Profile image
Story Views

Last Hour:
Last 24 Hours:

Per Strandberg: The physical forces driving UAH global temperature

Thursday, October 13, 2016 10:47
% of readers think this story is Fact. Add your two cents.

I presented here a while back my research using an Artificial Neural Network analyzing ENSO.

Learn more here

I’m going to write here how it all started, but first I like to show my updated recent ENSO data and forecast which I presented at the recent climate conference in London.


Fig 1: ENSO result from my ANN. Training period is from 1979 and up to 2005. The testing period is from 2005 and up to the end of 2015. From 2015 and up to the end of 2022 it is a forecast. The red line is the real ENSO value and dark line is the result I got from the ANN. As you can see the dark line is from the average values from ensemble.


Fig 2: Zoomed up version of the previous graph. As you can see my prediction indicate that the current La Niña is going to deepen and reach it strongest values somewhere around February, March or April.

Something about the background that led me to start to investigate ENSO.  I had learned about Henrik Svensmarks theory of galactic cosmic rays and cloudiness which I taught was a good alternative to the theory of CAGW. At the same time, I saw in the media more and more extreme claims about the coming global warming catastrophe. Eventually I decided to investigate this for myself. I was more interested in looking for data than to read peer review papers. One tool I used was Google’s image function. You search for data and instead of looking for papers you look for graphs in the images function. Then you look at the paper that graph comes from. I quickly found out that the sun’s electromagnetic activity as a climate driver was greatly underestimated.

I had worked with ANN before and know it’s math, so I decided to test it for climate data which I think it is well suited for. I was interested in if CO2 effect could be studied this way and also if GCR according to Svensmark could be effecting the global temperature. One benefit with all the research that has been made in this field is that data are easily collectable is download from the internet. I collected all sorts of possible data that could have effect on global temperature.  I then build an ANN, I put it all together and tested with the in-data going from the previous month and going back 3 years as inputs in the in-going neurons. I then played around with this data and looked if I could get convergence. I tested the network against the global temperature as well as the derivate value of the global temperature. I analyze the result of the derivate value from the network against the derivate of global temperature as measured from UAH as this gives a higher frequency and better statistics. I then started to investigate each parameter individually by registering if they diverge or converge. If they converge I registered the error variance value “the result” for each.

I knew that ENSO affects global temperature, therefore I put its influence on the derivate value of the global temperature to be 100%.

Here are the different parameters in ascending in percent value: SST 110.2%, ENSO 100%, LOD 68.5%, Solar wind speed 49.5%, SOI 45.8%, Kp Magnetic Index 27.4%, Solar wind temperature 26.3%, AMO 22.4%, Ap Geomagnetic index 13.3%, Solar wind density 9.7%, Sunspot number 4.8%, F10.7 radio flux.  4.0%. The exact percent value is not the important point here, but rather the order.

And here are parameters that diverge. In order words, I didn’t get any correlations. Interplanetary magnetic field IMF, Neutron counter -> Galactic cosmic radiation, TSI, PDO.

I would like to stress again that this is results from my ANN on the relations between these parameters and the derivate value of the global temperature as measured from satellite. In other word, this is for short term variations and from pulse like effects. Nevertheless, there are several interesting things that can be said about it.

SST is the global sea surface temperature. That SST has more influence on the temperature than ENSO, makes sense.

LOD Length Of Day which is the same as small changes in Earth’s rotation is affecting changes in Earth’s global temperature anomaly. This may sound a little weird but it makes sense because ENSO and LOD are correlated to each other. In other word, LOD is an ENSO signal. But, why are they correlated?  Well, the most logical reason must be that they are connected by tidal forcing. This was a reason for why I started to investigated possible tidal forcing on ENSO.

Solar wind data and Kp and Ap all show high correlation to the temperature variations. These are all related to electromagnetic variations of the Sun. The correlations with these parameters to temperature variations are all hard evidence that electromagnetic variations of the Sun have an effects global temperature variations.

Next comes AMO or Atlantic Multidecadal Oscillation which is an index over temperature anomaly in the North Atlantic. There are other Oceans indexes which I didn’t examine.

Of less important are the radio flux at F10.7 and the sunspot numbers.

And then there are other parameters that show no linkage to short global temperature changes at all. Among them are TSI which is a measure of the amount of heat beaming down to Earth from the Sun. Note also that the PDO (Pacific Decadal Oscillation) show no linkage to temperature derivate value. There is a reason for this. The definition of PDO is temperature anomaly value north of the tropic in the Pacific Ocean when SST value have been subtracted. This doesn’t mean that it has a long term effect. GCR as measured from the Neutron counter at Oulu university show no correlation with the temperature variations. In fact, my results show that the linkage between solar activity and short term temperature variations is dominated by variations in solar wind and in changes in Earth’s magnetic field.

I have found Artificial Neural Networks or NN for short is a powerful technique for analyzing climate data which depends in many inputs and is subjected to time delays of varying degrees. Alternatives such as statistical regression analysis, frequency analysis and the use of dynamic models to analyze this type of data are all insufficient.

