Preliminary studies towards the use of IASI PC products in the Météo

Document NWPSAF_MF_VS_005
Version 1.0
20/2/15
Preliminary studies towards the use of IASI
PC products in the Météo-France global data
assimilation system
Stephanie Guedj, Vincent Guidard, and Jean-Francois Mahfouf
Doc ID
Preliminary studies towards the Version
use of IASI PC products in the Date
Météo-France global data
assimilation system
: NWPSAF-MF-VS-005
: 1.0
: 20/2/15
Preliminary studies towards the use of IASI PC products in the
Météo-France global data assimilation system
Stephanie Guedj, Vincent Guidard, and Jean-Francois Mahfouf
CNRM-GAME/METEO-FRANCE and CNRS
42 av. Gaspard Coriolis, 31057 Toulouse, FRANCE
This documentation was developed within the context of the EUMETSAT Satellite
Application Facility on Numerical Weather Prediction (NWP SAF), under the Cooperation
Agreement dated 29 June 2011, between EUMETSAT and the Met Office, UK, by one or
more partners within the NWP SAF. The partners in the NWP SAF are the Met Office,
ECMWF, KNMI and Météo France.
Copyright 2015, EUMETSAT, All Rights Reserved.
Change record
Version
Date
Author / changed by
1.0
20/2/15
Stephanie Guedj,
Vincent Guidard, and
Jean-Francois Mahfouf
Remarks
Preliminary studies towards the use of
IASI PC products in the Météo-France
global data assimilation system
Report : NWP-AS14-P01
Stephanie Guedj, Vincent Guidard*,
and Jean-Francois Mahfouf
CNRM-GAME/METEO-FRANCE and CNRS
42 av. Gaspard Coriolis, 31057 Toulouse, FRANCE
*correspondig author: [email protected]
February 20, 2015
1
Abstract
A pre-processing chain has been set up in the Météo-France research environment to get Eumetsat BUFR of IASI PC scores and reconstruct radiance spectra
from these PC scores. Reconstructed data have been compared with original data
and globally showed a good agreement. Nevertheless, some discrepancies occuring
either over cloudy areas or over land indicating that some numerical errors may
occur either in the coding of the BUFR files or in the reconstruction of the spectra
in Météo-France pre-processing.
When ingested in the assimilation process of the global model ARPEGE, reconstructed BTs have the same impact on analyses and forecasts as the original
BTs. No impact is seen on the bias correction, which means that assimilating
reconstruted BTs introduces no bias in the system. The standard deviations
of observations minus simulations from model fields are smaller for reconstructed
BTs than for the original BTs, which is consistent with the a posteriori Desroziers
diagnostic. Especially in band 1, reconstructed BTs are less ”noisy” and could be
assimilated with smaller σo (which means with a larger weight). But it appears
that the PCA technique introduces inter-channel correlation in the observation
errors.
Results are consistent with those obtained at the Met-Office.
2
Contents
1 Work plan
5
2 Introduction
6
3 Monitoring of the original and reconstructed set of IASI radiances
3.1 Pre-processing of IASI Radiances . . . . . . . . . . . . . . . . . . . . .
3.2 Objective comparison between original and reconstructed radiances . .
3.3 Radiance simulation skills . . . . . . . . . . . . . . . . . . . . . . . . .
3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4 Assimilation experiments of reconstructed radiances
4.1 Impact on the innovations . . . . . . . . . . . . . . . .
4.2 Effects on cloud detection . . . . . . . . . . . . . . . .
4.3 Analysis and forecast skills . . . . . . . . . . . . . . . .
4.4 Potential changes in observation error correlations . . .
5 Conclusion and future plans
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7
7
7
10
12
.
.
.
.
14
14
15
15
16
19
3
Acknowledgements
This research is funded by the NWP-SAF of Eumetsat. The authors would like to thank
Pascal Brunel from Météo-France for his help with RTTOV. The authors warmly thank
Eumetsat teams, in particular Tim Hultberg for his help on PC scores. Many thanks
to Jean Maziejewski for his help to revise the manuscript.
