Data assimilation in global ocean analysis and forecasting system, for Marine applications :
focus on the Tropical Atlantic
Marie Drévillon, Elisabeth Rémy, Eric Greiner, Charly Régnier, Jean-Michel Lellouche
and the Mercator Ocean team
Overview
➢ Short description of Mercator Ocean’s global (re)analyses system
➢Main strengths and limitations in the Tropical Oceans
➢… with some illustrations
➢More information on Copernicus Marine Service CMEMS, GODAE and OSEs to follow in tomorrow’s presentation
pace
Satellite observations (surface, repetitive) In-situ observations (description at depth, sparse)Models (3D, assimilating all observations)
The past (long data time series)The present (current oceanic conditions)The future (forecast)
Blue ocean (physics : currents, T and S …)White ocean (sea ice)Green ocean (chlorophyll, CO2, oxygen, pH, …)
Monitoring the Marine Environment is INTEGRATING:
Cliquez et modifiez le titre
Global (re)analysis system in short
Use of NEMO ORCA025 (1/4°) 75L + LIM
Forced with ERA-interim reanalysis (large scalecorrection of precipitations and radiative forcing) and climatological runoffs
Multivariate assimilation of SST, SLA, in situ T/S profiles (and monovariate for Sea Ice)
3D T and S 3DVAR large scale bias correction
restoring to climatology at Gibraltar strait, Bab el Mandeb strait, and south of 60°S below 2000m
Focus on the altimetry era: 1992-now
Evaluation protocole from GODAE/GSOP/ORA-IP
Ocean reanalysis « GLORYS2V4 » 1993-2016 Ocean analyses 2007-now
Use of NEMO ORCA12 (1/12°) 50L + LIM
Forced with ECMWF IFS analyses and climatologicalrunoffs
Multivariate assimilation of SST, SLA, in situ T/S profiles (and monovariate for Sea Ice)
3D T and S 3DVAR large scale bias correction
Weak assimilation of EN4 climatology below 2000m
Adaptive observation errors for SLA and SST
Phased with homogeneous HR atmospheric forcing availability 2007-now
Evaluation protocole from GODAE/MyOcean/CMEMS
Cliquez et modifiez le titre
Global (re)analysis system in short
Use of NEMO ORCA12 (1/12°) 50 L+ LIM
Forced with ERA-interim reanalysis (large scalecorrection of precipitations and radiative forcing) and climatological runoffs
Multivariate assimilation of SST, SLA, in situ T/S profiles (and monovariate for Sea Ice)
3D T and S 3DVAR large scale bias correction
Weak assimilation of EN4 climatology below 2000m
Adaptive observation errors for SLA and SST
Focus on the altimetry era: 1992-now
Evaluation protocole from GODAE/GSOP/ORA-IP
Ocean reanalysis « GLORYS12V1 » 1993-2016 Ocean analyses 2007-now
Use of NEMO ORCA12 (1/12°) 50L + LIM
Forced with ECMWF IFS analyses and climatologicalrunoffs
Multivariate assimilation of SST, SLA, in situ T/S profiles (and monovariate for Sea Ice)
3D T and S 3DVAR large scale bias correction
Weak assimilation of EN4 climatology below 2000m
Adaptive observation errors for SLA and SST
Phased with homogeneous HR atmospheric forcing availability 2007-now
Evaluation protocole from GODAE/MyOcean/CMEMS
New in 2018: HR reanalysis GLORYS12, increased consistency
between reanalysis and NRT analysis
A very BIG dataset!
