The charts below are parallel to those shown in the previous post. Most of the month-to-month variability in the distribution of temperature seems to be random, but there is an exception in July and August, when the negative AMO phase shows a modest association with unusually warm conditions. When the AMO is in the bottom quartile, July and August are warmer than normal about 75% of the time. Presumably they are also drier than normal, but I haven't looked at precipitation yet.
The charts below show similar behavior on a daily timescale for the negative AMO phase in summer, especially in July, and there are a few other interesting but less statistically significant features. Note that the AMO phase is currently positive, having entered a long-term positive phase in 1995.
How do we know that July is significant? How can we quantify this?
ReplyDeleteThis is really making me wonder what a multivariate fitting would show for AMO, PDO, ENSO and all of the other alphabet indices.
I can think of three things that connect an Alaskan July and the AMO: 1. Back-propagation from a blocking pattern. 2. Forward propagation of AMO enhancements to Alaska which would be amazing. 3. Not AMO at all. Just that AMO and July have a common source perhaps in the Pacific.
Eric, there would be many ways of testing significance - some I'm surely not aware of. If we consider a binomial distribution, the chance of having 5 or less out of 21 years on the same side of normal (as in July and August) is 1.3%. With 12 months of the year, the chance of one month meeting this criterion just by random chance is 14%, but the probability of two or more months like this is only 1%. So I think we can say the July-August result for negative AMO is certainly significant.
DeleteAnother approach would be to re-sample repeatedly from the historical temperature distribution and find out how often an anomaly this large is observed in a sample of 21 years.
Some problems with multivariate regression are the rather small sample size for multiple predictors (85 years) and the non-linearity of the response to each predictor, i.e. the positive phase sometimes doesn't have the opposite influence of the negative phase. There's also the problem of non-linear interference, e.g. negative AMO might affect Alaska quite differently in positive or negative PDO; the combined effect is not a simple linear addition. Having said that, I think there is potential for useful investigation in the area of combined index influences... it's really a data mining problem.
Now we're getting into a global coupled general circulation model and CM2.X analyses. Thermohaline circulation stuff and enhanced El Nino teleconnections.
ReplyDeleteI like Eric's #3 for the above reasons, but without a definitive substantiation.
Gary
I like #3 too; correlation doesn't require causation. And the Atlantic and Pacific temperature variations are certainly not independent.
Delete