Acronyms
AROME
ARPEGE
BT
CrIS
ECMWF
GNSS
IASI
IRS
ITCZ
MTG
NWP
PC
PCA
RTTOV
SAF
STD
TIROS
TOVS
Applications of Research to Operations at MEsoscale
Action de Recherche Petite Echelle Grande Echelle
Brightness Temperature
Cross-track Infrared Sounder
European Centre for Medium Range Forecasts
Global Navigation Satellite System
Infrared Atmosperic Sounding Interferometer
InfraRed Sounder
InterTropical Convergence Zone
Meteosat Third Generation
Numerical Weather Prediction
Principal Component
Principal Componet Analysis
Radiative Transfer for TOVS
Satellite Application Facility
Standard Deviation
Television InfraRed Observation Satellite
TIROS Operational Vertical Sounder
4
1
Work plan
The activity will be to examine the use of the real-time Eumetsat IASI PC scores
in order to reconstruct IASI radiances and to assimilate them in the Météo-France
ARPEGE global model using a 4D-Var data assimilation system. Data monitoring of
reconstructed radiances will be implemented and assimilation experiments performed
over a two month period. These two aspects will be compared to the current operational
practice at Météo-France where 314 original IASI radiances are monitored and about
123 assimilated. The main components of the work are as follows.
1. Install and test the various tools and data (available from the NWP SAF) needed
to reconstruct the radiances on local computers. The computational efficiency of the
reconstruction of the original spectrum will be evaluated for operational purposes. This
could lead to the selection of a reduced number of PC scores to reconstruct the IASI
radiances.
2. Monitor the original set of selected IASI radiances and the reconstructed ones
through the 4D-Var assimilation system “screening”. Special attention will be paid to
low peaking channels and to channels sensitive to water vapour since they have rather
large variability. The variational bias correction scheme will be applied in the 4D-Var
to the reconstructed radiances and resulting biases will be compared to those from the
original radiances. The comparison will be restricted to clear sky pixels.
3. Examine the specification of observation errors. Reconstructed radiances should
have lower error values (noise reduction) but these errors should be correlated between
channels (and this could require an artificial increase of diagonal errors). Diagnostics
will be performed on background and analysis departures to estimate observation error
variances and correlations using the a-posteriori Desroziers method.
4. Prepare report for publication on the NWP SAF web-site. The conclusions from
this study will provide guidance on the strength and weaknesses of the reconstructed
IASI radiances in various aspects of a global NWP data assimilation system (e.g. bias
correction, cloud detection, channel selection, error specification). Recommendations
will be given on the use of PC scores for the monitoring and the assimilation of reconstructed radiances both from a data provider perspective and from a data user
perspective. The study will be undertaken using the same version of the RTTOV radiative transfer model to simulate real and reconstructed radiances, with an implicit
assumption that it is not a dominant source of differences.
5
2
Introduction
This report focuses on the assimilation of reconstructed radiances from PC compression
in the framework of NWP applications. In the last few years, with each generation of
satellite instruments, the data volumes have increased dramatically. The Infrared Atmospheric Sounding Interferometer (IASI; Cayla, 2001) is an infrared Fourier transform
spectrometer which was first launched on the MetOp-A satellite in October 2006. It is
the first such instrument which is truly operational, and has 8461 channels covering the
spectral interval from 645 to 2760 cm−1 at a resolution of 0.5 cm−1 (apodised). IASI
radiances are currently being assimilated operationally at a number of NWP centres
(e.g. Hilton et al., 2011; Guidard et al.,2011). A significant challenge in the coming
decade will be to accommodate MTG-IRS, the hyperspectral sounder on Meteosat Third
Generation (MTG), scheduled to launch in 2021.
Rationale
The ability to assimilate hyperspectral sounding data in the form of reconstructed radiances is likely to be very useful for making improved use of IASI and CrIS data in
the NWP context and for preparing for MTS-IRS data. Indeed, the recent growth in
satellite radiances from infrared hyperspectral instruments available to the NWP community raises issues due to the huge increase in data volumes both for assimilation and
storage. The dissemination cost is also becoming an issue for satellite agencies. An
efficient representation of these satellite radiances is therefore highly desirable. The
Principal Component Analysis (PCA) is an appealing method for compressing information without significant losses. The PCA has also filtering properties that could be
of interest by reducing the noise from the original data. Since February 2011, Eumetsat has been operationally disseminating IASI PC scores through EumetCast. Even
though the direct assimilation of PC scores could be possible, as experienced recently
by ECMWF, it would necessitate, for many NWP centres, a significant investment to
adapt many aspects of their data assimilation systems. An interesting alternative is
the direct assimilation of reconstructed radiances. Such activity has started at the Met
Office but also needs to be undertaken by other NWP centres. Since the generation of
PC scores relies on various hypotheses regarding the size of the training database (that
could lead to representativeness issues) and the noise normalisation (that requires the
specification of an error covariance matrix), real and reconstructed radiances could have
different properties. It is therefore important to investigate whether the choices made at
Eumetsat in order to establish the current real time IASI PC scores are compatible with
the requirements of NWP centres that would like to assimilate reconstructed radiances.
