GRIB files

[franksingleton]

Frank, I've been trying to understand for years why the mesoscale models on let's say WindguruPro are much closer to actual coastal winds than GFS, which on the day almost always seems to be about 1 force too low. Let's take the Devon & Cornwall coast where I am now (Dartmouth to Falmouth). Are the higher res. models taking account of topography as you say, or sea breeze, or what? The fact is that over many years (in West Wales as well) they always seem to give higher forecast values and these are closer to actual in many if not most cases.

The performance of all meso-scale models will depend, ultimately, on the large scale forecast ie they have to take in data from global models or, better still, run globally on a meso-scale size grid. The latter is not possible at present because of computer power.

Meso-scale models do represent topography better than the GFS simply because the grid points are closer together. However, all models can only represent weather and topography on a scale of about 4 or 5 grid lengths. On one of my pages I give the case of the isle of Man. To represent the shape and topography of the IOM would need a very close grid. As far as I am aware the Windguru that you pay for uses a grid of around 10 km. It can only represent topography on a scale of about 40 or 50 km.

The free version of Windguru is, as far as I know, just the GFS interpolated. Hence, I would not expect the free version to be any better than the GFS. I would expect some improvement on the GFS with the paid for version BUT only when the global forecast is good and, usually, when there is a fairly static situation. That is why the SWS will often do better than the GFS through the Gibraltar Strait but be no better in predicting the change from E to W winds through the Strait. That is a result of large scale – “synoptic” scale patterns.


Does that help?

 

[bluedragon]

If I understand you correctly Frank the meso-scale models take their data from the coarser global data (say GFS) and then refine it at closer grid points, using more than just interpolation. That's clear. But why do the data figures almost always come up with a higher wind speed once the resolution has been increased? I've been using Windguru Pro for some years now, and always have the GFS data on the screen at the same time as the high res. model for any given location. Over many years and many hundreds of observations, the WRF 9km data is almost always higher, often by 5-10knots in active weather systems. Only when we have slack pressure gradients as today does the data start to line up. I don't understand why higher resolution itself leads to higher wind predictions in coastal waters...yet it always seems to.

 

[elton]

I've noticed an irrational dislike or dismissal of zyGrib, mainly from people who've never used it. I've been using it for two years (both Windows and Linux versions) and I can confirm it is robust, fully functional, simple to use, simple to install, and as accurate as any other GRIB viewer.

 

[franksingleton]

If I understand you correctly Frank the meso-scale models take their data from the coarser global data (say GFS) and then refine it at closer grid points, using more than just interpolation. That's clear. But why do the data figures almost always come up with a higher wind speed once the resolution has been increased? I've been using Windguru Pro for some years now, and always have the GFS data on the screen at the same time as the high res. model for any given location. Over many years and many hundreds of observations, the WRF 9km data is almost always higher, often by 5-10knots in active weather systems. Only when we have slack pressure gradients as today does the data start to line up. I don't understand why higher resolution itself leads to higher wind predictions in coastal waters...yet it always seems to.

Am at anchor just now. Power limited. I will come back ASAP. meanwhile, have a glance at my page on NWP. It may help.

 

[franksingleton]

I've noticed an irrational dislike or dismissal of zyGrib, mainly from people who've never used it. I've been using it for two years (both Windows and Linux versions) and I can confirm it is robust, fully functional, simple to use, simple to install, and as accurate as any other GRIB viewer.

There can be no difference in forecasts between zyGrib, UGrib, Sildocs, MailASail, XCWeather etc becaise all these are using the GFS. I like xyGrib as it is robust software.

If bandwidth is restricted, I prefer Saildocs using email with a repeat request facility.

 

[franksingleton]

Am at anchor just now. Power limited. I will come back ASAP. meanwhile, have a glance at my page on NWP. It may help.

OK. I have some power now and can sit at my keyboard.

First, as I said, the free Windguru is simply a straight interpolation from the GFS. But you are using the PRO version.

The GFS calculates on a ~25 km grid although it only publishes data on a 50 km grid. Using a 25 km grid means that everything is, effectively, smoothed. For winds shown at upwards of F3, I normally assume one extra force. If I see a F6 forecast, I plan for F 7 or 8.

Most, probably all, commercially run meso-scale models start from an analysis that is interpolated from the GFS. They do their computations over a limited area and, therefore, have to be continually changing the boundaries using the GFS to do that. As far as I am aware, and I have asked most but not always had replies, none uses any more detailed data such as at my page http://weather.mailasail.com/Franks-Weather/Weather-Observations-Nwp. The “big boys” ie the UK NAE, the US NAM, USN COAMPS, HIRLAM all make use of the vast amount of data to get good detailed analyses.

The WRF and other models used by the commercial modellers are good models and will model topography within the limits that apply to all models ie they can only represent weather and topography on a scale of around 4 – 5 grid lengths. What this all means is that these models are likely to improve on the GFS near land and particularly where topography is significant. Over the open sea, I can see little reason why they should do noticeably better than the GFS. They have input no extra weather data. They are not benefitting from being able to model land effects – except near the coast.

It is fair to say that a meso-scale model should improve on a good GFS forecast but will not rectify a poor GFS forecast. Also the longer the period of the forecast the more that what is happening in any area will depend upon what is happening far away. Also, the lifetimes of small weather detail is short. The UK only runs its NAE out to 48 hours and ( I think only use the output to about 36 hours.)

 

[LadyInBed]

If bandwidth is restricted, I prefer Saildocs using email with a repeat request facility.

The size of the download is (I assume) dependent on the size of the area you have selected to download.

What area around your position do you tend to download to give you a reasonable view as to what is going on?

 

[franksingleton]

The size of the download is (I assume) dependent on the size of the area you have selected to download.

What area around your position do you tend to download to give you a reasonable view as to what is going on?

Download size depends on the area and grid spacing that you choose ie 0.5, 1 or even 2 degrees. It depends on the number of times that you request and the number of elements es wind = 2 elements, presure, rain etc are one each.

When in the W Med, I used to get the area from the Spanish coast to Italy and the S French coast to Africa. I chose wind only. Using saildocs, I was able to order dataevery 6 hours for 36 hours ahead and every 12 hours after that to 5 days. The total email plus GRIB file was under 20 kb.

You really have to experiment at home to see which best suits your needs in terms of file size and area of interest. Someone on an Atlantic crossing might be happy with winds at 2 degree intervals over a large area. Here, on the N coast of Spain, were I pushed for baddwidth, I might be using Saildocs for forecasts every 3 hours, 0.5 degree spacing for 36 hours aand 12 hourly thereafter. As it is, I have an Orange dongle so no problem. I can be profligate!

 

[bluedragon]

Meso-scale forecasts

Thanks Frank. I'm not sure I understand why "smoothing" of the GFS data seems to underestimate the wind speed, but at least your observations seem to agree with mine. Add on at least one wind force for GFS above F3 seems to be a good rule of thumb.

 

[franksingleton]

Thanks Frank. I'm not sure I understand why "smoothing" of the GFS data seems to underestimate the wind speed, but at least your observations seem to agree with mine. Add on at least one wind force for GFS above F3 seems to be a good rule of thumb.

Try my http://weather.mailasail.com/Franks-Weather/Grid-Length-Resolution with the example of the Isle of Man. Or simply draw a shape on paper, see how many grid points you need to define the shape reasonably well. Compare the ratio of grid spacing to the size of the shape. Do simple Y or N for inside the shape or outside..

Then think about defining a gradient.

That may help.