New global Mean Dynamic Topography from a GOCE geoid model, altimeter measurements and oceanographic in-situ data

Transcription

1 New global Mean Dynamic Topography from a GOCE geoid model, altimeter measurements and oceanographic in-situ data MH Rio, S. Mulet -1 -

2 INTRODUCTION The Mean Dynamic Topography (MDT) is a key reference surface for the optimal exploitation of altimeter data. It is the missing component that allows to estimate the ocean absolute dynamic topography (ADT) and the corresponding absolute geostrophic surface currents from the altimeter Sea Level Anomalies (SLA): ADT=MDT+SLA It may be written as the difference between an altimeter Mean Sea Surface (MSS=mean sea level above a reference ellispoid) and a geoid height relative to the same reference ellipsoid. MDT=MSS-Geoid However, due to the spectral differences of both surfaces (the MSS is known at few kilometer accuracy; present satellite-only geoid models resolve the with centimetric accuracy geoid scales of km (GRACE) to 100km (GOCE), spatial filtering is needed. MDT information at shortest scales may be brought by combination to oceanographic in-situ information as ARGO floats and drifting buoys velocities (Rio et al, 2004; 2005;2007;2011) - 2 -

3 INTRODUCTION Rappel de la méthodologie Direct Method MDT=MSS-Geoid filtering Synthetic Method The short scales of the MDT (and corresponding geostrophic currents) are estimated by combining altimetric anomalies and in-situ data Large scale MDT=First guess Multivariate Objective Analysis High resolution MDT - 3 -

4 MSS used for first guess computation Geoid model used for First Guess computation: CMDT CNES-CLS09 MSS CLS01 EIGEN-GRGS.RL02.MEAN based on 4 1/2 years of GRACE data CMDT CNES-CLS13 MSS CLS11 EGM-DIR-R4 based on 7 years of GRACE data and 2 years of reprocessed GOCE data Filtering used for First Guess computation: Buoy velocities dataset Optimal filter (~400 km) 15m drogued SVP drifters Period Optimal filter ~125km SD-DAC SVP drifters, with or without the drogue - Period Corrected for Wind slippage in case of drogue loss Argo floats surface velocities KEOPS SVP drifters are also included Ekman model T/S data Parameters fitted over the Parameters fitted over the period, period, by latitude, year, and month (3 by longitude, latitude and month (3 months months moving window) moving window) Computation of an Ekman model at 0m and at 15m depth CTD, ARGO Pref variable CTD (Cora3.4), ARGO Pref variable 200/400/900/1200/1900 Period /400/900/1200/1900 Period Resolution Global, ¼ (no Méditerranean) Global 1/4 (including the Mediterranean Sea)

5 Computation of the MDT first guess MDT=MSS CNES-CLS11 EGM-DIR-R4 RAW DIFFERENCE cm - 5 -

6 Computation of the MDT first guess MDT=MSS CNES-CLS11 EGM-DIR-R4 OPTIMALLY FILTERED cm - 6 -

7 The first guess spatial scales 100 km 200 km 150 km optimal - 7 -

8 Computation of the MDT first guess MDT=MSS CNES-CLS11 EGM-DIR-R4 OPTIMALLY FILTERED cm - 8 -

9 Computation of the MDT first guess First guess used for the CNES-CLS09 MDT computation (GRACE data) cm - 9 -

10 Use of in-situ oceanic measurements to improve the MDT at scales < ~125 km (u a,v a ) (u,v) η =h h (u,v) h geoid At each position r and time t for which an oceanographic in-situ measurement is available: dynamic height h (r,t) or surface velocity u(r,t),v(r,t) - the altimetric height/velocity anomaly is interpolated to the position/date of the in-situ data. - the in-situ data is processed to match the physical content of the altimetric measurement. h - the altimetric anomaly is subtracted from the in-situ height/velocity = u93 99 = uinsitu u' 93 v = vinsitu v' hinsitu h'

11 Synthetic mean heights from the CORA3.4 T/S profiles Dataset used for the CNES-CLS09 MDT computation cm

12 Drifting buoy data processing: U = U + U + U + U + U + buoy geost ekman tides inertial stokes U ageost _ hf Modelization of Ekman currents 3 days low pass filtering

