Support Sailonline

If you haven't already - join the SAILONLINE YACHT CLUB!

Please also consider making a donation - all amounts are greatly appreciated!

Board » General Discussion » Weather - Brain Storm

Page: 1 2 3 Next

Recent chat in the M2O 2020 about the weather that we import into SOL has prompted this new thread, hoping that concerns and ideas may gravitate towards some sort of consensus or conclusion. Whether anything can be done with such a conclusion is another matter.
Remember: SOL is a 'cooperative' and everything that happens is done by volunteers. Any change in the fetching of weather is likely to require changes to the software (the racing engine) that runs on the server. This is no sinecure, and the volunteer resources that we have that can attempt this are scarce.
I'll get the ball rolling with a separate post about the status quo.
Clipping along
Currently, we have the capability to fetch three sources for weather with the following 'tags' - global_gfs, nam_awip12 and wrfems, updating every six hours.
global_gfs is what we use most often and it is rough-and-ready. I'm not hugely familiar with meteorologist terminology but I think it is what is called 'synoptic', which I think means it is primarily based on surface pressure variation; in other words, there is no modelling of local coastal effects. The highest resolution we can fetch is 0.5 x 0.5 degrees, in other words the data - that is then interpolated by the server (or a router) - is given at nodes set 0.5 degrees of longitude and latitude apart.
0.25 degree resolution is in fact available, but we can't fetch it. With 0.25 resolution the amount of data describing the wind over space and time increases by a factor 4, and obviously routing or SOP-ing through this becomes commensurately trickier.
I'll write a second note later.
Clipping along
Thanks for setting up this brainstorm, Jan. My concern is: what happens if GFS (or other WX sources we use) start updating every 3 or 4 hours? I have the impression that even with the every 6 hour updates it is already difficult for some people to combine SOL with their professional or family life (as well as sleep). So, maybe we can discuss that in one go.
But in keeping with my first question I want to submit this article which takes a good look at where weather forecasts could be going in the coming years. Some techniques seem to pave the way for accurate forecasts that are 8 times longer than we now have: https://metro.co.uk/2019/09/18/will-weather-forecast-ever-100-correct-10751641/
That is a very interesting article, Dan, but I suppose it all depends on what NOAA consider 'accurate' or 'correct', when it is quoted to state 'a seven-day forecast is accurate around 80% of the time' or 'a five-day forecast is correct approximately 90% of the time'. One can imagine that the assertion is correct about something like 'it will rain in The Peloponnese tomorrow', but, as we know from experience, it doesn't imply that the wind strength and direction forecast for 120 hours ahead at a specific lat/lon posn will be re-forecast as exactly the same 6 hours on, now 114 hours ahead.
I would like to think that that sort of accuracy will remain impossible for very many years yet. So, that aspect of our game - forecast volatility and second-guessing the met service - is not much at risk, IMO.
I do agree with you that once gribbed free-issue forecasts become available with a frequency of every 3 hours, it would reduce the attractiveness of our game if the game were to fetch these every 3 hours as well. I see little reason (though not none) for simply keeping our own fetch frequency at 6 hourly; especially if 'accuracy' has improved anyway.
Indeed, I would suggest it could mean an opportunity to reduce our fetch frequency to every 9 hours, which would give everyone around the world a midnight WX every couple of days.
Clipping along
My agenda is not so much a concern with ever more and better forecasts, but a concern with the rough-and-ready nature of global_gfs, our standard 'go to' weather. The question and the challenge I want to raise is 'Can we do better?'.
For the long distance ocean races, this question is not so important, as, out in the middle of the ocean, modulation for whatever should change little to a forecast, and because guessing what is the best thing to do given the uncertainty about what the wind will really be a week ahead will continue to be a challenge for SOLers.
It is the races of say 3 to 4 days maximum that I feel/find are or have become somewhat mathematically deterministic. You have to have the tools to do the math, but once you have such tools working for you, a race becomes primarily a "do" exercise.
Clipping along
I'm sorry to spoil the party but I'd prefer to always sail on WRF weather. Obviously that's not "feasible" in the most straightforward manner for races over large areas. However, if it would be possible to limit calculating the wrf forecast only to areas where there are boats, where the boats can reach, and only calculate very short-term wrf, like 12h-15h (mostly to cover across a single wx update failure), it could bring down the amount of processing power needed for the forecasting.

