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.

0 0 votes
Article Rating

Discover more from Watts Up With That?

Subscribe to get the latest posts sent to your email.

313 Comments
Inline Feedbacks
View all comments
September 11, 2013 6:07 am

To Geoff Sherrington:
It’s in plain English if you already understand the technical terminology.
*resigns self to spending a few hours reading up on signal processing techniques*
To Ian E:
You have it backwards. This is observing the wriggling trunk and trying to work out what the elephant looks like.
To all WUWTers:
I have seen doubts expressed in some places on the applicability of techniques developed for electronic signal processing when applied to the climate. Straw man? Valid?

September 11, 2013 6:08 am


>Nick Stokes said How can you establish a 170 year periodicity from 113 years of data?
Short answer: you can’t. You need at least two samples per period from a sinewave, in order to reconstruct it from a sampled set of measurements. (In the same sense that you need two sample from a line to reconstruct it) http://en.wikipedia.org/wiki/Shannon%E2%80%93Hartley_theorem
The sequence of odd harmonics in the processed waves suggests that windowing artifacts remain, as the convolution of sine waves (which are always infinite in extend from an abstract mathematical viewpoint) multiplied by a sampling window (“square wave”) would suggest.http://en.wikipedia.org/wiki/Convolution_theorem
In AR modeling, the predictive modeling results depend on the time series being stationary, i.e. constant mean and variance. But historical temperature records are not stationary, and in fact it is the non-stationary part of the record that you are trying to model. So results of predicting values in the future based on values in the past will generally not be credible.
http://en.wikipedia.org/wiki/Wold's_theorem
😐

Jeff Patterson
September 11, 2013 6:11 am

Nick Stokes says: How can you establish a 170 year periodicity from 113 years of data?
In theory, you need just two samples per period. Just as two points define a line uniquely, there is only one way to draw a sine wave between two points as long as you know the points are separated in time by no more than 1/2 the shortest period (or equivalently twice the highest frequency) sine wave present in the data. if the condition is not met, the high frequency components masquerade as low frequency components in a process referred to as aliasing. That is why we remove the high frequency components by filtering before decimating (re-sampling) the data.

Pamela Gray
September 11, 2013 6:12 am

Ric, don’t ya love “back in my day” times? Want to do an Anova, or Coanova? Take an MSDOS computer with a blank screen and program it (I programmed mine to say good morning to me every time I turned it on). Or grab a calculator, the clunky dark cream colored kind with sharp square corners, and do it that way. Hell, just use a pencil and paper. I remember getting all twitterpated when I got a little Mac SE with Statview installed on it. I thought it was Christmas every day.

richardscourtney
September 11, 2013 6:14 am

Friends:
I second the request of Mike Lewis at September 11, 2013 at 5:41 am; viz.

If the same method was applied to data just through 1980, would it accurately predict the warming in the 80′s and 90′s?

I note the strong caveat in the above essay concerning the possible inappropriateness of extrapolation of the obtained curve. However, if it is assumed that the curve fit does capture the form of the temperature changes, then failure of the method to predict the temperature curve after 1980 would imply the temperature changes are internal variability of the climate system and are not externally forced.
Richard

Jeff Patterson
September 11, 2013 6:17 am

IanE says:
September 11, 2013 at 5:52 am
‘With four parameters I can fit an elephant, and with five I can make him wiggle his trunk.’ John von Neumann.
As the stochastic fitting exercise showed, JvN could only only do this .6% of the time if all he had to work with was harmonically related sine waves.

zentgraf2
September 11, 2013 6:18 am

A good intellectual exercise and perhaps more valid than the 73 worldwide GCM’s. If by 2040 the forecast pans out, we can save a lot of tax money. No need for an understanding of climate physics or chemistry. Just have a small group of DSP mathematics folks hidden away in a skunk works.
I am joking ,of course. Nevertheless I yearn for a creditable effort that would create severe competition for the circular thinking climate science community to reorient their thinking towards a comprehensive understanding of the natural forces in climate.

