ENSO forecast based on tidal forcing with an Artificial Neural Network
Investigation submitted by Per Strandberg
Here on this page, you are going to find evidence that tidal forcing is one of the most important, if not the most important driver for ENSO variations. That tidal forcing could be the main explanation for ENSO variations was something I stumbled upon when I examined possible ENSO drivers.
After the previous results which I got when I was using an Artificial Neural Network ANN and where I did an analysis of the correlations between the global mean temperature and possible forcing drivers, which can be viewed here, I turned my attention to the ENSO index by looking into the Multivariate ENSO Index (MEI).
One thing I found when I analyzed the result from correlations to ENSO was that there is a strong correlation between variations in Earth’s rotations both to the global mean temperature and to the ENSO index. What we are talking about here are small variations in Earth’s rotations, which are in the order of milliseconds.
One other factor with correlation to the ENSO is of course SOI, but I also found correlations to SST, PDO, and the Kp and Ap indexes.
From this, I concluded that either it is ENSO which is driving changes in Earth’s rotation or it changes in Earth’s rotation, which is causing variations in ENSO or more likely it is some combination of both.
Proof that ENSO and variations of Earth’s rotation are proportionally correlated to each other has been known for some time. This can be seen here.
The mechanisms which tie ENSO and variations in Earth’s rotation together are caused by sea current changes, changes in trade winds or by displacements of water between the equator and slightly higher latitudes. This all makes sense.
The water currents in the northern hemisphere follow a clockwise pattern, and in the Southern hemisphere they follow a counterclockwise pattern because of the Coriolis effect. The trade wind and the currents near the equator are moving to the west. However the Current closest to the equator called the equatorial counter current move to the east. Still deeper at depth down to 200 meters at the equator an ever stronger current is moving to the east.
The behavior of this current of the Equatorial Pacific is shown on this page by Bob Tisdale.
The only mechanism by which ENSO can be driven by changes in Earth’s rotation is by variations in the tidal force.
My next step was to try to include tidal forcing in my ANN.
I then got three problems, which I had to overcome.
Firstly: I had to find data over the position and distance to the Moon and to the Sun. Eventually, I found software from which I could get this data, although it gave limited information and I was only able to print out time and position when the Sun and the Moon were closest and farthest from the Earth and with the Moon I could also calculate the time and position of the new moon, the full moon and the moon nodes. The Moon nodes are the location where the Moon cross over the ecliptic plane.
Secondly: I had to find the formula for the tidal force vector and implement this into my software.
Thirdly: I had to figure out what features in the tidal forcing which could affect ENSO. I had to experiment with different configurations based in my limited and rather crude data. To do this, I had to make complicated trigonometrically calculations in order to get the right value of the tidal force. Eventually, I got good correlations between ENSO and the tidal forcing. By this time, I had figured out which features in the tidal forcing that were causing this correlation. However it was not a direct correlation with ENSO, rather it was a correlation with the derivate signal of ENSO, i.e. it was affecting the rate of change of ENSO. The correlation to the change of rate in Earth’s rotation, on the other hand, is direct. This means that tidal forcing is causing the rate of Earth’s rotation to either speed up or speed down. The rate of rotation is then responsible for changes of the ENSO index. One reason was that it was difficult to identify, which features, which cause correlations. This was because each tidal forcing point I use the sum of monthly calculations. The size of the tidal forcing changes each and every day and how to summarize this data the right way into useful functions, which can be used to construct values that could create good correlations were difficult.
Of course, the tidal force is not the only factor which drives ENSO, but it is the most influential factor.
To test if that would be the case I ran my network with the right tidal forcing data. I also included feedback loops back in the network from the output ENSO values to some of the input nodes. After some testing and individual adjustments of the internal components in the artificial network, I got good results. Following on my earlier experiment of the ANN on the mean global temperature I trained the ANN from late 1978 up to the end of 2004. I used the time from 2005 up to the late 2011 for test the calculations, in order to find the minimal error function.
 |
This is the result I got. The exciting thing with this result is that it is possible to make forecasts for much longer times into the future. Today’s predictions use computer models and are only able to make credible predictions 4 to 5 months into the future. While in my case, using my ANN calculations based on tidal forcing can be made for forecasts for an almost unlimited time because the Moon and Sun’s positions into the future are known in advance. Although, I have to stress that with the predication so far it is not possible to get the last figure right. Currently, it is only possible to make an estimate with a relative high likelihood at any date if ENSO are going to be positive, negative or neutral. However, as can be seen here the predictions are not always correct. The main large El Niño events of 1982 and 1998 can clearly be seen, but the large magnitude of these events can not be predicted.
 |
I later made a ENSO forecast from late 2011 up to 2020. I cannot show the result here because of proprietary reasons. This picture shows the test period and some of the forecast which ends in early 2013.
