Discussion about models,patterns and seasonal forecasts

[cycloneye]

Weather will NEVER be able to be predicted more than 2 weeks in advance as the equations of the atmosphere cannnot be solved precisely by the models, they are approximated. By the end of 14 days, the errors have simply grown too large

which is why pattern recognition is ALMOST ALWAYS more effective in meadium range forecasting, than just taking numerical model output at face value. and analoging is much more effective in seasonal forecasting as compared to staring at a long-range climate model. case in point, the AGCM's preformance prior to the 2001-02 winter. which was laughable at best.

the GFS (AVN/MRF) is a better example of the inconsistencies in meadium range model preformance. beyond 240hrs, the output is for the most part useless.

And also FLGUY worse than the medium range models are the long range model data that are more useless because for the same reasons you said.

[Stormsfury]

Weather will NEVER be able to be predicted more than 2 weeks in advance as the equations of the atmosphere cannnot be solved precisely by the models, they are approximated. By the end of 14 days, the errors have simply grown too large

which is why pattern recognition is ALMOST ALWAYS more effective in meadium range forecasting, than just taking numerical model output at face value. and analoging is much more effective in seasonal forecasting as compared to staring at a long-range climate model. case in point, the AGCM's preformance prior to the 2001-02 winter. which was laughable at best.

the GFS (AVN/MRF) is a better example of the inconsistencies in meadium range model preformance. beyond 240hrs, the output is for the most part useless.

And the long range model data is way useless because for the same reason you said FLguy.

Exactly again ... which refers me to my original post on this thread ...

TODAY IS A GOOD EXAMPLE OF WHY HUMANS ARE STILL NEEDED IN THE FORECAST PROCESS...AND HOW EASY IT IS TO BECOME RELIANT ON MODELS AND FORGET THAT THEY CANNOT HANDLE EVERY SCENARIO BY THEMSELVES. from the GSP AFD in regards of JUST WHERE THE CAD would penetrate ...

And I agree with Derek as well 100%. And just remember, there are smaller scale features that sometimes are totally missed or misread by guidance and just that ALONE can cause large-scale errors the further out you go ...

And for anyone else, remember, that the model guidance that we currently have, some run on respectable resolutions, some aren't ... generally do NOT do well with CAD situations (The ETA is good at AG - ageostrophic gradient, but particularly horrible in dealing with TC's) ... The GFS did much better this past season in tropical cyclone genesis in 2003, but again, comes down to all the biases found in the models ... and these flaws are subject to human interpretation, and sometimes, that can be just as flawed as the model guidance outputs, especially, in MR and LR timeframes.

SF

[FLguy]

agreed with all of you. what it esscentially boils down to is model biases.

[cycloneye]

agreed with all of you. what it esscentially boils down to is model biases.

FLGuy a question about models. Do all the models have biases and if some have more biases than others or all are the same?

[Stormsfury]

Luis

This should help out quite a bit ...

http://www.hpc.ncep.noaa.gov/html/model2.shtml#biases

[cycloneye]

Luis

This should help out quite a bit ...

http://www.hpc.ncep.noaa.gov/html/model2.shtml#biases

Thank you Mike for the link that has a very complete info about all the models that have biases.

[Derek Ortt]

Even if the models could resolve every small scale feature, it is still a mathematical impossibility to solve the equations to a point that would enable one to make 100% accurate forecasts

[FLguy]

Even if the models could resolve every small scale feature, it is still a mathematical impossibility to solve the equations to a point that would enable one to make 100% accurate forecasts

exactly. the MESOETA and MM5 are a good example of that. their higher resolution and smaller grid scale can be useful in trying to pinpoint where heavier precipitation will fall, but still cant resolve most forms of convective and conditional instability. in order to make an assessment of where banded snowfall is likely to develop (CSI banding) one has to look for areas of weak positive or Negative EPV within an environemnt with sufficient vertical motions to saturate the unstable layers. thats the only way to tell where banded snowfall is likely to develop. the MESOETA or GFS arent going to pinpoint that for you because the model's grid system isnt small enough to resolve where CSI bands set up.

the same goes for convective instability, a mesoscale model cannot tell you where a thunderstorm cell is going to develop 6-12 hours in advance, but it can tell you where the higher CAPE values are likely to be, or where lapse rates are the best to support convective development.

that is why its so importnat to understand how to accurately find the problems with the model output and then use your knowledge of the pattern to correct them.

