2017 Weather to the Azores

[Palarran]

Originally Posted by barnakiel

I would propose that (?) there is no way any software can give you anything like even marginally reliable routing beyond X days' horizon(?). Not even in a boat as big and fast as a Lagoon 500, I believe. We can attempt to establish what the above X is.

However, the exercise you want is easy to do with any wx routing package, you can do the following, e.g.:

DAY ONE: set the points and pick up the polar, DERATE according to how the boat is sailed/loaded/equipped, DERATE according to the sea-state, download the grib, run the model, check if you agree, if you do: MARK DAY ONE END POINT.

DAY TWO: use the marked above waypoint as your boat position, repeat the whole exercise (download fresh gribs, etc).

Do the above, say daily, and you get to see a what it would be like in real life. It takes about two or three weeks of daily (or any other period) attention.

The above routine will work as if you have sailed an actual passage. But it must be done on a regular daily or other periodical basis - for you can download the gribs daily and run the exercise at a future point - only if you are very keen on manipulating the routing software. And not all software allows you to manipulate this much either.

On reading the weatherfax chartlets: It is indeed a very very easy thing. The air around the Low and High cells moves in its distinct manner - direction and speed related to center, etc. The other visible features are fronts and on a front there is a specific set of wx features too - typical cloud formations, wind shifts, sea state conditions. So this is like a general plan telling you the train is coming and what kind of train it is. The grib data interpreted by any software will tell you then how fast the train is moving and also at what angle it will hit you.

It takes not more than a week of intensive study to be quite literate with interpreting a weather fax image. And it is best done while you are there - so that you can see the features shown in the wxfax as they pass around you - the change in air clarity, the change in cloud formations, the wind shifts, the lightening and the wind turn on the ridge and the acceleration and the wind shift on the front. Etc.

Wxfax images are a great tool to predict the best days to watch out for a green flash when you are in the Caribbean - for you want to know when the pre-front clearing comes. A grib file will not tell you this. But looking at the horizon every morning will. The beauty of using various sources and seeing what they actually attempt, and can/cannot, describe.

Can you possibly let me know what passage time your software generated for your boat? I am surprised you got full passage covered as my longest gribs show 10 days data and beyond 4 days I no longer use this data for any planning. But even with the max 10 days' grib data, vastly unreliable as it is, I do not think my packages (MaxSea and qtVlm) could predict a Lagoon 500 getting from the West Indies to the Azores.

Choice of departure time and timing is dependent on how strong the boat and crew are: North Atlantic speed records are often attempted at times when no cruising boat should leave the harbour. The rest is choices of the skippers who are ultimately responsible for safety of their crews.

Cheers,
b.

I totally agree with you on the accuracy window. Predict Wind has a 14 day model. Sure, it's not accurate which is why you would need to run it every day.

Are you on Tenerife? If so, maybe I can hire you to do a 4 hour synoptic weather class (I don't have a week to spare) for me before we leave next December. I have a slip reserved in Marina San Miguel.

[estarzinger]

Originally Posted by barnakiel

Can you possibly let me know what passage time your software generated for your boat?

That is included in the table in his post 11 - around 12 days +- 1 day depending on which of the 4 models he used.

I am surprised you got full passage covered as my longest gribs show 10 days data

You can get longer than 10 day gribs. NOAA themselves generate 16 day operational forecast.

and beyond 4 days I no longer use this data for any planning.

You are probably short changing yourself. Very large features (like la Ninia and the jet stream) are (typically) stable (at least somewhat) beyond a 4 day horizon and thus your routing can be better than just using historical climatic average picture. You can 'edge' in the right direction and fine tune it along the way.

But even with the max 10 days' grib data, vastly unreliable as it is, I do not think my packages (MaxSea and qtVlm) could predict a Lagoon 500 getting from the West Indies to the Azores.

as mentioned, his route is 12 days . . . . 8.5kt average . . . . I think that is possible but would guess it is a factory or racing polar and his polar needs to be tuned down a little to reflect cruising speeds.

I would propose that (?) there is no way any software can give you anything like even marginally reliable routing beyond X days' horizon(?).

Well, lets divide this up into 4 time buckets. Bucket #1 is somewhat accurate local details . . . . generally 4 days, Bucket #2 is somewhat accurate major system details . . . . generally 10 days (note: this differs by ocean - in the southern ocean systems persist longer than in smaller more complex areas - but even in the less stable areas you can often pick up hints of major system issues 10 days out but I could agree with 6,7,or 8 days in this bucket in complex oceans if someone really wanted to debate it), Bucket #3 is somewhat accurate macro features . . . . generally 16 days, and Bucket #4 is climatic synoptic average . . . . essentially forever (or until the earth's basic climate changes). You can use all 4 buckets for some amount of routing information/skill (eg better than just guessing or just going straight line).

