Zezo forum:

Post by NOR7034 | 2017-12-12 | 18:22:47

Are we about to see a new situation where the VR and NOAH weather forcasts differs? For us in the north NOAH expects the wind to turn to approx. 270 degrees in about 2 hours. No such change is visible in the VR forcast. The difference will be around 10 degrees in a few hours. Or is it just me?

45 Comments

Post by YourMomSA | 2017-12-12 | 18:31:28

I agree that something looks wrong. Zezo is telling me to gybe in an hour or so, and the angle it's telling me to gybe to is 12 degrees off of what will actually happen in VR if I proceed with the gybe. Not sure if the problem is in Zezo or VR, but I'm guessing VR.

(I assume you mean NOAA, and you're referring to NOAA's GFS model)

Post by Rostef | 2017-12-12 | 18:33:12

I have the impression, the wind we see in the game is still the gfs files from this morning.
And the result is quite different :-)

Post by zezo | 2017-12-12 | 18:39:45

The VR wind data looks fine on the outside right now
Tue, 12 Dec 2017 16:32:31 GMT
Tue, 12 Dec 2017 16:30:03 GMT

Also seems to fine on the inside because the proper file is downloaded:

https://static.virtualregatta.com/winds/live/20171212/12/006.wnd

It seems that the last version of the interface tries to sove traffic by not refreshing all wind data all of the time but is overdoing it and missing the updates.

Post by MichelleWhy | 2017-12-12 | 18:36:24

I'm a bit concerned as well. From an earlier routing it told me to gybe in about 3 hours. A more recent routing says in an hour. Both wouldn't make much sense if the wind does not shift. Worried. Gybing angle of like 90 degrees doesn't look right to me :P

Post by feriber | 2017-12-12 | 18:36:49

It's very strange.
In the chart if you put your mouse over a dot, it tells yoy one wind direction. But if you move it slightly, the it tells you the chart wind direction, with almost 10 degrees difference.

Edit: Now it has suddenly changes, as it seems to be using 06z(12z+213)GSF file now, while 10 minutes ago was using 06z(12z+177) GSF file.

Post by zezo | 2017-12-12 | 18:41:48

In my chart wind map is exact only in the current moment and then on every 3-hour mark, so you can expect only black dots on thick dark blue lines (and on the line exactly between them) to match the other display.

Post by NOR7034 | 2017-12-12 | 18:41:41

Sorry, - NOAA is the weather forcast provider. Looking 12 or 24 hours ahead things just differ more and more. And with the weather system approaching behind us it would be nice to know how we get on track again....

Post by MichelleWhy | 2017-12-12 | 18:49:46

Yes, everything back to normal now.

Post by NOR7034 | 2017-12-12 | 18:54:41

Thank You Zezo, whatever You did. With gfs.t12z+237h we are back on track! I am so happy to be sitting in my living room now when I see what the real boats will be fighting with the next 48 hours....

Post by zezo | 2017-12-12 | 19:01:57

There is this thingie that VR interpolates between the old and the new weather model for the +6h point and the router does not.

What happens is this, times in UTC.

at 15:00 we all use the old data.
at 15:30 new data starts coming out
The data for 18:00 is 20% new model and 80% old model at VR. Here it is all the old
If there is a big difference between the model runs it starts to become visible
at 16:30 the VR mix is 50:50 old/new data
I switch to the new data
Things start to converge and by 18:00 they are exactly the same again.

90% of the time all of the above can't be noticed because the difference is small, but now and then some low pressure system decides to change direction abruptly or you happen to be in the center of a High with very unpredictable winds.

Post by toxcct | 2017-12-12 | 22:13:28

16:60, what hour is that Cvetan ?! :D

Post by zezo | 2017-12-12 | 22:15:28

It's about time to open a beer at our latitudes, possibly not the first one for the day ;-)

Thanks. Will fix it right after I hit Enter now.

Post by YourMomSA | 2017-12-12 | 19:05:07

Thanks for the outstanding explanation!

My interpretation is... If VR and Zezo have dramatic differences during an update, don't make any sudden reactions. Be patient (and stay awake, if necessary), and wait for them to converge before making a decision.

