best wind predictor sites

franksingleton said:
Re #12.

A number of organisations run numerical weather prediction models with smaller grid lengths. See http://weather.mailasail.com/Franks-Weather/Grid-Length-Resolution and other pages of my site.

What these people do not tell you is that they usually start with GFS output at 50 km spacing but use no more observational data. I am probably rather jaundiced but without observational data it seems a nonsense to try to make detailed predictions. It is like a doctor making a detailed prognosis after a cursory diagnosis.
Click to expand...

Could you comment on the Windfinder superforecast please. They quote a 12km resolution.
I sail the Solent and the forecast has seemed to me the most accurate available, it even, on occasions, seems to forecast the western Solent sea breeze fairly accurately.
 
quick reality check for burnham on crouch. at 19.00 today wrf 9 km windguru says south west 4 beaufort gusting 5. Actual readout from the yacht harbour weather feed is south west and from the display looks more like a 3 gusting 4. however, up nearer black point or donw nearer the Roach it may well be a bit breezier.
 
Windguru as get accurate forecasts for Anglesey and the Gulf of Gökova my two main sailing areas
 
alahol2 said:
Could you comment on the Windfinder superforecast please. They quote a 12km resolution.
I sail the Solent and the forecast has seemed to me the most accurate available, it even, on occasions, seems to forecast the western Solent sea breeze fairly accurately.
Click to expand...

Windfinder is, to the best of my knowledge one of those that runs meso-scale models without using analysing observational data on the scale of their grid. I say that for two reasons. First, I have asked a number of firms what data they use. None has said that they use anything more than grid point GFS forecasts or analyses. Secondly, I have some idea of the computer and manpower resources required to set up and run detailed models using the data necessary to run a model with a proper analysis.

The Met Office dowes run a model with nearly the same grid length – the NAE. You can see it on Weatheronline. They say, and I agree, that it does little if anything more than their global model.

Remember that no model can do better than represent weather and topography on a scale of about 5 grid lengths. The Solent is well below the resolution of a 10 km model. To cope with the Solent sea breeze, for example needs a model with a grid of 1 to 2 km. That requires a very detailed analysis, including weather radar imagery.
 
For the Irish Sea, North Channel & St George's Channel I find Met Eireann's 3 hour forecast pretty good. It gives a good impression of how the wind will vary across an area.

http://www.met.ie/forecasts/3hour.asp

I also use UGRIB & xcweather, but add large bucketfuls of salt. The North Channel being a classic example where it underestimates the funnelling along the channel.
 
lpdsn said:
For the Irish Sea, North Channel & St George's Channel I find Met Eireann's 3 hour forecast pretty good. It gives a good impression of how the wind will vary across an area.

http://www.met.ie/forecasts/3hour.asp

I also use UGRIB & xcweather, but add large bucketfuls of salt. The North Channel being a classic example where it underestimates the funnelling along the channel.
Click to expand...


Ugrib and XCWeather are the same data. Different presentation. You get no more information.

The GFS uses a grid length of about 27 km. It cannot represent weather on a scale of less than 100 to 130 km. It cannot give detail through the North Channel. Human forecasters add experience and knowledge of model behaviour.

Use GRIBs in conjunction with GMDSS services for short term use.

Use GRIBs for longer trepanning up to or 6 days ahead. Look for consistency from one day’s computer run to the next. That gives a good idea when the models are getting it broadly correct. We will be crossing from St Peter Port to Dartmouth tomorrow (4 August.) On 27 July, we identified this as a possible window for us. At our age, we like a daylight crossing, enough wind to sail but not too much. We have watched the GFS daily since then They have been consistent.
 
franksingleton said:
Ugrib and XCWeather are the same data. Different presentation. You get no more information.
Click to expand...

I know. If you look carefully, XCWeather tells you which run of the GFS it is based on. Usually one less recent than that available via UGRIB.

I use the one that is convenient to use at the time (i.e. I don't have UGRIB installed at work).

XCWeather does however show you the current conditions, which can be valuable, even when at sea. You can use it, for example, to see if an expected change has already reached a buoy or airfield 100M away. Very handy when you're taking advantage of short weather windows.
 
DipperToo said:
XC Weather works pretty well. If you have an Android mobile phone, then I have found the App Marine Forecast from Meteo Consult works pretty well. Gives wind, direction, gusts, swell, tide times etc. https://play.google.com/store/apps/details?id=com.lachainemeteo.marine.androidapp

I think they may do an iPhone app as well but not sure.
Click to expand...


I have a fairly comprehensve set of GRIB Apps aat http://weather.mailasail.com/Franks-Weather/Grib-Files-Apps, Any additions, please??

