Correlation of Accumulated Cyclone Energy and Atlantic Multidecadal Oscillations

Guest essay by Michel de Rougemont

Various sources, scientists publishing their opinion in the media, claim that Tropical Storm Harvey, recently landed in Texas, is one more signal of the influence of global warming on such catastrophic events. These claims are based on model calculations. Let’s examine the facts.

In the Atlantic Ocean, sea surface temperature oscillations are observed as a multidecadal cycle (AMO).

Figure 1 Atlantic Multidecadal Oscillations (AMO). Monthly series of sea surface temperature. Smoothing by centered running averages AMO_13m, over 13 months, AMO_7y, over 7 years,  AMO_7ym: of each month of the year, over 7 years.


The total energy accumulated each year by tropical storms and hurricanes (ACE) is also showing such a cyclic pattern.

NOAA’s Hurricane Research Division explanations on ACE: “the ACE is calculated by squaring the maximum sustained surface wind in the system every six hours (knots) and summing it up for the season. It is expressed in 104 kt2.” Direct instrumental observations are available as monthly series since 1848. A historic reconstruction since 1851 was done by NOAA (yearly means).

Figure 2 Yearly accumulated cyclone energy (ACE) ACE_7y: centered running average over 7 years


A correlation between ACE and AMO is confirmed by regression analysis.

Figure 3 Correlation ACE=f(AMO), using the running averages over 7 years. AMO: yearly means of the Atlantic Multidecadal Oscillations ACE_7y: yearly observed accumulated cyclone energy ACE_calc: calculated ACE by using the indicated formula.


Regression formula:



Thus, a simple, linear relation ties ACE to AMO, in part directly, and in part with an 18 years delay. The correlation coefficient is astonishingly good.

Origin and mechanisms directing AMO are quite unknown. Therefore, any speculation aiming at associating cyclone energy with other phenomena, as e.g. anthropogenic climate warming, would need a clear and cut irrefutable proof. Model speculations cannot serve as proof without full validation.

Published on MR’s blog., submitted to WUWT by the author.

clip_image014About the author:

Michel de Rougemont, chemical engineer, Dr sc tech, is an independant consultant.

In his activities in fine chemicals and agriculture, he is confronted, without fearing them, to various environmental and safety challenges.

His book ‘Réarmer la raison’ is on sale at Amazon (in French only)

He maintains a blog, as well as a web site dedicated to the climate


He has no conflict of interest in relation with the subject of this paper.

0 0 votes
Article Rating
Newest Most Voted
Inline Feedbacks
View all comments
September 4, 2017 9:18 am
Tom Halla
September 4, 2017 9:30 am

As the cycling of the AMO predates any rise in CO2, this is one of those vague “natural causes” that the IPCC dismisses, at least since 1950.

Reply to  Tom Halla
September 4, 2017 10:41 am

That is a very good point, see my graph and the linked article below.
IPCC uses selection bias by concentrating the conclusions on “the last 50 years” or “the latter half of the 20th century”. This is dishonest and unscientific since they are just preselecting a period with a fairly monotonic rise and brushing away an equally important earlier rise and a cooling period. Clearly these are both highly significant in assessing what is natural and what is possibly attributable to human causes.

September 4, 2017 9:39 am

The AMO cycle follows the solar activity cycle, at least since 1900. I suspect that there is a pumped resonance at work.

Reply to  ShrNfr
September 4, 2017 10:36 am

Really ? not so much since 1990.

September 4, 2017 10:05 am

This is very similar in essence to my “Ace in the hole” article which was published on Judith Curries site last January. Although this analysis much less thorough and uses the ubiquitous, crappy running mean as a filter.

Reply to  Greg
September 4, 2017 10:25 am

The correlation coefficient is astonishingly good.

Astonishment is a personal reaction , not an objective scientific measure. Firsly you need to assess what level of correlation would be significant before being astonished.
Since you are running the correlation test on data filtered and distorted by 7y running mean it is questionable what the result shows. You should at least run the correlation using the annual data ( since that is base resolution of annual ACE totals ).
One key point is that having used a 7 rather heavy filter you have severely reduced how impressive any correlation is since you have very few independent data points.
Also each time you add in another explanatory dataset ( eg the arbitrarily delayed AMO ) you again lessen the astonishment value of getting a good fit since you have intorduced two more arbitrary fiddle factors: the scaling and phase shift.
Yes, there is a basic underlying similarity. This is not astonishing since hurricanes derive their energy from SST and this is the basis of the simplistic claims the more “global warming” will cause more cyclones.

