Barcelona, 2015 Climate Prediction and Climate Services Virginie Guemas and the Climate Forecasting Unit 9 February 2015

Climate timescales and climate prediction Meehl et al. (2009) Focus on sub-seasonal, seasonal, interannual and decadal timescales

Climate system predictability  Memory on interannual to centennial timescales in the ocean  Memory on seasonal to interannual timescales in the sea ice and land surface  External radiative forcings (solar activity, greenhouse gases, aerosols)

Decadal climate prediction exercise Nov 2000 Nov 2001 Nov 2002 Nov 2003 Nov 2004 Nov 2005 Nov 2006 Forecast time 5 years Core Tier 1 Forecast time 1 year

Methodology Observations … until member prediction started 1 Nov 1960 Experimental setup : 1 grid-point

Methodology Observations member prediction started 1 Nov member prediction started 1 Nov 1960 Experimental setup : 1 grid-point

Methodology Observations … until member prediction started 1 Nov member prediction started 1 Nov member prediction started 1 Nov 1960 Experimental setup : 1 grid-point

Methodology Observations member prediction started 1 Nov 2005 … until member prediction started 1 Nov member prediction started 1 Nov member prediction started 1 Nov 1960 Experimental setup : 1 grid-point … every 5 years …

Methodology Observations member prediction started 1 Nov 2005 … until member prediction started 1 Nov member prediction started 1 Nov member prediction started 1 Nov 1960 Experimental setup : 1 grid-point Focus on averages over forecast years 2 to 5 … every 5 years …

Methodology Observations member prediction started 1 Nov 2005 … every 5 years … … until member prediction started 1 Nov member prediction started 1 Nov 1965 Experimental setup : 1 grid-point Focus on averages over forecast years 2 to 5 Ensemble-mean 5-member prediction started 1 Nov 1960

Methodology … until 2009 Experimental setup : 1 grid-point As many values as hindcasts for both the model and the observations to compute skill scores. Ex : correlations

Typical decadal forecast skill – IPCC AR5 Doblas-Reyes et al. (2013) Nature Communications (Top row) Root mean square skill score (RMSSS) of the ensemble mean of the initialised predictions and (bottom row) ratio of the root mean square error (RMSE) of the initialised and uninitialised predictions for the near-surface temperature from the multi-model CMIP5 experiment ( ) for (left) 2-5 and (right) 6-9 forecast years. Five-year start date interval. Added-value from initialisation Skill

Typical seasonal forecast skill Correlation of the ensemble mean for the ENSEMBLES multi-model (45 members) wrt ERA40-ERAInt (T2m over ) and GPCP (precip over ) with 1- month lead T2m JJA T2m DJF Prec JJAPrec DJF

Some open fronts Work on initialisation: generate initial conditions (e.g. for sea ice, ocean). Compare different initialisation techniques (e.g. full field versus anomaly initialisation) Improving model processes: Inclusion and/or testing of model components (biogeochemistry, vegetation, aerosols, sea ice) or new parameterizations, model parameter calibration, increase in resolution Calibration and combination: empirical prediction (better use of current benchmarks), local knowledge. Forecast quality assessment: scores closer to the user, reliability as a main target, process-based verification, attribution of climate events with successful predictions, diagnostics of model weaknesses with failing predictions More sensitivity to the users’ needs: going beyond downscaling, better documentation (e.g. use the IPCC language), demonstration of value and outreach.

Initialization : in-house sea ice reconstructions  NEMO3.2 ocean model + LIM2 sea ice model  Forcings : DFS4.3 or ERA-interim  Nudging : T and S toward ORAS4, timescales = 360 days below 800m, and 10 days above except in the mixed layer, except at the equator (1°S-1°N), SST & SSS restoring (- 40W/m2, -150 mm/day/psu)  Wind perturbations + 5-member ORAS > 5 members for sea ice reconstruction 5 member sea ice reconstruction for 1958-present consistent with ocean and atmosphere states used for initialization Guemas et al (2014) Climate Dynamics

Initialization : in-house sea ice reconstructions Reconstruction IceSat Too much ice in central Arctic, too few in the Chukchi and East Siberian Seas October-November Arctic sea ice thickness Guemas et al (2014) Climate Dynamics

Sea ice reconstruction – extraction of variability modes Clustering methods more robust than EOF analysis + account for nonlinearities. Tools available in s2dverification R package Fučkar et al (2015) ClimateDynamics k-means cluster analysis of reconstructed sea ice thickness (SIT)

