January -- The Cet Lone Wolf -- See Why

[Roger J Smith]

CET analysis shows January has warmed much differently from the other eleven months

________________________________________________________________________________

__________

This analysis seems to belong more in climate change than historical weather. Using my data set of CET monthly anomalies (1659-present), I looked at each month in the more recent period since 1901 to make this point a lot easier to visualize.

There may well have been large differences in how the various months varied from long-term averages before the 20th century. However, since many believe that the later part of the 20th century began a period of "global warming" I wanted to examine how the twelve months had responded to this warming signal. The analysis uncovered something rather unexpected. January has a much different signal to the other eleven months, as the following graph illustrates.

CET_108Y.xls

Using a 21-yr r.m. of the long-term anomaly values, I compared January with the other eleven months, generating 88 data points starting with 1901-21 and ending with 1988-2008. The first note we should make is that the whole period, 1901-2008, was on the average 0.363 C deg warmer than the average of the 350-year period. That value is shown on the graph as a horizontal line that shows the years of the 20th century for reference (the last data point is about 1998). I shifted these four years to the left because program conversion seems to push labels about that far to the right (in case you're checking up on me). So in the data table, for example, "10" appears in 1906 but it shows up on the graph around 1910, and this is the data point for 1910-1930.

Now, if you look at the graph for just a few moments, you'll quickly notice that January has a much different signal from the rest of the year. These other eleven months don't look a lot different from one another, we could take a look at each of them against January and spot some differences, but essentially, January turned out a lot different from all the other months.

January warmed up from the long-term average (of 3.2 C) earlier in the 20th century, then plunged down to quite cold levels around 1940 to 1955. After that, it struggled up to lead the parade of warming, which is not that surprising since January has a higher standard deviation than most months (probably twice that of the summer months).

However, the other months (and this includes all of them to some extent) did not warm up as much as January before the 1940s, and then some of these months actually warmed while January was cooling. The spring months in particular warmed in this cold-January period. Others held steady, for example, February. Then these other months began to warm but when you look at particular cases, some started a little later than others.

You can play around with the graph if you can use Excel, the monthly average columns start with January in N, and run through to December in Y. Then in Z you find the average of all months except January, that's the column on the graph, so N and Z are on the graph that you can see here.

The basic story here is that January went through some more dramatic changes in the 20th century than the rest of the year. And it makes me wonder, if January is this unstable, then perhaps it is about to revert to the colder modality. The last time that happened, the rest of the year did not follow suit, although February seemed to try to do this about seven years later (1947-63 is a period of cold Februaries, while 1940-55 was more the period of cold Januaries).

[BLAST FROM THE PAST]

Roger

Anything in your research that may indicate a reason for this?

BFTP

[Chris Knight]

CET analysis shows January has warmed much differently from the other eleven months

________________________________________________________________________________

__________

... January has a much different signal to the other eleven months, as the following graph illustrates.

Hi, Roger.

I think that using the same visual scale for January anomaly alone and for the mean anomaly of the other 11 months of the year is a little misleading.

I have replotted January in blue, the mean anomaly for the whole 12 months in red and for the mean anomaly of the other 11 months in black, all to the same relative scale, leaving your original series plots in grey. It can be seen that the whole 12 month series is the additive curve of the January and 11 month series curves.

January's anomaly from the whole baseline period certainly had a greater contribution to the whole in the first 25 years of the series, and the latter 35, whereas over the middle 30, the anomaly was so near to zero it made little difference to the annual anomaly. The mild Januaries of 1916, 1921 and 2007 and 2008, and the cold run of Januaries in the early 40s must help shape the curve when included in 21-year running means.

But, perhaps this is an anthropic signature, relating to relative prosperity and industrial output in the early 20th century, our post WWI slowdown, depression and austerity until the end of rebuilding after WWII, in terms of extended heat island effects, during the mid winter.

(As an OT aside, I can't work out how you do the data labels, unless I am missing something very obvious - can you let me know please?)

[cheeky_monkey]

i dont understand your graph with reference to jan as plotted in blue...shows it flatlined around the 350year mean value..which isnt right???

[Roger J Smith]

Okay, I see some of the point of comparing January at scale with the average of eleven other months, however, the comparison month by month looks more like the visual I first presented. January is different from 1901 to 1960, basically ... while it starts out quite mild for 35-40 years and then goes quite cold for the next twenty, most of the other months stay in the colder climate signal that had accumulated through the 19th century, then started to warm in the 1940s (except February which had a later cold trough), then cooled slightly before joining January. Looking again at February, I would say it cooled in phase with January then got even colder in the 1950s and early 1960s before warming.

My point should be clarified that the more recent period is the most similar part of the comparison. January warms rapidly like most other months do. However, some flatten out as you can see by comparing any two months. December in particular did most of its warming in the 1980s and then started to flatten out. I don't think I meant to imply that January was making a larger contribution than other months to trends, although clearly it was making a different kind of contribution to trends from 1901 to 1960. If all the other months had followed January, we would be talking about a return to the warmth of the early 20th century after the cold 1940s and early 1950s. I guess all the cold January cases in the period 1979-87 got averaged out by mild cases in that upward climb.

