2
L. Chaumat, O. Lezeaux, P. Prunet, B. Tournier
F.-R. Cayla (SISCLE), C. Camy-Peyret (LPMAA) and T. Phulpin (CNES)
Study supported by CNES
ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008

Outline
2
9 Efficient extraction of the CO
2 signal from the IASI spectrum: specific data processing algorithm
9 Quantitative analysis of the information content for the definition and specification of an optimal product
9 First validation of the retrieval on IASI spectra
NOV-3678-SL-5797 ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 2

Methodology - Signal processing: data filtering
The atmospheric spectrum measured by IASI exhibits a quasi-periodic line structure
NOV-3678-SL-5797 ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 3

Methodology - Signal processing: data filtering
Atmospheric spectrum re-sampling on a periodic base built from the spectral transitions of the CO
2 lines
Application of a Discrete Fourier
Transform (DFT) to 16 specific spectral windows
Mean signal
NOV-3678-SL-5797
Fundamental
First Harmonic mean signal, fundamental and harmonics: pseudo-data
ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 4

Methodology - Signal processing: data filtering
Characteristics of the selected spectral windows
Window designation
Representative wave number (cm
-1
)
Limits (cm
-1
)
Number of spectroscopic lines
Number of corresponding
IASI samples (N)
9 51
W6 804.7
812.48413
W7 946.1
956.1851
15 91
13 78
15 88
9 50
11 64
13 87
13 65
15 103
17 85
15 88
9 52
17 117
15 79
17 118
13 66
NOV-3678-SL-5797 ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 5

Methodology - Signal processing: data filtering
Signal to noise ratio (SNR) for the mean signal of DFT compared to the SNR in the spectrum space
Windows DFT SNR DFT SNR / √ N Spectrum SNR Windows DFT SNR DFT SNR / √ N Spectrum SNR
W1
W2
W3
W4
W5
W6
W7
W8
1074.8
150.5
138.0
1357.7
142.3
165.2
1825.5
206.7
150.5
142.3
206.7
282.7
358.9
323.4
312.3
300.8
138.0
165.2
206.0
216.9
296.0
333.0
304.6
296.3
W9
W10
W11
W12
W13
W14
W15
W16
1815.9
2559.6
831.3
647.6
57.8
64.8
1076.4
689.9
178.9
277.6
88.6
89.8
5.3
7.3
99.1
84.9
173.7
283.1
71.2
94.4
3.3
9.8
103.1
89.6
¾ The Fourier space analysis method allows information merging by increasing the SNR with a factor √ N (number of window samples)
¾ The pseudo-data equivalent to super-channels: data compression
NOV-3678-SL-5797 ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 6

Analysis from simulated and real IASI data
1-DVAR information content analysis
Studies from
9 Synthetic spectra and Jacobians
(4A/OP, poster A11)
9 Specific set of real IASI spectra y Ξ IASI spectrum or pseudo-data
9 IASI data CO
2 channel selection:
1282 spectral samples (SEL)
9 IASI pseudo-data: 48 elements
(3 harmonics x 16 windows) (DFT)
Extended retrieval studies for
9 Specification of the retrieval strategy
9 Definition of the optimal product
9 Optimisation of the data processing
NOV-3678-SL-5797 ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 7

Analysis from simulated and real IASI data
Specific IASI data set : external calibration mode
¾ Redundancy of IASI pixels acquisition at a given location
NOV-3678-SL-5797
9 Processing of 25 spectra measuring the same atmospheric state (1.4 km between 2 consecutive spectra over clear-sky sea)
9 Use of a low-noise mean spectrum for retrieval optimisation tests
ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 8

Retrieval strategy
Analysis from simulated and real IASI data
Different parametrisation tests
9 CO
2 plus surface temperature (T
S
)
9 CO
2
, T
S plus column of contaminating species
9 CO
2
, T
S plus profiles of contaminating species
9 CO
2
, T
S
, plus proper transport of species profile errors
NOV-3678-SL-5797 ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 9

Information content analysis
SEL DFT
Analysis from simulated and real IASI data
3-levels profile retrieval errors 2-levels profile retrieval errors
Definition of the optimal product : retrieval error
9 CO
2 mean VMR : 2.3 ppmv, driven by measurement error
9 2 pieces of information profile : tropo: 5 ppmv, strato: 10 ppmv
9 3 pieces of information profile : low tropo: 35 ppmv, high tropo: 10 ppmv, strato: 15 ppmv
NOV-3678-SL-5797 ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 10

Analysis from simulated and real IASI data
Definition of the optimal product : weighting function
9 Mean VMR, SEL
9 Mean VMR, DFT
9 3 pieces of information profile
2 pieces of information profile should be optimal
DFT slightly degrades the information content
NOV-3678-SL-5797 ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 11

Analysis from simulated and real IASI data
Time series of CO
2 retrieval from IASI data set: optimisation of processing
Data processing
9 Classical channel selection (SEL)
9 DFT processing
Temporal and vertical stability of the DFT processing
Filtering of elements which pollutes the
CO
2 signal
NOV-3678-SL-5797 ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 12

Analysis from simulated and real IASI data
Summary (Analysis from simulated and 25 real IASI data)
9 CO
2 mean VMR, as well as CO reasonable accuracy
2 stratospheric and tropospheric partial VMR, can be retrieved from IASI spectrum with a
9 An optimised processing for the CO information compression and filtering
2 retrieval is defined :
¾ DFT data processing ensures efficient and stable retrieval via
¾ CO
2
¾ T/H
2 is retrieved simultaneously with surface information
O auxiliary information from IASI retrieval should be used (no optimal in our experiments)
¾ Trace gases error is properly considered
9 A full product characterisation is provided by the retrieval algorithm (error estimate, weighting function)
NOV-3678-SL-5797 ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 13

Study in progress
Work in progress
9 Extended validation on IASI data
¾ Global statistics obtained from about 800 inversions on homogeneous scenes classified as “clear over sea” at different locations compared to the theoretical analysis
• Noise retrieval (precision: relative standard deviation)
Method
CO
2 product
VMR
2 layers (TR-ST)
DFT
Theory
1.14 %
2 %
Measures
5.5 % 2.8 %
0.89 %
5.8 %
Theory
0.72 %
1.5 % 3.5 %
SEL
Measures
0.50 %
2.5 % 9.7 %
• These preliminary results confirm previous results:
– The DFT method needs to be optimised for the mean VMR
– The DFT method is more robust and stable to retrieve 2 pieces of information profile
NOV-3678-SL-5797 ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 14

Outlook
9 Extended validation on a representative data set
9 Impact studies of IASI CO inverse modelling
2 product through
9 Possibility to improve CO
2 product through spatial and temporal averaging or specific processing
9 Coupling with efficient cloud processing to increase the availability of IASI accurate information
¾ Poster B04: “Processing of IASI heterogeneous scenes”
NOV-3678-SL-5797 ITSC-XVI: Angra dos Reis, Brazil, 7-13 May 2008 15