Choice of departure time and timing is dependent on how strong the boat and crew are.

Hmmm disagree (to a point) . . . . departure time and timing should look at the weather forecast, not just the boat and crew. Leaving the Caribbean, you 'should' be able to pick a weather window which gives you a decent run north, to the turn (to the west) point . . . or about 6 days give or take. So by looking at the weather you can pick to make the first half (roughly) rather easier. On departure no tool (software or human) will be able to tell you exactly what systems you will face when you get there, but you can definitely update and adjust as you progress.

Note: I am not intending to be argumentative here. But I believe you are undervaluing both gribs and longer range routing (grib or human).

I shore routed two vendee boats, and used these tools on all our own passages (on our 2nd boat they did not exist during our first voyage - when we used wefax and SSB voice forecasts).

............

[estarzinger]

I might add . . my personal feeling is that, in the day of gribs, the most important meteo skill the cruising skipper can develop is to be able to assess a particular grib's confidence . . . eg how likely is this to actually play out this way. With some patterns you can assess the likelihood is quite high and you can then do very clever routing. With other patterns you can assess that there is a lot of uncertainty in how the particular/specif set of patterns/systems will play out and then you want to 'play it safe' and route to only the really major features and not to the details.

The NOAA guys I have talked to agreed with this thought. They themselves think about the confidence level of forecast this way, and they have automated tools to help them assess confidence, but (to my knowledge) those tools are not available to the public (and they also chew up quite a bit of computer power)

edit . . . . another thing I might note about all this . . . is that you tend to have much more opportunity to do 'clever' routing on North/south oriented routes (like our run from Iceland to cape horn) than on east/west oriented routes.

[donradcliffe]

Interesting stuff. I have a friend with a boat in Lanzarote who wants to come to the US in in the next month. I did it one April and had to go down as far as 14N to get wind. Stopped in St John and Ft Lauderdale, and it was a long boring trip, but nothing got broken. Would you suggest he take the southern route, or wait til Mid May and take a more direct route?

[barnakiel]

For Palarran:

Some digital experiments done here:

- based on historical (Pilot Charts, vpp) data: 12 days,
- based on rolling on of 10 days' data (grib, MaxSea) : 14 days,
- based on 4 days' data (grib, qtVlm) : 4 days at 25% of the direct distance. ("=16 days")

All of the above a generic L50 cat, light, no sea state derated. Engine on if boat speed under sails is less than 6kts, under engine speed: 6 kts.

The tracks drafted by the MaxSea are very much like the ones you posted (this is because my version of MaxSea rolls-on existing grib data).

The tracks assumed by qtVlm in the 4 days' experiment (no time rolling of gribs) way more direct - way more E`ly early on in the passage.

The tracks drafted by vpp (optimisation for shortest passage time) - nearly direct.

Cheers,
b.

[barnakiel]

Originally Posted by Palarran

(...)

Are you on Tenerife? If so, maybe I can hire you to do a 4 hour synoptic weather class (I don't have a week to spare) for me before we leave next December. I have a slip reserved in Marina San Miguel.

I am in Las Palmas. I do not deal with synoptics. I only use forecasts and data prepared by others (mostly NOAA, in the N Atlantic basin).

If you are sailing via Las Palmas simply get hold of me and let us swap and compare the skills, tools and wx sources we use.

In Decemeber we might be in-between so then next opportunity is once you land in the West Indies.

BTW re the 14-day (or any other long term) model: try to avoid the temptation to use the 14 days' data IF you at all consider the data is of doubtful usability. Simply download a short data range that you trust (this is why I mentioned the X in my longer post above) and see where this shorter data puts you. I say this as any data routing software will 'make assumptions about assumptions' - so a 4 days' data may show you heading West while a 10 days' data on the same routing model will show you to go East ... And why should you go East then if you assume the 10 days data to be of poor usability in the first place?

This is at times visible when you use a software that draws the best route in real time of the calculation (e.g. qtVlm does). You can see how the model "changes its mind" which may (and does) lead to favouring the further (and less relaible) data over the shorter (and more relaible) data. But there is no way round this fact, if one wants a "10-days' routing".

A-ha. ;-)

Cheers,
b.

[barnakiel]

Originally Posted by estarzinger

b: and beyond 4 days I no longer use this data for any planning.

e: You are probably short changing yourself

Have a look at my post at #3. It gives some insight into why I see some limitations and why I generally lean towards seeing long term models to be of disputable value for wx routing.

