Weather Forecast Models

[mvweebles]

Unless you know why the euro model was correct, it could be dumb luck.

The financial world is full of advisors who get it right once in a while. Is it brilliance, or is it the rooster taking credit for the sunrise? I don't know.

Forums like this are full of posts about how someone got their ass handed to them because a forecast was wrong. My thinking is that you need to see the pressure systems to give you highbeams on what might change. Perhaps all you need to do is just choose the Euro model and forget about learning any meteorology. But best would be having access to the underlying data - where is it? Would allow a mariner to anticipate much better. Using Ian as an example, if Euro was unwavering, was for a reason. What was the reason? For GFS, they were blunt about the 3d chess of pressure systems that broke down and caused rhe storm to move. Made perfect sense. What were the underlying forecast models of Euro?

Does beg the question, if the Euro model is so accurate, why doesn't the GFS model just use the Euro model and call it good?

Peter

 

[Jc180]

Unless you know why the euro model was correct, it could be dumb luck.

The financial world is full of advisors who get it right once in a while. Is it brilliance, or is it the rooster taking credit for the sunrise? I don't know.

Forums like this are full of posts about how someone got their ass handed to them because a forecast was wrong. My thinking is that you need to see the pressure systems to give you highbeams on what might change. Perhaps all you need to do is just choose the Euro model and forget about learning any meteorology. But best would be having access to the underlying data - where is it? Would allow a mariner to anticipate much better. Using Ian as an example, if Euro was unwavering, was for a reason. What was the reason? For GFS, they were blunt about the 3d chess of pressure systems that broke down and caused rhe storm to move. Made perfect sense. What were the underlying forecast models of Euro?

Does beg the question, if the Euro model is so accurate, why doesn't the GFS model just use the Euro model and call it good?

Peter

Two points and then will call it a night.
1. The Windy App does provide graphic displays or layers including surface pressure, wind, waves, precipitation, etc which are actually outputs from the model you select. So the Euro model data you are seeking is right there.
2. Suspect some politics involved in US Govt favoring their own model over the foreign competition.

 

[OldDan1943]

I rely on the “cone” and it’s movement as it moves closer to shore.

 

[mvweebles]

Two points and then will call it a night.

1. The Windy App does provide graphic displays or layers including surface pressure, wind, waves, precipitation, etc which are actually outputs from the model you select. So the Euro model data you are seeking is right there.

2. Suspect some politics involved in US Govt favoring their own model over the foreign competition.

No, Windy does not provide the data, it provides its algorithmic interpretation (and interpolation) of both models. In a sense, Windy is a "cover band."

When boaters complain about weather forecast being wrong, usually carries cynical comments about how it was forecast at x, but they saw 2x. They never know why, just that it happened. No idea what really happened. Did they look at the weather for Saturday 4 days prior and were surprised? Did they misread? I can tell you one thing that didn't happen - they didn't track a pressure system.

I have a fairly strong sense where NOAA gets their data. And I know where to access it. I can dig as deep as I want. I have no idea where to access the underlying analysis of the Euro data. If it's more accurate, why? What do they do differently? Where do they get their data?

Peter

 

[Alaskan Sea-Duction]

Good info here. Crossed Dixon yesterday. Windy was spot on. Following seas, comfortable cruise

 

[mvweebles]

Interesting article on hurricane Ian and forecast models.

https://www.tampabay.com/hurricane/...ne-models-hurricane-ian-gave-us-some-answers/

A couple takeaways

First, hurricane modeling is much different than daily weather forecasting due to the stakes involved. Separately, I know that after Andrew in 1992, a lot of effort was put into improved forecasting.

Second, of the roughly 27 hurricane models out there, GFS may have been further off the mark than Euro, but the Euro model didn't a great job either. A UK model was better for Ian.

Finally, according to the article above, Euro tends to be a bit more accurate, but not always. Of the 27 forecast models for hurricanes, GFS too tops in 2021.

