1. Argo Delayed-Mode Salinity Data

2. Acknowledgments Taiyo Kobayashi, FORSGC, Japan Ron Perkin, IOS, Canada
Brian King, SOC, UK
Lars Böehme, IfM Kiel, Germany
Rebecca McCreadie, BODC, UK

3. Introduction Salinity is one of the main measurements made by
Argo profiling floats, but ...

4. How to calibrate without an absolute reference?
… the sensors sometimes give measurements that have artificial trends.
DRIFT
OFFSET θ θ S S
Problem:
How to calibrate without an absolute reference?

5. Argo data quality control elements
Real-time (RT) data stream
Function: Apply agreed RT QC tests to float data.
Assign quality flags.
Users: Operational centres, data assimilation,
researchers needing timely data.
Timeframe: hrs after transmission.
Who/Where: Perform by National Data Assembly
Centres.
Data from floats
Delayed-mode (DM) data stream
Function: Apply accepted DM procedures to float data.
Provide statistically justified corrections using accepted
methods. Provide feedback to RT system.
Users: All needing adjusted data with error estimates.
Timeframe: 6-12 months after transmission.
Who/Where: Perform by PIs with DM Agencies
or Regional Centres.

6. What is done in real-time for salinity data?θ
• Range test
• Spike test
• Gradient test
• Density inversion test • • • • • • • • • • S
Action: “Bad” float salinity data are flagged.
Result: “Good” float salinity data have a “good”
relative vertical profile, but may still
contain artificial drifts and offsets.

7. What is done in delayed-mode for salinity data?
Basic theory …
θ-S spread in a 10°x10° box around 25°S,155°W
• By using reference
datasets and statistical
methods, we can estimate
what salinity should be at
float locations. θ S
from WOD 2001

8. • Confidence of statistical salinity estimates is
θ-S spread in a 10°x10° box around 55°N,45°W
• Confidence of statistical
salinity estimates is
dependent on reference
data and regional
variability. θ S
from WOD 2001

9. Action: Float salinity are checked against the least
• Weighted least squares
fits are obtained from the
“good” float salinity data
and the statistical salinity
estimates. θ S
Action: Float salinity are checked against the least
squares fits for artificial drifts and offsets.
Significant drifts and offsets are adjusted.
Result: Delayed-mode salinity data
with error estimates.

10. What are the reference datasets?
Need datasets with:
• good quality data;
• good spatial coverage;
• recent in time.
All post-1985 CTD data from WOD 2001
World Ocean Database 2001 +
Additional recent CTD data

11. … depends on what region the float is in.
What is the confidence of delayed-mode salinity data?
… depends on what region the float is in.

12. Example in the tropical Pacific.

13. • Float WMO ID 3900059 salinity data have significant artificial drift
• Float WMO ID salinity data have significant artificial drift. Drift is removed in delayed-mode by weighted least squares fit to statistical salinity estimates from reference data.
Before delayed-mode adjustment
After delayed-mode adjustment
Profile no

14. Time series along θ = 5.5°C (~800 dbar)
• Artificial salinity drift is 0.25 PSS-78 in one year. Delayed-mode adjustment uncertainty is less than ±0.006 PSS-78.
Time series along θ = 5.5°C (~800 dbar)
drift = 0.25 S
uncertainty < ±0.006
Profile number

15. Example in the subarctic North Pacific.

16. Time series along θ = 2°C (~1700 dbar)
• Float WMO ID has artificial salinity drift of 0.05 PSS-78 in two years. Delayed-mode adjustment uncertainty is less than ±0.005 PSS-78.
Time series along θ = 2°C (~1700 dbar)
uncertainty < ±0.005 S
drift = 0.05
Profile number

17. Example near Japan.

18. Time series along θ = 2°C (~1900 dbar)
• Float WMO ID
was recovered. Deployment
CTD confirmed artificial
salinity offset of 0.02 PSS-78.
Delayed-mode adjustment
uncertainty is less than
±0.006 PSS-78. S
Profile number
For more Pacific examples, see Kobayashi et al.
“New Pacific historical dataset for the Argo standard scheme of delayed-mode quality control and its performance of salinity correction”

19. Modifications for the North Atlantic (Boehme/Send)
Complex water mass structure and basin bathymetry, rapid timescales of
Change require different approach to salinity mapping using historical data
Do not mix different water masses in vertical (even if Temp homogeneous) and account for Temp inversions 
Database not saved on isotherms
windowing in the vertical
Only specific isotherms are used in mapping
Avoid mixing different regimes (here boundary current and basin interior) in historical data selection 
Selection based on spatial distance D and fractional distance in potential vorticity f/H
Use the same distance weighting for covariance functions and mapping

20. In North Atlantic, deep salinities change interannually
In North Atlantic, deep salinities change interannually. Thus a very recent CTD cast must get much larger weight than 1 year old one 
Determine timescales from appropriate data, e.g. moorings (here 250 days)
Can also obtain better (and f/H weighted) spatial scales for salinity mapping
years
Decision process:
start with hypothesis that float is good and correct only if ΔS > 2 ×Error
Assume a float has either constant offset or linear drift
Apply linear fit ΔS=a+bt
Calculate uncertainty of a and b estimates and correct only if a or b larger than 2× their uncertainty

21. In the South Indian Ocean
• Looking at salinity anomaly at different depths can help discern artificial drift;
• King, “The detection of subsurface θ-S changes”

22. Salinity anomaly on Theta levels

23. In the Southern Ocean
• Isolating water masses that are common across the Polar Front;
• McCreadie et al, “Delayed-mode QC of Argo float salinity in the Southern Ocean”

24. There is more than one salinity value for the same q level
There is more than one salinity value for the same q level. Temperatures warmer than the sub-surface maximum can be associated with a wide range of salinities which interferes with the calibration.
Float

25. Conclusion
• With improving float technology, float sensors can be expected to remain stable for the lifetime of the float.
• With improving reference datasets and statistical methods, Argo delayed-mode salinity data can achieve accuracy of 0.01 PSS-78 or better in most oceanic regions.

26. Where to find Argo delayed-mode salinity data?
A novel feature of the Argo data format …
PSAL PSAL_QC
PSAL_ADJUSTED PSAL_ADJUSTED_ERROR PSAL_ADJUSTED_QC
Real-time data
Delayed-mode data

27. Argo Global Data Assembly Centres
Coriolis at IFREMER
USGODAE at Monterey