Assimilation of Cross-track Infrared
Sounder radiances at ECMWF
Reima Eresmaa, Tony McNally and Niels Bormann
European Centre for Medium-range Weather Forecasts, Reading, UK
Approach
CrIS instrument
The effect of spectral resolution
We aim at operational assimilation
of Cross-track Infrared Sounder (CrIS)
radiances in the global NWP system
of ECMWF. Experience gathered from
earlier hyper-spectral radiances,
including those from the Atmospheric
Infrared Sounder (AIRS) and Infrared Atmospheric Sounding Interferometer
(IASI), is applied where possible: the
experimental assimilation of new
radiances is limited to cloud-free
channels over sea and sea ice, and the
emphasis is put on the efficient use of
temperature sounding channels in the
15 μm CO2 absorption band.
In contrast to the operational use
of IASI radiances, a large number of
spectrally-adjacent CrIS channels are
assimilated using a relatively aggressive
specification for observation error
standard deviation. This is hoped to
compensate for the poorer spectral
resolution of CrIS instrument. A nondiagonal observation error covariance
matrix is applied to account for the
effect of the signal apodization.
CrIS is a Michelson interferometer
attached to the Suomi-NPP polar-orbiting
satellite launched into an afternoon orbit
in October 2011. The infrared sounder
provides radiance information on
1,305 channels in the wavenumber
range 650 – 2550 cm-1. Field-of-View
diameter is 13.5 km in nadir and global
coverage is produced every 12 hours.
In comparison with earlier hyper-spectral
infrared sounders, lower instrument
noise levels of CrIS compensate for the
effect of the relatively poor spectral
resolution. While past experience
from the assimilation of AIRS and IASI
radiances provides a natural starting
point for the use of CrIS radiances,
modifications are required to properly
account for the different trade-off
between noise and spectral purity.
We aim at maximizing the
meteorological information extracted
A AIRS
I IASI
C CrIS
1.2
1
0.8
0.6
0.4
0.2
0
A
A
A
A
A
A
AA
AA
AA
A
A
AA
A
A
CIII A
A
A
A
A
CIIIII A
AAA AA A
AAAA A
I
I
I
I
A
C IIIIIIIIIIIA
A
II III A
AIA A
A
C IIIIA
A
IIIIIII AIII IA I I A A
A
A
A
I
I
A
I
A
I
I
I
I
I
C
I
I
I
C
C
CC
CCCC CCCCCCCCCC C C C C C
CCCCCCCCCCC
CCC
800
A
II II
IIIAI A
AIII
AA
A
IIA
A
AA
AA
A
A
A AA
C C C
A C
C
AI C
I
AI C
C
CC
10–1
10–1
C
C
667.50
A
667.50
100
100
Pressure (hPa)
OmB St. Dev. (K)
1.4
I I
I
I
A
1000
1200
1400
Wavenumber (1/cm)
1600
677.50
101
677.50
680.00
680.00
102
Figure 1 Global background departure
standard deviation on selected AIRS (blue), IASI
(red) and CrIS (black) channels.
736.25
740.00
103
0
0.1
101
102
736.25
740.00
0.2
0
0.1
Jacobian (K/K)
0.2
Figure 2 Temperature Jacobians on selected
IASI (left) and CrIS (right) channels in a tropical
reference atmosphere.
0
25
22
200
20
ATMS channel number
AIREP T
400
TEMP T
TEMP Q
800
12
NPP ATMS
Departure (K)
Metop-A
AMSU-A
1000
100
Before
After
Gaussian
10
1
0
25
50
75
100 125
Channel rank index
Figure 7 Smoothed (line) and unsmoothed
(dots) background departures on vertically-ranked
channels in two single cases. Channels diagnosed
cloud-contaminated in each case are shown in red.
R.Eresmaa@ecmwf.int
Northern
extratropics
Z
T
W
Tropics
R
Z
T
W
Southern
extratropics
R
Z
T
W
R
200
hPa
500
hPa
850
hPa
-1 -0.5
+0.5 +1
Normalized OmB sdev (%)
Figure 5 Normalized global background
departure standard deviations for some
conventional (left) and space-borne
microwave radiance (right) observations in
a 90-day assimilation experiment. Negative
values imply improved background fit
when CrIS radiances are assimilated.
-2
-3
7
5
is only 20% of that on a high-peaking
stratospheric channel.
The low noise level of CrIS results in
increased sensitivity to contamination by
undetected clouds. Therefore, parameters
of cloud detection need to be tuned more
stringently for CrIS than for earlier hyperspectral sounders. In an appropriatelytuned detection scheme cloud-free
background departures constitute a nearly
symmetrical and Gaussian histogram.
Count
Departure (K)
-1
11
9
10
6
-1.0 -0.5
+0.5 +1.0
Normalized OmB sdev (%)
5
13
14
8
1000
4
NPP ATMS
18
900
3
Metop-A
MHS
1
0
240
220
220
650 675 700 725 750 775
Wavenumber (1/cm)
650 675 700 725 750 775
Wavenumber [1/cm]
Figure 3 Excerpts of IASI (left) and CrIS (right)
long-wave spectra.
Figure 6 Medium-range forecast impact from
adding CrIS radiances in the ECMWF 4D-Var
system. Z, T, W, and R refer to geopotential,
temperature, vector wind, and relative humidity,
respectively. Green (red) indicates positive
(negative) impact. Dot size indicates the
magnitude of the impact.
