Applying Regression Switching to Global Warming Data

The BIS.Net Team BIS.Net Team

Since it is not possible to directly test whether greenhouse gases do cause global warming, the best that can be done is to analyse the correlation between greenhouse emissions and global temperature. However, even this has limitations.

Of course, correlation does not establish causality. The correlation may be due to other related factors. However, absence in correlation can throw doubt on the relationship.

Correlation between global temperatures and greenhouse emissions has already been used as supportive evidence by science. However, the analysis technology publicly available is surprisingly over simplistic, hence a machine powered more modern analysis was performed on the data,using the Change Analysis App.


The following analysis is based on data that was freely available on the internet on sites such as



The findings are as good as the reliability of the data provided

The science is very complex and based on decades of research. We have not seen the full extent of the research. All we can do is analyse the correlation of greenhouse emissions with global temperature based on the publicly available data.

Some sections of the data are known to be unreliable. Monthly data computed before 1880 data was sparse. More and more data became available after 1880.

There have been many changes in testing and sampling over time. This can also affect patterns.

Sampling problems

Average global temperatures cannot possibly be true atmospheric averages. To accurately estimate the average atmospheric temperature requires measuring stations devices systematically distributed throughout the relevant parts of the atmosphere. Where are the measuring devices in the atmosphere?

Weather satellites which can take measurements in the atmosphere and surface do not measure temperatures directly they are only inferences and hence subject to inference errors.

Temperature anomaly

Deviations from reference are used to demonstrate global warming. The reasons given are that this is more accurate than using absolute values and it is a better way of comparing regions with vastly different climates.

The argument is a good argument, but bear in mind that anomalys lose perspective of the change relative to the total temperature variation. A one-degree change at the north pole may be disastrous, but a one-degree change in the Sahara may have little effect.

A one-degree change may have little consequence when the annual temperature range is 50 degrees compared to an annual temperature range of 10 degrees.

Existing comparisons of temperature anomaly with Green-House Emission

LOWESS smoothing or simple regression is commonly used to demonstrate global warming.

Figure 1 Global temperature anomalies since 1880

Lowess smoothing makes the presentation more appealing. But it is only a presentation tool, without statistical basis for identifying statistically significant trends and thus provides no additional useful information.

Simple regression, such as shown below over the whole data set, does not accurately model the data as the time series does not consist of a single linear trend.

Figure 2 Biasing an upward trend by fitting a linear line to a data set that is not mono linear.

To provide evidence of the link between global warming and greenhouse gases scientist have compared the Figure 1 and 2 data charts to emissions charts, such as below.

Figure 3 An example of an emissions chart

Comparing Figure 3 with Figure 1 over the same date range shows that both emissions and temperature have increased over the same period. Both were low at the beginning and both were high at the end.

Such a simple comparison does not provide scientific statistical evidence that greenhouse gases are responsible for global warming. They could be coincidental.

An alternative analysis

Modern machine powered algorithms can identify statistically significant trends which can be used to evaluate the strength of a relationship between global temperature anomalies and greenhouse emissions. Applying the technology, the following output was obtained.

Figure 4: (Top Chart) Greenhouse Emissions (Bottom Chart) Land + Ocean temperature anomaly using air temperature above sea ice from 1850 onwards.

The algorithms themselves are irrelevant. A visual inspection shows that the algorithms correctly found the underlying significant temperature trends. These trends are vital to determine if perhaps science is wrong.

The emissions graph does show non-linear changes which are not accurately reflected by linear trends. Although a more advanced algorithm could be used this was deemed unnecessary as the conclusions below are around global warming which throw doubt on manmade global warming.

Section 1 below is prior to 1880 where data was less limited. This explains the greater dirtiness of the patterns. Likewise, section 2 is dirtier where the data is not neatly randomly distributed about the underlying trends. Nevertheless, the trends are statistically significant.

Both section 2 and 4 show a drop-in temperature of a similar magnitude drop. The drop in Section 4 has been attributed to aerosols by the scientific community, but that cannot be 100% proven. The fact that there is a similar drop in section 2 does open the possibility that there was another factor involved.

The upward trends in section 1 , 3 and 5 are remarkably similar in slope, taking into account expected sampling variation in the slopes.

Figure 5 The rapid increase in global temperature started as early as 1850 at a time when greenhouse emissions were very low relative today.

The analysis shows that global warming may have commenced as early 1910 and possibly as early as 1850. The greenhouse levels at that time were low.

The fact that there were similar slopes of warming in sections 1, 3 and 5 points to the possibility that global warming is due to another reason. Perhaps there is natural global warming which a reduction in emissions would not prevent.


Scientists have built a record of past climates using ancient records from earths past climates from ice core samples. It is not 100 percent reliable method and has many holes, but it does provide a disturbing insight if we assume that there is reasonable accuracy in the technology.

The image below is for the last 800,000 years.

Figure 6 Temperature anomalies over an 800000-year time period estimated from ice core samples.

0 is the current value. Looking at the cycle time between maximums we are due for global warming, irrespective of greenhouse emissions. We may just be experiencing a natural event, if we can believe in the indirect estimates from ice core samples.

Assuming the estimates are reliable, what is disturbing is the fact that for the last 400,000 years the maximum has increased considerably from the 400000 period before. There were no carbon dioxide levels similar today.

NASA acknowledges that there were periods warmer than today, confirmed by the image.

This means that there could be further increases in temperature and there is nothing we can do about it, other than prepare ourselves

Scientist are taking issue with the fact that in the past century alone the temperature alone rose by .7 degrees, ten times faster than the average rate of ice age recovery rate. That does not support greenhouse emissions. We don’t know if we had such localized increases during past natural upturns. The resolution of the information is too low to tell.

‘Random’ Effects cause of global warming

Everything about the earth is variable. Temperature can NOT remain constant.

There is a statistical phenomenon called random walk. If the true global average this year is now 20 degrees Celsius it will unlikely be the same next year because there are so many variable natural events (changes in ocean current, wind, solar activity …). Next year the temperature will either be a little higher or it will be a little lower. If its 20 this year, next year it may be 20.2 or it may be 19.7.

Even though we would expect temperature over time to fluctuate around 20 degrees without a long-term change, this does not have to be the case if there are no buffering mechanisms. A random walk would start if next year’s temperature depends on this year’s temperature and there is an equal chance of the temperature being above or below this year i.e. no forces such as greenhouse gases driving temperature higher (if a fact), or aerosols driving temperature lower (if a fact).

It is easy to simulate how temperatures can change. The results of such a random walk are shown below. The first image shows the true global warming trend and the second trend that was due to a simulated random walk. The pattern is similar showing that even pure random events which have nothing to do with mankind can cause the temperature changes we are experiencing.

No assertion is made that this is the cause. The objective is to simply point out that real life is far more complicated and that there could be other explanations


There exists some evidence (but nowhere near conclusive) that current global warming is due to a natural cycle which we have no control over. If this is true, then politicians need to decide the worth in greenhouse emission reductions and instead prepare for even higher global temperatures.

The data throws considerable doubt on the conclusion that global warming is caused by greenhouse gases. But it is important to realize that just as scientists have not conclusively proved that they are right the analysis has not conclusively proved that they are wrong. The analysis hinges on the data and since we had no input in how the data was obtained and how it was processed we will not make the mistake of dogmatically stating the conclusion is right.

Also, if it has been proven that currently there is no greenhouse effect, this does not mean that in the future emissions will not reach a threshold that triggers man made global warming.