Windy and PredictWind are much better than what was available just a few years ago. But if you're cruising involves more than 72 hour passages in exposed waters, you will be much more nimble if you track pressure systems with a basic understanding of synoptic charts. Otherwise, there will be a constant draw to leave day after tomorrow.

Peter

 

[heysteve]

For those using PredictWind, they have a Validation page where you can compare model accuracy. It has limited functionality but is interesting. Here is the page for La Paz.

In the SoC, I've found the NAM and HRRR models best probably because of their high resolution (lots of microclimates here). Unfortunately, they only go out a couple days so for longer term trends I've found the ECMWF model best. The GFS model just doesn't seem that accurate here in comparison to the alternatives.

I typically just use Windy to view the various models though I do like the PW split screen for comparing forecasts.

 

[rgano]

I tend to believe most models when the predicted time is about one hour away.

 

[Davil1]

So two things that have worked for me.
1) I have the buoys from Florida to Nova Scotia on my PC and I watch them as I approach the areas - wind speed and direction - wave heights and direction for the last 6 hours to see if they are growing or shrinking.
2) I then look at the forecast and ask if the forecast is off by 10 knots in the wrong direction would I want to be out there?

If the answer is no - I stay where I am and wait...
Stay Safe.

=========================
sorry, dumb question
how do I get them?
thanks

 

[Marco Flamingo]

I’ve found the “gust” predictions far more accurate. If I’m happy with the gust forecast on the worst model, I’ll probably go.

That is what I've started doing. I'm sure there is some algorithm that decides what is a wind speed and what is a gust. Also when do the gusts become frequent enough to be the wind speed? If the forecast is 5 to 7, but the gust is 20 to 25, I'm not doing the crossing.

 

[Davil1]

=========================
sorry, dumb question
how do I get them?
thanks

================
what is the link for

buoys from Florida to Nova Scotia ?

thanks

 

[murofrelwart]

================
what is the link for

buoys from Florida to Nova Scotia ?

thanks

https://www.ndbc.noaa.gov/obs.shtml...351&zoom=3&type=oceans&status=r&pgm=&op=&ls=n

 

[Davil1]

great
thank you

Have this

https://www.wpc.ncep.noaa.gov/#page=ovw

 

[cafesport]

For the new guys here, there was a member named Richard who was a retired Air Force weatherman based out of Alaska. He became interested in trawlers and crossing oceans.

For Richards crossings I furnished him numerous ship readings for wind direction speed barometric pressure air and sea temps from my Sirius weather subscription. With that info he created his own weather charts and then modified them based on his local out the window observations. Those in combination with the published discussions kept the big surprises away. Here is a link to what Richard used to recommend when asked what he used.

https://www.trawlerforum.com/forums/s30/what-do-you-use-weather-28392-2.html#post580582

 

[tpbrady]

Richard and I were together in Wrangell for a couple of years before he decided to sell his boat and move to a small farm near Austin. Always fun to talk to on a variety of subjects. His comment about modifying forecasts based on his local observations is what I have settled on. The problem with that method if you don't have internet access and have to depend on NOAA weather radio, is that you have to depend on NOAA weather radio. It's something that hasn't changed in almost half a century. They only issue forecasts every 12 hours, repeat them on no apparent schedule, follow it with observations that are incomplete, and start the process over again. I can't count the number of times my VHF reception is marginal and just when the observation I want to hear is coming up, it fades to static and I have to wait 5 minutes or so for it to repeat. The whole world is digital except for weather radio. On the radio channels that they use, I could send the entire forecast and all the observations in a couple minutes as a data stream, much like how Sirius radio does it.

That's not going to happen until long after I sell my boat, but Starlink has changed that. With Windy I can see all the observations I need and compare them to the forecasts using several different models and pick the one that best fits the real world. I would like to say I could use the NWS Juneau Office web site for the observations, but it is so disorganized as to be useless. Windy has all the observation locations I am aware of from multiple sources including occasional ship reports presented on a map so you know where they are. The NWS forecasters may be good, but the web developers should be supervised by someone trying to use their web site.