BTW You got lucky with Vendee as nowadays they are not allowed to be shore routed. (I think)

I agree with your comments. There are very many ways to skin a cat and wx forecasting has much to do with predicting stock prices (ask my partner what she thinks about both my jobs and hobbies - just do not let me know what she says, for I already know) ;-)

Hugs,
barnakiel

[Palarran]

Some verry good posts. I'd like to thank you guys for spending the time. I doubt my average speed over such a long time would be 8.5 knots. The boat could do it, but I'm not that good of a sailor and kind of a sissy at nights and when boat speed goes over 10 knots.

One thing I"m willing to do is print the last 6 days of this forecast, then I'll wait 3 days, move my start point, and compare day by day for 6 days out and see what the changes are. That's pretty easy to do. If anyone has another idea to check how the models change let me know.

In regards to the data, since we, and most other sailors, will be using Iridium, I won't be able to pull the data at sea the same as on shore with broadband. So my plan is to choose one model only and shorten up the time frame so the download doesn't go over 1/2 hour. We are not buying the Iridium Go until a month before our crossing so at that point I can get a better idea of data sizes and transmission speeds.

[estarzinger]

Originally Posted by barnakiel

Have a look at my post at #3.


I am quite familiar with the graph in post 3. It (essentially) says that commercial forecast (singular) are better than climatology averages out to 9 days. . . .or about twice the horizon of your 4 days. And there is still 'positive skill' (the technical term for value beyond some baseline) beyond 9 days (described in my buckets 3 and 4).

No point in debating this . . . I will simply say that I have found significant value in forecasts and routing beyond 4 days. They have allowed us to do some clever routing, and take advantage of features we would not have been able to, and it was well beyond 'luck'.

wx forecasting has much to do with predicting stock prices

I am both a math phd (quite familiar with chaos theory), and a university of Chicago finance MBA (quite familiar with the random walk). Both wx and stock prices are interesting areas for 'forecasting', but undeniably they both do benefit from skilled 'forecasting'. The pure fact that Quants rule wall street now make that point very precisely.

It is a separate point that the unskilled will be better of in an index fund, than paying high fees to marginally skilled 'managers'. That just means there are A LOT of charlatans, and even the non-charlatans are able to extract fees in excess of the value they deliver.

edit: my vendee work was 'off the boat', paid to explain the 'play by play' (eg why did boat 1 go north today but boat 2 go east and who is actually ahead) to the audience. That has always been allowed.

.............

[barnakiel]

Originally Posted by estarzinger

b: I am surprised you got full passage covered as my longest gribs show 10 days data

e: You can get longer than 10 day gribs. NOAA themselves generate 16 day operational forecast.

Do they also generate publicly available 16 day gribs? Where can I get them?

THX
b.

[estarzinger]

Originally Posted by barnakiel

Do they also generate publicly available 16 day gribs? Where can I get them?

THX
b.

You can use Jim's grib viewer program ViewFax to get 16 days. (I might note it was set that way because Stan wanted 16 days for his own routing work ) I am guessing you own airmail - the viewfax program is included in that package so you probably already have it.

I am honestly a bit out of date atm on NOAA access. I was using a custom script/data base query that either Stan Honey or Jim Corenman wrote for me to pull them. However, it no longer works after they restructured their archives. I think (not 100% sure) that you can actually get 30 day GFS's using a custom query.

Predict wind is pulling 12 day GFS's (and 3 other models).

Edit: I might comment that I personally found the euro model to generally be better at longer term system tracks than GFS. However it is harder to access. I had paid commercial access back when I was doing weather, but that has expired. Not sure what access possibilities exist today.

Edit 2: NOAA does a lot of work on forecast accuracy. They are quite open and upfront about it all. One thing that is interesting (but not all that surprising) is that there have been quite signifivant steady improvements in accuracy . . . for example http://www.nhc.noaa.gov/verification/verify5.shtml

[barnakiel]

Originally Posted by estarzinger

e: No point in debating this . . . I will simply say that I have found significant value in forecasts and routing beyond 4 days.

(...)

e: The pure fact that Quants rule wall street now make that point very precisely.


(...)

Yes.

No doubt having some long term predictions is better than having nothing. Acting upon them is however far less binary. Or, at least (my opinion) it is far less emotionally / mentally binary. We have the longer data (agreeably a good thing) and then some will decide to employ it while others will remain skeptic (the latter being a personal thing, alike risk seeking/avoidance profile of investors).

We all draw from our experience, and mine was drawing upon the longer term data too heavily in the past which led me to some wrong opinions. Having found that, I looked for how I got to those wrong opinions and how I could potentially try to avoid repeating the same behaviours again. And hence my statement on making decisions about where the reliable/unreliable cutoff point is. That's why I say let's try to find what the X is (it may/will be a different figure for different skippers / personalities that's sure).

And on the Quants (my layman view, I have never studied them): they will easily beat an ignorant and they will possibly at least match an educated but inefficient investor. Will they ever match Buffet? Hmmmm... they may potentially outdo Warren - except he has been around for 86 years now. So we may have to wait another 50 years or so to see how things compare.

