Accuracy of Grib weather data

Mark-1 said:
Hi Frank,

Was I looking at the right explanation in my post 16? If so I need a bit of spoon feeding sorry!

http://www.ybw.com/forums/showpost.php?p=3882363&postcount=16
Click to expand...

Because of the GRIB spacing of ~27 km for the GFS, it cannot capture detail below about 130 - 150 km size. Although there may be a broad area of strong winds, the strongest will be fairly localised. What you will get, if you could get the 27 km data, would, in effect, be values smoothed to that resolution. The strongest winds would be missed. As it is, we only get the data at about 50 km so the smoothing effect is more marked.

Is that clear? I realise that I have lived and worked with NWP models back in my )long past) working life but still do as an active travelling cruising sailor who uses GRIBs continually when cruising. What seems blatantly obvious to me may not be so to others without that background and experience.

I have discussed GRIBs with ocean going cruisers, RTW racers and people who do my kind of sailing. I have found virtual unanimity on their value - if used carefully. Using any weather forecast, care has to be exercised. There is no such animal as an "accurate" forecast. Weather is imprecise; there is always an element of chaos. But, there is no other way of tackling the problem. We can only use what we have and expect improvements but never total accuracy.
 
Last edited:
franksingleton said:
Although there may be a broad area of strong winds, the strongest will be fairly localised. What you will get, if you could get the 27 km data, would, in effect, be values smoothed to that resolution. The strongest winds would be missed.
Click to expand...

Thanks, I appreciate your time and patience.

So we're saying that lower wind areas are larger, wheras higher wind areas of the weather system are more localized so on average the higher wind areas are more likely to get missed???

And the reason the overestimates are more likely in winds over 15knots are that the peak areas of winds over 15 knots are more likely to be under the resolution and more likely to be missed???

So, for instance, a low would have small areas within it of higher wind speed which would be predictable if the resolution was right and the computing power was there?

How do forecasters factor these local peaks in for the inshore etc? The peaks must be invisible in the forecasters data as well which presumably is at the same resolution as the gribs. Clearly if there's a land mass they can have a good guess but out at sea is it just a case of the forecasted thinking "A low giving 20kt winds usually have local peaks of 30kts so I'll forecast 20kts to 30kts instead of 20kts?"???

Have I got the right end of the stick on any of this?
 
Mark-1 said:
Thanks, I appreciate your time and patience.

So we're saying that lower wind areas are larger, wheras higher wind areas of the weather system are more localized so on average the higher wind areas are more likely to get missed???

And the reason the overestimates are more likely in winds over 15knots are that the peak areas of winds over 15 knots are more likely to be under the resolution and more likely to be missed???

So, for instance, a low would have small areas within it of higher wind speed which would be predictable if the resolution was right and the computing power was there?

How do forecasters factor these local peaks in for the inshore etc? The peaks must be invisible in the forecasters data as well which presumably is at the same resolution as the gribs. Clearly if there's a land mass they can have a good guess but out at sea is it just a case of the forecasted thinking "A low giving 20kt winds usually have local peaks of 30kts so I'll forecast 20kts to 30kts instead of 20kts?"???

Have I got the right end of the stick on any of this?
Click to expand...

Just going away. Will try to rtemeber to answer later.
 
franksingleton said:
Rather than arguing on the basis of subjective impressions, get some numeracy into the argument.
Click to expand...
Someone, yourself perhaps, noted "Like medicine, a weather prognosis needs a diagnosis based on observation."

This thread started from observing that wind speed data fron Sandettie was being ignored by GFS. Why is the simple question?
 
franksingleton said:
I see several of the usual misunderstandings ....

The best way to check accuracy of GRIBs is to repeat the exercise that I have done on several occasions for talks. See http://weather.mailasail.com/Franks-Weather/Grib-Forecast-Examples.

......

An Alternative, but equally revealing test is to look at an 8 day forecast on day 1. On day 2, look at the 7 day forecast. On day 3, look at the 6 day forecast. Etc.. These will all verify on the same day. Two things outcomes happen. Either there first three or four will be consistent. If that is so, then the chances are that the 5 day forecast will be pretty good and you can plan ahead with come confidence.

If the First few are inconsistent, then you will not be able to plan ahead more than a day or two at a time.
.
Click to expand...


This is the sort of thing I would like to see forecasts include as a general background by way of indicating confidence that the forecast is good.... or imaginary?

a simple indicator to say confidence is high, medium or low would help to give the user a background... so today the confidence is low... "perhaps the weather will be much worse."
 
franksingleton said:
Can I make a plea. I do not at my laptop or tablet all day monitoring YBW. If anyone really wants me to add a bit of real knowledgee it would help if the sent me an email. It can be either using the YBW or my home email. "frank at franksweather.co.uk."

Or, is there a way of being automatically aleretd if certain topics arise?
Click to expand...

Well the obvious answer would be a separate weather forum as its common to power and sail but that suggestion has always fallen on deaf ears in the past.
 