6
3
Monitoring of the original and reconstructed set
of IASI radiances
3.1
Pre-processing of IASI Radiances
The operational Eumetsat stream of PC scores from Metop-A and Metop-B has been
taken from the EumetCast distribution. Auxiliary data, namely the Eigenvectors, Noise
and Mean for the 3 bands, have been downloaded from the Eumetsat website in HDF5
format. The routines to reconstruct radiances have been extracted from the NWPSAF PCA package and they have been adapted and included in Météo-France NWP
pre-processing tool. In order to save some computational time, only the subset of 314
IASI channels, which are routinely monitored in operations, are reconstructed. Then,
the reconstructed radiances undergo the same treatment as the original ones. Figure 1
shows an example of reconstructed versus original raw radiances. Comparable structures
and brightness temperature (BT) amplitude are found for both datasets after this preprocessing step.
Original radiances : channel 3002
Reconstructed radiances : channel 3002
260
260
o
60 N
250
o
60 N
250
30oN
240
30oN
240
o
o
230
0
o
30 S
o
60 S
180oW
120oW
60oW
0o
60oE
120oE
180oW
230
0
o
220
30 S
220
210
60 S
o
210
200
180oW
120oW
60oW
0o
60oE
120oE
180oW
200
Figure 1: Maps of original raw BT (left) and reconstructed BT (right) for the 20/11/2014
- 09UTC to 15UTC.
3.2
Objective comparison between original and reconstructed
radiances
The operational monitoring of raw radiances is actually performed in BT space. The
monitoring of reconstructed radiances is also done in BT space. Statistic are computed
for different conditions: land and sea surfaces using the ARPEGE land-sea mask, clear
and cloudy conditions using the AVHRR cloud cover product. Statistics are also applied
seperately on the 3 bands:
• band 1 : from 648,75 to 1142.5 cm−1 (temperature, window and ozone)
• band 2 : from 1142.75 to 1927.25 cm−1 (humidity)
• band 3 : from 1927.5 to 2646.5 cm−1
Among the pre-selected 314 channels, 165 are located in Band 1, 104 in Band 2 and
45 in Band 3. Figure 2 shows spectrum of BT mean and standard deviation (STD) for
the original and reconstructed radiances within the 6-hour time window centered on the
7
20/11/2014 at 00UTC, which means 80720 pixels × 314 channels. Very good agreement
is found between original and reconstructed IASI BT mean et standard deviation show
similar values. As mentioned in Table 1, larger errors affect reconstructed radiances over
land surface and in cloudy conditions, especially for band 3. Among the pre-selected
314 channels, 165 are located in band 1, 104 in band 2 and 45 in band 3. As shown in
Figure 3, relatively to band 1 which exhibits best scores, errors in band 3 are increased
by a factor close to 6 for global, 4 over land, 3 over sea, 5 in cloudy pixels and 4 in clear
sky conditions.
a) Original Radiances (OR) vs Reconstructed Radiances (RR) : MEAN
280
OR avg
RR avg
Mean Bt (K)
260
240
220
200
500
1000
1500
2000
2500
3000
−1
Wavenumber (cm )
b) Original Radiances (OR) vs Reconstructed Radiances (RR) : STDEV
25
OR stdev
RR stdev
Stdev Bt (K)
20
15
10
5
0
500
1000
1500
2000
2500
3000
Wavenumber (cm−1)
Figure 2: a) Mean and b) standard-deviation of original BT (red line) and reconstructed
(black line) as a function of the wavenumber, for data within the 6-hour time window
centered on the 20/11/2014 at 00UTC. The channel plotted are those used operationally
at Météo-France.
