Upcoming SlideShare
×

• 563 views

More in: Technology
• Comment goes here.
Are you sure you want to
Be the first to comment
Be the first to like this

Total Views
563
On Slideshare
0
From Embeds
0
Number of Embeds
0

Shares
11
0
Likes
0

No embeds

Report content

No notes for slide

Transcript

• 1. Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis N. G ökhan K asapoğlu Dept . of Electronics and Communication Engineering İ stanbul Technical University , Turkey
• 2. 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. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 3. SAR Data Assimilation R2 Dual Polarized SCWA SAR DATA Forecast Analysis Forecast SAR Feature Extraction Assimilated Observations SAR Forward Model DATA Assimilation Mode l N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 4. 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:
J obs J b
• where:
• x b 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
N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 5. Variational method of solution
• Use an iterative descent algorithm to find the analysis increment ∆x = x-x b 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
-> N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis J Analysis Variable
• 6. 0D-Var SAR Data Assimilation R2 Dual Polarized SCWA SAR DATA SAR Feature Extraction Assimilated Observations SAR Forward Model H(x,  ) DATA Assimilation (0D-Var), J(x) Analysis N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 7. Data Assimilation Formulation
• 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. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• C Band, 5.6 GHz
• HH and HV channels
• 500 km 2 coverage
• 100 m pixel spacing
N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 9. SAR Features r2_20090226_215200
• RadarSAT-2 SCWA SGF Product – HH channel
• 10. SAR Features
• RadarSAT-2 SCWA SGF Product – HV channel
• 11. SAR Features N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 12. SAR Features Observed HH-Variance from r 2_20090226_215200 Observed HH-Dissimilarity from r 2_20090301_220402
• Texture Feature Examples
N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 13. SAR Features Observed HV-Lee Filtered Image from R2_20090226_215200 Observed HV-Entropy from R2_20090301_220402
• Texture Feature Examples
N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 14. Forward Model IT h = 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. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 15. Simulated SAR Features Predicted HH from obs operator and image analysis Observed H H from RadarSAT-2 image
• r 2_20090 301 _2 20402
N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 16. Simulated SAR Features Predicted HV-Entropy from obs operator and image analysis HV-Entropy from RadarSAT-2 image
• r 2_20090 301 _2 20402
N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 17. Sea - Ice Open Water Discrimination Mean and Standard deviation of SAR feature (  0 HH Mean ) for Sea ice (Red) with IC>%95 and Open Water (Blue) versus incidence Angle N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 18. Sea - Ice Open Water Discrimination Mean and Standard deviation of SAR feature (  0 HV) for Sea ice (Red) with IC>%95 and Open Water (Blue) versus incidence Angle N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 19. Sea - Ice Open Water Discrimination Mean 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. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 20. Sea - Ice Open Water Discrimination Mean 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. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 21. Simple Forward Model 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. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 22. 0D-Var Analysis Results Background_Bias Background_Std Analysis_Bias Analysis_std - 0.074 4 0.19 9 -0.0667 0.194 F_ID: 3_4_7_8_9_11_13_23 X a =X b +  x 0D-Var Analysis Result X b : Background State; PM data only R2_20090226_215200 N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 23. 0D-Var Analysis Results Analysis Increment:  x R2_20090226_215200 R: HH Lee-Filtered Image G: HH Variance B: HV Mean N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 24. 0D-Var Analysis Results Analysis Increment:  x R2_20090226_215200 N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 25. 0D-Var Analysis Results Background_Bias Background_Std Analysis_Bias Analysis_std 0.1471 0.3057 0.088 9 0.264 5 R2_201002 21 _ 10 3028 X a =X b +  x X b : Background State; PM data only N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 26. 0D-Var Analysis Results R2_201002 21 _10 3028 Analysis Increment:  x HH H V N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 27. 0D-Var Analysis Results N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 28. Optimal SAR Feature Selection
• Statistical ly 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. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 29. Separability Measures and Discrimination Analysis N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 30. SAR Feature Selection N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis Best SAR feature combination selection Top-Down & Bottom-Up Strategies Analysis Bias as a selection criteria Feature Selection for Incidence Angle Intervals
• 31. Feature Selection for Incidence Angle Intervals N.G. K asapoğlu , IGARSS 2011, Vancouver,Canada July . 29 , 20 11 Synthetic Aperture Radar Data Assimilation for Sea Ice Analysis
• 32. Acknowledgements