Reliance on GRIB Files

[TheOffice]

Great thread!

I recently did my third Annapolis Bermuda race. I studied the GFS and Euro models as well as PW religously for weeks before the race. Drove the Admiral nuts!

In the end, a cold front was predicted to be near the mouth of the Chesapeake. Unfortunately, it was 50-100 miles south of the predicted location. Instead of a Gulf Stream crossing in a 15 knot SW, we have 35+ out of the Northeast. I would have never left the Bay if cruising under those conditions, but it was a race, so off we went. At the weather briefing I noticed the NOAA surface map showed the front south of the forecast models (where it actually ended up). When I asked the meteorologist from NOAA about it, he said that those charts are not meant to depict features that accurately.

[Dockhead]

Originally Posted by TheOffice

. . . . When I asked the meteorologist from NOAA about it, he said that those charts are not meant to depict features that accurately.

That's actually very interesting.

If it's true, it means that we really shouldn't be using GRIB charts without a considerable amount of interpretation based on what kind of systems are moving through.

[El Pinguino]

Originally Posted by conachair

Not me!! Gimme both, plus anything else available, 500Mb. might have a look at building a QFH antenna one of these days to get sat images as well, the more the merrier

Not actually sure what info is on the 500Mb which can't be seen on the MSLP ?

I would suggest buying an MX-137 https://www.wraase.de/hardware/antennas/ rather than building a QFH

Interpretation skills take a bit of time to develop but some days you just know to stay in bed...........

[belizesailor]

Originally Posted by Sedridor

Quite so. Lorenz published several papers from 1963-1965 demonstrating that due to the butterfly effect, errors in chaotic systems grow at highly variable rates. It's commonly generalized that these errors roughly double every three days, and that forecasts are theoretically unmanageable (even with excellent initial states) after two weeks as an upper bound. So we will almost certainly never see a reliable forecast system that goes out more than two weeks.

The more interesting observation from those papers, to me, is that the error growth is highly variable. So the trick with modern forecasting, IMO, is to develop the ability to determine when (and which) models are retaining skill and when they are not. Many in this thread have already touched on this.

As some others have mentioned, ensemble models are excellent tools for this. Hedging with multiple product overlays is essentially construction of a DIY ensemble model, albeit with fewer components.

For those who are unfamiliar with ensembles, the best illustration of their utility I know of is something called a spaghetti plot. You can see an example here:

https://cms.met.psu.edu/sref/ensembles/

Pick 500H and your favorite region (NOAM gives a good overview), and note how the isobars diverge forward in time. It is quite easy to see when (and which) models are losing agreement. I think this kind of analysis has potential for marine forecasts, but seeing as I don't have extensive cruising experience (yet!) I can't really speak to its merits.

If anyone has attempted to use ensembles as a statistical hedge for weather-based course adjustments, I would be very interested to hear about it.

Ensemble forecasts are something I have on my list to gain more experience with. One of my favorite weather sites (WeatherOnline) has extensive resources including access to ensembles from several models:

https://www.weatheronline.co.uk/cgi-...O=78762&ZOOM=0

The ECMWF site as a good laymans description of ensemble forecasts:

https://www.ecmwf.int/en/about/media...er-forecasting

[Snowpetrel]

I'd love it if gribs, and forecasts in general came with reliabilty or confidence data. It would be pretty easy to run a set of ensembles and extract some measure of the stability of the model and add it in as an overlay. (I see from the above links they call this the sharpness of the ensemble)

I tend to consider weather as being in a stable or a dynamic situation. Ie a big blocking high or a well established low is generally pretty stable, and predictable. A newly forming low tends to be much more dynamic and unpredictable and I'd be much more wary around them. Sometimes you get a big high breakimg down and a cluster of little lows, or the whole wx map will have hardly any defined features or tight isobars. This always makes me wary of something brewing in an unpredictable way. In these cases Gribs can only show one of many possible outcomes. I guess thats the power of an ensemble, to show you many possible outcomes/forecasts.

[estarzinger]

Originally Posted by Snowpetrel

I'd love it if gribs, and forecasts in general came with reliabilty or confidence data. It would be pretty easy to run a set of ensembles and extract some measure of the stability of the model and add it in as an overlay. (I see from the above links they call this the sharpness of the ensemble)

noaa has experimental products which show confidence (I'm not wild about how they do this, it is not how I personally evaluate 'confidence', but they are the pros ) and others that show all the features on wefax (types of fronts and lows and such) embedded in the 'normal' grib - I believe you can see these in their web portal (look for an experimental tab) but the difficulty on-board is that 'normal' grib viewers don't know how to read/display this extra data.

