmospheres,<br />
the reference data are a limited representati<strong>on</strong><br />
of the true state observed by the space-borne <strong>in</strong>strument<br />
because of unavoidable spatial and temporal n<strong>on</strong>-co<strong>in</strong>cidences<br />
between the correlative products. Pougatchev et<br />
al. [57] evaluated the errors <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 products<br />
versi<strong>on</strong> 4.3 with radio-s<strong>on</strong>des operated dur<strong>in</strong>g a dedicated<br />
validati<strong>on</strong> campaign at L<strong>in</strong>denberg, Germany.<br />
Between June and September 2007, Vaisala RS92-SGP<br />
radios<strong>on</strong>des were launched <strong>on</strong>e hour and <strong>five</strong> m<strong>in</strong>utes<br />
prior to <str<strong>on</strong>g>IASI</str<strong>on</strong>g> overpasses <strong>in</strong> order to sample the higher and<br />
lower troposphere simultaneously with <str<strong>on</strong>g>IASI</str<strong>on</strong>g> acquisiti<strong>on</strong>s<br />
(full documentati<strong>on</strong> <strong>on</strong> the campaign <strong>in</strong> the f<strong>in</strong>al report<br />
‘‘EUMETSAT Polar System Programme Atmospheric<br />
Sound<strong>in</strong>g Campaign’’). The study c<strong>on</strong>sidered <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 s<strong>in</strong>gle<br />
<strong>retrievals</strong> with<strong>in</strong> 100 km of L<strong>in</strong>denberg and evaluated the<br />
c<strong>on</strong>tributi<strong>on</strong> of the n<strong>on</strong>-co<strong>in</strong>cidences to the total error<br />
budget. In c<strong>on</strong>clusi<strong>on</strong>, the expected and the assessed<br />
errors <strong>in</strong> the operati<strong>on</strong>al <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 are <strong>in</strong> good agreement<br />
between 800 (700) hPa and the tropopause for temperature<br />
(water vapour) <strong>retrievals</strong>, with typical error standard<br />
deviati<strong>on</strong> of 0.6 K between 800 and 300 hPa and <strong>in</strong>creas<strong>in</strong>g<br />
up to 2 K at the surface. The bias evaluated aga<strong>in</strong>st<br />
radios<strong>on</strong>des oscillates with<strong>in</strong> 70.5 K. For relative humidity<br />
(RH), the s<strong>in</strong>gle retrieval error is below 10% RH<br />
standard deviati<strong>on</strong> between 700 and 300 hPa and up to<br />
15% closer to the surface. The bias is with<strong>in</strong> 710% RH.<br />
Larger errors <strong>in</strong> the boundary layer were assumed <strong>in</strong> this<br />
Table 6<br />
Validati<strong>on</strong> classes def<strong>in</strong>iti<strong>on</strong>.<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1359<br />
study to be partly due to <strong>in</strong>correct surface parameters<br />
(e.g. emissivity) and undetected clouds or haze.<br />
2.6.2. Global assessment of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 profiles v4 vs v5<br />
Global error estimate are found by <strong>in</strong>ter-comparis<strong>on</strong>s<br />
to ECMWF analyses fields, which we summarise here. A<br />
detailed presentati<strong>on</strong> of the validati<strong>on</strong> strategy and the<br />
analyses can be c<strong>on</strong>sulted from EUMETSAT validati<strong>on</strong><br />
report ‘‘Vertical Temperature and Humidity Profiles<br />
with<strong>in</strong> <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF v5: N<strong>on</strong>-Regressi<strong>on</strong> Tests and Validati<strong>on</strong><br />
Results’’ (EUM/MET/TEN/09/0448), available <strong>on</strong>-l<strong>in</strong>e<br />
at: www.eumetsat.<strong>in</strong>t. In view of the <strong>in</strong>troducti<strong>on</strong> of the<br />
versi<strong>on</strong> 5, the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 temperature and humidity profiles<br />
of versi<strong>on</strong> 4 and versi<strong>on</strong> 5 have been compared to NWP<br />
analyses from ECMWF <strong>in</strong>tegrated forecast system (IFS) to<br />
characterise the improvements. The reference data used<br />
here are provided <strong>on</strong> a 0.51 0.51 geographical grid and<br />
are represented <strong>on</strong> a vary<strong>in</strong>g hybrid 91 pressure level<br />
vertical grid, def<strong>in</strong>ed for each po<strong>in</strong>t with the local surface<br />
pressure. They are available for the synoptic times 00, 06,<br />
12 and 18 UTC. The s<strong>in</strong>gle <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 <strong>retrievals</strong> were matched<br />
with the closest ECMWF latitude/l<strong>on</strong>gitude grid po<strong>in</strong>t and<br />
the corresp<strong>on</strong>d<strong>in</strong>g NWP profile was <strong>in</strong>terpolated <strong>on</strong>to the<br />
L2 product grid.<br />
The results obta<strong>in</strong>ed with two validati<strong>on</strong> datasets are<br />
reported here. The first <strong>on</strong>e is composed of <str<strong>on</strong>g>IASI</str<strong>on</strong>g> products<br />
from 43 <strong>orbit</strong>s centred <strong>on</strong> the regular ECMWF analysis<br />
times. As a result, the associated reprocessed L2 v4 and v5<br />
<strong>retrievals</strong> are <strong>on</strong> average with<strong>in</strong> half-an-hour of the<br />
reference data. These 43 <strong>orbit</strong>s were randomly selected<br />
with weekly <strong>in</strong>tervals to span a n<strong>in</strong>e-m<strong>on</strong>th period runn<strong>in</strong>g<br />
from June 2007 to March 2008 and to equally<br />
represent the four analysis times aga<strong>in</strong>st which the<br />
validati<strong>on</strong> is performed. The sec<strong>on</strong>d data set addresses<br />
the full Earth coverage with c<strong>on</strong>secutive <strong>retrievals</strong> dur<strong>in</strong>g<br />
the period 19–24 March 2010. A time-<strong>in</strong>terpolati<strong>on</strong><br />
was applied to the vertically re-sampled analyses prior<br />
to comput<strong>in</strong>g the departures, <strong>in</strong> order to account for<br />
potential temporal n<strong>on</strong>-co<strong>in</strong>cidences. They both lead to<br />
c<strong>on</strong>sistent statistics and c<strong>on</strong>clusi<strong>on</strong>s. The statistics of <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
L2–ECMWF departures were analysed <strong>in</strong> n<strong>in</strong>e geographical<br />
and surface categories (descripti<strong>on</strong> <strong>in</strong> Table 6). They<br />
are reproduced here for the period 19–24 March 2010<br />
<strong>on</strong>ly, <strong>in</strong> Fig. 17 (temperature) and Fig. 18 (humidity).<br />
2.6.2.1. <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 temperature vs. ECMWF analyses. The<br />
temperature enter<strong>in</strong>g the f<strong>in</strong>al L2 products is nom<strong>in</strong>ally<br />
Class Label Surface pressure (hPa) Surface type Latitude Time<br />
1 North Pole (NP) o1050 Land and sea 4601 Day and night<br />
2 North Sea 4900 Sea [301; 601] Day and night<br />
3 North land 4900 Land [301; 601] Day and night<br />
4 Elevated terra<strong>in</strong> o900 Land [ 601; 601] Day and night<br />
5 Intertropical Sea 4900 Sea [ 301; 301] Day and night<br />
6 Intertropical land 4900 Land [ 301; 301] Day and night<br />
7 South Pole o1050 Land and sea o 601 Day and night<br />
8 South Sea 4900 Sea [ 601; 301] Day and night<br />
9 South land 4900 Land [ 601; 301] Day and night
1360<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
Fig. 17. Temperature vertical profile <strong>retrievals</strong>, departures from ECMWF analyses products (Retrieved—ECMWF). <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 <strong>retrievals</strong> shown <strong>in</strong> black and<br />
red corresp<strong>on</strong>d respectively to the PPFs v5.0.3 and v4.3.3. Dashed l<strong>in</strong>es denote the bias, while the standard deviati<strong>on</strong> and rms are plotted with th<strong>in</strong> and<br />
thick pla<strong>in</strong> l<strong>in</strong>es, respectively. The classes listed <strong>in</strong> Table 6 appear <strong>in</strong> order from left to right and top to bottom. (For <strong>in</strong>terpretati<strong>on</strong> of the references to<br />
colour <strong>in</strong> this figure legend, the reader is referred to the web versi<strong>on</strong> of this article.)<br />
selected from the OEM described <strong>in</strong> Secti<strong>on</strong> 2.4.2. The<br />
agreement between the temperature retrieval and the<br />
reference varies with the nature of the scene, the best<br />
matches occurr<strong>in</strong>g over ocean surfaces; outsides the<br />
Polar regi<strong>on</strong>s. The (<str<strong>on</strong>g>IASI</str<strong>on</strong>g>–ECMWF) departures <strong>in</strong>crease <strong>in</strong><br />
the boundary layer and are usually smaller for PPF v5<br />
than for PPF v4. The rms differences are typically 0.7–1 K<br />
(1–1.5 K) between 200 and 800 hPa, and <strong>in</strong>crease up to<br />
2 K (2.5 K) at lower atmospheric levels for PPF v5 (PPF<br />
v4). Over the c<strong>on</strong>t<strong>in</strong>ents, the PPF v5 still shows the<br />
same good statistics <strong>in</strong> the mid and upper troposphere<br />
while PPF v4 errors are larger by 1.5 K. Below<br />
800 hPa, however, both revisi<strong>on</strong>s exhibit similar errors,<br />
rang<strong>in</strong>g between 2.5 and 3.5 K <strong>on</strong> average. The same<br />
observati<strong>on</strong>s apply to the North Pole and the elevated<br />
surfaces. The highest disagreements are found over<br />
Antarctica, with rms differences exceed<strong>in</strong>g 4 K below<br />
600 hPa, where the models are also expected to have<br />
<strong>in</strong>tr<strong>in</strong>sically larger uncerta<strong>in</strong>ties.<br />
The bias oscillates vertically around 0 K for PPF v4,<br />
with an amplitude of approximately 1 K. This behaviour<br />
disappears with revisi<strong>on</strong> 5, whose products have a bias of<br />
about þ0.5 K rather c<strong>on</strong>stant with pressure level for n<strong>on</strong>polar<br />
land and ocean surfaces between pressure levels<br />
200 and 800 hPa. Closer to the surface however, the sign<br />
of the average departure changes and the bias can be as<br />
large as 3 K (for tropical land surfaces), sometimes even<br />
larger than the PPF v4 biases over land.
