# Changing temperature anomaly baselines

I wanted to check whether the choice of baseline can affect the calculation of  global temperature anomalies from station data. Each temperature index (GISS, Berkeley, CRU) uses different normalisation periods for calculating weather station temperature anomalies. I was surprised to discover that this choice makes no difference whatsoever to the results.

I used the new GHCN V4 which contains 27315 weather stations, and calculated the global average temperature anomaly relative to 5 different 30-year baseline periods using Spherical triangulation. Selecting different baselines restricts the analysis to those stations with sufficient data falling within those periods. Here are the results.

Global Land temperature anomalies calculated relative to 5 different baselines. The numbers in brackets are the number of stations contributing for each baseline period.

All the trends are very similar despite a factor of up to 8 difference in the number of stations used.  We can compare them all directly by offsetting each onto the same 1961-1990 baseline. To do this I simply scale each one by the offset difference between 1961-1990 (shown in ‘calc’ brackets).

All 5 baselines offset to the same 1961-1990 normalisation. The offsets are shown as Calc.

The results are surprisingly similar.  This means that the choice of baseline period is essentially arbitrary and does not affect the end result.

This entry was posted in AGW and tagged , . Bookmark the permalink.

### 17 Responses to Changing temperature anomaly baselines

1. Bryce F Payne says:

“All the trends are very similar despite a factor of up to 8 difference in the number of stations used….The results are surprisingly similar. This means that the choice of baseline period is essentially arbitrary and does not affect the end result.”
Central Value Theorem still stands.
There are enough stations in the available data to provide a statistically consistent picture.
Appear to be two periods of falling and two periods of rising temp.
Falling T periods are roughly half as long and half as steep as rising T periods, i.e., climate is warming.
Pattern suggests may be an upcoming period of falling T.
If not, then………
Unfortunately the whole data set includes only two falling/rising T cycles, and no longer term cycles.

2. oz4caster says:

For annual temperature anomalies, changing baselines simply shifts all the annual values up or down by a constant amount and does not effect trends. However, for monthly or daily temperature anomalies I have found that changing baselines can effect seasonal patterns, but does not effect long-term trends. Here is an example looking at daily NCEP/NCAR Reanalysis 1 temperature anomalies:
https://oz4caster.wordpress.com/2019/02/15/global-temperature-reanalysis-baseline-comparisons/

I am planning to post some additional results in the near future.

3. Mr Broccoli says:

Interesting graphs Clive. I see the second and more pronounced rising trend comes after Starfish Prime. Do you think that the disruption caused to the Van Allen belts has had a lasting influence? I remember at the time people speculated that it would have long term effects on climate.
Also
I understood that the had been a hiatus in warming this century. Why do the graphs not show this? Was there no leveling off?

4. Nick Stokes says:

Clive,
Changing anomaly base means adding an offset for each of the 12 months. I posted a table of the needed offsets for the major indices here. Because it is done by the month, the difference between months can change slightly. If there was a run of warm Mays in 1951-80, but cool Junes, then that get built into that base and subtracted. So in other years, that will induce a small artificial rise going from May to June, which might not happen with another base period.

These offsets are worked out after averaging. You are looking at also changing the population of stations depending on the availability of data in the period. Most majors (and I) have some way of making use of stations lacking data in the period. But even without that, there is such an excess of stations (esp in V4) that there is not much difference in changing subsets.

5. Gerry McIsaac says:

Calculating anomalies is simply coding the data to give each station an average of 0. For global temperature anomalies, the mean is set to zero but the variance is the same as that found for recorded temperatures. The difference in the response of graphing recorded temperature vs. time compared to calculated anomaly vs. time is just a matter of scale.

• Clive Best says:

Yes, except for sampling differences. Individual stations will be included or excluded depending on their having temperature measurements within the normalisation period.

6. Gerry Mcisaac says:

Valid point. That would suggest a global mean temperature calculation would be reliant on temperatures measured within the normalization period specified, i.e. 30 years. That would suggest in order to obtain an average global temperature for a period of say, 100 years, the normalization period should be extended to 100 years to include the entire time period in which the recorded temperatures were assembled. Still, just a matter of how you code the data as you have shown.

• Bindidon says:

Gerry Mcisaac

“That would suggest in order to obtain an average global temperature for a period of say, 100 years, the normalization period should be extended to 100 years to include the entire time period in which the recorded temperatures were assembled.”

