UMAC data callpage 1 of 16Global Ensemble Forecast System - GEFS Global Ensemble Forecast System Yuejian Zhu Ensemble Team Leader, Environmental Modeling.

UMAC data call page 1 of 16Global Ensemble Forecast System - GEFS

Global Ensemble Forecast System

Yuejian ZhuEnsemble Team Leader, Environmental Modeling CenterNOAA / NWS / NCEP

[email protected]

UMAC data call page 2 of 16Global Ensemble Forecast System - GEFS

Wh

ole

A

tmo

sph

ere

M

od

el

Unified Global Model

Application=

Ensemble+

Reanalysis+

Reforecast

NGGPSUnified Global Coupled Model

“GFS” “GEFS” “CFS”

Actionable weather

Week 1 through 4-6

Seasonal & annual

1 y 2 y 4 y Update cycle

3 y 20-25 y 1979 - present Reanalysis

6h 6-24h ??? cycling

WCOSS WCOSS WCOSS ? where

UMAC data call page 3 of 16Global Ensemble Forecast System - GEFS

Current Status of GEFS

• Initial analysis/perturbations– GFS/GSI analysis– BV-ETR (Breeding Vector – Ensemble Transform with Rescaling)

• Toth, Z., and E. Kalnay, 1997: Ensemble forecasting at NCEP and the breeding method. Mon. Wea. Rev., 125, 3297–3319.

• Wei, M., Z. Toth, R. Wobus, and Y. Zhu, 2008: "Initial Perturbations Based on the Ensemble Transform (ET) Technique in the NCEP Global Operational Forecast System" Tellus 59A, 62-79

• Model– GFS model (see GFS reforences)

• Relocation– TS relocation since 2005

• Liu, Q., S. J. Lord, N. Surgi, Y. Zhu, R. Wobus, Z. Toth and T. Marchok, 2006: Hurricane Relocation in Global Ensemble Forecast System, Preprints, 27th Conf. on Hurricanes and Tropical Meteorology, Monterey, CA, Amer. Meteor. Soc., P5.13.

• Stochastic physics– STTP (stochastic total tendency perturbation) since 2010

• Hou, D., Z. Toth, Y. Zhu, W. Yang and R. Wobus, 2012: A Stochastic Total Tendency Perturbation Scheme Representing Model- Related Uncertainties in the NCEP Global Ensemble Forecast System, Submitted to Tellus-A (Dec. 2010)

• Configuration– 20+1 members, four times per day, out to 16 days.

UMAC data call page 4 of 16Global Ensemble Forecast System - GEFS

GEFS Upgrade (Q4FY15)

V10.0.0 (OPR) V11.0.0 (PARA)

GFS Model Euler, 2012 Semi-Lagrangian, 2015

Resolution 0-192 h T254 (52km) L42 (hybrid) TL574 (34km) L64 (hybrid)

Resolution 192-384 h T190 (70km) L42 (hybrid) TL382 (52km) L64 (hybrid)

Computational Cost105 nodes (Integration + post-process)

310 nodes 1st segment250 nodes 2nd segment

Execution time 55 min35 min 1st segment30 min 2nd segment

Output resolution 1O x 1O 0.5O x 0.5O for 0-8 days 1O x 1O the rest

Output frequency 6h 3h the first 8 days; 6h the rest

UMAC data call page 5 of 16Global Ensemble Forecast System - GEFS

GEFS Upgrade (Q4FY15)

• Moving from BV-ETR approach to EnKF– A major scientific shift

• Unification DA and Ensemble Generation– Direct link to the hybrid 3D-Var EnKF DA system

• Perturbations are 6h forecasts EnKF with adjustments:– Tropical Storm Relocation– Centering of the perturbations on the ensemble control analysis

• Stochastic perturbation (STTP) upgrade– Fine-tune amplitude for changes in model and perturbation

method– Turn off surface pressure perturbations for tropical

• to reduce the spread growing of geopotential height

UMAC data call page 6 of 16Global Ensemble Forecast System - GEFS

GEFS Upgrade (Q4FY15)

6

685 cases

Northern Hemisphere 500hPa Geopotential

Height

77%

9.52d

9.20d

75%

Summary from nearly 2 years retrospective runs:

1. GEFSv11 improves forecast skill (60%) about 8 hours

2. GEFSv11 enhances AC score from 75% to 77% for day-7 forecast

T2m for Northern American

With bias correction, GEFSv11 reduces

RMS error for all leads

Summer precipitation for CONUS

Improve reliability

UMAC data call page 7 of 16Global Ensemble Forecast System - GEFS

00UTC

Opr: T254L42(55km)

Para: T574L64(33km)

06UTC

Thick blue: ensemble

mean

Bimodality?

