GFS model frustrations

[Rocky Rascovich]

Greetings! Hoping all is well and your trying not to get too frustrated
over the season thus far.

I have a question concerning the GFS model and for that matter, other models in there consistency and eventual accuracy as of late.

In my 20+ years of chasing and my ever growing thirst for more knowledge
on severe weather forecasting, I'm curious to why it seems this year more
than ever the GFS model has shown an overall dismal performance.

I spoke with Jim Leonard yesterday and he informed me that the model will
undergo some improvement by the end of next month..My question is what
exactly is being done to improve the reliability of the GFS. I know that
no model is perfect, however I've noticed the European has had not a bad
track record lately. What parameters of the GFS should I question as far
as moisture, upper level wind flow, temps, etc? I know that basically its
all a crap shoot past seven days...I know earlier in the year, just about
every anticipated artic outbreak that was to make it to Okla. did'nt
materialize or was greatly modified, my guess was that the GFS could'nt
calculate the lack of snowcover.

Are there any other models besides the Euro which I can at least trust a
tad more? Speaking of the European model...three days running now it
paints a potential nice scenario for the following weekend!

Onless this is a good thread to start on this subject, email me privately
about this.

Thanks! Rocky&family

 

[Nic Wilson]

I'll let others with more numerical model expertise cover what specific initialization problems and computations the GFS struggles with, but I thought I would add that it is well known for the GFS to have a cold bias as seen in the first link below. I consult these NCEP charts regularly to track which medium-range model is handling the set-ups better as of late. The ECMWF has by far been the most consistent this spring, but it even struggled with this latest system. Unfortunately, the ECMWF only has limited free access due to the privatization of the national meteorological organizations in the EU. You have to be in the US meteorological inner-circles to have access to the entire suite of data. Hopefully the planned improvements for the GFS will help because NCEP can't be satisfied with its performance.

500 mb Height Bias
http://wwwt.emc.ncep.noaa.gov/gmb/STATS/ht...raphics/mnz.gif

500 mb Correlation Coefficient
http://wwwt.emc.ncep.noaa.gov/gmb/STATS/html/acz5.html

 

[Glen Romine]

Originally posted by Rocky Rascovich+--><div class='quotetop'>QUOTE(Rocky Rascovich)</div>

...., I'm curious to why it seems this year more than ever the GFS model has shown an overall dismal performance.
[/b]

I don't think I know enough about this model's performance to be able to answer this with authority, but I was abe to find a web site with model statistics that may at least address some of your concerns. If you check the link below, and you can get a feel for the long term trends in performance for a number of global models.

http://www.emc.ncep.noaa.gov/gmb/STATS/htm...l/aczhist6.html

It is a busy plot, so focusing on the top portion only shows model performance of 5 day forecasts of 500 mb heights for the Northern Hemisphere. The GFS model is shown by the solid black line, with the ECMWF model shown by the solid red line. Better performance is when the line is closer to 1. So, we see that historically the ECMWF is performing better than all of the other models, which is summarized by the average 'scores' in the lower right corner, with ECMWF getting a .861, UK model getting .82, and the GFS at .819 coming in third. If you look at the yearly trends, there is typically a dip in performance during the summer months, and improved performance during the winter. I don't see any obvious difference between performance this year and the previous, but the longer term trend is for slow improvement.

It is probably also worth noting that the model performance varies based on which cycle you are looking at. 00Z runs are better for all models, but even the 12Z ECMWF performs better than the 00Z GFS. You can see that here:

http://www.emc.ncep.noaa.gov/gmb/STATS/htm...tml4x/acz5.html

Originally posted by Rocky Rascovich@

..My question is what exactly is being done to improve the reliability of the GFS.

There is a plan to increase the model resolution beyond 84 hours - such that it runs at the same resolution (~35 km) out to 180 hours. So, some improvement in the representation of waves within this period will be possible. What is believed by some is that the far more sophisticated method used for initializing the European model is what gives it better accuracy, but there are so many other possible factors that it is difficult to know for sure - but regardless their system is generally better, particularly at longer range. Also, as Nic noted, the GFS has a cold bias, which partially explains why the forecast cold blasts were muted as you noted.

<!--QuoteBegin-Rocky Rascovich

Are there any other models besides the Euro which I can at least trust a
tad more?

As show above, the UK model is overall comparable in performance to the GFS, so may be worth a look in addition to the ECMWF. I would suggest though that you try reading the diagnostic model discussion of the long range - as these folks watch the trends in these models on a regular basis, and provide their expert guidance into which model appears to have a better handle on the forecast for the US region. Unfortunately, it is in somewhat cryptic language that takes some time to get used to with all the acronyms and met terminology.

Medium range:
http://www.nws.noaa.gov/data/WNH/PREEPD

Shorter range:
http://www.nws.noaa.gov/data/WNH/PMDHMD

More GFS model statistics:
http://www.emc.ncep.noaa.gov/gmb/STATS/STATS.html

Hope this helps.