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1361<br />
Fig. 18. Same as previous figure, for water-vapour profiles <strong>in</strong> volume mix<strong>in</strong>g ratio.<br />
These departures were stratified aga<strong>in</strong>st the view<strong>in</strong>g<br />
angle to assess potential angular variati<strong>on</strong>s of the retrieval<br />
quality. The rms differences show an <strong>in</strong>crease of 0.5 K from<br />
the Nadir to the swath edge <strong>in</strong> the mid-troposphere for v4<br />
products, <strong>in</strong> both the bias and the standard deviati<strong>on</strong>,<br />
which is no l<strong>on</strong>ger present <strong>in</strong> the v5. An angular effect<br />
however rema<strong>in</strong>s <strong>in</strong> the bias closer to the surface with v5,<br />
with amplitude of about 1 K at 940 hPa. As the ECMWF<br />
analyses are <strong>in</strong>dependent of the scan angle, this effect is<br />
attributed to the retrieval scheme. The reas<strong>on</strong>s still need to<br />
be <strong>in</strong>vestigated, they could be due for <strong>in</strong>stance to scan<br />
angle dependent performances <strong>in</strong> the cloud detecti<strong>on</strong>.<br />
2.6.2.2. <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 humidity vs. ECMWF analyses. Tests showed<br />
that the water-vapour profiles retrieved <strong>in</strong> the low<br />
troposphere with the OEM are essentially dom<strong>in</strong>ated by<br />
the a priori (see discussi<strong>on</strong>s <strong>in</strong> Secti<strong>on</strong> 2.6.1), yield<strong>in</strong>g large<br />
systematic errors. The f<strong>in</strong>al choice of retrieval for the <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
L2 product is c<strong>on</strong>figurable and, <strong>in</strong> the case of the watervapour,<br />
it was therefore decided for versi<strong>on</strong> 5 to select the<br />
first guess profiles retrieved with the EOF regressi<strong>on</strong> [19]<br />
<strong>in</strong>troduced <strong>in</strong> Secti<strong>on</strong> 2.4.1. The water vapour <strong>retrievals</strong><br />
perform differently depend<strong>in</strong>g <strong>on</strong> the scene and the overall<br />
water vapour c<strong>on</strong>tent. As with the temperature profiles,<br />
the retrieved profiles (with v4 and v5) and the modelled<br />
profiles are <strong>in</strong> good agreement down to 700–800 hPa, with<br />
an rms of about 10% relative humidity (RH) and no bias.<br />
C<strong>on</strong>sistently with the c<strong>on</strong>clusi<strong>on</strong>s drawn from the<br />
validati<strong>on</strong> with s<strong>on</strong>des, the departures are larger <strong>in</strong> the<br />
boundary layer, with an rms of the order of 20% RH. The<br />
retrieved relative humidity profiles for both PPF v4 and v5<br />
lie <strong>in</strong> a different range than the ECMWF analyses, usually<br />
with a much smaller dynamic. This is typically the case <strong>in</strong><br />
the Northern Hemisphere over oceans where the modelled
1362<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
Fig. 19. Time series of the temperature errors at 700 hPa (bias: pla<strong>in</strong> l<strong>in</strong>e, standard deviati<strong>on</strong>: dash-l<strong>in</strong>es) with different satellite products (blue: AIRS,<br />
purple: <str<strong>on</strong>g>IASI</str<strong>on</strong>g>/NOAA, gold: <str<strong>on</strong>g>IASI</str<strong>on</strong>g>/EUMETSAT, green: UCAR, red: GFS forecastts) as evaluated aga<strong>in</strong>st collocated <strong>in</strong>-situ s<strong>on</strong>de measurements. (http://www.<br />
star.nesdis.noaa.gov/smcd/opdb/poes/NPROVS.php). (For <strong>in</strong>terpretati<strong>on</strong> of the references to color <strong>in</strong> this figure legend, the reader is referred to the web<br />
versi<strong>on</strong> of this article.)<br />
humidity at 980 hPa varies from 20% to 90% RH while PPF<br />
v5 (v4) spans a shorter 55–80% RH (30–70% RH) range. The<br />
best agreement is obta<strong>in</strong>ed <strong>in</strong> the tropics [301S; 301N] over<br />
land with a correlati<strong>on</strong> coefficient of 0.76 for PPF v5 and<br />
0.58 for PPF v4. The overall performances of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
humidity products v5 are otherwise comparable to the v4.<br />
2.6.2.3. M<strong>on</strong>itor<strong>in</strong>g aga<strong>in</strong>st s<strong>on</strong>des. The operati<strong>on</strong>al EUMET-<br />
SAT <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 products have been m<strong>on</strong>itored at the U.S.<br />
Nati<strong>on</strong>al Oceanic and Atmospheric Adm<strong>in</strong>istrati<strong>on</strong> (NOAA)<br />
<strong>in</strong> the scope of the NOAA Products Validati<strong>on</strong> System<br />
(NPROVS) which compiles datasets of collocated radios<strong>on</strong>de,<br />
drops<strong>on</strong>de and appended numerical weather predicti<strong>on</strong><br />
(NWP) data for comparis<strong>on</strong>s to satellites products [58].<br />
Each day, the statistics (mean and standard deviati<strong>on</strong>) for<br />
temperature and water vapour at selected pressure levels are<br />
computed and stored <strong>in</strong> view of m<strong>on</strong>thly and l<strong>on</strong>g-term<br />
trend visualisati<strong>on</strong>s for, am<strong>on</strong>g other <strong>in</strong>struments, the<br />
<strong>in</strong>frared sounders <str<strong>on</strong>g>IASI</str<strong>on</strong>g> and AIRS L2 products. Fig. 19,<br />
generated by the NPROVS <strong>on</strong>-l<strong>in</strong>e tool (http://www.star.<br />
nesdis.noaa.gov/smcd/opdb/poes/NPROVS.php, last accessed<br />
23/01/2012), shows the error and l<strong>on</strong>g-term trend for<br />
temperatures retrieved at 700 hPa from different spaceborne<br />
<strong>in</strong>struments. The operati<strong>on</strong>al EUMETSAT <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
product is displayed <strong>in</strong> gold, the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 produced at NOAA<br />
<strong>in</strong> purple, the AIRS L2 temperatures <strong>in</strong> blue and the <strong>retrievals</strong><br />
from the COSMIC (C<strong>on</strong>stellati<strong>on</strong> Observ<strong>in</strong>g System for<br />
Meteorology, I<strong>on</strong>osphere & Climate) missi<strong>on</strong> <strong>in</strong> green. In<br />
additi<strong>on</strong>, numerical weather predicti<strong>on</strong>s are shown <strong>in</strong> red.<br />
The black arrows <strong>in</strong>dicate the <strong>in</strong>troducti<strong>on</strong> of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF<br />
v5. The improvement with respect to the former versi<strong>on</strong> as<br />
compared to the s<strong>on</strong>de measurements appears clearly <strong>in</strong> the<br />
bias (pla<strong>in</strong> thick l<strong>in</strong>e), which drops from about 1 K to around<br />
0.25 K, as well as <strong>in</strong> the standard deviati<strong>on</strong>, which is reduced<br />
by approximately 1 K. The <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 v5 products also perform<br />
very well <strong>in</strong> comparis<strong>on</strong> to the other satellite products. These<br />
c<strong>on</strong>clusi<strong>on</strong>s apply at 500 hPa. At 300 hPa and higher, versi<strong>on</strong><br />
5 does not br<strong>in</strong>g an improvement, with <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 products<br />
hav<strong>in</strong>g good quality <strong>in</strong> comparis<strong>on</strong> to the other satellite<br />
products: the bias towards the s<strong>on</strong>des ranges between 0<br />
and 0.5 K and the standard deviati<strong>on</strong>s is around 1.5 K. At<br />
850 hPa, the departures are larger: the standard deviati<strong>on</strong> is<br />
around 2.5 K for all satellite products and EUMETSAT <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
temperature shows larger biases larger than NOAA’s<br />
temperatures, by 1–1.5 K <strong>on</strong> average, at the level of the<br />
biases characterised for AIRS products. As for the watervapour,<br />
the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 v5 is comparable to v4, as c<strong>on</strong>cluded <strong>in</strong><br />
the previous secti<strong>on</strong>, with departures from <strong>in</strong>-situ<br />
measurements usually slightly larger than for the other<br />
satellite products analysed <strong>in</strong> NPROVS. The reas<strong>on</strong>s for<br />
these relative larger errors are still under <strong>in</strong>vestigati<strong>on</strong>s and<br />
current developments address a different c<strong>on</strong>figurati<strong>on</strong> (e.g.<br />
channel selecti<strong>on</strong>, land surface emissivity <strong>in</strong>puts) as well as<br />
the jo<strong>in</strong>t use of <strong>in</strong>dependent <strong>in</strong>formati<strong>on</strong> from the microwave<br />
measurements, less sensitive to cloud-c<strong>on</strong>tam<strong>in</strong>ati<strong>on</strong>, to<br />
c<strong>on</strong>stra<strong>in</strong> the <strong>retrievals</strong> <strong>in</strong> the lower troposphere.<br />
2.7. The atmospheric compositi<strong>on</strong> products<br />
2.7.1. Oz<strong>on</strong>e products and reference datasets<br />
The <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 O 3 product c<strong>on</strong>sists of a total and three partial<br />
columns from ground to 478.54 hPa ( 6km), 222.94hPa
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1363<br />
( 12 km) and 132.49 hPa ( 16 km). The vertical profile<br />
result<strong>in</strong>g from the OEM (see Secti<strong>on</strong> 2.4.2) is<strong>in</strong>tegratedto<br />
form these columnar amounts and is <strong>in</strong>cluded <strong>in</strong> the f<strong>in</strong>al<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 product. Only the total column is currently distributed<br />
as an operati<strong>on</strong>al product. The partial columns are still<br />
subject to further validati<strong>on</strong> and development.<br />
The validati<strong>on</strong> of the total column (TC) retrieved from<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements <strong>in</strong>volved comparis<strong>on</strong>s to O3 <strong>retrievals</strong><br />
generated by the EUMETSAT SAF <strong>on</strong> oz<strong>on</strong>e and Atmospheric<br />
Chemistry M<strong>on</strong>itor<strong>in</strong>g (O3MSAF) from the Global<br />
Oz<strong>on</strong>e M<strong>on</strong>itor<strong>in</strong>g Experiment-2 (GOME-2) <strong>in</strong>strument, <strong>on</strong><br />
<strong>Metop</strong>-A. This was d<strong>on</strong>e as part of the EUMETSAT <strong>in</strong>ternal<br />
calibrati<strong>on</strong> and validati<strong>on</strong> activities <strong>on</strong> EPS and was completed<br />
by an external study performed by the ‘‘Laboratoire<br />
Atmospheres, Milieux, Observati<strong>on</strong>s Spatiales’’ (LATMOS),<br />
France. GOME-2 is a nadir view<strong>in</strong>g scann<strong>in</strong>g ultraviolet and<br />
visible spectrometer cover<strong>in</strong>g 240–790 nm with spectral<br />
resoluti<strong>on</strong> vary<strong>in</strong>g from 0.26 to 0.51 nm that allows the<br />
daytime retrieval of O 3 am<strong>on</strong>g other species, such as NO 2,<br />
SO2, BrO, OClO and CH2O. It has a ground pixel size of<br />
80 40 km 2 and a swath width of 1920 km comparable to<br />
that of <str<strong>on</strong>g>IASI</str<strong>on</strong>g>, which provides an almost daily global coverage.<br />
For more details, see Munro et al., 2006 (‘‘GOME-2 <strong>on</strong><br />
<strong>Metop</strong>: From <strong>in</strong>-<strong>orbit</strong> verificati<strong>on</strong> to rout<strong>in</strong>e operati<strong>on</strong>s’’,<br />
proceed<strong>in</strong>gs of EUMETSAT Meteorological Satellite C<strong>on</strong>ference,<br />
2006, Hels<strong>in</strong>ki). The accuracy of the GOME-2 O3 TC<br />
product had been assessed us<strong>in</strong>g ground-based <strong>retrievals</strong><br />
from Dobs<strong>on</strong> and Brewer measurements as references [59].<br />
It is shown to have little bias, vary<strong>in</strong>g with locati<strong>on</strong>, seas<strong>on</strong><br />
