More on Persistence

I suspect that not everyone finds this as interesting as I do, but nevertheless here's a follow-up on seasonal temperature persistence in and around Alaska, this time from a map perspective.  Using ERA5 reanalysis data, I calculated the correlation of consecutive monthly temperature anomalies from 1950 through 2020, with the linear trend (specific to each month) removed.  A positive correlation means that the sign of the anomaly (i.e. above or below trend) tends to persist from one month to the next, but a negative correlation indicates that it tends to reverse.

It is usually a safe rule of thumb that weather and climate anomalies tend to be "persistent" even over land - there is a positive autocorrelation - so it's a surprise to see that the overall correlation is slightly negative for a portion of east-central Alaska and an adjacent zone in northwestern Canada.

This is actually the only place on the planet that has a negative month-to-month correlation, according to ERA5 data for this particular historical period.  There are plenty of regions with very low correlations, but this small region just to the north of Eagle is the only place with inverse persistence.  Here's a map for the Northern Hemisphere extratropics.

The tendency for temperature anomalies to reverse sign is mostly found in the winter for interior and eastern Alaska.  Here's the December-January correlation:

The rest of the year is added below.  The maps confirm the observation I made in the previous post: for the state as a whole, persistence is very much heightened in April-May and in July-August.  One might say those pairs of months are temperature twins: they tend to resemble each other in terms of departure from normal.

The other striking point is how much higher persistence is in summer than in winter for the maritime southern regions, and for late summer and early autumn near the Arctic Ocean.  Clearly the warm season temperature anomalies are dominated by slowly-varying ocean temperature regimes in locations close to open water, whereas relatively chaotic atmospheric flow patterns tend to control the month-to-month temperature changes during winter or over ice-covered ocean.

Seasonal Persistence

As spring gets under way in Alaska, temperature anomalies tend to become more persistent from week to week and from month to month.  What I mean is that colder than average - or warmer than average - weather tends to stick around more in spring than in winter; but winter temperatures are more variable over the course of weeks and months.

The chart below shows some evidence to back up this claim.  I've taken the NOAA/NCEI monthly temperatures for Alaska as a whole, removed the 1950-2024 trend, and calculated the month-to-month similarity of departures from normal.  The orange columns show the traditional correlation coefficient of adjacent monthly anomalies, and the blue columns show a "persistence index" that I defined here: the index takes a value of 1 if the adjacent monthly anomalies always have the same sign (perfectly persistent), and a value of 0 if the anomalies always reverse sign from month to month (perfectly anti-persistent).

The only pairs of months with fairly strong month-to-month persistence of statewide temperature anomalies are April-May and July-August, although there's a secondary peak of modest correlation in October-November.  Month-to-month persistence from November though March is remarkably low.

Looking at weekly data from Fairbanks shows a more prominent peak of persistence in the autumn (early October at a one-week lag), and there's also a clear peak in mid-April.  These two peaks are undoubtedly related to the persistent impact of snow cover anomalies: if there's more snow than usual at those transition times, it tends to remain cold, but if snow is lacking, it tends to remain warm.

It's interesting to note a pronounced dip in persistence in early June for Fairbanks, but I can't immediately think of an explanation for that.

In Anchorage we see much more of the late summer persistence that characterizes the statewide temperatures.  It's tempting to attribute this to the persistent effect of sea surface temperature anomalies around the western and southern parts of the state, but I'm not sure why the effect would show up more prominently at just that time of year (late July, early August).

The NCEI data for the Cook Inlet climate division confirms July-August as peak season for temperature persistence in the South-Central region, but the correlations are also significantly positive throughout spring and early summer.

Hopefully readers agree that "secondary" climate statistics like this provide interesting nuance and subtle insight into seasonal climate: there's a lot more to climatology than just the progression of normals and averages.  The next step in this analysis will be to calculate persistence from the gridded data so that we can examine the spatial distribution for each time of the year.

Persistence

Continuing with July's theme of wet weather, western and southern parts of Alaska (except Southeast) have remained much wetter than normal so far this month.  Here's a "percent of normal" analysis based on NWS precipitation estimates for the latest available 14-day period.  Flood watches and warnings are out for several regions in the west. 

The wet pattern illustrates the idea that weather patterns often tend to persist for weeks or even months, but as we saw earlier in the summer, a dramatic reversal can occur too:

Thinking about this, I started wondering if weather patterns have become more or less "persistent" over the decades in Alaska.  A simple measure of persistence can be constructed by counting the frequency with which the precipitation (or temperature, wind etc) departure from normal reverses sign from month to month.

For example, if the pattern flips from warm to cold and back again every month of the year, then this persistence index is zero: the monthly anomalies have opposite signs in consecutive months.  But if all 12 months are warmer than normal, then the index is 100%: no sign reversals occurred between consecutive months.

Using statewide data from NOAA/NCEI, and using a trailing 30-year mean as "normal", the result looks like this:

Higher numbers correspond to more month-to-month persistence.  The first thing that strikes me here is how non-persistent statewide temperatures are; I would have expected higher persistence for temperature, given the large influence of nearby (and very persistent) ocean temperature anomalies on Alaska climate.  But having said that, the atmospheric circulation pattern governs a very large fraction of the temperature variability from month to month, so perhaps it shouldn't be surprising.

It's also interesting to see that persistence was relatively low for both temperature and precipitation in around 2005-2015 (the chart uses a 10-year running average), but temperature persistence has increased substantially in recent years.

The recent prevalence of month-to-month persistence is partly related to the sharp uptrend in statewide temperatures: if it's "warm all the time", then the persistence index will be high.  However, it's worth noting that the sharp increase in temperatures in the 1980s did not produce an equally dramatic rise in temperature persistence.  Here's a chart of actual decadal temperatures and precipitation; both have increased over the last 75 years.

To counter the effects of trend on the persistence index, it seems worthwhile to detrend the data.  After doing this with a simple linear trend (calculated for each month of the year separately), the persistence index looks like this:

We still see that temperature anomaly persistence has increased a lot in recent years, but the detrended perspective suggests that the 1970s and early 1980s were similarly persistent.

The precipitation result is quite interesting, with a slow and steady rise of persistence into the 1990s, a sharp change to more "flip-flopping" patterns after 2005, and then a recovery to a more normal situation recently.

How about summer and winter separately?  The chart below shows results for June through August (with the June values compared to May, so that June can be included).  Here we see a big spike in temperature persistence in the 1990s, but this is likely because summers started becoming much warmer then.  Historically cool summers became a rare breed starting in about 1987.

As for winter, the recent uptick in temperature persistence (see below) must be related to the rapid warming trend, although again it's curious that the warming in the 1980s didn't involve a similar increase in persistence - evidently there were still plenty of colder winter months interspersed with extremely warm months in that period.

Finally, the decrease in winter precipitation persistence this century looks like a "real" change compared to the more persistent regime of the 80s and 90s.  In the last 20 years or so, a wet winter month has been more likely than not to be followed by a dry month, and vice versa, i.e. the statewide monthly precipitation is anti-persistent in winter.  I find this interesting and not a little surprising.