1_kasapoglu_igarss11.ppt
TRANSCRIPT
Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
N. Gökhan Kasapoğlu
Dept. of Electronics and Communication Engineering İstanbul Technical University, Turkey
Outline
• Introduction• 0D-Var SAR Data Assimilation • SAR Features• SAR Forward Model• Sea Ice Open Water Discrimination• A Simple Forward Model• SAR 0D-Var Analysis Results• Optimal SAR Feature Selection
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
R2 Dual PolarizedSCWA SAR DATA
Forecast
Analysis
Forecast
SAR Feature Extraction
AssimilatedObservations
SAR Forward Model
DATA Assimilation
Model
SAR Data Assimilation
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Variational data assimilation
• Basic idea: Bayes theorem is used to update pdf of model forecast (x) using all available observations (y)
• If all pdfs Gaussian, maximum likelihood estimate minimizes the function:
)()(2
1)][()][(
2
1])|[ln()( 11
bT
bT HHCpJ xxBxxyxRyxyxx
Jobs Jb• where:• xb is the short-term ice-ocean model forecast used as the background state• B is the background error (forecast error) covariance matrix• y is the vector of observations• R is the observation error covariance matrix• H maps the analysis vector into observation space
)(/)()()( yxx|yy|x pppp
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Variational method of solution
• Use an iterative descent algorithm to find the analysis increment ∆x = x-xb that minimizes the cost function J (requires gradient of J)
• If obs operator is linear, then J is quadratic and possible to easily find unique minimum
• Otherwise, obs operators can be periodically re-linearized during minimization
• Can be difficult to find the global minimum of a non-quadratic cost function (when H highly nonlinear)
• Grey is original J with nonlinear H, Red curves are from linearizing H
→
)()][(][
)(
)()(2/1)][()][(2/1)(
11
11
b
bb
HH
J
HHJ
xxByxRx
xx
xxBxxyxRyxx
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
J
Analysis Variable
R2 Dual PolarizedSCWA SAR DATA
SAR Feature Extraction
AssimilatedObservations
SAR Forward Model H(x,)
DATA Assimilation (0D-Var), J(x)
0D-Var SAR Data Assimilation
Analysis
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Data Assimilation Formulation
)()(2
1),(),(
2
1)( 11 bTbT xxBxxxHyRxHyxJ
• y : the observed SAR features
• H : the forward model, H(x,) is the predicted SAR feature
• x : ice concentration
• : incidence angle
• R : observation error covariance matrix
• B : background error covariance matrix
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
RadarSAT-2 ScanSAR data
• RadarSAT-2 SCWA Dual-polarized data– C Band, 5.6 GHz– HH and HV channels– 500 km2 coverage– 100 m pixel spacing
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
SAR Features
r2_20090226_215200
• RadarSAT-2 SCWA SGF Product – HH channel
r2_20090301_220402
RADARSAT-2 Data and Products © MacDonald, Dettwiler and Associates Ltd. (2009) - All Rights Reserved.
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
SAR Features
• RadarSAT-2 SCWA SGF Product – HV channel
r2_20090226_215200 r2_20090301_220402
RADARSAT-2 Data and Products © MacDonald, Dettwiler and Associates Ltd. (2009) - All Rights Reserved.
