GFS UPgrade

[franksingleton]

NOAA is about to implement a major upgrade to the GFS, probably on 19th March. There will be a doubling in number of levels in the vertical. This better representation of the structure of the model should give better definition of the jet stream and near the surface with a consequent improvement if forecasts. This brings it into line with ECMWF, but, as far as I can see, no change on the horizontal grid of about 13 km as opposed to 9 km for ECMWF and 10 for the UK. Much of the upgrade involves better representation of the physical processes that are so difficult to quantify in models. There are improvements to the wave modelling.
Experimentation and evaluation of the model upgrades covered part of the 2018 hurricane season and the whole period from May, 2019, to the present. Speaking from (bitter) experience in the game, however much is done in advance, the real testing will come after implementation operationally.
Hopefully, there will be few hitches in implementation but it would not be surprising if the various third parties did not have some hitches. This will be so for the various free suppliers such as XyGrib, PocketGrib. SailGrib, MailASail, Windy.com, Ventusky, PassageWeather, XCWeather as well as the commercial providers such as PredictWind, Squid, LuckGrib.

 

[PhillM]

Thanks Frank, for those of us not properly versed in the mechanics, the nod that things on the free sites may not all be as represented due to tech issues, is really helpful.

 

[franksingleton]

Thanks Frank, for those of us not properly versed in the mechanics, the nod that things on the free sites may not all be as represented due to tech issues, is really helpful.

I think that the paid for sites might just as easily be affected.

 

[lustyd]

Speaking from (bitter) experience in the game, however much is done in advance, the real testing will come after implementation operationally.

Are you able to elaborate on why this may be so? Presumably we still have all of the input data going back many years, so why would it make a difference if we're predicting past weather or future? The model doesn't know the outcome in either instance, and in both cases we measure the model by the weather that happens following the prediction.
What am I missing?

 

[sailaboutvic]

Although the weather forecasting have come a long way from when I starting sailing many many moons ago its still hard to predict more then a few days and for them to get it right ,
we waiting for weather window to move on and in the last few days we seen Prediction of the part of the Med we need to cross from 56kts to 32 kts yesterday it drop to 25kts to day its back up to 45kts this is for forecaster in two days time ,
I guess by thursdays there be no wind at all and we be motoring.
Unless the weather is settled 24hours is about the best you can relie on .

 

[franksingleton]

Are you able to elaborate on why this may be so? Presumably we still have all of the input data going back many years, so why would it make a difference if we're predicting past weather or future? The model doesn't know the outcome in either instance, and in both cases we measure the model by the weather that happens following the prediction.
What am I missing?

Not sure that I follow your question. There are two aspects. First is that NOAA have made some changes to the interface used by third parties. These may cause problems, hopefully not bu just possible.
Secondly, modifications of this scale to NWP program suites may have bugs not yet found.
Past wearher has nothing to do with it. Models start with knowing what is happening now and apply the laws of physics. It is an initial value problem.
What is amazing to me, is the ability to predict often with some confidence for a week or more ahead. It is not always possible but, at least, forecasters do say that the situation is uncertain when that is the case.

 

[lustyd]

I thought you were referring to the model being tested rather than the interfaces. That makes more sense!

 

[franksingleton]

I thought you were referring to the model being tested rather than the interfaces. That makes more sense!

Sorry if I have not been clear.
There have been changes to the interface through which third parties access GFS data. I now understand that these are fairly minor and should not create problems.
There have been extensive changes to the model itself. In particular, they will compute at 127 levels rather than the current 64 and the highest level will be at around 80 km. That brings it into line with ECMWF in those respects although, I understand that the horizontal grid will remain at about 13 km. The model has been extensively tested. However, in my experience, there can still be problems that only appear when a model goes truly operational.
These changes should and probably will be good news but time will tell.