Glen

 

[Darren Addy]

I'm just a newbie and you have a ton of more experience and knowledge on the subject than I do, but I'm not sure how one can do what you are asking - especially as far out as you are asking to do it.

These are numerical models based on a multiplicity of variables. Any tiny change to one of the variables can lead to massive differences as you step into the future (the farther you project, the greater the differences). I can understand how models can be tweaked over time and improvements can be made in the overall accuracy of the model. I don't see how an individual can diagnose a model from where we sit, make corrections and see a better result to our own 7 day forecast.

If I'm not mistaken, models are calculated using supercomputers. I have no idea how an individual human could apply and calculate a "correction".

In some ways, we seem quite spoiled by the advances that science has made in forecasting and tend to underestimate the task. What was said by Gedzelman in his excellent (1980) book "The Science and Wonders of the Atmosphere" still holds pretty much true:

\"A truly staggering effort has gone into improving our weather forecasts, but somehow the atmosphere yields its secrets only with great reluctance. The improvement in weather forecasts over the last decade has been disproportionately small compared to the vast improvements in observations, communications, computer power, mathematical techniques, and understanding of the physics of the atmosphere. . . . if we make a mistake in measuring the wind or temperature, and the like, or if we leave out some seemingly unimportant weather feature. . . or even if we round off the computations, then the mistake will eventually grow so large that our prediction will be worthless. Experience and theory indicate that small errors tend to double within two or three days. \"

Darren Addy
Kearney, NE

 

[mikegeukes]

Another model to look at is the DGEX:

Operational Links:
http://www.wxcaster.com/conus_offhr_models.htm
http://www.wxcaster.com/dgex.htm
http://wwwt.emc.ncep.noaa.gov/mmb/mmbpll/dgexhome.ops/

Other Links:
http://rammb.cira.colostate.edu/visit/dgex.html
http://www.nws.noaa.gov/ost/ifps_sst/DGEX.html

Mike

 

[Justin Turcotte]

EFFECTIVE MAY 31 2005...BEGINNING WITH THE 1200 COORDINATED
UNIVERSAL TIME /UTC/ RUN...THE NATIONAL CENTERS FOR ENVIRONMENTAL PREDICTION /NCEP/ WILL UPGRADE THE GLOBAL FORECAST SYSTEM /GFS/ TO IMPROVE ITS ACCURACY AND PRODUCTS.

THIS UPGRADE INVOLVES INCREMENTAL IMPROVEMENTS TO THE ANALYSIS AND FORECAST OF THE GFS. DETAILS OF THESE CHANGES INCLUDE:

ANALYSIS CHANGES:

- INCREASE HORIZONTAL RESOLUTION TO T382.
- ASSIMILATE AQUA AIRS AND AMSU-A BRIGHTNESS TEMPERATURES.
- MODIFY THINNING ALGORITHM FOR BRIGHTNESS TEMPERATURES.
- ADD SURFACE MICROWAVE EMISSIVITY MODEL OVER SNOW AND ICE.
- MODIFY BRIGHTNESS TEMPERATURE QUALITY CONTROL ALGORITHM
TO CHECK FOR THE PRESENCE OF CLOUDS IN IR FIELDS OF VIEW.

FORECAST MODEL CHANGES:

- HORIZONTAL RESOLUTION INCREASED FROM T254 (~55 KM) 64 LAYERS TO T382 (~35KM) 64 LAYERS FOR FORECAST HOURS 0-180.
- HORIZONTAL RESOLUTION INCREASED FROM T126 (~100 KM) 42 LAYERS TO T190 (~70 KM) 64 LAYERS FOR FORECAST HOURS GREATER
THAN 180.
- ADDITION OF THE NOAH LAND SURFACE MODEL WITH 4 SOIL LAYERS.
- NEW SEA ICE MODEL.
- REDUCED VERTICAL DIFFUSION.
- ENHANCED MOUNTAIN BLOCKING.

THIS UPGRADE WILL NOT CHANGE GFS PRODUCTS SENT OVER NOAAPORT. NCEP CENTRAL OPERATIONS IS CURRENTLY RUNNING A REAL-TIME PARALLEL RUN OF THE T382 GFS. RESULTS FROM THESE PARALLEL RUNS CAN BE VIEWED AT /USE LOWER CASE LETTERS/:

HTTP://WWW.NCO.NCEP.NOAA.GOV/PMB/NWPARA/ANALYSIS/

 

[mikegeukes]

Biases of Forecast Models:
http://www.crh.noaa.gov/lmk/soo/docu/models.htm
http://www.hpc.ncep.noaa.gov/mdlbias/biastext.html
http://www.hpc.ncep.noaa.gov/mdlbias/index.html

Note: Some of the stuff, may be outdated.

Mike