Predictwind: weather routing

Koeketiene

Koeketiene

Well-Known Member
Joined
24 Sep 2003
Messages
18,456
Location
Le Roussillon (South of France)
www.sailblogs.com
Looking for the previous experience of some users.
Predictwind offers various for their weather routing (PWG, PWE, ECMWF, GFS, UKMO, ...) - and the significant differences between their forecasts/routings struck me.

As always, a man with a watch knows the time but a man with two watches is never sure.

Which model have you found to be the most reliable over time or is it really a matter of take your pick and hope for the best?

Many thanks
 
We don't use Predictwind, but the following applies to any weather model: I've found the accuracy of weather models is extremely regional. Meaning there isn't one best model for everywhere - which model gets it right most often depends on where you are. Especially for coastal areas where local effects play a big role, which must be known, documented and entered into the model. It's not unusual for a Greek fisherman to accurately tell you tomorrow's weather (based on decades of his and his peers observations) while the weather model is completely wrong (because nobody has collected and entered the local effects into the model).

The rest varies with quality of input data, meaning the weather observations, ocean temperature measurements, elevation, albedo, etc. - and in areas where there are fewer weather stations the quality drops. In areas where different weather systems regularly meet (e.g. south Peleponese) all models seemed to fail wildly.
 
I am no expert, but do use PredictWind (though not automated routing).
Without doing proper objective analysis it is very difficult to say one model is better than another, even in specific locations. That’s why I use PreductWind to look at and compare models.
The general rule is if all in agreement, then there is a fairly stable weather pattern and can use with reasonable confidence (generally where I sail this means either very good news, of a high pressure spell, or bad news of being pasted with gales).
If a lot of variability between forecasts then can’t rely on any of them beyond the next day.
 
Yngmar said:
The rest varies with quality of input data, meaning the weather observations, ocean temperature measurements, elevation, albedo, etc. - and in areas where there are fewer weather stations the quality drops. In areas where different weather systems regularly meet (e.g. south Peleponese) all models seemed to fail wildly.
Click to expand...

I agree with your point about local variables.
Wouldn't use Predictwind for coastal cruising.

Am planning to leave Brittany for the Azores this Sunday/Monday and when looking at all the models there are significant differences when it comes to the Bay of Biscay in the first half of next week.
 
Koeketiene said:
Would you care to elaborate?
Heading for the Azores this Sunday or Monday.
Click to expand...
The weather changes continually, so do the models. You can get Predictwind to do a route planning based on the models predition for the next 7 days say. That route is only valid for the first day, maybe. After that you need to be able to run the model with the latest forcast each day, starting from your new position each day. For that you need satcoms to run Predictwind out in the ocean.
 
RAI said:
The weather changes continually, so do the models. You can get Predictwind to do a route planning based on the models predition for the next 7 days say. That route is only valid for the first day, maybe. After that you need to be able to run the model with the latest forcast each day, starting from your new position each day. For that you need satcoms to run Predictwind out in the ocean.
Click to expand...

Gotcha - guess I misunderstood your post.
The 'big mistake' using Predictwind whilst unable to receive updates because of the lack of satcom.

Got an Iridium Go of fleabay for the trip.
Had a dry run two weeks ago - Predictwind routing updates do come in but can take quite a while to download (15-20 mins).
 
Koeketiene said:
Gotcha - guess I misunderstood your post.
The 'big mistake' using Predictwind whilst unable to receive updates because of the lack of satcom.

Got an Iridium Go of fleabay for the trip.
Had a dry run two weeks ago - Predictwind routing updates do come in but can take quite a while to download (15-20 mins).
Click to expand...
Don't Predictwind have a special service for satcom users, to reduce the data requirements
 
Some comments on global models. ‘Official” models freely available are ECMWF, GFS- USA, GEM - Canada, ICON - DWD, ARPEGE - Meteo France. These use grids of size 0.1 degree lay/Lon for ECMWF and the UK, 0.125 for the rest, except ARPEGE which uses a variable length grid averaging about 15km. The PWE and PWG use grids of 50 km. All models have built-in smoothing of about 4 or 5 grid lengths. This is to avoid spurious developments and computational instability. So, the “official” models should be able to resolve detail down to between 50 and 70 km. PWE/PWG can only resolve down to about 200-250km. Don’t look for detail from global models. Don’t bother with PWG/PWE.

