ForecastWatch.com July 2006 Customer Newsletter
Welcome to the July 2006 Newsletter of ForecastWatch.com. You are receiving this email as a subscriber to a ForecastWatch.com product or by signing up for ForecastWatch news. Please see the bottom of this email for instructions on how to unsubscribe from this newsletter.
One thing I've been recently curious about is how stable forecasts are and how that is reflected in the accuracy of a forecast. By stable I mean how much a forecast for a given day changes over time. For example, the forecast for Columbus, Ohio (http://www.forecastadvisor.com/Ohio/Columbus/43235) for this Saturday, July 22, has changed a little. This past Sunday, the forecast said it was going to be 90 degrees with a chance of rain. Today, the forecast is calling for 80 degrees and partly cloudy. However, the forecast for today, July 20, has changed very little over time. It's always called for a chance or rain or thunderstorms, and the high temperature forecast has only varied from 88 to 91 degrees. Will that forecast, on average, be more accurate than the forecast for Saturday?
Since past forecasts are hard to find unless you write them down yourself, we've added a feature to ForecastAdvisor (http://www.forecastadvisor.com) that will allow you to look at past forecasts for any upcoming date. On a forecast page, just click on the graphic of any forecast and the previous forecasts for that day will pop up. It's more information than you can find elsewhere, and maybe it will help.
I've started to do some analysis to see if more stable forecasts are more accurate on average. You can see some initial results in the blog entry titled "See Previous Forecasts" which can be found here: http://www.forecastadvisor.com/blog/2006/07/see-previous-forecasts.html
I've been having an interesting discussion with the National Weather Service about finding forecasts by zip codes on their main Weather.gov (http://www.weather.gov) web site.
Zip codes are a ubiquitous way people enter in location information. It's a short-cut, and it's used EVERYWHERE. For example, if I enter in "85065" (a Phoenix zip code, right at the airport) on Accuweather's, Intellicast's, CustomWeather's MyForecast.com site, The Weather Channel website (http://www.weather.com), Wunderground (http://www.wunderground.com), etc. I get the forecast for Phoenix as you expect. You don't if you enter that zip code on the NWS main page (http://www.weather.gov).
If you enter a zip code into any Store Locator service on any retailer's website, you get back reasonable results. Best Buy (http://www.bestbuy.com), Walmart (http://www.walmart.com), etc. Try it. You get Phoenix-area Best Buys and Phoenix-area Walmarts. If I type in "85065" for a Google (http://www.google.com) search, the first result is "Map of 85065" (http://maps.google.com/maps?oi=map&q=85065) which calls up a Google Map of downtown Phoenix, just west of the Sky Harbor Airport.
The National Weather Service, for many western zip codes, returns bizarre results. Type in 85065 at Weather.gov (http://www.weather.gov) and you get Quartzsite, Arizona, which is about 130 miles away from where everyone else says it is. Why can't the National Weather Service get it right?
You might say I'm being too hard on the NWS. After all, on the Weather.gov website, it does say "Local forecast by 'City, St'", with no mention of zip codes. If that's the case, then why do zip codes work at all? Enter a Boston zip code, you get Boston. Enter a Fort Myers, Florida zip code you get Fort Myers. In fact, on many of the regional NWS websites, like Wilmington (http://www.erh.noaa.gov/iln/) it says "Local forecast by 'City, St' or zip code". Why the inconsistency?
One argument the NWS had was that "zip codes were never intended to describe a geographic area". Of course that's correct, but that's an empty argument. The fact is, a zip code can be used to roughly describe a general location, and that it what it is used for. It's short hand, and while it's not precise, it's precise enough for all but the most extreme cases. And if the NWS really believed that, why do zip codes work for MOST of the zip codes (save some western region ones)?
You can read more at "Is There an NWS Web Issue?" at http://www.forecastadvisor.com/blog/2006/03/is-there-nws-web-issue.html . Please send us your comments (just reply to this email if you'd like) about this and if you agree or disagree.
We've audited and loaded the June 2006 accuracy data into both ForecastWatch (http://www.forecastwatch.com) and ForecastAdvisor (http://www.forecastadvisor.com). The three year trend for 1-5 day out June high temperature absolute error is 3.77 degrees in 2004, 3.60 in 2005, and 3.60 also in 2006 (however standard deviation of error increased from 4.88 in 2005 to 4.93 degrees in 2006).
Bias was 0.63 degrees in 2004 (meaning forecasts were on average 0.63 degrees too warm), 0.09 degrees in 2005, and 0.18 degrees in 2006.
Climatologically (for the stations we track), average temperatures were 1.75 degrees below 1971-2000 averages in 2004, 1.25 degrees above those averages in 2005, and 1.47 degrees above those averages in 2006. The climate data calculated is incidental, as climate is used as an unskilled forecast, and it is only for about 800 discrete generally high population areas, but it trends well with NWS figures. NWS comparisons against the 1901-2000 normals showed 0.5 degrees below long-term climate normals in June 2004, 1.2 degrees above in 2005, and 2.5 degrees above in 2006.
Please let us know what you think at ideas@forecastwatch.com . We welcome your ideas and suggestions to make our services better and more useful.
- Copyright 2006 Intellovations, LLC.
