Digital Signal Processing analysis of global temperature data time series suggests global cooling ahead

This DSP engineer is often tasked with extracting spurious signals from noisy data. He submits this interesting result of applying these techniques to the HadCRUT temperature anomaly data. Digital Signal Processing analysis suggests cooling ahead in the immediate future with no significant probability of a positive anomaly exceeding .5°C between 2023 and 2113. See figures 13 and 14. Code and data is made available for replication. – Anthony

Guest essay by Jeffery S. Patterson, DSP Design Architect, Agilent Technologies

Harmonic Decomposition of the Modern Temperature Anomaly Record

Abstract: The observed temperature anomaly since 1900 can be well modeled with a simple harmonic decomposition of the temperature record based on a fundamental period of 170.7 years. The goodness-of-fit of the resulting model significantly exceeds the expected fit to a stochastic AR sequence matching the general characteristic of the modern temperature record.

Data

I’ve used the monthly Hadcrut3 temperature anomaly data available from http://woodfortrees.org/data/hadcrut3vgl/every as plotted in Figure 1.

clip_image002

Figure 1 – Hadcrut3 Temperature Record 1850-Present

To remove seasonal variations while avoiding spectral smearing and aliasing effects, the data was box-car averaged over a 12-month period and decimated by 12 to obtain the average annual temperature plotted in Figure 2.

clip_image004

Figure 2 – Monthly data decimated to yearly average

A Power Spectral Density (PSD) plot of the decimated data reveals harmonically related spectral peaks.

clip_image006

Figure 3 – PSD of annual temperature anomaly in dB

To eliminate the possibility that these are FFT (Fast Fourier Transform) artifacts while avoiding the spectral leakage associated with data windowing, we use a technique is called record periodization. The data is regressed about a line connecting the record endpoints, dropping the last point in the resulting residual. This process eliminates the endpoint discontinuity while preserving the position of the spectral peaks (although it does extenuate the amplitudes at higher frequencies and modifies the phase of the spectral components). The PSD of the residual is plotted in Figure 4.

clip_image008

Figure 4 – PSD of the periodized record

Since the spectral peaking is still present we conclude these are not record-length artifacts. The peaks are harmonically related, with odd harmonics dominating until the eighth. Since spectral resolution increases with frequency, we use the eighth harmonic of the periodized PSD to estimate the fundamental. The following Mathematica (Mma) code finds the 5th peak (8th harmonic) and estimates the fundamental.

wpkY1=Abs[ArgMax[{psdY,w>.25},w]]/8

0.036811

The units are radian frequency across the Nyquist band, mapped to ±p (the plots are zoomed to 0 < w < 1 to show the area of interest). To convert to years, invert wpkY1 and multiply by 2p, which yields a fundamental period of 170.7 years.

From inspection of the PSD we form the harmonic model (note all of the radian frequencies are harmonically related to the fundamental):

(*Define the 5th order harmonic model used in curve fit*)

model=AY1*Sin[wpkY1 t+phiY1]+AY2*Sin[2*wpkY1* t+phiY2]+AY3*

Sin[3*wpkY1* t+phiY3]+AY4*Sin[4*wpkY1* t+phiY4]+AY5*

Sin[5*wpkY1* t+phiY5]];

vars= {AY1,phiY1,AY2,phiY2,AY3,phiY3,AY4,phiY4,AY5,phiY5 }

and fit the model to the original (unperiodized) data to find the unknown amplitudes, AYx, and phases, phiYx.

fitParms1=FindFit[yearly,model,vars,t]

fit1=Table[model/.fitParms1,{t,0,112}];

residualY1= yearly- fit1;{AY1→-0.328464,phiY1→1.44861,AY2→-

0.194251,phiY2→3.03246,AY3→0.132514,phiY3→2.26587,AY4→0.0624932,

phiY4→-3.42662,AY5→-0.0116186,phiY5→-1.36245,AY8→0.0563983,phiY8→

1.97142,wpkY1→0.036811}

The fit is shown in Figure 5 and the residual error in Figure 6.