Now, as to the label question, I'm not sure what labels you're asking about. The problems with labels come because I have all my research files on "aseasyas" which I cannot upload to net-weather, so I have to convert them. When I convert them, some of my labels shift to the right about three or four data points. So if I have specified "above" for the curve label for January, I have to pick a spot where an almost-peak occurs 3-4 years before an actual peak, then the conversion places it over the actual peak, but if the near-peak is any amount below the actual peak the label sits at that level. So far, haven't been able to solve this problem. The year data labels that I mention on the arbitrary line .363 C (the period mean) actually didn't shift to the right as far as I expected they would. Apparently the conversion for numbers is smaller than letters.

Those numbers are probably closer to 07, 17, 27 etc. They only shifted one unit to the right. But in any case they are only there to give an overview of the trends. (Edit ... I fixed the positions of these year labels in the graph posted below. But please note what the numbers mean. 10 means this is the data point 1910-30. 20 means 1920-40. etc.)

If people can download the chart and manipulate it on their own computer, then I would say leave range A or 1 as is (January) and change range B or 2 to any other month you want to compare. January's 88 data points are in column N. Any other month is to the right of that, for example, July is in column T. What I have in range C or 3 is the arbitrary mean.

I have to admit, I haven't gained as much familiarity with Excel as some of you may have, I know aseasyas like the proverbial back of the hand, but once I convert a file to excel, I don't know where half of the applications go ... finding my label ranges or the curve symbol choices for example, is a step I have not yet taken. Whatever, the graphs essentially show the same thing in both programs. One other thing that appears in the conversion is that first data point in the x range, that long number that appears lower left outside the graph. That range is defined as void in the aseasy program and no such number appears (it's the value of the first data point in range A). The conversion also defines arbitrary scales for the y axis which aseasyas asks you to define. If I leave those in, the conversion tries to plot these limiting values in the graph.

If you do this month by month, you'll see my point more clearly. If I get requests I may post these eleven graphs one after the other, but I assume it's just as easy for you to do it on your own computer.

Actually, what I will do is to post the comparison of January to March here. Now the scaling of the two is not a factor. However, I wonder if this scaling question is a false paradigm. Think of this, if all eleven months had the same signal, then their average would look the same as any of their individual curves, so where in that would you need to scale January by reducing its variability by a factor of eleven? Now the other months don't look exactly the same, but they have a broad similarity so I think the reduction should perhaps be more like one in three or four, not one in eleven, to give the right visual comparison.

In any case, here's January vs March... I corrected the year labels so they appear slightly to the right of the first graph, and in the right spot now.

Anyone doing those other comparisons should use this post as the other data are also in this file.

CET_JAMA.xls

(Fred, I have some preliminary ideas about cause related to theory. It will take quite some time to investigate these ideas, possibly three to six months -- I don't plan to spend the whole time on this either, but I will give it some thought. What about you? Any ideas? )

[Chris Knight]

i dont understand your graph with reference to jan as plotted in blue...shows it flatlined around the 350year mean value..which isnt right???

Hi Cheeky_monkey

The anomalies for each month are variations on either side of an arbitrary mean, in this case the mean for each calendar month between the years 1659 and 2008. This gives 12 baselines, one for each month, the deviations from which 21-year running means are calculated for each of the twelve months.

This would give this spaghetti graph, which is difficult to interpret:

No one can fail to see that the running mean for January appears to be higher at the beginning of the period, then gets lost in the general mass of lines towards the end.

Roger then compares this January single month curve to that of the mean of the other eleven months, i.e. (FEB+MAR+APR+MAY+JUN+JUL+AUG+SEP+OCT+NOV+DEC)/11, which is not actually comparable.

In the graph I posted, the red curve is (JAN+FEB+MAR+APR+MAY+JUN+JUL+AUG+SEP+OCT+NOV+DEC)/12 ,

the black curve is (FEB+MAR+APR+MAY+JUN+JUL+AUG+SEP+OCT+NOV+DEC)/11 * 11/12 ,

and the blue curve is JAN/12.

Why is this?

It is because

(JAN+FEB+MAR+APR+MAY+JUN+JUL+AUG+SEP+OCT+NOV+DEC)/12 =

JAN/12+FEB/12+MAR/12+APR/12+MAY/12+JUN/12+JUL/12+AUG/12+SEP/12+OCT/12+NOV/12+DEC/12

QED

I guess my comment re the labels is irrelevant, since it is a feature of "aseasyas", of which I am not familiar. It seems to be tricky to label in a similar way in Excel. Thanks Roger

[Roger J Smith]

Thanks Chris, the spaghetti graph is not necessarily that difficult to follow, too bad there aren't twelve distinct colours but you can certainly see that October warms up anomalously around the 1950s ahead of other months, for example.