Tomahawk said:
This is the sort of thing I would like to see forecasts include as a general background by way of indicating confidence that the forecast is good.... or imaginary?

a simple indicator to say confidence is high, medium or low would help to give the user a background... so today the confidence is low... "perhaps the weather will be much worse."
Click to expand...

In principle, it can be and is done - using model ensembles. See http://www.metoffice.gov.uk/learning/science/days-ahead/how/ensemble. When the BBC forecaster is very certain, they will tell you about the next week. When they are uncertain, the forecaster will often say so.

For forecasts that we use they could give “certainty" factor and that has been discussed in the met office over the years dating back to my time. They have always resiled from doing so for daily forecasts because many in the general public would not understand. Just think about the long range – seasonal or monthly forecasts and how these were always misqupoted. People on this forum would for the most part, no doubt, use the figures sensibly. Many would not.

I doubt that there is any way that they cou;d be attached tp GRIBs – which is why I always advise the approach that I suggested. It is a kind of poor man’s ensemble.
 
ParaHandy said:
Someone, yourself perhaps, noted "Like medicine, a weather prognosis needs a diagnosis based on observation."

This thread started from observing that wind speed data fron Sandettie was being ignored by GFS. Why is the simple question?
Click to expand...


To predict the weather somewhere, you have to know about the weather everywhere. To predict the wind at Sandettie for more than an hour or two ahead, you have to know far more than what is happening at Sandettie.

The GFS (as the UK and all other major Met services) start with a 6 hour forecast from the last data time. They then merge that with all the data at whatever time, place and height using a form of 4D analysis/data assimilation. This has been and still is a major problem. As much computer power goes into this process as into the forecast itself. One of the reasons for improvement in forecasting has been the enormous effort put into data analysis.
 
Mark-1 said:
..............

....................
Have I got the right end of the stick on any of this?
Click to expand...

I seem not to have explained myself very well. Or, its is late at night and I may not have grasped you question fully.

If you have, for example, a strong NW airstream behind a cold front, the average gradient might correspond to, say, 30 KN. Within that there will be areas and times when the wind is above or below 30 kn. The computer model cannot resolve these to better than about 130 to 150 km size. So, if I am planning ahead and looking at a forecast that says 25 knots, my immediate assumption is that if I go out, I may well bet F7 winds,

If a depression has a tight center, then it will be a matter of luck whether or not there is a grid point near the centre. My IOM example at http://weather.mailasail.com/Franks-Weather/Grid-Length-Resolution gives some idea about the limitations of a 25 km grid. Remember that the GFS only provides data o a grid of about 50 km .

For inshore forecasts, the forecaster, until fairly recently, has used experience and native wit coupled with the UK 25 km model. Now, the UK has a meso-scale model using a 1.5 km grid. If you look at their free App, you can see the area for which it is run. Quite small. Around the boundaries, they have to update continuously from the global 25 km model. Within the area, they can start with a very detailed analysis.

There are several reasons why even such a model has its limitations.
1. It depends on the large scale pattern.
2. Small weather details have short lifetimes so that small detail in the analysis may last only a few hours.
3. Detail such as convection cannot be modelled – there is no way of knowing where or when the next shower will form.
4. Prediction of sea breezes effects will be greatly dependent on cloud cover, itself difficult to get right in detail.
5. Some of the topography that drives sea breezes is too small to be resolved even by a 1.5 km model. For example, the UK model is capable of modelling the Isle of Wight effect but not the Tor bay or Plymouth effects.
6. Headlands such as Start Point and Portland Bill are too small to be modelled in detail.
I could probably dream up a few more but, I hope that you can see why I usually say something rude when I see claims of high precision forecast to 1 km resolution anywhere in the world.

Having said all that, all weather is difficult to predict and wind is no exception. In all my years of sailing, I can only remember one passage when the wind hardly varied. Admittedly, it was night time but from S Sicily to Gozo the wind direction and speed was uncannily steady. Usually, direction and speed vary greatly –as we all know. The atmosphere is in a continual state of flux. As soon as there is a pressure gradient the air moves but, as it moves, the pressure gradient will change so the wind changes.

One example that I usually give of the problem of wind is think of a W F4 up the Channel. If the pressure at Southampton is 1000, the pressure on the latitude of St Malo will be 1004. If the pressure at Southampton falls by 1 hPa, ie the total weight of air above Southampton has fallen by 0.1%, then the wind will be F6. If it falls by 0.2% the wind would be F6. I hope that that gives some idea of the fine balance between a F4m 5, 6. I sometimes marvel at how good wind forecasts are – although I complain as much as anyone!

Apologies for the length. An apparently simple question about weather rarely has a simple answer.
 
franksingleton said:
I doubt that there is any way that they could be attached to GRIBs – which is why I always advise the approach that I suggested. It is a kind of poor man’s ensemble.
Click to expand...

Techically there would be no problem - simply attach a reliability number to each grid point (perhaps the variance of the ensemble parameter being checked at different forecast times?) and output it as another layer in the GRIB file. Low variance (normalized against the actual values) = high reliability and vice-versa. GRIB is merely a data format for raster data, and GRIB itself doesn't care what's in the different layers. Of course, the various programs people use to view the data would have to be amended to make use of the data, but I don't see it as a problem for GRIB as such.
 