Note, however that the calculations from this graph, the ENSO index uses ENSO feedback values which all are from estimated ones. Those are not the real ENSO values.
Now Look at: the previous graph with the whole time span from 1979 up to 2011! On this graph, look at the beginning at the 3 first years from the start of 1979. These 3 first years have all exceptional good correlation to the real ENSO values. The difference with the start values in this case is that I use real ENSO values for the feedback values in the network calculations which are going into the calculations with values before the graph begins. This is because in my ANN, I use for every calculation point values which goes 3 years back in time and I must use real input values for values before my first calculated value.
If I would make a forecast for the next years using current real ENSO values from 3 years back and up to the current date, my forecast would be greatly improved and would be much better than forecast made with current computer models. There were 2 important events that happened the years just after 1979. The first was the eruption of El Chichon in 1982 in Mexico. The second was the unusually strong el Niño event between 1982-1983.
My calculated values after 1982 of ENSO tend to come out of phase after and around 1982, and the ANN seems not to be able to handle strong El Niño’s very well. After some years, the estimated ENSO value deteriorate somewhat mainly because of errors in the feedback caused by the inertia in Earth’s rotation. In contrast to computer model forecasts, I don’t use data from the Tropical Atmosphere Ocean TAO network which is a NOAA measurement network of buoys in the tropical Pacific that deliver real time data which feeds these ENSO models with real time data.
 |
Here is a result from the same program, but as input it uses variations in the Earth’s rotation instead of tidal forcing. As you can see, the correlation to the Earth’s rotation makes the result much better. However in contrast to tidal forcing, future changes in Earth’s rotation is unknown.
 |
Here is another graph from the same program with feedback but this time the input signal is only from SOI Southern Oscillation Index. As expected SOI is closely related to ENSO.
 |
Here is a repeat with the same program, but this time with a combination of tidal forcing, changes in Earth’s rotation, SOI, Kp and the Ap indexes. As you can see, this result is similar to that of the previous with only SOI.
The next step I plan to take is to use the ANN with real time data and make more accurate ENSO predictions for the next 3 to 4 years in to the future. I also want to test with real ENSO input data for several time periods in order to evaluate statistically how the good predictions can be based on real time ENSO feedback input data for the beginning.
After that, I want to improve on my result by using more precise and accurate tidal calculations. I have found a program from which I can make precise calculations of the Moon and the Sun on a daily basis. Other factors I plan to look into are the mechanism of the Kelvin wave, Walker circulation and MJO which all should influence ENSO to some degree.
So far I have only looked at ENSO. I can easily switch to SOI, NOI and changes in Earth’s rotation and use that as an output for predicting ENSO with the ANN.
Conclusion from my result is that tidal forcing is as a major factor in ENSO forcing. I now have gotten new questions. Compared with other causes how important is the effect from tidal forcing? Is it possible to find an increasing effect from tidal forcing by improving the tidal data I use? Is it, for example, possible to identify tidal forcing as the cause for the strong El Niño of 1982 and 1998?
It may be possible to get better ENSO results by using predictions based on SOI, NOI, Earth’s rotations or by starting from tidal forcing only. I’ll test and see. Also, ENSO and SOI are parameters for which there exists long historical data records. By, using a longer time span for training and testing, the accuracy of predictions based on ANN should be improved.
I acknowledge that it is not easy to find correlations between tidal forcing without testing out the right feature and by using am ANN. However I do find it very strange that no scientist to my knowledge has been looking into a possible connection between tidal forcing and ENSO in any depth.
As can be seen from what the IPCC writes about ENSO predictions, they do not have a clue. The current data models that are in use can only predict with any accuracy 4 to 5 months into the future. When it comes to the ENSO drivers, these researchers think chaos theory and random noise are the mechanisms which explains the causes of ENSO changes.