and view<strong>in</strong>g geometry (e.g. solar zenith and satellite<br />
scann<strong>in</strong>g angles) between 1% and 1% generally, with<br />
larger systematic errors <strong>in</strong> the tropics ( 2–3%). The<br />
standard deviati<strong>on</strong>s typically range from 4–5% with high<br />
correlati<strong>on</strong> score ( 0.94) between the satellite and the<br />
ground-based products.<br />
2.7.2. Intercomparis<strong>on</strong> results<br />
We extract here<strong>after</strong> a summary of the methodology<br />
and the outcome of the external validati<strong>on</strong> study performed<br />
by George and Clerbaux (LATMOS) <strong>in</strong> 2010. <str<strong>on</strong>g>IASI</str<strong>on</strong>g> O 3<br />
TC from the m<strong>on</strong>th of August 2009, reprocessed with the<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF v5, were compared with available observati<strong>on</strong>s<br />
from the thermal <strong>in</strong>frared satellite missi<strong>on</strong>s Aqua/AIRS [7]<br />
and from GOME-2/<strong>Metop</strong>. The reader is referred to the<br />
f<strong>in</strong>al report (George and Clerbaux, July 2010, ‘‘Validati<strong>on</strong><br />
Study for <str<strong>on</strong>g>IASI</str<strong>on</strong>g> Trace Gas Retrievals’’) for full details. The<br />
comparis<strong>on</strong> was performed for cloud-free <strong>retrievals</strong> (as per<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 flags) <strong>on</strong> m<strong>on</strong>thly averages over a 11 11 latitude/<br />
l<strong>on</strong>gitude grid. The correlati<strong>on</strong> between the O3 TC from<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> and with the two reference products is about 0.9. The<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> products are <strong>on</strong> average 1.5% higher than the operati<strong>on</strong>al<br />
product from GOME-2 (day) and approximately also<br />
1.5% lower than the AIRS (day and night) <strong>retrievals</strong>. The<br />
best agreement is found <strong>in</strong> the Southern Hemisphere<br />
(latitude <strong>in</strong> [451S; 151S]), where EUMETSAT <str<strong>on</strong>g>IASI</str<strong>on</strong>g> products<br />
show a slight positive bias of 0.35% aga<strong>in</strong>st GOME-2 O 3 TC<br />
and 0.43% (0.37% night) aga<strong>in</strong>st AIRS products. The correlati<strong>on</strong>s<br />
are of 0.97 and 0.95 (0.96 at night) with GOME-2 and<br />
AIRS products, respectively.<br />
In the background regi<strong>on</strong>s with low c<strong>on</strong>centrati<strong>on</strong>s,<br />
the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> O3 is larger than <strong>in</strong> the GOME-2 and AIRS<br />
products. In the <strong>in</strong>ter-tropical band (latitudes between<br />
[151S; 151N]), the bias <strong>in</strong> <str<strong>on</strong>g>IASI</str<strong>on</strong>g> products is of 2.1% and the<br />
departures are with<strong>in</strong> 0% and 5%. The correlati<strong>on</strong> is of<br />
0.93. In c<strong>on</strong>trast, the range of differences with AIRS is<br />
much broader, spann<strong>in</strong>g a 10–15% <strong>in</strong>terval. The correlati<strong>on</strong><br />
coefficient between these <str<strong>on</strong>g>IASI</str<strong>on</strong>g> and AIRS data samples<br />
is of 0.69 <strong>on</strong>ly. These m<strong>on</strong>thly average results are c<strong>on</strong>sistent<br />
with a pixel-to-pixel <strong>in</strong>ter-comparis<strong>on</strong> study performed<br />
at EUMETSAT for the period 19–24 March 2010 at<br />
latitudes below 701. Fig. 20 shows the maps (GOME-2–<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g>) bias and standard deviati<strong>on</strong>. Quantitatively, <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
products have an overall positive bias of approximately<br />
2.6%, with standard deviati<strong>on</strong>s below 3%. Larger discrepancies<br />
are observed over the Sahara, with s gett<strong>in</strong>g as<br />
high as 6–9%, which are assumed to result from <strong>in</strong>accurate<br />
surface parameter sett<strong>in</strong>gs and possible dust c<strong>on</strong>tam<strong>in</strong>ati<strong>on</strong>;<br />
<strong>in</strong>vestigati<strong>on</strong>s are still <strong>on</strong>go<strong>in</strong>g <strong>in</strong> this area. As<br />
reported <strong>in</strong> the validati<strong>on</strong> of GOME-2 products with<br />
ground-based data, the average relative difference<br />
(GOME-2 <str<strong>on</strong>g>IASI</str<strong>on</strong>g>) varies with latitude from about 4% (<str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
higher) around the Equator and <strong>in</strong> Southern tropical<br />
latitudes to little or no bias (o1%) at Northern midlatitudes.<br />
Not presented here, the departures assessed as a<br />
functi<strong>on</strong> of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> scan positi<strong>on</strong> and showed disc<strong>on</strong>t<strong>in</strong>uities<br />
of the order of 1% which are attributed to scan<br />
angle dependencies, known and corrected for <strong>in</strong> the<br />
reference product [59].<br />
In a sec<strong>on</strong>d step, the operati<strong>on</strong>al <str<strong>on</strong>g>IASI</str<strong>on</strong>g> O3 TC was<br />
compared to the research products generated at the ‘‘Université<br />
Libre de Bruxelles’’ (ULB) <strong>in</strong> collaborati<strong>on</strong> with the<br />
LATMOS [25,60]. They are based <strong>on</strong> the retrieval software<br />
Fast Optimal Retrievals <strong>on</strong> Layers for <str<strong>on</strong>g>IASI</str<strong>on</strong>g> (FORLI) [61],<br />
implement<strong>in</strong>g an OEM <strong>after</strong> Rodgers [45], and the radiative<br />
transfer model Atmosphit. Independent validati<strong>on</strong> of this<br />
product were carried out with ground-based <strong>retrievals</strong> by<br />
Ant<strong>on</strong> et al. [62] over the Iberian Pen<strong>in</strong>sula as well as with<br />
<strong>in</strong>-situ s<strong>on</strong>de measurements (G. Dufour, 2011, <strong>in</strong> discussi<strong>on</strong>:<br />
‘‘Validati<strong>on</strong> of three different scientific oz<strong>on</strong>e products<br />
retrieved from <str<strong>on</strong>g>IASI</str<strong>on</strong>g> spectra us<strong>in</strong>g oz<strong>on</strong>es<strong>on</strong>des’’, Atmos.<br />
Meas. Tech. Discuss., 4, 5425–5479, 2011, doi:10.5194/<br />
amtd-4-5425-2011). In [62] FORLI-O3 <strong>retrievals</strong> are shown<br />
to overestimate the O 3 TC by 4.4%, with standard deviati<strong>on</strong>s<br />
of error of 3– 5%. Dufour, 2011, evaluated the global error<br />
budget to approximately 5%. The correlati<strong>on</strong> with reference<br />
O3 TC is high <strong>in</strong> both studies, usually above 0.9.<br />
EUMETSAT reprocessed v5 and FORLI O 3 TC were<br />
directly <strong>in</strong>ter-compared for the whole m<strong>on</strong>th of August<br />
2009, <strong>on</strong> a pixel-to-pixel basis. The two products correlate<br />
well, with a global correlati<strong>on</strong> coefficient larger than 0.9.<br />
The overall (EUMETSAT–FORLI) biases (and associated<br />
standard deviati<strong>on</strong>) are 1.7% (3.6%) for day time and<br />
2.1% (3.6%) for night time. EUMETSAT O 3 values are<br />
generally smaller than FORLI and, <strong>in</strong> the light of FORLI<br />
validati<strong>on</strong> results referred to previously, have less bias<br />
with respect to the true O3 c<strong>on</strong>centrati<strong>on</strong>s. The best<br />
correlati<strong>on</strong>s are observed <strong>in</strong> the Southern Hemisphere<br />
(latitudes [ 451; 151]) where high O3 c<strong>on</strong>centrati<strong>on</strong>s<br />
are measured, with correlati<strong>on</strong> coefficients of 0.96 (for<br />
day and night time). In this regi<strong>on</strong>, the mean biases (and<br />
associated standard deviati<strong>on</strong>) are 2.2% (2.7%) for day<br />
time and 2.5% (2.6%) for night time. Specific topical case
1364<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
Fig. 20. Maps of the (GOME-2–<str<strong>on</strong>g>IASI</str<strong>on</strong>g> v5) O 3 relative bias (top) and standard deviati<strong>on</strong> (bottom) for the period 19–24 March 2010.<br />
studies were performed <strong>in</strong> 51 51 latitude/l<strong>on</strong>gitude<br />
boxes, address<strong>in</strong>g regi<strong>on</strong>s of high O 3 c<strong>on</strong>centrati<strong>on</strong>s<br />
(e.g. al<strong>on</strong>g Polar vortex) and background O3 regi<strong>on</strong>s (such<br />
as the mid-Atlantic, Tropical forests and Africa), as well as<br />
over polluted (Ch<strong>in</strong>a) and urban (Teheran and San Francisco)<br />
areas. They c<strong>on</strong>firmed the global figures, but with<br />
larger amplitude with high O3 load (bias 3.5%) as<br />
compared to O 3 background locati<strong>on</strong>s (bias 1.8%).<br />
As part of the C<strong>on</strong>t<strong>in</strong>uous Development and Operati<strong>on</strong>s<br />
Phase (CDOP-2) of the O3M-SAF (o3msaf.fmi.fi), the<br />
FORLI-O3 algorithm will be <strong>in</strong>tegrated to the operati<strong>on</strong>al<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 process<strong>in</strong>g cha<strong>in</strong>. Additi<strong>on</strong>al validati<strong>on</strong> work is<br />
foreseen, <strong>in</strong>clud<strong>in</strong>g comparis<strong>on</strong> to oz<strong>on</strong>e s<strong>on</strong>des and<br />
additi<strong>on</strong>al satellite products.<br />
2.7.3. Carb<strong>on</strong> m<strong>on</strong>oxide 1<br />
The operati<strong>on</strong>al CO product <strong>in</strong> <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 v5 is the total<br />
column amount retrieved with the artificial neural<br />
1 The studies <strong>in</strong>volv<strong>in</strong>g real <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements were performed<br />
with the pixels 3 and 4 <strong>on</strong>ly. Because of an <strong>in</strong>ter-pixel difference <strong>in</strong> the<br />
L1C radiances (which is not discussed <strong>in</strong> this paper), mostly affect<strong>in</strong>g the<br />
CO and N 2O regi<strong>on</strong> and to which the CO retrieval was sensitive, the<br />
network <strong>in</strong>troduced <strong>in</strong> Secti<strong>on</strong> 2.4.1. As such, it does not<br />
<strong>in</strong>clude vertical sensitivity and error estimates, as can be<br />
obta<strong>in</strong>ed with the OEM for other parameters. Similar to O3<br />
<strong>in</strong> Secti<strong>on</strong> 2.7.1, its performance was first assessed <strong>in</strong>ternally<br />
at EUMETSAT and the validati<strong>on</strong> was completed by an<br />
external study performed by LATMOS. We present here a<br />
summary of the rati<strong>on</strong>ale and outcome of the performance<br />
assessments characteris<strong>in</strong>g the updated CO product which<br />
became operati<strong>on</strong>al <strong>on</strong> 14/09/2010, with the release of the<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 v5. This addresses the theoretical accuracy as well<br />
as <strong>in</strong>ter-comparis<strong>on</strong> with other satellite products.<br />
2.7.3.1. Theoretical performances. The theoretical performances<br />
can be assessed with the database used to teach<br />
the artificial neural network. An overview of its compositi<strong>on</strong><br />
and associated results is given here, a more detailed<br />
(footnote c<strong>on</strong>t<strong>in</strong>ued)<br />
operati<strong>on</strong>al producti<strong>on</strong> of <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 CO (and N 2O) TC <strong>on</strong> pixels 1 and 2 was<br />
<strong>in</strong>terrupted with the release of the PPF v5. The radiance <strong>in</strong>ter-pixel<br />
difference was reduced with an update of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L1c process<strong>in</strong>g cha<strong>in</strong><br />
<strong>on</strong> 07/02/2011 and the producti<strong>on</strong> of L2 <strong>retrievals</strong> for the four IFOVs<br />
resumed <strong>on</strong> 14/03/2011.
