Using GRIBs and other Objective Forecasts

About this page

A summary of information and advice on using forecasts.

Related pages


On this page -

Introduction

Using weather information is the most difficult, least understood, yet the most critical, of all the many sailing skills. From time to time, I see GRIB services that claim to give “…. weather information that is accurate up to the nearest mile” or “the most comprehensive and easy to use global weather information service” or “the most detailed sailing weather forecasts ever produced” or “the most accurate” or that provides “location-specific weather information” or “animated, up to date, local weather forecast anywhere in the world” or uses “the most advanced weather forecast system.”

We all wish that it could be so easy and, as an old pro, I know that I might be accused of being somewhat cynical but I would like to put all such services into perspective and, maybe, clear the fog a little.

The reality is very different. No easy solutions for getting and using forecasts exist; there are no short cuts and no one stop shops. In this page I try to give guidance on the utility and usefulness of GRIB coded forecasts. The theme is to understand the limitations and apply some thought to their use. Interpreting weather information requires knowledge, patience and experience.

All forecasts are just forecasts; they are not statements of fact. It is my view that if you do not understand these simple facts and are not prepared to put some of your own thought into forecast interpretation, then you are probably better sticking to rowing boats on the Serpentine - in settled weather, of course!

GRIB files and large scale weather prediction

Increasingly sailors are using GRIB files (GRridded Information in Binary code) to obtain forecasts, particularly of wind vectors over an area or, sometimes, for specific locations. Despite the frightening name, GRIB files are simply series of values of wind, pressure etc at grid points. They are direct output from Numerical Weather Prediction computer programs in which equations, describing the physical processes in the atmosphere, calculate changes in pressure, wind, temperatures, cloud etc on a 3-D grid for several days ahead in very short time steps (a few minutes). It is a scientific, mathematical model, an approximation of reality and not a statistical or pattern matching method. Greatly improved forecasting of lows and highs for the next few days has resulted from developments in NWP over the past 50 years.

National weather services run NWP models globally, inputting and outputting data with a horizontal spacing around 20 nautical miles (1/3 degree lat by 1/2 degree lon). This ?grid length? describes weather and topographic features of roughly five times the grid length or 100 NM size, sufficient for most area or general forecasting purposes.

Reduced grid lengths will become possible as computer power increases but be limited, ultimately, by data availability. Despite satellites, data buoys, etc, we will never define the atmosphere precisely.

As someone who was writing programs in the late 1950s, I find the computer power involved to be awe inspiring. However, it is easy to be mesmerised by the technology and forget that, however scientific the method, the results will always be imprecise forecasts and not statements of fact. So much for the science, let us get on to the practicalities.

GRIB was developed initially for national weather services to exchange and store large data volumes in a compact form. Massive GRIB data sets freely available, every 6 hours, provide data at a (slightly degraded) 1/2 degree spacing and at three hour intervals for several days ahead. Private organisations both non profit making and commercial can access GRIB files from the US National Center for Environmental Prediction, General Forecast System. GRIB files from other services, eg Météo France, may be available on repayment. Subsets of these files and products from them are supplied to users for combination with base charts on a computer.

An example is shown below.

Sample GRIB

A file providing data as this example, 12 hourly from T=0 to T=120 hours (five days) was about 17 kb. The cost of the request email and the reply using a mobile phone and GPRS was less than 10p (not roaming). For those who are not sure about wind arrows, along the bottom line can be seen the forecast wind at the position noted. NB GRIB files tend to underestimate stronger winds. For planning purpose, I add one Beaufort force. An experienced round-the-world professional navigator adds 10% in the tropics, 20$ in temperate latitudes and 25% in the Southern Ocean.

GRIB services on offer

Firms offering products using basic GRIB files sometimes make extravagant claims which cannot be justified. Words like ?accurate? and ?detailed? or ?location specific? make me wonder how they define accuracy - and measure it. I have yet to find any firm that has published statistics to support such claims. From my professional experience, I know that forecast verification is extremely difficult and that justifying accuracy claims is always questionable.

Beware claims made by forecasters to be providers of accurate or detailed or location specific forecasts either for large areas globally for several

days ahead or for a large number of specific locations. These are almost certain to be simple spatial and temporal interpolation of data from global GRIB files. Remember that, at best, these will only define weather on a 100 to 150 NM scale and provide indicative rather than definitive forecasts; the user may be misled into thinking that he is getting more real information than is the case.

This does not detract from the usefulness of the information. It does have implications on the way the data are used. Some GRIB services that I have used or looked at are described elsewhere and are [Grib-And-Objective-Forecasts-Reviewed|]]listed on another page .

Small scale weather

We sailors experience weather on small space-time scales, one to five miles and an hour or less. Predictions of use to us are, and always will be, well nigh impossible. To clarify that rather sweeping statement, I am referring to wind prediction rather than rain or cloud. For the latter, weather radar and satellite imagery provide real time data on a resolution of around 5 km. Comparable observations for wind are not on the horizon although there are satellite systems that deduce wind over the sea, but only from low flying polar orbiters, once per orbit, in areas where there is no rain and are not near coasts.