All bands (N=314) correlation
bias
STD
global
1.0000
0.0182 0.0447
1.0000
0.0268 0.0360
land
1.0000
0.0423 0.0735
1.0000
0.0271 0.0529
sea
1.0000
0.0143 0.0354
1.0000
0.0278 0.0321
cloudy
1.0000
0.0271 0.0417
1.0000
0.0432 0.0385
clear
1.0000
0.0141 0.0476
1.0000
0.0140 0.0371
Table 1: Correlation, bias and standard-deviation of original versus reconstruced BTs
for all bands for a night-time period (6-hour window centered on 20/11/2014 at 00UTC)
and a day-time period in italic (6-hour window centered on the 20/11/2014 at 12UTC).
The selected channels are those used operationally at Météo-France (N=314).
8
BAND 2
BAND 3
0.1
BAND 4
0.1
0.1
BIAS
STD
BIAS
STD
0.09
0.09
0.08
0.08
0.08
0.07
0.07
0.07
0.06
0.06
0.06
0.05
Bt error (K)
0.09
Bt error (K)
Bt error (K)
BIAS
STD
0.05
0.05
0.04
0.04
0.04
0.03
0.03
0.03
0.02
0.02
0.02
0.01
0.01
0.01
0
global
land
sea
clear
cloudy
0
global
land
sea
clear
cloudy
0
global
land
sea
clear
cloudy
Figure 3: Bias and standard-deviation of original versus reconstructed BTs on the daytime period (6-hour window centered on 20/11/2014 at 12UTC) for band 1, band 2 and
band 3. Statistics are applied over a subset of observations for various conditions (all,
land, sea, clear cloudy).
9
Special attention is paid to low peaking channels and to channels sensitive to water vapour since they have rather large variability. 3 channels were selected: channel
226 (701.25 cm−1 , Temperature), 1191 (942.5 cm−1 , Window) and 3002 (1395.25 cm−1 ,
Humidity) may give good and representative examples. Their weighting functions are
plotted in Figure 4.
400
600
800
0
0.2
0.4
0.6
dtau/dln(P)
0.8
1
200
Pressure (hPa)
200
1000
IASI canal 3002
IASI canal 1191
Pressure (hPa)
Pressure (hPa)
IASI canal 226
400
600
800
1000
200
400
600
800
1000
0
0.2
0.4
0.6
dtau/dln(P)
0.8
1
0
0.2
0.4
0.6
dtau/dln(P)
0.8
1
Figure 4: Weighting functions for 3 representative channels : 226 (701.25 cm−1 , Temperature), 1191 (942.5 cm−1 , Window) and 3002 (1395.25 cm−1 , Humidity)
Figure 5 shows maps of original and reconstructed BTs averaged over 10 days (2030/11/2014) for the 3 selected channels. The averaged difference is also presented. Maps
show similar structures and amplitudes. Positive bias is drawn close to the ITCZ indicating that the use of reconstructed radiances produces higher value of BTs than the
original radiances in this region. This effect may be related to the presence of clouds
(see section 4.2).
Time series of mean and standard deviation differences (reconstructed minus original) are plotted in Figure 6 for the 3 selected channels. Statistics were applied over data
filtered for different conditions: global, land, sea, clear and cloudy. In all conditions
(global) data, reconstructed BTs exhibit a warm bias and less variability with respect
to orignal BTs. The high-tropospheric temperature channel provides smaller differences
than both other channels (between -0.1 and 0.1). The highest differences concern the
window channel, over land surface. Large biases between original and reconstructed
BTs can reach up to 0.5 K. Water Vapour channel statistics show less differences for
the clear sky class.
3.3
Radiance simulation skills
Both datasets were simulated using the RTTOV radiative transfer model (v10) over 10
days (from 20/11/2014 to 30/11/2014). Atmospheric profiles (temperature, humidity,
ozone) as well as surface parameters (emissivity, skin temperature) were extracted from
the operational ARPEGE system. Input atmospheric and surface fields are the same
for both simulation experiments. Simulations are calculated with a clear sky assumtion.