13 Ekman theory Drifting buoy data processing: Modelling Ekman currents 45 π z D u = e * τ* e π 2 ± e ρfd e π z D v e * τ* π 2 e = e ρfd e Model Rio and Hernandez, 2003 u buoy u alti cos( π + π z) 4 D e sin( π π + z) 4 D e β θ u e = βτe Band pass filtered 30 hours 20 days iθ Wind stress from ERA INTERIM β and θ are estimated through least square fit Dataset used for the CNES-CLS09 MDT computation: SVP Drifting buoys flagged as DROGUED by the SD-DAC for the period

14 Drifting buoy data processing: Modelling Ekman currents β and θ computed over the global ocean by year Strong dependency of β and θ parameters with time Increase with time of parameter β Decrease with time of θ Direction of Ekman currents closer to wind direction Rio et al, 2011 This was due to a failure in the SVP buoy drogue loss detection system: Undetected undrogued drifter, directly advected by the wind in addition to surface currents, pollute the dataset Grodsky et al, 2011; Rio et al, 2012, Lumpkin et al, 2012

15 Drifting buoy data processing: Modelling Ekman currents at 15m depth β and θ computed over the global ocean by year Drogued drifter only (SD-DAC update) Strong dependency of β and θ parameters with time Increase with time of parameter β Decrease with time of θ Direction of Ekman currents closer to wind direction Rio et al, 2011 This was due to a failure in the SVP buoy drogue loss detection system: Undetected undrogued drifter, directly advected by the wind in addition to surface currents, pollute the dataset Grodsky et al, 2011; Rio et al, 2012, Lumpkin et al, 2012

16 Drifting buoy data processing: Modelling Ekman currents at 15m depth JANUARY JUNE β β θ θ

17 Drifting buoy dataset: Number of drogued versus undrogued data Content of the updated SD-DAC SVP drifter dataset Num buoy velocities Drogue ATTACHED Num buoy velocities Drogue LOST Number of velocity measurements STRONG interest of using undrogued buoy velocities to improve data coverage However, undrogued buoys are advected by surface Ekman currents, not 15m Ekman currents! the direct action of wind (wind slippage) These 2 effects must be modeled and removed

18 Drifting buoy data processing: Modelling Ekman currents at the surface Use of the surface velocities derived from the ARGO floats (YOMAHA database covering the period Number of ARGO buoy velocities

19 Number of SVP buoy velocities Drogue ATTACHED Number of SVP buoy velocities Drogue LOST Number of ARGO buoy velocities

20 Drifting buoy data processing: Modelling Ekman currents at the surface DROGUED SVP drifters OLD DROGUED SVP drifters UPDATED

21 Drifting buoy data processing: Modelling Ekman currents at the surface UNDROGUED SVP drifters DROGUED SVP drifters OLD DROGUED SVP drifters UPDATED

22 Drifting buoy data processing: Modelling Ekman currents at the surface UNDROGUED SVP drifters ARGO floats DROGUED SVP drifters OLD DROGUED SVP drifters UPDATED

23 Drifting buoy data processing: Modelling Ekman currents at the surface 15m JANUARY surface β β θ θ

24 Drifting buoy data processing: Modelling Ekman currents at the surface 15m JUNE surface β β θ θ

25 Drifting buoy data processing: Wind slippage correction Method from Rio et al, 2012, JAOT Correlation between the drifter velocity and the wind Drog lost Total drifter velocity vs wind Geostrophic drifter velocity vs wind Geostrophic drifter velocity minus α Wind vs Wind α (%)

26 Comparison between drifting buoy velocities and altimeter velocities (derived from SLA+First Guess MDT) Drogued SVP buoys Corrected for 15m Ekman currents Undrogued SVP buoys Corrected for 15m Ekman currents Mean zonal differences Undrogued SVP buoys Corrected for surface Ekman currents Root Mean Square zonal differences cm/s cm/s

27 Comparison between drifting buoy velocities and altimeter velocities (derived from SLA+First Guess MDT) Drogued SVP buoys Corrected for 15m Ekman currents Undrogued SVP buoys Corrected for 15m Ekman currents Mean zonal differences Undrogued SVP buoys Corrected for surface Ekman currents and wind slippage Root Mean Square zonal differences cm/s cm/s