So we'd be having sort of composite wx with a base layer of 0.25x0.25 gfs to visualize longer term forecast and act as a backup wx. Which parts of the base layer would be selectively replaced by wrf would be dynamically determined as the race progresses.

Whether we have computation capability to do that (or would be able to get the necessary processing capacity) I have know idea but I certainly would want to know how big area would need to be calculated. Even in ocean races boats tend to keep in packs (although wrf itself might impact the dynamics that typically lead to it with gfs). The biggest problem in this approch IMHO is the long tail of boats because they'd require disproportionate amount of the all processing.

If that turns out still infeasible, some additional filter could be applied which allows using wrf only under conditions where wx seems unstable (based on the corner wind components it's possible to estimate this I think) or when there's landmass.

It's very complex to realize technically though, we cannot even run the 0.25x0.25 gfs part of that currently so it may remain wishful thinking.

I agree with bonk that 80% reliability is really a joke. Even 90% is not really something I'd boast about. It's worth to remember the error starts to rapidly accumulate. It's unlikely to be dropping linearly but more like larger error also leads to much larger error accumulation (and thus uncertainty).

I've never seen the definitions of that kind of % metric anyway (so if somebody has pointers to some scientific papers evaluating forecast long-term accuracy, please, I might read myself). But I wouldn't be surprised if it turns out quite bogus or irrelevant to measure what kind of accuracy we're talking here.

Comparing 0.25x0.25 wx grid with 1x1 (I forgot to make the arrow density the same so please disregard that difference):



--- Last Edited by ij at 2020-09-07 22:18:39 ---
Just a minor note from some ECMWF validation report now that I came across one: in 10m wind at 5 days the frequency of large errors is around 4% (+/-1 due to year cycle) now so it's in ballpark with those 90/80% numbers. (And in temps slightly worse.)

GFS is likely somewhat worse than that given ECMWF is generally considered slightly better although both can be shown to be better in selected cases.

What is means to routing accuracy is hard to know without looking deeper into the report (if it's told at all). The distribution of the large errors over time matters too much. But if distribution would be uniform (it very likely isn't), one would very likely hit one large deviation in almost every 2 weeks long race. And how bad the large error % gets when consider beyond 5 days, no idea.


--- Last Edited by ij at 2020-09-09 21:01:19 ---
Hi Bonk and Ij,

I don't know how accurate the weather models will become. Even the experts don't agree on that question. We can probably just agree that they keep getting better and look at possible consequences for SOL simulations.

I agree with Jan on the possible problem that shorter races are moving closer to "fixed solution". If there is only 1 (or even no) WX change during a race we are on a fixed weather map where good math tools can figure the winning route and then it just becomes a question of who executes that route more optimally. But some SOLers like that kind of challenge, so I wouldn't take that away.

If long term weather forecasts get more accurate then even the medium term (3-4 day) races would become more like sprint races where execution of near optimal route is more important than anticipation of weather changes.

In the long ocean races it will not be anytime soon that weather models can give us a near optimal route for more than 2 or 3 days. But if WX updates become more frequent then we would have to think about SOLers' sleep.
We can decide to skip some updates and only update the maps that SOL uses every 6 or 8 hours. But even in that case some players would gain an advantage by looking at the GFS updates we skip, because they will already show in what direction the weather is changing. Perhaps it would be more useful to change to a more detailed, but less frequently changing, source of weather data as ij suggests...

--- Last Edited by Zorba777 at 2020-09-10 08:47:58 ---

Page: 1 2 3 Next

Please login to post a reply.


