GHCN Data Analysis: A Simple Approach

Guest Post by Andrew Chantrill

I recently decided to see what information could be obtained from the unadjusted GHCN temperature data. Simple averaging of all the data does not work because stations come and go. For example the ending of a temperature record in Barbados and the simultaneous starting of one in Siberia would introduce a ‘step change’ downwards.

Accordingly data from each reporting station is usually allocated to a geographic ‘cell’ on the surface of the Earth to ensure all parts of the globe are equally represented.  Each cell is analysed and a trend calculated, and then the results for all cells combined to give a global result. Whilst attractive, this method is very time consuming, and decisions on the weighting given to each station provide an opportunity for claims of bias.

I decided on a simple method that eliminates the ‘step change’ problem without incurring the effort in allocating each station to a cell.

In essence I calculated the temperature change, year on year, for each station. These changes are averaged for each year, and these averages are then summed to ‘integrate’ the annual anomalies back to a long-term anomaly. Here’s a summary of what was done for anyone who would like to try it:

1. Downloaded v2.mean.z file (from here), and opened it in Excel. Note: You need to be running either Excel 2007 or 2008 as the file is some 595,000 rows.
2. Extracted the station number, year and monthly temperature data into separate columns. Eliminated rows which did not have a full 12-months data.
3. Calculated annual average temperatures for the remainder.
4. Ran a pivot table report with years across the top and stations down the side, and temperatures in the body of the table.
5. Calculated the temperature changes one year to the next, in a table which looks very like the pivot table.
6. Averaged the temperature changes for each station reporting for each year.
7. Summed the temperature changes to ‘integrate’ the changes back to ‘temperature anomaly’.
8. Calculated an 11-year rolling average (centred on year 6).

The results are plotted agains the number of stations used in reporting temperature to GHCN in any given year (Figure1).

Figure 1. Temperature changes in data averaged from the GHCN database since 1700.

The unadjusted result suggests that current temperatures were previously seen in the 1730s.

Whilst open to criticism of under- and over-representing parts of the globe (most of the stations are US-based, for example) it does include all the data available and weights each station equally.

[I saw Andrew’s link to his graph under comments in Jeff Id’s post on CRU data analysis and I wanted to show it here mostly as it is a slightly different approach than many people have taken. VJ]

This entry was posted in Uncategorized. Bookmark the permalink.

2 Responses to GHCN Data Analysis: A Simple Approach

1. tonyb says:

I have written previously about the similarities between todays temperatures and the 1730’s. Here are a few extracts;

Defining climate trend as a 30-yr trend one can plot the CET trends for a sliding 30-yr window:
http://www.leif.org/research/CET1.png

The following, condensed from the records of the Hudson Bay company, demonstrate that climate change is not a new phenomena.

“Over the fifteen years between 1720 and 1735, the first snowfall of the year moved from the first week of September to the last.

This comes from the extensive weather records of Thomas Jefferson;the warm weather of the early 1700’s has given way to intense cold then another period of warmth