Presentation on theme: "1 March 2007 NAEFS Upgrade Yuejian Zhu, Richard Wobus, Mozheng Wei, Bo Cui and Zoltan Toth Environmental Modeling Center NOAA/NWS/NCEP Acknowledgements:"— Presentation transcript:
1 March 2007 NAEFS Upgrade Yuejian Zhu, Richard Wobus, Mozheng Wei, Bo Cui and Zoltan Toth Environmental Modeling Center NOAA/NWS/NCEP Acknowledgements: Qingfu Liu, DingChen Hou, Mark Iredell and Stephen Lord EMC Luke Lin, David Michaud, Joey Carr and Brent Gordon NCO
2 Planned Changes - Summary 1.Increasing NAEFS (NCEP/GEFS) membership 80 perturbations in cycling (see schematic plot) Replaced operational 56 perturbations in cycling 20 perturbed long forecasts (16-d) in each cycle Replaced operational 14 long forecasts in each cycle 2.NAEFS (NCEP/GEFS) post-product Post-process 20 ensemble forecasts instead of operational 14 ensemble forecasts for: Bias correction Anomaly forecast
3 T00Z 56m T06Z 56m T18Z 56m T12Z 56m 6hrsNext T00Z Re-scaling Up to 16-d T00Z 80m 6hrs T06Z 80m T12Z 80m T18Z 80m Up to 16-d Re-scaling 6 hours breeding cycle Production 6 hours breeding cycle Planned Change Next T00Z Re-scaling
4 CurrentPlan Model GFSGFS (current) Initial uncertainty ETR Model uncertainty None Tropical storm Relocationsame Daily frequency 00,06,12 and 18UTCsame Hi-re control (GFS) T382L64 (d0-d7.5) T190L64 (d7.5-d16) same Low-re control (ensemble control) T126L28 (d0-d16) 00,0612 and 18UTC same Perturbed members 14 for each cycle20 for each cycle Forecast length16 days (384 hours)same ImplementationMay 30 th 2006March 27 th 2007 GEFS configurations
5 Early study indicated: (by Mozheng Wei) ET20/80 give most skillful probabilstic forecast when compared to 10m operation and 10m ET
6 Planned Changes 1 Increasing NAEFS (NCEP/GEFS) membership. –From current 14 increasing to 20 members per cycle. Tuning initial perturbations. Using 80 cold start initial perturbations (on Jan. 25) –From 24-h forecasts and many dates. –To have large spread of sampling –This change is intended to improve ensemble based probabilistic forecast over all and to support NAEFS (North American Ensemble Forecast System) project. –Results: Improving probabilistic skills. Not much improvement for ensemble mean (expected).
7 ENS_s – operation ENS_b – large sprd ENS_d – small sprd Doted lines – 1-day spread Day-5 PAC scores Doted lines-spread Cold start from 00UTC Nov Large initial spread has more skill
8 Different period-more runs for large initial spread (cold start) ENS_s – operation ENS_c – large sprd Cold start from 00UTC Nov. 7 th 2007 NH 500hPa ROC Tropical 500hPa ROC SH 500hPa ROC
9 20m has more prob skills than 14m ENS_s – operation ENS_b – large sprd ENS_d – small sprd NH 500hPa RPSS SH 500hPa ROC Tropical 500hPa ROC
10 Retrospective experiments/NCO real time parallel Retrospective experiments. –Summer period: August 22- September (Hurricane season). –Winter period: November 7-December –20 ensemble members. –Using current GFS. NCO real time parallel. –Starting since January Full evaluations (follow up above experiments). –Summer: –Winter: –NCO real time: Expect to implement (March ).
15 NCO real time parallel NH 500hPa geopotential height NCO real time parallel NH 2 meter temperature
16 Conclusions of the 20m experiments Ensemble mean: –No significant different (do not expect). –No degradation (not worse – very important) Ensemble distribution (probabilistic forecast): –Significant improvement from 20 ensemble members. For ROC, EV, Reliability. Not for RPSS BSS. Initial perturbation for cold start: –Large cold start initial spread is (JIFed) From backup perturbation based on 24h forecasts in many dates. Giving sufficient spread in ET. Quickly scaling down to the expected level (1-2 days). Able to catch up large uncertainty cases. Degrading the skills a little for short lead-time in the first few cycles. –Small cold start initial spread is From backup perturbation based on 6h forecasts in one day. No degradation at early cycling. Not able to catch up large uncertainty system for longer lead-time. Because error grows very slowly.
17 Planned Changes 2 Details for Slide 2 NAEFS (NCEP/GEFS) Post Products –Bias corrected members of joint MSC-NCEP ensemble NCEP –Current: 14 (perturbed forecasts) + 1 (control forecast) = 15 forecasts at each cycle –Plan: 20 (perturbed forecasts) + 1 (control forecast) = 21 forecasts at each cycle MSC –Same: 16 (multi-model, multi-physics forecasts) + 1 (control forecast) = 17 forecasts at 00UTC and 12UTC cycle only 35 of NAEFS variables Bias correction against each center’s own operational analysis –Weights for each member for creating joint ensemble (equal weights right now) Weights don’t depend on the variables Weights depend on geographical location (low precision packing) Weights depend on the lead time –Climate anomaly percentiles for each member NCEP –Current: 14 (perturbed forecasts) + 1 (control forecast) = 15 forecasts at each cycle –Plan: 20 (perturbed forecasts) + 1 (control forecast) = 21 forecasts at each cycle MSC –Same: 16 (multi-model, multi-physics forecasts) + 1 (control forecast) = 17 forecast at 00UTC and 12UTC cycle only 19 of NAEFS variables
18 NH Mean Sea Level Pressure NH 2 Meter Temperature NH 500hPa Height NCO real time parallel: Accumulated bias (absolute values) for past one month Black – raw forecasts Red – bias corrected forecasts
19 Summary of Subjective Analysis CPC –We do not see any cause for concern that an increase in the ensemble will have any negative impact on the CPC operational products. Given that, we fully support this implementation to increase the ensemble size. HPC –There was little improvement over the standard GFS ensemble forecast. This subjective evaluation was noted on both mass fields and model QPF. –HOC recommends implementation OPC –Little difference noted between operational and parallel data sets. –OPC recommends implementation AWC –While AWC supports the idea to increase the number of members in the ensemble to represent better the envelope of possibilities for various forecast parameters, we are unable to provide a more rigorous assessment due to the lack of resources to ingest the data and routinely view the available fields