clip_image010

Figure 5 – Harmonic model fit to annual data

clip_image012 clip_image014

Figure 6 – Residual Error Figure 7 – PSD of the residual error

The residual is nearly white, as evidenced by Figure 7, justifying use of the Hodric-Prescott filter on the decimated data. This filter is designed to separate cyclical, non-stationary components from data. Figure 8 shows an excellent fit with a smoothing factor of 15.

clip_image016

Figure 8 – Model vs. HP Filtered data (smoothing factor=3)

Stochastic Analysis

The objection that this is simple curve fitting can be rightly raised. After all, harmonic decomposition is a highly constrained form of Fourier analysis, which is itself a curve fitting exercise that yields the harmonic coefficients (where the fundamental is the sample rate) which recreate the sequence exactly in the sample domain. That does not mean however, that any periodicity found by Fourier analysis (or by implication, harmonic decomposition) are not present in the record. Nor, as will be shown below, is it true that harmonic decomposition on an arbitrary sequence would be expected to yield the goodness-of-fit achieved here.

The 113 sample record examined above is not long enough to attribute statistical significance to the fundamental 170.7 year period, although others have found significance in the 57-year (here 56.9 year) third harmonic. We can however, estimate the probability that the results are a statistical fluke.

To do so, we use the data record to estimate an AR process.

procY=ARProcess[{a1,a2,a3,a4,a5},v];

procParamsY = FindProcessParameters[yearlyTD["States"],procY]

estProcY= procY /. procParamsY

WeakStationarity[estProcY]

{a1→0.713,a2→0.0647,a3→0.0629,a4→0.181,a5→0.0845,v→0.0124391}

As can be seen in Figure 9 below, the process estimate yields a reasonable match to observed power spectral density and covariance function.

clip_image018 clip_image020

Figure 9 – PSD of estimated AR process (red) vs. data Figure 9b – Correlation function (model in blue)

clip_image022 clip_image024

Figure 10 – 500 trial spaghetti plot Figure 10b – Three paths chosen at random

As shown in 10b, the AR process produces sequences which in general character match the temperature record. Next we perform a fifth-order harmonic decomposition on all 500 paths, taking the variance of the residual as a goodness-of-fit metric. Of the 500 trials, harmonic decomposition failed to converge 74 times, meaning that no periodicity could be found which reduced the variance of the residual (this alone disproves the hypothesis that any arbitrary AR sequences can be decomposed). To these failed trials we assigned the variance of the original sequence. The scattergram of results are plotted in Figure 11 along with a dashed line representing the variance of the model residual found above.

clip_image026

Figure 11 – Variance of residual; fifth order HC (Harmonic Coefficients), residual 5HC on climate record shown in red

We see that the fifth-order fit to the actual climate record produces an unusually good result. Of the 500 trials, 99.4% resulted in residual variance exceeding that achieved on the actual temperature data. Only 1.8% of the trials came within 10% and 5.2% within 20%. We can estimate the probability of achieving this result by chance by examining the cumulative distribution of the results plotted in Figure 12.

clip_image028

Figure 12 – CDF (Cumulative Distribution Function) of trial variances

The CDF estimates the probability of achieving these results by chance at ~8.1%.

Forecast

Even if we accept the premise of statistical significance, without knowledge of the underlying mechanism producing the periodicity, forecasting becomes a suspect endeavor. If for example, the harmonics are being generated by a stable non-linear climatic response to some celestial cycle, we would expect the model to have skill in forecasting future climate trends. On the other hand, if the periodicities are internally generated by the climate itself (e.g. feedback involving transport delays), we would expect both the fundamental frequency and importantly, the phase of the harmonics to evolve with time making accurate forecasts impossible.

Nevertheless, having come thus far, who could resist a peek into the future?

