Received: from zonker.hpl.hp.com (zonker-fddi.hpl.hp.com) by opus.hpl.hp.com with ESMTP (1.37.109.18/15.5+ECS 3.3+HPL1.1) id AA238137438; Thu, 3 Apr 1997 20:17:18 -0800 Return-Path: <poulton@zonker.hpl.hp.com-DeleteThis> Received: (from poulton@localhost) by zonker.hpl.hp.com (8.7.1/8.7.1) id UAA11606 for wind_talk@opus.hpl.hp.com-DeleteThis; Thu, 3 Apr 1997 20:17:17 -0800 (PST) Date: Thu, 3 Apr 1997 20:17:17 -0800 (PST) From: Ken Poulton <poulton@zonker.hpl.hp.com-DeleteThis> Message-Id: <199704040417.UAA11606@zonker.hpl.hp.com-DeleteThis> To: wind_talk@opus.hpl.hp.com-DeleteThis Subject: Re: Accuracy of the NWS morning forecasts? Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit
> Does anybody have any idea of the accuracy of the NWS forecasts? They
> were dead wrong today.
I did a study of the SFO afternoon forecasts several years ago and got
about a 5 knot standard deviation of prediction minus measurement.
In lay terms: they were wrong by more than a sail size (+-3 knots) half
the time! Today they were wrong by 5 knots (from 3 to 5 PM).
Sounds pretty bad, huh? The NWS folks say that forecasting wind in this
area is hard because of the terrain and the strong interaction with the
marine layer ("fog"). I suspect that a 5 knot std dev is just fine
for most purposes and its just us windsurfers who are hyper-sensitive to
the difference between 21 knots (big fun on my 5.4!) and 14 knots (forget it).
I tried to see what I could do for a prediction. I found the parameters
with the best corelation to windspeeds, but I was never able to put
these together to make a prediction that was substantially better. I
think what may be needed is a neural net program that can learn the
various patterns involved. Figuring out where the fog lies is part of
the data it should have, which is another problem in itself.
Anyway, I have years of data if anyone wants to make a run at
making an automated predictor.
> What is the most accurate source for forecasts.
No idea. I use the SFO forecast, look at the marine forecast, sometimes
read the forecast models discussion, look at the key current readings,
and still get skunked sometimes.
> I'm trying to
> determine when to bring my gear with me to work.
That's easy. Every day!
Ken Poulton
poulton@opus.hpl.hp.com-DeleteThis
Meteorology, n.: A psychology of the science of finding a pattern in the
randomness of weather behavior in order to forecast future random
weather behavior events. -- Jonathan Leffler
This archive was generated by hypermail 2b29 : Mon Dec 10 2001 - 02:31:31 PST