Evolution of the Mercator Océan system, main

components for reanalysis and forecast

Y. Drillet, J.M. Lellouche, O. Le Galloudec, M. Drévillon, G. Garric, R. Bourdallé-

Badie, C. Bricaud, J. Chanut, G. Reffray, C.E. Testut, L. Parent, E. Remy, J. Beuvier,

C. Regnier, E. Dombrowsky, B. Tranchant, M. Benkiran, E. Greiner, B. Barnier

2

Plan

• Brief overview of the Mercator systems

• Available real time and reanalysis systems

• Performance of the systems

• Ocean modelling

• Atmospheric forcing

• Vertical mixing

• Vertical coordinates

• Data assimilation

• Sea Ice assimilation

• Incremental Analysis Update

• Bias correction

• Ongoing and futur works

• Conclusions

3

MERCATOR OCÉAN SYSTEMS

4

Operational systems running today and reanalysis

Global:

• Global 1°, Operational since mid 2011 real time for seasonal forecasts

application . Replacing a previous 2°(started in 2003).

• Global ¼°, Operational since 2005 real time, daily service. Provides

the physics to the BGC system

• GLORYS2V3 Global ¼° reanalysis. From 1993 to 2012.

• Global 1/12°, Operational since 2010 real time, daily service

Regional

• Natl + Med 1/12°, Operational since 2003 real-time, daily service,

embedded in the Global ¼°, provides IC and BC to the IBI system

• IBI 1/36°, operational since Dec 2011 real-time, daily service, include

tides, aimed at delivering service to coastal systems

• IBIRYS1V1 IBI 1/12° reanalysis over North East Atlantic area including

tide, pressure forcing, and online biogeochemistry. From 2002 to 2012.

5

Elapsed time for operational forecast

(weekly scenario including 2 analysis and 1 forecast)

Tim

e in

ho

urs

500 Gflops 1 Tflops

10 Tflops

Systems update

50 Tflops 1 Pflops

Computational power available at Météo France

6

Evolution of performance scores

« Quo Va Dis ? » Bulletin available online

http://www.mercator-ocean.fr/eng/science/Qualification-validation2

Lellouche et al, 2013. Ocean Science.

Present version

SLA

Salinity

Temperature

7

OCEAN MODELLING

8

brief history

• 2003-2005 : The community model is called NEMO, it is choosen by several

groups in Europe. This version is based on OPA9 including free surface and

partial step, ice model (LIM) and tracer module (TOP). It is freely distributed.

• 2005-2006 : NEMO include mesh refinement (AGRIF), several vertical mixing

(KPP) and advection (UBS) schemes, IO module

• 2007-2008 : non linear free surface with variable volume, AGRIF for passive

tracers, new open boundary module (bdy, obc), multi category ice model

(LIM3), surface module, interpolation on the fly

• 2008 : Consortium agreement signed between CNRS, Mercator Océan,

MetOffice, NERC

• 2009-2010 : tidal mixing, light penetration, variable volume with time splitting,

GLS vertical mixing scheme, runoff, diurnal cycle

• 2011 : Consortium agreement includes INGV and CMCC

• 2011-2013 : new pressure gradient for s-coordinate, semi implicit bottom

friction, new bulk formulation, tidal potential forcing, XIOS parallel output server

9

physics and parameterisations in standard configurations:

code : NEMO 3.4 (OPA9+LIM) Grid: ORCA at 1/4°, 1/12°, 1/36° 50 levels (1 to 450 m) or 75 levels (1 to

200m) z coordinates with partial step Parameterizations: Filtered or explicit free surface; TVD; Lap. Iso. on tracers ; biharmonic on momentum; TKE or K-ε turbulent closure scheme

=

FORCING: Bulk: CORE + analytic or explicit

diurnal c.

Atmospheric Fields: 3h ERAinterim or operational ECMWF Precipitations corrected towards GPCP Long & short wave corrected towards GEWEX

Initialization: Levitus 2005 Or global

reanalysis Or real time

analysis

10

Mean 2009 after 20 yrs simulation,

comparison with ARMOR salinity.

• Fresh bias largely corrected

• Still fresh in Northern Atlantic

• Too salty in 20-30°lat band

• Impact of correction at depth

• Still too fresh in Atlantic Ocean.

• Too salty in Pacific Ocean

Atmospheric forcing impact on salinity

Corrections implemented – Rainfalls fluxes

ARMOR

No correction

Correction

Pacific

Ocean Indian

Ocean

Atlantic

Ocean

0 m

1500 m

The correction is :

• Local

• Large scales

• Based on GPCP

No correction are

applied :

• Northward 65°N

• For small value of

heat flux or

precipitation

Method induces :

• No change of

interannual signal.

• No change of

synoptic patterns

(cyclones).