Huh. Not all that much related to wx ;- but an interesting drift anyways.

Cheers,
b.

[barnakiel]

Originally Posted by estarzinger

(...)

Well, lets divide this up into 4 time buckets. Bucket #1 is somewhat accurate local details . . . . generally 4 days, Bucket #2 is somewhat accurate major system details . . . . generally 10 days (note: this differs by ocean - in the southern ocean systems persist longer than in smaller more complex areas - but even in the less stable areas you can often pick up hints of major system issues 10 days out but I could agree with 6,7,or 8 days in this bucket in complex oceans if someone really wanted to debate it), Bucket #3 is somewhat accurate macro features . . . . generally 16 days, and Bucket #4 is climatic synoptic average . . . . essentially forever (or until the earth's basic climate changes). You can use all 4 buckets for some amount of routing information/skill (eg better than just guessing or just going straight line).

(...)

But this is not what PredictWind does. Or is it?

I tried to test ride PredictWind (about half a year ago) and back then their free to test version did not do weather routing. They also insisted on using my email which I found intrusive back then. Now that I am Bill.Gates@gmail.com I might give them another try.

So, to achieve a (far cry from, but a strong try otherwise) what you suggest I am actually using 3 different pieces of software. And nothing wrong with that (I think) for when one source is not available, some amount of fail proofing is acquired in this otherwise more time consuming process. (Still, if the Internet is down, all the tools are down, except for the long term climatic averages that are of little use to a skipper chased by a storm).

My (default but very flexible) stand is that a forecast can be correct only in one way (by being spot on or very close to it) but it can be wrong in countless ways. Do not take it literally though: I am just trying to point to the fact that without knowing how the specific model is built (and without having deep understanding, let alone knowledge) of the tools used to build the model (mostly plain maths, but at times complex stat analysis tools) it is not possible to say what level of probability to expect (vs. what level of reliability is found empirically, like in that material quoted at #3).

To put it in different words, it is not even a typical (two event) Bayesian inference - for we do not have just one possible false positive but rather a set of possible false positives. This may get out of hand easily, unless the brains that build these models match Bayes' brain (few do, imho).

And two out of four quoted models are not even publicly available for our (that is scientific, not mine) examination. Someone is trying to tell us something about we do not know what. I am ready to take this back when someone post a link to the models and their explanation by their makers.

So it all boils down to my choosing to not make bets when the odds of winning (the forecast being correct) are slim (=long term, how long? - make your own X). Possibly because the risks associated with a loss (the nature doing her thing, the boat sinking) far outweigh any (at time pecuniary) benefits of being right 9/10. The missing 1 does make a huge difference to how I handle (the same, after all) data.

Oufff. I must sound like a Gollum/Ironman (more of the Gollum thing though) hybrid ;-) by now.

Have fun, sail safe, throw in any pieces of info that might be valuable to people making this passage this year and in the future.

Cheers,
b.

[Palarran]

Well, I don't know about all that. I did think about what Starzinger said about the weather model and thought I should clarify that the weather picture being represented in my snips were based off the PWG model. The routes (yellow, green, red, and blue) are individually based on their specific model, but the background wind shown is from a specific selection of forecast model.

[estarzinger]

Originally Posted by barnakiel

a forecast can be correct only in one way (by being spot on or very close to it) but it can be wrong in countless ways.

This is perhaps a key point where we differ.

My perspective is that ALL have some information content - and that none (of the sort we are talking about here) are exactly 100% accurate.

The user's challenge is to correctly extract the information content.

This is why I said above I believe the key skill to develop for a cruising navigator is the ability to judge the confidence level in a specific forecast.

In short, I do that by looking at three things (1) is the pattern consistent over time (eg did yesterday's multi day run look similar to today's), (2) are different models showing similar pattern (eg do GFS and Euro look similar), and (3) does the particular specific pattern lend itself to confidence (eg isolated systems of distinct pressure tend to have much higher confidence pattern than either large areas of slack pressure or complicated system interactions).

If you conclude that the pattern has high confidence you can route close to systems and take advantage of the details. However if you conclude it has low confidence then you stick closer to the rumb line, pretty much ignore the micro details and just avoid the big bad stuff.

When looking at the tail end of a long range forecast you are obviously going to place less confidence in the micro details. You are looking simply to see if you should edge to the left or to the right to be better positioned to take advantage of things that might develop. Again, my experience is that there is often sufficient information content and confidence to make such a choice (and the vendee and volvo Nav's agree).

Also, as I said, it is easier to play cleverly with systems when you are sailing a north/south axis than an east/west axis.

In any case, I have had my say on that . . . .