Munehiko Yamaguchi 1 21 August 2014 (Thu.) Multi-model ensemble forecasts of tropical cyclones using TIGGE World Weather Open Science Conference Montreal, Canada Collaborators: Frederic Vitart 2, Simon Lang 2, Linus Magnusson 2, Russell Elsberry 3, Grant Elliott 4, Masayuki Kyouda 1, Tetsuo Nakazawa 5, Koji Kuroiwa 5 1: Japan Meteorological Agency 2: European Centre for Medium-Range Weather Forecasts 3: U.S. Naval Postgraduate School 4: Bureau of Meteorology in Australia 5: World Meteorological Organization

Outline 1.Introduction of TIGGE What is TIGGE? What is the benefit of using TIGGE? 2.Ensemble tropical cyclone activity prediction Motivation, Verification Methods, Results - Single model ensemble - Multi-model ensemble (e.g. Multi-centre grand ensemble) 3.Multi-center ensemble predictions for Hurricane Sandy, Cyclones Phailin and Typhoon Haiyan 4.Summary 5.Future directions

What is TIGGE? Past Present Future Research Phase TIGGE (started 2006) Cyclone XML (started 2008) Goal: Enhanced use of ensemble prediction for operational purposes Operational Phase Various projects to demonstrate the value of ensemble prediction have been conducted. North Western Pacific Tropical Cyclone (TC) Ensemble Forecast Project (NWP-TCEFP) Severe Weather Forecasting Demonstration Project (SWFDP)

What is the benefit of using TIGGE? Past Research Phase TIGGE (started 2006) Cyclone XML (started 2008) MCGE EPS at CMC EPS at NCEP EPS at JMA EPS at ECMWF EPS at XXX TIGGE makes it possible to construct a new ensemble, that is Multi-Center Grand Ensemble (MCGE). MCGE is an ensemble of ensembles of various NWP centers.

Track Prediction for Typhoon SOULIK (2013) Blue portion of the tracks is the Day 1 forecast and the green, orange, and red portions are the Day 2, Day 3, and Day 4 forecasts.

Ensemble Size = 207

Black line is the observed track. The number on the black line indicates day(s) from the initial date.

Track Prediction for Typhoon FITOW (2013)

Ensemble Size = 207

Typhoon SOULIK Init.: UTC Typhoon FITOW Init.: UTC What is the benefit of using MCGE? MCGE products provide forecasters with additional information on the forecast uncertainty and increase the level of confidence in the forecast.

Systematic verification of MCGE The relative benefits of MCGE over single model ensemble (SME) are investigated from both deterministic and probabilistic perspectives. 58 TCs in the western North Pacific from 2008 to 2010 are verified. 1.TC strike probability Reliability is improved in MCGE, especially in the high-probability range. MCGE reduces the missing area by about 10 %. 2.Confidence information When multiple SMEs simultaneously predict the low uncertainty, the confidence level increases and a chance to have a large position error decreases. 3.Ensemble mean track prediction The position errors of 5-day predictions by the MCGE-3 are slightly smaller than that of the ensemble mean of the best SME although the difference is not statistically significant. Yamaguchi, M., T. Nakazawa, and S. Hoshino, 2012: On the relative benefits of a multi-centre grand ensemble for tropical cyclone track prediction in the western North Pacific, Q.J.R. Meteorol. Soc., 138,

NWP-TCEFP website Main Page ( MRI/JMA operates a website of NWP-TCEFP where the MCGE products of TC tracks are available. Send to to get ID and password

Ensemble tropical cyclone activity prediction

Frequency of days from TC genesis to the landfall - Japan- Days Frequency (%) Courtesy of Hirose and Fudeyasu (Yokohama National Univ.)

Frequency of days from TC genesis to the landfall - Philippines- Courtesy of Hirose and Fudeyasu (Yokohama National Univ.) Days Frequency (%)

Motivation TC genesis and subsequent track forecasts are important from a perspective of early warnings. Such forecasts would be particularly important for countries located in the lower latitude where a lead time from TC genesis to the landfall is relatively short. Although the performance of ensemble TC predictions has been studied well, the verification samples are usually limited to prediction cases where TCs exist at the initial times (i.e. TC strike probability prediction). There are few studies that verify TCs created during the model integrations on the medium-range time scale (i.e. TC genesis and subsequent track prediction).

Methodology 1/2 TC activity (i.e. TC genesis and subsequent track ) ensemble forecasts from short- to medium range time scales (0 – 14 days) are verified at each TC basin below. ecpatlwnpnin sinausspc The latitude is limited up to 25 ˚N or S to focus on TC activity in the lower latitude.

TC activity ensemble forecasts are verified within a 3 day time window, which is applied over a forecast length of 2 weeks. For the ensemble forecasts, the TIGGE data set from ECMWF, JMA, NCEP and UKMO is used. The initial time is 12 UTC only. As the forecast length of JMA is day 9, verification for JMA is up to a time window of 6 – 9 days. Day …… Methodology 2/2

Example: TC activity probability maps -Haiyan- Observation (0 or 100%) Initial time of the forecasts: 2013/10/31 12 UTC (about 4 days before the genesis and 8 days before the landfall over the Philippines) Time window: 2013/11/05 12 UTC – 2013/11/08 12 UTC (T+5days – T+8days) Climatological TC activity of this initial time and this forecast time window TC activity probability maps A threshold distance of 300 km is used to determine whether observed or forecast TCs affect a grid point.