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1365<br />
Fig. 21. CO absolute (left) and relative (right) tra<strong>in</strong><strong>in</strong>g errors. The overall statistics are displayed <strong>in</strong> black while the fitt<strong>in</strong>g Gaussian and associated<br />
numbers are shown <strong>in</strong> red. (For <strong>in</strong>terpretati<strong>on</strong> of the references to color <strong>in</strong> this figure legend, the reader is referred to the web versi<strong>on</strong> of this article.)<br />
presentati<strong>on</strong> and discussi<strong>on</strong> <strong>on</strong> the performance is available<br />
<strong>on</strong>-l<strong>in</strong>e <strong>in</strong> EUMETSAT Technical Note EUM/MET/TEN/09/<br />
0232 cited <strong>in</strong> Secti<strong>on</strong> 2.4.1. The tra<strong>in</strong><strong>in</strong>g base approximately<br />
c<strong>on</strong>ta<strong>in</strong>s 200,000 patterns made up of atmospheric<br />
state vectors and their associated synthetic <str<strong>on</strong>g>IASI</str<strong>on</strong>g> spectra<br />
computed with RT<str<strong>on</strong>g>IASI</str<strong>on</strong>g>-5.3. The atmospheric temperature,<br />
humidity surface pressure and w<strong>in</strong>d comp<strong>on</strong>ents are based<br />
<strong>on</strong> the climatological database from Chevallier, 2001, cited<br />
<strong>in</strong> Secti<strong>on</strong> 2.4.2.3. As trace gas profiles are not part of this<br />
dataset, they were generated to cover the whole range<br />
of expected situati<strong>on</strong>s with random variati<strong>on</strong>s around<br />
standard profiles. In the case of CO, the vertical distributi<strong>on</strong>s<br />
are based <strong>on</strong> 43 orig<strong>in</strong>al profiles sampled by D.<br />
Cunnold, 2001 (pers<strong>on</strong>al communicati<strong>on</strong>) from the<br />
MOZART 3D chemical transport model calculati<strong>on</strong>s [63].<br />
To generate a realistic and c<strong>on</strong>t<strong>in</strong>uous set of scenarios for<br />
CO, the selected profile was subsequently randomly either<br />
left unchanged or varied by add<strong>in</strong>g up to half of the<br />
variability (max m<strong>in</strong>) of the mix<strong>in</strong>g ratio <strong>in</strong> the basic 43level<br />
modelled vertical distributi<strong>on</strong>s. By c<strong>on</strong>structi<strong>on</strong>, the<br />
tra<strong>in</strong><strong>in</strong>g set, which can be seen as the background <strong>in</strong> the<br />
OEM, is characterised by a high vertical correlati<strong>on</strong>. As for<br />
surface parameters, elevated areas are by essence <strong>in</strong>cluded<br />
<strong>in</strong> the tra<strong>in</strong><strong>in</strong>g set which also comprises a wide range of<br />
surface types. Over water, the surface emissivity was<br />
computed analytically [15,16]. Ground emissivities were<br />
derived from the MODIS UCSB emissivity library as<br />
follows: based <strong>on</strong> the spectra of pure surfaces, composite<br />
surfaces have been generated with random c<strong>on</strong>tributi<strong>on</strong>s<br />
from up to three different types, exclud<strong>in</strong>g however<br />
comb<strong>in</strong>ati<strong>on</strong>s like snow/ice at tropical temperatures.<br />
Random noise as per the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>in</strong>strument noise<br />
characteristics was added to the <strong>in</strong>put radiances <strong>in</strong> order<br />
to regularise the tra<strong>in</strong><strong>in</strong>g [64]. Neither clouds nor aerosols<br />
were <strong>in</strong>cluded to compute the synthetic radiances such that<br />
the networks learnt pure clear cases <strong>on</strong>ly, which are also<br />
subsequently their nom<strong>in</strong>al doma<strong>in</strong> of validity.<br />
The teach<strong>in</strong>g database was split <strong>in</strong>to a tra<strong>in</strong><strong>in</strong>g and a<br />
c<strong>on</strong>trol set. To avoid overtra<strong>in</strong><strong>in</strong>g, i.e. the net becom<strong>in</strong>g too<br />
specific to the tra<strong>in</strong><strong>in</strong>g patterns and los<strong>in</strong>g its generalisati<strong>on</strong><br />
ability, <strong>on</strong>e m<strong>on</strong>itors the retrieval error of the c<strong>on</strong>trol set and<br />
stops the learn<strong>in</strong>g before it starts diverg<strong>in</strong>g from the tra<strong>in</strong><strong>in</strong>g<br />
error. The correlati<strong>on</strong> between retrieved and target columns<br />
is 0.99 and the l<strong>in</strong>ear relati<strong>on</strong> is very close to unity (slope<br />
0.97). Due to the n<strong>on</strong>-l<strong>in</strong>ear nature of the MLP, the errors<br />
are not statistically Gaussian-like, as can be seen <strong>in</strong> Fig. 21.<br />
The absolute rms error typically lies between 0.2 and<br />
0.24 10 18 molecule/cm 2 (Fig. 21, left) and is <strong>in</strong>dependent<br />
of the column density itself (not shown here). As a result, the<br />
relative errors are higher for the th<strong>in</strong>nest columns. For total<br />
columns higher than 0.7 10 18 molecule/cm 2 , the relative<br />
error ranges from 7% to 11% standard deviati<strong>on</strong> with small<br />
bias below 1%. These figures are c<strong>on</strong>sistent with an error of<br />
11% as characterised <strong>in</strong> a <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 CO product assimilati<strong>on</strong><br />
experiment <strong>in</strong> the chemistry and transport model MOCAGE<br />
[65] (El Amraoui 2011, pers<strong>on</strong>al communicati<strong>on</strong>).<br />
2.7.3.2. Intercomparis<strong>on</strong>s with other satellite products.<br />
Follow<strong>in</strong>g the approach of George et al. [66], the<br />
characteristics of the operati<strong>on</strong>al CO product were<br />
evaluated aga<strong>in</strong>st other space-borne CO products, with<br />
l<strong>on</strong>ger operati<strong>on</strong>al and validati<strong>on</strong> heritage. Namely, they<br />
are <strong>retrievals</strong> from the Measurement Of Polluti<strong>on</strong> In The<br />
Troposphere (MOPITT) <strong>in</strong>strument, from Aqua/AIRS and<br />
from the Tropospheric Emissi<strong>on</strong> Spectrometer (TES)<br />
<strong>on</strong>board Terra. We summarise here<strong>after</strong> the results of the<br />
comparis<strong>on</strong>s aga<strong>in</strong>st MOPITT CO as obta<strong>in</strong>ed at EUMETSAT<br />
and <strong>in</strong> the frame of the external study performed by<br />
George and Clerbaux (LATMOS). The full details can be<br />
c<strong>on</strong>sulted <strong>on</strong>-l<strong>in</strong>e (www.eumetsat.<strong>in</strong>t) <strong>in</strong> the EUMETSAT<br />
technical Note EUM/MET/TEN/09/0232, and <strong>in</strong> the<br />
validati<strong>on</strong> report (George, July 2010) cited <strong>in</strong> Secti<strong>on</strong> 2.7.2.<br />
MOPITT is part of the polar <strong>orbit</strong><strong>in</strong>g Terra satellite<br />
payload and was launched <strong>in</strong> December 1999. The ground<br />
pixel size is larger than for <str<strong>on</strong>g>IASI</str<strong>on</strong>g>, 22 22 km 2 at Nadir, and<br />
has a swath width approximately half the size of <str<strong>on</strong>g>IASI</str<strong>on</strong>g>’s,<br />
allow<strong>in</strong>g a global coverage <strong>in</strong> 3 days. Terra and <strong>Metop</strong> are<br />
<strong>on</strong> sun-synchr<strong>on</strong>eous Polar <strong>orbit</strong>s, with slightly different<br />
parameters and a descend<strong>in</strong>g node at 10.30 and 09.30,<br />
respectively, such that correlative <str<strong>on</strong>g>IASI</str<strong>on</strong>g> and MOPITT
1366<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
Fig. 22. CO total column retrieved from MOPITT (top) and <str<strong>on</strong>g>IASI</str<strong>on</strong>g> (bottom) dur<strong>in</strong>g 25–31 August 2008.<br />
sens<strong>in</strong>g were distant by <strong>on</strong>e hour <strong>on</strong> average <strong>in</strong> this study.<br />
The MOPITT CO versi<strong>on</strong> 3 products were used <strong>in</strong> this<br />
work. They are based <strong>on</strong> the observati<strong>on</strong>s <strong>in</strong> the 4.7 mm<br />
CO fundamental band <strong>on</strong>ly and exploit the same spectral<br />
regi<strong>on</strong> as the <strong>retrievals</strong> from <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements. The<br />
profiles are retrieved <strong>on</strong> 7 vertical levels us<strong>in</strong>g a n<strong>on</strong>l<strong>in</strong>ear<br />
optimal estimati<strong>on</strong> method <strong>in</strong>clud<strong>in</strong>g a unique and<br />
global a priori [67]. They are distributed with the <strong>in</strong>tegrated<br />
total column, an estimate of vertical sensitivity<br />
(c<strong>on</strong>tributi<strong>on</strong> of background a priori vs. retrieved <strong>in</strong>formati<strong>on</strong>)<br />
and a cloud mask (no <strong>retrievals</strong> <strong>in</strong> cloudy pixels).<br />
The degrees of freedom for signal, i.e. the number of<br />
<strong>in</strong>dependent pieces of <strong>in</strong>formati<strong>on</strong> <strong>in</strong> the retrieved profiles,<br />
are usually maximal <strong>in</strong> the tropics, where they do<br />
not exceed 2, and drop down to 1 or even below 1 at mid<br />
and Polar latitudes [68].<br />
In the assessment performed at EUMETSAT, we compared<br />
s<strong>in</strong>gle <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>retrievals</strong> to L3 Daily MOPITT CO total<br />
column, gridded <strong>in</strong> 11 11 l<strong>on</strong>gitude/latitude b<strong>in</strong>s for<br />
daytime and night time sens<strong>in</strong>g separately. Departures<br />
were computed for the whole m<strong>on</strong>ths of August and<br />
November 2008 between each cloud-free <str<strong>on</strong>g>IASI</str<strong>on</strong>g> IFOV (as<br />
per <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 flags) and the average retrieval <strong>in</strong> the corresp<strong>on</strong>d<strong>in</strong>g<br />
day/night nearest MOPITT product grid po<strong>in</strong>t.<br />
The MOPITT CO profiles with average c<strong>on</strong>tributi<strong>on</strong>s from<br />
the background a priori larger than 50% were discarded<br />
from the <strong>in</strong>ter-comparis<strong>on</strong>. This mostly occurs at higher<br />
latitudes or at night, when the surface cools down and has<br />
no thermal c<strong>on</strong>trast with the boundary layer [25]. Asan<br />
example, we present qualitative comparis<strong>on</strong> MOPITT vs.<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> CO for the last week <strong>in</strong> August 2008 <strong>in</strong> Fig. 22. The<br />
ma<strong>in</strong> CO patterns are observed <strong>in</strong> both products: biomass<br />
burn<strong>in</strong>g <strong>in</strong> Africa and South-America, as well as polluti<strong>on</strong><br />
<strong>in</strong> Eastern Asia, transported out over the Pacific and<br />
between North America and Europe [69]. Differences<br />
between the two products are visible <strong>in</strong> the Sahara, off<br />
the coast of California and west coast of South America.<br />
The latter discrepancy is likely caused by undetected<br />
clouds <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 process<strong>in</strong>g as there is no or little<br />
counterpart <strong>in</strong> the MOPITT products, which are generated<br />
for clear-sky <strong>on</strong>ly. Quantitatively, the global correlati<strong>on</strong> is<br />
of 0.82 and 0.79 for August and November, respectively.<br />
The statistics were broken down <strong>in</strong> latitude bands and<br />
day/night c<strong>on</strong>diti<strong>on</strong>s. The results for August 2008 are<br />
repeated <strong>in</strong> Table 7, the c<strong>on</strong>clusi<strong>on</strong>s are similar for<br />
November 2008. The best agreement is found between<br />
301S and 601N; especially <strong>in</strong> the tropical regi<strong>on</strong>s with a<br />
correlati<strong>on</strong> as high as 0.86 and a small relative bias below<br />
3%. Overall, the standard deviati<strong>on</strong> (MOPITT–<str<strong>on</strong>g>IASI</str<strong>on</strong>g>) CO TC<br />
range is 0.17–0.25 10 18 molecules/cm 2 , or 8–15% <strong>in</strong><br />
relative terms. <str<strong>on</strong>g>IASI</str<strong>on</strong>g> CO TC underestimates MOPITT <strong>retrievals</strong>:<br />
the bias varies with the latitude and larger systematic<br />
differences from 10% to 20% are observed <strong>in</strong> the North<br />
Pole and at higher Southern latitudes, exceed<strong>in</strong>g even 30%<br />
over Antarctica. They corresp<strong>on</strong>d to areas with colder<br />
surface temperature and lower thermal c<strong>on</strong>trast, where<br />
the respective <strong>retrievals</strong> are expected to be more <strong>in</strong>fluenced<br />
by the background a priori. The comparis<strong>on</strong> was<br />
extended by George (2010) to February 2009 and performed<br />
<strong>on</strong> m<strong>on</strong>thly averages. The results c<strong>on</strong>firmed the<br />
characterisati<strong>on</strong> of an average bias of about 10% as<br />
compared to MOPITT CO (<str<strong>on</strong>g>IASI</str<strong>on</strong>g> CO lower) and high correlati<strong>on</strong><br />
coefficients around 0.9 <strong>in</strong> the tropics for all three<br />
m<strong>on</strong>ths.