We assume the periodicity is externally forced and the climate response remains constant. We are interested in modeling the remaining variance so we fit a stochastic model to the residual. Empirically, we found that again, a 5th order AR (autoregressive) process matches the residual well.

tDataY=TemporalData[residualY1-Mean[residualY1],Automatic];

yearTD=TemporalData[residualY1,{ DateRange[{1900},{2012},"Year"]}]

procY=ARProcess[{a1,a2,a3,a4,a5},v];

procParamsY = FindProcessParameters[yearTD["States"],procY]

estProcY= procY /. procParamsY

WeakStationarity[estProcY]

clip_image030 clip_image032

A 100-path, 100-year run combining the paths of the AR model with the harmonic model derived above is shown in Figure 13.

clip_image034

Figure 13 – Projected global mean temperature anomaly (centered 1950-1965 mean)

clip_image036

Figure 14 – Survivability at 10 (Purple), 25 (Orange), 50 (Red), 75 (Blue) and 100 (Green) years

The survivability plots predict no significant probability of a positive anomaly exceeding .5°C between 2023 and 2113.

Discussion

With a roughly one-in-twelve chance that the model obtained above is the manifestation of a statistical fluke, these results are not definitive. They do however show that a reasonable hypothesis for the observed record can be established independent of any significant contribution from greenhouse gases or other anthropogenic effects.

The climate data they don't want you to find — free, to your inbox.
Join readers who get 5–8 new articles daily — no algorithms, no shadow bans.
0 0 votes
Article Rating
313 Comments
DesertYote
September 11, 2013 8:47 am

Yay Agilent! (E44xx) I was wondering when someone would apply modern Information Theory or Signal Processing techniques to this problem, let alone modern statistical methods. I have yet to see any study doing bootstrap 🙁

DirkH
September 11, 2013 8:58 am

I like it a lot. A quantification of the likelihood that the model has predictive skill. That’s more than the IPCC gives you.

DirkH
September 11, 2013 9:01 am

Jim G says:
September 11, 2013 at 8:13 am
“Or any more. Statistical masturbation, at its best, in both cases. Should have left it in the “boxcar” and shipped it out.”
That’s completely unfair. Jeff has argued convincingly why there is reason to believe that the periodicity is real.

Pippen Kool
September 11, 2013 9:10 am

Can you “predict” the last 25 years of your data set analyzing just the years before ?
I mean, using your same methods , can you predict the warming in the last few decades?

Slartibartfast
September 11, 2013 9:11 am

Get off that tangent.