« error » SSS without correction

« error » SSS with correction

-1 psu +1 psu

11

Vertical mixing scheme in global configuration, impact on

stratification and SST, 3D error (model-Levitus05)

TKE

+

Solar penetration rgb bands

TKE

+

Solar penetration 2 bands

K-ε

+

Solar penetration 2 bands

3 global ¼° experiments

12

NEMO (v 3.4) vertical coordinates

Hybrid systems:

Coordinates defined thanks to a “smoothed”

bathymetry enveloppe to reduce pressure gradient

errors.

z

z+part.

cells

s

s-z

s-z + partial

cells

Uncoded yet

NEMO

60 s levels

(Song and

Haidvogel

steching)

NEMO 75 levels

hybrid system

Max slope =5%

13

2400m

, z-v

coord

.

2400m

, hybrid-v

coord

.

Impact of vertical coordinate on the overflow

Bottom

salinity

Depth

of

max

salinity

14

DATA ASSIMILATION

15

- Development steps of SAM2 : - [2003-2005] : First developments of the SAM2 kernel

- [2004-present] : Introduction of new parametrisations for the operational applications

(new background error, adaptive scheme, IAU, …)

- [2007-present] : Deployment through all Mercator operational configurations

- Baseline of ocean observation datasets : - along-track sea level anomalies;

- vertical T/S profiles;

-Global SST maps

- Sea ice concentration

- Assimilation scheme

– Based on a multivariate SEEK filter

– Innovation are calculated using “3DFGAT”

- Forecast error covariances Pf computed with a pseudo ensemble

- IAU (Incremental Analysis Update) scheme implemented

- Bias correction scheme implemented in all systems

SAM2 Data Assimilation System

16

Assimilation of Sea Ice Concentration : Main features of the GLORYS2v3 simulation

Context - French Reanalysis project produced at Mercator

- Based on Mercator operational systems + tuning and developments

- Produce reanalysis spanning the “altimetric + ARGO" era 1992-2012

Model - Nemo 3.1, LIM2-EVP

- Global ¼, 75 levels

- 1992-2012

Sea Ice Concentration

from CERSAT (IFREMER)

Assimilation - 2 separate SAM2 analyses, 7 days cycle

- Ocean Analysis (SLA, InSitu Data from CORA3.2,

SST) , IAU on (h,T,S,U,V)

- Ice Analysis (SIC), IAU on (SIC)

- SIC Error: 1% open ocean, linear from 25% to 5%

for SIC values between 0.01 and 1

- Temperature and salinity bias correction using

Argo (3DVar method)

17

CERSAT

Assimilation of Sea Ice Concentration : Impact in Antarctica with GLORYS2V3 System

Sea Ice Concentration on 15th September 1992 (assimilation start in December 1991)

Sea Ice Concentration Misfits to Observation (CERSAT)

on 15th September 1992

Sea Ice Concentration RMS misfits

G2V3-NOASSIM/ICE

G2V3-ASSIM/ICE

Jan 1992 Sep 1992 May 1993

Jan 1992 Sep 1992 May 1993

GLORYS2V3-NOASSIM/ICE Global ¼°, 75 levels

GLORYS2V3-ASSIM/ICE Global ¼°, 75 levels

18

• SAM2 scheme : All operational systems since 2009

- 7 days cycle

- 1 Analysis using 3D modes at time t=4.0

- 3D model update at time t=4.0

- IAU during 8 days

IAU weight function : increasing during 1 day, constant during 6 days and

decreasing during 1 day

SAM2 Data Assimilation System : Incremental Analysis Update method

• Objectives: Better control timely the analysed trajectory and improve the backward (and

the forward) propagation of the information from the observation

19

Global ¼ (ORCA025)

SAM2 Data Assimilation System : Comparison Forecast run vs Analysed run (Best)

RMS T RMS S

Altimetry

SST

Analysed Run

2nd Run

Forecast Run

1st Run

20

Mercator Data Assimilation System : Temperature and salinity bias correction using Argo

Due to Argo network good spatial coverage, it is possible to perform bias

correction for Temperature and Salinity.

Salinity innovation at 130m Exemple for Sep.-Nov 2007

Bias method correction :

Step 1. Collection of innovations (T&S) over the past 3 months Step 2. Analysis of the bias (3DVAR method) Step 3. Model correction using a Incremental Analysis Update (IAU) method

21

Mean misfits on Global ¼ (PSY3)

Mercator Data Assimilation System : Temperature and salinity bias correction using Argo

S

T

No bias correction With bias correction

22

Ongoing and futur works for next systems version

• Modelisation

• New NEMO options or modules in global configuration

(Agrif, tide, numerical scheme …)

• Improvment of ocean atmosphere interaction

• Multi category sea ice model (LIM3)

• biogeochemistry

• Assimilation

• Adaptative observation errors based on Desroziers criteria

• Assimilation of new observations (SSS, high resolution

SST, velocity, ocean color, sea ice …)

• Atmospheric forcing correction

• Evolution of the assimilation system including 4D error

covariances, anamorphosys transformation, ensemble

approach

23

Conclusions

• Computational power increases but systems are

more and more complex and costly. Resolution,

ensemble methods, biogeochemistry, atmosphere

coupling are costly.

• Highlight system improvement is not

straightforward. Error of systems with assimilation

are close to the observation errors specified in the

system. Independent observations are rare.

Intercomparison of systems, new diagnostics,

statistics, metrics and link with applications are

useful.