Even with perfect observation and prediction systems (neither possible), predictability will always be a major constraint. Even were weather

observed on, say, a one km (1/2 NM) grid, it is unlikely that the detail observed by that network would still be there by the time a user obtained the information, and certainly not by the time a short period forecast was produced.

While large weather features, with long lifetimes, are predictable, at least in general terms, small features, with short lifetimes, are much less so. Taking an extreme, gusts can only be forecast when about to hit. A shower, with its associated winds, is not and never will be predictable, say, six hours ahead. Forecasters have no option but to use ill-defined words, much criticised by users, such as "locally" or "at times."

High resolution forecasts

Using input from global models, more detailed short period predictions can be made over parts of the globe, using smaller grid lengths. As well as input from the large scale global models, most important is high quality meteorological and topographic data on a resolution comparable to that of the model. This requires very significant resources including the ability to interpret a variety of data sources. Some of this will come from satellite and radar imagery. The Met Office is currently trialling a limited, UK area model using 4 km (2 NM) grid lengths although even this only defines weather on a 10 NM scale. Their operational fine scale model uses a grid length about 6 NM (0.11 degree lat/lon) and covers from central Europe across the Atlantic to the eastern USA. The results are only used to 36 hours ahead and it can only define weather on about a 30 NM scale.

Claims relating to high precision forecasts, from national weather services or private companies, should be assessed in the light of the preceding paragraph. The US gives free access to GRIB files of fine scale forecasts, at 6 NM (0.1 degree lat/lon) spacing around North America and Météo France makes some fine scale forecasts available on repayment. Otherwise,

fine scale predictions are made by private firms running good fine scale models with topographic data sets to match, but starting with GRIB file data at the 1/2 degree (30 nm) spacing interpolated down to the grid spacing of the model. In my opinion, there are two shortcomings to such fine scale forecasts.

First, there is unlikely to be any input of small scale weather detail. Without such input models cannot take into account small areas of cloud that may affect the heating of the land and, thus, delay or weaken sea breezes. They would have no data relating to areas of thundery rain moving northwards from France.

Secondly, the models can take several hours, at lest six, to adapt to the sudden injection of topographic detail on the very much finer scale than the global models. The first six or so hour’s predictions may show little improvement over the global scale model. In fact, there even may be some reduction in quality.

And my assessment of them

For these reasons, I think these commercial models may well only give a large amount of data and little real extra information. Looking at output from some models I have been unable to detect much detailed wind structure comparable with the grid spacing. However, my opinions can only be proved or disproved by a thorough, independent and rigorous trial. This could be of forecasts for light vessels and data buoys around the UK and Biscay. The forecasts could be from the US GFS GRIB files and high resolution forecasts from one or more commercial firms. The Météo France Navimail system offers both global and fine scale output. As a national weather service they have the resources to start from a good detailed analysis. They should be included in any trial.

Does anyone know of a University department wanting projects for its students in computing and statistics?

Anyone who thinks that I am being unduly pessimistic about these fine scale models should read a very good article on The history and future of numerical weather prediction in the Met Office by Brian Golding, Kenneth Mylne and Peter Clark, published in November, 2004 in Weather, the popular magazine of the Royal Meteorological Society. The need for fine scale data is stressed as is the need for subjective input. The final section of this fascinating article takes a look at very high resolution models down to 1 km grid lengths. The authors anticipate that sufficient computer power to carry out short operational predictions over the more populous parts of the UK could be available by the end of the decade.

Copies of this article can be obtained from through Wiley InterScience.

So, how does the sailor get detailed local forecasts?

Sailors like to have some idea about local weather conditions and it would be very convenient were these to be reliably available either from national weather services or the private sector. I argue that they cannot and that the only answer to the question, in a nutshell, is by hard work; they will not come nicely wrapped up with pink ribbon. No currently available objective, mass produced, allegedly location specific or very detailed forecast can predict the Torbay or Plymouth sea breezes, or the effects of the Isle of Wight. No current operational forecast, nor even the model under trial by the Met Office, can cope with the topography of features such as St Catherine?s Point, Portland Bill or Start Point. Location specific forecasting is a myth ? whoever is the provider.

GMDSS inshore waters forecasts on VHF or NAVTEX 490 kHz, and sea area forecasts on NAVTEX 518 kHz, marine MF/SSB, national radio, INMARSAT-C and HF/SSB all come from national weather services such as the UK Met Office and Météo France. They are produced from subjective, expert assessment of NWP output, The interpretation of these forecasts, especially near coasts, requires experience and nous. An experienced sailor (or one who has read and understood books by Alan Watts and David Houghton, among others) will know that -

  • Wind will blow around headlands and along steep coastlines rather than over them.
  • Wind will be accelerated round headlands and through straits.
  • Heating tends to reduce pressure over land, shaping isobars and winds accordingly. That is how sea breezes form.
  • Night time cooling tends to increase pressure over land giving offshore winds, especially down slopes and valleys.
  • Winds often increase by day and decrease by night.
  • Gusts are likely near showers.