8/

September 11, 2013 9:15 am

In the last year in a series of posts at
http://climatesense-norpag.blogspot.com
I have laid out a method of climate forecasting based on recognising quasi repetitive- quasi cyclic patterns in the temperature and other relevant climate-driver data time series.Pattersons post illustrates a useful approach to recognising possible patterns.Interestingly his forecast cooling until 2040 followed by warming until about 2070 is generally similar to my projections..However I forecast further cooling beyond 2070 based on incorporating the 1000 year solar millennial cycle into the forecast.
We should not expect mathematical precision in this type of forecast because of the changing resonances between the quasi cyclic rate processes which integtrate into the temperature climate metric.It would however be very useful if Patterson could extend his analysis to the 2000 year proxy record of Christiansen and Ljundqvist 2012 which is the basis for most of my projections for the next several hundred years .The following simple relationships should be noticed.Pattersons 171 year cycle = 3x PDO+/- 3X 171 =513, 6 x 171 = 1026. These intervals relate to the Roman Climate Optimum ,the Dark Ages ,the MWP,the LIA and the Modern Peak.The shape of the curve is not a simple sine curve. The Christiansen record of the last 1000 years is the most obvious go-by for the future trends
.Here is part of my original post on this subject at the above link on 1/22/13
“e) The real controlling factor for the immediate future is where the present day stands relative to the approximate 1000 year solar cycle peak
f) Having looked at numerous reconstructions I suggest that that of Christiansen and Ljungqvist 2012 is the most likely to be closest to reality. : http://www.clim-past.net/8/765/2012/cp-8-765-2012.pdf
Discussion.
The key working hypothesis is that the solar cycle from 1000- 2000 may repeat and we may see a pattern of temperatures from 2000 – 3000 which is similar to that from 1000 – 2000. Fig.5 from the Christiansen paper is shown above.The solid lines are the 50 year moving averages and the dashed red lines are the upper and lower quantiles.
Inspection of Figure 5 – both the moving average and the annual data suggests the following.
1) The millennial peak is sharp – perhaps 18 years +/-. We have now had 16 years since 1997 with no net warming – and so might expect a sharp drop in a year or two – 2014/16 -with a net cooling by 2035 of about 0.35.Within that time frame however there could well be some exceptional years with NH temperatures +/- 0.25 degrees colder than that.
2) The cooling gradient might be fairly steep down to the Oort minimum equivalent which would occur about 2100. (about 1100 on Fig 5) with a total cooling in 2100 from the present estimated at about 1.2 +/-
3) From 2100 on through the Wolf and Sporer minima equivalents with intervening highs to the Maunder Minimum equivalent which could occur from about 2600 – 2700 a further net cooling of about 0.7 degrees could occur for a total drop of 1.9 +/- degrees
4)The time frame for the significant cooling in 2014 – 16 is strengthened by recent developments already seen in solar activity. With a time lag of about 12 years between the solar driver proxy and climate -see:
http://adsabs.harvard.edu/full/2005ESASP.560…19U we should see the effects of the sharp drop in the Ap Index which took place in 2004/5 in 2016-17. This estimate is quite independent from the estimate made from Fig5.
Conclusions1) It seems reasonably probable – say 60/40 that the NH will cool by about .35 degrees by 2035.
2) We should be able to check the accuracy of this forecast by 2018 -20.
3)The forecast of a 1.2 degree drop by 2100 is little more than a mildly interesting suggestion at this time.
4)The idea of a Maunder Minimum equivalent at 2600 – 2700 is highly speculative.
5)Contrary to the forecasts made here, the Livingston and Penn solar data are suggesting a possible Maunder type Minimum in the near future.Given our ignorance of solar physics this is entirely possible. In this case a much more rapid cooling would occur with very serious consequences to the global food supply and the world economy.
6) Global cooling will take place concurrently with that of the NH but because of the great extent of the southern oceans the global cooling will be significantly less – maybe +/- 50 % and there will also be considerable regional variability. in both hemispheres.
7) There is no reason to expect damaging global warming.Cooling is more likely .To prepare for it, all ethanol and biofuel subsidies and mandates should be abolished.Renewable energy and electric car subsidies are economically wasteful and accomplish nothing.There is no reason to control CO2 emissions, indeed some extra CO2, while having little effect on temperature, might aid farm productivity . 25% of the increased crop yields in the 20th century was due to the CO2 increase

lemiere jacques
September 11, 2013 9:16 am

the problem is you can use this method with so many curves…
so take any kind of data. price of wheat oranges or .., well anything..cut the last part of the data do the same analysis and..see if it can “foresee ” anything…i am pretty sure no…
i don’t buy that..because you don’t have any idea if the harmonics are related to something ,corespondind to a oscillating phenomena ,or just …a way to fit the curve.
May be this method can be interesting to find out if some frequancy you find smell “physical”

Jim G
September 11, 2013 9:21 am

DirkH says:
September 11, 2013 at 9:01 am
Jim G says:
September 11, 2013 at 8:13 am
“Or any more. Statistical masturbation, at its best, in both cases. Should have left it in the “boxcar” and shipped it out.”
“That’s completely unfair. Jeff has argued convincingly why there is reason to believe that the periodicity is real.”
Define “unfair”. Extreme data pre-massage is evident to the extent of data torture. We cannot be skeptical of warmist models and not be skeptical of this. Time will tell if it predicts well as we know that the AGW models do not.

JerryC
September 11, 2013 9:34 am

So if we believe the modeling at all, then this is telling us that from 2020 – 2075 we can expect below average temperatures on average. How exactly are they planning to explain that away?

Steven Devijver
September 11, 2013 9:35 am

@jeffery, two questions:
1/ why not apply these techniques in hindcast mode? That would be really interesting.
2/ would it be possible to provide this code in a stand-alone form so that others could play with it. I’m a developer, I would be happy to help out.

September 11, 2013 9:39 am

Patterson

The residual _is_ stationary. That’s the point. We can certainly drive an AR process with stationary (and in this case also nearly Gaussian) noise + sine waves and get a valid projection.