Figure 7 shows scatterplots of reconstructed BTS minus simulated BTs versus original
10
Original radiances : channel 226
Diff
Reconstructed radiances : channel 226
0.5
o
o
60 N
o
60 N
30oN
30oN
o
230
o
40 N
225
o
0
0
30 S
o
30 S
60oS
60oS
o
220
215
o
180 W
o
120 W
o
60 W
o
0
o
60 E
o
120 E
o
180 W
o
180 W
Original radiances : channel 1191
o
120 W
o
60 W
o
0
o
60 E
o
120 E
60 N
60 N
30oN
30oN
0o
0o
o
30 S
o
30 S
o
60 S
180 W
120 W
60 W
0
o
60 E
280
270
260
250
o
120 E
o
180 W
230
o
180 W
o
60 N
o
30 N
o
120 W
o
60 W
o
0
o
60 E
o
120 E
o
180 W
0
0
30oS
60oS
60oS
180 W
120 W
o
60 W
o
0
o
60 E
o
120 E
o
180 W
o
o
−0.5
0.5
40oN
o
0
0
o
40 S
80oS
245
240
235
230
180 W
o
120 W
o
60 W
o
0
o
60 E
60oE 120oE 180oW
o
120 E
o
180 W
o
80 N
40oN
0o
0
40oS
80oS
180oW 120oW 60oW
0o
60oE 120oE 180oW
Figure 5: Mean original BTs (left column), reconstructed BTs (middle column) and
difference [reconstructed - original] (right column) over 10 days (20-30/11/2014).
11
−0.5
0.5
250
o
0o
Diff
255
o
30oS
o
60 E 120 E 180 W
o
180oW 120oW 60oW
o
o
30 N
o
0
80 N
Reconstructed radiances : channel 3002
o
60 N
o
o
240
Original radiances : channel 3002
o
o
180 W 120 W 60 W
290
o
60 S
o
80oS
Diff
o
o
0
o
40 S
Reconstructed radiances : channel 1191
o
o
0o
o
o
180 W
300
o
80 N
−0.5
0.4
0.3
0.3
0.3
0.3
0.2
0.2
0.2
0.2
0.2
0.1
0
−0.1
−0.2
0.1
0
−0.1
−0.2
0.1
0
−0.1
−0.2
0.1
0
−0.1
−0.2
−0.3
−0.3
−0.3
−0.3
−0.4
−0.4
−0.4
−0.4
−0.5
−0.5
−0.5
−0.5
20 21 22 23 24 25 26 27 28 29 30
November 2014
20 21 22 23 24 25 26 27 28 29 30
November 2014
20 21 22 23 24 25 26 27 28 29 30
November 2014
[RR−OR] (K)
Time series [RR−OR] chan: 226, CLOUDY
0.5
0.4
[RR−OR] (K)
Time series [RR−OR] chan: 226, CLEAR
0.5
0.4
[RR−OR] (K)
Time series [RR−OR] chan: 226, SEA
0.5
0.4
[RR−OR] (K)
Time series [RR−OR] chan: 226, LAND
0.5
0.1
0
−0.1
−0.2
−0.3
−0.4
−0.5
20 21 22 23 24 25 26 27 28 29 30
November 2014
20 21 22 23 24 25 26 27 28 29 30
November 2014
Time series [RR−OR] chan: 1191, CLOUDY
0.5
0.4
0.4
0.3
0.3
0.3
0.3
0.2
0.2
0.2
0.2
0.2
0.1
0
−0.1
−0.2
0.1
0
−0.1
−0.2
0.1
0
−0.1
−0.2
0.1
0
−0.1
−0.2
−0.3
−0.3
−0.3
−0.3
−0.4
−0.4
−0.4
−0.4
−0.5
−0.5
−0.5
−0.5
20 21 22 23 24 25 26 27 28 29 30
November 2014
20 21 22 23 24 25 26 27 28 29 30
November 2014
20 21 22 23 24 25 26 27 28 29 30
November 2014
[RR−OR] (K)
Time series [RR−OR] chan: 1191, CLEAR
0.5
0.4
[RR−OR] (K)
Time series [RR−OR] chan: 1191, SEA
0.5
0.4
[RR−OR] (K)
Time series [RR−OR] chan: 1191, LAND
0.5
[RR−OR] (K)
Time series [RR−OR] chan: 1191, GLOBAL
0.5
MEAN Dif
0.4
STD Dif
0.3
0.1
0
−0.1
−0.2
−0.3
−0.4
−0.5
20 21 22 23 24 25 26 27 28 29 30
November 2014
20 21 22 23 24 25 26 27 28 29 30
November 2014
Time series [RR−OR] chan: 3002, CLEAR
0.5
Time series [RR−OR] chan: 3002, CLOUDY
0.5
0.4
0.4
0.4
0.3
0.3
0.3
0.3
0.2
0.2
0.2
0.2
0.2
0.1
0
−0.1
−0.2
0.1
0
−0.1
−0.2
0.1
0
−0.1
−0.2
0.1
0
−0.1
−0.2
−0.3
−0.3
−0.3
−0.3
−0.4
−0.4
−0.4
−0.4
−0.5
−0.5
−0.5
−0.5
20 21 22 23 24 25 26 27 28 29 30
November 2014
20 21 22 23 24 25 26 27 28 29 30
November 2014
20 21 22 23 24 25 26 27 28 29 30
November 2014
[RR−OR] (K)
Time series [RR−OR] chan: 3002, SEA
0.5
0.4
[RR−OR] (K)
Time series [RR−OR] chan: 3002, LAND
0.5
[RR−OR] (K)
Time series [RR−OR] chan: 3002, GLOBAL
0.5
MEAN Dif
0.4
STD Dif
0.3
[RR−OR] (K)
[RR−OR] (K)
[RR−OR] (K)
[RR−OR] (K)
Time series [RR−OR] chan: 226, GLOBAL
0.5
MEAN Dif
0.4
STD Dif
0.3
0.1
0
−0.1
−0.2
−0.3
−0.4
20 21 22 23 24 25 26 27 28 29 30
November 2014
−0.5
20 21 22 23 24 25 26 27 28 29 30
November 2014
Figure 6: Time series of mean and standard deviation differences (reconstructed minus
