Projected weather/course app

[Paul L]

Originally Posted by CassidyNZ

No, when I say I do one download a day, I mean I don’t do it twice, not that I only do one model.

Depends where I am too. If I’m ashore and prepping to depart, I download on standard PredictWind and I use PWG and PWE. On a passage I use PredictWind Offshore, I download the same models.

However, I use PWG as the primary and PWE as a comparison. If PWE differs widely, PWG is my sole reference. Interestingly, the forecast in my area for the next 10 days shows both models as nearly identical, would give a high level of confidence.

Interesting. So why use the pw modified models instead of the original GFS and Euro models?

[belizesailor]

Originally Posted by Paul L

You aren't including the cost of getting the additional data offshore. In tough conditions it can take a long and tedious time to download all of the available info. For me, I feel like it is more important to quickly ( at relative term) get a decent Grib and make your decisions. And perhaps put my effort into making sure you the crew has a good hot meal for the upcoming episode.

I download what is effective and easy offshore and so far haven't had the need to try and fill in with more predictions, etc. I do often listen to HF nets to get actual conditions ahead of us that other passagemakers are seeing. This also comes with cost as some Nets can drone on for a long time and be hard to hear.

Yes, there is a cost in terms of time and maybe money. If you've got your hands full aboard it sure is handy to have someone ashore, with better access to resources and not busy running a boat, to handle some of this for you...whether pro or experienced amatuer. Especially true if single/short handed in challenging conditions.

[barnakiel]

Choosing the best model is not easy.


One can say - 'get the one that fits best'. OK. But do you have access to all models while you are out there? Few have.


This is one of the points where a shore based human wx router is good use: I get boat local data in the morning and log it. Then I download data from two models, compare with boat data. OK. Now we are talking. I may choose one model over the other on Monday then shift the preference on Tuesday. etc. Much easier in a dry office with broadband than in a boat - especially if the wx is nasty or the boat is small.


There ARE major difference btw models, unfortunately. Unfortunately, once we get going we may be locked to one service. But an onshre router can shift at any moment they find a model is not relevant or less reliant than the alternative.



Last time from Gudeloupe to Canary Islands (non-stop), our Garmin In-Reach gave us forecasts (based on OCENS, according to Garmin website) that were way off the actual conditions. Too way off to be truly useful.


I relied on Garmin OCENS forecasts too much and in result missed Sao Miguel by some 100 miles. Something that would probably never have happened if I had sailed 'raw' - using only the radiofax chartlets that we used on all earlier passages.


We learn as we go. As sailors, wx routers and consumers.


Cheers,
b.

[belizesailor]

Originally Posted by CassidyNZ

....

But reality is that once you’re out there, it is what it is. Weather prediction is only really helpful to plan departure - once under way, it only helps to diminish the surprises. But if your departure planning is up to snuff, surprises are in any case diminished.

True that the typical cruising sailboat has limited options once underway, but you do have some. Having up-to-date weather details not only helps avoid, or at least be prepared for, surprises, but can be a tremendous help with feasible routing decisions. Example: on the way to the Azores in May-June we worked our way into a sweet spot along the NW edge of the Azores high with lovely moderate SW winds and a 4 knot (yeeha!) favorable current.

[conachair]

Originally Posted by barnakiel

This is one of the points where a shore based human wx router is good use: I get boat local data in the morning and log it. Then I download data from two models, compare with boat data. OK. Now we are talking. I may choose one model over the other on Monday then shift the preference on Tuesday. etc.

Very hard to see how you can tell anything about the accuracy of different models with so little actual data coming back......... Without extensive plotting to determine subtle trends it's hard to see how there's enough to go on. Or if you can then hats off - that's some going!!

Then with such little data to go on how do you know if your guess was right or it's just luck within the noise in the forecasts & extensive noise in the tiny amount of data you've got back? I've spent a lot of time by hand looking at model data against actual, and gave up - (my, anyway ) human bias is just too much to cut through the noise and see actual data trends. Only thing which has been useful is python doing all the work to plot the forecast against actual

On the plus side, once python got involved the forecasts are actually remarkably accurate, in the few locations I've played with anyway, this is UK met office forecast against actual in scilly isles, picked as no land around & data is from 10m, good going most of the time even 4 days out, 5 days ahead was asking a bit much. >>


Back on the thread, try this > open windy.com & pick an offshore data buoy then try forecasting a few days ahead and see how well you do from the different models & synoptics. Might as well get some practice in. And buy these and read them twice
https://www.amazon.co.uk/gp/product/...t_bibl_vppi_i2
https://www.amazon.co.uk/gp/product/...t_bibl_vppi_i5

[Pete17C]

Originally Posted by CassidyNZ

The one voyage that I did in which I used “sources of information” I got information from a renowned weather person in California for forecasting from the US westwards across the Pacific. When I got to Rarotonga I included forecasting from a well-known weather guru in New Zealand. The forecasting received from both led me directly into the only (life threatening) major storm I have experienced at sea.

