National Aeronautics and Space Administration
AIRS v. 5 Temperature and Water Vapor Retrievals
Characterization and Error Assessment
Nikita Pougatchev, Van Dang, Evan Fishbein, Bill Irion, and Bjorn Lambritsen
Jet Propulsion Laboratory, California Institute of Technology
Introduction
Concept of Validation Assessment Model for satellite retrievals
We present the characterization and error assessment for
the AIRS v. 5 temperature and water vapor retrievals. We use
dedicated radiosondes for the reference data and Validation
Assessment Model as the tool for error assessment. The
geographic coverage is from tropics to Alaska. In addition to
the estimates of error biases and covariances we infer
averaging kernels from the real measurements data.
Validated Sounder
True
State
xsat
x=x
ˆ ˆ exp +ε
Validation
Output
xˆ -xˆbe =εˆ
x̂be =F(x̂cor)
Validation Assessment Model
The total error depends on instrumental and geophysical
factors. That requires the End-to-End error analysis in the
sense that front end input, i. e. Earth-Atmosphere, as well as
final products to be included into the consideration.
True
State
xcor
Methodology
Temporal Non-coincidence Errors
Direct comparison of the retrievals to radiosondes mapped
onto the AIRS (100 levels) vertical grids
Linear error analysis:
STD non-coincidence error
RMS
Averaged between 800 - 300 mb
Sodankyla, Lindenberg,
and Southern Great Plane ARM
STD non-coincidence error
RMS
Averaged between 800 - 300 mb
Sodankyla, Lindenberg,
and Southern Great Plane ARM
3.0
30
2.5
25
2.0
1.5
Sodankyla
Lindenberg
SGP ARM
1.0
Non-coincidence error
rmserror
error (% RH)
std
rms
std error
error (K)
xˆ - xsonde = x0 + A(xtrue - x0 ) + ε - xsonde
Averaging Kernel - smoothing
x̂cor(xcor ) εcor
Correlative Measurements
0.5
x 0 - xtrue = B(xsonde - xsonde ) + ξ
20
15
Lindenberg
Sodankyla
SGP ARM
10
5
0.0
0
0
1
2
3
4
5
6
7
8
9 10 11 12 13
0
1
2
Time non-coincidence (h)
3 4 5 6 7 8 9 10 11 12 13
Time non-coincidence (h)
Standard Deviation and Bias Humidity
Temperature
Bias ln(AIRS)-ln(Sondes)
400
600
700
700
800
900
900
1000
1000
0.5
1.0
1.5
2.0
2.5
3.0
0.0
Sonde Variance
100
300
600
800
0.0
200
500
0.5
1.0
1.5
2.0
2.5
3.0
std (K)
Pressure (mb)
500
400
800
200
S expexted = (I - A)S sonde (I - A)T + S noise
400
900
-2
600
-1
0
1
800
900
0
1
2
3
4
5
200
300
600
700
800
900
900
1000
0.2
0.4
0.6
0.8
1.0
0.0
0.2
0.4
0.6
0.8
1.0
Sonde Variance
600
TWP
Toulouse
SGP
NSA
Lindenberg
Average bias
700
1000
300
Sexpexted = (I - A)Ssonde (I - A)T + S noise
400
-1.0
-0.5
0.0
0.5
1.0
bias (ln(layer column))
500
600
700
TWP
Toulouse
SGP
NSA
Lindenberg
800
900
1000
6
500
900
200
2
400
800
std (ln(layer column))
100
TWP
TLoulouse
SGP
NSA
Lindenberg
1000
100
Assessed Error
500
800
0.0
bias (K)
700
400
700
1000
500
300
400
600
Pressure (mb)
Pressure (mb)
300
300
1000
TWP
Toulouse
SGP
NSA
Lindenberg
Average
700
200
500
500
600
100
200
Pressure (mb)
300
400
100
Pressure (mb)
300
100
Pressure (mb)
200
Expected Error
Bias T (AIRS-Sondes)
100
200
Pressure (mb)
Pressure (mb)
Assessed Error
Expected Error
100
0.0
std (K)
0.2
0.4
0.6
0.8
1.0
std (ln(layer column))
Retrieval of Averaging Kernels from Correlative Measurements
Averaging Kernel is Correlation Matrix between Retrieval and True State
Approach
ˆ = S S −1
A
ˆ
xx
x
S x and S x̂x
300
300
400
400
500
600
700
700
800
800
900
900
-0.05
500
600
700
800
900
1000
1000
0.00
0.05
0.10
Reported AIRS v. 5 temperature averaging
kernels underepresent the retrievals
Errors estimated with retrieved averaging
kernels are in good agreement with the ones
assessed from correlative measurements
0
-0.05
0.15
0.00
0.05
0.10
0.15
1
2
3
4
5
6
std (K)
Averaging Kernel
Averaging Kernel
 is retrieved averaging kernel
www.nasa.gov
600
1000
are sample cross – and autocovariances
Jet Propulsion Laboratory
California Institute of Technology
Pasadena, California
500
200
200
Pressure (mb)
x
400
Pressure (mb)
ˆ
xx
300
100
Trace=6.7
100
Trace=5.1
200
Pressure (mb)
T
ˆ T } = S xx
E{xx
ˆ = AE {xx } = AS x
S = AS
Temperature std Error
Averaging Kernels Retrieved
Lindenberg
Averaging Kernels Reported
Lindenberg
100
References
Sexpexted =(I-A)Ssonde (I-A)T +Snoise
Total Assessed Error
Total Expected Error - Rterieved AvKe
Total Expected Error - Reported AvKe
1. N. S. Pougatchev, “Validation of atmospheric sounders by correlative measurements,” Appl. Opt. 47(26), pp. 4739-4748
(2008).
2. N. Pougatchev, T. August, X. Calbet, T. Hultberg, O. Oduleye, P. Schlüssel, B. Stiller, K. S. Germain, and G. Bingham, IASI
temperature and water vapor retrievals – error assessment and validation, Atmos. Chem. Phys., 9, 6453-6458, 2009,
www.atmos-chem-phys.net/9/6453/2009/ access 2009.
This work was carried out at the Jet Propulsion Laboratory, California Institute of Technology under a contract with the National Aeronautics and Space Administration. Copyright 2010 California Institute of Technology. Government sponsorship acknowledged