Optimal Spectral Decomposition (OSD) for GTSPP Data Analysis

Transcription

1 Optimal Spectral Decomposition (OSD) for GTSPP Data Analysis Peter C Chu (1),Charles Sun (2), & Chenwu Fan (1) (1) Naval Postgraduate School, Monterey, CA pcchu@nps.edu, (2) NOAA/NODC, Silver Spring, MD Charles.Sun@noaa.gov GTSPP Meeting, Oostende, Belgium, 5-7 May 2010.

2 Observational Data

3 It is an urgent need to establish monthly varying (T, S, u, v) gridded dataset from GTSPP and Argo trajectory data Oceanographic and Climate Studies

4 Ocean Data Analysis Classical Method Fourier Series Expansion

5 Joseph Fourier Fourier was obsessed with the physics of heat and developed the Fourier series and transform to model heat-flow problems.

6 Fourier Series Expansion For a rectangular region (L x, L y ), the basis functions are sinusoidal functions. i x f ( x, y) aij sin sin L b i j x y cos i x cos j y ij i j Lx Ly j y L

7 For the Dirichret boundary condition : f = 0 at the boundaries i x f ( x, y) aij sin sin L j y L i j x y The dots represent the Observations.

8 Linear Algebraic Equations for the Coefficients a ij ob ob i x f ( x1, y1 ) aij sin sin L j y L ob ob 1 1 i j x y ob ob i x f ( x2, y2 ) aij sin sin L... j y L ob ob 2 2 i j x y ob i x f ( xm, ym ) aij sin sin L j y L ob ob ob M M i j x y

9 Determination of Spectral Coefficients (Ill-Posed Algebraic Equation)

10 Known a ij Analyzed Field i x f ( x, y) a sin sin L j y ij L i j x y

11 For the Neumann boundary condition at the boundaries n f 0 i x j y f ( x, y) bij cos cos L L i j x y The dots represent the Observations.

12 Linear Algebraic Equations for the Coefficients a ij ob ob i x f ( x1, y1 ) bij cos cos L j y L ob ob 1 1 i j x y ob ob i x f ( x2, y2 ) bij cos cos L... j y L ob ob 2 2 i j x y ob i x f ( xm, ym ) bij cos cos L j y L ob ob ob M M i j x y

13 Known b ij Analyzed Field i x f ( x, y) b cos cos L j y ij L i j x y

14 For General Ocean Basin Generalized Fourier Series Expansion

15 Spectral Representation Fourier Series Expansion m Basis functions (not sinusoidal) c any ocean variable

16 Determination of Basis Functions (1) Eigen Functions of the Laplace Operator (Data and Model Independent) (2) Empirical Orthogonal Functions (Data or Model Dependent)

17 Eigen Functions of Laplace Operator Basis Functions (Closed Basin) Ψ k Streamfunction Φ m T, S, Velocity Potential

18 Basis Functions (Open Boundaries)

19 Boundary Conditions

20 Spectral Decomposition M 0 m m m 1 T( x, t) T ( x) c ( t) ( x) M 0 m m m 1 S( x, t) S ( x) d ( t) ( x)

21 Benefits of Using OSD (1) Don t need first guess field (2) Don t need autocorrelation functions (3) Don t require high signal-to-noise ratio (4) Basis functions are pre-determined before the data analysis. They are independent on the data.

22 Optimal Mode Truncation

23 Vapnik (1983) Cost Function

24 Optimal Truncation Gulf of Mexico, Monterey Bay, Louisiana- Texas Shelf, North Atlantic Kopt = 40, Mopt = 30

25 Determination of Spectral Coefficients (Ill-Posed Algebraic Equation) This is caused by the features of the matrix A.

26 Rotation Method (Chu et al., 2004) Well-Posed The matrix S is determined by

27 Errors 0 48 l 1 Tˆ T ( x) T D ( x) T ( x) l l uˆ 24 0 n n n 1 u( x, t) C A k ( x) u( x) u ( x) T, u errors

28 Noise-to-Signal Ratio Error Estimation αβ, α β ( P) ( P) ( T, T T ') ~ 0.1

29 OSD Applications (a) Baroclinic Rossby Waves in the tropical North Atlantic at mid-depth (Argo Trajectory Data) (b) Synoptic Current Reversals on the Texas- Louisiana Continental Shelf (Surface Drifting Buoys) (c) Monterey Bay Surface Circulation (CODAR) (d) Synoptic (T, S) Fields for Pacific

30 (a) Baroclinic Rossby Waves in the tropical North Atlantic at middepth (ARGO)

31 References Chu, P.C., L.M. Ivanov, and O.M. Melnichenko, 2005: Fall-winter current reversals on the Texas- Lousiana continental shelf. Journal of Physical Oceanography, 35, Chu, P.C., L.M. Ivanov, O.M. Melnichenko, and N.C. Wells, 2007: On long baroclinic Rossby Waves in the tropical North Atlantic observed from profiling floats. Journal of Geophysical Research Oceans, 112, C05032, doi: /2006jc These papers can be downloaded from:

