Jul 17 2008
A Small ‘Problem’ with GisTemp
Yesterday I was messing around with the temperature time series, again. It’s kind of a fun thing to do, even if it is mostly a big waste of time. This time it was triggered by a post by a post at Climate Progress, Eighth warmest June on record means Great Ice Age of 2008 is STILL over. The monthly report that was released recently was published by NOAA, not NASA. While I started looking at the NOAA data, it became evident that the NASA data was more interesting. I wanted to present the temperature data in a way besides the typical time series.
First I needed a couple sample time series that looked similar to the temperature time series that we’re all familiar with. I’ve plotted those below. The top panels and bottom left panel are time series with a prescribed slope but with differing levels of white noise. The bottom right panel is a time series with no slope, with a 12 ‘month’ periodic signal and white noise.
 |
(All plots linked to bigger versions.)
The purpose of these time series is to test the visualization to see if the results from random time series look similar to those from the real temperature time series. The method used to present the temperatures is to see if temperatures have been declining into a new ‘ice age’. To do this, I sorted the temperatures. I then plot the date as a function of index, where the index is the n-1th warmest month. For instance, the date associated with the zeroth index has the warmest monthly temperature anomaly. This is the blue curve in the plots below.
The yellow curve is the temperatures sorted by temperature. This was mostly done to make sure the sorting of the dates was done correctly, but it also offers an easy way visualize that both the temperatures and dates are rising in a positively correlated manner.
 |
Notice that when there is little noise in the temperatures (top left panel) that the warmest month will always be the most recent month. As more noise is added to the temperature time series (top right, bottom left), the blue date curve has more noise added as well. But there is still a positive correlation between the date and temperature, as expected.
Now let’s look at the 3 climate metrics that span back to 1880, GisTemp, HadCruT, and NCDC. I’ve also plotted the average of the 3 climate metrics in the graph below. The panels are not labeled to show just exactly how similar the 3 metrics are to each other. It would be difficult, but not impossible, to distinguish between the 3 based just on this graph.
 |
The horizontal axis here is kept the same as those above, and is not in years but in months since the first date.
 |
Here are the sorted date plots. The configuration of the panels is the same as the graph above, if you want to see if you correctly identified them. What should be apparent is that approximately the last 100 months have been the hottest 100 months on record. There are a few months that deviate from the hot trend, such as the early months of 2008. However, to say that we are in an ‘ice age’ would be silly.
The ‘problem’ with the GisTemp record is that there appears to be a pseudo-periodicity when the temperatures are sorted. I tried to identify what period it was, but was unable to do so. Hence the pseudo-periodicity language in the previous sentence. But it’s obvious that this pattern is not reflected in either of the temperature time series produced by NCDC or the Hadley Centre.
When all three are plotted on the same graph, it’s not immediately clear that GISS is different than the other two. Again, there is no legend to show that visually, they appear almost identical.
 |
GISS are the black points, and after closer inspection it appears that most of the GISS values are clumped in 3-5 point groups unlike either NCDC or Hadley. I have no idea if this is a real problem or just an artifact of sorting. However, I was not able to reproduce it using either white, red, or periodic noise [not shown].
(1 votes, average: 1 out of 5)
Loading ...Related Posts:
13 Responses to “A Small ‘Problem’ with GisTemp”
To reduce spam, comments are automatically closed 30 days after the last comment. If you would like to comment on any closed thread, please use the contact form at the top of this page.
Looks to me like it’s just quantization — the Gistemp data are only 2 significant figures, the others (Hadcrut at least) have 3. Your periodicity is the gaps between temperatures.
“I then plot the date as a function of index, where the index is the n-1th warmest month.”
So you’re plotting dates against temperature ranks — is that right? (The explanation is a bit confusing, so I just want to make sure.)
If so, that gives me some ideas. I’m thinking along the lines of flipping it around, and plotting ranks against dates. Or plotting signed ranks against dates, where the signs are taken relative to a hypothesized trend or flat line…
Correct. I thought about plotting it the other way around - ranks against dates, but thought it would be too confusing.
OT: One shell game practitioner identifies another. Others are slower on the uptake.
Since 1990 an increasing number of stations have gone “missing”, and it seems an increasing number of records for existing stations have gone missing as well. See http://www.climateaudit.org/?p=2703.
GISS essentially uses monthly, seasonal, and annual averages to infill the missing data. I think this process will reinforce the periodicity you are seeing.
John Goetz: I think it’s the inactivists who tried to make the data go away and then pretend that it’s GISS’s fault. So if anything, your explanation should be flipped around.
Looking at the graphs after hom. adjustment on IJI”S website, I noticed that Marysville is on average about 2 degrees warmer. I’m sure the hom. adjustment gets a chunk of the problem, but not nearly oll of the problem.
Mike C:
The homogeneity adjustment (which gets a chunk of the problem, in any case) is just part of the story.
The other part — answering Goetz’s comment — is that Watts’s surfacestations.org home page mysteriously decided to start the (pre-adjustment!) Marysville graph after 1900, even though data before that were also available. Someone’s trying to make some data go away, and it’s not the GISS.
I’m still not sure I can wrap my mind around what the quasi-periodicity may mean. Literally, it’s saying that certain ‘phases’ of the ranks correspond to certain intervals in time, but I’ll be blessed if I can find a useful interpretation of that.
(Perhaps, as Josh said, it’s indeed just an artifact produced by sorting temperature values with 2 decimal places.)
The range of GISS temp is from -0.87 to +0.86 to two decimal places and gives only 174 unique results. Since there are over 1500 data points it is obvious that there are going to be lots of ties when the results are put in rank order.
Perhaps you could do a few runs adding a random number to three decimal places between 0.000 and 0.009 to each element and plotting the new graph. If the clumping does not disappear then you will have found something interesting.
Bi ITI,
That is because Anthony downloaded that graph before GISS switched to the new data set a short while back.
Mike C:
Interesting… I stand corrected then.
Hey,
I’ve just been banned from ‘Open Mind’ for expressing doubts about GSMT and GISS coverage.
What does that say about scientists open minds?
Dave Andrews:
Well, I was banned from Free Republic for writing facts about Gore’s energy usage. I just had to rewrite it on my own blog; no big deal. You should try that some time.