original) for several conditions (global, sea, land, clear and cloudy) over 10 days (2030/11/2014), for the 3 selected channels.
BTs minus simulated BTs, for various classes. A better agreement is found in clear sky
conditions.
3.4
Conclusion
Reconstructed and original BTs show a good agreement. One has to point out differences
occuring in cloudy areas, which could be due to problems in the computation of PC
scores in those areas. This could be due to a numerical accuracy in conding in the
BUFR format or in the training database. The numerical representation problem could
also occur during the reconstruction process included in Météo-France pre-processing.
12
Figure 7: Scatterplots of reconstructed BTs minus simulated BTs versus original BTs
minus simulated BTs, columns represent various conditions, one line per channel. Statistics over 10 days (20-30/11/2014).
13
4
Assimilation experiments of reconstructed radiances
Twin assimilation cycles were carried out: the first one assimilating IASI raw BTs
(hereafter RAW) and the other one assimilating reconstructed BTs (hereafter REC).
The cycles begin at the analysis date 7 November 2014 at 18UTC and end at the analysis
date 7 December 2014 at 18UTC. The standard evaluation procedure has been used to
compare the two cycles with each other.
4.1
Impact on the innovations
Innovations are the differences between the observed values and the values simulated
from model fields. Time series of averages (also called biases) and standard deviations
of the innovations are drawn for the 3 reference channels (Figure 8).
0.4
0.4
0.3
0.3
0.2
0.1
0.2
IASI-B - raw
IASI-A - raw
IASI-B - rec
IASI-A - rec
0
[Kelvin]
[Kelvin]
0
0.1
-0.1
IASI-B - raw
IASI-A - raw
IASI-B - rec
IASI-A - rec
-0.2
-0.3
-0.1
-0.4
-0.5
-0.2
-0.6
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
days from the first assimilation
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
days from the first assimilation
[Kelvin]
(a) channel #0226
(b) channel #1191
2
1.9
1.8
1.7
1.6
1.5
1.4
1.3
1.2
1.1
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5
IASI-B - raw
IASI-A - raw
IASI-B - rec
IASI-A - rec
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
days from the first assimilation
(c) channel #3002
Figure 8: Time series of averages (solid lines - lower curves) and standard deviations
(solid lines with bullets - upper curves) for the innovations (observations minus model)
before bias correction. Original BTs are in blue (resp. red) for IASI-B (resp. IASI-A)
and reconstructed BTs are in cyan (resp. orange) for IASI-B (resp. IASI-A).
RAW and REC exhibit similar biases for all channels, both for IASI-A and IASI-B.
The impact of assimilating reconstructed radiances on the variational bias correction
14
(VarBC) implemented within the assimilation system is neutral. REC has smaller standard deviations than RAW for temperature channels (both atmospheric and window
channels), whereas standard deviations are similar for water vapour channels. These
findings are consistent with the Principal Componant Analysis technique which is supposed to reduce the noise, leaving biases unchanged.
4.2
Effects on cloud detection
Two algorithms are used in the assimilation process: the McNally and Watts technique
-hereafter cloud detect- (McNally and Watts, 2003), which flags each channel of each
pixel clear or cloudy, and the CO2-slicing technique (Chahine, 1974; Guidard et al.,
2011), which diagnoses the top pressure and fraction of a single-layer cloud in the pixel.
