Thanks for this. well explained. anthony.barnston noted last post "the U.S. is divided into over 200 squares that are approximately equal in area.." and that "one-third of the total number of squares is expected to be hits just by luck" since there are 3 possible outcomes (lets ignore the equal chances, although if a particular area is statistically correlated to equal chance outcomes, perhaps they should get a tax break). So while using "33.3% chance" as a "no skill" baseline is obviously good practice, a 14 by 14 grid of random numbers between (-1,0, and 1) never really matches the spatial pattern of reality... What is the verification score if the computer generated forecasts (which do tend to match the spatial data at times), are assigned to observed outcomes across different years (i.e. 2014 forecast matched with 2006 outcome)? just curious about whether the mean would be zero.