Sandy

Ini: 20121022; Fcst: 8-day

UMAC data call page 8 of 16Global Ensemble Forecast System - GEFS

GEFS Short-term Plan

• Initial perturbations– Use EnKF analysis directly, not 6-hour forecasts

• Model– GFS model with tuning parameters for lower resolution

• Relocation– TS relocation with tuning TS amplitude of perturbations

• Stochastic physics– Introduce well-established stochastic tendency perturbations

• SKEB, SPPT and SHUM

• Configurations– 20+1 members, four times per day, out to 16 days.– Continue to investigate the benefits from different ensemble

membership (size)

UMAC data call page 9 of 16Global Ensemble Forecast System - GEFS

GEFS Long-term Plan

• Toward unified global system– Full coupled system

• Uniform resolution– Consider uniform resolution after introduce high resolution global

ensemble for weather (out to 7-10 days)

• Ensemble size (or membership)– Continue to investigate the benefits of different ensemble membership

• Cover week-1, week-2, week 3&4 and monthly forecast– See slides (#10-15) for GEFS extend forecast– Introduce land surface uncertainties (perturbations)– Introduce ocean uncertainties (perturbations)

• Upgrade every 2 years• 20-25 years reanalysis and GEFS reforecasts for every

upgrade.

UMAC data call page 10 of 16Global Ensemble Forecast System - GEFS

Global Ensemble Forecast System

Extend to 35 days (experiments)

Yuejian ZhuEnsemble Team Leader, Environmental Modeling CenterNOAA / NWS / NCEP

[email protected]

UMAC data call page 11 of 16Global Ensemble Forecast System - GEFS

GEFS Extend Forecast• Short term plan (1-2 years)

– Atmospheric model only– Introduce SST forcing to assimilate MJO and others– Ocean and land surface perturbations

• In testing – may not include in short-term implementation plan– Stochastic physics

• Will use the same stochastic physic scheme as GEFS 0-16 day forecast.– Configurations

• 20+1 members, twice per week, from 16 days to 35 days.

• Long term plan (3-5 years)– Unified global forecast system– Fully coupled system

• Atmosphere-Ocean-Land-Ice-Wave-Aerosol – Initial perturbations

• EnKF DA for atmosphere and ocean– Model uncertainties

• Stochastic physics– Configurations

• 20+1 members (?), uniform resolutions, once per day, out to 45 days

UMAC data call page 12 of 16Global Ensemble Forecast System - GEFS

GEFS Extend Forecast

• Experiments– Period: Sept. 1st 2013 – Feb. 28th 2014 (6 months)

• Weak MJO period– GEFS v11 (2015)

• 20+1 members, 33km for day 0-8, 50km for day 8-16, 70km for day 16-35– Three different configurations have been tested

• CTL – default GFS/GEFS setting – start with RTG SST analysis, relaxed to climatology (90 days e-folding)

• RTG – using RTG analysis forcing GEFS every 24 hours (AMIP)• CFS – using CFSv2 predictive SST forcing GEFS every 24 hours)

– Two additional configurations are on going• CFS* - using CFSv2 predictive SST (bias corrected) anomaly forcing GEFS

every 24 hours• Full coupling with MOM4 (or MOM6)

– Evaluations• Tropical – MJO, precipitation• Extratropical – 500hPa height, surface temperature and precipitation• Extreme weather patterns – Blocking, polar vortex

UMAC data call page 13 of 16Global Ensemble Forecast System - GEFS

1. RTG-SST (AMIP) run has good MJO forecast skills2. CTL-SST runs has less MJO forecast skills for first 2 weeks

WH-MJO fcst skills for GEFS(6 months average)

UMAC data call page 14 of 16Global Ensemble Forecast System - GEFS

Evolution of error difference (500hPa height)5-day running mean (09/01/2013 – 2/28/2014)

UMAC data call page 15 of 16Global Ensemble Forecast System - GEFS

Lea

d d

ay =

19

Lea

d d

ay =

23

RMSE SPREAD

RMS error and spread distributions for 500hPa height of CTL-SST20130901 – 20140228 (6 months) 5-day running mean

For different lead times

Uncertainty is Well represented

Over dispersion ?

UMAC data call page 16 of 16Global Ensemble Forecast System - GEFS

GEFS Reforecast

• Consider latest 20-25 years– Goes with every model upgrade– Goes with new re-analysis– Better to start from 1998 (AMUSU data started)

• Configurations– Option 1: use “white paper” recommendation– Option 2: consider NWC’s request (current impossible)

• Real-time or retrospective– Resource dependence

• Benefits– National centers: WPC, CPC, OPC, NWC– Regions: WFO