The ratio of cloudy data obtained with the cloud detect is higher for REC than for
RAW (cf. Figure 9). The ratio increase is about 0.02-0.04 for channels in the lower
stratosphere - upper troposphere.
1
0.1
RAW
REC
0.9
0.09
0.08
0.07
0.7
ratio difference
ratio of cloudy channel
0.8
0.6
0.5
0.4
0.06
0.05
0.04
0.03
0.3
0.02
0.2
0.01
0.1
0
0
-0.01
0
40
80
120
160
200 240 280 320
IASI channel number
360
400
440
480
520
(a) comparison between RAW and REC
0
40
80
120
160
200 240 280 320
IASI channel number
360
400
440
480
520
(b) difference REC minus RAW
Figure 9: Ratio of data flagged cloudy with the McNally and Watts technique for each
channel in the long-wave temperature region.
The CO2-slicing technique exhibits neutral impact on retrieved cloud top pressures
(cf. Figure 10). Retrieved cloud fractions are more often equal to 1 with REC than
with RAW. Which is consistent with the results on the McNally and Watts technique.
Both techniques use bias-corrected innovations as inputs. Even though the innovations and the bias correction applied to them are similar between RAW and REC (as
seen in previous paragraph), the small differences seem sufficient to modify the cloud
characterisation techniques which are very sensitive.
4.3
Analysis and forecast skills
Statistics (averages and standard deviations) on innovations (observations minus model)
and residuals (observations minus analysis) have been calculated for the whole set of
observations assimilated in ARPEGE (radiosoundings, aircraft data, winds derived from
satellite imagery, scatterometer, GNSS radio-occultation, radiances from polar-orbitting
15
10
5
RAW
REC
9
RAW
REC
8
4
frequency (%)
frequency (%)
7
3
2
6
5
4
3
1
2
0
0
1
0
50 100 150 200 250 300 350 400 450 500 550 600 650 700 750 800 850 900 950 1000 1050
cloud top pressure (hPa)
0
0.1
0.2
(a) cloud top pressure
0.3
0.4
0.5
0.6
cloud fraction
0.7
0.8
0.9
1
(b) cloud fraction
Figure 10: Comparison of cloud parameters retrieved using the CO2-slicing technique
for RAW (black lines) and REC (red lines).
and geostationary satellites, etc.). Figure 11 gives some statistics for temperature observations from the radiosoundings. Differences between statistics for RAW and REC
are hardly visible (same conclusion for all other observations). Impact of assimilating
reconstructed radiances instead of raw radiances appears to be negligible on the analyses
and 6-hour forecasts. Thus, no scores have been calculated on longer term forecasts.
exp - ref
STD.DEV
-8
13
-4
9
34
4
-4
-8
Pressure (hPa)
5.0
10
20
30
50
70
100
150
200
250
300
400
500
700
850
1000
10
8
-8
-33
-41
-57
2
0
0.6
1.2
1.8
2.4
nobsexp
BIAS
16047
76633
81115
95080
102880
101380
115099
105302
100400
92698
107672
125246
167026
174430
170656
59957
3
5.0
10
20
30
50
70
100
150
200
250
300
400
500
700
850
1000
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
Figure 11: Bias (right) and standard deviation(left) statistics for radiosounding temperature over the whole month. RAW is in red, REC in black. Solid lines represent
statistics for observations minus 6-hour forecast and dashed lines represent statistics for
observations minus analysis. Column nobsexp is the number of radiosounding observations assimilated for each vertical range, column exp-ref is the difference between REC
and RAW.
4.4
Potential changes in observation error correlations
A posteriori diagnostics have been calculated to evaluate the observation errors. The
so-called Desroziers method (Desroziers et al., 2005) has been applied to both raw and
reconstructed BTs. This technique can provides the full observation error covariance
16
matrix (the R matrix), but only for channels that have been assimilated. The R matrix
can be splitted into observation error standard deviations (σo ) and correlation matrix.