They do understand the value of sharing both of those types of information with users, just have not figured out practically how to do it for the low bandwidth cruisers.

[jackdale]

For an over-the-top reliance on gribs, try using racing software. Put in your polars and sail selection, then add your grib data. Your sail selection, course to steer and optimum speed in 3 hour increments is all available to you.

Anytime I have used such software, I have learned to steer the wind I am in.

[Snowpetrel]

Originally Posted by estarzinger

noaa has experimental products which show confidence (I'm not wild about how they do this, it is not how I personally evaluate 'confidence', but they are the pros ) and others that show all the features on wefax (types of fronts and lows and such) embedded in the 'normal' grib - I believe you can see these in their web portal (look for an experimental tab) but the difficulty on-board is that 'normal' grib viewers don't know how to read/display this extra data.

They do understand the value of sharing both of those types of information with users, just have not figured out practically how to do it for the low bandwidth cruisers.

Thanks Evans, I'll have a play with them. Could be very useful when preplanning ashore at least, or to pass on via a satphone call to a friend.

I guess reliability is one issue, then there is the probability of a dangerous outcome due to a changing forecast. I suppose this this is just standard risk managment theory as used in businesses, OH&S, and real life.

Ie take the 98 Sydney Hobart, the forcast was low confidence at the start, but the contestants weren't really aware of this, so many did not critically re-evaluate the weather on passage. The situation with the explosive low formation in Bass Strait was always on the cards, just less likely than more moderate development. So as a sailer does the low confidence/lack of sharpness (fuzzyness/blurryness?) Imply a better or a worse outcome for us? In the sydney hobart it was a dramatically worse outcome.

For me I am most interested in tipping points, ie some situations have two probable and likely outcomes depending on how stuff developes. Eg maybe a low has a 30% probability of forming and deepening rapidly, Vs a 70% probability that it will form but quickly dissipate? The gribs should always show the 70% probability. But it would be great to be able to see the Ensemble results and the chance of the 30% probability outcome and defend against it as well. I guess thats where a very good weather router comes in.

Another issue these days is working with modified base model data. Many third parties modify the base model data to improve coastal forecasts. Eg Meteye in Australia uses ensemble and blended models typically from GFS and/or ECMWF, then breaks the larger grids down to 6km squares and applies land effects and local heating info to get seabreezes and then outputs the more detailed final data. I think willyweather does something similar, as does predictwind.

So the coastal data can differ significantly between these improved (or tweaked) models and the base Model such as ECMWF or GFS. So if I am using the base models I have to mentally allow for coastal effects, but I have to be careful not to over apply them to the already improved models. It can be hard to know exactly what tweaks have been made or what models are blended in these cases.

[cyan]

Originally Posted by estarzinger

noaa has experimental products which show confidence

This is really good information, thanks!
I have often wondered why I cannot even find any ATTEMPTS at confidence. Perhaps I'm not looking hard enough?

Graphical GRIB software might just be too pretty. When magically stepping into the future by 3 hour increments, I think many newbie sailors get lured into the "time machine fantasy". Move the slider and simply watch EXACTLY what Poseidon has in store for you. The GRIB data for the next hour looks super-detailed. Typically accurate. Yay. The GRIB data for 5 days from now looks super-detailed. Typically a crap-shoot. Boo.

The problem can be one of perception. The wonderful colors and arrows for 5 days out look just as wonderful as today's detailed display. In engineering, this is similar to the difference between precision and accuracy. The two should not be confused. The computer model displays show the same "precision" for the 5-day picture as they do for the current picture. I feel this can be deceiving, even when you understand the significant difference. (or when you are drinking beer while on shore doing your forecast analysis, waiting for that "window")

When a hurricane track is predicted, the math models are often shown with color-coded probability bands that fan out farther from the storm. Even the general public can understand that. Why is it, then, the typical GRIB displays don't include more of this type of "confidence" indication?

