Lessons learned with EC-Earth on the predictability of the Earth system within the CMIP6 exercise and beyond

�Pablo Ortega (BSC)

On behalf of the Climate Prediction Working Group

Virtual EC-Earth meeting, 16-18 November 2021

A new decadal prediction system based on EN4/ORAS5 (BSC)

New

Old

New - Old

• Old full-field DCPP system based on ORAS4 cannot be used for new forecasts
• New initialization method assimilates ORAS5 SST and SSS and EN4 T,S in the subsurface to circumvent the non-stationarity biases in ORAS5
• The new method seems to partly correct the loss of skill in the Central North Atlantic

​

​

R. Bilbao, J. Acosta Navarro, I. Ayan, P. A. Bretonniere, V. Lapin, F. López Martí, P. Ortega, V. Sicardi, E. Tourigny

Anomaly correlation for surface air temperature

Reducing biases in Arctic sea ice (DMI)

T. Tian, R. Davy*, S. Yang and S.M. Olsen

​

by improving the representation of SSHF through leads

Leads in Central Arctic cover 1.2% of the ocean during winter, but account for more than 70% of the upward heat fluxes

GCMs generally have too coarse a resolution (~100 km) to simulate leads explicitly

Experiments: Two sets of 50-year simulations with both warm (2015) and cold (1985) fixed forcing, only differing in the activation of lead parameterization (Lead/Ctrl).

DMI has introduced in EC-Earth a new parameterization (developed by NERSC*) to amplify the sensible heat flux through lead when SIC >70% .

Future work:

Disentangle the added value of leads in a warm climate in pairs of seasonal and decadal forecasts

Use of satellite sea ice data in parameterization

Large sea ice cover (1985)

Small sea ice cover (2015)

Bias Bias reduction

Bias Bias reduction

Testing sea ice freeboard assimilation in coupled mode (UCLouvain)

F. Massonnet, S. Fleury, F. Garnier, E. Blockley, P. Ortega, J. C. Acosta Navarro, L. Ponsoni and F. Klein

Assimilation Method

First results (September 2012)

Deterministic Ensemble Kalman Filter (DEnKF, Sakov et al. 2008)

NERSC implementation adapted to EC-Earth3 (Massonnet et al. 2015)

​

​

DEnKF updates ocean/sea ice fields that are not observed, by exploiting covariances in the forecast ensemble. Atmospheric fields are not updated

Note: the observed 2012 September sea ice conditions were affected by the « great » cyclone of early August 2012, which is not predictable.

• Assimilation reduces, but does not eliminate, the biases in the Chukchi- Beaufort-East Siberian Seas by prescribing thinner ice.
• Constraining SIC/SIT/SIFB alone seem to be similarly efficient

Added value of assimilating SIC on seasonal prediction (BSC)

J. Acosta Navarro, J. García-Serrano, V. Lapin, P. Ortega (to be submitted)

Improvements in summer prediction skill

Start month: May Reforecast period: 1992-2018 Ensemble size: 30 members

Sea Ice Concentrations

Geopotential Height at 500 hPa

Quantile-matching method for initializing decadal hindcasts (CNR)

D. Volpi, V. Meccia, V. Guemas, P. Ortega, R. Bilbao, F. Doblas-Reyes, A. Amaral, P. Echevarria, R. Mahmood and S. Corti (2021, Front. Clim.)

​

Near surface temperature anomaly correlation (AC)

AC in the Labrador Sea

Ref: GISTEMP

Mixed layer depth

Barotropic streamfunction

Experimental set-up

10 ensemble members

5 forecast years

Start-dates: Nov 1960-2014

Comparison with a full field initialised (FFI) hindcasts

and 15 historical simulations

The quantile matching (QM) in a nutshell

The initial state is the model state that matches the observed distribution at initialisation time

QM allows the model to be on its attractor (reducing the initial shock and the drift)

Three different approaches for decadal prediction (SMHI)

T. Kruschke, P. Karami, S. Wang, F. Schenk and T. Koenigk

(contributions to CMIP6-DCPP, EUCP and ArcPath)

Standard Resolution

(1° ocean, T255 atmos)

High Resolution

(1/4° ocean, T511 atmos)

Coupled Assimilation

(1° ocean, T255 atmos)

SMHI & DMI set up a quasi operational decadal prediction system with EC-Earth (v3.3.1.1)

15 members, initialized annually on 1 November throughout the period 1960-2020

Anomaly initialization for ocean + sea-ice concentration and thickness (new method was developed)

Full-field initialization for atmosphere

Tian et al., 2021

Karami et al., In prep

Same initialization approach as in SR

forecast period 1990-2004, 10-member; 5 years and 2 months

​

​

​

​

​

​

​

​

​

​

HR shows improved skill over SR

Skill difference HR-SR

10 historical-SSP2-4.5 runs: 1850-2030

​

5 assimilation runs started from 5 historicals with SST restored to HadISST1 anomalies: 1900-1949

(2 runs continued until 2015)

​

5 assimilation runs started from previous ones, with SST restored to HadISST1-anomalies + atmospheric relaxation to 6h ERA5-anomalies (surface near winds, surface pressure): 1950-2020

​

Hindcasts to be run soon

On the path towards a HR decadal prediction system (BSC)

A. Carreric, S. Palomas, M. Castrillo, M. Donat, P. Ortega and F. Doblas-Reyes

1) Tuning of EC-Earth3.3-HR

3) Seasonal Hincasts

It has reduced strong Arctic sea ice biases in EC-Earth3P-HR

​

It does not show collapses of Labrador Sea convection as in EC-Earth3.3-LR

​

2) Producing HR reconstruction

Same approach as for the LR decadal predictions (with a small change in snow conductivity to compensate for a warm bias in ERA5)

Reforecast period: 1990-2015 (summer)

15 members (May to September)

10 members (May to December)

​

​

​

​

​

​

​

​

​

​

​

​

​

Skill levels higher than for SEAS5

4) Reduced DCPP system

Skill for predicting Labrador Sea SST

Normal Reduced

​

1960-2020 1960-2020

(every yr) (every 2rd yr)

​

10 members 5 members

​

10 forecast yrs 3 forecast yrs

​

TOTAL (6100 yrs) TOTAL (465 yrs)

Hindcasts to be run soon

​

Depending on the results we could later run more members/start dates/forecast years

A first decadal hindcast of carbon cycle with EC-Earth-ESM (BSC)

J. Acosta-Navarro, E. Tourigny, R. Bernardello, V. Sicardi, E. Exarchou, V. Lapin

3) Seasonal Hincasts

EC-Earth-3.3-ESM (concentration driven) Ensemble size: 5 (+10 in production)

Reforecast period: 1980-2020 Forecast length: 7 years

Next steps: using emission driven configuration of EC-Earth3.3-ESM

Skill of global ocean CO2 flux anomalies

Skill of global land CO2 flux anomalies

ref: GCB2020

Skill of global land+ocean CO2 fluxes

Ocean initialization: As in new EC-Earth3-GCM system (only physics assimilated) → 3 cycles to get present CO2 uptake

Land initialization: Offline land-surface reconstruction with LPJ-GUESS → forced with ERA20C/ERA5 (prior/after 1950)