Tropical ocean in Mercator Ocean analyses
➢ Tropical oceans are a key area for ocean atmosphere interaction : Météo-France seasonal forecasting system oceanic initial conditions are derivedfrom Mercator Ocean analyses (cf Magdalena Balmaseda)
➢ Many other Marine applications require high quality high resolutionoceanic information in the tropical oceans: ▪ Defense▪ Fisheries, marine resources…▪ maritime safety, commercial ships routing▪ Marine renewable energies
pace
Coastal & marine
environment
Maritime safety
Marine
resources
Weather, climate & seasonal forecasting
Areas of benefit
Tropical ocean in Mercator Ocean analyses
• Data assimilation gives a compromise in between a model solution (first guess) and all sources of observations. The analysis is close to available observations on average -> the observing system is at the center, the more (QC) observations, the better
• Currently, scales smaller than ~¼° and ~1 day are not constrained -> in progress
• Errors cumulate where there are less observations/constraint: at depth, salinity
• Tropical oceans specificities: ➢ larger zonal correlations scales than in higher latitudes -> taken into
account➢ rapid wave propagation and strong vertical shear -> more difficult to
constrain➢ Issues with constraining equatorial dynamics with altimetry, MDT errors ->
large errors in currents
Sea surface temperature variability
SST variability is well represented from the weekly to the interannual scale. It isconstrained by atmospheric forcings and assimilated SST (OSTIA in real time, NOAA ¼° analyses in reanalyses)
Monthly SST average anomaly (°C, black line and color shading) in the nino3.4 boxDashed line is NOAA CPC nino3.4 index-> see CMEMS Ocean State Report, JOO 2017 -> Ocean Monitoring Indicators to appear on CMEMS catalogue in 2018, including Tropical Atlantic Boxes
Map of SST biases (NRT analyses)
NRT analyses are
too warm (~0.5°C) on
average with respect
to assimilated OSTIA
But bias with respect to in situ is
different ->
inconsistencies (foundation SST, in
situ depth etc…)
Lack of Trop atl in situ observations
Surface salinity
Surface salinity biases are reduced in the HR reanalysis with respect to ¼°reanalyses
GREP product: ensemble mean and standard deviation from 4 reanalysesORAS5, CGLORS, GLOSEA5, GLORYS2V4
Interannual variability of SSS
Hovmuller 2010 – 2015 of Pacific Ocean surface salinity 2°N-2°S for reanalysis ¼ ° (left) , and NRT analyses (right)seasonal cycle is removed
-> Influence of atmospheric forcings
Sea Level
RMS errors are very small on average (< 4 cm)
Significant biases persist in the Tropical Pacific.
RMS errors are large in highly variable areas. In the Tropical Atlantic RMS errors are large in the North Brazil Current.
Surface Currents climatology
Surface currents position and variability are well captured thanks to altimetry
GLORYS12V1 average zonal velocity 1993-2014 CMEMS INS TAC drifters average zonal velocity 1993-2014
m/s m/s
Surface Currents climatology
GLORYS12V1 average meridional velocity 1993-2014 CMEMS INS TAC drifters average merid. velocity 1993-2014
m/s m/s
Surface currents bias
U drift innovation in 2008-2013 (psy4v3r1)
m/s
Surface currents are not constrained directly by observations, and errors in winds or vertical physics, or MDT can induce large errors in currents -> equatorial divergence issues
Lack of observations for validation (here 5 years of drifters velocities are needed to produce an error map with global coverage)
Currents at depth
(from ARGO parking depth)
GLORYS12V1
Currents at depth
reanalysis
NRT analysis
Pirata O°E 23°W
Transports
From Mignac et al, OSD 2018
Spread in transports estimates -> linked withlack of near coastalmeasurements + DA tunings near the coasts
Green Ocean
BGC models with data assimilation of ocean colour:Encouraging results but still a long way to goFirst evaluations/calibrations with bio argoplanned in 2018
NO ASSIM ASSIM
DATA
Year 1995: Annual mean of chlorophyll concentration
GLORYS2V3 : Chlorophyll after 3 years
Assimilation of tropical moorings’ data
Tropical moorings have high temporal frequency but low spatial sampling -> underdetermined estimation problem for fast tropical waves, -> need of filtering of the data model misfits to remove unresolved scales.
Mean and RMS observation-analysis error to in situ temperature observations in the Nino 3 region : with the
TAO assimilated in red, without in blue.
Continuous improvement of analysis system
for marine applications
Case study of the search for the AF447 wreckageDrevillon et al, 2013 Clim. Dyn.-> under-observed conditions
With current NRT analysesWith NRT analyses available in 2010
Wreck was found near « ACARS » point
Ensembles of forward trajectories initiated from all points
inside the search area
Short distance score = minimum distance from all debris
is found for trajectories starting from those points
Conclusions
Strength and limitations of ocean analyses for marine applications:
• temperature and salinity variability and state are well captured, especially near the surface
• Surface currents variability is well captured
• Equatorial currents are too strong, especially at depth
• Only large scale information is extracted from current observation system
need for observations:
• All observations are valuable -> towards the use of high resolution observations
• HF moorings are essential for validation/calibration, and useful for DA
• More observations at depth will help reduce system biases and improve the capacity to capture trends
• More coastal observations + better taken into account could improve circulation
Perspectives
Perspectives:
• Data assimilation improves the average accuracy, but can induce spurious high frequency phenomena (gravity waves, recirculation cells) -> need for more process oriented validation of experiments with and without data assimilation
• Small scales (<1 day and < ¼°) are unconstrained -> improvements expected first from HR SST assimilation using “4D” approach, HF mooring observations impact will be evaluated
Part of this work is planned in Atlantos project and/or GODAE OSE-val TT
On the longer term
Ocean-Atmospheric Boundary Layer-waves coupling at high resolution (1/36°)
Ensemble runs -> uncertainty estimates, ensemble DA