Well, ideally, the residuals should be pure white noise, with mean=0
Overall I really like your experiment. Well done. A good teaching moment for us all!
But I have used similar ARIMA-based estimators myself to predict these kinds of time-series (stock-market) and managed to find a lot of “useless” predictability (i.e. in the linear sense). It always turned out that the event sequences that I couldn’t predict (the “innovations”) were mostly the same events that would have made me a rich man if my predictor worked the way I hoped it would.
To make these models perform better you need to know a lot more about the hidden processes which generate the observable outputs.
😐

MikeN
September 11, 2013 9:42 am

170 year period data concluded from a time period of less than a full cycle?
All that means is that the increase in temperatures is then turned into a decrease on the other side of the wave.

September 11, 2013 9:45 am

It has been said that if it is good science, it does not require a statistical massage. I heartily endorse that sentiment.

September 11, 2013 9:46 am

People are confused about “decimation” — think down-sampling for an analogy.
See the documentation in GNU Octave and Matlab for an explanation of decimation… Here is one link.
http://octave.sourceforge.net/signal/function/decimate.html
It might make more sense than — or less perhaps if you don’t follow this type of stuff..

RC Saumarez
September 11, 2013 9:49 am

Jim G and others.
I can take broadband noise and then pass it through a low-pass system. Provided the system has persistance, I will get a signal containing trends and this may appear quasi-periodic. I can fit a Fourier series to this signal and calculate an “input” that will create the output for a short length of recorded signal.
However, the input is noise and therefore the small segment of “input” is simply one snapshot of its underlying process. It is inevitable that if you do this you will find want appear to be cycles. If you repeat the “experiment”, you will find different periodicities in each repeat.
What one cannot do is assume that this snapshot will repeat – it is simply one sample of a noise record.
If you look at figure 13, it is the 21st century prediction and is simply a recapitulation of the 20th century with a basic triangular waveform so that the 20th and 21st centuries connect.
The logic underlying this post is, to my mind, hopelessly flawed.

Greg Goodman
September 11, 2013 9:51 am

“Digital Signal Processing analysis of global temperature data time series suggests global cooling ahead”
Let’s see a plot of the next 170 years to see what this method suggests.
My guess is that it’s going to ramp down to about -0.4 and rise steadily upwards with a couple of bumps in the middle.
I promise I have not try plotting it, just guessing…..

September 11, 2013 9:51 am

crosspatch says:
September 11, 2013 at 8:26 am
The problem with this sort of analysis, while interesting, is that it can only show what HAS happened and not what WILL happen. Would an analysis of the opening of the 1812 Overture predict the finale? About the only thing I am prepared to say with any certainty is that the next 1000 years is more likely to be cooler than the last 1000 years than it is likely to be warmer.

Sure it could! And in 1812’s case (as with many [most?] other pieces of classical music) the intro comes back in the finale in the tonic key. This is the entire point of prediction, and human nature, to find (and create) patterns/expectation/reconciliation….then do it again.

September 11, 2013 9:54 am

If I understand correctly what he is doing the “BoxCar and the Decimation” are more about creating yearly “buckets” (or average/smoothed values) so that yearly values can be determined and examined. The same techniques are in R and widely used there for the same purposes — sometimes named differently.
Probably better explained in “The little book of Time Series” than I can do
http://a-little-book-of-r-for-time-series.readthedocs.org/en/latest/
hth

September 11, 2013 9:57 am

I don’t have time this week. If I remember this discussion next week I will run it through a somewhat different set of processes and see If I get the same periodic repeats.

Greg Goodman
September 11, 2013 9:58 am

oops, hadn’t seen the caption on fig 13, I must have started skimming well before I got that far. Guess what if the previous 170 years isn’t strangely similar.
As RC Saumarez says, this is nothing more than the extrapolation fallacy.

