1. 1st Argo Science Workshop
Design Requirements for an Argo Float Array in the Indian Ocean Inferred from Observing System Simulation Experiments
1st Argo Science Workshop
Tokyo
Andreas Schiller
Susan Wijffels
Gary Meyers
CSIRO Marine Research

2. Status of Argo Floats in Indian Ocean (September 2003):
(~170 floats by end of 2003, ~450 in 2005)
Courtesy Helen Phillips,
CSIRO Marine Research

3. Questions
Are typical horizontal velocities at parking depths small enough so
that Argo floats will be quasi-stationary over their anticipate
lifetime of about 3 years?
What are the required temporal and spatial sampling intervals for
capturing seasonal and longer-term variability?
Using Argo floats, can we simultaneously measure intraseasonal
and longer-term variability in the Indian Ocean?
Are there areas in the Indian Ocean that need to be more
frequently sampled than other areas?

4. Current Argo Mission Recording T & S as float rises
Surface
Recording T & S
as float rises
- 500 m m
Profiling from 2000m every 10 days, drifting at 2000m

5. Proposed Argo Mission
Surface
- 500 m
5 days m
Profiling from 500m every 5 days, profiling from 2000m every 20 days

6. Anomaly Composites SST Model Anomaly Composites SST (Reynolds obs.)
Schiller and Godfrey, Journal of Climate, 2003 °C

7. What are Typical Advection Velocities at Parking Depths?
Simulated Lagrangian Floats at z=2000m : Jan 82 – May 94
Max. Velocity:
< 1.2 cm/s
(< 32km/month)
Numbers indicate
distances covered
by floats over period
of 12 years,5 months

8. Simulated Lagrangian Floats at z=500m : Jan 82 – May 94
Max. Velocity:
< 3 cm/s
(< 76 km/month)

9. Spatial Sampling Temporal Sampling Complete Observations = 200km
= 3days
= 600km
= km
= 1200km
= km
= 6 days
= 9 days
= 21 days
Temporal
Sampling

10. Spatial Sampling Temporal Sampling Complete Observations = 200km
= 3days
= 600km
= km
= 1200km
= km
= 6 days
= 9 days
= 21 days
Temporal
Sampling

11. Spatial Sampling Noise added Temporal Sampling Complete Observations
= 200km
= km
= 3days
= 600km
= km
= 1200km
= km
= 6 days
= 9 days
= 21 days
Temporal
Sampling

12. Seasonal Variability of Temp. at 100m Depth (21-day sampling)
Variability Complete Obs.
RMS
Variability Argo (21 days)
Signal-to-Noise Ratio Argo

13. Intraseasonal Variability of Temperature at 100m Depth
Variability Complete Obs.
6-day Sampling day Sampling day Sampling
Variability
RMS
SNR

14. Variability of Temperature at 80ºE
Variability Complete Obs.
Variability Argo (6-day sampling)
Colour: Intraseasonal,
Contour lines:
Seasonal signal
RMS
Signal-to-Noise Ratio Argo

15. Equatorial Averages (5ºS- 5ºN) of Temporal* Sampling Strategies
Seasonal Intraseasonal ARGO ARGO
Variability Variability RMS Signal:Noise Ratio
Complete
Obs.
Argo
Complete
Obs.
Argo °C °C °C
* 0-500m: 6 days, m: 21 days. Full model resolution: = 200km
= km

16. = 600km = 1200km Spatial Sampling Temporal Sampling X A B C
Complete Observations
= 200km
= km
= 3 days
= 600km
= km
= 1200km
= km
= 6 days (0-500m)
[and 18 days
( m)] X
= 9 days (0-2000m)
= 15 days (0-2000m) A
~ 2500 floats B
~ 300 floats
Temporal
Sampling
End of 2003: ~ 170 floats C
~ 80 floats

17. Equatorial Averages of Temporal and Spatial* Sampling Strategies
Seasonal Intraseasonal ARGO ARGO
Variability Variability RMS Signal:Noise Ratio
dotted:
9 days
sampl.
Complete
Obs.
Argo
Complete
Obs.
Argo °C °C °C
* 0-500m: 6 days, m: 21 days. Model resolution: = 600km
= km

18. Basin Averages of Temporal and Spatial* Sampling Strategies
Seasonal Intraseasonal ARGO ARGO
Variability Variability RMS Signal:Noise Ratio
dotted:
9 days
sampl.
Complete
Obs.
Argo
Complete
Obs.
Argo °C °C °C
* 0-500m: 6 days, m: 21 days. Model resolution: = 600km
= km

