THE BOOK--Playing The Percentages In Baseball

Haven’t really explored it much but http://www.weatherspark.com looks interesting

I believe Weather Underground just shows the National Weather Service forecasts. 

I actually tried collecting forecast data from different places for a couple months last year.  But I got bogged down in how to evaluate the data (what’s the difference between “partly sunny” and “mostly cloudy”?) and eventually just gave up.  Maybe I’ll take a look it later and see if I learned anything useful.

There was this a few years ago
http://www.omninerd.com/articles/Internet_Weather_Forecast_Accuracy
I don’t know of any followup.

Bill/#2, I agree that you can’t parse the data without reliable definitions. Aviation weather forecasts (aviationweather.gov), which are a product of NOAA (weather.gov) are well defined, but I’m not sure how they relate to other products. Cloud cover for pilots is defined as 1/8ths of cloud cover, but it seems that “mostly sunny” and “mostly cloudy” are mostly relative until they’re defined.

0/8ths: Clear
1/8 to 2/8ths: Few
3/8 to 4/8ths: Scattered
5/8 to 7/8ths: Broken
8/8ths: Overcast

Link: http://en.wikipedia.org/wiki/METAR

Found this site which reports forecaster accuracy percentages for the past month and year for any city/zip-code:

http://www.forecastadvisor.com/

For Atlantic City, NJ, it claims that the National Weather Service was (just barely) the most accurate over the past year (with an accuracy rate of just 57.05%).  It reports The Weather Channel has been the most accurate for Atlantic City for the past month.

Disclaimer: according to the FAQ, this site was created by a company, which if you follow the link, sells weather accuracy software to some of the forecasters being graded.

When I tried various sites a few years back, I found the National Weather Service provided the most accurate forecasts. I live in an area adjacent to and just east of the Rockies, so the weather can change rapidly and seemingly unexpectedly. The only shortcoming I’ve found is a minor one: The daily high temperature forecast is usually 5 degrees below what actually occurs.

http://www.weather.gov

I have nothing to add here except to say that the Bill Ayers moniker was clever.

I showed this to a friend of mine who works for the National Weather Service in California.  Here are some of his bullet points and notes.

- Cloud coverage descriptions are standardized within each agency but different between them (NWS vs AccuWeather).  Ask 10 people on the street to describe the sky (partly cloudy etc...) and you are likely to get 10 different answers.

- If you type in the weather forecast for Los Angeles at different websites (AccuWeather, WeatherCom, NWS etc...) you are likely getting readings from different places.  Where micro climates are a factor like in California you could get a big difference in temperature forecasts.  Imagine someone coming to visit a friend in LA and types in the weather for Santa Monica and the forecast says a high of 72 degrees, but it turns out their friend really lives in Woodland Hills (12 miles away as the crow flies) and the temperature there is 104 degrees.  Conclusion would be the weather people suck when in fact the forecasts were correct.

- Most sites don’t list a 0% chance of rain unless the chance is 15% or greater.  There could be a 10% chance of rain and it would be listed as 0%.  Surprise when it does rain.

- Temperature ranges of 4 or 5 degrees between sites is considred normal.  How many people can differentiate between 4 or 5 degrees and if so does it really matter?

- The NWS just started a team of “social scientists” to better understand all this stuff in the hopes of providing better and more relevant data to everyone.

- Radio and TV forecasts are not reliable.  By the time the forecast makes it to the TV/Radio weatherperson the “story” has changed, just like having 3 people tell the same story.  These weather people are often not trained meteorologists and are on TV for their boobs and or hair.  A station manager often will purposely change the weather forecast to sensationalize a story or perhaps they are sponsoring an outdoor event the next day and don’t want to scare people away with a 20-30% chance of rain forecast.

Any question let me know and I will pass them on.

I’ve heard from the head of NCAR that the weather channel has the best overall forecasting models out there, but obviously different models may be superior in specific regions.  Weather bug relies on a lot of stations placed at schools, etc., and so is well known for having data issues.

And re: Xei/8:  Yep, the cloudy, etc. descriptions are different for each site.  Another big one is as follows:  Say you have an area 100 miles on a side, and a 100% chance of rain over half the area.  Is that a 50% chance of rain? (NWS) or a 100% chance of rain (Weather channel?  Or a judgement call (prob. really the weather channel) All in the definitions.