28 Comparison between drifting buoy velocities and altimeter velocities (derived from SLA+First Guess MDT) Drogued SVP buoys Corrected for 15m Ekman currents Mean zonal differences Undrogued SVP buoys Corrected for surface Ekman currents and wind slippage Root Mean Square zonal differences cm/s cm/s

29 Comparison between drifting buoy velocities and altimeter velocities (derived from SLA+First Guess MDT) Drogued SVP buoys Corrected for 15m Ekman currents Argo floats Corrected for surface Ekman currents Mean zonal differences Undrogued SVP buoys Corrected for surface Ekman currents and wind slippage Root Mean Square zonal differences cm/s cm/s

30 Calculation of synthetic mean velocities From drogued SVP drifters only From undrogued SVP drifters only cm/s

31 Consistency check of using only drogued SVP drifters ( DROGUE ATTACHED ) versus only undrogued SVP drifters ( DROG LOST ) versus both drogued+undrogued SVP drifters ( DROGUE ALL ) COMPARISON TO INDEPENDENT SURFACE VELOCITIES RMS differences between the ARGO floats surface velocities and altimeter derived velocities (expressed in % of Argo floats velocity variance) MDT CNES-CLS13 DROGUE ALL MDT DROGUE ATTACHED RMS U 44.6 < 45.2 ~ 45.0 RMS V 52.4 < 53.4 ~ 53.2 MDT DROGUE LOST MDT DROGUE ATTACHED MDT DROGUE LOST MDT differences 1cm RMS cm

32 Final set of synthetic mean velocities: SVP-DROGUE ON + SVP-DROGUE OFF+ARGO U V ErrU ErrV cm/s cm/s

33 The GOCE only MDT (First Guess)

34 The CNES-CLS13 MDT

35 The GOCE only MDT (First Guess)

36 The CNES-CLS13 MDT

37 The CNES-CLS13 MDT Errors cm cm/s

38 VALIDATION: Comparison to other existing MDT solutions CNES-CLS09-CNES-CLS13 MDT Mean difference=0.2cm RMS difference =3.5cm

39 VALIDATION: Comparison to other existing MDT solutions Ebauche-CNES-CLS13 Mean=-0.01 RMS=1.9 MAXIMENKO-CNES-CLS13 Mean=44.2 RMS=5.1 DNSC08-CNES-CLS13 Mean=3 RMS=4.4 GLORYS2V1-CNES-CLS13 Mean=43.5 RMS=

40 VALIDATION: Comparison to other existing MDT solutions Comparison to independent velocities SVP velocities aquired in real time from September 2012 to September 2013 Processed as the DT SVP drifter velocities to remove wind slippage and Ekman currents cm/s

41 VALIDATION: Comparison to other existing MDT solutions Comparison to independent velocities SVP velocities aquired in real time from September 2012 to September 2013 Processed as the DT SVP drifter velocities to remove wind slippage and Ekman currents RMS differences between the SVP velocities aquired in real time from September 2012 to September 2013 and altimeter derived velocities (expressed in % of drifter velocity variance) velocities MDT- CNES- CLS13 MDT CNES- CLS13-V0 MDT CNES-CLS09 MDT GLORYS2V1 MDT MAX08 MDT GOCE (First Guess) RMS U RMS V

42 VALIDATION: Calculation of integrated quantities: The Meridional Overturning Circulation in the North Atlantic Altimetry : Field of absolute geostrophic surface currents weekly - 1/3 based on MDT+SLA Armor3D : 3D T/S fields weekly - 1/3 - [0-1500]m g z i u(z = zi) = u(z = 0) + ρf z = g z i v(z = z i ) = v(z = 0) ρf z = 0 0 ρ'(z)dz y ρ'(z)dz x Surcouf3D 3D geostrophic current fields weekly ( ) 1/3-24 levels from 0 to1500m

43 VALIDATION: Calculation of integrated quantities: The Meridional Overturning Circulation in the North Atlantic CNES-CLS09 Mean interior geostrophic transport calculated over Sv when using the CNES-CLS09 MDT -12 Sv with the CNES-CLS13 MDT. CNES-CLS13 The RAPID-MOCHA array (Kanzow et al., 2007) gives an independent estimate of -14 Sv thanks to moorings that monitor temperature, salinity and currents

