That would be an interesting study. Take weather data and forecast data from the last n years and look at how good various sources are at prediction over various regions and timescales.

I'm sure it's been done?