Products and Forecasts - Examples and experiences Yuejian Zhu and Hong Guan Ensemble Team...

Products and Forecasts

- Examples and experiences

Yuejian Zhu and Hong Guan

Ensemble TeamEMC/NCEP/NWS/NOAA

Presents for NWP Forecast Training ClassMarch 31, 2015, Fuzhou, Fujian, China

Background• NDGD

– National Digit Guidance Database• Guidance's are produced from NWP output, post process and et al.

• NDFD– National Digit Forecast Database

• Forecasts are issued from WFO forecasters based on all different guidance

• Coverage and resolution– CONUS – 2.5km– Alaska – 3km– Hawaii – 2.5km– Puerto Rico – 2.5km– Guam – 2.5km

Elements Projections

1 12-hour Probability of Precipitation (PoP12) Every 12 hours, out to 168 hours

2 Apparent Temperature Every 3 hours out to 72 hoursEvery 6 hours out to 168 hours

3 Dew Point Every 3 hours out to 72 hoursEvery 6 hours out to 168 hours

4 Hazards Every hour out to 72 hoursEvery 6 hours out to 120 hours

5 Maximum Temperature Every 24 hours out to 168 hours

6 Minimum Temperature Every 24 hours out to 168 hours

7 Quantitative Precipitation Amount Every 6 hours out to 72 hours

8 Relative Humidity Every 3 hours out to 72 hoursEvery 6 hours out to 168 hours

9 Significant Wave Height Every 6 hours out to 120 hours

10 Sky Cover Every 3 hours out to 72 hoursEvery 6 hours out to 168 hours

11 Snow Amount Every 6 hours out to 48 hours

12 Temperature Every 3 hours out to 72 hoursEvery 6 hours out to 168 hours

13 Weather Every 3 hours out to 72 hoursEvery 6 hours out to 168 hours

14 Wind Direction Every 3 hours out to 72 hoursEvery 6 hours out to 168 hours

15 Wind Gust Every 3 hours out to 72 hours

16 Wind Speed Every 3 hours out to 72 hoursEvery 6 hours out to 168 hours

US NWS NDFD forecast elements and projections (2014)

High Impact Weather and Extreme Forecast

Hurricane Katrina (2005)

Hurricane Katrina (2005)

Early prediction:Friday – August 26NHC’s prediction

ECMWF ensemble forecast: Strike

probability from Friday – August 26

0-120 hours120km radius

Typhoon Morakot (2009)

Courtesy of Bill Kuo

Courtesy of Bill Kuo

10

Resolution makes difference for Typhoon Morakot

Ini: 2009080600

T126 ensemble

T190 ensemble

Ini: 2009080700

T126 ensemble

T190 ensemble

Most models do notmake right forecasts

Typhoon Megi (2010)

2010101512 2010101612

2010101712 2010101812

1. NCEP2. CMC3. ECMWF4. 3-ENS

Derecho (2012)

Date(s) June 29–30, 2012Duration ~18 hours (10:00 AM-4:00 AM)Track length ~800 MilesPeak wind gust 91 mph (Fort Wayne, Indiana)Largest hail 2.75 Inches (Bismarck, Illinois)Fatalities 28 fatalitiesDamage $2.9 billion[1]Areas affected United States Midwest,

United States Mid-Atlantic

A moderate risk area was issued by the SPC for the areas downstream of the derecho. It was later expanded eastward.

Derecho (2012)

Sandy Case Study for NEXT GEFSPeriod: 10/22 – 10/28/2012

Named: 10/23/2012

Yuejian ZhuEMC/NCEP

September 15 2014

Acknolegements: Dingchen Hou, Xiaqiong Zhou and Jiayi Peng

00UTC

20121022 (8 days)

Opr: T254L42(55km)

Para: T574L64(33km)

06UTC

Thick blue: ensemble

mean

Bimodality?

