Global Ensemble Implementation

Zoltan Toth, Richard Wobus and Yuejian Zhu

There is an main change for ensemble forecast started from 29 April 2003 at T12Z, which is an ensemble mask (spread/deviation) rescaled to 90% (Northern Hemisphere), 40% (Southern Hemisphere), and 50% (Tropical) from former operational mask levels. In order to make this implementation, the experiments have been done to test the skill improvement by extending high resolution (will be next implementation) and mask rescaling. The selected experiments results will be attached which includes RMS errors, PACs, Brier skill scores, Economic values, ROC areas, Hurrican track verification, etc. We like to thank the helps from Tim Marchok and Lacey Holland to verify hurrican tracks. Please send your message to either one of Zoltan.Toth@noaa.gov, Richard.Wobus@noaa.gov and Yuejian.Zhu@noaa.gov if you have any question.

Experiments Setup:

Period: (1) 20020825-20020930 (37 days)

(2) 20021201-20030114 (45 days)

Test: breeding cycling, model resolutions, smoothing, mask (spread) rescaling.

List two best experiments of them:

1). s: operational global ensemble

2). i: reduceing initial spread by factors: 0.8(NH), 0.3(SH),0.5(TR)

3). j: reduceing intial spread by factors: 0.9(NH),0.4(SH),0.6(TR)

Following experiments:

Period: (3) 20020825-20021009 (46 days)

Test: rescaling masks only.

1). o: operational global ensemble

2). p: reduceing initial spread by factors: 0.9(NH), 0.4(SH),0.5(TR)

3). q: reduceing intial spread by factors: 0.9(NH), 0.4(SH), 0.42(TR)

Selected results from above experiments

Probabilistic evaluation

1.NH Economic Values (1)

2.NH Economic Values (2)

3.NH Brier Skill Scores (1)

4.NH Brier Skill Scores (2)

5.SH Economic Values (1)

6.SH Economic Values (2)

7.SH ROC areas (1)

8.SH ROC areas (2)

9.SH Brier Skill Scores (1)

10.SH Brier Skill Scores (2)

Ensemble mean evaluation

1.NH PAC (Die-off) (1)

2.NH PAC (Die-off) (2)

3.NH RMS (Err-grow) (1)

4.NH RMS (Err-grow) (2)

5.NH RMS (Err-grow) (3)

6.SH PAC (Die-off) (1)

7.SH PAC (Die-off) (2)

8.SH RMS (Err-grow) (1)

9.SH RMS (Err-grow) (2)

10.SH RMS (Err-grow) (3)

Hurrican tracks evaluation

1.Figure (3)

RMS errors day by day comparison

1.Table (2)