Before and after that voyage I did my own routing/forecasting and managed several intercontinental passages in total safety. My last 1100nm crossing was done in 7 days without tacking even once (port tack from Fiji all the way to Russell in the Bay of Islands) with only the odd line squall to contend with.

So forgive my scepticism.

There comes a point in a sailor’s life where personal experience and attention to detail trump the information provided by experts who have no exposure to the consequences of getting it wrong. “Good weather routers” are generally not on board when their forecast turns up woefully inaccurate.

Users of PredictWind will know that the software proposes routing based on four different models. They mostly differ from one another sometimes by very significant margins. Most users will have long since realised that one of those models mostly conforms to their thinking and that is the one they consistently reference. On the odd occasion where all the models tell the same story, confidence in the forecasting shoots up but don’t hold your breath for that eventuality.

Multiple models simply confuse me. But then, maybe I’m just not clever enough to tell the good model from the poor one.

The good model is the one that holds true, watch them a few days before your voyage, and the one that seems to be hitting it right, is the one to put more value on.

If the forecast is getting it right currently, you can assume it will continue to do so (until the next forecast comes out, then reevaluate).

And when the view out the window doesn't match the forecast (is it better or worse?), you use your own judgment (will it continue to be better or worse than forecast?).

Then sail (or not) your weather.

[jmh2002]

Originally Posted by conachair

...try this > open windy.com & pick an offshore data buoy then try forecasting a few days ahead and see how well you do from the different models & synoptics.

Originally Posted by Pete17C

The good model is the one that holds true, watch them a few days before your voyage, and the one that seems to be hitting it right, is the one to put more value on.

Windy also has a feature called "Observation vs Forecast": https://community.windy.com/topic/75...orecast?page=1

This enables you to compare what the "Weather Station Observed" with what the various "Forecast Models Predicted" over the last few days.

This works will all of the Models that Windy uses (location dependant): ECMWF, GFS, NEMS, ICON-EU, AROME, NAM and enables the user to check how often the Forecast 'got it right', so to speak.

To use this feature:
- open Windy to the desired location, eg:
https://www.windy.com/-Outdoor-map-m...62.123,5,p:off
- select Reported Wind at bottom right to display all Weather Stations
(I also used the Outdoor Map Layer for clarity, but it's not essential)
- select the Weather Station, eg: I chose Bermuda Airport
- select the Forecast Model, at bottom left
- select the desired period in the Timeline
- the grey floating box displays the Observation vs Forecast
(and some basic colour coding to indicate Right / Ok / Wrong)





Hope this helps too

[barnakiel]

Originally Posted by conachair

Very hard to see how you can tell anything about the accuracy of different models with so little actual data coming back......... Without extensive plotting to determine subtle trends it's hard to see how there's enough to go on. Or if you can then hats off - that's some going!!

Then with such little data to go on how do you know if your guess was right or it's just luck within the noise in the forecasts & extensive noise in the tiny amount of data you've got back? I've spent a lot of time by hand looking at model data against actual, and gave up - (my, anyway ) human bias is just too much to cut through the noise and see actual data trends. Only thing which has been useful is python doing all the work to plot the forecast against actual

On the plus side, once python got involved the forecasts are actually remarkably accurate, in the few locations I've played with anyway, this is UK met office forecast against actual in scilly isles, picked as no land around & data is from 10m, good going most of the time even 4 days out, 5 days ahead was asking a bit much. >>
Back on the thread, try this > open windy.com & pick an offshore data buoy then try forecasting a few days ahead and see how well you do from the different models & synoptics. Might as well get some practice in. And buy these and read them twice

Not surprisingly then, I do not make any assumptions on models' accuracy.


To avoid going into tech details (I assume interesting for few and boring for most members), I will give a most simplified, (but otherwise correct) example of one possible procedure:


- morning session boat spot you get boat data wind NE 20 kts, veering,



- data model A shows morning spot conditions N 20kts, outlook: veering,

- data model B shows morning spot conditions NE 25kts, outlook: veering,



At this point the crew, or their land based router may (all other things being equal) elect a route based on model B, and sail conservatively.


In reality then, except for few truly mind bending cases, choosing the route is a straightforward process - because each boat is limited by her target wpt, distance to target, land masses, obstacles, polar performance and crew objectives and strength.

Wx routers do not forecast wx. The job is to:


- collect, digest and forward wx info to the crew,
- assist the crew in route choice decisions.


I think it is likely that each router may have a slightly different attitude to how many data sources to monitor and how to elect the ones for the job at hand. Possibly much the same like any other person acting in fields strongly dependent on decision making in conditions of uncertainty.


Body count tells you the rest of the story.



Cheers,

b.