Next Race: 00d 00h 00m

Current Races:

Sailonline's Tall Ships Re-Run 2023 - Lisbon to Cadiz
Sailonline's Tall Ships Races 2023 - Lisbon to Cadiz
Race starts: Oct 01st 14:00 Registration will open soon

▶ Flash

Cairns to Darwin 2023
This race was originally Leg 7 of the A3 series - AGage's Around Australia, which SOL dedicated to AGage's memory. This time it is the ninth race of the All-Round Championship 2023 Series - 1405nm in Orange 125’ Catamaran from Cairns, the hub of Australia's tropical holiday zone, to Darwin, the frontier with Asia.
Race #1724
INFO by brainaid.de
WX Updates:
0430 / 1030 / 1630 / 2230
RACE CLOSE: Saturday,
September 30 at 2300 UTC.
Race starts: Sep 21st 05:00 Registration Closed

▶ Flash

Carib Rum Run 2023

Welcome to the Caribbean and Sailonline's annual Pirate race, running with International Talk Like a Pirate Day on 19 September! This is a 1759nm slalom through the rum islands of the Caribbean in our glorious Falkun superyachts! Our mission is to collect barrels from the region's distilleries and carry them to the rum-parched beaches of Cancun, Mexico. Join SYC before race end and ye befuddled scallywags can disguise yer boat for this race only with a pirate name!
Race #1698
INFO from brainaid.de
WX Updates:
0430 / 1030 / 1630 / 2230
Ranking: SYC
RACE CLOSE: Tuesday,
October 3 at 2300 UTC.
Race starts: Sep 16th 15:00 Registration Closed
▶ Flash

Sinbad Braves the Bay of Bengal 2023
“Sinbad, you sorry sailor, why have you not told me of this great sea, this giant bay, stretching hundreds of thousands of qasbah east of Al-Hindi to fabled lands where tea and rice and rubber is abundant?” and Sinbad replied “Ahh, umm, shall I go have a look. May I take the Judel Vrolijk 62 again, it’s going to be a long voyage, perhaps a million and one half qasbah (c 3000nm)? I’ll start in Chittagong and follow the coast to the southern tip of the Hindus’ land, round Serendip, head east into the vastness of the seas, and if I’m lucky (no pun intended, oh Mighty One) I shall find land again and then head north and see where I end up”. “Fair winds, Sinbad”, said the Sultan.
Race #1641
INFOby brainaid.de
Judel Vrolijk 62 PARTICULARS
WX Updates:
0430 / 1030 / 1630 / 2230
Ranking: SYC - SVS
RACE CLOSE: Saturday,
September 30 at 2300 UTC.
Race starts: Sep 13th 08:00 Registration Closed

▶ Flash

Maryland to Alaska 2023
Welcome to this September edition of 2023 OCCH. This route was first sailed in 2020. At the time it was inspired by Matt Rutherford's single-handed Red Dot Expedition around the Americas in 2012, SOLer Renegade proposing a half-around from Maryland over the North into the Pacific and back through the Panama Canal, in 5 legs, 5 years later. Leg 2 across the North West Passage proved to be particularly challenging and so to create a proper Ocean Race, SRC combined Leg 2 with Leg 1 to give you a race of at least 4000nm from Maryland to Alaska, now to be sailed in our magnificent Archer 78.
Race #1719
INFOby brainaid.de
WX Updates:
0430 / 1030 / 1630 / 2230
October 1 at 2300 UTC.
Race starts: Sep 01st 14:00 Registration Closed

▶ Flash

Go to race archive

SYC Ranking

  1. Sailonline Yacht Club Member WRmirekd
  2. Sailonline Yacht Club Member FreyjaUSA
  3. Sailonline Yacht Club Member Pit8008
  4. Sailonline Yacht Club Member Sax747
  5. Sailonline Yacht Club Member rafa
  6. Sailonline Yacht Club Member Vida_Maldita
  7. Sailonline Yacht Club Member Kipper1258
  8. Sailonline Yacht Club Member sassy63
  9. Sailonline Yacht Club Member HappyHour
  10. Sailonline Yacht Club Member AL53_SSIta

View full list


Mobile Client

SYC members have the benefit of access to our mobile/lightweight web client!

The mobile client