Prediction of extreme events at Sub-seasonal to Seasonal lead times Frédéric Vitart, Laura Ferranti, Ivan Tsonevsky ECMWF

INDEX S2S database MJO/Weather regimes in S2S database Case studies: 2010 Heat Wave over Russia Tropical Cyclone PAM Extreme Forecast products and Verification

WWRP/WCRP Sub-seasonal to Seasonal (S2S) Prediction Project Teleconnections (C. Stan and H. Lin) Madden-Julian Oscillation (D. Waliser and S. Woolnough) Monsoons (H. Hendon) Sub-Projects Africa (A. Robertson and R. Graham) Extremes (F. Vitart) Verification and Products (C. Coelho) Research Issues Predictability Teleconnection O-A Coupling Scale interactions Physical processes Modelling Issues Initialisation Ensemble generation Resolution O-A Coupling Systematic errors Multi-model combination Needs & Applications Liaison with SERA (Working Group on Societal and Economic Research Applications) S2S Database

S2S database 3-week behind real-time forecasts + re-forecasts (up to day 60) Common grid (1.5x1.5 degree) Data archived with a daily frequency (sub-daily for total precip/max and min 2mtm) in GRIB2 at ECMWF and CMA About 80 parameters, including: 3D fields (u/v/w/z/t/q) on 10 pressure levels (up to 10 hPa) Surface fluxes Sea Surface temperature Sea-ice cover (fraction) Snow depth/density/snow fall/snow albedo

WWRP/WCRP S2S Database Time-range Resol. Ens. Size Freq. Hcsts Hcst length Hcst Freq Hcst Size ECMWF D 0-46 T639/319L91 51 2/week On the fly Past 20y 2/weekly 11 UKMO D 0-60 N216L85 4 daily 1993-2015 4/month 3 NCEP D 0-44 N126L64 4/daily Fix 1999-2010 1 ECCC D 0-32 0.45x0.45 L40 21 weekly 1995-2014 BoM T47L17 33 1981-2013 6/month JMA D 0-34 T319L60 25 1981-2010 3/month 5 KMA 1996-2009 CMA D 0-45 T106L40 1886-2014 CNRM T255L91 1993-2014 2/monthly 15 CNR-ISAC 0.75x0.56 L54 40 HMCR D 0-63 1.1x1.4 L28 20 10

Sub-seasonal prediction of extreme events Prediction of large-scale, long lasting events (> 1 week): Heat/cold waves Droughts Flooding Prediction of statistics of small scale events, for example: Tropical cyclones Tornadoes

MJO influences on large floods of the West Coast of North America Madden Julian Oscillation MJO PHASES Locations of large floods during 1985–2010 MJO influences on large floods of the West Coast of North America From C. Zhang, BAMS 2013

Bivariate Correlation with ERA Interim – Ensemble Mean 1999-2010 re-forecasts All Year DJF

MJO Teleconnections (S2S re-forecasts) EI 0.48 Z500 Composites Phase 3 + 3 pentads NDJFM BoM 0.15 CMA 0.14 HMCR 0.13 NCEP 0.32 ISAC 0.25 CNRM 0.15 UKMO 0.28 JMA 0.22 ECCC 0.21 ECMWF 0.31

Predicting skill associated with the Euro-Atlantic Regimes: NAO - NAO + Bom Bom Blocking Atlantic Ridge Bom Bom Ens. Meand anomaly correlation as a function of forecast lead time for the prediction of the 4 EA regimes. The scores are computed for the common period of reforecast covering 12 years (1999-2010) we have applied a 5 days running mean prior the Acc computation. The ACC for NAO- drops below 0.5 between 11-15 days while for the BL ACC drops about 2-3 days earlier (9-12 days for BL) this results is consistent with other studies.

Russian Heat Wave July-August 2010 Worst heat wave on record over the past 33 years (Hoag, Nature 2014) Estimated 55,000 deaths Wildfires, smoke, worst drought in nearly 40 years, and the loss of at least 9 million hectares of crops ERA interim 2mtm anomalies 1-7 August 2010

Russian Heat Wave 2010 1% 5% 4 July 1 August 15 August ERA Interim Day 1-7 Day 8-14 Day 15-21 1% 5% 4 July 1 August 15 August WEEK1: time evolution of heat wave well predicted WEEK2 and 3: Onset and decay predicted one week too late Timing of maximum well predicted

2mtm anomalies over Russia – ECMWF reforecasts 1-7 August 2010 1% 5% Lead time d26-32 d22-28 d19-25 d15-21 d12-18 d8-14 d5-11 d1-7

2mtm anomalies over Russia – S2S reforecasts 1-7 August 2010 UKMO (17 Jul) BOM (16 Jul) ECMWF (18 Jul) NCEP(16 Jul) ECCC(21 Jul) JMA(20 Jul) HMCR(20 Jul) CNRM(15 Jul) ERA Interim

2mtm anomalies over Russia – S2S models reforecasts 1-7 August 2010 July 18 17 15 18 21 20 20 16 18 20

Tropical Cyclone PAM Case Study Formed March 6, 2015 Dissipated March 22, 2015 Hit Vanuatu islands on 13 March 2015 Most intense tropical cyclone of the south Pacific Ocean in terms of sustained winds and regarded as one of the worst natural disasters in the history of Vanuatu.