Tom

 

[DDW]

PredictWind has their own models, also you can see the ECMWF and GFS and several others. The graphic depiction adds no value other than visualization, in theory the ECMWF viewed on Windy and PredictWind will be the same (and from observation, they are). The GFS model is pretty old and seems to be a little less reliable than the ECMWF in my experience. PredictWind' algorithm (PWx) is run with both GFS and ECMWF initialization. In areas where they have the 1km matrix, it seems to be significantly more accurate, as it accounts for local variations and terrain much better than the 8km matrix which both GFS and ECMWF use. Unfortunately they do not have that except in populated areas (like Puget Sound) as it is very expensive to run.

Starlink would let you get these predictions everywhere, as long as Elon allows it. I've not submitted myself to his whimsy yet. For places where VHF is not reliably available (it's been about half the time on this trip for me) SiriusXM has been pretty good. They show a GFS grib and also have the NOAA and Environment Canada marine forecasts. When I am way north anchored in a bay with high hills to the south, I can't get it, out in the channels away from the hills it has worked.

 

[NSM]

Most balloon sounding locations provide data twice a day. Hence the update schedule for most models. And there aren’t many sounding locations. Only 2 in Washington State and two in Oregon (I think). Makes for a spatially and temporally sparse data set from which to run models. Some modelers may provide forecasts with more granularity but it may not be any more precise.

And which model works best for each microclimate may vary…

 

[Cigatoo]

Another glitch in forecasting for coastal cruisers are seas running in and around inlets. Here in Rhode Island / southeast Ma you have to deal with body’s of water colliding. Point Judith, Block Island Sound, Long Island Sound, Rhode Island Sound, Buzzards Bay, Cape Cod Canal, and Vineyard Sound can all be a real mess under otherwise benign weather conditions. Almost impossible to forecast sometimes. You look at tidal currents and wind then hope for the best as there are no buoys in close to shore showing how the seas are running. The “washing machine” effect can really kick your butt.

 

[Fletcher500]

I generally look at everything previously mentioned, including a subscription to Bouy Weather, which are the same guys who started Surfline years ago (surf forecaster known to predict swells better than others). I throw out the low and high, and generally find I get the median which is usually good data, and much appreciated. We are fortunate to have it compared to what people previously had to deal with before the technology came on line.

 

[Alaskan Sea-Duction]

The USCG has an app, among other things, you can pull up real bouy infomation. Very helpful. Environmental Canada also has real time information from light houses.

 

[tpbrady]

Virtually all observation locations in Southeast Alaska are on land and not all are US Government. The only app I have found that includes all observations from any source including ship stations is Windy. I am sure there are others.

Tom

 

[sjisailor]

For most of us, I imagine the most crucial boating forecasts are for 0-2 days out. For those GFS has no advantage over NAM and HRRR for US waters. GFS offers 3-16 day forecasts but with increasing uncertainty as the days add up. (Plus global.)

I would say the ECMWF is roughly on par with HRRR for 0-2 days, but in the PNW the ECMWF often gives bizarre and extreme forecasts in the 5+ day range.

Rather than trying to pick a winner, however, I agree with several posts that it is helpful to check multiple models, and then decide based on the mission and one's own tolerance.

Also, a GRIB viewer app like LuckGrib is useful for real weather geeks but requires some investment of time to learn.

 

[mvweebles]

.....Rather than trying to pick a winner, however, I agree with several posts that it is helpful to check multiple models, and then decide based on the mission and one's own tolerance.

Sjisailor - using this post as an example. As you mention, common response is to view multiple models.

How exactly does "model shopping" work? Do you take the most benign forecast because that's the one you like, or the most dire one because that's the one you fear?

In the old days, cruisers would huddle in certain places waiting for a wx window - FLL to go to Bahamas; Huatulco to cross Teujentepec. Invariably, the least confident would set the pace - the wx was always better if they waited another day (or more).

How does model shopping help? Or do you just wait for a cosmic convergence for them all to align? Consulting multiple sources reminds me of a saying from my dearly departed father: "a man with a clock always knows what time it is. A man with two is never sure."