However Cerveny, R. S. and J. A. Shaffer (2001) et al. in the report, The Moon and El Niño, Geophys. Res., writes about the Moon cycles and ENSO, where they find correlations between the solar cycles and ENSO.
To me at least, it seems that the solution to long range ENSO prediction has for a long time been right in front of the eyes of these researchers, but nobody has taken up the challenge to figure it out.
I see the same reason why the climate community at large has not studied tidal forcing as an explanation for ENSO variations and why none TSI solar forcing as an additional cause for climate forcing ignored. The primary reason is that they have had their education in meteorology, atmospheric physics, thermodynamics or in computer science. Most of them are specialists in a few narrow disciplines, and as such they prefer only to apply knowledge from the fields they know. They are not generalists and display strong resistance for applying knowledge from other area from which they lack knowledge. Then add to that group thinking, peer pressure and lack of funding for research in alternative causes of climate change and this explains the current one-sided situation. This is one of reason that predictions made with computer simulations are failing.
ANN are seldom used in climate science. There are some exceptions. One is research done by Dr William Hsieh from the University of British Columbia who uses this technique for ENSO predictions, but to my knowledge without using any tidal forcing. To learn more about how ANN works and how I have implemented this technique in climate investigation, Click here
As ENSO variation tends to be caused by changes in the trade winds and that changes in the trade winds could well result in changes in rotation speed, I’m not seeing much here. The trade winds flow opposite the rotation of the earth so strong trade winds would tend to slow down the rotation a smidge and weak trades allow it to speed up a little. Seems like a sort of “water is wet” conclusion to me. You can think of that water piling up in the West Pacific Warm Pool as a result of the brake being applied by the trade winds and the sloshing of that water back east during El Nino as a result of releasing that “brake”.
Excellent. Please continue with the very important plans you have. I believe you have materially advanced the case that tidal forces are a major player in earth’s climate.
An earlier paper on this subject:
Keeling and Whorf (2000): The 1,800-year oceanic tidal cycle: A possible cause
of rapid climate change
http://www.pnas.org/content/97/8/3814.full.pdf
Many misconceptions might be attributed to language barriers.
It is the thought that counts.
The author refers to a more complete exposition that is in progress. I await that, my curiosity piqued by this post. All calculational details (those trigonometric functions, how derivatives of data are calculated) need to be supplied. The test of the model is its ability to make accurate predictions of the future. Computational Neural Networks have weaknesses common to all high-dimensional multivariate time series methods; sometimes they produce good results, and sometimes they do not even produce unique results with the same data.
Willis-
In your post on Moon wind, I was wondering if you considered the possibility that the slight temperature change due to the moon was caused by the slight change in gravity experienced by the atmosphere that results in a slight modulation in the adiabatic lapse rate. I think it would tend to show up under dry conditions.
(a rewrite of my 9:54 am comment)
The word “causation” as in “Correlation does not imply causation” is missing from this article.
You seem to have assumed that changes in the rate of Earth rotation causes changes to what ENSO measures without investigating whether it could be the other way round, or whether both are influenced by something else.
(Chart: Deviation of day length, 1965-2010), shows changes that amount to no more that 2 milliseconds per day per month and average out to less than 0.5 milliseconds per day per year.
It is far more likely to me that the redistribution of 10^21 kg of ocean water at various temperatures, of which the ENSO Index is only a measure, is a cause of changes to the Earth’s moment of inertia and therefore is a cause of TINY changes to the length of day. Earth Tides caused by the sun and moon are also an element of the angular moment inertia. Couple earth tides with complicated lunar orbit eccentricity and node precession and there are many known dissonant elements affecting moment of inertia and Length of Day.
What in fact to you mean by “Tidal Forcing”? Yes, there are tidal forces, but you seemed to be selective about which to use. I had figured out which features in the tidal forcing that were causing this correlation. Tells us how each liter of ocean water can tell which feature to heed. You were quite vague about that “Tidal Force Vector” This was because each tidal forcing point I use the sum of monthly calculations. The size of the tidal forcing changes each and every day and how to summarize this data the right way into useful functions, which can be used to construct values that could create good correlations were difficult. What comes to mind is “With four parameters, I can fit an elephant….“-von Neumann.
