It’s been nearly a year since I looked at the Arctic Oscillation data. One reason I haven’t paid closer attention to this is because it doesn’t show the cyclical patterns that AMO, PDO, and ENSO do. I haven’t run a correlation analysis on the data (yet) to determine whether or not it appears to depend more on regional temperature, or whether it seems to drive the regional temperature, but it doesn’t appear – at least in the short term – that there is a clear cycle that we can hang our hat on and say with any certainty that certain conditions can or cannot be expected over the next few years.

The same kind of analysis is done here as presented in my previous posts. I do have a correction to make on the long-term sine wave, however. In my previous post I made an observation that the long-term sine wave suggested a pattern for the Arctic on a 9500 year cycle. That calculation pulled the wrong values. The new fitting and corrected calculation indicates a sine wave with a full cycle completed in 368 years.

Even that number is nothing I’d hang my proverbial hat on. Trying to speak to the length of a cycle that is hundreds or thousands of years old on the basis of 60 years of data is a suspect exercise. I only point it out because I mentioned it as a point of interest in my previous post. I now see that the comparison is not apt and that particular point of interest is meaningless. I apologize for the confusion.

Since the last update, we saw a stretch of positive anomalies in 7 of the next eight months. The last three anomalies have been negative. The anomalies for June and July were both less than -1.3000.

The best-fit curve itself is scaled by a factor of 2.929. Whereas the AMO, for example, ranged between +/-0.20, the Arctic ranges between +/-4.00, but mostly between +/-3.00. Thus, the higher scale factor. As mentioned, the curve itself is quite flat, fitted to reveal a 367 year cycle.

There is little vertical shift required, so the zero line is right about where it should be based on the dispersion of the data. The shift is a mere -0.0031, which is close enough to zero to call it that.

One interesting thing I noted in looking through the data was the average squared distance from the curve in different time periods – a variance of sorts, not from the overall mean, but from the best-fit curve. Here are the time-periods and the average variance value:

1950-1954: 0.6603

1955-1959: 1.0570

1960-1964: 0.9414

1965-1969: 1.2891

1970-1974: 0.5337

1975-1979: 1.1788

1980-1984: 0.7463

1985-1989: 1.2549

1990-1994: 1.1941

1995-1999: 0.7523

2000-2004: 0.7847

2005-current: 0.7929

I wish I could tell you if that has any deep meaning. But what I can tell you for sure is that the period-to-period deviations around the curve over the last 15 years shows the most consistent limited fluctuation values in the data. A couple periods were lower, but they are bookended by much higher values. I have no idea if this is an indicator of anything in particular, but I thought it to be an interesting observation.

Again, I present the smoothed charts. The longer-term averages have a lot of autocorrelation, and the spike in average is driven and sustained by 3 pretty high anomalies in the early-mid 1990s. The overall trend of the average is upward because of the combination of those anomalies and the dropping out of some lower anomalies in the 1970s. It’s kind of interesting to see that show up in the longer-period averages since the raw data chart doesn’t seem to show that as much. However, part of the reason for this is the scale. The scale on the longer-term average charts is much lower (+/-0.5 vs. +/-4.0) so the trend looks steeper than it probably is. That said, the 10-year average is what it is, and it is definitely higher now than it was 40 years ago, though it is quite a bit lower than the peak averages of a decade ago.