There is no local experience built into these models. Verification/assessment is done on the large scale. An ECMWF/WMO web page shows routine verifications. These show that ECMWF outperforms the rest. The UK UM is slightly better than the rest. Meteo France is the least good. However, ECMWF runs two hours later and, depending on third party source, you may get it even later. Windy.com can be well over two hours later. Because it has no short term operational pressures, it can afford to get more data in order to produce better predictions over the longer period. For some sailing needs, it is less useful than, say, the GFS which updates 6-hourly compared to 12 hourly for ECMWF.

Using Windy.com, you can see comparison between some models up to 5 days. In my experience, differences are usually small over that period. That is what I would expect from knowledge of ensembles. Wind speeds will usually be within around 5 knots of each other and meaningful directions, a few degrees. If one model is an outlier, then trust none. The odd man out could well be best on the day. Effectively, you are seeing too small am ensemble to make useful deductions. Anyone telling you to use a consensus, does not understand ensembles. Meteociel has a facility to show tabulated values of ensembles for a few models.

Limited area models are, in my view, not very useful for most sailing purposes but might help in the short term. Again, no local fisherman knowledge is included. Remember that small weather details are too short lived for meaningful, deterministic prediction. They should cope with sea breezes and other topographical effects but may not do so I, for example, cloud cover is not well forecast.

Too much is made of local experience. My late colleague, David Houghton of Olympic forecasting fame in the 70s and 80s, did a survey. If a local said “so-and-so always happens” it really meant less than half the time!
 
Last edited:
RAI said:
The weather changes continually, so do the models. You can get Predictwind to do a route planning based on the models predition for the next 7 days say. That route is only valid for the first day, maybe. After that you need to be able to run the model with the latest forcast each day, starting from your new position each day. For that you need satcoms to run Predictwind out in the ocean.
Click to expand...

I agree entirely. The forecasts change substantially from day to day - regardless of which you use. So, really, you need to download every day and use the next 24 hrs as your guide.
 
KeelsonGraham said:
I agree entirely. The forecasts change substantially from day to day - regardless of which you use. So, really, you need to download every day and use the next 24 hrs as your guide.
Click to expand...
In general the main computer models can be quite consistent from day to day. When the are so it is a good sign that the forecast is on the right lines. http://weather.mailasail.com/w/uploads/Franks-Weather/ninedaystppdart.png shows forecasts on successive days for a point in mid Channel. If there are major changes from day to day then the situation is less predictable than normal.

Do not take a majority view if one forecast is an outlier. Those making such suggestions do not understand ensembles.The odd man out may be best on the day.

Remember that all models smooth out detail less than 4 or 5 grid lengths. That is why models usually under-predict the strongest winds.

As a matter of interest, go to the Meteociel site. Look for ensembles. It shows how forecast winds can vary over their 30-member GFS ensemble. This gives an indication of the uncertainty In wind and other elements. From this, looking at the Windy.com comparison facility, each of their 4 or 5 forecasts is usually within the uncertainty range of the others. When that is not so, trust none.
 
Koeketiene said:
They do - but at 2.8K download speed, it still takes a while.

To put things into perspective: 2.8K download speed is 10 times slower than the first dailup modems back in the stone age (they were 28.8K).
Click to expand...

Actually my first modem was 9.6kbs
 
franksingleton said:
In general the main computer models can be quite consistent from day to day. When the are so it is a good sign that the forecast is on the right lines. http://weather.mailasail.com/w/uploads/Franks-Weather/ninedaystppdart.png shows forecasts on successive days for a point in mid Channel. If there are major changes from day to day then the situation is less predictable than normal.

Do not take a majority view if one forecast is an outlier. Those making such suggestions do not understand ensembles.The odd man out may be best on the day.

Remember that all models smooth out detail less than 4 or 5 grid lengths. That is why models usually under-predict the strongest winds.

As a matter of interest, go to the Meteociel site. Look for ensembles. It shows how forecast winds can vary over their 30-member GFS ensemble. This gives an indication of the uncertainty In wind and other elements. From this, looking at the Windy.com comparison facility, each of their 4 or 5 forecasts is usually within the uncertainty range of the others. When that is not so, trust none.
Click to expand...
Fully agree.
When crossing Biscay we look for perfect model consistency and none complex weather. Anything else and risks of a bad crossing increase
 
You must log in or register to reply here.

Share this page

Top