The “astonishing” part is the 10y hurricane drought, just when we are supposed to be in the WARMEST DECADE EVAH , which you fail to even comment on. Not only is there no land-falling major hurricanes, there is a crash in total ACE across the whole N. Atlantic.

With some less distorting filters and using non-detrended “AMO” you find an even better match and the crash of ACE during the hottest periods become even clearer, and no need to use arbitrary and physically unjustified 18 y lags of the data.comment image
SST is an important factor but hurricanes are not that simple.

Reply to  Greg
September 4, 2017 11:01 am

The evidence is that hurricanes don’t like “plateaux” or “pauses”. If N. Atlantic ACE is waking up from a 10 year slumber this is probably an indication that we are entering a cooling period like the post war cooling in the N. Atlantic .

Reply to  Greg
September 4, 2017 11:02 am

BTW those bumps in the post 1975 warming period are 9.1 years long. This is very likely to be a lunar forcing.

Gary Pearse
Reply to  Greg
September 4, 2017 3:18 pm

Greg, very interesting. It seems that climate science feels itself more sciency when it does a load of statistics on data. Your plot is even better without. Are the Ssts ‘adjusted’ figures a la NOAA-Nasa sources, though?
Continual adjustments of data must ultimately destroy any legitimate correlations.

Reply to  Greg
September 4, 2017 4:11 pm

Gary, the data link is on the graph, ERSSTv3 , go and dig what you can find.
Data sources and refs at the bottom of my article if you prefer:
AFAIK there has been no attempt on either side to make these two datasets agree with each other. ( You will recall that Chris Landsea threatened to sue the IPCC if they did not take his name off AR4 because they totally ignored his input about hurricanes. ).
This graph is one thing that gives me a little more confidence that HadSST adjustments are not too far wrong at least in the post WW11 drop. most of which was an adjustment, later adopted by the other players.
Spectral analysis of the cross-correlation shows peaks at 9.1y and about 60y.comment image
The 9.1 is a strong indication of a lunar forcing. For some odd reason we readily accept that the moon affects tides and shifts huge volumes of water horizontally but we never hear any discussion of it affect on climate through longer lunar periods. ( Other than Scafetta who gets ridiculed for even suggesting it ).
There was also a paper from BEST on which Judith Curry was an author which similarly reported finding 9.1 year periodicity in both AMO and PDO.

Reply to  Greg
September 4, 2017 9:50 pm

Greg, what is the horizontal axis in this graph?

Wim Röst
Reply to  Greg
September 4, 2017 11:52 pm

Greg September 4, 2017 at 10:25 am: “SST is an important factor but hurricanes are not that simple”
Depressions are formed as cold air meets warm air. More or less like eddies are formed in the oceans when cold water meets warm water as we see near Newfoundland.,41.67,1657/loc=-65.181,40.162
Hurricanes are large scale tropical depressions.
To get hurricanes formed, we need warm and cold air near to each other and a wind that combines the warm and the cold air. We need that situation in the area where hurricanes are starting to develop: in front of West Africa.
Air adapts to sea surface temperatures. Present SST are for example shown here:,25.09,740/loc=-29.529,16.341
We find SST anomalies up to 2.1 °C warmer around the Capeverdian islands and up to 0.8 °C colder northwest of the Canary Islands. A northeaster wind connects.
To me it seems that the occurrence of this situation stimulates the present development of hurricanes.

Reply to  Greg
September 5, 2017 10:43 am

“A historic reconstruction since 1851 was done by NOAA (yearly means).”
Reconstructions are not real-time data.
Strike one.
NOAA is not to be trusted due to their pro-global warming bias
and false claims that 2016 was a few hundredths of a degree C. warmer
than 2015 — far less than any reasonable margin of error,
Strike two.
There is no good reason to trust “a historic reconstruction” “done by NOAA”.
This article has to be dismissed as a logical theory based on questionable data.
In addition, only two complete AMO cycles may not be enough data for a confident conclusion.