The correlation to the change of rate in Earth’s rotation, on the other hand, is direct. This means that tidal forcing is causing the rate of Earth’s rotation to either speed up or speed down. Tidal forces do affect the moment of inertia, and therefore length of day. This is not new.
The rate of rotation is then responsible for changes of the ENSO index. That does not follow. You must do more to show causality. Even if ENSO and LOD changes are tightly coupled, they may both be driven by a common action. Can you find a way to drive ENSO via changing sun and moon’s tidal forces without first going through LOD effects?
It is such a pleasure to read the workings and comments of well grounded realists. Where else, in the blogosphere, can one find such a rich concentration and enjoy such conversation? I can’t wait to read Bob Tisdale’s valuable opinion and contribution, to these reported results. GK
In 1967, as a graduate student, I stood on the top of an active volcano (Irazu) in Costa Rico and saw the Pacific ocean to one side and the Atlantic to the other. Over the course of the summer, I spent time on both coasts. The tides on the pacific side were huge- probably 8-10+’. The tides in the Caribbean were almost unnoticeable. I wasn’t studying tides, so only took it as an observation. Just sayin. Although I agree with Willis’ observations, I like new ideas and openness, both of which I see here. Push onward please.
There is a sort of up and coming fashion these days about “tidal forces” playing some sort of major role in everything from climate to volcanism. While I do believe that tidal forces might play some role, I don’t believe it is significant. If a tidal difference sets off a volcano or earthquake, then it was right on the hairy edge of “going” without it anyway. I think the changes in rotation are an effect of changes in the trade winds which also cause changes in ENSO but I will be interested to see what else is coming on the subject.
” I later made a ENSO forecast from late 2011 up to 2020. I cannot show the result here because of proprietary reasons”
============================
If you can’t show results you should not publish. Is that so hard to understand? Really, this statement tells me all that I need to know.
Question: do the fortnightly peaks and troughs in LOD correspond to spring and neap tides, or to max/min declination between the moon and the earth’s axis? –AGF
Investigation submitted by Per Strandberg
Here on this page, you are going to find evidence that tidal forcing is one of the most important, if not the most important driver for ENSO variations. That tidal forcing could be the main explanation for ENSO variations was something I stumbled upon when I examined possible ENSO drivers.
The global temperature function is determined from i.) the solar tides and ii.) the ENSO or ONI impedance geometries.
This is evident by the fact that if one subtracts the time shifted ONI function by +0.45 years from the global temperature function, the resulting function shows mostly all of the solar tide function:
http://www.volker-doormann.org/images/oni_cleaned_rrs_temps.gif
s. also: http://www.volker-doormann.org/images/uah_2010_2013.gif
V.
I disagree that tidal influence has anything to do with ENSO. ENSO is made possible by the fact that in the Pacific both equatorial currents are blocked and as a result the warm water they carry piles up to form the Indo-Pacific Warm Pool, the warmest water on earth. This does not happen in the Atlantic because the horn of Brazil directs both equatorial currents north and there is no place for any water to pile up. When the Indo-Pacific Warm Pool is full gravity flow starts east along the equatorial counter-current and an El Nino wave forms. It can be observed by satellites. As it reaches South America it runs ashore and spreads out north and south about twenty degrees. This warms the air above it, warm air rises, interferes with trade winds, mixes with the westerlies, and causes global temperature to rise . That is when we notice that an El Nino has started. But any wave that runs ashore must also retreat. As the El Nino wave retreats water level behind it drops by half a meter, cold water from below rises up to fill the vacuum, and a La Nina has started. As much as the El Nino raised the global temperature the La Nina will now lower it. They always occur in pairs and talk of El Nino-like or La Nina-like periods is just pure nonsense. The natural period of the oscillation is determined by the dimensions of the ocean basin and is about five years. In practice this can vary because there are other things happening in the ocean that can influence timing. El Nino has existed as long as the present equatorial current system has existed which is to say since the Panamanian Seaway closed. Nino3.4 is a good observation post because it sits there in the middle of the equatorial counter-current and records the presence of the El Nino waves before they reach South America. If you had read my book “What Warming?” you would know all that by now and I wouldn’t have to repeat it.
