Wednesday, September 19, 2012

Weather Whys Wednesday: Long Term AO/NAO Forecasting Troubles

Last week, I posted our Phillyweather.net Facebook page, asking if anyone had some topics they'd like to see covered for Weather Whys Wednesday columns. We got a few responses (you can post one there if you have an idea, as I've bookmarked the status and will check it periodically).

One of the topics that seemed to stand out was the predictability (or lack of) of the NAO and AO, two oscillations we use, especially in winter, to help determine if we're setting up for Arctic cold or snowstorms. Tom posted a great primer on the NAO leading into last winter. Now, even reading that alone, you see why trying to predict these things on a seasonal basis is near impossible. We can certainly draw conclusions from things, such as summer NAO, and they can be rather firm conclusions. But what will happen in reality is another matter entirely. As for the AO, it sort of goes hand in hand to some extent, and I discussed this a bit last December.

Three month running mean of the NAO
since 1950. Credit: NOAA CPC
So the question a lot of you have is: Why are the NAO and AO not particularly predictable with much skill on more than a couple weeks lead time?  I think one mistake made by a lot of amateur meteorologists (and even those of us in the professional community are sometimes guilty of it too) is that we put too much emphasis on the AO and NAO as numeric values. For instance, we look at monthly values of these oscillations and try to find matches and similarities based on patterns and trends of the numbers. As I mentioned in the previous paragraph, yes, you certainly can use that information to draw some conclusions. But what gets ignored is the bigger picture. Where is the NAO block? Is it on the western part of Greenland or closer to Europe? What's dominating the pattern globally? Nationally? What are conditions downstream from the cold air supply (is there snowpack or open prairie)? The physical meteorology of the NAO and AO can tell you much more than the numeric value alone.

But on the whole, the reason that predicting the tendency of the AO/NAO at long lead times is so difficult is the same reason predicting the forecast for any given location at long lead times is difficult. The AO and NAO are basically just numeric values assigned to measurements of air pressure related over distances. It is challenging to predict air pressures at any given location at lead times of longer than 10-15 days, let alone 3, 4, 5, etc. months in advance. We can make some assumptions about things, but sometimes those will be incorrect. As pointed out in the link above, we know that when the NAO is staunchly negative in summer (like last summer), it tends to linger toward that tendency in winter (eight out of the 14 lowest summer -NAOs prior to last winter led to -NAO dominated winters). But obviously just because the odds stack that way, if you end up with another signal that dominates in the winter, you can throw those stats and odds out the window (as we had to last winter). And in the majority of years, when the NAO/AO don't particularly shine strongly in summer, it's tough to make judgments at all about what that means going into winter.

The patterns over Greenland and the Arctic change so quickly (changing the NAO/AO tendencies), similar to how our own weather changes so quickly sometimes. If you look at charts or tables of the NAO over time, they're incredibly noisy. While there is research suggesting that indices like that tend to behave in cycles, the noise within those cycles and its impacts on weather are less predictable and explainable than, say, El Nino or La Nina. Ocean temperatures tend to change more slowly with time and behave better, so as a consequence, their cycles are slightly easier to predict. I personally think forecasting the AO and NAO (or any teleconnection index) at longer lead times will only improve as operational weather forecasting improves with time. But unfortunately, we seem to be plateauing in our ability to operationally forecast better at longer lead times.