Reply to  Greg
September 5, 2017 12:10 pm

Greg, what is the horizontal axis in this graph?

Sorry, it is less clear out of the context of the linked article. It should have been labelled explicitly.
It is spectrogram showing frequency and the units are “per year”, so p2 at about 0.11 per year, corresponds to a period of 9.07 years. p1 is close to 60 years.

Reply to  Greg
September 5, 2017 12:23 pm

Wim Rost:

We find SST anomalies up to 2.1 °C warmer around the Capeverdian islands and up to 0.8 °C colder northwest of the Canary Islands. A northeaster wind connects.
To me it seems that the occurrence of this situation stimulates the present development of hurricanes.

What you say may well be true of what creates this intial depressions. That may be a key factor in explaining the hurricane drought if the initial depressions are not forming.
On the other hand SST is one factor which make a tropical depression develop into a major hurricane. Moist air is lighter than dry air and thus rises, drawing in surrounding air to replace it. This air is ( appears ) deflected by coreolis forces bending its path to the right in NH, giving rise to the anti-clockwise rotation. This wind also invokes more evaporation leading to more of the less dense moist air …. this is a positive feedback situation leading to an ever stronger storm.
There are many other factors such as wind shear which tends to break up the cyclonic structure. This is why it is not a simple case of warmer water leading to more and more powerful storms. Though there is a clear correlation.

Reply to  Greg
September 5, 2017 1:02 pm

daveburton: “Greg, what is the horizontal axis in this graph?”
Sorry it should have been explicitly labelled. It is a spectrgram showing frequencies, given in “per year” units. p2 at about 0.11 thus corresponds to 9.01 years, p1 is close to 60 years.

Reply to  Greg
September 6, 2017 11:35 am

Thanks, Greg.

Reply to  Greg
September 4, 2017 1:30 pm

If you want to know what is wrong with runny mean filters just look at what is going on in Fig.2 in the 80s adn 90s. The 7y filtered average is totally out of phase with annual data it is supposed to be derived from : peaks where there were troughs and vice versa. That alone is degrading the correlation.
If you have 140y of annual data and you do a 7y runny mean, you now only have 20 truly independent data. That makes the R2 values much less “astonishing” since the level of correlation which could arise in similar lengths of random data is quite high.
A ( proper ) low-pass filter is fine as a visualisation aid but you should do the correlation test on the original data.
Also why use a detrended AMO index? That is implicitly accepting that the trend part is “known” to be of human origin, why else would you remove part of the data.
In fact if you want to understand hurricanes you need to look at them in relation to actual SST, not “detrended” SST. It is SST which is the primary source of energy. As can be seen in my graph, using N. Atl SST with the trend intact, provides a much better fit than inexplicably detrended AMO with a 18y lagged echo added back in for fun.
Have at ’em , Occam.

Gary Pearse
Reply to  Greg
September 4, 2017 3:20 pm

Hey I asked just this question above your comment here!

Reply to  Greg
September 5, 2017 2:08 pm

Spectral analysis of the cross-correlation shows peaks at 9.1y and about 60y.

Since, unlike autocorrelation, cross-correlation is not an even function of time-lag, it’s by no means clear what “spectral analysis” has been performed here. Is it the cosine transform, the sine transform, or some combination of both? In any event, this is not a proper cross-spectrum analysis, whose result is complex-valued and the spectral relationship between the power densities of the two variables is specified by the dimensionless coherence and the cross-spectral phase.
BTW, the abcissa should be labeled “frequency,” specified here by cycles per year.

September 4, 2017 10:06 am

There is also curious coincidence with (15 years?!) delayed with the atmospheric pressure as measured at Reykjavik

Reply to  vukcevic
September 4, 2017 10:08 am

or in english : There is also a curious coincidence with the (15 years?!) delayed atmospheric pressure as measured at Reykjavik

Reply to  vukcevic
September 4, 2017 10:30 am

Vuk’ this would be easier to assess visually if you presented both data sets in the same way: a continuous line. Have you used cross-correlation to determine the lag at which the max correlations occurs?

Reply to  vukcevic
September 4, 2017 10:43 am

Hi Greg
…might revisit at the end of the current hurricane season

Reply to  vukcevic
September 4, 2017 10:52 am

Thanks but that also masks the periods where one plot drops behind the other. If you do an update how about a straight plot of the two variables?