Peter

 

[sjisailor]

Sjisailor - using this post as an example.

How exactly does "model shopping" work? Do you take the most benign forecast because that's the one you like, or the most dire one because that's the one you fear?

In the old days, cruisers would huddle in certain places waiting for a wx window - FLL to go to Bahamas; Huatulco to cross Teujentepec. Invariably, the least confident would set the pace - the wx was always better if they waited another day (or more).

How does model shopping help? Or do you just wait for a cosmic convergence for them all to align?

Peter

It should not be model shopping, but using multiple fallible sources of information.

When the models disagree, it is a "game theoretic" choice. One has to make a judgment about the upside and downside consequences relative to the situation. If it's just me, and I very much need to get somewhere then I would accept worse conditions than when there are others on board for a pleasure trip.

If that's still uncertain then the answer defaults to "no go". (Of course uncertainty always remains!)

 

[mvweebles]

I guess it really comes down to what style cruising folks are doing. My guess is 98% of the folks using weather forecasts are making day hops. - maybe a run from Florida to Bumini or such. Not 48hr-72hr runs. In my opinion, it takes a bit more understanding of the movement of the pressure systems, not just what the output of the algorithms may be.

 

[rslifkin]

Sjisailor - using this post as an example. As you mention, common response is to view multiple models.

How exactly does "model shopping" work? Do you take the most benign forecast because that's the one you like, or the most dire one because that's the one you fear?

In the old days, cruisers would huddle in certain places waiting for a wx window - FLL to go to Bahamas; Huatulco to cross Teujentepec. Invariably, the least confident would set the pace - the wx was always better if they waited another day (or more).

How does model shopping help? Or do you just wait for a cosmic convergence for them all to align? Consulting multiple sources reminds me of a saying from my dearly departed father: "a man with a clock always knows what time it is. A man with two is never sure."

Peter

Sometimes it's a matter of how they align, or how they're changing each time they update. Other times, it's that yesterday and today's forecast from model X were much closer to the actual conditions, so I'm more inclined to trust that model over the others for tomorrow in this location.

Recent example was this past Sunday. NOAA forecast was for 10 - 15 kts from the Northeast becoming 15 - 20 from the Northwest. They were right about the wind shift, but the actual wind speed was in the 4 - 7 kt range all day. The free "blend" model from Sailflow was much, much closer to reality.

 

[Hippocampus]

Random thoughts

During the pandemic there were less ships at sea and fewer airplanes flying over it. The usual sources I used for passage quality went way down. In statistics there’s a concept of sampling error. In the grossest terms the fewer samples the greater the possibility of error.
Also in statistics there’s a concept of deviation from the mean. If you have a system which has wide fluctuations naturally it’s harder to make predictions than a system which has less deviations. It requires more samples and those samples need to be closer in time in order to have an accurate model created . A recent example is the forest fires in eastern Canada causing rapid temperature and wind fluctuations in New England over the last couple of days. We talk about MMCC but don’t fully consider that indirectly is causing wide rapid (days not epochs) fluctuations in weather. That creates much more difficulty in weather prediction. We have not increased sampling adequately so predictions have suffered. This has particularly hit farmers hard. Inability to predict frost nights has made cranberry growers in my area a much more strenuous undertaking.
Gribs show behavior in relatively large fields. The scale maybe appropriate for airplane pilots traveling high above land rapidly but is problematic for low speed vessels like ocean going sailboats or coastal boats traveling on the surface. Behavior in coastal settings is effected by hyperlocal factors. Refractivity of nearby land. Height of near by land. Depth of water and hence its color. Conformation and its effect relative to the front.
Because of these factors all too frequently I’ve sat in 30-40kts when the girb field has said 10-20. For the sailor all weather is local. This is particularly true when coastal where they’re maybe dramatic differences just a mile or three away. The girb is not granular enough to be of much help in many coastal settings.the predictions are based on assumptions for a larger field than is useful in that setting unless you are near shore (25-50nm off shore) . When truly coastal or inshore or inland waters they can be quite misleading. Changing the magnification of your view doesn’t improve the data.
I find placing and looking at multiple small boxes on the NWS zonal along my expected transit line much more helpful. Also look at the chart and google earth so I can think about how local geography will effect weather and waves on my path.
For wind waves and swells if looking at a girb field and coastal you need to apply another level of analysis. Need to think how that general non specific information of the girb prediction will interact with your local geography. How things will reflect, refract and interact. Same with compression zones between islands or at the mouths of rivers with high banks. Same with wind. Prime example is coastal mountains and downdrafts.
So if you’re using gribs you need to have a very jaundiced eye. You really don’t care how the girls field looks like. You do care about the wind and waves right where you are and are going to be. Example-on our recent transit from Hilton head sc to Deltaville va the predict wind and windy looked the same day after day for wave height and wind speed/gusts. 90% of the days were small craft from nws. But we had a pleasant, low stress trip with the SeaKeeper turned on only for a couple of days when the day included crossing sounds where I predicted unpleasant chop would occur. I did glance at windy but it was the least important in deciding travel days or the decision to turn on the SeaKeeper in the morning. Totally agree for nearshore and offshore it’s worthwhile to look at all the models. Personally prefer arrows so use passage weather which is free (I give a donation annually). Find you lose details with the fancy colors. But I’m taken aback by how many use to pay walled commercial products as their sole source of information.