To exemplify this problem, let’s take a look at solar wind data. The temperature derivate response from changes in the solar wind variations is a complex one with different response times. I believe that the solar wind directly influences things like AO and NAO in the northern hemisphere. This is exemplified by phenomena such as sudden stratospheric warming. The response time for this is short usually about a month. But, thing gets more complex when we include the solar winds effects on ENSO. The function of El Niño is that it works as a ventilation mechanism releasing heat from the pool of warm water in western Pacific thru what is called Kelvin Waves which moves warm water toward the East. If this water reaches the surface in eastern pacific we can get an El Niño. This is not always the case as this can be blocked by cold upwelling and the warm water is then being dispersed. These mechanisms are influenced by variations in the trade winds. The normally easterly trade winds in the tropical pacific push warm surface water to the west raising the sea level there and filling it up with a growing pocket of warm water. When what is called the JMO Julian Madden Oscillation is positive in the western pacific in an area called the Kelvin Wave Generation Area then the trade winds change direction and release the water pressure which has built up the higher water level in this area and Kelvin Waves are usually generated. JMO indicates an area in the tropics with enhanced convections. This area of convections moves counter clockwise around the equator. This convection area makes an orbit around the Earth during a period of about 30 to 90 days. I expect that JMO variations is mainly driven by lunar gravitational Perigee pulses and from variations in the electromagnetic variations of the sun, but I haven’t had time to analyze this yet. So part of the changes in solar wind affect the ENSO index while it has an effect on the trade winds. So the solar wind affects the trade wind, say after one month. This lead to the generation of a Kelvin Wave after an additional month. Then this kelvin Wave reaches the surface in the Eastern Pacific “sometimes” after about 3 months and we get an El Niño. The effect from the added humid air during El Niño then affect the global temperature anomaly after about 5 months. The result in my NN that is generated from solar wind variations is thus from an aggregated sum of different time lags which includes it response from its effect on ENSO.

This kind of correlations is almost impossible to do with other methods. I guess, if one use, for example, statistical linear regression and analyses values from individual time lagged months it could be possible to find some weak correlation.

So why do not more people use NN to analyze similar complex relationship where climate and weather related phenomena seems to be ideal?

One reason is that it is complex and takes time to work with NN. And while there it is possible to buy NN software from the selves it is not easy to use especially when the user is unfamiliar with the underlining mechanism.

It is more of a kind of handicraft skill that one need to utilize the full functionality of an NN.

In my case it was about 4 years ago that I discovered that the connection between lunar cycles and ENSO variations was thru the seemingly chaotic variations during individual Lunar Perigee gravitational pulses which was the main driver of ENSO variability. I have still a way to go until I’m completely satisfied with my result. When that is done I can quickly examine other related ENSO indexes and other types of climate data. I just switch from the MEI ENSO to whatever parameter I want to investigate without adding extra program code.

During the following 4 years I have improved and tested different ideas thru successive baby steps. I have now very good results and as I believe that I also have created good forecast for the current expected development of ENSO several years into the future.

While, I’m not here going into how NN works in detail, I’m going to describe some basic principles how it’s works. You find easily information how NN works on the internet. Neural network is built around asymptotic transfer functions in a network. In order to make it work, the network has first to be trained. In my case I use an initial training period where thru successive modification of several hundreds of weights which regulates hundreds of asymptotic transfer functions. The goal with an algorithm I use is to minimize the variance value between the calculated output from the network and the real ENSO value calculated for each month in the training period. At the same time during this calculation I use a test period where I also calculates the variance value in the same way except I use the same weight values which has been calculated in the training period. Not the test part. The goal is to not only minimize the variance value during the training time period but also to minimize the variance for the test period. The variance for the test period is usually converging in the beginning but eventually it starts to diverge because of statistical noise after many iterations. At that time or just a time before that time, the weight values are saved and a recalculation based in the weight values can be made for the ENSO including for a forecast period. I have thus been successful in identifying the underlining forces that is behind ENSO variability by eliminating noise sources or suppress them and by amplifying data which contains correlations between in and output of the NN. I have employed innovative and unorthodox methods so that the NN can differentiate between noise and signals which contain correlations. Nothing in set in stone.

The fact that I have identified the main drivers of ENSO variability, and yet the mainstream climate community has not, raises an interesting question: why haven’t they included variations in the electromagnetic solar activity in their GCM models? After all, if you look at correlations between electromagnetic solar activity and variations in global temperature anomaly on a decadal scale you find a good fit. But first you have to compensate for the current temperature data manipulation by GISS.


Note: On the temperature spike during the recent El Niño.

While the strength during the resent El Niño was weaker than that of the El Niños of 1982-83 and 1997-98 the recent El Niño was a slow starter with near Modoki El Niño during 2014-15 which warmed the surface in the pacific and tropical air masses. Similar effect to ENSO exist also in the tropical Indian Ocean and tropical Atlantic. The response time is different and weaker. Because this El Niño’s had a slow start, which also included a Modoki phase (westerly displace El Niño), all the tropical water was warmer than normal. This was not the case during the previous stronger El Niños which started quicker and from negative ENSO values.


We encourage you to Share our Reports, Analyses, Breaking News and Videos. Simply Click your Favorite Social Media Button and Share.

Report abuse


Your Comments
Question   Razz  Sad   Evil  Exclaim  Smile  Redface  Biggrin  Surprised  Eek   Confused   Cool  LOL   Mad   Twisted  Rolleyes   Wink  Idea  Arrow  Neutral  Cry   Mr. Green

Top Stories
Recent Stories



Top Global

Top Alternative



Email this story
Email this story

If you really want to ban this commenter, please write down the reason:

If you really want to disable all recommended stories, click on OK button. After that, you will be redirect to your options page.