Initialization : sea ice data assimilation Observations (e.g., ice concentration only) 1. Model forecasts 2. Analysis The ensemble Kalman filter: a multivariate data assimilation method for smoother initialization Francois Massonnet

Initialization : sea ice data assimilation Francois Massonnet OBS Importance of multivariate initialization for seasonal sea ice prediction

Initialization : sea ice data assimilation Francois Massonnet Fully-coupled sea ice data assimilation in EC-Earth: the next challenge  What are the perturbations required to generate adequate spread in EC-Earth during the forecast steps of the assimilation run ?  Should the atmosphere be updated when sea ice observations are assimilated?  Can we afford to run the EnKF with less members (CPU time is limited) ?

The climate prediction drift issue Observed world Retrospective prediction ( hindcast ) affected by a strong drift, need for a-posteriori bias correction Time Predicted Variable (ex. Temperature) BIAS Biased model world Danila Volpi

Testing bias correction methods – percentile matching Bias-corrected ECMWF S4 forecasts for November with start date in November over One-year-out cross-validation applied. Method 1: Simple = Per-pair Method 2: Percentile Matching Bias corrected forecast Uncorrected forecast Observation Veronica Torralba

Bias correction and calibration ECMWF S4 predictions of 10 m wind speed over the North Sea for DJF starting in November. Raw output (top), bias corrected (simple scaling = per-pair, left), ensemble calibration = percentile matching (right). One-year-out cross-validation applied. Veronica Torralba

Developing a new bias correction method IC (Initial conditions) bias correction method (green) accounts for the dependence of the climate prediction drift on the observed initial conditions through a linear regression -> lower forecast error Fučkar et al (2014) Geophysical Research Letters Tools available in s2dverification R package

The climate prediction drift issue  Issue : Distinction between climate drift and climate signal  Hypothesis : If the model climate is stable (no drift), the simulated variability is independent of the model mean state within the range of current model biases and closer to the observed variability than when mixed with the drift  Testing the hypothesis : Allowing the climate model biases but constraining the phase of the simulated variability toward the contemporaneous observed one at the initialization time : Anomaly Initialization (AI) Danila Volpi

The climate prediction drift issue Observed world Biased model world Retrospective prediction ( hindcast ) affected by a strong drift, need for a-posteriori bias correction Time Predicted Variable (ex. Temperature) BIAS Retrospective prediction with anomaly initialization Danila Volpi

Anomaly versus full-field initialization EC-Earth2.3, 5 members, start dates every 2 years from 1960 to 2004 NOINI : historical simulation FFI : Full-field initialization from ORAS4 + ERA OSI-AI : Ocean and sea ice anomaly initialization with corrections to ensure consistency rho-OSI-wAI : Ocean and sea ice weighted anomaly initialization to account for the different model and observed amplitudes of variability + (density, temperature) Instead of (temperature, salinity) anomaly initialisation Volpi et al (2015) Climate Dynamics

Anomaly versus full-field initialization RMSE AMO. Ref: ERSST RMSE PDO. Ref: ERSST RMSE sea ice area. Ref: Guemas et al (2014) RMSE sea ice volume. Ref: Guemas et al (2014) NOINI FFI OSI-AI rho-OSI-wAI Volpi et al (2015) Climate Dynamics

Anomaly versus full-field initialisation Experiment with the minimum SST RMSE Forecast year 1Forecast years 2-5 Volpi et al (2015) Climate Dynamics

Ensemble generation : Stochastic perturbations DJF one-month lead time bias for the 10-metre zonal wind (m/s) from EC-Earth3 T255/ORCA1 hindcasts over (10-member ensembles) with the standard forecast system and with SPPT. (blue = reduction in bias). Control |SPPT|-|Control| Lauriane Batté

Impact of initialization : CMIP5 decadal predictions Predictions Historical simulations Observations Atlantic multidecadal variability (AMV) Global mean surface atmospheric temperature CMIP5 decadal predictions. Global-mean t2m and AMV against GHCN/ERSST3b for forecast years 2-5. Doblas-Reyes et al. (2013) Nature Communications