Regarding other uses of Neural Networks in climate science – they have been used in at least one IPCC-referenced climate model – see Knutti et al.: “Probabilistic climate change projections using neural networks”, http://www.up.ethz.ch/people/plattner/Publications/knutti03cd.pdf.
I have doubts about the legitimacy of neural network use in models – they have little to no explanatory power and even if they seem successful for some data runs, they can fail unpredictably.
Caz in BOS says:
Nice analysis, but you know very well that accurate prediction is the ultimate test.
Soo true. I’m working on it.
rgbatduke says:
The only mechanism by which ENSO can be driven by changes in Earth’s rotation is by variations in the tidal force.
Only mechanism? Driven by changes in Earth’s rotation? Neither of these is either proven nor (IMO) provable, certainly not from observing a correlation.
Sorry. I should have been more precise. Either variations in the Earth’s rotation are caused by mechanism in the Earth’s atmosphere and oceans or variations in Earth’s rotation is causing changes in ENSO.
Leron says:
Are you sure that the Mass differentials of the Bulges of warm and cool water during ENSO cycles across the surface of the pacific could not be the cause of the orbital differences detected.
I believe bulges along the same latitude should have little effect on variation on Earth’s rotation. It should only occur when water is displaced between different latitudes that there are large effects on the rotation as this forces changes in Earth’s rotation to keep up with the angular momentum.
Willis Eschenbach says:
Color me unimpressed … the claims made in this paper are waaaay stretched.
I won’t comment on each of your comments.
Instead I am going to make some general points.
This what is published on WUWT is not peer-reviewed and I have never claimed that it is.
It is out of personal interest I have made this research. I’m not associated with any academic institutions or have contact with any, although I have a M.Sc. degree in Applied Physics and Electrical Engineering.
However based what I have found in my research, I’m now convinced that tidal forcing is the main forcing for ENSO fluctuations.
For all practical purposes I’m not going to disclose in detail precisely how I proceeded or what type of optimization I made or what specific feature which it is that is responsible for the connections between the tides and ENSO variations.
The only comment I’m going to make on that issue is that the optimizations which worked best make full sense based on physical principles.
After all, this discovery has a certain monetary value. Not necessarily for me so much, but sure for others.
Of course I’m going to continue to improve upon the results and I’m going to produce real time ENSO forecast, which should give the ultimate proof.
I’m also interested to have my worked peer-reviewed by working together with some institutions. But, hey I’m a denier, after all!
Lots of problems with this study. The main issue is the data source. Try using JPL for the lunar, Earth, Sun coordinates.
Show the prediction for the next 10 years. Then there is something to talk about.
Per Strandberg (@LittleIceAge) says:
January 20, 2013 at 2:45 pm
In that case, I’ll make the general comment that you have refused to deal with the scientific issues I have highlighted. Very, very bad sign.
Per, perhaps you don’t know this, but as soon as you refuse to respond to valid scientific objections, your paper goes in the trash can. Seriously.
I don’t care if it is peer reviewed. I care whether you are willing to defend your work.
Yes, we know that. You’ve stated it over and over. It is clear that you are convinced. The problem is, you think the fact you are convinced should convince us … it doesn’t, no matter how many times you repeat it.
In that case, I don’t have a clue what you are doing posting on WUWT. This is a scientific site, surely you know that. By refusing to answer questions about your work, you are not acting as a scientist in the slightest, and your scientific opinion should be ignored until you are willing to be honest, transparent, and responsive to objections.
Ask me if I care.
Oh, please, not this coy nonsense about how you’d tell us but someone somewhere might make money off of it. I’m sorry, Per. I made a mistake. I thought you were a scientist. I didn’t realize you were playing businessman.
Since you refuse to answer questions, you are useless to us and to science in general. However, you may indeed have a bright future as a businessman. So run along now, and make some money, there’s a good fellow.
w.
Per Strandberg (@LittleIceAge) says: January 20, 2013 at 2:45 pm
……
rgbatduke says:
The only mechanism by which ENSO can be driven by changes in Earth’s rotation is by variations in the tidal force.
Only mechanism? Driven by changes in Earth’s rotation? Neither of these is either proven nor (IMO) provable, certainly not from observing a correlation.
……..
Sorry. I should have been more precise. Either variations in the Earth’s rotation are caused by mechanism in the Earth’s atmosphere and oceans or variations in Earth’s rotation is causing changes in ENSO.
