Mio Matsueda (University of Oxford) Tetsuo Nakazawa (WMO)

Transcription

1 GIFS-TIGGE WG (12-14 June, 2013) Early warning products for extreme weather events using operational medium-range ensemble forecasts Mio Matsueda (University of Oxford) Tetsuo Nakazawa (WMO) with thanks to Richard Swinbank and Ken Mylne (UKMO)

2 The TIGGE Museum 5,300 visitors since Apr 2011 or Google tigge mio Updated everyday with a 2.5-day delay Real time products spaghetti plots MJO forecast blocking forecast early warning of severe weather EPS meteogram forecast skills model biases spaghetti plots MJO forecast blocking forecast early warning of severe weather Products are available for past forecast cases after October 2006.

3 Early warning products for high impact weather using TIGGE extreme events heavy rain strong winds high/low temp. thresholds for prob. forecasts(90, 95 &99) 14 regions (including 5 SWFDP regions) past forecast cases with observation (Oct current) forecast lead time (up to +15 days)

4 Early warning products for high impact weather using TIGGE 11 regions (no SWFDP regions) Centres (grand or single-centre ensemble) forecast lead time (up to +11 days) No past forecast cases (the latest forecasts only) Many accesses from developing countries!

5 the occurrence probability for grand ensemble Example: ensemble forecasts of surface temperature at Exeter ECMWF's 99th percentile ECMWF EPS (51 mem.) 5 T2m Climatological percentiles derived from NWP models differ from each other. Each model's climatological percentile value is used for a definition of extreme event. T2m 15 (=5+6+4) members predict 305K 310K 315K extreme NCEP's 99th percentile NCEP EPS (21 mem.) 6 305K 308K 310K 315K extreme UKMO's 99th percentile UKMO EPS (24 mem.) 305K 4 310K 312K 315K T2m extreme a higher value than each model's climatological 99th percentiles (310, 308, and 312K). Then, occurrence probability of extreme high temperature is defined as 15.6%(=15/( )).

6 How climatological PDFs are estimated from the TIGGE data A climatological PDF used here is: calculated for each EPS using TIGGE data (all members in each EPS) during October 2006 to January 2011 defined at each grid point for each calender day in each lead time with the 31-day time window. Example: A climatological pdf for 72-hr ECMWF ensemble forecast verified on 16th January is made from all the 72-hr ECMWF forecasts (members) verified on 1st - 31st January in 2007 to day time window 2007 ensemble forecast No bias correction. +72hr +48hr x 5 years +24hr January Total number of samples ( Initial date of forecast ) for ECMWF: 31days x 5yrs x 51mems = 7905

7 Examples of early warning for extreme weather events Russian heatwave (JJA 2010) Pakistan floods (July 2010) Hurricane Irene (August 2011) Hurricane Sandy (October 2012)

8 Russian heatwave (JJA 2010) 4th August Wildfires brought heavy smog The heatwave killed at least 15,000 people, and brought wildfires, smoginduced health injury, and huge economic loss. Moscow New maximum record of 39! Moscow 1,600 drowning deaths! China!? Not flu!

9 Russian heatwave (JJA 2010) + 5-day forecast

10 Pakistan floods (July 2010) ECMWF Newsletter (No. 125, 2010)

11 Pakistan floods (July 2010) + 7-day forecast

12 Hurricane Irene (August 2011) 12Z 28 Aug. Irene (27 August, 2011) From NASA's web Landed! (12Z 27 Aug.) Central Grand Station, NY, USA (27 August, 2011) Kill Devil Hills, North Carolina, USA (25 August, 2011)

13 Hurricane Irene (August 2011) + 5-day forecast

14 Hurricane Sandy (October 2012) Early warning (28 October, 2012) Hurricane Sandy (28 October, 2012) New England, USA (30 October, 2012) Queens, New York City, USA (30 October, 2012) Staten Island, New York City, USA (1 November, 2012)

15 Cyclone Sandy (October 2012) + 6-day forecast

16 Verification reliability diagram for precipitation Obs: GSMaP +3days +5days +9days +15days

17 Verification reliability diagram for high temperature Obs: ERA-Interim +3days +5days +9days +15days

18 Verification reliability diagram for wind speeds Obs: ERA-Interim +3days +5days +9days +15days

19 Summary The ensemble-based early warning products for extreme weather events are introduced. The early warning products are based on operational medium-range ensemble forecasts from four of the leading global NWP centres: ECMWF, JMA, NCEP, and UKMO. The ECMWF EPS shows the best forecast reliability in the single-centre ensembles. The construction of a grand ensemble by combining four single-centre ensembles can improve the forecast reliability regarding probabilistic forecasts of extreme events, up to a lead time of hr. The grand ensemble can provide more reliable forecasts than single-centre ensembles. This results from a fact that the best performing ensemble is case dependant.