Today’s weather gem from ./. A bad trip to a waterpark (it was wet :lol: ) prompted someone to analyze the accuracy of various online weather forecasts and write this article.
ForecastAdvisor.com does a similar thing for lots of sites. I have a link to the San Luis Obispo page in the Forecasts menu on SLOweather.
http://www.forecastadvisor.com/California/SanLuisObispo/93405
Interestingly, the NWS is almost always among the worst here.
Thats really strange, I have found that my NWS forecast office in Sterling, Va. does an outstanding job at predicting the weather compared to my local TV stations, WeatherChannel, etc… I’m not so sure of those percentages… 
Weather channel misses the boat completely most of the time for our area. NWS does a very good job here.
There’s a saying in Louisiana…“If you don’t like the weather now just wait 30 minutes for something else”. I don’t think anybody does an accurate forecast for here. Mostly you just walk outside and look at the Weather Stone. You know, “If it’s wet there’s rain, if it’s gone there was a tornado.” Much more accurate than any forecaster :lol:
Well, we have to keep in mind that such an analysis is very specific to that location and that location only.
Also, he screwed the stats when he selected The Weather Channel as his “live” temperature source. Temperature, as we all know is very fluid, even for just a few miles, it can vary at mid-day by 1-3 degrees. That right there is going to completely invalidate the stats.
Then, of course, we have to consider that most sources are actually DIRECTLY using NWS data… with no modifications other than a time delay. NWS METARs report in whole celsius numbers. This adds another +/- 1.8F variance in the temperature.
In my opinion the guy just doesn’t know that much about how weather works and how it is measured.
(P.S. I read /. too!)
I was interested to run across this topic, having done a great deal of research in developing and testing WXSIM (www.wxsim.com). Over the years, I’ve collected roughly 400 forecasts, many out to 72 hours, and for various sources, including the NWS coded cities forecasts, various NWS MOS products, WXSIM, and even myself! This takes great care, as the definitions of high and low temperatures involve specific time windows. For MOS (and I think most other competing sources) this means overnight lows from 7 PM to 8 AM local standard time and highs from 7 AM to 7 PM local standard time, here in Atlanta anyway. The verification numbers are specifically from KATL. These are reported to the nearest degree Fahrenheit, even in METAR reports, as there is an extended section for translating Celsius temperature including tenths back into original whole number Fahrenheit degrees.
Generally, I’ve found lows and highs to have about the same accuracy, with summer slightly more accurate than winter. Most sources seem to have improved slightly (with some data going back 10 years) over time. In the latest study I did (225 forecasts in 2005-2006), The Weather Channel was the overall winner, followed immediately and closely by WXSIM in the short range, and by NWS beyond at 36 hours and beyond. The NWS humans fairly consistently beat their own MOS products, but by quite a small margin (there are real “diminishing returns” with attempts to improve on already fairly good data). In the studies in which I included myslef, I generally beat all other sources, as I could look at all of them, including multiple WXSIM scenarios/runs, and come to an informed consesnus of sorts.
In short, at least some forecast sources are really quite good most of the time. Many times, I’ve heard people gripe about blown forecasts. Sometimes this is in fact true, but more often, I’ve found that the complainer in fact mis-heard or mis-remembered the forecast!
Brief gripe of my own: some TV or radio sources are very vague about which night the low temperature is for. For example “tomorrow, a high of 58 and a low of 35” may mean the 35 is “tonight” or “tomorrow night”. I regard the low as usually belonging to the morning of the day in question, but in some cases, it’s not clear!
Tom