Error Assessment and Validation of the
IASI Temperature and Water Vapor
Profile Retrievals
Validation by Radiosonde Correlative
Measurements
N. Pougatchev, T. August, X. Calbet, T. Hultberg, P. Schlüssel, B. Stiller
K. St. Germain and G. Bingham
ITSC-16 Angra dos Reis
May 7, 2008
Objectives
• Scientific/Methodological – Testing of NPP CrIS
Validation Assessment Model by its application to
validation of IASI L2 data against radiosondes
• Scientific/Utilitarian – Assessment of the IASI
Temperature and Water Vapor retrieval errors in the form
that can be utilized by the community – regionally specific
Covariance and Bias
Validation by Correlative Measurements
True
State
xsat
Validation
Target
Validated Sounder
ŷsat = ysat (xsat )+εsat
Validation Goal:
Estimate the Error εsat
yˆ sat − y val = ε̂sat Validation output
yval =F(yˆ cor )
Validation Model
True
State
xcor
ŷcor = ycor (xcor )+εcor
Correlative Measurements
Validation Issues
Why do We Need Validation Model
Why We Can NOT Use Correlative Data As Is
• Characteristic Difference– validated sounder and correlative
measurements sample atmosphere differently.
• State Non-Coincidence – correlative measurements are at
different time and location.
Validation Model reconciles the issues by modeling best
linear estimate of the satellite measurements
and assessing the errors
y val =Byˆ cor =ysat(xsat )+εval
IASI Validation Study
• Validation Data Set – radiosondes at Lindenberg
(Germany, 52.21o N, 14.12o E, 112 m a.s.l ). Dedicated launches 1
hour prior and at the overpass time; and synoptic times (0,
12, 6, and 18 UTC)
• Validated parameters – Atmospheric Temperature and
Water Vapor Vertical Profiles.
• Validated System – IASI characterized by averaging
kernels.
• Validated Data Set – EUMETSAT v. 4.2 retrievals; cloud
clear; ± 1O Lat. and Long. about Lindenberg
Overpasses and Sonde Launches
Overpasses Lindenberg (Germany) site
AIRS(AQUA) and IASI (METOP-A)
0/24 UST
18 h UST
6 h UST
12 h UST
Temperature Non-Coincidence Error
Free Troposphere
RMS non-coincidence error
Averaged between 800 - 300 mb
Sodankyla, Lindenberg,
and Southern Great Plane ARM
3.0
rms error (K)
2.5
2.0
1.5
Sodankyla
Lindenberg
SGP ARM
1.0
0.5
0.0
0
1
2
3
4
5
6
7
8
9 10 11 12 13
Time non-coincidence (h)
Averaging Kernels Temperature
Averaging Kernels
Temperature retrieval
Lindenberg June-August 2007
100
Sum of diagonals=14.18
Pressure (mb)
200
300
400
500
600
700
800
900
1000
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
Averaging Kernels
AvKe_IASI_LND_2007.JNB
Temperature Retrieval Error
Variances/rms
Temperature
IASI Retrieval Errors and Variances
100
Expected retrieval - xˆ exp
xˆ exp = xa + A(x true - xa ) + ε
Sexp = (I - A)S x (I - A)T + Sε
Expected covariance − AS x A T + Sε
Pressure (mb)
Expexted error - xˆ exp - x true
200
300
Variances
400
Errors
500
600
700
800
900
1000
0
1
2
3
rms (K)
4
5
Total expected error
Total actual error
Repeatability/ retrieval noise
Variance actual retrievals
Variance sondes
Variance expected retrievals
6
Temperature Retrieval Error
Covariance Matrices
Actual
Total Error Temperature Retrieval
Correlation Matrix
100
100
200
Pressure (mb)
200
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
300
400
500
600
700
800
900
1000
100
Actual
Pressure
(mb)
300
400
500
600
700
800
900
1000
100
Expected
Pressure (mb)
Total Error Temperature Retrieval
Rows of Correlation Matrix
1000
Total Error Temperature Retrieval
Rows of Correlation Matrix
16
14
14
12
12
Altitude (km)
16
Altitude (km)
Pressure (mb)
Expected
Total Error Temperature Retrieval
Correlation Matrix
10
8
6
1000
10
8
6
4
4
2
2
0
0
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
Temperature Retrieval Error
Bias
Temperature Retrieval
Bias against Radiosondes
100
Pressure (mb)
200
Sonde launch time
about IASI overpass
300
0 min
54 min
0-54 min
1.5-3.6 h (synoptic times)
400
500
600
700
800
900
1000
-2
-1
0
Bias (K)
1
2
Relative Humidity Non-Coincidence Error
Free Troposphere
RMS non-coincidence error
Averaged between 800 - 300 mb
Sodankyla, Lindenberg,
and Southern Great Plane ARM
30
rms error (% RH)
25
20
15
Lindenberg