One good (to me) reason to use GRIBs, whether as a Tablet App, FTP (zyGrib or UGrib) or as email is the automatic saving. Also, so that I can select areas, times, elements that I want. Otherwise, the various browser versions are just the same as the GRIBs.

Forecasts by such as Météo Consult are almost certainly GFS or based on the GFS.
 
I looked at the Météo consult App but found it a pain to try to set up on my iPad and gave it up.

The wind data is almost certainly GFS or based on GFS. Any automated system that offers short period forecasts for small areas should always be treated with care. I include those pictorial automated forecasts produced by national met Services. Météo Consult forecasts appear in many marinas on the continent. Maybe in the UK but we are only just back home.

At times they can be dangerously wrong and should always be used in conjunction with GMDSS forecasts that are written by a thinking human forecaster.

At times I have seen winds much under forecast. More seriously visibility has been over forecast. Recently a boat left Audierne for Ile de Sein. I commented that fog was forecast – 50 metres – by Météo France on VHF. “We have looked at the forecast.” They had. It was Météo Consult on the marina office window saying Good Visibility. Ile de sein had a day of thick fog. Caveat emptor.
 
franksingleton: What is your opinion about WRF model (9 km resolution for UK coast) and how would you compare it to GFS?

Thanks
 
The reason I am interested in Franks opinion is cause multiple sources did comparisons of GFS and WRF and WRF won - I also observe better predictions from windguru WRF9 model as well, but this could be just my subjectivity..

From windguru:

WRF model

The Weather Research and Forecasting (WRF) Model is a next-generation mesocale numerical weather prediction system designed to serve both operational forecasting and atmospheric research needs. It is an evolutionary successor to the MM5 model. The effort to develop WRF has been a collaborative partnership, principally among the National Center for Atmospheric Research (NCAR), the National Oceanic and Atmospheric Administration (the National Centers for Environmental Prediction (NCEP) and the Forecast Systems Laboratory (FSL), the Air Force Weather Agency (AFWA), the Naval Research Laboratory, Oklahoma University, and the Federal Aviation Administration (FAA). More info at the WRF model website.

We are running WRF model for most Europe and Mediterranean. Canary Islands, Madeira and part of Morocco's Atlantic coast are also covered by WRF with 9 km resolution. Another smaller WRF 9 km resolution domain covers Northern Red Sea with popular spots in Egypt, and also Israel, Lebannon and south of Cyprus. Argentina, Chile and Uruguay are covered by WRF 12 km. The initial and boundary conditions come from GFS. If everything goes smooth WRF updates 4 times per day and produce forecast for 78 hours to the future in 1 hour step. Forecasts include wind speed and direction, wind gusts, temperature, total cloud cover and precipitation.

The highest resolution model we are running now is 3 km. Such a high resolution requires massive computing so the covered area can't be very large. WRF with 3 km covers Czech republic, forecasts are for next 48 hours and update 4 times per day. Another small 3 km domain covers one of the most famous areas for wind & kitesurfing in Europe - Tarifa / Strait of Gibraltar.
Click to expand...

this is one of the studies done in israel I believe

http://www.mmm.ucar.edu/wrf/users/workshops/WS2008/presentations/8-4.pdf
 
Last edited:
parito said:
franksingleton: What is your opinion about WRF model (9 km resolution for UK coast) and how would you compare it to GFS?

Thanks
Click to expand...


There really is little difference between models. As you no doubt know, WRF is a model under continual development as are all models.

More important is the data input. Most firms running meso-scale models start with the GFS analysis at T = 0. This provides a staring point. The GFS forecasts are used to update boundary conditions at each time step.

Although the GFS uses a grid length around 27 km, it only provides free output data on a 0.5 degree grid, ie around 50 km. Further, it calculates at about 67 levels but only provides data at (my guess) 15 to 20 levels.

Few if any of the various firms running these detailed (meso-scale) models have access to the vast mass of (mainly satellite) data nor the resources to incorporate them into detailed analyses. To use an analogue, to provide a detailed forecast without a detailed analysis is akin to a doctor giving a detailed prognosis from a cursory glance at the patient.

The only way in which these various models may outperform the GFS (or other models eg the UK global) is in modelling topography. Even then, you have to remember that any model can only define features of about 4 or 5 grid lengths in size.

So, using a model with a 9 km grid, do not expect any detail of size less than about 45 km. And that presumes a good analysis.

Given that they start with a crude analysis, you will understand that I do not use any of these models myself. While I would never recommend that they should not be used, I never recommend that they are used.