Jeff Patterson
September 11, 2013 10:03 am

I appreciate the response posted here. rather than address each in turn, let me make some general comments on the method.
First, the concerns about aliasing are IMHO unwarranted. The monthly data record is highly low-passed, being averaged in both space and time. Also, presumably the climate process, containing as it does many integrative elements, imposes its own filtering of high frequency input variances. The power spectral density of the unaltered Hadcrut data shows the high frequency floor to be down some 60dB (1/1000) from the spectral peaks we are trying to extract. If one understands the aliasing process in the time domain, a moment’s reflection should reveal the mathematically provable theorem that the amplitude of an alias cannot exceed the amplitude of it source. Hence aliasing effects here are insignificant.
On filtering: In harmonic decomposition, filtering must be avoided like the plague because it alters the amplitudes and phases of the periodicity we’re attempting to extract. Loss of spectral resolution is also an issue, especially when trying to extract cycles whose periods exceed the record length. However, the monthly data has a strong annual component that must be removed prior to decomposition because the underlying assumption of the method is that all periodicities are harmonically related. Fortunely, if you know the period of the offending cycle there is a solution, Box care filtering as described above. This filter is ideal in that it places its transfer function zeros precisely on top of the spectral peak being removed. It is in essence a high-Q notch filter which has the least impact on the remaining spectrum. It is true that this filter has high sidelobes but in this case that’s fine because the power spectral density at the lobe peaks is extremely low as evidenced above.
On record periodization: First note this is only a test to verify that observed spectral peaking is not due to the discontinuity at the record end points- the harmonic decomposition is performed on the original, unperiodized data).
Fourier analysis (which is used here only to perform the PSD estimate) gets around the required infinite length record by simple assuming the record repeats ad infinitum. If the endpoints don’t line up (withing the average step-size of the sample), the discontinuity created by stitching the records together (that’s not how an FFT works but in principle the effect is the same) creates a periodic error that shows up as spectral peaks. We can test for this effect by subtracting a line that connects the data record endpoints and dropping the last point from the result to avoid repeating the zero-error point at the ends. The result is a record which has no discontinuity distinguishable from the noise present in the data. If the PSD peaks remain after this process, it indicates the spectral peaks are not record-length artifacts.
One reader asked “why use the eight harmonic?”. Because there are more cycles present of the eight harmonic and so it provides a better estimate of the fundamental.
Fig. 10 spaghetti graph.
What was allowed to vary between trials? What stayed the same?
Figure 10 is output of the AR process used to create “climate-like”. Each path is a 113 point sample of the process for a new randomized state and random, white, input function of the variance derived by the AR process estimation function.

September 11, 2013 10:15 am

Jeff Patterson
Best Regards and thanks for your post. I repeat the request made in my post at 9:15 Could you possibly find the time to do a similar analysis of the 2000 year proxy record referred to in that comment.I think all the basic data is available on the web It might be a very illuminating exercise.

James at 48
September 11, 2013 10:19 am

If we started to fairly and objectively consider ringing, overshoot, beat frequency and other of these sorts of behaviors with respect to the global climate system, some interesting and perhaps frightening possibilities might start to become more apparent and would open the door for a whole new tone in the conversation.

Jeff Patterson
September 11, 2013 10:25 am

I forget to add a comment on forecasting:
In retrospect I wish I had left this section out, even though I thought I had presented the appropriate caveats from the start. Unsurprisingly, readers naturally gravitate to projections and by doing so miss the point of the exercise which was not to try to predict the future but rather to show that a plausible alternative hypothesis (actually two, harmonic forcing and/or resonances in the climate system) provides a good fit to the observed data and that such a good fit is unusual. I have neither the time or expertise to complete the exercise – namely providing a physical explanation for the hypothesis and empirically testing the result.
Some general comments about modeling:
Information theory tells us nothing can be learned from a calculation. Knowledge is created by surprise (entropy) and no calculation should surprise us. All modeling, including GCMs, no matter how complex are simply calculations. The results are completely determined by the input and initial state. They are thus only useful in the hypothesis-forming portion of the scientific method (something the climate modelers seem to have forgotten). Once formed, the hypothesis must be verified by empirical observation.
Again, thanks to all for the useful comments. I’ll try and follow up with the suggestion to run the method on a subset of the data and post the results. My thanks to Mr. Watts for providing this site and to all for the interesting conversation.

Jonathan Abbott
September 11, 2013 10:28 am

I think DayHay makes the most important point.

1 3 4 5 6 7 13