Impact of sea ice initialization Predictions with EC-Earth2.3 started every November over with ERAInt and ORAS4 initial conditions, and our sea-ice reconstruction. Two sets, one initialised with realistic and another one with climatological sea-ice initial conditions. Ratio RMSE Init/Clim hindcasts 2-metre temperature (months 2-4) RMSE Arctic sea-ice area Guemas et al (2015) Geophysical Research Letters

Impact of sea ice initialization Predictions od NAO with EC-Earth2.3 started every November over with ERAInt and ORAS4 initial conditions, and a sea-ice reconstruction. Two sets, one initialised with realistic and another one with climatological sea-ice initial conditions. Javier Garcia-Serrano

Impact of land surface initialization Difference in the correlation of the ensemble-mean near-surface temperature (top) and precipitation (bottom) from two experiments (JJA), one using a realistic and another a climatological land-surface initialisation. Results for EC-Earth2.3 started every May over with ERAInt and ORAS4 initial conditions and our sea-ice reconstruction. Prodhomme et al (2015) Climate Dynamics

Impact of land surface initialization JJA precipitation in 2003 (top row) and near-surface temperature in 2010 (bottom row) anomalies from ERAInt (left) and experiments with a climatological (centre) and a realistic (right) land-surface initialisation. Results for EC-Earth2.3 started in May with initial conditions from ERAInt, ORAS4 and a sea-ice reconstruction over Prodhomme et al (2015) Climate Dynamics

Impact of increasing the resolution Mean SST (K) systematic error versus ERAInt for JJA one-month lead five-member predictions of EC-Earth3 T255/ORCA1 and T511/ORCA025. May start dates over using ERA-Interim and GLORYS initial conditions. EC-Earth3 T255/ORCA1EC-Earth3 T511/ORCA025 Chloe Prodhomme High – Low resolution

Predictions of DJF NAO with EC-Earth3 low and high resolution and ECMWF S4 started in November over with ERA-Interim and GLORYS initial conditions and five-member ensembles. Correlation of the ensemble mean on top left. EC-Earth3 T255/ORCA1 ECMWF S4 EC-Earth3 T511/ORCA025 Impact of increasing the resolution Lauriane Batté

Hurricane frequency predictions Average number of hurricanes per year estimated from observations and from EC-Earth CMIP5 decadal predictions. The correlation of the ensemble mean for the initialized, uninitialized and statistical predictions are shown with the 95% confidence intervals. Louis-Philippe Caron CMIP5 predictions Ec-Earth full- field initialized Ec-Earth anomaly initialized CMIP5 historical Persistance

Attribution of extreme events How has anthropogenic activity changed the odds of extreme events? Southern African drought (2002/2003) and flood (1999/200) Climate change has increased the risk of dry winter seasons and reduced the risk of wet winter seasons. Fraction of attributable risk FAR=1-P ALL /P NAT P ALL,NAT = Probability of observing the event using all forcings and natural forcings only. Omar Bellprat

Global mean Sea Surface Temperature Predictions of the XXI st century hiatus Forecast years 1 to 3 from climate predictions initialized from observations Observations (ERSST) Guemas et al (2013) Nature Climate Change EC-Earth2.3 CMIP5 decadal climate predictions capture the hiatus

Predictions of the XXI st century hiatus Observations EC-Earth historical simulations starting from 1850 preindustrial control simulations Forecast years 1 to 3 from EC-Earth climate predictions initialized from observations Crucial role of initialization from observations in capturing the plateau Guemas et al (2013) Nature Climate Change

Predictions of the XXI st century hiatus Ocean heat uptake (0-800m excluding the mixed layer) at the onset of the plateau Guemas et al (2013) Nature Climate Change Plateau explained by increased ocean heat uptake

Global Framework on Climate Services

Climate services: renewable energy Lienert and Doblas-Reyes (2013) Journal of Geophysical Research

Progress on open fronts Work on initialisation: more advanced data assimilation (ex: EnKF, coupled assimilation) to generate initial conditions, use of new observations and reanalyses, better ensemble generation. Improving model processes: Impact of aerosols, interactive vegetation, prediction of biogeochemistry, more efficient use of computing resources, drift reduction, leverage knowledge from modelling at other times scales Calibration and combination: estimation of uncertainty Forecast quality assessment: attribution of climate extremes (drought, sea ice minima and maxima), analysis of ocean, sea ice and land sources of predictability, role of external forcings More sensitivity to the users’ needs: going beyond downscaling, better documentation (e.g. use the IPCC language), demonstration of value and outreach.