Red arrow means good

forecast

12UTC

20121022 (7.5 days)

Opr: T254L42(55km)

Para: T574L64(33km)

18UTC

Thick blue: ensemble

mean

Bimodality?

Opr: T254L42(55km)

Para: T574L64(33km)

20121026 (4 days)

00UTC

06UTC

Opr: T254L42(55km)

Para: T574L64(33km)

20121028 (2 days)

00UTC

06UTC

Short Summary for Sandy case• Higher resolution and new model improve the

forecast skill for most lead-time, especially for longer lead-time (day 7-8).

• Higher resolution and new model have excellent predictability around 8-days

• Bimodality of forecast tracks is clearly for early lead-time – around 30-32N

• Very good forecasts for short lead-time (less than 4-5 days) of both production and parallel

• Problem/concern:– Forecast inconsistency from cycle to cycle since initial

condition changes, especially for Oct. 23 - 24

March 6 2013 – winter storm “Saturn”Heavy, wet late season snowstorm likely to paste D.C., Mid-Atlantic Wednesday

Tuesday night to Wednesday night for potential of at least 5 inches of snow for entire region*

From 11:47 AMUpdate: 1:55PM

FEDERAL OFFICES in the Washington, DC, area are CLOSED. Emergency and telework-ready employees (employees with approved individual telework agreements) required to work must follow their agency’s policies, including their approved individual telework agreement

All Howard County public schools and offices are closed today, Wednesday, March 6, 2013. All evening activities in schools, both school-sponsored and community-sponsored, are canceled. This includes high school athletic practices and games.

Review of 2015 NE Blizzard01/26 12UTC – 01/27 12UTC 2015

Yuejian ZhuEMC/NCEP/NWS

Jan. 27 2015

Acknowledgements: Hong Guan, Yan Luo and Xiaxiong Zhou

ESRL’s EFI maps

WFO forecast (48 hours accumulation)

CCPA 24 hours accumulation (mm)ENDING 12 UTC 20150127 ENDING 12 UTC 20150128

Ensemble Forecast - Uncertainty

Small and large uncertainty.

1 day (large uncertainty) = 4 days (control) = 10-13 days (small uncertainty)

Toth and Zhu (2001)

Ensemble Based Probabilistic Quantitative Precipitation Forecast (PQPF)Valid: 2015012612 – 2015012712 Amount 24hr >25.4mm (1 inch)

PROD PARA

Initial Time

2015012400

2015012500

2015012600

NCEP GEFSPROD/PARA

Forecast difference

Ensemble Based Probabilistic Quantitative Precipitation Forecast (PQPF)Valid: 2015012612 – 2015012712 Amount 24hr >25.4mm (1 inch)

Initial Time

2015012100

2015012300

2015012200

PROD ECMWFPARA

Ensemble Based Probabilistic Quantitative Precipitation Forecast (PQPF)Valid: 2015012612 – 2015012712 Amount 24hr >25.4mm (1 inch)

PROD PARA

Initial Time

2015012400

2015012500

2015012600

ECMWF

Here is the differenceEC’s ensemble predicted 50%+ chance for 1 inch prcip over

NY city

NCEP GEFS failed on Oct. 3-4 2014

• First fail – 18UTC Oct. 3rd 2014 – Friday– No product for GEFS, SREF, Wave and NAEFS

• Second fail – 12UTC Oct. 4th 2014 – Saturday– Delayed 3hrs for all products and downstream

• Where is the problem?– Failed on first integration time step

• Point to short wave radiation

– In fact, the problem is longwave radiation code• secdiff(j) = a0(j) + a1(j) * exp( a2(j)*pwvcm) < 0.0• It was never happened before

– Protect – cost?• Morning of Oct. 10th 2014 – Friday

– Announced we found a problem

The art of good forecasting

•

经验预报 Meteorologist

最好的预报 Modified NWP forecast

A

A2 持续性预报 Persistence最好的数值预报

World Best NWP