Annual Temperature Plots
Cooling the Past

The question is - What is the overall effect of adjusting the data?

It is frequently claimed (by skeptics) that the primary cause of "global warming" is the way data is adjusted. The analysis presented below basically supports that claim.

To reproduce this analysis, use the GHCN Temperature Plotter tool and

Basic setup | Average plots - More data, Fewer sites The trends | Histograms | xy-Plots | Maps | Discussion | Notes on the images

Basic setup

The default baseline for most analyses (not this one) is 15 years of data between 1961 and 1990.

For this analysis, do the following for both the raw and adjusted datasets.

This should select 683 sites in the raw dataset and 615 sites in the adjusted dataset. This difference is assumed to be because, even though the number of sites selected in each should be the same, the number of years reported within the two date ranges is different.

These are 2 examples of sites present in both datasets, but not included in the filtered selections. Sites of this type are easy to find - just toggle between the 2 datasets using the provided radio buttons and see which sites (dots) disappear. There should be 68!

In both cases, a negative trend in the raw data was converted to a positive trend after adjustment. However, because the adjusted data is missing several years present in the raw data, the filter criteria removed the adjusted site from this analysis. The software is able to remove these sites from both datasets, but this page was written before that feature was added (which is, in fact, the reason it was added) and removing those sites makes almost no difference in the analysis presented here. (I thought it would make the comparison more fair and had to try it to see if it made a difference.)

On the other hand, the fact that the adjusted data is missing data points (years) where the raw dataset has data indicates a far more serious problem.

To recreate these images

The Search by Name function on the Station Data tab can be used to locate the stations. Using this technique, you will have to search separately on each dataset - raw and adjusted.

Average plots

Using the filters defined above, these plots compare the average anomaly and actual temperatures for the selected stations. For these plots, the average temperature (or anomaly) is plotted for each year.
When the raw data was adjusted, the temperature trend for the entire period was changed The difference between the two plots is surprising - based on simple math, the 2 raw trends should be identical and the 2 adjusted trends should be identical. However, these differences are because there are large holes (lots of missing years) for some of the sites,

The anomaly plot makes it clear that the relative values of the raw data were not changed from 1900 to about 1940 and the actual temperature plot makes it clear that the actual values were not changed from 2000 to 2014.

However, it is clearly obvious that the 1900 to 1940 temperatures were decreased about 0.4°C - thus the adjustments increased the computed rate of temperature increase.

These images were produced via the Hi/Lo button.

More data, Fewer sites

There are many ways to select which sites to use. The standard for this page just requires data during the baseline period (1910 to 1920) and 2000 to 2014. Using a more restrictive filter (one with fewer holes in the data) produces different results - but I did not expect this specific difference. (On the Basic Filters tab, check all the options and press the Clear filters and Set defaults button.) Using all the default filters with the 1910 to 1920 baseline
When long term, high quality, temperature records are modified like this, I think it is necessary to investigate why.

The rest of the images on this page use the less restrictive filter described in the Basic setup section above.

The trends

I was quite surprised to see that the trends for the anomaly data were different than when the actual temperatures were used. As mentioned above, this is due to missing data - many sites do not have data for each year.

The plots are generated by averaging the temperatures, or anomalies, for each year and then fitting a trend line thru the results. Because the number of temperature values (number of sites) varies from year to year, and because each site has its own baseline used to calculate the anomaly, the annual average of the baselines varies slightly. It is this that causes the trends to vary. Thus, the anomaly trend for the raw data is different from the temperature trend for the same raw data. (Same for the adjusted data.)

The application also provides a full linear regression (via the Trend Lines tab) using all the data for the selected sites, not just the annual averages. Even that produces slightly different slopes for anomalies vs temperatures.

As a further test, I selected only the sites with data for every year - then the slopes (trends) were identical.

You can argue that it would be better to simply average the individual trends - perhaps, but I think a histogram provides more information.

Does any of this matter? - Well, when the choice of method gives results that vary by 15% or so, I don't think it is reasonable to make a big deal out of a change of only 0.1% (something the press does on a regular basis).


The linear plots in the previous section show basic trends. However, the data form all the sites is averaged per year before finding the trend. This completely hides the fact that some sites are cooling and the "Global" warming isn't global. Adding a standard deviation to the plots helps (there are options for that on the Std Dev tab), but it sort of hides what is really happening. The Histogram plots look at the adjustments made to each station before the data is averaged.
Notice that the histogram makes it obvious that most of the raw data with cooling trends was changed to produce warming trends.

These images were produced via the Histograms tab.


The xy-plots provide a way to display the changes between the raw and adjusted datasets in some parameter for each reporting station.
It should be obvious that some sites had the past values cooled, while a much smaller number had it warmed. While the average cooling was about 0.04°C/decade, it is obvious that a significant number of sites were cooled a lot more than that.

As expected, when the adjustments cooled the past, the rate of temperature increase increased.

I checked these differences against all the available x-axis options, but did not see any significant trends worth showing here other than the change in trend verses the raw trend. (Not shown) However, the histogram in the previous section makes that much clearer than the corresponding xy-plot.

By the way, on the xy tab, the charts that plot adjusted vs raw data or the difference between the raw and adjusted data use only the sites selected in the adjusted dataset - it does not matter what sites are selected in the raw dataset. (Every site in the adjusted dataset is also in the raw dataset .. but not vice versa.) For most other plots - it does matter!

These images were produced via the xy tab using automatic bin sizes.


This data can also be displayed as a colored dot map (where colors show which sites are warmer and which are colder after the adjustments were made). Even when it is zoomed, it is hard to see what is happening on the following image. It is better to use the application since it allows you to zoom the map (rather than just an image). The USA detail is provided as an example.

These maps were produced by selecting Baselines on the Raw vs Adj tab and 7 levels on the Bin Colors tab.

The maps can also be used to show trends (Map Trends tab) and the change in trends (Raw vs Adj tab). In each case, the map is updated as soon as the selected date range changes. (Hint - use the mouse wheel and the shift key with the date fields.)


Of course, these few images don't prove anything one way or the other. However, they do demonstrate clearly (at least in my opinion) that the skeptics have a point - That the adjustments appear to be the major cause of the apparent temperature increases.

They also make it clear why skeptics want to see exactly how and why the adjustments were done. From my experience, there is not enough data available for anyone outside the field to validate the adjustments, and there is more than enough data to question the techniques used.

By the way, a zero line (Hi/Lo plot on the Plot Controls tab) lies completely within 2 standard deviations of the data - strongly suggesting that the null hypothesis (the Earth is not warming) should be accepted.

These images were produced using version 0.15 of GHCN Temperature Plotter tool and the configurations specified above.

Notes on the images

Note: All the images on this page can be zoomed by simply using the mouse wheel.
Double click to toggle full size to default size

Author: Robert Clemenzi
URL: http:// / Science_Facts / Annual_Temperature_Plots / Cooling_the_Past.html