ECMWF or GFS

[Seven Spades]

If you look at the forecasts for the weather next week or the South Coast, on Monday the ECMWF has strong winds from the south west moving to the North West on Tuesday. The GFS has very light winds coming from the North East or even ENE. Which is likely to be more accurate? I have never seen the two forecasts being so different.

 

[Porthandbuoy]

I use the Predict Wind app. They have 6 models for their weather forecasts. Generally they are in agreement up to 3 days ahead. By the 7th day they tend to disagree, probably due to some butterfly in the Amazon flapping its wings. When the models agree you can be pretty confident the forecast is reasonably accurate. GFS is usually the first one to diverge from the others in my experience, while ECMWF tends to be the most accurate in the longer term.

 

[TiggerToo]

while ECMWF tends to be the most accurate in the longer term.

the key is in the "M". They have been at it for a long time.

 

[Bobc]

The two forecasts being so different suggests that the pattern is not easily readable. You need to watch all of the models and wait for them to start agreeing with each other.

 

[GHA]

Which is likely to be more accurate?

No one knows.
The gefs-spag 500mB can be useful to get a hint of how sensitive the models are, it runs lots of times with tiny differences in the input. Usually it's pretty solid for a good few days but right now seems the models are touchy, so guesswork that far out just what will happen.
MAG Image Animation

Predictwind do a nice dual display as well - PredictWind.com

GFS has a low tracking a bit further south & dissipating, will be interesting to see what happens next few days, monday midday>>

 

[lustyd]

the key is in the "M". They have been at it for a long time.

I generally assume that the M means they are less good at the short range stuff because they optimise for medium range forecasts. As such ECMWF seem good at next week while GFS seems better at tomorrow as a general rule.
Also, it's quite hot in Reading right now, so don't trust a word from ECMWF, they're probably sunbathing with a cold beer.

 

[Piddy]

I generally assume that the M means they are less good at the short range stuff because they optimise for medium range forecasts. As such ECMWF seem good at next week while GFS seems better at tomorrow as a general rule.
Also, it's quite hot in Reading right now, so don't trust a word from ECMWF, they're probably sunbathing with a cold beer.

Actually they are all working from home. Our daughter works for them!

 

[lustyd]

Actually they are all working from home. Our daughter works for them!

I had a feeling they were just phoning it in

 

[dunedin]

I use the Predict Wind app. They have 6 models for their weather forecasts. Generally they are in agreement up to 3 days ahead. By the 7th day they tend to disagree, probably due to some butterfly in the Amazon flapping its wings. When the models agree you can be pretty confident the forecast is reasonably accurate. GFS is usually the first one to diverge from the others in my experience, while ECMWF tends to be the most accurate in the longer term.

I also use Predict wind and try to compare models. But, certainly for NW Scotland, there have been huge differences for most of the past few weeks - worst was gust predictions of 4 vs 36 knots Just one day ahead. For me, and purely subjectively, it has been Met Office best, then GFAs and largely ignoring ECMWF as most often wrong. But could depend on,location.

Certainly agree that models agreeing is good news, models disagreeing is a warning of instability and beware. Just a lot of “beware” recently up here.

 

[TiggerToo]

Actually they are all working from home. Our daughter works for them!

getting ready to sell up, and move to Darmstadt or some such, then....

 

[GHA]

But could depend on,location.

The windy station reports let you compare actual data with the models which can be very interesting & sometimes flag up breezy headlands etc when going to pastures new

 

[Daydream believer]

I had a feeling they were just phoning it in

Which reminds me of 20+ years ago taking a big green one down my oillies as I was trying to hank the storm jib onto my Stella just off Walton. Thames CG were, at the same time transmitting an inshore update of something -like variable 3 or less smooth or slight. At the time I thought, " Surely someone could just take a look out the b...y window now & then"

 

[glynd]

If you look at the forecasts for the weather next week or the South Coast, on Monday the ECMWF has strong winds from the south west moving to the North West on Tuesday. The GFS has very light winds coming from the North East or even ENE. Which is likely to be more accurate? I have never seen the two forecasts being so different.

I understand ECMWF uses smaller cell sizes than the US GFS model (at the moment anyway) so possibly benefits from that?

 

[lustyd]

Smaller grids don't necessarily produce better results, just more data so I wouldn't use that as a yardstick. In theory resolution does help but only if the model is good enough to make use of it.
Also remember that more data does not mean more, or even any, information!

 

[Piddy]

getting ready to sell up, and move to Darmstadt or some such, then....

Bonn mostly but with the super computer in Bologna.. Not selling up as the MOD owns the site in Shinfield. Whilst not funded directly by EU (it's weather organisations in Europe), it doesn't make sense to have it here (to them, at least).

 

[Buck Turgidson]

I've come to the conclusion that the actual weather doesn't pay much attention to the forecasts and therefore neither do I. These days I prefer a general overview of where the systems are right now and roughly where they are going. rather than looking at small grid long forecasts.
Best way to know what will happen is to look up!

 

[TiggerToo]

Bonn mostly but with the super computer in Bologna.. Not selling up as the MOD owns the site in Shinfield. Whilst not funded directly by EU (it's weather organisations in Europe), it doesn't make sense to have it here (to them, at least).

Indeed, it doesn't. We are a backwater.

 

[TLouth7]

I have never seen the two forecasts being so different.

They actually offer very similar patterns of weather, they just disagree on how fast the low will move. The large difference in predicted wind at a given point in time is a result of the real rapid change in wind that you get when a low moves directly over a region.

As for which is better, I prefer the ECMWF but that might just be a pro-European bias.

 

[Roberto]

From the MetOffice Extended Outlook for Niton areas

On Friday evening, a low is expected to be centred over the far north of France 1006 with a ridge of high pressure extending from the Azores across the NW of the UK. A second low over Iberia 1010. Pressure remaining relatively low on Saturday across France and Iberia. An Atlantic trough will move east across far western waters of Niton and Shannon later in the day. By Sunday afternoon a complex low develops across the United Kingdom although confidence for detail is currently very low. The current forecast has an area of low pressure centred Scotland 1000 by the early afternoon on Monday with relatively high pressure extending eastwards across Fitzroy and Biscay. Gales possible Friday night into Saturday morning across far southeast Fitzroy. Across Sole and Shannon occasional gales may affect these areas at times through the period, with the greatest risk affecting Shannon

 

[franksingleton]

ECMWF has a finer grid, 9 or 10 km vs 13 for GFS. It also has more levels in the vertical. Therefore, it can define the atmosphere more accurately to start off a forecast. It runs about two hours later than the operational, nationally run models.

It produces output to users twice a day. Although it should, in principle be better in the short term - 3 to 5 days - the gain is small on that time scale and negated by the more frequent updates of say, GFS and ICON. Its remit is medium range so the later start allows it to produce better results. For 16 hours in 24, GFS and ICON forecasts available to us are based on more recent data Than the latest ECMWF available, Between about 0700/1900 and 1100/2300 UTC, ECMWF output will be based on the same initial time but have a better starting analysis. In practice, it probably does not matter which you use for forecasts up to 4 or 5 days. Personally, I prefer the operstional models and more recent data. Beyond that time, ECMWF should be the better bet.

PS. Apologies for the late input. Being on the move, I missed this one.