Xei: thanks for that.

Here’s one to ask: why when they make the hourly forecast, do they show “25% chance of rain” each hour for practically the whole day, BUT the chance of rain for that day is ALSO 25% chance.

For example, say that the Yankees have a 25% chance of scoring in the first three innings (one hour) and a 25% chance of scoring in the next 3 innings (one hour) and a 25% chance of scoring in the last 3 innings (one hour), we wouldn’t then say they have a 25% chance of scoring for the whole game!

We’d say 75% of NOT scoring x 75% of NOT scoring x 75% of NOT scoring = 42% chance of NOT scoring, or 58% chance of scoring during the game.

Now, we can say that because each slice of 3 innings is (fairly) independent.

Obviously, rain doesn’t work like that.  But still, to have a 25% chance of rain for each and every hour would imply, to me, at least a 35% chance of raining AT SOME POINT during the day.

Or, are they simply reporting the “maximum hourly chance of rain”?

Tangotiger/10, I think this Wiki page has a good explanation:
http://en.wikipedia.org/wiki/Probability_of_precipitation

If the forecast is for 20% chance of rain for the entire day and also 20% for each hour, then the way I interpret it is that IF those 20% rain cloud(s) emerge in the forecast region at 9am, then there is a 100% chance(*) that they will still be present somewhere in the forecast region at 10am, 11am, noon, etc.

(*)I assume that “100%” could be, say 95%, since the figures are always shown in increments of 10%.  Your example of 25% and 35% may be spot on, but the forecaster might round both numbers to 30%

I hope it’s of some help to offer the following:

As noted above, NOAA offers aviationweather.gov, probably the most popular free weather website for aviation. One of the tabs there is “TAFs” which are weather forecasts for certain US airports and their five statute mile vicinities. Plug in an airport ID (remembering to put a “K” in front of it (so JFK becomes KJFK)) and you get the local forecast. Up comes a coded statement which you can get in the “Translated” format if you so select. I find that most other weather forecasts map very closely to the data found here.

Occasionally, one will see a “PROB30” or “PROB40” code in a given TAF line but, during my relatively short time of being a pilot (<5 years), I have seen these rarely. As shown in the “Aviation Weather Services Handbook,” a free pdf publication offered by the FAA that serves as a guide to the website, the PROB30 means 30% to <40% and PROB40 means 40% to <50%. Once the probability is 50% or more, for a given time period, it becomes the actual forecast. If the probability is <30%, it’s discounted to 0. Like the Holy Hand Grenade, there is a small set of numbers to be used: these are they: 0, 30%, 40%, 100%

If your weather curiosity inspires you to spend a lot of time at the site, you’ll soon discover that the NOAA weather forecasters tend to deal with very large probability ranges, at least as such forecasts serve the aviation community. I rarely view any other weather source than AWS. The lack of super-fine X% granularity keeps me from trapping myself into the false precision of single-point forecasting.

My $0.02.

For those fortunate enough to live in the greater DC or Philly areas here are too fantastic websites for weather. They boil the weather down to the simple fact that it is highly unpredictable. However, by looking at history and the multiple models available to forecast you can come up with several likely scenarios. Also, they are very thin on the weather hysteria that too often precedes big storms.

http://www.washingtonpost.com/blogs/capital-weather-gang

http://www.phillyweather.net/

#10. He said there are a lot of holes in the math for the hourly rain percentages vs chance of daily rain.  He said his office only breaks it down every 6 hours instead of hourly.  Hourly rain forecasts have huge margins of error (esp when further out in time), but give the people what they want.  The hourly rain percentages gets back to the “social scientists”.

In summary, there is room for improvement and some sites are a little loose with the facts math-wise.  Horse shoes and hand grenades.

If you’re into forecasting, check out probcast.com.  Right now it only works for the Seattle area, but it’s pretty cool and I believe it’s the future of forecasting.  It doesn’t pretend to give you an exact forecast.  It give you a projected high (82 for tomorrow) and an 80% confidence interval that the temperature will be between 85 and 78.

I live in Denver and the weather here is very erratic.. but personally, I would like a nice Android weather app that included a current wind speed reading if not a weather forecast. Makes it easier to figure out whether or not it’s book-reading weather.