Sodankyla
SGP ARM
10
5
0
0
1
2
3 4 5 6 7 8 9 10 11 12 13
Time non-coincidence (h)
Averaging Kernels Relative Humidity
Averaging Kernels
RH retrieval
Lindenberg June-August 2007
100
Sum of diagonals=10.14
Pressure (mb)
200
300
400
500
600
700
800
900
1000
-0.1
0.0
0.1
0.2
0.3
0.4
Averaging Kernel
AvKe_IASI_LND_2007.JNB
Relative Humidity Retrieval Error
Variances/rms
Relative Humidity
IASI Retrieval Errors
100
100
200
Pressure (mb)
200
Pressure (mb)
Relative Humidity
IASI Retrieval and Sonde Variances
300
400
300
400
500
500
600
700
800
900
1000
600
700
800
900
1000
0
10
20
rms (% RH)
30
Repeatability
Total expected error
Total actual error
Variance RH Lindenberg
0
10
20
rms (% RH)
30
Actual retrieval variance
Expected retrieval variance
Sonde profile variance
Relative Humidity Retrieval Error
Covariance Matrices
Actual
Total Error RH Retrieval
Correlation Matrix
100
100
200
200
-0.4
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
300
400
500
600
700
800
900
1000
100
1000
300
400
500
600
700
800
900
1000
100
16
Actual
Total Error RH Retrieval
Rows of Correlation Matrix
16
14
12
12
Altitude (km)
14
10
8
6
1000
Pressure (mb)
Pressure (mb)
Altitude (km)
Pressure (mb)
Expected
Total Error RH Retrieval
Correlation Matrix
Expected
Total Error RH Retrieval
Rows of Correlation Matrix
10
8
6
4
4
2
2
0
0
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
Relative Humidity Retrieval Error
Bias
RH Retrieval
Bias against Radiosondes
100
Pressure (mb)
200
Sonde launch time
about IASI overpass
300
0 min
54 min
0-54 min
Inter-sondes interpolation
400
500
600
700
800
900
1000
-20
-10
0
Bias (% RH)
10
20
IASI and AIRS Performance
Temperature
Validation T retrievals vs. radiosondes
IASI (Lindenberg) and AIRS (ARM SGP)
IASI - red lines; AIRS - black lines
100
(AIRS - Tobin et al., JGR 2006, 111, D09S14, doi:10.1029/2005JD006103)
200
300
Pressure (mb)
Variance of the sondes
400
rms
500
600
bias
700
800
900
1000
-2
-1
0
1
2
T (K)
3
4
5
6
IASI and AIRS Performance
Relative Humidity
Validation Water Vapor retrievals vs. radiosondes
IASI (Lindenberg) and AIRS (ARM SGP)
IASI - red lines; AIRS - black lines
100
(AIRS - Tobin et al., JGR 2006, 111, D09S14, doi:10.1029/2005JD006103)
200
Pressure (mb)
300
400
rms
Variance of the sondes
500
600
bias
700
800
900
1000
-20
0
20
40
60
80
H2O %
This is NOT Relative Humidity!!!
100
Conclusions
• At Lindenberg state dependent error dominates the total
error for both for Temperature and Relative Humidity.
That implies that retrieval errors should be characterized
regionally and seasonally specifically.
Conclusions
• Temperature
– Our Validation Assessment Model and radiosondes allow to
account for non-coincidence error and finite vertical resolution of
IASI and assess retrieval errors accurately:
– Variances/rms of a single FOV retrieval are ≤1K between 800 –
300 mb with increase to 2 K in tropopause and at the surface
– Bias against radiosondes oscillates within ± 0.5K between 950 –
100 mb.
– Actual vertical resolution apparently lower than expected in midtroposphere and tropopause regions
Conclusions
• Relative Humidity
– Complex spatial structure of highly variable RH field does not
allow to assess retrieval errors accurately:
– Variances/rms of a single FOV retrieval are between 6 - 18 % RH
in 800 – 300 mb with best performance at 600 mb
– No statistically significant bias was detected between 800 – 300
mb.
– Actual vertical resolution apparently lower than expected in midtroposphere region
– Techniques other than radiosondes, e. g. NAST-I in combination
with drop-sondes, are needed for accurate RH retrieval error
assessment.
Conclusions
• IASI and AIRS retrieval performance are in reasonable
agreement with direct comparison needed for more
definitive conclusion
• The Validation Assessment Model analysis of temporal
variations of T and RH at Lindenberg and ARM SGP sites
reveals that Lindenberg is better for T validation (synoptic
time launches can be used) and ARM is better for RH
validation.
• This approach can be used for Cal/Val planning through
optimal validation tools and sites selection and
characterization.