Surface Marine Wind Retrieval in Non-Precipitating Regions

1
Surface Marine Wind Retrievalin
Non-Precipitating Regions
Harold Ritchie, Richard Danielson, and Michael
Dowd Canadian Meteorological Centre and Dalhousie
University

• Nonlinear regression (2D-Var) approach
• Radarsat-1 synthetic aperture radar (SAR)
• Validation using ship and buoy winds
• SAR error estimates
• Conclusions

2
Regression Approach

• The assimilation of SAR data
• into models depends partly on
• whether errors in SAR-wind
• information can be quantified
• the identification of conditions
• for which SAR data improve
• marine wind forecasts.
• Errors can be explicitly quantified using
  nonlinear regression.
• These may be related to physical processes (e.g.,
  wave tilt,
• precipitation impact) or satellite
  characteristics (e.g., beam
• mode, incidence angle).

3
Regression Approach
SAR backscatter cross
section
SAR errors Numerical Model
winds
Model errors
4
Regression Approach

Hersbach (2003), Vachon and Dobson (2000)
SAR backscatter cross section

SAR errors Numerical Model winds
Model
errors
Radarsat-1 incidence angle bias correction
5
Regression Approach

• CMOD is first used to remove the incidence angle
• dependence of the SAR obs ( ). This allows R
  to
• be positive definite.
• J is generally a function of the estimated winds
  (x)
• and the unknown error covariances (R and B).
• Here, error covariances are assumed to decay
• exponentially with a length scale of 150 km and
  B
• error variances are fixed at 1 m2/s2 (only R
  varies).

6
Radarsat-1 SAR

• Polar orbiting every 100 minutes at 800 km
• C-band SAR (5-cm wavelength horizontally
  polarized)
• First ScanSAR to use multiple beam modes to
  obtain
• 50-m resolution over swaths of 400 km
• We employ 609 acquisitions from June 2004 to
• July 2005 at 6.4-km resolution

7

• 400-m SAR Acquisition
• (Koch 2004 smoothing)
• masking over land

Backscatter (dB)
8

• 400-m SAR Acquisition
• (Koch 2004 smoothing)
• masking over land
• along beam seams

Backscatter (dB)
9

• 400-m SAR Acquisition
• (Koch 2004 smoothing)
• masking over land
• along beam seams
• over sea ice

Backscatter (dB)
10

• 400-m SAR Acquisition
• (Koch 2004 smoothing)
• masking over land
• along beam seams
• over sea ice
• where retrieved wind speed
• would be less than 3 m/s or
• greater than 33 m/s

Backscatter (dB)
11

• 800-m SAR Acquisition
• (Koch 2004 smoothing)
• masking over land
• along beam seams
• over sea ice
• where retrieved wind speed
• would be less than 3 m/s or
• greater than 33 m/s

Backscatter (dB)
12

• 1.6-km SAR Acquisition
• (Koch 2004 smoothing)
• masking over land
• along beam seams
• over sea ice
• where retrieved wind speed
• would be less than 3 m/s or
• greater than 33 m/s

Backscatter (dB)
13

• 3.2-km SAR Acquisition
• (Koch 2004 smoothing)
• masking over land
• along beam seams
• over sea ice
• where retrieved wind speed
• would be less than 3 m/s or
• greater than 33 m/s

Backscatter (dB)
14

• 6.4-km SAR Acquisition
• (Koch 2004 smoothing)
• masking over land
• along beam seams
• over sea ice
• where retrieved wind speed
• would be less than 3 m/s or
• greater than 33 m/s

Backscatter (dB)
15
Ship and Buoy Validation
GTS ship/buoy obs (CDC web archive)
16
Ship and Buoy Validation

• GTS ship/buoy obs
• (CDC web archive)
• vertical adjustment to
• 10-m using Walmsley
• (1988) or logarithmic
• profile requires obs
• heights (WMO Pub 47)

17
Ship and Buoy Validation

• GTS ship/buoy obs
• (CDC web archive)
• vertical adjustment to
• 10-m using Walmsley
• (1988) or logarithmic
• profile requires obs
• heights (WMO Pub 47)
• taken within 90 min of
• an acquisition

18
Ship and Buoy Validation

• GTS ship/buoy obs
• (CDC web archive)
• vertical adjustment to
• 10-m using Walmsley
• (1988) or logarithmic
• profile requires obs
• heights (WMO Pub 47)
• taken within 90 min of
• an acquisition
• valid within a radius of
• 5-50 km, depending on
• proximity to land

19
Retrieval Example
SAR Backscatter (dB)
Precipitation Region
20
Retrieval Example
SAR Backscatter (dB)
Normalized by CMOD
Precipitation Region
21
Retrieval Example
Normalized by CMOD
15-km Hourly Model Winds
xb and CMOD(xb)
22
Retrieval Example
Retrieval
15-km Hourly Model Winds
xb and CMOD(xb)
23
Error Estimates
SAR error variance is reduced (as expected)
Errors appear Gaussian
24
Error Estimates
Wind speed (and direction) errors are unchanged
Errors appear Gaussian
25
Error Estimates
Wind Speed (m/s) Retrieval (mean / std) Error (bias / std) Number of Collocations
Precip Region 11.0 / 4.4 0.6 / 3.4 1542
No Precip 9.5 / 3.1 0.2 / 2.1 1542

• Precip regions have higher error standard
  deviation
• (with slightly stronger wind speeds)

Wind Speed (m/s) Retrieval (mean / std) Error (bias / std) Number of Collocations
High Incidence 7.8 / 3.2 0.0 / 2.2 3520
Low Incidence 7.7 / 3.0 0.0 / 2.5 3580

• Low incidence angle regions (with no precip)
  have
• higher error standard deviation

26
Conclusions

• If errors in ship and buoy obs can be
• neglected, then the regression approach
• permits a distinction between errors with
• and without precipitation and at high and
• low incidence angles.
• A more sophisticated approach considers
• ship and buoy errors (which may be larger
• than corresponding SAR or model errors).
• The B and R error covariance matrices can
• also be improved.

27
Radarsat-1 Incidence Angle Bias
28
Spatial Error Correlation relative to ship and
buoy wind speed and backscatter (using CMOD)
29
CMOD (C-band model) empirically relates wind and
Bragg scattering from waves.