19. Estimating α: Time Series at 0º N, 81º E (Reppin et al.,1999)
White Noise
Error (ºC)
Estimating α: Time Series at 0º N, 81º E (Reppin et al.,1999)
Model-Observation:

20. White Noise: Signal-to-Noise Ratios of Argo Temperature
Seasonal
Seasonal
Basin
Equator
Intraseasonal
Intraseasonal
Error Factor α
Error Factor α
0-500m: 6 days, m: 21 days. Model resolution: = 600km
= km

21. White Noise, Large-Scale Smoothing, Asynchronous Sampling:
Signal-to-Noise Ratio of Argo Temperature
Seasonal
Seasonal
Basin
Equator
Intraseasonal
Intraseasonal
Error Factor α
Error Factor α
0-500m: 6 days, m: 21 days. Model resolution: = 600km
= km

22. White Noise, Large-scale Smoothing, Asynchronous Sampling:
Signal-to-Noise Ratio of Argo Temperature
Seasonal
Seasonal
Basin
Equator
Intraseasonal
Intraseasonal
Error Factor α
Error Factor α
0-500m: 9 days, m: 21 days. Model resolution: = 600km
= km

23. Summary (1)
Velocities at Parking Depths (~30kms/month at 2000m, ~80kms/month at 500m)
small compared to propagation of intraseasonal signals
 assume stationary floats (less valid for seasonal and longer signals).
Simulated Variability of Temperature on Intraseasonal-to-Seasonal Time Scales
(Equator and Basin):
(a) Temporal and Spatial Sub-Sampling (both on “typical” scales):
temporal sampling causes RMS errors of order 0.1ºC for both
seasonal and intraseasonal time scales
spatial sampling significantly increases RMS errors for both seasonal
signal (≤ 0.9ºC) and intraseasonal signal (≤ 0.1ºC)
 Seasonal RMS error of mixed spatial/temporal sampling almost completely
determined by spatial RMS error (SNR ≥ 10).
However: Intraseasonal RMS error due to temporal sampling is as large as
intraseasonal RMS error due to spatial sampling (0.5 ≤ SNR ≤ 5).

24. Summary (2)
(b) with additional noise (within reasonable range of error estimates):
Seasonal SNR ≥ 1
Intraseasonal SNR: 0.1 ≤ SNR ≤ 1*
(c) with asynchronous profiling and large-scale smoothing:
Seasonal SNR increases by ≈ 10-30%
Intraseasonal SNR almost unchanged (≤ 10%)
Caveat:
Results are based on single OGCM. Model deficits exist due to:
Limited horizontal and vertical resolution (non-eddy resolving)
Errors in forcing fields
Errors in model parameterizations/physics (e.g., lack of : internal wave
energy, tidal, inertial and other high frequency noise).
* simulated intraseasonal signals are only half the size of observed amplitudes
Observed intraseasonal SNRs are likely to be larger than simulated

25. Recommendations:
To capture variability on intraseasonal-to-seasonal scales in the Indian
Ocean with Argo Floats the results suggest that
spatial sampling of the ocean is of crucial importance everywhere irrespective of dynamical regime  ARGO floats should be deployed such that they resolve spatial scales of interest. A minimum requirement for resolving ISO in the ocean is given by the spatial scales of intraseasonal variability, i.e.
temporal sampling becomes particularly important in dynamically active areas such as the WBC and the equatorial domain of the ocean. High frequency sampling is required to maintain an acceptable SNR on intraseasonal timescales (strongly depends on “noise” level)
 to capture intraseasonal variability in the Indian Ocean the
minimum sampling interval in the upper ocean should be
5 days or less.

26. The End

27. Supplementary Material

28. Composites of Observation-Based Intraseasonal Wind Stress (FSU/NCEP) and Precipitation (CDIAC MSU)Anomalies
dynes/cm^2
mm/day

29. Impact of Sampling Error on Ocean Mixed-Layer Heat Budget:

30. Spatial resolution Δx, Δy (averages) Temporal resolution Δt (averages)
Noise
added
(1)
Complete Obs.
200km,
50-100km
3 days No
(2)
Exp. 6D
6 days
(3)
Exp. 9D
9 days
(5)
Exp. 21D
21 days
(6)
Exp. 3DX
600km, km
(7)
Exp. 3DXX
1200km, km
(8)
Exp. 6DX
(9)
Exp. 21DX
(10)
Exp. 3DN
0-3x
WOA
error
(11)
Exp. 6DXN
0-3x WOA
(12)
Exp. 21DXN