No. Of course it shouldn’t, at least within the anomaly model chosen by Clive (and by me too), in which only those stations contribute which have sufficient data within the reference period.

In your case, that would mean that you de facto exclude all stations whose lifetime does not include the 100 year reference period.

This is not a good idea, and is the reason why all professionals working on anomaly construction use a model independent of reference periods.

See for example:

Homogenization of Temperature Series via Pairwise Comparisons (2009)
http://www-personal.umich.edu/~eoswald/Menne2009_QC_algorithm_4USHCNv2.pdf

Rgds, J.-P. D.

7. Bindidon says:

Clive Best

This is for me a very interesting post, as I did some similar observations when ‘doing my very best’ last year in processing NOAA’s GHCN daily record (what you did too at about the same time, if I well remember):

https://drive.google.com/file/d/1a36CaZqdv9vQak5Rx4UplRIFum4YuKBS/view

vs.

https://drive.google.com/file/d/1bWUVAYII5NANG9Na4atzYEAJgTsZXa-W/view

But however, the situation becomes a bit different when choosing 1891-1920 as reference period, because many stations active at that time aren’t active right now. I don’t have the chart at hand, but the difference is much more perceptible: the number of GHCN daily stations moves here from about 36000 down to less than 1000.

My impression is that the GHCN V4 data set contains much more data for earlier times as does GHCN daily.

As I would like to avoid using Menne’s concept inherently married to homogenisation, I’ll try using a concept of a sliding, overlapping reference period of say 10 years, and to construct a time series wrt a specified main period out of all these small intervals, by shifting the local monthly anomalies by the climatology differences between each period and the main one.

The motivation to do this came from a trial to adapt the GHCN V3 / daily ware to a similar processing of the sea level data from

https://www.psmsl.org/data/obtaining/rlr.monthly.data/rlr_monthly.zip

Rgds, J.-P. D.

8. Gerry Mcisaac says:

“In your case, that would mean that you de facto exclude all stations whose lifetime does not include the 100 year reference period.”
There is no exclusion or inclusion of data.
What ever normalization period is used, the plots presented show anomalies from 1845 to present based on different thirty year periods. When a single defined thirty year period is used, they overlap.
The difference is in how you code the data. Using different average values will result in different anomaly values. Using a set average value will result in the same anomaly values.
Here’s a question.
If so much attention is given to the number of relevant stations with regard to the accuracy of the data. How can you represent a time period from 1845 to present using a 30 year normalization period from 1961-1990? There are definitely a large number of stations included or excluded during this time period.

• Bindidon says:

I don’t understand your comment.

On the one hand you write

“There is no exclusion or inclusion of data.”

“Using different average values will result in different anomaly values. Using a set average value will result in the same anomaly values.”

On the other hand:

“There are definitely a large number of stations included or excluded during this time period.”

What exactly do you mean?

9. Brendan says:

This data looks like it has been adjusted. The cooling from, 1940 to 1970 has been flattened out and the globe’s temperatures have been going sideways since 1998, not up in a vertical line like you have here. Your 1998 to 2018 chart that you have in your next post “January 2019 global average temperature remains unchanged – 0.73 C” shows the flat line temperature, totally different to this chart in this post.

10. Bryce Payne says:

Seems there are a lot of comments that miss what I think are the more important points of Clive’s results. It appears that regardless of the time frame or the number of stations providing data the chronological pattern is for all practical purposes indistinguishable. If I had time, my approach would be what is the minimum amount of data, and which are the essential stations, from which data is necessary to provide a reliable indication of the global average temperature (anomaly)? I have long been concerned that statistically it should not be necessary to have huge numbers (10 of thousands) of reporting stations to get a reliable estimate of the global average temperature. Clive’s recent findings seem to support that statistically based expectation. After all of this it remains the case that allowing for occasional “random” variations, there is a definitive global warming trend that appears to amount to about 1.5 deg C over the last 150 years or so, and about 2/3 (1 deg C) of that warming occurring in the last 40 years or so.

• Bindidon says:

Bryce Payne

It’s a bit late for a more detailed answer, but concerning your question

“what is the minimum amount of data, and which are the essential stations, from which data is necessary to provide a reliable indication of the global average temperature (anomaly)?”

I propose you to read an excellent article written 8 years ago by Nick Stokes:

https://moyhu.blogspot.com/2011/03/area-weighting-and-60-stations-global.html