Experience teaches us about such effects; nous helps us apply that experience.

It must always be remembered that relatively small changes in pressure can make all the difference between a boring force 2-3, a pleasant force 3-4, a delightful 4-5 or even an exciting 5-6.

With the wind scales on RYA MetMaps it is easy to see that a westerly force 4 up the Channel will occur with pressures of, say, 1000 hectoPascal (millibars in old money) at Southampton and 1004 at St Malo, a 4 hPa difference, A fall of pressure to 999 hPa (0.1%) at Southampton, gives a 5 hPa difference and a force 5. If followed by a rise of 1 hPa at St Malo, increasing the difference to 6 hPa, the wind will reach force 6.

Singleton's Seven Safety Suggestions

  1. Keep working at using forecasts. There are no short cuts, no easy solutions, no one stop shops.
  2. Weather does not know itself to within one Beaufort Force. Neither does the forecaster nor his models. For many reasons there will be local, unpredictable, often short-lived variations in wind.
  3. Inconsistency between forecasts implies uncertainty. That should colour judgements. This applies to forecasts both for days and hours ahead
  4. Better to be in port wishing you were at sea than at sea wishing you were in port. Plan ahead; use the strengths of forecasts to avoid being in the wrong place at the wrong time and having to take unnecessary risks.
  5. Use synoptic charts and GRIB files, or products from them, mainly for planning purposes. For the first 24-48 hours, only use them as aids to interpreting GMDSS forecasts, not as standalone tools. Remember that GRIBs tend to underestimate winds; I normally add one Beaufort force on to winds of 10 knots or more. Adding 20% to the speed in temperate latitudes is a good guide.
  6. Never go to sea without studying GMDSS forecasts for at least the last 24 hours. Always read the latest forecast in the context of its predecessors. On passage, keep monitoring the forecasts.
  7. Latest actuals let you monitor the current forecast. Monitoring forecasts gives advance warning of weather developments ? and changes to the forecast. These are not synonymous.

Finally

ALWAYS remember that forecasts are usually pretty good in general terms for the next few days but poor in detail over the next few hours. Distrust any detailed forecasts for areas or specific locations whether for a few hours or a few days ahead. The only certainty in any forecast is the date ? as long you do not engage with the theologians!

Do I always get it right? No! Weather is like that. My aim is to minimise the chances of getting it wrong.

Remember the old adage. Plan for the worst and hope for the best. The worst may not happen

And a cautionary note

GRIB files and other forecast products derived from them have some practical limitations. First, they are produced automatically with no human scrutiny. Secondly, as I have said earlier, any numerical weather prediction model can only describe weather to a scale of about 5 times the grid length. Thirdly, small scale weather features have short lifetimes and may not be predictable at a useful time ahead.

Some information providers make all this very clear, notably Saildocs, GlobalMarineNet and WindGuru. Others are ambivalent and some make quite unrealistic claims for their services whether on the large global or small local scale. Below I show the warning issued by Saildocs. Users of all GRIB or other objective forecasts should take heed.

Saildocs GRIB warning, reproduced here with acknowledgement to Jim Corenman

"This grib file is extracted from a computer forecast model. While such computer data can provide useful guidance for general wind flow, there are limitations which must be understood. What you are receiving is a weather prediction generated by a computer run by NOAA/NCEP (GFS, WW3 models) or the US Navy (COAMPS, NOGAPS) and downloaded and processed by Saildocs (a service of Sailmail). The network is complex, and any computer network is subject to hardware and software failures or human error which can affect accuracy or availability of data.

In particular, if our servers were not able to download a current data file then the grib-file may be based on old data. The file information is shown above and also contained in the file itself. Also remember that grib data is not reviewed by forecasters before being made available. You are getting a small part of the raw model data that the forecasters themselves use when writing a forecast, and it is your responsibility to make sure that the data is consistent with your local conditions and with the professionally-generated forecasts (e.g. text bulletins and weather-fax charts).

Grib data also has limitations along shore, where local effects often dominate and may not be adequately modeled. In addition these models cannot provide adequate prediction for tropical systems, frontal activity or convergence zones. For example, while global models can provide useful data on the likely track of hurricanes, they grossly underestimate the strength of hurricanes because of their small size compared to the model grid. For hurricane/cyclone forecasts, carefully monitor the appropriate warning messages and do not rely on grib data from any source. That all said, grib data can provide useful guidance not available elsewhere. Understand the limitations and use the data carefully. Grib data should be considered supplemental to other forecasts, and not be relied upon in lieu of professionally-generated charts or forecasts."


Acknowledgement

The text is largely based on articles written for the RYA quarterly magazine and the bi-monthly magazine of the Cruising Association, Cruising. The detailed technical articles by members are among the many benefits of CA membership. ]


Home Page