This post presents an R based chart of global GISS land and sea temperature anomaly data for the 1880-2009 period with both the long term trend and the individual decadal trends. Links to the source data file and the R script are provided so that readers can reproduce/ improve on this analysis.

## Global Temperature Trends

I have a number of posts on global temperature trends (see my Climate Trends page for links). As I have studied climate change (see my ProcessTrends.com page for chronology of my learning curve) I have found that I needed to sharpen my trend analysis because climate trends are affected by annual, decadal and geologic time scale cycles as well as natural variation. El Nino and volcanoes, for example, play a major role in year to year variation in temperature trends (see my previous posts here, here and here).

In my George Will post, I showed why it is dangerous to over-interpret a short term, 10 year trend.

In this post, I show the decadal trends for the same data set I used in the George Will post.

**GISS Anomaly Decadal Trend Rates **

GISS provides an on-line file of monthly temperature anomalies since 1880 at this link. Here’s a sample of the file:

While this file is a great resource, it takes some work to develop an automatic way to update trend charts for do-it-yourselfers. Excel users need to use VBA.

As an aspiring R user, I needed help from LearningR to get the R script needed to automatically update the GISS monthly data. Thanks again to LearningR for the help.

Here’s my trend chart with decadal trend rates:

The R script for this automatic chart generation is doing quit a bit:

- Reading the raw GISS data file
- Extracting monthly data from 1880 to most current value
- Plotting overall data
- Calculating overall trend, plotting trend line and displaying rate
- Calculating decadal trend rates, displaying color coded trend lines for decade, and displaying color coded trend rate

Here’s a link to my R script.

I’m sure that my LearningR colleague will follow up with a ggplot2 version of this chart. Are there any Excel users who want to reproduce/ enhance this chart in Excel?.

Pingback: Understanding the Science of CO2’s Role in Climate Change: 1 – Introduction « Climate Charts & Graphs

I need to calculate decadal trends to monthly air temperature and precipitation since 1965 to 2008. I can to use this script or same? please let me know.

Best regards

Juan

I have included a link to my R script. You are welcome to use it. I suspect you will need to make some modifications to handle both air temperature and precipitation.

Let me know if you run into any difficulty with my script.

Pingback: Segmented Regression of GISS Temperature Trends « Charts & Graphs

Pingback: ggplot2: Decadal Trend Rates in Global Temperature « Learning R

I think Jon’s point about smoothing makes sense. By showing piece-wise lines by decade (that are pretty noisy), you push the viewer into thinking that there are meaningful subtrends, and therefore a spline smoothed model or lowess regression might make more sense.

The big green line pushes the idea that this is an inexorable trend, but the shorter segments undermine the hypothesis that a periodic trend could not revert.

The decadal trends provide a useful way to show calculated trend rates over time. A lowess trend is great, however, it does not provide the actual trend rates over time.

It would be interesting to see the change point analysis of this data. I’ll have to read up on this.

Actually, I didn’t mean to imply that the linear trendline wasn’t useful. For my own use, I usually fill up a couple worksheets with different ways to look at data and different ways to try to make sense of relationships.

A rough eyeball of the raw data in your chart even hints that there might be four straight segments (1880-1920, 1920-1940, 1940-1975, 1975-2009).

I’m working my way up to change point analysis (CPA), an analytic method to identify changes in either mean or variation in a series. Here’s a link to an Excel post I made on this a while back.

CPA is based on CuSum analysis, a well know SPC technique.

Stay tuned for additional posts on this.

Are the lines color coded by slope? That’s a nice touch.

The single regression for the entire data set (the green line) does not appear to be too great a fit. This is where a quadratic fit might be worthwhile. Or some kind of smoothing (LOWESS or moving average).

The overall trend line shows a slope of 0.565 oC per century. The adjusted r-squared is 0.627, pretty good from my standpoint.

This means that 62.7 % of the temperature anomaly variation is explained by year – month. Considering the inter decadal variations, the trend is very strong.

I find smoothing techniques useful to give a sense for trends, trend lines are useful to get a measure of the rate of change.

The decadal trend line and rate note color coding (red for increasing, blue for decreasing) helps to give a quick visual sense of the shifting pattern in decadal trends.