Post by zezo | 2017-12-12 | 19:21:46

Exactly. You have to wait for the full week of data to get meaningful forecast for a long route anyway, and that happens at about the time of the changeover to the new model.

In the beginning I was applying the new model right from the start, first thing at 15:30 and that would be more accurate in the long run, but would cause bigger differences in displayed numbers at first.

Post by Ursus Maritimus | 2017-12-12 | 20:01:08

best would be not to match real weather forecast, but VR weather forecast ;)

hope the real boats are not navigating on zezo.org ...

Post by Sir_Petrus | 2017-12-12 | 21:47:37

Now I know why I was the only one to gybe!
Can not take a nap anymore...

Post by Flying Dutchman | 2017-12-12 | 22:58:31

Well Sir Petrus, I gybed too:) and Agoodnight sleep.....
Now I'm confused!!!
Gybe or not to gybe?
Assume we will know in 24 hours....:)

Post by Sideshow | 2017-12-12 | 23:06:25

The thing that doesn't make sense to me is why the 1st 6 hours of wind changed between the 1st refresh at 15:30 and the final one? I've noticed this a couple of times this leg, which doesn't feel like normal behaviour for the model. I'm not sure exactly when it switched, but sometime after about 15:45 I'd guess.
When you say 20% new model and 80% old, I've always taken that to mean you the closest 20% of time periods are 100% new and the later 80% of time periods are 100% old.
Might just be worth a check.
Thanks as always

Post by zezo | 2017-12-12 | 23:35:42

I missed a detail.

The weather between 15:30 and 18:00 is a triple interpolation.

1) It's the usual linear interpolation between 15h and 18h gfs
2) The 18h gfs itself is interpolation between 6Z+12 and 12Z+6 files.

So the weight of the 12Z+6 data is something like 5% at 15:30, because the entire 18h data point has 20% weight and 20% of that come from the 12Z+6

The intention is good - not to have abrupt changes even if model changes a lot. Even with wind coming from the opposite direction you won's see more than 5% jump in wind data from one iteration to the next.

The problem with that approach is that it can also affect the data between 18 and 21h (if not executed properly. I don't say if it is or is not, but a slightly naive approach could calculate the interpolated wind data for 18h using the current (16h) data point and then show you the 20h winds using that data interpolated with the 21h data) Once again, that's a pure speculation.

Now, I'm not 100% positive on all of the above. Being 100% positive would mean that I've recreated it myself and confirmed that results match within 1% error margin and I haven't. But I'm pretty confident that it's the way it's supposed to be.

Post by zezo | 2017-12-12 | 23:41:20

It is possible to confirm the above assumptions just by observing the numbers, but it's very hard to do in reality, because:

1) The game shows no position for the wind marker, therefore you should select visible grid confluence points for your samples.

2) Displayed result is rounded to one digit speed/ one degree direction which introduces at least 1% error.

3) You have to find good suitable unstable forecast conditions, and that does not happen on every update.

Given all of the above, you can sit down and write the predicted wind for few geographical points, for few points in time, every hour on the hour between 15Z and 21Z, decompose the data into U/V components and see how the numbers add up.

For me it would be easier to recreate the entire logic and see if it works for my current boat position while playing the game.

Post by Flying Dutchman | 2017-12-13 | 00:59:27

Cvetan,
I try to use your charts with predicted positions.
Now (2340) it tells me agan to tack to SE?
Dont get much of all the above zeros and ones talk..... Am I doing something wrong in using your charts?

Post by zezo | 2017-12-13 | 01:03:43

Now there is new GFS model out, having the same problems as the ones 6 hours ago. And that happens every 6 hours. Such is the nature of weather forecasting itself, other differences are small relative to the actual forecast change.

Post by Ursus Maritimus | 2017-12-13 | 01:25:37

so why did you just tack if the WF is not so reliable at the moment? it changed about 20 minutes ago and tells me now to gybe 6 times tonight.

not gonna follow that advise as it would burn all my programming cards, so just one gybe tonight arond 03:20 CET to hopefully end up in more or less the same spot and not loose to much ground

Post by zezo | 2017-12-13 | 01:28:08

Got enough PR cards myself ;-)

I've been quite busy in real life recently and will be playing this leg almost in robot mode following the router without much thought.