September 4, 2017 10:15 am

Model speculations cannot serve as proof without full validation.

Our beloved Naomi Oreskes tells us that is impossible. link In a reasonable world that should put the kibosh on alarmists trying to prove their point using models … but no …

September 4, 2017 10:38 am

AMO unsmooth, long: Standard PSD Format – monthly

September 4, 2017 10:44 am
Reply to  chaamjamal
September 4, 2017 10:58 am

Why this obsession with “trends” in climatology. The result heavily depends upon when you start and when you finish. If looking at data with a large circa 60y variability this can be very misleading.
Pacific has different length periodicity. Trends are near worthless in this context but the whole discussion for the last 30 years seems to be based on “trends”.
Hey fellas, this is science, not econometrics. Get beyond line fitting in Excel.

Pamela Gray
September 4, 2017 11:08 am

The failure in anthropogenic climate research is not ruling out confounding factors, IE the real factor(s) driving both increasing CO2, globing warming, and historically significant weather events that are of natural origin. And here’s the rub, we don’t know enough about natural variation to rule it out. Why? The AGW craze sucked all the money for such research down a rabbit hole.

Reply to  Pamela Gray
September 4, 2017 2:10 pm


September 4, 2017 11:35 am

First sentence states”…Tropical Storm Harvey recently landed in Texas…” .It was a cat 4 that’s a hurricane. No more comments!

Steve R
September 4, 2017 12:27 pm

Does ACE account for variations in areal coverage of storms? Or is it based only on intensity and time?

Reply to  Steve R
September 4, 2017 12:46 pm

ACE = accumulated cyclone energy ; accumulated meaning it is an annual total of all storms in that year.
If you want to know more Google is your friend.

Reply to  Greg
September 4, 2017 1:03 pm

… or read the article and follow the links provided !

Reply to  Greg
September 5, 2017 4:11 pm

Per recent announcements, Google isn’t anyone’s “friend” anymore…

Gary Pearse
September 4, 2017 3:38 pm

Are we satisfied that Harvey was a Cat 4, or do they revisit this stuff once everything’s calmed down? It wasn’t as big a blow as Cat 4, the rain notwithstanding. They seemed to use aircraft measurement from above where every hurricane is a Cat 4.

Reply to  Gary Pearse
September 4, 2017 5:10 pm

I am no expert but I agree with Gary’s question. I was in Broward County during Andrew. Found this at…
” An automated station at Fowey Rocks reported 142 mph sustained winds with gusts to 169 mph (measured 144 ft above the ground), and higher values may have occurred after the station was damaged and stopped reporting. The National Hurricane Center had a peak gust of 164 mph (measured 130 ft above the ground), while a 177 mph gust was measured at a private home. Additionally, Berwick, LA reported 96 mph sustained winds with gusts to 120 mph.”
I don’t recall any measured wind speeds near these for Harvey.

David A
Reply to  Macusn
September 4, 2017 5:41 pm

There were none. Carla top 5 wind gusts vs Harvey show the Carla gusts were about 40 mph stronger!

David A
Reply to  Macusn
September 4, 2017 5:42 pm

Going by ground based wind gusts and wind speed, Harvey is the weakest Cat 4 in U.S. history.

September 4, 2017 5:39 pm

Three of us already have credit with the Library of Congress on the Correlation of Sunspot Activity and Accumulated Cyclone Energy. My work is at See Greek Conference. As I was looking at my work, sunspot activity easily correlates into your charts.

Loren C. Wilson
September 4, 2017 7:57 pm

ACE does not properly account for the size of a storm, just the max wind speed. Why can’t we use a method that integrates area times velocity, now that we have entered the computer age?

September 5, 2017 1:34 am

Then we all agree Dave it was not within the tropical stormcategory, it was a hurricane . Some are trying to downgrade it to a 3.

Reply to  Ron
September 5, 2017 11:54 am

However, the majority of the damage is from Tropical Storm Harvey (which it became after landfall, like every hurricane that actually moves over a major land mass).

September 13, 2017 6:21 am

Greg September 4, 2017 at 4:11 pm
Spectral analysis is the way to go, not regression. But a longer record is needed, possibly from proxies.

%d bloggers like this:
Verified by MonsterInsights