Conachair said:
Very interesting thread, ta for taking the time to post, Frank.

Came across this, which is quite interesting, and some accuracy stats. Winter is better than summer, never knew that..

http://www.accuweather.com/en/weather-blogs/weathermatrix/why-are-the-models-so-inaccurate/18097
http://www.emc.ncep.noaa.gov/gmb/STATS/STATS.html
Click to expand...

To be blunt, it is a pretty half baked bit of journalism. The first question is a loaded one of the “have you stopped beating your wife,” I could equally have asked, “Why are NWP models not more accurate?”

On any particular occasion one model may outdo the rest. That is probably as much due to analysis rather than models themselves, or partly so, at least. The ECMWF uses a shorter grid length; they put more effort into data analysis/assimilation. They can do that because, unlike the NOAA GFS, the UK Met Office, HIRLAM or Météo France, they do not have an immediate imperative. Their remit is medium range. My spies tell me that ECMWF are slightly ahead of the UK and the UK slightly ahead of the rest but that there really is little to choose. The ECMWF normally does better at 3” Days. Their H+84 is as good as the UK H+72.
Those findings are based on RMS of surface pressure over the N Hemisphere.

All the major NWP centres monitor their and other models. The modelling community is very tight knit and there is much interchange of ideas. Any modeller worth his salt know about all the models in use. In essence, they do the same job working with all the physics that drive the atmosphere. Differences arise in detail such as just how what approximations are mad and what estimates are used for the many factors that cannot be measured or calculated exactly.

To say that a NWP model is an algorithm is a somewhat naive statement. They are massive suites of programs starting with data analysis and assimilation, itself taking as much computer power as the forecast.

To say that they tweak the models is misleading. The modellers are always looking to refine the data analysis and making best use of new data sources. They are always trying to improve the ways in which physical processes are represented, And so on. Every time a modification is made, they check that the model has benefitted.

“Chaos” needs is far more important than “Of course, if you believe in chaos theory, there are limits to what computing power will buy us...” implies.

The point about ensembles is to counter the effects of chaos, a point not brought out. It is little use, as Weatheronline does, just to give one chart from an ensemble. You have to see all the outcomes – 24 of them for the UK.
 
AntarcticPilot said:
Techically there would be no problem - simply attach a reliability number to each grid point
..............
Click to expand...



The problem that you overlook is that ensembles are run for a specific area. The UK runs its ensembles looking at W Europe. The critical area into which small analysis variations are introduced may be over, say, N Canada.

Someone running an ensemble for, say the US W Coast, may have to be putting in analysis variations in a different area...

One logistical type of problem with probabilities is, “what does the user do with them?” A couple of years ago, in a snow situation over the UK, there was that most difficult case of warm front rain coming over cold ground and air. Will the falling rain cool the air by evaporation enough to give snow over London? The meso-scale ensemble gave a 20% risk. What does the forecaster do?

Strictly, he should have passed on that advice. But, just think of the consequences. What would the councils have done? Had all their staff on stand-by at enormous cost? Have 20% of staff on stand-by? Would each council have made the same decision? The press would have had a field day whatever happened. . ANOTHER SNOE FORECAS WRONG – AGAIN!

In the event, the chief (it was senior in my day) forecaster made the right call. No mention of snow for London. I am sure that he had a sleepless night.
 
I use GRIBs from Mailasail which are quick and easy to download. These, I think, are GFS gribs.

Is there as easy access to GRIBs from other models?
 
franksingleton said:
....

One logistical type of problem with probabilities is, “what does the user do with them?” A couple of years ago, in a snow situation over the UK, there was that most difficult case of warm front rain coming over cold ground and air. Will the falling rain cool the air by evaporation enough to give snow over London? The meso-scale ensemble gave a 20% risk. What does the forecaster do?
....
In the event, the chief (it was senior in my day) forecaster made the right call. No mention of snow for London. I am sure that he had a sleepless night.
Click to expand...

My interpretation of that is 100% indication of precipitation and 80% chance it will remain as rain..

To me 80% seems like a high level of confidence... which was borne out by the outcome..
 
franksingleton said:
One logistical type of problem with probabilities is, “what does the user do with them?”
Click to expand...

I do 'more clever' routing (go closer to features and further off the GS track) when I think the forecast accuracy is high, and 'less clever' routing when I think it is low. I would consider some departure weather window's "open" if I knew the accuracy was high, but closed if it was uncertain. It would be very helpful to me to have some indication of what the met office thinks the accuracy/probability is.
 
Tomahawk said:
My interpretation of that is 100% indication of precipitation and 80% chance it will remain as rain..

To me 80% seems like a high level of confidence... which was borne out by the outcome..
Click to expand...

I know that and so do many others. Had the press got hold of the forecast of 20% chance of snow we would have heard accusations of hedging bets, some would have latched onto the word snow and made a meal of it. I really cannot guess what the councils would have done and doubt that they would have all done the same thing. Our press, as Leveson has said, leaves much to be desired.
 
You must log in or register to reply here.

Share this page

Top