Post by Flying Dutchman | 2017-12-13 | 01:09:29

So whats more acurate at the moment? Follow the VR router or your router:)
Yours seems to sail me into low winds if I look at the VR forecast

Post by zezo | 2017-12-13 | 01:24:34

Have you reloaded the game interface in the past hour or so? There should no be much difference at this point, but you may be observing the previous VR winds.

BTW comparing the TWS/TWD numbers from the extension(game engine actually) vs. the game interface vs. the router should give you some idea which one is off and how much.

Post by Slapdash | 2017-12-13 | 01:41:54

Hmmmm..... I wonder how I can tie a rope to the stern of Zezo's boat.... and let him drag me around the southern ocean....:-)

Thanks for the explanation.... hell... thanks for everything!!

Post by Slapdash | 2017-12-13 | 01:41:55

ooops...

Post by Flying Dutchman | 2017-12-13 | 01:50:07

reloaded both VR and yours. Still yours tells me to tack to the SE......so will give it a try:)

Post by nsp | 2017-12-13 | 04:15:18

Just a Note (to add to others in my list):

It doesn't mean anything practical now with imediate consequences, because VR isn't going to change it overnight. And if and when they do it they would do it with Zezo or a solution similar to the one Cvetan implemented.

But one thing is for sure.

The interpolation is fake weather, not real weather. What VR sells is I quote "Use the real weather and cross the finish line first".

With this sentence VR is opening the door for every player to use any weather forecast provider from everywhere in the world.

1 . When Interpolating weather charts, if it is possible to have real weather online, VR is breaking the contract they have with players because they are not giving them real weather.

This, together with bugs on interpolation routin or algorythm or cycle or whatever, leads sometimes to players making errors and taking options impossible to recover later in the race when things get to normal. And this implies responsability from the part of VR.

2. On the other hand if it's not possible to have real weather, and knowing that what VR is selling is real weather 4 times for really short periods and interpolations in between, so giving you fake weather, making it impossible to use any weather forecast provider...in this case VR is also breaking the contract with their clients..more they are saying they are selling something they know they can't give them. It's a bit worse in the grade of default due to the knowledge in that case that it's impossible to have real weather online in the game.

That together with the aberration of the 4 knts treshold. Wind is what it is. Whether is a lot or nothing we have to deal with it.

Now: Million dollar question to the tech experts, Cvetan & Co.: Is it possible to have real weather in the game without interpolation, even if it is with a little delay?

If yes: you know the answer.

If not: You also know the answer.

In any of the situations VR is in default with their clients!!!

Sometimes to assume a humble position ackowlege limitations, communicate with clients and have/create some empathy do miracles, but this guys seems to don't have read the book

Post by zezo | 2017-12-13 | 07:01:40

What VR does is as close as practically possible to "real life data", excluding the obvious bugs where your client shows you old cached files, and the 4kt limit, but even the limit does not matter much as we saw in the previous leg.

In order to display "Real" data you'd have to deploy a gloval network of BILLIONS of sensors, which is not PRACTICALLY possible, even if you had the billions of $$$ needed to do it.

Don't remember if I've sent you the link, but there was actually a lawsuit revolving around this question. It's a long and interesting read.

In brief:

1. There is a patent concerning "Competing virtual boats against real boats using real weather data"
2. LiveSkipper were holding the rights to that patent
3. VR licensed the patent
4. VR failed to pay their license fees
5. LS sued because of that
6. VR filed counter claim
7. Case was dismissed

One of the points of VR defense was precisely that it's impossible to use "Real" weather data unless it's some kind of local event where you stream live weather data from boats and network of weather sensors.

Post by nsp | 2017-12-13 | 08:02:55

Your weather directly from Noaa has more or less intepolation (meaning manipulation) than VR?

And Cvetan, what you just wrote just confirm what I said. VR knows they can't give us real weather, so they are lying to their clients as there is no real weather unless 4 times a day for short periods and with delay, So, and selling us a fake product.

It would be better to say...." As close as it gets to real weather " but this people never learn.