[cycloneye]

The models are good for guidance but when you want to know what will happen beyond 7 days it is more difficult to make a precise forecast from that far out.

[Stormsfury]

The models are good for guidance but when you want to know what will happen beyond 7 days it is more difficult to make a precise forecast from that far out.

And that's where pattern recognition and human elements MUST factor in ... weather will NEVER have a 100% guarantee until it's actually occurring right at that moment.

[MGC]

Necessity is the mother of invention. Given enough time and desire, humanity can resolve any problem it faces. Long range weather forecasting will one day be reality. It will take some time as the technology has yet to be refined. I can remember when the first color TV arrived on my block and now I'm receiving my TV via satellite. We are still in the infancy of long range weather forecasting so please be patient. Build it and they will come........MGC

[Derek Ortt]

It will be possible when we can solve 500,000 linear equations with the same number of knknowns to a point when the resulting rounding and computer errors do not diverge to the point so that the output is as much of a pile of dung as a pile of dung. There is no mathematical method known to do this, only approximations are available. Until a more reliable method is developed, the computer input will be limited by this point

[MGC]

I agree Derek. There are many limiting factors currently which prevents models from a reasonable degree of accuracy. Given time and money these limiting factors will be overcome. Of course it will require government cooperation on a global scale which might well be the hardest factor to overcome.......MGC

[FLguy]

It will be possible when we can solve 500,000 linear equations with the same number of knknowns to a point when the resulting rounding and computer errors do not diverge to the point so that the output is as much of a pile of dung as a pile of dung. There is no mathematical method known to do this, only approximations are available. Until a more reliable method is developed, the computer input will be limited by this point

mathematically speaking, im not even convinced its possible.

[Stormsfury]

mathematically speaking, im not even convinced its possible.

I totally agree with that ... especially since meteorology can defy guidance at a moment's notice. IMHO, I'd rather see the focus continue to shift back to the SR, and studies to even more trying to pinpoint tornado development to a specific area for incredible lead times, etc, and that's something that we cannot quite do as of yet ... though, we've come a hell of a long way and undoubtedly seen countless lives saved due to early warnings, but the fact is, there are still some surprises.

SF

[FLguy]

and as long as there are still mathematial imperfections in numerical modeling, there will never be perfect model output. the other side of the problem lies with RAOB coverage.

[cycloneye]

But I think that there has been progress forecasting hurricane tracks and Isabel was a good example.That 5 day forecast that for the first time TPC did officially not experimental was IMO A+ regarding powerful Isabel's track.We knew that Florida was not going to be a target of Isabel when she was still east of the leewards thanks to the 5 day forecast.So in that regard there has been progress but much more has to be done to make an exact forecast but as MGC said let's be patient because time will come that will make the forecasts more exact than what they are now.

[Stormsfury]

Isabel was definitely a forecasting success in regards to nailing down a specific area. Claudette was also well forecasted, and all the guidance 5 days out generally targeted Texas/Mexico ... and was narrrowed down to South Central Texas ...

SF

[Stephanie]

The majority of the models had Isabel pegged at hitting the EC in the MidAtlantic. The biggest problem was with the strength which FORTUNATELY was alot less than what she had originally grown to.

[cycloneye]

The majority of the models had Isabel pegged at hitting the EC in the MidAtlantic. The biggest problem was with the strength which FORTUNATELY was alot less than what she had originally grown to.

Yes Steph last year the forecast of the tracks was very good as I mentioned with Isabel and SF mentioned Claudette but still about forecasting intensity there is a ways to go and an example was Isabel at it's early formation how the forecast of the intensity was not so good that Isabel went from a wave to a hurricane in 2 days.