September 11, 2013 6:27 am

@Me

Short answer: you can’t. You need at least two samples per period from a sinewave, in order to reconstruct it from a sampled set of measurements. (In the same sense that you need two sample from a line to reconstruct it) http://en.wikipedia.org/wiki/Shannon%E2%80%93Hartley_theorem

As soon as I hit the “Post Comment” button I realized that my reponse was wrong because the 170-year wave in question is indeed “over-sampled”, not under-sampled.
But the reason I was so sure this couldn’t be done is that I’ve tried to reconstruct these kinds of “over-sampled partial waves” myself and have never been successful at it. But perhaps I going at it in the wrong way.
😐

kb
September 11, 2013 6:29 am

“extracting spurious signals from noisy data”
Hopefully this means “extracting meaningful signals” or “removing spurious signals”…

Editor
September 11, 2013 6:29 am

John Day says:
September 11, 2013 at 6:08 am


>Nick Stokes said How can you establish a 170 year periodicity from 113 years of data?
Short answer: you can’t. You need at least two samples per period from a sinewave, in order to reconstruct it from a sampled set of measurements. (In the same sense that you need two sample from a line to reconstruct it) http://en.wikipedia.org/wiki/Shannon%E2%80%93Hartley_theorem

This all applies to periodic sampling of a periodic signal. That’s not what we have here, we have many samples of part of a single cycle.
Had this been similar to Nyquist sampling, there would be either one or two samples of the entire 113 years. Instead, the sampling rate is far higher than that and there are hundreds of samples to work with.

Greg Goodman
September 11, 2013 6:31 am

I’ll parse this in more detail later but initial reactions:
1. HadCruft is a land + ocean amalgam. Rate of change of land is twice that of SST, how will this affect freq decomposition.
2. “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 ” Running mean filter _introduce_ spurious signals. http://climategrog.wordpress.com/2013/05/19/triple-running-mean-filters/
3. A 12m filter it NOT an appropriate anti-aliasing filter for 12m decimation
4. HadCruftX is probably aliased already even in the monthly data series further filtering will not fix it once it’s done: http://judithcurry.com/2011/10/18/does-the-aliasing-beast-feed-the-uncertainty-monster/
5. Series of add harmonics suggests a triangular waveform likely due to non stationary data. The first and last point trick is new to me and I can’t see this being any guarantee of the conditions needed for FFT. Linear detrending of whole dataset or taking first diff would be more usual.
It’s an interesting idea and I don’t want to totally dismiss the result without a closer look, but this seems like a fair list of processing errors so far.

RC Saumarez
September 11, 2013 6:36 am

This appears to be an example of the well known fallacy of Fourier Extrapolation.

Luther Wu
September 11, 2013 6:39 am

Thank you, Mr. Patterson for your fine work.
Fortunately, you’ll need spend no effort trying to construct a compensatory microcode.
Bit of a busman’s holiday, is that it?

G. Karst
September 11, 2013 6:42 am

If the above statistical analysis is valid ( as it appears)… why not submit it for publication and peer review. Surely the gatekeeping practices, of climate research, have somewhat abated since the climategate revelations. Or is this, just a pre-review before submission. GK

Jeff Patterson
September 11, 2013 6:46 am

Here’s a hopefully approachable summary:
A time record can be represented in the time domain (like the Hadcrut plot) or the frequency domain (like the power spectral density plots in the post). The frequency domain looks like an old fashioned radio dial, with the high frequencies to the right and the low frequencies to the left. A radio station is assigned a frequency, unique from all others in the area, so that its signal only appears one place on the dial and is not interfered with by other stations. If we were do do a spectral analysis (i.e. a PSD plot) of the AM band, we would see a peak at each frequency where a station is broadcasting and a valley where no station was present.
We can transform from one domain to the other without distortion via the Fourier Transform. It uses every possible frequency which are all harmonically related to the frequency at which the data was sampled (i.e. monthly in the case of Hadcrut3) to exactly represent the data in the frequency domain. To reconstruct the 113 samples in my decimated record exactly would require 226 variables (113 frequencies and their phases). We show that we excellent results using just 5 harmonically related frequencies where instead of using the lowest frequency of the FFT, we use the frequency with the highest peak as the fundamental. Actually, the 5th harmonic doesn’t add much and most of the fit is captured with just 3 harmonics.
The great thing about sine waves is that they repeat which means if they are really present, the model can be used to predict the future trend. The bad thing is that it is easy to get fooled if the data is noisy. The second half of the paper examines the likelihood that we were fooled. By generating some random sequences from a process designed to produce data that in character (peak-to-peak changes, intermediate slopes etc.) matches the temperature record, we can perform the same harmonic decomposition we did to the temperature data and see how many times the fit is as good as we achieved with the actual data. In this case, only .6% of the 500 trials were as good, which places the probability that we were fooled at about 8%

September 11, 2013 6:47 am

Hmmm, based on the data, we must embrace the possibility that increasing the concentration of atmospheric CO2 includes cooling as a possible outcome. Oh boy, who would have guessed?

