On the relative importance of SST, Argo and altimetry for an ocean reanalysis

1
On the relative importance of SST, Argo and
altimetry for an ocean reanalysis

• Peter Oke
• May 2007
• CSIRO Marine and Atmospheric Research

2
Introduction

• Bluelink a partnership between the
• Bureau of Meteorology, CSIRO and
• the Royal Australian Navy
• Talk Outline
• Ocean model
• Data Assimilation
• Bluelink ReANalysis, BRAN
• Observing System Experiments
• Ocean Reanalyses and GHRSST

3
Ocean Forecasting Australia Model, OFAM

• Global configuration of MOM4
• Eddy-resolving around Australia
• 10 m vertical resolution to 200 m depth
• Surface fluxes from ECMWF (for reanalyses)

every 10th grid point shown
4
Bluelink Ocean Data Assimilation System, BODAS

• Ensemble OI
• model-generated covariances
• like least squares fitting to model-based
  anomalies
• 72-member ensemble of model anomalies from a
  10-year run
• Multivariate assimilation system
• Assimilates observations of SLA, SST, in situ T
  and S

5
Bluelink ReANalysis, BRAN

• BRAN1.5
• 1/2003 6/2006
• Forced with ECMWF forecast fluxes
• Assimilates observations once per week
• Assimilates SLA from Jason, Envisat and GFO (T/P
  with-held)
• Assimilates AMSR-E SST
• Assimilates T and S from Argo and ENACT database

6
BRAN1.5 SST and SLA
7
Comparisons with with-held T/P altimetry AMSR-E
SST

• ?SLA RMS errors 8-10 cm
• ? SST RMS errors 0.7o

8
Observing System Experiments (analysis only)
vertical projection of surface observations
Plan view of sea-level increments
Cross-section of temperature increments
9
Observing System Experiments
2003
10
RMS Temp Error over top 1000 m w/
No Assim
No SST
No Altimetry
No Argo/XBT
11
What data to assimilate?A perspective from an
SST user

• ?What data should I assimilate? L2P L3 L4
• Assimilation or validation against 1/4o Rey SST
  are not encouraging
• (eclipse.nodc.noaa.gov/pub/OI-daily/NetCDF/1993/AV
  HRR)
• Revert to AMSR-E or Pathfinder SST

12
What data to assimilate?A perspective from an
SST user
13
BRAN1.5 vs 1/4o Rey SST anomalies
BRAN1.5 assimilates AMSR-E SST
14
BRAN2.1 vs Composite SST anomalies
BRAN2.1 (currently running) assimilates 54 km
PATHFINDER and AMSR-E SST
15
Conclusion

• BRAN realistically reproduce the 3-d time-varying
  mesoscale ocean circulation around Australia
• Each component of the GOOS has a unique and
  important contribution to the forecast skill of
  upper ocean temperature
• Assimilation of SST is essential if you want to
  get SST right!
• Which GHRSST products for reanalysis?

16
Thankyou
17
BRAN1.5 Short-range forecast skill of
SST?Comparisons with AMSR-E
18
BRAN1.5 Comparisons with with-heldalong-track
altimetry
BRAN Nowcast
Aviso
BRAN 7d Forcst
BODAS
19
Observing System Experiments (analysis only)
vertical projection of surface observations
Plan view of sea-level increments
Cross-section of temperature bkgnd (grey)
analysis (black-colour)
20
Comparisons with independent T/P altimetry

• Comparisons between BRAN1.5 and with-held T/P
  altimetry
• ? RMS error of 8-10 cm
• ? anomaly correlations of 0.6

21
Comparisons with AMSRE

• Comparisons between BRAN1.5 and AMSR-E SST (every
  7th day is assimilated)
• RMS error of 0.7o
• anomaly correlations of 0.7

22
Observing System Experimentscomparisons with
Reynolds 1/4o SST (left) and CLS 1/3o SLA

• For the 2003 - GOOS
• the exclusion of SST has the greatest impact on
  forecast skill of SST
• the exclusion of SST has minimal impact on
  forecast skill of SLA

23
Observing System Experiments

• Metric
• Depth average (0-1000 m) of the RMS error in
  potential temperature

For the 2003 - GOOS ?each component of the GOOS
has a unique and important contribution to the
forecast skill of upper ocean temperature ?each
component has comparable impact on the forecast
skill of the upper ocean temperature
24
Bluelink Ocean Data Assimilation System, BODAS

• Ensemble OI
• 72-member ensemble of model anomalies from a
  9-year run
• Multivariate assimilation system

Single point assimilation
25
Bluelink Ocean Data Assimilation System, BODAS

• Ensemble OI
• 72-member ensemble of model anomalies from a
  9-year run
• Multivariate assimilation system
• Assimilates observations of SLA, SST, in situ T
  and S
• Spatially resolved estimates of representation
  error for observations

26
Bluelink Ocean Data Assimilation System, BODAS

• Ensemble OI
• 72-member ensemble of model anomalies from a
  9-year run
• Multivariate assimilation system
• Assimilates observations of SLA, SST, in situ T
  and S
• Spatially resolved estimates of representation
  error for observations
• OFAM initialised using simple nudging

27
BRAN1.5 vs Mapped SLA
28
Conclusion

• BRAN1.5 realistically reproduces the 3-d
  time-varying meso-scale ocean circulation around
  Australia

29
Observing System Experiments

• Experiment design
• With-hold each component of the observing system
• 6-month integration (1st half of 2003)
• compare to with-held observations
• treat BRAN1.5, with all observations assimilated,
  as the truth

30
Bluelink Ocean Data Assimilation System, BODAS

• Ensemble OI (like an ensemble Kalman filter, with
  a stationary ensemble of anomalies)
• Typically use a 72-member ensemble of anomalies
  about the models seasonal cycle to approximate
  background error covariances

31
Observing System Experiments (analysis only)
vertical projection of surface observations
Plan view of sea-level increments
Cross-section of temperature increments
32
Observing System Experiments (analysis only)
vertical projection of surface observations
Plan view of sea-level increments
Cross-section of temperature bkgnd (grey)
analysis (black-colour)
33
Observing System Experimentscomparisons with
Reynolds 1/4o SST (left) and CLS 1/3o SLA

• For the 2003 - GOOS
• the exclusion of SST has the greatest impact on
  forecast skill of SST
• the exclusion of SST has minimal impact on
  forecast skill of SLA

34
BRAN1.5 vs Reynolds ¼o SST