32 Tropical North Atlantic (4 o -24 o N) Important Transition Zone Meridional Overturning Circulation (MOC) (Rahmstorf 2006)

33 MOC Variation Heat Transport Variation Climate Change

34 Are mid-depth (~1000 m) ocean circulations steady? If not, what mechanisms cause the change? (Rossby wave propagation)

35

36 Argo Observations (Oct-Nov 2004) (a) Subsurface tracks (b) Float positions where (T,S) were measured

37 Circulations at 1000 m estimated from the original ARGO float tracks (bin method) April 2004 April 2005 It is difficult to use such noisy data into ocean numerical models.

38 Boundary Configuration Basis Functions for OSD

39 Basis Functions for Streamfunction Mode-1 and Mode-2

40 Circulations at 1000 m (March 04 to May 05) Bin Method OSD

41 Fourier Expansion Temporal Annual and Semi-anuual

42 Fourier Expansion Temporal Annual and Semi-anuual

43 Optimization

44 Annual Component

45 Semi-annual Component

46 Time Longitude Diagrams of Meridional Velocity Along 11 o N Annual Semi-Annual

47 Time Longitude Diagrams of temperature Along 11 o N Annual Semi-Annual Annual Semi-Annual 550 m 950 m

48 Characteristics of Annual Rossby Waves Western Basin Eastern Basin Western Basin Eastern Basin

49 (b) Synoptic Current Reversals on the Texas-Louisiana Continental Shelf (Surface Drifting Buoys)

50 Ocean Velocity Observation 31 near-surface (10-14 m) current meter moorings during LATEX from April 1992 to November 1994 Drifting buoys deployed at the first segment of the Surface Current and Lagrangian-drift Program (SCULP-I) from October 1993 to July 1994.

51 Moorings and Buoys

52 LTCS current reversal detected from SCULP-I drift trajectories.

53

54 LTCS current reversal detected from the reconstructed velocity data December 30, 1993 January 3, 1994 January 6, 1994

55 EOF Analysis of the Reconstructed Velocity Filed EOF Variance (%) 01/21/93-05/21/93 12/19/93-04/17/94 10/05/94-11/29/

56 Mean and First EOF Mode

57 Mean Circulation 1. First Period (01/21-05/21/93) 2. Second Period 12/19/93-04/17/94) 3. Third Period (10/05-11/29/94)

58 EOF1 1. First Period (01/21-05/21/93) 2. Second Period 12/19/93-04/17/94) 3. Third Period (10/05-11/29/94)

59 Calculated A1(t) Using Current Meter Mooring (solid) and SCULP-1 Drifters (dashed)

60 8 total reversals observed Uals ~ alongshore wind

61 Results Alongshore wind forcing is the major factor causing the synoptic current reversal. Other factors, such as the Mississippi- Atchafalaya River discharge and offshore eddies of Loop Current origin, may affect the reversal threshold, but can not cause the synoptic current reversal.

62 (c) Monterey Bay Surface Circulation (CODAR)

63 CODAR

64 Monterey Bay

65 30 cm/s 37.0 o N 30 cm/s 36.9 o N 36.8 o N 36.7 o N 36.6 o N o W o W o W o W o W o W o W o W o W o W Place for comments: left - radar derived currents for 17:00 UT December 1, 1999 right reconstructed velocity field.

66 Preliminary Comparison of OSD and OI Mean ( ) Visual comparison of the Optimal Spectral Decomposition (OSD)-derived SST (left) to the NOAA Optimally Interpolated (OI) SST (right)

67 (d) Temporal and spatial variability of Pacific Ocean

68 Monthly Temperature (10 m) in the Pacific Ocean since 1990 (analyzed from GTSPP)

69 Monthly Temperature (100 m) in the Pacific Ocean since 1990 (analyzed from GTSPP)

70 Monthly Temperature (500 m) in the Pacific Ocean since 1990 (analyzed from GTSPP)

71 Monthly Temperature (1000 m) in the Pacific Ocean since 1990

72 Conclusions OSD is a useful tool for processing realtime velocity data with short duration and limited-area sampling especially the GTSPP/Argo data. OSD has wide application in ocean data assimilation.

73 What is next? It is an urgent need to establish monthly varying (T, S, u, v) gridded dataset from GTSPP and Argo trajectory data Oceanographic and Climate Studies

74 Steps for GTSPP Data Analysis (1) Change the current data format one file for one profile into new data format one file for profiles in a month from January 1990 for individual ocean basin (Atlantic, Pacific, Indian oceans). (2) Reconstruct the profile data with the same month and year into grid points using the Optimal Spectral Decomposition (OSD) method.

75 4D Global Velocity Data Reference + Geostrophic (T, S)