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1367<br />
Table 7<br />
Statistics of departures (MOPITT–<str<strong>on</strong>g>IASI</str<strong>on</strong>g>) CO total column <strong>in</strong> 10 18 molecules/cm 2 for different latitudes <strong>in</strong> August 2008 (bias, standard deviati<strong>on</strong>, correlati<strong>on</strong><br />
coefficient and sample size are stored <strong>in</strong> columns).<br />
August 2008<br />
Day Night<br />
Bias s r # Bias s r #<br />
[601N; 901N] 0.14 0.21 0.66 86106 0.22 0.17 0.69 4713<br />
[301N; 601N] 0.06 0.23 0.72 127351 0.01 0.25 0.72 104175<br />
[301S; 301N] 0.07 0.21 0.84 250628 0.04 0.22 0.86 237613<br />
[601S; 301S] 0.33 0.20 0.68 39146 0.29 0.21 0.63 37801<br />
[901S; 601S] 0.63 0.12 0.38 431 0.54 0.18 0.32 1795<br />
2.7.3.3. Intercomparis<strong>on</strong>s with FORLI-CO. The operati<strong>on</strong>al<br />
total column was also evaluated aga<strong>in</strong>st the research<br />
product retrieved from <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements with FORLI-<br />
CO at ULB/LATMOS [25]. FORLI-CO is a dedicated versi<strong>on</strong><br />
of the OEM research algorithm <strong>in</strong>troduced <strong>in</strong> Secti<strong>on</strong> 2.7.1<br />
[61], allow<strong>in</strong>g the retrieval of CO profiles under clear-sky<br />
assumpti<strong>on</strong>s. The FORLI-CO products were validated<br />
aga<strong>in</strong>st <strong>in</strong>-situ airborne measurements <strong>in</strong> the specific<br />
Polar c<strong>on</strong>text [70], where correlati<strong>on</strong> from 0.15 to 0.84<br />
and biases between 5% and 12% with respect to the<br />
reference data are reported. The best match is achieved<br />
<strong>in</strong> situati<strong>on</strong>s with favourable thermal c<strong>on</strong>trasts. George<br />
et al. [66] assessed their accuracy aga<strong>in</strong>st <strong>retrievals</strong> from<br />
other space-borne sensors (MOPITT, AIRS and TES) [66]<br />
and characterised a high correlati<strong>on</strong> coefficient, between<br />
0.83 and 0.94 between FORLI-CO and the respective<br />
correlative satellite products. Typically 2.5 DoFs were<br />
found <strong>in</strong> the tropics, decreas<strong>in</strong>g to 1.5 <strong>on</strong> average at<br />
mid-latitudes and down to 0.8 at higher latitudes,<br />
towards the Poles. The study c<strong>on</strong>firmed that Polar<br />
<strong>retrievals</strong>, especially over Antarctica, were essentially<br />
c<strong>on</strong>stituted by the background a priori. The ma<strong>in</strong><br />
difference between the FORLI-CO and the EUMETSAT<br />
operati<strong>on</strong>al CO TC is <strong>in</strong> the numerical method employed<br />
(OEM vs. ANN, respectively), which allows the <strong>retrievals</strong><br />
of profiles with error and vertical sensitivity estimates <strong>on</strong><br />
<strong>on</strong>e hand (FORLI) and is limited to the total column <strong>on</strong> the<br />
other hand (EUMETSAT). Another important difference is<br />
the process<strong>in</strong>g of cloud-c<strong>on</strong>tam<strong>in</strong>ated pixels with FORLI<br />
while the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PPF strictly limits the CO <strong>retrievals</strong> to<br />
cloud-free IFOVs. Indeed, FORLI implements a cloud test<br />
similar to the NWP test described <strong>in</strong> Secti<strong>on</strong> 2.2.1 with a<br />
less c<strong>on</strong>servative threshold and tolerates cloud<br />
c<strong>on</strong>tam<strong>in</strong>ated pixels with potential cloud fracti<strong>on</strong>s of up<br />
to 25% ([25] Secti<strong>on</strong> 3.2.1). As a result, the retrieved CO TC<br />
yield and therefore its coverage are much higher with<br />
FORLI-CO.<br />
In an external study performed at LATMOS <strong>in</strong> 2010,<br />
George and Clerbaux compared the two <str<strong>on</strong>g>IASI</str<strong>on</strong>g> CO products<br />
<strong>on</strong> a direct pixel-to-pixel basis, for the m<strong>on</strong>ths of August<br />
2008, November 2008 and February 2009. Only cloud-free<br />
pixels, as per <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 flags, were reta<strong>in</strong>ed for the <strong>in</strong>tercomparis<strong>on</strong>.<br />
A summary of the results is reported here,<br />
the details can be c<strong>on</strong>sulted <strong>in</strong> the validati<strong>on</strong> study f<strong>in</strong>al<br />
report. For the three m<strong>on</strong>ths <strong>in</strong>vestigated, EUMETSAT CO<br />
distributi<strong>on</strong>s match the FORLI distributi<strong>on</strong>s well. For data<br />
over all latitudes, the (EUMETSAT–FORLI) biases (and<br />
associated standard deviati<strong>on</strong>s) are of 2.1% (10.6%), 3.1%<br />
(11.5%) and 1.4% (15.2%), respectively, for August 2008,<br />
November 2008 and February 2009. We observe the best<br />
correlati<strong>on</strong>s <strong>in</strong> the equatorial regi<strong>on</strong> (latitudes between<br />
151S and 151N) with respective correlati<strong>on</strong> coefficients of<br />
0.93, 0.92 and 0.94. In this regi<strong>on</strong> the agreement is<br />
excellent with the slopes of the regressi<strong>on</strong> l<strong>in</strong>es close to<br />
1 and the biases (standard deviati<strong>on</strong>) are 1.4% (7.9%), 4.7%<br />
(8.2%) and 1.2% (7.3%), respectively, for August 2008,<br />
November 2008 and February 2009. The correlati<strong>on</strong> drops<br />
to 0.52 at boreal latitudes but rema<strong>in</strong>s around 0.8 at midlatitudes.<br />
In regi<strong>on</strong>s of low background CO c<strong>on</strong>centrati<strong>on</strong>s<br />
(Atlantic, Pacific) as well as over forests, EUMETSAT CO TC<br />
is c<strong>on</strong>sistently higher than FORLI-CO, with biases of 5–7%<br />
and departures standard deviati<strong>on</strong>s of 6.5–9%. In biomass<br />
burn<strong>in</strong>g regi<strong>on</strong>s, the bias between EUMETSAT and FORLI<br />
products is small; 1% (EUMETSAT lower) with 9%<br />
relative standard deviati<strong>on</strong>. In the c<strong>on</strong>text of high polluti<strong>on</strong><br />
(Ch<strong>in</strong>a), however, EUMETSAT CO TC is noticeably<br />
smaller than FORLI by 6.7% <strong>on</strong> average, with a standard<br />
deviati<strong>on</strong> of 11.6%. Some retrieval artefacts over the<br />
deserts observed <strong>in</strong> FORLI-CO maps, <strong>in</strong>duced by <strong>in</strong>accurate<br />
surface emissivity [70], are not seen <strong>in</strong> EUMETSAT<br />
products, which show a higher spatial coherence <strong>in</strong> these<br />
regi<strong>on</strong>s. C<strong>on</strong>versely, a few dubious high CO c<strong>on</strong>centrati<strong>on</strong>s<br />
were observed <strong>in</strong> EUMETSAT products off the<br />
Namibian and Californian coasts. The current assumpti<strong>on</strong><br />
attributes this feature to undetected low level clouds, to<br />
which FORLI-CO seems less sensitive.<br />
This good agreement between both <str<strong>on</strong>g>IASI</str<strong>on</strong>g> CO products<br />
was c<strong>on</strong>firmed <strong>in</strong> a recent case study evaluat<strong>in</strong>g the<br />
ability of space-borne <strong>in</strong>frared sounders to quantify the<br />
Russian fire CO emissi<strong>on</strong>s with validati<strong>on</strong> aga<strong>in</strong>st groundbased<br />
measurements and <strong>retrievals</strong> [71]. The FORLI-CO<br />
profiles have been m<strong>on</strong>itored and assimilated <strong>in</strong> the<br />
numerical model run <strong>in</strong> the scope of the M<strong>on</strong>itor<strong>in</strong>g<br />
Atmospheric Compositi<strong>on</strong> and Climate (MACC) project<br />
s<strong>in</strong>ce February 2009. MACC is the current pre-operati<strong>on</strong>al<br />
atmospheric service of the European GMES programme<br />
(www.gmes-atmosphere.eu, last accessed 23/01/2012). It<br />
provides data records <strong>on</strong> atmospheric compositi<strong>on</strong> for<br />
recent <strong>years</strong>, data for m<strong>on</strong>itor<strong>in</strong>g present c<strong>on</strong>diti<strong>on</strong>s and<br />
forecasts of the distributi<strong>on</strong> of key c<strong>on</strong>stituents for a few<br />
days ahead (Inness et al., 2009,‘‘GEMS data assimilati<strong>on</strong><br />
system for chemically reactive gases’’, ECMWF Technical<br />
Memorandum 587) [72]. With the <strong>in</strong>troducti<strong>on</strong> of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
L2 PPF v5, the CO TC generated at EUMETSAT was
1368<br />
m<strong>on</strong>itored <strong>in</strong> this c<strong>on</strong>text dur<strong>in</strong>g the whole m<strong>on</strong>th of<br />
September 2010. Fig. 23 (Inness, 2010, pers<strong>on</strong>al communicati<strong>on</strong>)<br />
illustrates the noticeable improvement of<br />
EUMETSAT CO product with the <strong>in</strong>troducti<strong>on</strong> of <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
v5 <strong>on</strong> 14/09/2010 as compared aga<strong>in</strong>st the MACC model<br />
first guess (blue) and analyses (red), as well as its overall<br />
good quality as compared to FORLI-CO total column.<br />
Future developments <strong>in</strong> this area will address the<br />
operati<strong>on</strong>al producti<strong>on</strong> of CO profiles with error estimate<br />
to allow the assimilati<strong>on</strong> <strong>in</strong>to numerical models. In that<br />
perspective, as part of the C<strong>on</strong>t<strong>in</strong>uous Development and<br />
Operati<strong>on</strong>s Phase (CDOP-2) of the O3M-SAF, the FORLI-CO<br />
retrieval will ultimately become part of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
processors operated at EUMETSAT central applicati<strong>on</strong><br />
facility (CAF).<br />
2.7.4. N2O, CH4 and CO2<br />
The N2O, CH4 and CO2 products planned <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2<br />
trace gases suite are not operati<strong>on</strong>al. These <strong>retrievals</strong> are<br />
still subject to research and development. The total<br />
columns are currently retrieved with <strong>in</strong>dividual artificial<br />
neural networks similar <strong>in</strong> c<strong>on</strong>cept to the network used<br />
for CO. They benefited from the same algorithm improvements<br />
<strong>in</strong>troduced <strong>in</strong> <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 PFF v5 and are currently<br />
produced <strong>in</strong> an experimental mode. Ricaud et al. [73]<br />
studied the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 v4 N 2O dur<strong>in</strong>g the m<strong>on</strong>ths of March–<br />
April–May 2008 and found a reas<strong>on</strong>able agreement<br />
between retrieved N 2O total column and transported<br />
patterns modelled with MOCAGE [73]. In order to extend<br />
these assessment studies, the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 trace gas products<br />
were recently reprocessed with the PPF v5 for the whole<br />
year of 2008 and the m<strong>on</strong>ths of February, May, August<br />
and November <strong>in</strong> the <strong>years</strong> 2009–2011.<br />
The synergistic use of <str<strong>on</strong>g>IASI</str<strong>on</strong>g> with microwave measurements<br />
from <strong>Metop</strong>/AMSU is anticipated to improve the<br />
separati<strong>on</strong> of temperature and the trace gas (CH 4,CO 2)<br />
<strong>in</strong>formati<strong>on</strong> present <strong>in</strong> the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> spectra [74,75].<br />
3. C<strong>on</strong>clusi<strong>on</strong>s and outlook<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
Fig. 23. Departures m<strong>on</strong>itor<strong>in</strong>g of <str<strong>on</strong>g>IASI</str<strong>on</strong>g> CO observati<strong>on</strong>s—MACC model (first guess <strong>in</strong> blue, analyses <strong>in</strong> red) <strong>in</strong> September 2010. Credits: A. Inness<br />
(ECMWF). (For <strong>in</strong>terpretati<strong>on</strong> of the references to color <strong>in</strong> this figure legend, the reader is referred to the web versi<strong>on</strong> of this article.)<br />
We have presented the structure of the operati<strong>on</strong>al <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
L2 process<strong>in</strong>g cha<strong>in</strong> versi<strong>on</strong> 5, which became operati<strong>on</strong>al<br />
<strong>on</strong> 14/09/2010; the <strong>in</strong>dividual retrieval modules and their<br />
algorithms and c<strong>on</strong>figurati<strong>on</strong>, a summary of the performance<br />
assessment through various <strong>in</strong>ternal and external<br />
validati<strong>on</strong> studies, and the current and future development<br />
activities <strong>after</strong> 5 <strong>years</strong> of operati<strong>on</strong>s at EUMETSAT. The<br />
validati<strong>on</strong> of the various retrieved geophysical parameters<br />
has been c<strong>on</strong>ducted with a wide range of satellite products<br />
(CALIOP, AATSR, MODIS, GOME-2, AVHRR, MOPITT, SEVIRI,<br />
etc.), with numerical weather predicti<strong>on</strong> and chemistry<br />
models (ECMWF, MOCAGE, MACC) and with <strong>in</strong>-situ measurements<br />
(radio-s<strong>on</strong>des, buoys etc.). Due to the evolv<strong>in</strong>g<br />
nature of the PPF, updates s<strong>in</strong>ce <strong>Metop</strong> launch and also<br />
upcom<strong>in</strong>g upgrades, systematic reprocess<strong>in</strong>g will take<br />