Joe Chang
September 11, 2013 6:49 am

Lets assume that the periodicities are fixed as opposed to variable. Given the 113 years of data, the longest periodicity extracted was 170 years. Lets us also assume that the shorter periodicities have been correctly extracted. So now, instead of projecting forward to take a stab at the future, how about projecting backward, past the LIA, MWP, RWP. Then with an estimate as to the magnitude and time of these events without concern on the decade scale wiggles, guess what the longer periodicities are? Sure this is torturing data …

Jeff Patterson
September 11, 2013 6:49 am

Greg Goodman says:
September 11, 2013 at 6:31 am
… Running mean filter _introduce_ spurious signals.
Exactly why I used box care (i.e. non overlapping) filtering, not running mean.

pochas
September 11, 2013 6:53 am

That fundamental of 170.7 years speaks strongly of the Jose cycle. If Patterson knew about it (I suspect he did) there is some danger of confirmation bias. I have the same bias. Fortunately the projected cooling is modest. I hope it stays that way. As for the usual comment that this is a curve-fitting exercise, in a world of junk “physics-based” math models, faulty theory, manipulated data and agenda-driven science, it’s really all we have.

September 11, 2013 6:54 am

The 170 year cycle length is suspiciously close to the approximate cycle length of the solar system barycenter. the 170 year cycle continues to show up in the works of many different authors. as does the 55-60 year harmonic of 170, which closely approximates the observed length of the PDO.
Based on what we know about tidal prediction, DSP is likely to be at least as accurate as the global climate models in predicting future climate, and quite likely a whole lot more accurate.
if one tries to model the ocean tides using first principles, as is done with the climate models, the results are garbage. earth’s tides would be miniscule based on gravitational attraction alone.
the earth’s tides are predicted to a high degree of accuracy by the motion of the celestial bodies. this is because the earth’s oceans have over thousands and millions of years, come to oscillate in phase with the motion of the celestial bodies.
it is completely likely that s that climate oscillates similar to the tides, only on a longer timescale, as a result of a small but regular harmonic forcings based on the motion of the celestial bodies over thousands and millions of years.
this suggests that we can predict climate with some hope of accuracy using methods similar to what the author has proposed here today.

Greg Goodman
September 11, 2013 6:54 am

This is similar in some ways to what I did 18 ago on Judith Curry’s site where I fitted just three , non harmonic frequencies to examine how Met Office processing had changed the data. (NB this was not intended as a prediction tool, just an evaluation of changes to the existing data).
: http://judithcurry.com/2012/03/15/on-the-adjustments-to-the-hadsst3-data-set-2/
see middle panels in figures 9 and 10.
It is interesting that I found hadSST (sea temp only) had _removed_ a 160 year periodicity. Maybe the combined use of Had + CRUT here, puts it back ;?

Rod Everson
September 11, 2013 6:57 am

The relevant line from a very long article: “Even if we accept the premise of statistical significance, without knowledge of the underlying mechanism producing the periodicity, forecasting becomes a suspect endeavor. ”
Let’s not duplicate the errors of our adversaries and assume we know more than we know.

Jeff Patterson
September 11, 2013 6:58 am

RC Saumarez says: This appears to be an example of the well known fallacy of Fourier Extrapolation.
HArmonic decomposition is much more constrained than fourier analysis but subject to some of same concerns about seeing periodicity that isn’t there. The last half of the paper address this issue directly.

Greg Goodman
September 11, 2013 6:59 am

JP “Exactly why I used box care (i.e. non overlapping) filtering, not running mean.”
Could you explain what you mean by “box car” in more engineering/mathematical language , maybe I misunderstood what you mean by railway analogy. 😉

rabbit
September 11, 2013 6:59 am

Jeffrey Patterson:
Have you considered Singular Spectrum Analysis (SSA) to extract the dominant modes of the temperature graph? This does not suffer from the windowing effects that a DFT suffers from, nor is it limited to a prescribed set of basis functions.