In a perfect world, I would be able to see the accuracy comparison of all available models for the past few days, or this time last year, etc. I would have it color-coded, let's say, to show things like color-coded wind barbs that range in color depending upon how accurate the model was for that position recently. I might have a phantom barbs (or a clever dual-symbol) at barb locations that stick out in red when there was an error, with older errors appearing faded like a persistent waveform display on an oscilloscope. You might see a region of past model errors that surround a cold front- alerting you to distrust predictions around cold fronts? Dunno.
I CERTAINLY would wish for a graphical change to indicate that I was looking at a future (less confident) picture- maybe barbs and isobars literally appear fuzzier as you march into the future?

In this thread, someone suggested manually tracking the past models with current real data to get a feel for reliability. Good heavens, where is the software to do all of that for you? Clearly I'm missing something on OpenCPN and the paid stuff too. If I simply have missed a product that does all of this, please inform me. Thanks in advance!

[Dockhead]

Originally Posted by jackdale

For an over-the-top reliance on gribs, try using racing software. Put in your polars and sail selection, then add your grib data. Your sail selection, course to steer and optimum speed in 3 hour increments is all available to you.

Anytime I have used such software, I have learned to steer the wind I am in.

Sure, but you should understand why it does this --

You need to have a plan which starts with the BEST GUESS about conditions along the way. That's what this software gives you. It's really valuable, but it's stupid to just blindly follow it. You have to adjust it for real vs predicted conditions, and you have to apply some interpretation to the risks in the forecast. And as Evans said, you have to apply knowledge of how to drive a boat through ocean weather.

To give a highly simplified example of this process --

When we cross the Channel, we have to use out predicted passage speed to get a CTS to deal with the tidal streams, and then we have to overlay any sailing tactics on top of this. The CTS is highly dependent on passage speed, and has to be adjusted along the way. You may also through off the plan with a change in sailing tactics, somewhat different wind, etc. But you need the plan because you need to know how to steer. For that you need a BEST GUESS to start with.

[james247]

Originally Posted by Snore

I believe, like navigation, one should not rely on only one source of data.



We have all been caught in weather that was not predicted- even by pros. But if you are using 2 sourc s of info, such as GRIBS and inReach weather, or a weather routing service and inReach— the odds of an “oh-my” moment should be lower. There is just too much at stake to trust in the infallibility of the weatherman

There old skills that are often forgotten and reliance on electronic data becomes all encompassing.
We used to follow our barometer with religious furore. They still work very well
A look out the window is also a good practice.
Seasonal wInd pie charts
Humidity, temperature and cloud formation give us an understanding of pending weather.
Grib charts are the bonus not the basics

[belizesailor]

Originally Posted by james247

There old skills that are often forgotten and reliance on electronic data becomes all encompassing.
We used to follow our barometer with religious furore. They still work very well
A look out the window is also a good practice.
Seasonal wInd pie charts
Humidity, temperature and cloud formation give us an understanding of pending weather.
Grib charts are the bonus not the basics

Having the knowledge to interpret local observations is a good skill, but of course it only gives you an idea of whats coming very soon.

Local observations can be even more effective if you have good local knowledge too, or have someone aboard with it. The old fishermem in Belize navigate by visual reference only, often thru very complex areas...most have never used a chart or compass...and would not have a clue what to do with a GPS. Same for weather, they know from local observations whats coming in 24 hours or less...which is all they need since they are not on long passages.

For example, if the winds go SSE and strengthen in Belize, then there is almost certainly tropical weather brewing, or at the very least a "bayama" (strong SE winds). No need to even look at a forecast...go find a hole to hide in. This local observation often gives you warning before the forecasts do...the forecasts then tell you just how bad its likely to be.

[Bleemus]

It is extremely important to cross check weather informations, specially when they come from a GRIB file. GRIB files are computer-generated, as opposed to the faxes, into which weather engineers put a lot of skills.*
Whenever there is a contradiction between a GRIB file and a fax, the fax should always be considered as the point of truth.

[jstevens]

"Whenever there is a contradiction between a GRIB file and a fax, the fax should always be considered as the point of truth."

Except maybe when the GRIB is based on later data than the Fax.

[Bleemus]

Originally Posted by jstevens

"Whenever there is a contradiction between a GRIB file and a fax, the fax should always be considered as the point of truth."

Except maybe when the GRIB is based on later data than the Fax.

Thank you for stating the obvious.