place at EUMETSAT to generate c<strong>on</strong>t<strong>in</strong>uous and c<strong>on</strong>sistent<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> L2 products records over the overall operati<strong>on</strong>al period<br />
of <str<strong>on</strong>g>IASI</str<strong>on</strong>g>. They will start with the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> L1C <strong>in</strong> 2013 <strong>after</strong> the<br />
commissi<strong>on</strong><strong>in</strong>g of <strong>Metop</strong>-B, and will be followed by the L2<br />
products.<br />
Significant improvements were obta<strong>in</strong>ed with versi<strong>on</strong> 5:<br />
<strong>in</strong> characterisati<strong>on</strong> of cloud products. The cloud top<br />
pressure is usually retrieved with<strong>in</strong> 50 hPa rmse as<br />
compared to ground-based radars. The correlati<strong>on</strong> is<br />
high (0.9) with the cloud Lidar CALIOP for Polar clouds.<br />
<strong>in</strong> cloud detecti<strong>on</strong>, with a positive impact <strong>on</strong> the SST.<br />
<strong>in</strong> characterisati<strong>on</strong> of the SST. A dem<strong>on</strong>strati<strong>on</strong>al L2P<br />
product has been created and distributed <strong>in</strong> the scope<br />
of the GHRSST. The SST is typically characterised by a<br />
cold bias of about 0.3 K and error standard deviati<strong>on</strong> of<br />
approximately 0.3 K.<br />
<strong>in</strong> LST. The <strong>retrievals</strong> show a correlati<strong>on</strong> of 0.9 with the<br />
SEVIRI LST derived at the LSA-SAF and the errors<br />
estimated are below 2.5 K rmse.<br />
<strong>in</strong> temperature profile <strong>retrievals</strong>, especially <strong>in</strong> the mid and<br />
upper troposphere where rms of departures computed<br />
aga<strong>in</strong>st ECMWF analyses are as low as 0.7 K over oceans;<br />
<strong>in</strong> oz<strong>on</strong>e total column. The operati<strong>on</strong>al O3 TC shows<br />
high correlati<strong>on</strong> with other calibrated satellite products<br />
from GOME-2 and low relative departures rang<strong>in</strong>g<br />
from 3% to 5%.<br />
<strong>in</strong> trace gas retrieval. Comparis<strong>on</strong> of the operati<strong>on</strong>al<br />
CO aga<strong>in</strong>st models and satellite products (MOPITT and<br />
FORLI-CO) c<strong>on</strong>firmed the improvement expected with<br />
synthetic data. The typical errors vary between 8% and<br />
15% (standard deviati<strong>on</strong>) with latitude dependent<br />
smaller biases.<br />
Day-2 developments are <strong>on</strong>go<strong>in</strong>g to further improve<br />
the products, with a particular focus <strong>on</strong> the retrieval of<br />
temperature and humidity <strong>in</strong> the boundary layer, where
the errors are larger. This is addressed with a different<br />
c<strong>on</strong>figurati<strong>on</strong> of the background <strong>in</strong>formati<strong>on</strong> <strong>in</strong> the OEM,<br />
<strong>in</strong> order to take advantage of the natural cross-correlati<strong>on</strong><br />
between temperature and humidity [76,77]. On the measurement<br />
side, recent work from Masiello et al. [78]<br />
<strong>in</strong>dicates that the <strong>in</strong>formati<strong>on</strong> c<strong>on</strong>tent of <str<strong>on</strong>g>IASI</str<strong>on</strong>g> is not<br />
sufficiently well exploited with the current channel sampl<strong>in</strong>g<br />
and that a different channel selecti<strong>on</strong> or the use<br />
of <str<strong>on</strong>g>IASI</str<strong>on</strong>g> pr<strong>in</strong>cipal comp<strong>on</strong>ents is needed to sound the<br />
temperature and especially humidity <strong>in</strong> the boundary<br />
layer [78]. Work has also started to optimise the characterisati<strong>on</strong><br />
of the measurement error described <strong>in</strong><br />
Secti<strong>on</strong> 2.4.2.2, <strong>in</strong> order to give a more adequate weight<br />
to the measurements. Synergistic process<strong>in</strong>g of microwave<br />
measurements (from AMSU and MHS/<strong>Metop</strong>)<br />
together with <str<strong>on</strong>g>IASI</str<strong>on</strong>g> radiances is <strong>in</strong>vestigated as a means<br />
to provide additi<strong>on</strong>al <strong>in</strong>dependent temperature/humidity<br />
<strong>in</strong>formati<strong>on</strong> and to support the explicit exploitati<strong>on</strong> of<br />
cloud c<strong>on</strong>tam<strong>in</strong>ated radiances, which RTTOV-10 now<br />
allows. The absolute validati<strong>on</strong> of water-vapour profiles<br />
still requires dedicated studies to create a more exhaustive<br />
accurate reference data set and to take account of the<br />
spatial and temporal n<strong>on</strong>-co<strong>in</strong>cidences which limit the<br />
validati<strong>on</strong> with <strong>in</strong>-situ radio-s<strong>on</strong>des measurements as<br />
discussed by Pougatchev et al. [57] dur<strong>in</strong>g the validati<strong>on</strong><br />
campaign at L<strong>in</strong>denberg. The absolute accuracy of the<br />
operati<strong>on</strong>al radios<strong>on</strong>de water-vapour measurements was<br />
discussed by Vömel et al. [79] and exhibit significant<br />
biases of the order of 6–8%, depend<strong>in</strong>g <strong>on</strong> the type of<br />
sensor used. More recently, Calbet et al. [80] showed the<br />
importance of us<strong>in</strong>g high-quality synthetic data, computed<br />
from temperature and humidity profiles measured<br />
with cryogenic frost-po<strong>in</strong>t hygrometers flown <strong>on</strong> s<strong>on</strong>des,<br />
for fitt<strong>in</strong>g the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements with simulated spectra.<br />
In the area of atmospheric compositi<strong>on</strong>, the FORLI-CO and<br />
O3 algorithms [25,61], allow<strong>in</strong>g profiles <strong>retrievals</strong> and<br />
error estimates, will be <strong>in</strong>tegrated to the operati<strong>on</strong>al <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
L2 processor <strong>in</strong> the scope of the O3M-SAF CDOP-2 phase.<br />
Acknowledgements<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1369<br />
The authors wish to acknowledge the c<strong>on</strong>tributi<strong>on</strong>s<br />
and the support of a large number of partners and<br />
cooperat<strong>in</strong>g <strong>in</strong>stituti<strong>on</strong>s. Without them this work would<br />
not have been possible. To menti<strong>on</strong> all <strong>in</strong>dividuals would<br />
be bey<strong>on</strong>d the scope of this paper, however we would like<br />
to acknowledge and appreciate the role of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
Sound<strong>in</strong>g Science Work<strong>in</strong>g Group, established jo<strong>in</strong>tly by<br />
CNES and EUMETSAT, which played a central role for this<br />
development. Claude Camy-Peyret is <strong>on</strong>e of its co-chairs<br />
from the very beg<strong>in</strong>n<strong>in</strong>g and his help and guidance is very<br />
much appreciated.<br />
The authors wish to thank CNES for the <strong>in</strong>vestigati<strong>on</strong>s<br />
and the reducti<strong>on</strong> of the <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>in</strong>terpixel differences.<br />
The authors acknowledge the work d<strong>on</strong>e by Maya<br />
George and Cathy Clerbaux (LATMOS) <strong>in</strong> the external<br />
validati<strong>on</strong> study of the CO and O 3 products.<br />
The satellite products from US missi<strong>on</strong>s were downloaded<br />
from the NASA EO portail, former Warehouse<br />
Inventory Search Tool, replaced as of January 2012 by<br />
Reverb/ECHO (http://reverb.echo.nasa.gov). MOPITT CO<br />
and CALIOP cloud products were obta<strong>in</strong>ed from the NASA<br />
Langley Research Center Atmospheric Science Data Center.<br />
The authors acknowledge ESA and IFREMER for the<br />
availability of the AATSR L2P products (ftp://ftp.ifremer.fr/<br />
ifremer/medspirati<strong>on</strong>/data/l2p/ats_nr_2p/esa/).<br />
References<br />
[1] Klaes KD, Cohen M, Buhler Y, Schlüssel P, Munro R, v<strong>on</strong> Engeln A,<br />
et al. An <strong>in</strong>troducti<strong>on</strong> to the EUMETSAT polar system. Bull Am<br />
Meteorol Soc 2007;88:1085–96, doi:10.1175/BAMS-88-7-1085.<br />
[2] Cayla F-R. <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>in</strong>frared <strong>in</strong>terferometer for operati<strong>on</strong>s and research.<br />
In: Ched<strong>in</strong> A, Chah<strong>in</strong>e MT, Scott NA, editors. High spectral resoluti<strong>on</strong><br />
<strong>in</strong>frared remote sens<strong>in</strong>g for Earth’s weather and climate<br />
studies, NATO ASI series, vol. I 9. Berl<strong>in</strong>: Spr<strong>in</strong>ger Verlag; 1993.<br />
[3] Blumste<strong>in</strong> D, Chal<strong>on</strong> G, Carlier T, Buil C, Hebert P, Maciaszek T, et al.<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>in</strong>strument: technical overview and measured performances.<br />
Proc SPIE 2004;5543:196, doi:10.1117/12.560907.<br />
[4] Hilt<strong>on</strong> F, et al. Hyperspectral earth observati<strong>on</strong> from <str<strong>on</strong>g>IASI</str<strong>on</strong>g>: <strong>five</strong><br />
<strong>years</strong> of accomplishments. Bull Am Meteorol Soc 2010;93:<br />
347–70, doi:10.1175/BAMS-D-11-00027.1.<br />
[5] Schlüssel P, Hultberg TH, Phillips PL, August T, Calbet X. The<br />
operati<strong>on</strong>al <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>Level</strong> 2 Processor. Adv Space Res 2005;36:<br />
982–8, doi:10.1016/j.asr.2005.03.008.<br />
[6] Camy-Peyret C, Eyre J. <str<strong>on</strong>g>IASI</str<strong>on</strong>g> science plan. ISSWG technical document;<br />
1998.<br />
[7] Chah<strong>in</strong>e M, et al. AIRS: improv<strong>in</strong>g weather forecast<strong>in</strong>g and<br />
provid<strong>in</strong>g new data <strong>on</strong> greenhouse gases. Bull Am Meteorol Soc<br />
2006;87:911–26, doi:10.1175/BAMS-87-7-911.<br />
[8] Loveland TR, et al. Development of a global land cover characteristics<br />
database and IGBP DISCover from 1 km AVHRR data. Int<br />
J Remote Sens<strong>in</strong>g 2000;21(6–7):1303–30.<br />
[9] Zhou DK, Larar AM, Liu X, Smith WL, Strow LL, Yang P, et al. Global<br />
land surface emissivity retrieval from satellite ultraspectral IR<br />
measurements. IEEE Trans Geosci Remote Sens<strong>in</strong>g 2011;49:<br />
1277–90, doi:10.1109/TGRS.20102051036.<br />
[10] U.S. Geological Survey’s EROS Data Center. Sioux Falls, South<br />
Dakota, /http://eros.usgs.gov/#/F<strong>in</strong>d_Data/Products_and_Data_A<br />
vailable/GTOPO30S.<br />
[11] Atk<strong>in</strong>s<strong>on</strong> NN, Hilt<strong>on</strong> FI, Ill<strong>in</strong>gworth SM, Eyre JR, Hultberg T.<br />
Potential for the use of rec<strong>on</strong>structed <str<strong>on</strong>g>IASI</str<strong>on</strong>g> radiances <strong>in</strong> the detecti<strong>on</strong><br />
of atmospheric trace gases. Atmos Meas Tech 2010;3:<br />
991–1003, doi:10.5194/amt-3-991-2010.<br />
[12] Karagulian F, Clarisse L, Clerbaux C, Prata AJ, Hurtmans D,<br />
Coheur PF. Detecti<strong>on</strong> of volcanic SO 2, ash, and H 2SO 4 us<strong>in</strong>g the<br />
Infrared Atmospheric Sound<strong>in</strong>g Interferometer (<str<strong>on</strong>g>IASI</str<strong>on</strong>g>). J Geophys<br />
Res 2010;115:d00l02, doi:10.1029/2009JD012786.<br />
[13] Clarisse L, Coheur PF, Chefdeville S, Lacour JL, Hurtmans D,<br />
Clerbaux C. Infrared satellite observati<strong>on</strong>s of hydrogen sulfide <strong>in</strong><br />
the volcanic plume of the August 2008 Kasatochi erupti<strong>on</strong>. Geophys<br />
Res Lett 2011;38:L10804, doi:10.1029/2011GL047402.<br />
[14] K<strong>on</strong>ovalov IB, Beekmann M, Kuznetsova IN, Yurova A, Zvyag<strong>in</strong>tsev AM.<br />
Atmospheric impacts of the 2010 Russian wildfires: <strong>in</strong>tegrat<strong>in</strong>g<br />
modell<strong>in</strong>g and measurements of an extreme air polluti<strong>on</strong> episode<br />
<strong>in</strong> the Moscow regi<strong>on</strong>. Atmos Chem Phys 2011;11:10031–56, doi:<br />
10.5194/acp-11-10031-2011.<br />
[15] Masuda K, Takashima T, Takayama Y. Emissivity of pure water and<br />
sea waters for the sea surface <strong>in</strong> the <strong>in</strong>frared w<strong>in</strong>dow regi<strong>on</strong>s.<br />
Remote Sens<strong>in</strong>g Envir<strong>on</strong> 1988;24:313–29.<br />
[16] Watts PD, Allen MR, Night<strong>in</strong>gale TJ. W<strong>in</strong>d speed effects <strong>on</strong> sea<br />
surface emissi<strong>on</strong> and reflecti<strong>on</strong> for the Al<strong>on</strong>g Track Scann<strong>in</strong>g<br />
Radiometer. J Atmos Oceanic Technol 1996;13:126–41.<br />
[17] Saunders RW, Matricardi M, Brunel P. An improved fast radiative<br />
transfer model for assimilati<strong>on</strong> of satellite radiance observati<strong>on</strong>s. Q<br />
J R Meteorol Soc 1999;125:1407–25.<br />
[18] Matricardi M, Saunders R. Fast radiative transfer model for simulati<strong>on</strong><br />
of <strong>in</strong>frared atmospheric sound<strong>in</strong>g <strong>in</strong>terferometer radiances.<br />
Appl Opt 1999;38:5679–91.<br />
[19] Zhou D, et al. Thermodynamic and cloud parameter retrieval us<strong>in</strong>g<br />
<strong>in</strong>frared spectral data. Geophys Res Lett 2005;32:L15805, doi:<br />
10.1029/2005GL023211.<br />
[20] Ackerman SA. Global satellite observati<strong>on</strong>s of negative brightness<br />
temperature difference between 11 and 6.7 mm. J Atmos Sci 1996;<br />
53:2803–12.