I have this file marked for further study, it is pretty complicated but I would put this idea out for discussion -- if the months show considerably different signals then would this not argue more for natural variability than AGW as the cause of the warming? If the months all respond the same way, or at least in some organized way, then perhaps vice versa.

So which do we see here, the evidence seems to be a growing randomness of signal around 1930 to 1960 and then a more unified warming but neither of these are absolute. This would argue for a period of natural high variance followed by a period of natural warming increasingly enhanced by AGW. Which is what I've thought subjectively for quite some time.

[BLAST FROM THE PAST]

(Fred, I have some preliminary ideas about cause related to theory. It will take quite some time to investigate these ideas, possibly three to six months -- I don't plan to spend the whole time on this either, but I will give it some thought. What about you? Any ideas? )

Hi Roger

Not really but a potential rhythm as we see a gap mid century where Jans were colder. I'm wondering if solar output and the PV set up is linked as it seems to be the PV that has scuppered Januaries for last 20 years in particular. [ there seems possible too for a rhythm within GWO research]. I must admit I remember [sF] I think bringing tis up about 2 years ago that January seems to have suffered more than most. The reason why I think you might have an answer in your research model is because I think due to our position on this planet we are in one of your timing lines. Has similar been found elsewhere?

Fred

[Android]

Given the lack of noise in each trend, I am suprised they are that far apart. Why are successive Octobers so much warmer in the middle of the graph than successive Septembers for example? And the beginning of the January trend just looks odd. Why would an artibary 31 day segment of the year have such a different trend than other 31 segments around it? The lack of noise suggests it isn't chance.

[Roger J Smith]

The smoothness probably reflects the 21-year running mean, this tends to eliminate a lot of shorter term signal noise.

The early period January warmth is partly due to the fact that January was a much colder month in general before about 1890, the long-term mean is 3.2 C which is quite a bit lower than the 20th century mean of about 4 C.

So the fact that the first 35-40 years of the century failed to produce many cold Januaries allowed this curve to start out quite a bit higher than the others, where every other month had its share of cold among the warmer values that were becoming more normal in the early 20th century.

As to the spurt of warmth in October around mid-century, that seems to correlate with the high frequency of hurricanes in those decades, so a stirred up Atlantic with a lot of poleward motion of air masses in general, probably the October 2005 and 2006 sort of pattern was often found there. It died out in the Novembers of the period. Another factor to consider would be low solar angle coming into play in that era of coal-smoke haze, this may have artificially depressed temperatures in the lower insolation months especially Nov to Jan.

I suppose it all comes down to testing the situation for randomness in general, if all months were in perfect lock-step that would defy the odds, so how much variance should one expect to find?

It's interesting how December has not gone along with many other months in the AGW period, it seemed to take a jump upwards a bit earlier then this trend flattened out more recently. December, perhaps, has done all the warming it can do already. And that's probably a good thing.

[Chris Knight]

One thing that strikes me here is that we are dealing with abstractions upon abstractions here:

- the calendar months are arbitrary periods of time*, roughly dividing any year into 12 unequal parts.

- monthly mean temperatures are at least two calculations away from any measured temperature - the sum of daily mean temperatures divided by the number of days of the month* - themselves being fictional, not mathematical, means - the sum of maximum and minimum temperatures divided by 2 over any day, rather than a true mathematical average.

- the daily HadCET data is a comparison of several sites, not always operational, interpreted by Manley from incomplete data over much of the early record, and even today, corrected against biases by various algorithms to provide mean temperatures within a certain defined area, and to account for missing or spurious data.

- in the case of a 21-year rolling mean smoothing, a couple of exceptional monthly means within a few years of each other can influence perhaps 20 years of the curve, especially towards the beginning and ends of the graph.

- our arbitrary monthly baseline means, to work out the monthly anomalies, are taken from the beginning of the data series to the end - which creates different monthly biases from other baseline periods. As an example, try any single year monthly means as a baseline, and see how the picture changes.

However, we can partially address the baselines problem by making the dataset self-consistent - using the baseline as the mean anomalies from the period 1901-2008

Back to the data: Dividing the "spaghetti graph" into seasons (another artificial division of time), may enable us to see similar trends in adjacent months, and the end and beginning months of the seasons.

I have done this with Roger's original data, in the excel sheet attached.

I have also produced another worksheet from the HadCET monthly mean temperature series - not the anomalies.

From this I have first produced 21 year running means for each month, and the annual series.

Next I produced the anomalies of these data from the means of each of the monthly columns, but this time, self-consistent with the baseline of 1901-2008.

This is plotted as "Anomalies spaghetti" Not only January stands out now, other months seem to have had their periods at the top of the tree, and it does not look random to me.

Because the anomalies are self-consistent, it is possible to get rid of the annual warming and cooling trends, to see if there is any underlying relationship in the monthly patterns masked by the overall warming and cooling.

Play with this:

CET_108Y_1A.xls