March 2015 MJO Forecasts starting on 26 Feb 2015

Modulation of tropical cyclone density anomaly by MJO MJO Phase 2-3 MJO Phase 4-5 MJO Phase 6-7 MJO Phase 8-1 OBS ECMWF NCEP JMA BoM Modulation of TC activity by the MJO. The plots show the anomaly of TC density as a function of MJO phase for 5 different S2S models and the multi-model combination for the time range day 10-32. The figure shows a remarkable agreement with observations and suggest that all the model simulate very well the modulation of TCs by the MJO. This is an encouraging result for TC prediction Multi

Probability of a TC strike within 300 km Prediction from ECMWF Probability of a TC strike within 300 km 09 March 2015 Day 1-7 02 March 2015 Day 8-14 23 February 2015 Day 15-21 16 February 2015 Day 22-28

Tropical Cyclone Pam case study Multi-model prediction of TC strike probability anomalies- 9-15 March 2015 (NCEP/ECMWF/BoM/JMA/CMA) 2015/02/19 day 19-25 2015/02/26 day 12-18 PAM case study: Tropical cyclone strike (within 300 km) probability anomaly (relative to model climatology) for the forecasts starting on 19 February 2015 (left panel) and 26 February 2016 (right panel) and verifying on 9-15 March 2015. During the 9-15 March 2015, Pam, the strongest tropical cyclone of the season, hit Vanuatu (black dot on the map) with devastating consequences. The multi-model forecasts (combination of NCEP/ECMWF/BoM/JMA and CMA) shows an increased risk of tropical storm strike in the area of Vanuatu at both time ranges. In fact the anomaly map is consistent with the impact of an MJO event (in this case it was the strongest MJO event on record) on TC density. During the week the MJO active phase was over the west Pacific, which is consistent with less TC activity over the Indian ocean and more TC activity over the western pacific, and the possibility of twin TC genesis (over the northern and Southern hemisphere). The model seems able to predict the increased risk of a twin TC.

Impact of Resolution - Tropical cyclone PAM - 9-15 March 2015 Probability of a TC strike within 300 km Day 12-18 Day 19-25 64km 32km 16km 110km+ SP Verification

Verification of extreme events in S2S database Main issues: Rarity of extreme events in observations Low number of ensemble members in most S2S re-forecasts Low frequency of ensemble reforecasts in some S2S models Short common re-forecast period 1999-2010 S2S real-time forecasts are more suitable for verification of extreme events, but period covered is small (archived only from 1st Jan 2015)

Relative Operating Characteristics (ROC) score Decile Probabilities – May 2015- April 2016 - NH ECMWF real-time forecasts Day 12-18 Probability to be in lower decile Probability to be in upper decile Persistence of previous week Forecast

2m temp Cumulative Distribution Function ensemble predictions for 29 June - 5 July 2015 Climate 15 June 2015 18 June 2015 22 June 2015 25 June 2015 (15-21d) (12-18d) (8-14d) (5-11d) Cumulative Distribution function. area = [ 50., -10. ,38, 15 ] Observed anomaly

2m temp Extreme Forecast Index forecast range:12-18days verifying 8-14 August 2016 Ncep Ecmwf JMA Ukmo http://www.ecmwf.int/en/research/projects/s2s/charts/s2s/

EFI skill assessment Preliminary results based on ECMWF system:

Example of Attribution: March 2013 Cold wave over Europe 2mtm anomalies ERA Interim ECMWF - strong MJOs NCEP - strong MJOs ECMWF - Weak MJOs NCEP - Weak MJOs Slide 34: An example of MJO impact on the Northern Extratropical weather. March 2013 was exceptionally cold over western Europe and the northern part of Asia. This extreme event coincided with an MJO propagation in the western Pacific, which could explain the negative NAO which developed in that period and which is conducive to cold temperature anomalies over western Europe (see slide 24). The ECMWF monthly forecast starting on 14 February 2013 was successful in predicting the onset of this event up to 4 weeks in advance, although the ensemble mean anomalies are weaker than the observed anomalies. To evaluate the impact of the MJO, a composite of the 10 ensemble members which predicted the strongest MJO event in the western Pacific (middle panel) and a composite of the 10 ensemble members which did not predict an MJO in the western Pacific (bottom panel) were produced. Results suggest that 10 ensemble members with the strongest MJO produced 2-meter temperature anomaly forecasts over the northern Extratropics consistent with analysis (top panel). On the other hand, the 10 ensemble members which did not predict the MJO event produced bad 2-metre anomaly forecasts over Europe. This result suggests that the MJO event played a key role in the successful prediction of the cold event over Europe.

Conclusions The S2S database can be a useful resource for case studies, skill assessment and also attribution of extreme events S2S models display skill to predict MJO up to 3-4 weeks. However impact on NAO is weaker than in re-analysis and skill to predict European blocking (important for summer heat wave prediction) and NOA- (important with winter cold waves) is limited to 2 weeks. Russian heat wave 2010: S2S model forecasts provided indications of an exceptional warm anomaly more than 10 days in advance. Tropical cyclone PAM: High probabilities of TC strike 2-3 weeks in advance, most likely because of the impact of the strongest MJO on record on TC activity Preliminary verification of ECMWF model suggests some useful skill for decile and EFI extended-range prediction

Regimes based on clustering of daily anomalies for 29 cold seasons (1980-2008) 500 hPa geopotential Obtain well-known Euro-Atlantic regime patterns The low frequency variability in European temperatures have been often associated with the frequency of occurrence of these well-known Euro-Atlantic flow regimes. The positive and negative phase of the North Atlantic oscillation the Scandinavian blocking and the Atlantic ridge. The large spatial scale and the low-frequency nature of such flow patterns are the key attributes for successful predictions at the sub-seasonal time-scale. In fact, circulation patterns like the NAO are often associated with global teleconnections through propagation of Rossby wave trains. ‘k means’ clustering applied to EOF pre-filtered data (retaining 80% of variance) m2s2

Heat Wave Prediction in ECMWF re-forecasts ERA Interim Monthly Forecast Day 12-18