Verification Metric Brier skill scores (BSS) for each time window at each basin are calculated. Verification period is January to December (June) in 2010 to 2013 for the basins in the northern (southern) hemisphere. Verified TCs are those with a tropical storm intensity or stronger (35 knots or stronger). Threshold wind values of 15, 20, 25, 30 and 35 knots are tested to define model TCs. In plotting the brier skill scores, a brier skill score giving the best performance among the 5 threshold values is selected for each NWP center and for each time window.

The best track date to create daily climatological TC activity is as follows. BasinData SourcePeriod atlNHC ecpNHC ausJTWC sinJTWC spcJTWC ninJTWC wnpJMA Climatological TC activity

Result - Intercomparison- Red: ECMWF, Blue: JMA, (up to 9 days), Yellow: NCEP Green: UKMO X axis: time window, Y axis: brier score (larger is better) ecpatlwnpnin sinausspc

Example: North Atlantic, Time window 6-9 days Initial time of the forecasts: 2010/09/09 12 UTC 1 day after the genesis of Hurricane Igor Approximately 3 days before the genesis of Hurricane Julia Approximately 5 days before the genesis of Hurricane Karl Karl Igor Julia

Sensitivity to the definition of model TCs Reliability Diagram Basin: ecp Time window: 2-5 days NWP: UKMO 35 knots 30 knots 25 knots20 knots15 knots Forecast probability Obs. Frequency y=x Obs > Fcst Obs < Fcst

Result -Relative benefit of MCGE 1/2- ecpatlwnpnin sinausspc Left: BSS of the best NWP center (Red: ECMWF, Blue: JMA, (up to 9 days), Yellow: NCEP Green: UKMO) Middle: MCGE3 Right: MCGE4 (up to 9 days)

Result -Relative benefit of MCGE 2/2- Reliability is improved in MCGE Reliability Diagram Basin: aus Time window: 4-7 days MCGE4 Forecast probability Obs. Frequency y=x Obs < Fcst

images are taken from wikipedia and bbc.co.uk Hurricane Sandy, Cyclone Phailin and Typhoon Haiyan

Hurricane Sandy (2012) Init: 2012/10/22 12UTC Init: 2012/10/24 12UTC Init: 2012/10/26 12UTC Init: 2012/10/28 12UTC

Cyclone Phailin (2013) Init: 2013/10/09 12UTC Init: 2013/10/05 12UTCInit: 2013/10/07 12UTC Init: 2013/10/11 12UTC

Typhoon Haiyan (2013) Init: 2013/11/02 12UTC Init: 2013/11/06 12UTC Init: 2013/11/04 12UTC Init: 2013/10/31 12UTC

Summary For TC track forecasts, MCGE products provide forecasters with additional information on the forecast uncertainty and increase the level of confidence in the forecast. TC activity predictions are evaluated using TIGGE data from ECMWF, JMA, NCEP and UKMO. Brier Skill Scores (BSSs) in wnp, ecp and atl basins are relatively larger than other basins and BSS is positive in week 2 forecasts in these basins. ECMWF has the largest BSS among the 4 centers in general. MCGE is more skillful than the single-model ensemble in general and the reliability is improved in MCGE. For recent high-impact TCs, Hurricane Sandy, Cyclones Phailin, and Typhoon Haiyan, MCGE predicted the landfall with high-confidence at least 5 days before the landfall.

TIGGE has made a great contribution to examine a new type of ensemble forecasting, that is, multi-center ensemble forecasting. The utility of such ensemble products has been confirmed through various research and forecasting demonstration projects (e.g., NWP-TCEFP and Sever Weather Forecasting Demonstration Project (SWFDP)) under the support of WMO. We should continue such efforts to promote the use of multi-center ensemble products in operational TC forecasting. Future directions 1/3

Future directions 2/3 According to Chan (2010), since Chan et al. (2002) paper, research on the physics of general TC motion has been almost non-existent, which suggests that most scientists are quite content with the current theories of TC motion. In reality, however, significant prediction errors still exist and there are prediction cases where the position error can exceed 1000 km over 3 days. Although the theories of TC motion might have reached a satisfactory level, our knowledge on the sources of prediction errors is still poor. Identifying the sources of large prediction errors and modifying NWP systems accordingly are of great importance to further improve our ability in forecasting TCs. Typhoon Megi initiated at 1200 UTC 25 th Oct Observed track Typhoon Conson initiated at 1200 UTC 12 th Jul MCGE-9 (BOM, CMA, CMC, CPTEC, ECMWF, JMA, KMA, NCEP, UKMO)

Future directions 3/3 Although the performance of TC forecasts has been studied well, the verification samples are usually limited to forecast cases where TCs exist at the initial times. Our knowledge on the performance in forecasting TC genesis and the subsequent track on a medium-range time scale is limited. TC genesis and subsequent track forecasts are important from a perspective of early warnings. Such forecasts would be particularly important for countries located in the lower latitudes.

Thank you for your attention

Supplementary slides

Case Study: Typhoon SON-TINH (2012) Black dots: detected ensemble storms from all ensemble members