0.9
raw radiances
reconstructed radiances
0.8
0.7
sigma_o [K]
0.6
0.5
0.4
0.3
0.2
0.1
0
700
800
900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000
wavenumber [cm-1]
Figure 12: Observation error standard deviations obtained from the a posteriori
Desroziers diagnostic
Standard deviations are shown in Figure 12. For those channels in band 2 (water
vapour channels), σo are rather similar between REC and RAW. σo in band 1 are dramatically lower for the reconstructed BTs than for the original BTs. When comparing
the correlation part of the R matrices (Figure 13), it turns out that this decrease in
standard deviations is linked to an increase of the inter-channel correlations. This is
consistent with what is known of the PC analysis technique.
17
-999
-1
-0.5
-0.25
0
0.25
0.5
0.7
0.8
0.9
2
-999
(a) raw radiances
-1
-0.5
-0.25
0
0.25
0.5
0.7
0.8
0.9
2
(b) reconstructed radiances
Figure 13: Correlation part of the R matrices obtained from the a posteriori Desroziers
diagnostic. On x-axis, wavenumbers increase from left to right; on y-axis, wavenumbers
increase from bottom to top. The temperature channels (band 1) are in the left lower
corner and the water vapour channels (band 2) are in the top right corner.
18
5
Conclusion and future plans
A pre-processing chain has been set up in the Météo-France research environment to
get Eumetsat BUFR of IASI PC scores and reconstruct radiance spectra from these PC
scores. Reconstructed data have been compared with original data and globally showed
a good agreement. Nevertheless, some discrepancies occuring either over cloudy areas
or over land indicating that some numerical errors may occur either in the coding of the
BUFR files or in the reconstruction of the spectra in Météo-France pre-processing. As
only one pixel out of 8 is monitored in the Météo-France global model, further studies
in the frame of the convective-scale model AROME could be of interest (all pixels are
monitored in AROME).
When ingested in the assimilation process of the global model ARPEGE, reconstructed BTs have the same impact on analyses and forecasts as the original BTs. No
impact is seen on the bias correction, which means that assimilating reconstruted BTs
introduces no bias in the system. The standard deviations of observations minus simulations from model fields are smaller for reconstructed BTs than for the original BTs,
which is consistent with the a posteriori Desroziers diagnostic. Especially in band 1,
reconstructed BTs are less ”noisy” and could be assimilated with smaller σo (which
means with a larger weight). But it appears that the PCA technique introduces interchannel correlation in the observation errors. As no inter-channel correlation is used
in ARPEGE assimilation for the time being, some non-neutral impact of assimilating
reconstructed BTs with the proper R matrix may be found. This study should be
performed again when the possibility of using a non-diagonal R matrix is available in
ARPEGE.
Results are consistent with those obtained at the Met-Office (F. Smith, pers.comm.).
19
References
Cayla, F. (2001). L’interféromètre IASI, un nouveau sondeur satellitaire à haute
résolution. La Météorologie 8ème série 32, 23–39.
Chahine, M. (1974). Remote sounding of cloudy atmospheres. i: The single cloud
layer. J. Atmos. Sci..
Desroziers, G., L. Berre, B. Chapnik, and P. Poli (2005). Diagnosis of observation,
background and analysis error statistics in observation space. Quart. J. Roy. Meteor. Soc. 131.
Guidard, V., N. Fourrié, P. Brousseau, and F. Rabier (2011). Impact of IASI assimilation at global and convective scales and challenges for the assimilation of cloudy
scenes. Quart. J. Roy. Meteor. Soc. 137, 1975–1987.
Hilton, F., R. Armante, T. August, C. Barnet, A. Bouchard, C. Camy-Peyret,
V. Capelle, L. Clarisse, C. Clerbaux, P.-F. Coheur, A. Collard, C. Crevoisier,
G. Dufour, D. Edwards, F. Faijan, N. Fourrié, A. Gambacorta, M. Goldberg,
V. Guidard, D. Hurtmans, S. Illingworth, N. Jacquinet-Husson, T. Kerzenmacher,
D. Klaes, L. Lavanant, G. Masiello, M. Matricardi, A. McNally, S. Newman,
E. Pavelin, S. Payan, E. Péquignot, S. Peyridieu, T. Phulpin, J. Remedios,
P. Schlüssel, C. Serio, L. Strow, C. Stubenrauch, J. Taylor, D. Tobin, W. Wolf,
and D. Zhou (2011). Hyperspectral earth observation from IASI: four years of
accomplishments. BAMS 93, 347–370.
McNally, A. and P. Wats (2003). A cloud detection algorithm for high-spectralresolution infrared sounder. Quart. J. Roy. Meteor. Soc. 129, 3411–3423.
20