-0.6
-0.4
-0.2
0
0.2
FG Departure (K)
0.4
0.6
Figure 8 Frequency distribution of background
departure on a mid-tropospheric sounding
channel before (black) and after (red) the cloud
detection re-tuning.
As compared with earlier
hyper-spectral sounders, CrIS
radiances are affected by a
unique trade-off between noise
characteristics and spectral
resolution. Therefore, experience
from the operational assimilation of
AIRS and IASI radiances is insufficient
to facilitate successful assimilation
of CrIS radiances.
We have put emphasis on strong
exploitation of cloud-free sounding
channels in the 15 μm CO2 absorption
band. We demonstrate a positive
medium-range forecast impact in
the extratropics from assimilating
CrIS radiances into the previouslyoperational IFS version at ECMWF.
However, similar level of performance
has not yet been produced in
the currently-operational IFS
version. Work is in progress to fully
understand the nature and impact
of CrIS radiance data in the ECMWF
4D-Var system.
The positive forecast impact from the
baseline configuration does not reproduce
in newer experiments based on a more upto-date IFS version (Cy40r1) and using higher
vertical resolution. Further optimization on
the use of CrIS radiances is being carried out
focussing on the following modifications:
•Pre-selection of the warmest FOV
•12 channels in the water vapour absorption
band added to the assimilation
•Humidity-sensitive channels removed from
the long-wave CO2 absorption band
•Stringent tuning of the cloud detection in
response to the low noise levels
•Observation error standard deviation
relaxed to 0.5 / 0.3 / 1.2 K
The following approaches will be considered
when designing future experiments to
improve the assimilation of CrIS radiances:
•Account for inter-channel error correlations
other than those introduced during the
signal apodization
• Use collocated imager information to assist
the cloud detection: VIIRS-based estimates of
cloud parameters are available for doing this
•Assimilate stratospheric-peaking channels
over land
•Study the impact of CrIS in a degraded
assimilation system, where no AIRS radiance
data is available
0.15
25
0.1
20
0.05
15
0
10
-0.05
Count (thousands)
750
800
850
Wavenumber (1/cm)
Conclusions
50
75
100 125
Channel rank index
240
OmB Mean or St. Dev. (K)
700
-2
-3
260
5
1
2
3
4
5 6 7
Pixel number
8
9
Figure 9 Pixel-number-specific background departure
statistics after Warmest-FOV-based data selection: mean
(black), standard deviation (blue) and count (green).
Statistics are shown for channel 61.
650
660.62
663.75
666.88
672.5
675.62
678.75
681.88
685
688.12
691.25
694.38
697.5
700.62
703.75
708.12
712.5
720
723.75
726.88
736.25
757.5
780.62
935
1442.5
1598.75
230
0
600
Figure 4 Long-wave CrIS spectrum
highlighting assimilated channels. Blue
bars indicate the assumed observation
error standard deviations.
250
McNally & Watts (2003): A cloud detection
algorithm for high-spectral-resolution infrared
sounders. Q. J. R. Meteorol. Soc., 129, 3411—3423.
-1
260
1
1025
900
457
210
173
139
124
119
113
101
94
87
82
77
72
67
62
57
52
47
42
37
28
23
18
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
−0.9
1
18
23
28
37
42
47
52
57
62
67
72
77
82
87
94
101
113
119
124
139
173
210
457
900
1025
0.2 K
270
Cloud detection
0
280
Channel number
290 0.4 K
AMSU-A / MHS channel number
Brightness temperature (K)
0.6 K
650
Pressure (hPa)
Used in experiments with the previously operational
IFS version (Cy38r2) in resolution T511 / L91
•Pre-select the middle FOV (pixel 5) from each
Field-of-Regard
•Assimilate 122 channels from the long-wave CO2
absorption band
•Set observation error standard deviations to 0.4 / 0.2 /
0.6 K
•Assume signal apodization to be the only source of
observation error correlation
•Assimilate data over sea and sea ice only
•Use RTTOV-10 for radiative transfer modelling
•Use variational bias correction with air-mass- and
scan-angle-dependent predictors
•Assimilate one FOV per thinning box (1.25° by 1.25°)
only
A consistently positive medium-range forecast
impact extending over the whole of troposphere is
found in the extratropics, while a negative impact is
seen in the tropical upper troposphere
1
280
The way forward
Baseline configuration
Cloud-contaminated channels are
rejected from the assimilation. Cloud
contamination is diagnosed using a
detection scheme of McNally & Watts
(2003) and it is based on brightness
temperature background departures in
a set of vertically-ranked channels in the
15 μm CO2 absorption band. Because
of cloud contamination, typical count
of usable data on a window channel
103
Brightness Temperature (K)
1.6
from apodized radiances by
using almost all channels in the
most critical spectral range at 670
–730 cm-1. Using a large number
of spectrally-adjacent channels
means that observation error
correlation introduced by the signal
apodization has to be taken into
account in the assimilation.
Furthermore, the low noise levels are
reflected in an aggressive specification
of observation error standard
deviations during the assimilation.
Channel number
Figure 10 Inter-channel observation error correlations as
diagnosed using the Hollingsworth – Lönnberg method
in a selection of 128 CrIS channels.