Interest of assimilating future Sea Surface Salinity measurements from SMOS and Aquarius missions in an operational ocean forecasting system

1
Interest of assimilating future Sea Surface
Salinity measurements from SMOS and Aquarius
missions in an operational ocean forecasting
system
Benoît Tranchant CERFACS/Mercator Ocean, Toulouse
(France)
Charles-Emmanuel Testut, Lionel Renault, Nicolas
Ferry
http//www.mercator.eu.org
2
What can be expected from SSS data products for
DA in an operational context?

• Improving error statistics
• of SSS
• of SST/SSS relationships
• Improving observation operators with different
  SSS products
• ? Simplification of observation operator
  Resolution in space and time corresponding to the
  scales resolved by the model (i.e. model
  dependent)
• Need for stronger links between data segment
  operations and data assimilation developments
• Impact studies OSE, OSSE

3
OSSE Overview

• The main objectives were to
• Understand the most efficient way to assimilate
  SSS satellite data in order to extract the best
  reliable information in the context of the
  Mercator Ocean forecasting system
• Evaluate the potential impact of two different
  observing systems
• Know the level of the observation error from
  which associated SSS data have a significant
  influence on the data assimilation system SAM2.
• Method
• Performing Observing System Simulation
  Experiments (OSSEs) with simulated AQUARIUS and
  SMOS SSS data over 1 year (2003).
• Sensitivity studies to level products L2/L3
• Sensitivity studies to observation errors
• Accuracy of SMOS level-2 products? Observation
  errors specification ?
• Sensitivity studies to observing systems
• Relative skills of SMOS and AQUARIUS products?
  Incremental benefit of their combination?

4
Main characteristics of SMOS and Aquarius
missions
Science Satellite Mission SMOS (ESA) Aquarius (NASA and the Space Agency of Argentina )
Scientific Objectives Observation of Soil moisture and SSS Observation of SSS
Measurements goals - Accuracy of 0.5-1.5 PSU for a single observation - Accuracy of 0.1 PSU for a 10-30 days average and for an open ocean area of 200 km x 200 km - Global monthly 150-kilometer resolution SSS maps with an accuracy of 0.2 PSU
Temporal and spatial resolutions Global coverage every 3 days and 45 km resolution Global coverage every 7 days and 150 km resolution
5
Simulated SSS L2 products Instantaneous SSS at
pixel scale

Map of the difference (retrieved reference SSS)
for 10th of January

Algorithm to characterize the L2 SSS error, see
Boone et al., 2005 or Obligis et al., 2008.
Reconstructed SMOS L2 SSS error for January 2003
6
Characteristics of simulated SSS data
Simulated SSS products computed by CLS Level (spatial and time resolution) Observation error range (RMS in PSU)
SMOS L3P Level 3 SMOS (map of 200kmx200km, 10 days) 0,02 - 0,5
SMOS L2P Level 2 SMOS (40kmx40km along tracks , daily) pixel scale 0,2 - 2,5
Aquarius L2P Level 2 Aquarius (100kmx100km along tracks, 1 daily) pixel scale 0,1 - 1,5
Aquarius L2
SMOS L2
SMOS L3
7
OSSE ingredients

• REFERENCE or CONTROL RUN
• Hindcast experiment with in-situ, SST and
  altimeter data assimilation over 2003.
• OPA model MNATL(1/3) covering North Atlantic
  from 20S to 70N.
• ECMWF daily forcing fluxes
• DATA ASSIMILATION SCHEME SAM2
• Based on a SEEK filter Reduced Order Kalman
  Filter (modal space)
• 3D multivariate background error covariances 140
  seasonal 3D modes (?,T,S) calculated from an
  hindcast experiment (7 years)
• Innovation vector FGAT method (SLA and in situ
  data), observation operator adapted for largest
  scales (SST and SSS)
• TRUTH
• The native sea surface salinity (SSS) located on
  the SMOS L2 data points
• The native SSS comes from the North Atlantic and
  Mediterranean high resolution (1/15) MERCATOR
  OCEAN prototype named PSY2V1 re-sampled at a 1/3
  (univariate assimilation of SLA, with a
  relaxation term to SST and SSS ).

8
1.Sensitivity to level products
Variance of difference (PSU2)
Mean of difference (PSU)
Spatial average of the mean in psu (left) and
variance in psu2 (right) of difference between
three different estimates control run or REF
(red dashed line), SMOS L3 (black solid line),
SMOS L2 (blue solid line) and truth every ten
days during the year 2003 for the overall domain

1. The assimilation of Level 2 SMOS products seems
   to be a better approach than the assimilation of
   Level 3 SMOS products.
2. The SSS constraint from SMOS L3 has a positive
   impact. Indeed, in comparison to the REF
   experiment, this simulation has both slightly
   reduced the bias and the variance of the
   difference with the truth

9
2.Sensitivity to observation errors
Main limitations of results The threshold found
in this study is only valid for these sets of
operational data using these sets of observation
errors into the MERCATOR ocean forecasting system
(1/3).
SMOS L2_1 1xerror
SMOS L2_2 2xerror
Variance of difference for SMOS L2_x Over 2003
Variance ofdifference (PSU2)
SMOS L2_0.5 0.5xerror
10 days unit

1. The initial observation errors associated with
   the SMOS L2 products given by CLS (Boone et al.,
   2005) are satisfactory
2. This level of observation error specification
   defines a threshold (minimum requirement) to have
   a significant impact on the MERCATOR operational
   forecasting system. It allows to reduce the
   difference from about 0.5 to 0.3 PSU rms.