Meso-scale models that I would recommend are those produced by national met Services. These start with global models on a 15 to 25 km grid. They use a vast amount of data. One of the best is the UK but they do not provide the data in a sailing friendly manner.

Countries that do issue meso-scale models that I would recommend are Denmark, Iceland, Norway, Poland, Spain, Croatia, Greece.

I hope that explains why I am not answering your question directly. Apologies for the length of this reply. I cannot give a simple answer to what is really a complex question.
 
Further to my long post above, some other points.

First, small weather details have lifetimes determined by whatever is their driving force. Convective cloud, several km to 20 or so km in size have short lifetimes as they are driven by convection. They are inherently unpredictable more than a very few hours ahead and even then, they have to be known about in detail.

Sea breezes are rather more predictable but the success of a forecast may depend on getting small detail such as cloud amount correct. A tall order even for the best models with the best data analysis possible.

Wind speed through straits should be well predicted – as long s the model grid is small enough – AND the large scale prediction has got the general flow right.

The way I work is to rely upon models, primarily the GFS, for planning ahead, ie identifying my next weather window. For short term, I rely upon a combination of what the GMDSS forecasts are saying couples with an overview from the GFS GRIB plus experience.

The lead time between collecting data, analysing them, producing forecasts and issuing them is such that, for my kind of cruising, I see no benefit from meso-scale models.

Were I enormously rich and racing a mega yacht, I would be wanting to commission very high resolution models with the maximum of data to give me predictions for up to 6 hours ahead.
 
franksingleton said:
The only way in which these various models may outperform the GFS (or other models eg the UK global) is in modelling topography. Even then, you have to remember that any model can only define features of about 4 or 5 grid lengths in size.

So, using a model with a 9 km grid, do not expect any detail of size less than about 45 km. And that presumes a good analysis.
Click to expand...

Frank, I notice that you always quote a best resolution for topographic features of 4-5 times the grid length. The usual Nyquist criterion would suggest twice the grid-length. I'm sure you are right, but could you explain this?

Incidentally, I note that another constraint on topography is sampling in the vertical dimension. If the models use 67 levels from sea-level to (I guess) the top of the stratosphere, then the vertical resolution is only about 1000m. Even if they only model the troposphere, then the vertical resolution is only about 500m (crude figures, which will depend on atmospheric conditions; I am aware that they don't use actual height as the Z dimension in modelling; I was the instigator of ISO 19111-2, which allows for that!). Given that a high mountain in the UK is only 1000m high, how much of the lower topography is actually representable in the models? Of course, the vertical resolution isn't constant; eavels are closer at low elevations than at higher ones.

Sorry for the technical digression!
 
AntarcticPilot said:
Sorry for the technical digression!
Click to expand...

My 4 to 5 grid lengths derives from a statement in a paper that I could provide if you wish. It referred to defining weather features. Basically, it says that a grid can only fully resolve weather features of size around 4 or 5 grid lengths. For that reason, weather features less than that size have to be damped out in order to avoid computational instability. That was a failing of early models.

By extension, weather determined by topography smaller than 4 or 5 grid lengths would have to be damped out. When I give talks, I usually show http://weather.mailasail.com/w/uploads/Franks-Weather/iom-25km-grid.jpg and http://weather.mailasail.com/w/uploads/Franks-Weather/iom-octagon.png. They show what the Isle of Man would be treated by a 25 km grid. IoM is roughly 30 x 10 km. A 2 km grid would give a reasonable description of its shape.

The top of the models (UK and GFS certainly, ECMWF, I am pretty sure is 80 km. The levels are not equally distributed. There is more definition near the surface and near the tropopause than elsewhere. ECMWF uses 91 levels.

I cannot explain the rationale except to say that there are rough relationships between horizontal and vertical grid lengths and time steps. ECMWF uses a horizontal grid of ~ 15 km and 91 levels. The UK uses ~ 25 km grid and 70 levels. The GFS uses ~ 27 km and 67 levels. Back in the 1970s, the Met Office operational model used ~ 100 km grid and 10 levels.

I do not have the theory for all this at my fingertips. I can only say that, over the past 60+ years a great deal has been learned through experience. These organisations and the other major players would not go to these degrees of complexity were it not worthwhile.

One point that I hope has come over, and I may be labouring it, is that it is quite easy to get hold of a model, say the WRF, bang in some data and get a forecast. The difficult part that these people do not realise, or ignore, is that it is greater problem to derive a good analysis to initialise their models. Hence my analogy with medical diagnosis/prognosis.
 
You must log in or register to reply here.

Share this page

Top