Post by zezo | 2017-12-13 | 08:35:11

Even NOAA is never "real weather"

The real weather is input to the model. Then the analysis of that data in order to fit it in a global picture and initialize the starting conditions for the forecast takes hours, so GFS is diverging from the real world for minimum of about 3 hours.

About the interpolation details - let's give credit where credit's due. The VR method is as good as you can get. There may be glitches and client-side display and caching issues but the methodology itself is good. It could be done in a slightly different way, but that would not make real difference as long as it's done in consistent way.

Post by Ursus Maritimus | 2017-12-13 | 08:42:09

indeed, it's weather FORECAST, but that in VR that forecast IS true weather for the next couple of hours after the weather update in VR.

Zezo tries to match as good as possible VR weather, where during changeover to new data difference occurence. Mostly to small to notice, but during strong weather changes there are differences

Post by karriv | 2017-12-13 | 09:43:57

Actually what is used in these virtual navigation games is a mathematical model of the weather we observe. The model is based on a very large (but not infinite) number of observations, and a mathematical model based on historic observations and knowledge of the dynamics of the atmospheric system, but to be clear, it is a model, not "real weather", even for the present state. If you think about it carefully, even the "present" weather, at the time of GFS release, is a forecast, because it takes time to generate the data.

Post by nsp | 2017-12-13 | 11:43:01

Cvetan, that procedure you described is what causes the delay. However that is the real weather with delay, without any interpolation, right? The Noaa's one i mean.

Wouldn't it be possible to use the Noaa source as direct source to the game weather without interpolartion, with a second delay even due to the necessary input to the new programm and model?

That would be real weather. (with delay) No fake weather!

The thing is, even now if I understand well, all the weather updated every 6 hours is 3 hours old (minimum). All the rest is fake weather.

Post by Sir_Petrus | 2017-12-13 | 11:42:24

Something strange happens.
For the same calculatios, the same inputs must give the same output.
So where is the diference between both places, VR and Zezo?
I do not understand if the NOAA files are issue for everybody at the same time, why we got diferencies.
Of course I don't want to go thru a conspiracy teory, like they do on propose to prejudice the Zezo calculations.

By the way, is it possible to give some kind of indication whem zezo is updtated?
Something like red/yellow/green leds?

Post by Inicio | 2017-12-13 | 11:56:02

You can read this:
http://zezo.org/forum.pl?tid=6109

Post by zezo | 2017-12-13 | 12:38:56

I haven't made all possible effort to duplicate the VR method exactly.

That does not make the VR method less valid. It makes me a bit lazy ;-)

Post by zezo | 2017-12-13 | 12:36:34

@nsp Everything is interpolated. You start with weather forecast data for every integer degree of the map every 3 hours.

So from the data for 46S/43E, 47S/43E, 46S/44E, 47S/44E for 15h and 18h you have to come with supposed wind data at 46� 17,8 S 43� 56,2 E ar 15:40.

That's called interpolation.

Post by Sir_Petrus | 2017-12-13 | 14:10:00

Just to confirm
VR is using UTC time?

Post by zezo | 2017-12-13 | 14:46:17

Yes, UTC behind the scene. Client presents it to you in local format, but the servers and everything use UTC internally.

Post by Flying Dutchman | 2017-12-13 | 22:53:13

Well I was a little busy so trying to catch up on you data pro's'.
Learning a lot of this data analyses.
IF I understand it correctly I better follow VR weather iso zezo at times with fast changing weather....for the short term....As we are playing in the VR.

Something I did not do last night however :))) I choose your Cvetan :)
For now it got me in a pretty good position.

Using the zezo charts for the long term. The visualisation of the polar chart forecast is realy usefull to me.

Regret is becomes so commercial with VR

Fast Sailing gentlemen ( and ladies)

Post by zezo | 2017-12-13 | 22:57:41

VR is the ultimate authority. They move your boat and place islands in the chart. Errors happen. NOAA model runs get delayed now and then. VR messes the updates once in a month. My service is not guaranteed 99.99% uptime either.

Post by Flying Dutchman | 2017-12-13 | 23:16:02

Sure I understand. Will not hold you countable:)
Just helps to look ahead:) Great Work. Will see tghis leg where it brings me.
Sure will come back to you when it makes me finish with the first 500 as a newbie