More on October and Winter Temperatures in Fairbanks

Continuing on the thread that Richard of Georgia asked of "Do October temperatures have predictive value value for the upcoming winter?"

A few days I presented a scatter plot of mean October temperatures vs. mean December through February temperatures, and the results were not encouraging.

Here I abstract away from the actual mean temperatures and give you the rank correlations for the 82 winters since 1930-31 (1 being the coldest and 82 being the warmest).  So, each point represents a pairing of the rank in the 82 winters of October mean temperature and the corresponding rank of the following mid-winter. As you can see, the correlation is pretty weak. However, just looking at the ten coldest Octobers (far left on the x-axis), what I see is that the following mid-winter has not been among warmest, while with the warmest Octobers (far right on the x-axis), there are no really cold mid-winters. So, what I get out of this analysis is that if there is an unusually cold or warm (top ten) October, then past performance suggests that the following mid-winter will not be among the extremes of the opposite sign, i.e. a very cold October is most unlikely to be followed by very warm mid-winter, (but could easily be above normal) and vice versa. Outside of the extremes, I find no useful signal at all.

I did look at the 1946-1976 negative PDO phase and the 1976-2007 positive PDO phase. During the negative PDO, the ranked correlation was slightly higher (0.19) than the full 82 years, while during the positive PDO phase the ranked correlation was an abysmal 0.08.