1370<br />
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371<br />
[21] Ackerman SA. Remote sens<strong>in</strong>g aerosols from satellite <strong>in</strong>frared<br />
observati<strong>on</strong>s. J Geophys Res 1997;102:17069–79.<br />
[22] Stowe LL, Davis PA, McCla<strong>in</strong> EP. Scientific basis and <strong>in</strong>itial evaluati<strong>on</strong><br />
of the CLAVR-1 global clear/cloud classificati<strong>on</strong> algorithm for<br />
the Advanced Very High Resoluti<strong>on</strong> Radiometer. J Atmos Oceanic<br />
Technol 1999;16:656–77.<br />
[23] Le Cun Y, Bottou L, Orr GB, Müller K-R. Efficient backprop. Neural<br />
networks: tricks of the trade, lecture notes <strong>in</strong> computer science,<br />
vol. 1524. Spr<strong>in</strong>ger; 1998 [p. 546. DOI:10.1007/3-540-49430-8_2].<br />
[24] Wan Z, Ng D, Dozier J. Spectral emissivity measurements of landsurface<br />
materials and related radiative transfer simulati<strong>on</strong>s. Adv<br />
Space Res 1994;14(3):91–4.<br />
[25] Clerbaux C, Boynard A, Clarisse L, George M, Hadji-Lazaro J, Herb<strong>in</strong> H,<br />
et al. M<strong>on</strong>itor<strong>in</strong>g of atmospheric compositi<strong>on</strong> us<strong>in</strong>g the thermal<br />
<strong>in</strong>frared <str<strong>on</strong>g>IASI</str<strong>on</strong>g>/MetOp sounder. Atmos Chem Phys 2009;9:6041–54, doi:<br />
10.5194/acp-9-6041-2009.<br />
[26] Corlett GK, Bart<strong>on</strong> IJ, D<strong>on</strong>l<strong>on</strong> CJ, Edwards MC, Good SA, Horrocks LA,<br />
et al. The accuracy of SST <strong>retrievals</strong> from AATSR: an <strong>in</strong>itial<br />
assessment through geophysical validati<strong>on</strong> aga<strong>in</strong>st <strong>in</strong> situ radiometers,<br />
buoys and other SST data sets. Adv Space Res 2006;37(4):<br />
764–9.<br />
[27] Ichoku C, Kaufman YJ, Remer LA, Levy R. Global aerosol remote<br />
sens<strong>in</strong>g from MODIS. Adv Space Res 2004;34:820–7.<br />
[28] Strabala KI, Ackerman SA, Menzel WP. Cloud properties <strong>in</strong>ferred<br />
from 8–12 mm data. J Appl Meteorol 1994;33:212–29.<br />
[29] Wei H, Yang P, Li J, Baum BA, Huang H-L, Platnick S, et al. Retrieval<br />
of semitransparent ice cloud optical thickness from atmospheric<br />
<strong>in</strong>frared sounder (AIRS) measurements. IEEE Trans Geosci Remote<br />
Sens<strong>in</strong>g 2004;42:2254–67.<br />
[30] Menzel WP, Smith WL, Stewart TR. Improved cloud moti<strong>on</strong> w<strong>in</strong>d<br />
vector and altitude assignment us<strong>in</strong>g VAS. J Clim Appl Meteorol<br />
1983;22:377–84.<br />
[31] Smith WL, Frey R. On cloud altitude determ<strong>in</strong>ati<strong>on</strong> from high<br />
resoluti<strong>on</strong> <strong>in</strong>terferometer sounder (HIS) observati<strong>on</strong>s. J Appl<br />
Meteorol 1990;29:658–62.<br />
[32] Kreuter A, Zangerl M, Schwarzmann M, Blumthaler M. All-sky<br />
imag<strong>in</strong>g: a simple, versatile system for atmospheric research. Appl<br />
Opt 2009;48(6):1091–7, doi:10.1364/AO.48.001091.<br />
[33] Feister U, Shields J. Cloud and radiance measurements with the VIS/<br />
NIR Daylight Whole Sky Imager at L<strong>in</strong>denberg (Germany). Meteorol<br />
Z 2005;14(5):627–39, doi:10.1127/0941-2948/2005/0066.<br />
[34] Hennemuth B, Weiss A, Bösenberg J, Jacob D, L<strong>in</strong>né H, Peters G,<br />
et al. Quality assessment of water cycle parameters <strong>in</strong> REMO by<br />
radar–lidar synergy. Atmos Chem Phys 2008;8:287–308, doi:<br />
10.5194/acp-8-287-2008.<br />
[35] W<strong>in</strong>ker DM, et al. The CALIPSO Missi<strong>on</strong>: a global 3D view of<br />
aerosols and clouds. Bull Am Meteorol Soc 2010;91:1211–29, doi:<br />
10.1175/2010BAMS3009.1.<br />
[36] W<strong>in</strong>ker DM, Hunt WH, McGill MJ. Initial performance assessment<br />
of CALIOP. Geophys Res Lett 2007;34:L19803, doi:10.1029/<br />
2007GL030135.<br />
[37] W<strong>in</strong>ker D, et al. Overview of the CALIPSO Missi<strong>on</strong> and CALIOP Data<br />
Process<strong>in</strong>g Algorithms. J Atmos Oceanic Technol 2009;26:<br />
2310–23, doi:10.1175/2009JTECHA1281.1.<br />
[38] Kim S-W, Chung E-S, Yo<strong>on</strong> S-C, Sohn B-J, Nobuo S. Intercomparis<strong>on</strong>s<br />
of cloud-top and cloud-base heights from ground-based Lidar,<br />
CloudSat and CALIPSO measurements. Int J Remote Sens<strong>in</strong>g 2011;<br />
32(4):1179–97, doi:10.1080/01431160903527439.<br />
[39] Menzel P, et al. MODIS global cloud-top pressure and amount<br />
estimati<strong>on</strong>: algorithm descripti<strong>on</strong> and results. J Appl Meteorol<br />
Climatol 2008;47:1175–98, doi:10.1175/2007JAMC1705.1.<br />
[40] Stubenrauch CJ, Chéd<strong>in</strong> A, Armante R, Scott A. Clouds as seen by<br />
satellite Sounders (3I) and Imagers (UISCCP). Part II: a new<br />
approach for cloud parameter determ<strong>in</strong>ati<strong>on</strong> <strong>in</strong> the 3I algorithms.<br />
J Clim 1999;12:2214–23.<br />
[41] Eyre J, Menzel WP. Retrieval of cloud parameters from satellite<br />
sounder data: a simulati<strong>on</strong> study. J Appl Meteorol 1989;28:267–75.<br />
[42] Schlüssel P, Goldberg M. Retrieval of atmospheric temperature and<br />
water vapour from <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements <strong>in</strong> partly cloudy situati<strong>on</strong>s.<br />
Adv Space Res 2002;29(11):1703–6.<br />
[43] Hadji-Lazaro J, Clerbaux C, Thiria S. An <strong>in</strong>versi<strong>on</strong> algorithm us<strong>in</strong>g<br />
neural networks to retrieve atmospheric CO total columns from<br />
high resoluti<strong>on</strong> Nadir radiances. J Geophys Res 1999;104:<br />
23841–54.<br />
[44] Turquety S, Hadji-Lazaro J, Clerbaux C, Hauglusta<strong>in</strong>e DA, Clough SA,<br />
Cassé V, et al. Operati<strong>on</strong>al trace gas retrieval algorithm for the<br />
Infrared Atmospheric Sound<strong>in</strong>g Interferometer. J Geophys Res<br />
2004;109:D21301, doi:10.1029/2004JD004821.<br />
[45] Rodgers C. Inverse methods for atmospheric sound<strong>in</strong>g. World<br />
Scientific; 981-02-2740-X [p. 238].<br />
[46] Collard AD. Selecti<strong>on</strong> of <str<strong>on</strong>g>IASI</str<strong>on</strong>g> channels for use <strong>in</strong> numerical weather<br />
predicti<strong>on</strong>. Q J R Meteorol Soc 2007;133:19771991b.<br />
[47] Aires F, Prigent C, Rossow WB. Temporal <strong>in</strong>terpolati<strong>on</strong> of global<br />
surface sk<strong>in</strong> temperature diurnal cycle over land under clear and<br />
cloudy c<strong>on</strong>diti<strong>on</strong>s. J Geophys Res 2004;109:D04313, doi:10.1029/<br />
2003JD003527.<br />
[48] Trigo IF, M<strong>on</strong>teiro IT, Olesen F, Kabsch E. An assessment of remotely<br />
sensed land surface temperature. J Geophys Res 2008;113:<br />
D17108, doi:10.1029/2008JD010035.<br />
[49] Seemann SW, Borbas EE, Knutes<strong>on</strong> RO, Stephens<strong>on</strong> GR, Huang HL.<br />
Development of a global <strong>in</strong>frared land surface emissivity database<br />
for applicati<strong>on</strong> to clear sky sound<strong>in</strong>g <strong>retrievals</strong> from multi-spectral<br />
satellite radiance measurements. J Appl Meteorol Climatol 2007;<br />
47:108–23.<br />
[50] D<strong>on</strong>l<strong>on</strong> C, et al. The global ocean data assimilati<strong>on</strong> experiment<br />
high-resoluti<strong>on</strong> sea surface temperature pilot project. Bull Am<br />
Meteorol Soc 2007;88:1197–213.<br />
[51] Smith DL, Delderfield J, Drumm<strong>on</strong>d D, Edwards T, Mutlow CT, Read<br />
PD, et al. Calibrati<strong>on</strong> of the AATSR <strong>in</strong>strument. Adv Space Res<br />
2001;28:31–9 [8105].<br />
[52] Ill<strong>in</strong>gworth SM, Remedios JJ, Parker RJ. Intercomparis<strong>on</strong> of <strong>in</strong>tegrated<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> and AATSR calibrated radiances at 11 and 12 mm. Atmos<br />
Chem Phys 2009;9:6677–83, doi:10.5194/acp-9-6677-2009.<br />
[53] Noyes EJ, M<strong>in</strong>nett PJ, Remedios JJ, Corlett GK, Good SA, Llewellyn-<br />
J<strong>on</strong>es DT. The accuracy of the AATSR sea surface temperatures <strong>in</strong><br />
the Caribbean. Remote Sens<strong>in</strong>g Envir<strong>on</strong> 2006;101(1):38–51.<br />
[54] D<strong>on</strong>l<strong>on</strong> CJ, M<strong>in</strong>nett PJ, Gentemann C. Toward improved validati<strong>on</strong><br />
of satellite sea surface sk<strong>in</strong> temperature measurements for climate<br />
research. J Clim 2002;15(4):353–69.<br />
[55] M<strong>in</strong>nett P. A numerical study of the effects of anomalous North<br />
