Windy.com. Wow!!

[belizesailor]

Originally Posted by jmh2002

I posted this earlier in the thread as one way to compare forecast model accuracy over time for a given location:

Yes, there are techniques for doing this comparison, but I dont know of anyone who has invested the time to do an in depth comparison of the forecast tools to actuals. For example, is PW's PWE (which uses ECMWF data as initial state) significantly more accurate than ECMWF alone.

[Paul L]

Originally Posted by belizesailor

Yes, there are techniques for doing this comparison, but I dont know of anyone who has invested the time to do an in depth comparison of the forecast tools to actuals. For example, is PW's PWE (which uses ECMWF data as initial state) significantly more accurate than ECMWF alone.

Certainly the major model providers do this long term verification, GFS and EC models. The updated GFS model wasn't released till it was run along with the old model for an extensive period and shown to be better.

Unlikely a small outfit like PW puts the money and research into verifying their model changes over the world range they apply them to.

Verifying the models by taking a current land station observation and comparing to say a 3 day old forecast is a little bit interesting, but doesn't tell you much about the model quality where a cruiser most needs it, offshore. Using the ASCAT satellite runs and comparing to days old model predictions might be more useful.

[belizesailor]

Originally Posted by Paul L

Certainly the major model providers do this long term verification, GFS and EC models. The updated GFS model wasn't released till it was run along with the old model for an extensive period and shown to be better.



Unlikely a small outfit like PW puts the money and research into verifying their model changes over the world range they apply them to.



Verifying the models by taking a current land station observation and comparing to say a 3 day old forecast is a little bit interesting, but doesn't tell you much about the model quality where a cruiser most needs it, offshore. Using the ASCAT satellite runs and comparing to days old model predictions might be more useful.

Yes, a comprehensive comparison would be beyond the resources most sailors would be willing/able to invest...hmmm...maybe some form of "crowd" verification could work?

But, spot comparisons, actual usage experience, and less exhaustive comparisons I think do have some value. For example comparing NESDIS vs Copernicus currents, with a dose of experience, along the Panamanian coast is a pretty obvious evaluation.

[conachair]

Originally Posted by belizesailor

Yes, a comprehensive comparison would be beyond the resources most sailors would be willing/able to invest...hmmm...maybe some form of "crowd" verification could work?

But, spot comparisons, actual usage experience, and less exhaustive comparisons I think do have some value. For example comparing NESDIS vs Copernicus currents, with a dose of experience, along the Panamanian coast is a pretty obvious evaluation.

Not too difficult in Python to do something...

This is live data against forecast for a UK Metoffice measuring station on the Isles Of Scilly , chosen as it's away from land and quite flat, plus anenometer is 10m. Forecast data is also met Office. The mouse wheel can zoom X or Y axis if you scool the wheel with the mouse just below x-axis or to the left of the y-axis. Legend top left turns different forecast on/off.
Remarkable just how accurate it is most of the time. Some flat lines in there as I never got round to automating it and forgot some days to run the script.



Scilly Wind – My First website

One day might get round to doing similar with GFS & openweather
https://openweathermap.org/

Almost no one has any idea about forecast accuracy, our ape brains just love to see patterns which aren't there Plotting lots of web available data is probably as good as you'll get for the vast majority of us.