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
SAR Features
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Observed HH-Variance from r2_20090226_215200
Observed HH-Dissimilarity from r2_20090301_220402
SAR Features
• Texture Feature Examples
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Observed HV-Lee Filtered Image from R2_20090226_215200
Observed HV-Entropy from R2_20090301_220402
SAR Features
• Texture Feature Examples
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Forward Model
0123
456789
2cos..).1(
...SAR
aaICaWSaIC
SATSDaSATaSDaIThaIThaaICfeature
ITh = ice thickness (from CIS image analysis chart)IC = ice concentration (from CIS image analysis chart) = incidence angle (know)SAT = surface air temperature (from GEM)SD = snow depth (from GEM)WS = wind speed (from GEM) = angle between instrument angle (know) and wind direction (from GEM)
= “optimal” model coefficients estimated from “training data”
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Predicted HH from obs operator and image analysis
Observed HH from RadarSAT-2 image
Simulated SAR Features
• r2_20090301_220402
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Predicted HV-Entropy from obs operator and image analysis
HV-Entropy from RadarSAT-2 image
Simulated SAR Features• r2_20090301_220402
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Sea-Ice Open Water DiscriminationMean and Standard deviation of SAR feature (0 HH Mean) for Sea ice (Red) with IC>%95 and Open Water (Blue) versus incidence Angle
19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49-20
-10
0
10
20
30
40
50
60
70
0 H
H M
ea
n
Incidence Angle (deg)
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 490.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0 HV
Incidence Angle (deg)
Sea-Ice Open Water DiscriminationMean and Standard deviation of SAR feature (0 HV) for Sea ice (Red) with IC>%95 and Open Water (Blue) versus incidence Angle
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 490.5
1
1.5
2
2.5
3
3.5
0 H
V E
ntr
op
y
Incidence Angle (deg)
Sea-Ice Open Water DiscriminationMean and Standard deviation of SAR feature (0 HV Entropy) for Sea ice (Red) with IC>%95 and Open Water (Blue) versus incidence Angle
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 490
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0 H
V D
ata
Ra
ng
e
Incidence Angle (deg)
Sea-Ice Open Water DiscriminationMean and Standard deviation of SAR feature (0 HV Data Range) for Sea ice (Red) with IC>%95 and Open Water (Blue) versus incidence Angle
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Simple Forward Model
)1()()(SAR ICαtpICαtpfeature)H(IC,α OowOiceO
H : Observation operator (forward model operator)IC : Ice Concentration i : Incidence Angle
o = floor (i) o : Rounded Incidence Angle o = 19,20,...,49 for SCWA
Number of Incidence Angle quantization level: 31
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0D-Var Analysis Results
Background_Bias Background_Std Analysis_Bias Analysis_std -0.0744 0.199 -0.0667 0.194
F_ID: 3_4_7_8_9_11_13_23
Xa =Xb+ x0D-Var Analysis ResultXb: Background State; PM data only
R2_20090226_215200
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Analysis Increment: x
R2_20090226_215200R: HH Lee-Filtered ImageG: HH VarianceB: HV Mean
0D-Var Analysis Results
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Analysis Increment: x
R2_20090226_215200
0D-Var Analysis Results
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Background_Bias Background_Std Analysis_Bias Analysis_std 0.1471 0.3057 0.0889 0.2645
R2_20100221_103028
Xa =Xb+ x
0D-Var Analysis Results
Xb: Background State; PM data only
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9R
2_20
1002
21_1
0302
8
Analysis Increment: x0D-Var Analysis Results
HH
HV
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
0D-Var Analysis Results
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Optimal SAR Feature Selection
• Statistically independent features • Forward model generalization capability • Sea-Ice and open water tie points distributions • Analysis results statistics • Weights of factors listed above x
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Separability Measures and Discrimination Analysis
Tk
ii
Cxiw xxkS
i
1
)()()(
k
i
TiiiB nkS
1
))(()(
11 )( bwSStrd
)(
)(2
w
b
Str
Strd
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Tjijijiijjiij mmmmTrTrdd 1111
3 2
1
2
1
)1(2 8/4
ijdTij edd
ji
ji
jijiT
ji mmmmBD2
ln2
1
28
11
BDij eJJM 12
SAR Feature Selection
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Best SAR feature combination selection
Top-Down & Bottom-Up Strategies
Analysis Bias as a selection criteria
Feature Selection for Incidence Angle Intervals
Feature Selection for Incidence Angle Intervals
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Acknowledgements
•Canadian Ice Service (CIS)
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011
S
ynth
etic
Ap
ert
ure
Ra
da
r D
ata
Ass
imila
tion
fo
r S
ea
Ice
An
aly
sis
Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
N. Gökhan Kasapoğ[email protected]
Thank you for your attention!
N.G. Kasapoğlu, IGARSS 2011, Vancouver,Canada July. 29, 2011