Comparatif Modèles GFS et ECMWF

[uranium]

Copier/coller des infos car l'insertion du lien ne fonctionne pas..anyway.

 

Weather Model Différences between the GFS (American) and ECMWF (European)

Posted At : January 6, 2014 10:27 PM | Posted By : Ria Persad

Most energy professionals receive weather information stemming from government models. The two government numerical weather prediction models most commonly employed in the U.S. energy markets are the GFS (American) and the ECMWF (European) ensembles. I receive many questions on the differences between these two models, so after a long literature search, corroborated by my own experience, I compiled this quick reference for energy professionals and weather enthusiasts. These generalities will not work every time, as there will be exceptions, but I have done my best to summarize the most current findings from a number of prominent scientists. If you are aware of new or different information than what is here, please feel free to email me at rpersad@statweather.com and I will update this list. If you are interested in long-range prediction models (beyond the 16-day range of the GFS and ECMWF ensembles), please go to WWW.STATWEATHER.COM to learn more about long-range forecasting.

DIFFERENCE BETWEEN THE GFS (American) AND ECMWF (European) MODELS - Northern Hemisphere

MAXIMUM SKILL RANGE (WHEN MODELS ARE BETTER THAN CLIMATE NORMALS)

GFS: Days 1 to 8 (Summer and Fall) and Days 1 to 9 (Winter and Spring).

ECMWF: One more day than GFS.

Implication: GFS’s accuracy is similar to that of yesterday’s ECMWF model run.

Reason: ECMWF is run at higher resolution, with better observational data and statistical post-processing.

PATTERN SHIFTS

GFS does a good job of predicting pattern shifts 37% of the time.

ECMWF does a good job of predicting pattern shifts 61% of the time.

(“Good job” means anomaly correlation coefficient of 0.9 or better, skillful at synoptic changes.)

After 5 days, the GFS will tend to only detect large scale pattern shifts.

After 6 days, the GFS will tend to only detect the very largest scale global pattern shifts.

ECMWF is half a day ahead at catching synoptic pattern shifts and is more accurate.

WINTER WEATHER SKILL

1 Day Ahead: GFS is better

2 Days Ahead: Average of GFS and ECMWF is best

Days 3 and Beyond: ECMWF has higher skill

YEAR ROUND SKILL (IN GENERAL)

Days 1 to 5: GFS and ECMWF are comparable

Days 6 and on: ECMWF significantly higher skill (gap between the 2 models increases substantially)

SEASONALITY

Greatest model errors for GFS and ECMWF are in Winter, which are double the errors in Summer.

ECMWF is ~10% more accurate than GFS in Summer months and ~20% more accurate in Winter months.

MODEL QUIRKS

GFS
• Once a month, GFS has “dropouts”, where the forecast is a major fail, a major outlier, and very divergent from the ECMWF, which does not have dropouts. GFS scientists are trying to fix dropouts through better initializations. One proposal is to use ECMWF initializations (better quality satellite observations), which would lower dropouts by 90%, but solutions are taking time.
• Can run a cold bias in the Eastern U.S. (but OK in the West).
• Ensembles can be over-confident (low ensemble spreads can still have low forecast accuracy).
• Tends to have a warm bias in the upper troposphere.
• Tends to have a cold bias for afternoon temperatures during warm months.

ECMWF
• Tends to underestimate heavy precipitation events, but GFS has no such bias.
• Tends to have greater model bias; GFS tends to have greater model absolute error.
• Tends to have a cold bias in the stratosphere.
• Tends to have a warm bias for morning temperatures.

DIFFERENCE BETWEEN GFS (OPERATIONAL) and GFS (ENSEMBLE)

GFS Operational is run at a higher resolution (more precision) than the GFS Ensemble. The GFS Ensemble uses as its “control” or “base run” a low resolution (watered-down) version of the GFS Operational run (to save on computing resources), then perturbs or varies this control “base run” slightly to produce the various members of the ensemble. As a result, sometimes the GFS Operational is different from the GFS Ensemble…and can even be more accurate at times. In general, use the ECMWF as an indicator as to whether to lean more heavily to the GFS Operational or the GFS Ensemble.

REFERENCES

Alpert J, Carlis D, Ballish B, Kumar VK. NCEP GFS Forecasts From ECMWF Analysis. American Meteorological Society Supported Meeting; 2012

Breivik Ø, Aarnes OJ, Abdalla S, Bidlot JR. Wind and Wave Extremes over the World Oceans From Very Large Forecast Ensembles. Proceedings of the 13th International Workshop on Wave Hindcasting. Banff, Canada; 2013

Buizza R, Houtekamer P, Pellerin G, Toth Z, Zhu Y, Wei M. A Comparison of the ECMWF, MSC, and NCEP Global Ensemble Prediction Systems. Monthly Weather Review. 2005; 133, 1076–1097

Curry J. U.S. Weather Prediction: Falling Behind. www.judithcurry.com. 2012

Fan Y, Van Den Dool H. Bias Correction and Forecast Skill of NCEP GFS Ensemble Week-1 and Week-2 Precipitation, 2-m Surface Air Temperature, and Soil Moisture Forecasts. Weather and Forecasting. 2011; 26: 355-370

Hotta D, Kalnay E, Ota Y, Miyoshi T. Ensemble Forecast Sensitivity to Observations (EFSO) and Proactive Quality Control. Cooperative Institute for Climate and Satellites, University of Maryland; 2013

Kumar VK, Ballish BA, Jacobs S, Kempisty K, Guan S. NCEP GFS forecast divergence versus ECMWF. 92nd Annual American Meteorological Society Meeting; 2012

Kumar VK, Alpert JC, Carlis DL, Ballish BA. A Sensitivity Study of the November 25, 2011 GFS Dropout Using the GSI Hybrid EnKF vs GSI 3DVAR. 92nd Annual American Meteorological Society Meeting; 2013

La Rue JA. Comparing Numerical Model's Days 3, 4, 5 and 6. National Weather Association; 2001

Pegion P, Whitaker J, Hamill T. A Comparative Evaluation of NCEP and ECMWF Methods for Estimating Model Uncertainty. Meeting of the National Centers for Environmental Prediction; 2012

Petersen D, Brill K, Novak D, Hogsett W, Klein M. WPC Winter Weather Desk Operations and Verification; 2013-2014

Yang F. NCEP/EMC Global Model Experimental Forecast Performance Statistics. http://www.emc.ncep.noaa.gov. 2014

Yang F. Review of GFS Forecast Skills in 2012. IMSG - Environmental Modeling Center, National Centers for Environmental Prediction

Yang F. Review of NCEP GFS Forecast Skills in 2011 and Beyond. The 46th CMOS Congress and the AMS 21th NWP and 25th WAF conferences. Montréal (Canada); 2012

Yang F. GFS Forecast Verification. NEMS/GFS Modeling Summer School 2013. Environmental Modeling Center National Centers for Environmental Prediction. College Park, Maryland

[Mizar]

Merci pour cette étude !!!

 

Concernant le GFS... Comment il fût bien meilleur dans les années passés et ce, juste avant qu'il ne le change pour une nouvelle version. Comment il a été bon pour ne pas dire ''excellant'' surtout lors de l'Hiver de 2007-2008.

 

Depuis ce temps, c'est beaucoup plus difficile.