[CassidyNZ]

Originally Posted by Paul L

Interesting. So why use the pw modified models instead of the original GFS and Euro models?

In the time I’ve used them the PW models have shown to be more stable and more consistently correct. Checking back on past predictions for current events is the best way to establish accuracy and mostly the PWG predictions show that they were correct even when made 7 - 10 days before.

[CassidyNZ]

Originally Posted by Pete17C

And when the view out the window doesn't match the forecast (is it better or worse?), you use your own judgment (will it continue to be better or worse than forecast?).

Yes, the view out the window. . . . . Unfortunately where I live the view out the window is valueless. I can have glassy smooth water and sunshine in the bay below my home now and in 30 minutes 35kn SE and pouring rain and 3 hours later 15kn NE and back to sunshine.

Besides, I’m not really interested in the weather now at home, it’s more about 3 days time when I’m 500nm from refuge.

[Paul L]

Originally Posted by CassidyNZ

In the time I’ve used them the PW models have shown to be more stable and more consistently correct. Checking back on past predictions for current events is the best way to establish accuracy and mostly the PWG predictions show that they were correct even when made 7 - 10 days before.

I'm surprised you think the two PW models are more accurate than the GFS or Euro model. Both the EU and the US spend 10s of millions of dollars on developing, maintaining and testing historical accuracy. Then a tiny company PW takes these two models and with completely unpublished methods and testing somehow improves them. I've always thought these two PW models were much more marketing than meat.

[barnakiel]

I wanted to add to earlier point where someone noted a wx router was wrong with a rally take off.


I do not know that rally but I can tell you ARC which is a big rally leaves ON DATE (not on weather). I have only ever seen them skip the date (fixed half a year ahead) once in maybe 10 years. I also seem to remember they did not make any adjustments when TS Delta hit the islands (not sure of this but I believe I am correct).


Big rallies are commercial events = all about money (for the organizers and the narrow circle of service suppliers). And time is money. Also rallies do not employ wx routers - there may be 200 boats on a rally, this would imply creating 200 routes a day, sending 200 scheds twice a day. Not going to happen. Rallies are far more likely to employ a pro wx forecaster, I think. And this will create a general wx info for the whole area, not spot info for each particular boat and no route advice for each boat (possibly no route advice at all).


A wx router will typically work in a one to one relationship with each client.


b.

[unbusted67]

Originally Posted by Paul L

I'm surprised you think the two PW models are more accurate than the GFS or Euro model. Both the EU and the US spend 10s of millions of dollars on developing, maintaining and testing historical accuracy. Then a tiny company PW takes these two models and with completely unpublished methods and testing somehow improves them. I've always thought these two PW models were much more marketing than meat.

Yeah, I am with you. I don't buy it. Those two models are like putting a thin layer of syrup over some very well thought out pancakes.

[CassidyNZ]

Originally Posted by unbusted67

Yeah, I am with you. I don't buy it. Those two models are like putting a thin layer of syrup over some very well thought out pancakes.

Does a thin layer of syrup improve pancakes or make them less desirable?

See, it’s hard to argue with success. I have been using PW for quite a while now and haven’t been caught in bad weather since I started. As long as the “thin veneer” that PW places over the base data keeps working for me I’ll keep using it.

YMMV

[Paul L]

Originally Posted by barnakiel

I wanted to add to earlier point where someone noted a wx router was wrong with a rally take off.


I do not know that rally but I can tell you ARC which is a big rally leaves ON DATE (not on weather). I have only ever seen them skip the date (fixed half a year ahead) once in maybe 10 years. I also seem to remember they did not make any adjustments when TS Delta hit the islands (not sure of this but I believe I am correct).


Big rallies are commercial events = all about money (for the organizers and the narrow circle of service suppliers). And time is money. Also rallies do not employ wx routers - there may be 200 boats on a rally, this would imply creating 200 routes a day, sending 200 scheds twice a day. Not going to happen. Rallies are far more likely to employ a pro wx forecaster, I think. And this will create a general wx info for the whole area, not spot info for each particular boat and no route advice for each boat (possibly no route advice at all).


A wx router will typically work in a one to one relationship with each client.


b.

B
Well the rally mentioned above did not start on a particular date. It was made clear to everyone that they should make their own decision on when to leave. The router gave daily briefings for about 10 days or so. A number of boats left days early, most had skippers who were expereinced on this route. Tbey had much better trips. The large bulk of the boats left after getting a briefing from the router recommending it. These are the boats that got in trouble, some seriously.

I heard the briefing and went away hearing, OK it could be 24 hrs of hard sailing then a sleigh ride down for next 6 or 7 days. Sounded good to me. The first 36 hours were far more than hard sailing and most of the rest of the trip was mixed winds and far, far from a sleigh ride. This had nothing to do with a rally enforced schedule. Oh well, can't hit it right everytime.