10
3.Sensitivity to observing systems

• Three possible explanations
• The daily data coverage is very different between
  these two products
• ?stronger constraint of SSS SMOS L2 compared to
  the AQUARIUS L2 Products.
• The observation error associated to Aquarius L2P
  is effectively lesser than that of SMOS L2P but
  not for the same surface.
• The spatial resolution the Aquarius L2P are only
  able to constraint the scale associated to the
  Aquarius grid.
• ? one part of the signal associated to small
  scales (lt 100 km) is not taken into account.

Variance ofdifference (PSU2)
The combination of the two L2 Products has a
weak impact in comparison to the SMOS L2
simulation
11
3.Sensitivity to observing systems
Focus on the data coverage

• Each ocean grid point is observed by
• Aquarius measurements every 7 days
• SMOS measurements every 3 days

The decorrelation scales (in days) corresponding
to a time correlation of 0.4 from a re-analysis
at 1/3 is generally less than 4 days
(atmospheric exchanges) in the North Atlantic.
CONCLUSION A full coverage in the North Atlantic
at a sufficient time frequency (4-5 days) is
usefull to the SSS assimilation problem in a
eddy-permitting model (spatial resolution lt1/3)
Temporal decorrelation scales from a re-analysis
(MNATL 1/3) over 11 years (Greiner et al., 2004)
day
12
To sum up
What is the real gain () of assimilating
remotely sensed SSS data (comparison to the REF
experiment) in term of RMSE
REF SMOS L3 SMOS L2 SMOS L2_0.5 SMOS L2_2. AQUARIUS AQUARIUS SMOS L2
-- - 20 - 36 - 41 - 20 - 10 - 37
RMSE The root mean square of error/difference
(RMSE) between assimilation experiments and
truth averaged overall the domain.
13
Error balance

• This figure shows the time evolution of the mean
  and the RMS of the SSS increment for all
• experiments.
• The mean of the increment is quite close to zero
  for all the experiments.
• The RMS of the SSS increment has the same
  behaviour/amplitude in REF, in SMOS L3 and SMOS
  L2 . ? constraint coming from the assimilated SSS
  (SMOS L3 and SMOS L2) is relatively relevant to
  improve the SSS increment pattern.
• These results show
• SSS observation error variance and particularly
  its ratio with regard to the error of the other
  assimilated data sets seems relatively consistent
  there are a compromise between SSS, SLA and SST
  increments
• Our scheme takes into account the new SSS
  constraint coming from another data source even
  if it is far from other operational data, ?
  Contributions of new SSS data to the SSS
  increment have not disrupted the existing
  equilibrium between all errors.

14
Conclusions

• What is the best strategy to optimally use the
  future SMOS and Aquarius data in the context of
  ocean prediction systems, from the perspective of
  monitoring the mesoscale ocean circulation?
• The use of the synthetic SMOS L2 product gives
  satisfactory improvement in the model results,
  since it provides a measurable impact of the
  quality of ocean analyses from operational
  systems.
• The SSS observation error variance as specified
  by Boone et al., (2005) and particularly its
  ratio with regard to the error of the other data
  sets assimilated seems appropriate.
• The impact of the Aquarius L2 Products is weak
  compared to the SMOS L2 Products. The combination
  of the two L2 Products had thus a small effect on
  final results
• BUT
• Simulated SSS data comes from SSS field
  relatively far from the other assimilated data
  (operational data).
• The assimilation system does not correct any
  fluxes, in particular the E-P fluxes ?
  Underestimate information coming from SSS.

15
For more informations

• Tranchant, et al. (2008), Expected impact of the
  future SMOS and Aquarius Ocean surface salinity
  missions in the Mercator Ocean operational
  systems New perspectives to monitor ocean
  circulation, Remote Sensing of Environment, 112,
  pp 1476-1487.
• Obligis et al.. (2008) Benefits of the future Sea
  Surface Salinity measurements from SMOS.
  generation and characteristics of SMOS
  geophysical products, IEEE Trans. Geoscience and
  Remote Sensing, vol. 46, issue 3, 746-753. 
• Tranchant et al.(2008), Data assimilation of
  simulated SSS SMOS products in an ocean
  forecasting system, Journal of operational
  Oceanography, Vol. 2008, No 2, August 2008., pp
  19-27(9).