Data assimilation and Data fusion
in a regional simulation
Nina Hoareau1 Marta Umbert1 Antonio Turiel1
Joaquim Ballabrera2 Marcos Portabella2
1: Institut de Ciències del Mar, Barcelona (SPAIN) 2: Unitat de Tecnologia Marina, Barcelona (SPAIN)
2
PLAN
• Introduction
•What is Data fusion?
•What is Data assimilation?
• Results and Validation
• Conclusion
Importance of SSS
• Water cycle intensity (Hosoda et al., 2009)
• Deep convection (Dickson et al., 2002)
• Ocean Ventilation (CLIVAR SWG, 2007)
• ENSO prediction (Ballabrera-Poy et al., 2003)
• Surface currents (Acero-Schertzer et al., 1997)
• Carbon cycle (Lefèvre et al., 2009)
• Heat storage (Ganachaud et al., 2003)
• Barrier Layer (Lukas et al., 1991)
• Surface Jet (Roemmich et al., 1994)
• Data Assimilation (Cooper, 1988)
Essential for Climate knowledge
MOORING ARRAYS
Measure of SSS
MOORING ARRAYS
Measure of SSS SSS and Soil moisture, Nov 2011
Level 1 Level 0 Level 2 Level 3 Level 4
Data Assimilation
and Data Fusion
Measurements Observations Global maps Global maps Raw data
Data processing chain
The SMOS mission
6
Data Fusion Blending method
7
• Singularity exponents can be accurately estimated from a single image using a novel new image processing technique.
• Singularity exponents are related to the flow dynamics: they trace streamlines
MW SST
SMOS mission products L4: Data Fusion
• Singularity exponents can be accurately estimated from a single image using a novel new image processing technique.
• Singularity exponents are related to the flow dynamics: they trace streamlines
MW SST
SMOS mission products L4: Data Fusion
• Singularity exponents can be accurately estimated from a single image using a novel new image processing technique.
• Singularity exponents are related to the flow dynamics: they trace streamlines
MW SST
SMOS mission products L4: Data Fusion
• The analysis of the output of different numerical simulations confirm that SST and SSS have the same exponents (synergic effect)
Assumption: singularity exponents are the same for any scalar.
• Singularity exponents can be accurately estimated from a single image using a novel new image processing technique.
• Singularity exponents are related to the flow dynamics: they trace streamlines
MW SST
SMOS mission products L4: Data Fusion
This implies that, locally, there is a linear relation between SSS and SST gradients:
• The analysis of the output of different numerical simulations confirm that SST and SSS have the same exponents (synergic effect)
Assumption: singularity exponents are the same for any scalar.
SSS SST
• Preliminary study shows improvement;
• SST images can hence be used to improve spatial correlations of SSS maps.
SMOS mission products L4: Data Fusion
SSS SST SSS
• Preliminary study shows improvement;
• SST images can hence be used to improve spatial correlations of SSS maps.
SMOS mission products L4: Data Fusion
Data Assimilation in NEMO-OPA
14
15
After Launch of SMOS satellite on November 2009 …
-> Presence of RFI (Radio Frequency Interferences) at L-band (1.4 GHz)
deteriorates the satellite data retrievals. More important than the presence of the
continent coastline.
Why this region ? Low annual and interannual SSS variability => Help the
calibration activities of SMOS .
The Macaronesian Region
Mourre et al. (2008)
16
Physical domain
Boundaries 45ºW-5ºW, 15ºN-44ºN
Grid size 128 x 100 x 31
Partial step Z-coordinates
Spatial resolution 1/3º (33 km at the
equator)
Time step 1800s (48 time steps/day)
The Tarfaya Simulation (Mourre et al, 2008)
Atmospheric Forcing (NCEP-NCAR)
Daily Wind stress, 10m Wind
speed, 2m Air temperature
Monthly Precipitation rate, Cloud
cover and Humidity
boundary conditions
Open boundary Seasonal data
(MERCATOR) Parameterization
Horizontal turbulent
diffusivity Laplacian, 300 m2 s-1
Horizontal turbulent
viscosity Bilaplacian, -1.2 10-11 m4
s-1
Deep vertical diffusion Laplacian, 1 10-6 m2 s-1
Surface vertical diffusion Laplacian, 1 10-4 m2 s-1
Vertical turbulent mixing TKE model
No Relaxation for SST and
SSS
)( 0 xxPhysicsdt
dx
Data assimilation Scheme
• Relaxation term is added into the equation of evolution of a prognostic variable (in our case, T and S).
• The nudging term tends to reduce exponentially the distance of the
model towards the observations.
Relaxation coefficient (s-1) Prognostic variable T, S, U, V.
Observations
) ( s -1-1 day210.5 6 ) ( s -1-1 month110.5 7
SST nudging coefficient: SSS nudging coefficient:
Data for 2011
• Maps of SST (1º x 1º) from Reynolds and
Smith (1994). • 7 days.
• Maps of SSS (1º x 1º) from SMOS-L3 • 10 days.
Validation: ARGO in-situ
19
December 2011
Assimilation Fusion
RMS: Experiments-ARGO
Conclusion • We have implemented two data algorithms to produce Level 4 SSS
maps from SMOS.
• Improvements by assimilating SMOS data have been found in regions far from coasts or RFI sources.
• Unrealistic results are due to the lack of data, or in regions were the data was not flagged as bad, but it had to be, providing information for the lower levels.
• This is an ongoing work and we plan to:
Data Assimilation: – Implement the EnKF – Use Singularity Exponents to validate the Data Assimilation outputs – Simultaneously assimilate additional data (T,S profiles, sea level, …)
Data Fusion: – Improve data fusion algorithms. – Use OSTIA SST (daily, 1/10º) as a template, which has a more realistic
singularity exponent maps.