Atlantic atmospheric c<strong>on</strong>diti<strong>on</strong>s <strong>on</strong> the <strong>in</strong>frared measurement of<br />
sea surface temperature from space. J Geophys Res 1986;91:C7.<br />
[56] O’carroll AG, Eyre JR, Saunders RW. Three-way error analysis<br />
between AATSR, AMSR-E, and <strong>in</strong> situ sea surface temperature<br />
observati<strong>on</strong>s. J Atmos Oceanic Technol 2008;25:1197–207.<br />
[57] Pougatchev N, August T, Calbet X, Hultberg T, Oduleye O, Schlüssel<br />
P, et al. <str<strong>on</strong>g>IASI</str<strong>on</strong>g> temperature and water vapor <strong>retrievals</strong> —error<br />
assessment and validati<strong>on</strong>. Atmos Chem Phys 2009;9:6453–8.<br />
[58] Reale T. ATOVS derived sound<strong>in</strong>gs us<strong>in</strong>g Noaa PROduct Validati<strong>on</strong><br />
System (NPROVS) datasets for comput<strong>in</strong>g first guess and sensor<br />
bias adjustments <strong>in</strong>dependent of NWP. In: Proceed<strong>in</strong>gs of the 16th<br />
c<strong>on</strong>ference <strong>on</strong> satellite meteorology and oceanography. AMS; 2009.<br />
[59] Loyola DG, Koukouli ME, Valks P, Balis DS, Hao N, Van Roozendael<br />
M, et al. The GOME-2 total column oz<strong>on</strong>e product: retrieval<br />
algorithm and ground-based validati<strong>on</strong>. J Geophys Res 2011;116:<br />
D07302, doi:10.1029/2010JD014675 [11 p.].<br />
[60] Boynard A, Clerbaux C, Coheur P-F, Hurtmans D, Turquety S, George<br />
M, et al. Measurements of total and tropospheric oz<strong>on</strong>e from <str<strong>on</strong>g>IASI</str<strong>on</strong>g>:<br />
comparis<strong>on</strong> with correlative satellite, ground-based and oz<strong>on</strong>e s<strong>on</strong>de<br />
observati<strong>on</strong>s. Atmos Chem Phys 2009;9:6255–71, doi:10.5194/<br />
acp-9-6255-2009.<br />
[61] Hurtmans D, et al. FORLI radiative transfer and retrieval codes<br />
for <str<strong>on</strong>g>IASI</str<strong>on</strong>g>. J Quant Spectrosc Radiat Transfer, doi:10.1016/j.jqsrt.2012.<br />
02.036, this issue.<br />
[62] Ant<strong>on</strong> M, Loyola D, Clerbaux C, Lopez M, Vilaplana JM, Ban<strong>on</strong> M,<br />
et al. Validati<strong>on</strong> of the <strong>Metop</strong>-A total oz<strong>on</strong>e data from GOME-2 and<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> us<strong>in</strong>g reference ground-based measurements at the Iberian<br />
Pen<strong>in</strong>sula. Remote Sens<strong>in</strong>g Envir<strong>on</strong> 2011;115:1380–6.<br />
[63] Brasseur GP, Hauglusta<strong>in</strong>e DA, Walters S, Rasch PJ, Müller J-F,<br />
Granier C, Tie XX. MOZART: A global chemical transport model for<br />
oz<strong>on</strong>e and related chemical tracers, part 1. Model descripti<strong>on</strong>. J<br />
Geophys Res 1998;103:28265–89.<br />
[64] Bishop CM. Tra<strong>in</strong><strong>in</strong>g with noise is equivalent to tikh<strong>on</strong>ov regularizati<strong>on</strong>.<br />
Neural Comput 1994;7(1):108–16, doi:10.1162/neco.1995.7.1.108.<br />
[65] Bousserez N, Attié JL, Peuch VH, et al. Evaluati<strong>on</strong> of the MOCAGE<br />
chemistry transport model dur<strong>in</strong>g the ICARTT/ITOP experiment. J<br />
Geophys Res 2007;112:D10S42, doi:10.1029/2006JD007595.<br />
[66] George M, Clerbaux C, Hurtmans D, Turquety S, Coheur P-F, Pommier<br />
M, et al. Carb<strong>on</strong> m<strong>on</strong>oxide distributi<strong>on</strong>s from the <str<strong>on</strong>g>IASI</str<strong>on</strong>g>/METOP<br />
missi<strong>on</strong>: evaluati<strong>on</strong> with other space-borne remote sensors. Atmos<br />
Chem Phys 2009;9:8317–30, doi:10.5194/acp-9-8317-2009.<br />
[67] Deeter MN, Emm<strong>on</strong>s LK, Francis GL, Edwards DP, Gille JC, Warner<br />
JX, et al. Operati<strong>on</strong>al carb<strong>on</strong> m<strong>on</strong>oxide retrieval algorithm and<br />
selected results for the MOPITT <strong>in</strong>strument. J Geophys Res<br />
2003;108(D14):4399, doi:10.1029/2002JD003186.
T. August et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 113 (2012) 1340–1371 1371<br />
[68] Deeter MN, Emm<strong>on</strong>s LK, Edwards DP, Gille JC, Drumm<strong>on</strong>d JR. Vertical<br />
resoluti<strong>on</strong> and <strong>in</strong>formati<strong>on</strong> c<strong>on</strong>tent of CO profiles retrieved by MOPITT.<br />
Geophys Res Lett 2004;31:L15112, doi:10.1029/2004GL020235.<br />
[69] Turquety S, Clerbaux C, Law K, Coheur P-F, Cozic A, Szopa S, et al.<br />
CO emissi<strong>on</strong> and export from Asia: an analysis comb<strong>in</strong><strong>in</strong>g complementary<br />
satellite measurements (MOPITT, SCIAMACHY and<br />
ACE-FTS) with global modell<strong>in</strong>g. Atmos Chem Phys 2008;8:<br />
5187–204, doi:10.5194/acp-8-5187-2008.<br />
[70] Pommier M, Law KS, Clerbaux C, Turquety S, Hurtmans D, Hadji-<br />
Lazaro J, et al. <str<strong>on</strong>g>IASI</str<strong>on</strong>g> carb<strong>on</strong> m<strong>on</strong>oxide validati<strong>on</strong> over the Arctic<br />
dur<strong>in</strong>g POLARCAT spr<strong>in</strong>g and summer campaigns. Atmos Chem<br />
Phys 2010;10:10655–78, doi:10.5194/acp-10-10655-2010.<br />
[71] Yurganov LN, Rakit<strong>in</strong> V, Dzhola A, August T, Fokeeva E, George M,<br />
et al. Satellite- and ground-based CO total column observati<strong>on</strong>s<br />
over 2010 Russian fires: accuracy of top–down estimates based <strong>on</strong><br />
thermal IR satellite data. Atmos Chem Phys 2011;11:7925–42, doi:<br />
10.5194/acp-11-7925-2011.<br />
[72] Elgu<strong>in</strong>di N, Clark H, Ordóñez C, Thouret V, Flemm<strong>in</strong>g J, Ste<strong>in</strong> O, et al.<br />
Current status of the ability of the GEMS/MACC models to reproduce<br />
the tropospheric CO vertical distributi<strong>on</strong> as measured by MOZAIC.<br />
Geosci Model Dev 2010;3:501–18, doi:10.5194/gmd-3-501-2010.<br />
[73] Ricaud P, Attié J-L, Teyssedre H, El Amraoui L, Peuch V-H, Matricardi<br />
M, et al. Equatorial total column of nitrous oxide as measured<br />
by <str<strong>on</strong>g>IASI</str<strong>on</strong>g> <strong>on</strong> MetOp-A: implicati<strong>on</strong>s for transport processes. Atmos<br />
Chem Phys 2009;9:3947–56, doi:10.5194/acp-9-3947-2009.<br />
[74] Crevoisier C, Nobileau D, Fiore AM, Armante R, Chéd<strong>in</strong> A, Scott NA.<br />
Tropospheric methane <strong>in</strong> the tropics—first year from <str<strong>on</strong>g>IASI</str<strong>on</strong>g><br />
hyperspectral <strong>in</strong>frared observati<strong>on</strong>s. Atmos Chem Phys 2009;9:<br />
6337–50, doi:10.5194/acp-9-6337-2009.<br />
[75] Crevoisier C, Chéd<strong>in</strong> A, Matsueda H, Machida T, Armante R, Scott<br />
NA. First year of upper tropospheric <strong>in</strong>tegrated c<strong>on</strong>tent of CO 2 from<br />
<str<strong>on</strong>g>IASI</str<strong>on</strong>g> hyperspectral <strong>in</strong>frared observati<strong>on</strong>s. Atmos Chem Phys<br />
2009;9:4797–810, doi:10.5194/acp-9-4797-2009.<br />
[76] Gambacorta A. An assessment of the tropical humidity-temperature<br />
covariance us<strong>in</strong>g AIRS. Geophys Res Lett 2008;35:<br />
L10814, doi:10.1029/2008GL033805 [5 p.].<br />
[77] Ross RJ, et al. Lower-tropospheric humidity–temperature relati<strong>on</strong>ships<br />
<strong>in</strong> radios<strong>on</strong>de observati<strong>on</strong>s and atmospheric general circulati<strong>on</strong><br />
models. J Hydrometeorol 2001;3:26–38.<br />
[78] Masiello G, Serio C, Ant<strong>on</strong>elli P. Inversi<strong>on</strong> for atmospheric thermodynamical<br />
parameters of <str<strong>on</strong>g>IASI</str<strong>on</strong>g> data <strong>in</strong> the pr<strong>in</strong>cipal comp<strong>on</strong>ents<br />
space. Q J R Meteorol Soc 2011. doi:10.1002/qj.909.<br />
[79] Vömel H, Whiteman DN, Lesht BM, Schmidl<strong>in</strong> FJ, Russo F. Absolute<br />
accuracy of water vapor measurements from six operati<strong>on</strong>al<br />
radios<strong>on</strong>de types launched dur<strong>in</strong>g AWEX-G and implicati<strong>on</strong>s for<br />
AIRS validati<strong>on</strong>. J Geophys Res 2006;111:D09S10, doi:10.1029/<br />
2005JD006083.<br />
[80] Calbet X, Kivi R, Tjemkes S, M<strong>on</strong>tagner F, Stuhlmann R. Match<strong>in</strong>g<br />
radiative transfer models and radios<strong>on</strong>de data from the EPS/<strong>Metop</strong><br />
Sodankylä campaign to <str<strong>on</strong>g>IASI</str<strong>on</strong>g> measurements. Atmos Meas Tech<br />
2011;4:1177–89, doi:10.5194/amt-4-1177-2011.<br />
[81] Menzel P, Strabala K. Cloud top troperties and cloud phase:<br />
algorithm theoretical basis document. ATBD-MOD-04, NASA Goddard<br />
Space Flight Center; 1997. 55 pp.