I would suggest it is more likely to be from the Earth’s core differential rotation outwards. The Earth’s core angular momentum (CAM) is unlikely to be shifted by ENSO or tidal oscillations.
Comparing the AMO to the CAM a degree of correlation can be established, particularly strong since the early 1900s.
http://www.vukcevic.talktalk.net/CAM-AMO.htm
(note; AMO blue scale is inverted)
@Willis
The man threw a theory out there, if you must disparage something, maybe it should be the theory.
None of this is/should be personal.
No matter the openings.
Many here question the usefulness of Artificial Neural Networks (ANN).
ANN is just a tool. It can be an effective tool, or it can be a confusing tool which doesn’t give useful results.
Many have bad experience with ANN. Often they use off the shelf software and are unfamiliar with the method or data that they work with are not useful for use in ANN.
I think myself that to be effective with ANN you have to get your hands dirty with ANN. I my case I have constructed the software myself and I have full control of what is happening with the program and its algorithm.
Per, when your tidal peaks coincide with solar minimum, you will get a bigger amplitude El Nino followed by a deeper La Nina. The 1983 El nino was big because of the El Chichon volcano. This robbed some power from the ’87 El nino which started just after solar minimum. The ’98 and 2010 El Nino’s followed soon after Solar Minimum. The Pinatubo eruption caused a sustained succession but with lower peak levels of El Ninos.
When power going into the ocean is reduced by volcanoes or solar minimum, the ocean uses the opportunity to lose some energy.
u.k.(us) says:
January 20, 2013 at 3:54 pm
Nonsense. The man didn’t “throw a theory out there”. He made a vague unsubstantiated claim that “tidal forcing” is a major factor in ENSO forcing. He provided no data or calculations to back it up.
Then, when I asked him some plain vanilla scientific questions, he flat-out refused to answer a single one of my questions.
At that point, my friend, he’s not acting as a scientist, and I have no further use for him.
Finally, how can I “disparage” his theory? He hasn’t put forth anything remotely resembling a theory, and he refuses to answer questions about his claims … there’s nothing there to disparage.
w.
u.k.(us):
Your post at January 20, 2013 at 3:54 pm says in total
Willis never needs help to defend himself, and I am not writing to defend him.
However, there are two fundamental problems with the discussed analysis and the argument between Willis and Per Strandberg illustrates both although Willis only states one of them.
Per Strandberg has conducted a study using a neural network to identify relationships. This is problematic by nature because one cannot know what a neural network is actually identifying.
(A well known and extreme example of this problem was a military use of a neural net to conduct image analysis. The system was shown numerous photographs of American and Soviet tanks. It was tested to determine if it had learned the difference by showing it other photographs of American and Soviet tanks and it identified them perfectly. But when tested using live images it failed completely. Upon investigation it was determined that the neural network had trained itself to identify the terrain in the images and all the soviet tanks had been photographed in central Europe but the American tanks had been photographed in the US.)
Hence, there is an underlying problem: in principle it is not possible for Per Strandberg to precisely state what his system has done because he used a neural net. Therefore, he cannot answer some specific questions about much of his method.
Willis asked specific questions about input data to the neural net which were in the form of “optimisations” and Per Strandberg refused to answer. It seems very likely that Per Strandberg cannot answer because his “optimisations” consisted of training the neural net then adjusting input parameters as a method to improve the training. This is a typical method for training a neural net (it is analogous to the military showing its system more images of tanks in my illustration).
Willis gave up in frustration because – as he said – it is not science if what is done cannot be precisely specified. But training a neural net is not a procedure which can be precisely specified.
More problematic is that the output of a neural net is not known with certainty. In this case, several posters have commented that the direction of causality is not determined. Indeed, as the military example shows, the output may not be indicating what one thinks one has trained the neural net to identify.
Hence, I am in agreement with Willis and this agreement is not “personal” to either of the discussants.
Richard
I agree with many of the criticisms that others have made. You have not shown causation, you have not shown a physical model that convinces me and many of the details are missing. I appreciate that English may not be your first language, but we need a lot more detail to evaluate your ideas. However, whether you have found a first order effect or an Nth order effect, what is interesting is if you have found a way to forecast ENSO. But I would like the detail filled in.
/ikh