44 VALIDATION: Expected impact on altimeter data assimilation in the Mercator-Ocean system Difference between the MDT currently used at Mercator- Ocean for SLA assimilation and the CNES-CLS13 MDT SLA innovation computed during the latest Mercator-Ocean reanalysis run (GLORYS2V3) Similarities between the two plots mean that the use of the CNES-CLS13 MDT will lead to improvements of the altimeter SLA assimilation into the Mercator-Ocean forecasting system

45 The Mediterranean Sea case The SOCIB-CLS MDT First Guess=MFS model mean Observations= drifters (Poulain et al, 2012) Result cm

46 CONCLUSIONS

47 - 47 -

48 PERSPECTIVES Improving the first guess Use of an improved geoid model The fifth and last GOCE release will be made available by ESA mid 2014 It will be based on data from the full duration of the mission (~3 years) including 6 months of data from a low orbit configuration, at the end of the mission we expect further improvement at scales down to 70-80km Use of an improved altimeter Mean Sea Surface Planned to be calculated in Improved filtering techniques Comparison of the optimal filter to other existing techniques (anisotropic diffusive filter (Bingham et al, 2010); Edge Enhancing Diffusion (Esanchez-Reales et al); singular spectrum analysis (SSA)-based filter (Ghil et al., 2002; Vianna et al, 2010)

49 PERSPECTIVES Improving the synthetic dynamic height estimation Removing the barotropic variability component of the altimeter SLA. Use of GRACE data? Adding the missing mean circulation at depth. Use of a mean surface from Argo displacement as (Willis et al), or (Ollitraut et al)?

50 PERSPECTIVES Improving the analyse objective parameters Calculating the variance and correlation scales from the observations <h,h>=f 1 (x 1 0,y 1 0) <u,u>=f 2 (x 2 0,y 2 0) x 1 0=x 2 0=x 3 0??? <u,v>=f 3 (x 3 0,y 3 0)

51 PERSPECTIVES Getting toward higher resolution? Positive impact expected from the use of the 5th GOCE geoid model Test a 2-step procedure: starting from the ¼ MDT as new first guess and apply the objective analysis on 1/8 mean heights and velocities Need for supplementary high resolution in-situ data Radar HF SAR derived velocities To come next: the «MDT options» kick-off

52 Ekman theory π z D u = e * τ* e Calculating the Ekman depth Ekman depth D E = Depth of frictional influence = Depth at which the direction becomes opposite to that at the surface and the speed falls to exp(-π)=4% of that at the surface. π 2 ± e ρfd e π z D v e * τ* π 2 e = e ρfd e β cos( π + π z) 4 D e sin( π π + z) 4 D e θ D E = π β (z= 0) 2 ρ f

53 Calculating the Ekman depth January June October m Ekman depth versus Mixed Layer Depth? The Ekman depth may differ from the mixed layer depth whose development depends on various factors as turbulence input (wind velocity), heat loss or gain (also salt), the stability of the underlying water and pre-existing stratification

54 Calculating the Ekman depth January June October m Mixed Layer Depth (temperature criteria) from the GLORYS2V1 Mercator system reanalysis

55 A case from the Keops-2 campaign Stirring patterns and Chl plume In Spring (October): Courtesy: F. D Ovidio - Geostrophic (altimetry) velocities are very consistent with the hypothesis that blooming waters are water coming from the plateau. - Adding Ekman velocities leads to largely overstimate the plume extension. Geostrophic velocities Geostrophic+Ekman velocities

56 A case from the Keops-2 campaign Stirring patterns and Chl plume Courtesy: F. D Ovidio In summer (January): - Geostrophic (altimetry) velocities underestimate the plume extension! - Adding Ekman velocities agrees with Chl maps. Geostrophic velocities Geostrophic+Ekman velocities

57 Winter/spring A case from the Keops-2 campaign Stirring patterns and Chl plume Courtesy: F. D Ovidio Late-spring/summer time Ekman layer Mixed layer Possible scenario: 1. There is an Ekman layer with a NE strong component. 2. Winter/spring time: the mixed layer is deeper than the Ekman layer: tracers diluted in the mixed layer have a geostrophic circulation. 3. late-spring/summer time: the mixed layer approaches the Ekman layer, and tracers also are structured by geostrophy+ekman October (spring) January (summer) D E D E MLD MLD m