31. Temperature Time Series at 0ºN, 81º E, z = 160m
Complete data
Black: model data
Red: Observations from Reppin et al., 1999
Sampling interval
3-days
Filtering
Seasonal (t 93 days) Intraseasonal (t < 93 days)
Sampling interval:
6-days

32. Spectra of Pot. Temperature at 0ºN, 80º 30’E, z=160m
Obs.: Reppin et al., 1999 (red) and model (black)
870

33. Observed u,v (Reppin et al
Observed u,v (Reppin et al., 1999; red) and Model (black) at 0ºN, 80º 30’E, z=25m

34. u,v Spectra at 0ºN, 80º 30’E (red: Obs. , Reppin et al
u,v Spectra at 0ºN, 80º 30’E (red: Obs., Reppin et al., 1999; black: Model)

35. What is the Required Temporal Sampling Interval for Capturing Seasonal-to-Interannual and Intraseasonal Variability?
Assumption:
Spatially stationary ARGO floats are available at each horizontal model grid point
Approach:
Sample model output (complete temperature available as 3-day
mean output) at different sampling intervals, e.g.
21 days ( m)
6 days (0-500m)
9 days (0-2000m)
Separate signal into longer (seasonal-to-interannual)
and shorter (intraseasonal) components
Calculate standard deviation of full data and sub-sampled data
Calculate RMS differences between full and sub-sampled data
Estimate signal-to-noise ratios (Standard deviation : RMS difference)

36. Cross Spectral Analysis of Temperature Time Series
at 0ºN, 81º E, z = 160m
Intraseasonal | seasonal SNRs: 0.7| | |1.0
Black: complete model
Red: sub-sampled (6days, Δx=600km, Δy=150km) with increasing noise levels
Error factor: α = α = α = 2.0

37. intraseasonal (colour), seasonal (isolines)
Variability at 8° S:
intraseasonal (colour), seasonal (isolines)
Java
equatorial
wave guide °C
C.I. = 0.5° C
Note
different
scales
9days
(shown):
S:N = 2:1
6 days:
S:N = 5:1
C.I. = 0.02° C

38. Variability at 4° N: 9days (shown): S:N = 2:1 6 days: S:N = 5:1
> 0.65°C
9days
(shown):
S:N = 2:1
6 days:
S:N = 5:1
Note changes in RMS
amplitudes:
intraseasonal RMS
( ) =
seasonal RMS
( ) >
> 0.35°C
> 0.55°C

39. Variability at 90° E:
9days
(shown):
S:N = 2:1
6 days:
S:N = 5:1

40. Where do we have to take measurements for capturing seasonal-to-interannual and intraseasonal variability (spatial sampling)?
Assumption:
Spatially stationary ARGO floats are available at each time step
Approach (similar to temporal sampling):
Sample model output at different spatial sampling intervals
9 days (0-2000m)
6 days (0-500m) [and 18 days ( m)]
Separate signal into longer (seasonal-to-interannual)
and shorter (intraseasonal) components
Anisotropic smoothing of signals
Calculate variabilities of full data and spatially sub-sampled data
Calculate RMS differences between full and sub-sampled data
Estimate signal-to-noise ratios

41. Equatorial Averages (5ºS- 5ºN) of Spatial* Sampling Strategies
Seasonal Intraseasonal ARGO ARGO
Variability Variability RMS Signal:Noise Ratio
Complete
Obs.
Argo
Complete
Obs.
Argo °C °C °C
* Full temporal resolution (3 days). Model resolution: = 600km
= km

42. White Noise: Signal-to-Noise Ratios of Complete Observations
Error (ºC)
, observed:
Basin
Equator
Seasonal
Seasonal
Intraseasonal
Intraseasonal
Error Factor α
Error Factor α

43. White Noise, Large-Scale Smoothing, Asynchronous Sampling:
Signal-to-Noise Ratios of Complete Observations
Basin
Equator
Seasonal
Seasonal
Intraseasonal
Intraseasonal
Error Factor α
Error Factor α

44. White Noise and Signal-to-Noise Ratios: Argo Salinity
Seasonal
Seasonal
Basin
Equator
Intraseasonal
Intraseasonal
Error Factor α
Error Factor α
0-500m: 6 days, m: 21 days. Model resolution: = 600km
= km

45. Related Activities in Pacific Ocean:
Ed Harrison, Gabriel Vecchi [PMEL], Tony Lee [JPL]