 

[heysteve]

Model resolution is often overlooked as well. The wind flow view that seems to be the default on both Windy and PW is the interpolation of the forecast points. In the open ocean this probably isn't an issue but if you're surrounded by islands or other coastal features it might be important.

Looking at Windy and PW, the GFS resolution is reported as 22 km/25km, ECMWF is 9km/8km, HRRR is 3km/3km, NAM is 5km/12km respectively. Both the PWG and PWE models are 1km. Note this is a precision issue, not necessarily accuracy

To better visualize the forecast resolution, you can go to the PW page for wind forecasts. In the upper right you have the dropdown for model selection. In that box you can choose to see flags instead of flow which will show you the forecast data points. I don't think Windy provides this feature.

 

[Fletcher500]

I am not saying the models are perfect and I have definitely taken my licks with conditions that were worse than predicted. Including 2 years ago getting caught by an unpredicted lightening storm and subsequent clearing winds that lead to a midnight exit from an offshore island cove. However, overall I think the resources available are very good.

The screenshot from Windy shows Wx from the various models for my area this week. Not a major difference between any of them. Again, I am not saying Wx tools are perfect and they can vary quite a bit but overall they are pretty darn good compared to what we had in the past and I am grateful to have it.

 

[Hippocampus]

Much truth here

Looking at Windy and PW, the GFS resolution is reported as 22 km/25km, ECMWF is 9km/8km, HRRR is 3km/3km, NAM is 5km/12km respectively. Both the PWG and PWE models are 1km. Note this is a precision issue, not necessarily accuracy (From post#58).

Post #59 is correct as well. Things are better than in the past at least out to 3 days. But still contend to need to remember what you read or was taught in a weather class and think-well here’s a spot I’d be more comfortable 5-10 miles further offshore or I’d better head in and use the headlands to break up the wind/waves or better hang on the anchor for another day even though it says it’s only 10-15 tomorrow because it’s still wind against wave.
Being dependent upon an outside source for your decisions is wrong. Need to have multiple sources . The difference between the models isn’t a big issue in my mind. They are variations in one source (computer modeling). Peter stressed the importance of reading the synoptic. I’d add the 500mb when thinking about when to time a vacation cruise and whether I’ll go north or south. I’d further add local observations and how they are trending and predicted to trend for day to day planning long distance transits. I still incorporate my observations. Pressure, what I see as clouds, movement, winds, local tide compared to charted prediction ( very wind dependent in parts of US SE). View the ascendancy of commercial girb products has lead to some people not using the skills we had and used in the past. Hence they are disappointed when there are incongruities.