May 27 2008
Comparing the time series of the major climate metrics
I’ve blogged before about the differences in the two satellite tropospheric temperature records. There are two major features of the difference time series: one is an apparent 1-year periodicity, and the other is a step in the year 1992. When calculating trends, the first has only a small effect. Unless you’re calculating trends for less than a few years, which would be silly. This post will take a look at the other issue. What are the implications for the change point in 1992.
An interesting side note is that both RealClimate and ClimateAudit have recently blogged about change points, and this post was written before before I had read them. Talk about blogospheric voodoo.
In case you haven’t seen the difference time series before, and didn’t click on the first link, I’ve included the graph below.
|
The red line is simply the mean of the time series before and after the prescribed change point. As was noted in previous points, I’m assuming the change occurred in January 1992. Visually this looks good, which is good enough for the Internet, right?
It is important to resolve the 1992 discrepancy because this accounts for almost all of the difference in trends between the RSS and UAH data. Since the beginning of their records (1979), the UAH trend has been 0.137 C/decade, while the RSS trend is 0.177 C/decade. By simply adding the difference between the two time series to the UAH data after the 1992, the UAH trend becomes 0.178 C/decade. Similarly, if the difference was added to the RSS temperatures before 1992, the RSS trend would decrease to near 0.137 C/decade (not actually calculated).
|
This plot show several temperature difference time series. The left column shows a climate metric subtracted from the UAH temperature, and the right column shows a climate metric subtracted from the RSS temperature, where the 3 climate metrics (GISS, Hadley, and NCDC) are represented by the rows.
It should be obvious that there is a lot more noise in the above 6 plots compared to the UAH-RSS data. There are no visually apparent steps in the data. As before, the red line shows the mean values before and after the prescribed change point. Notice that there is a larger change in the mean difference for UAH compared to RSS. Also, UAH has a negative step whereas RSS has a positive step. This means that the recent temperature from UAH are lower compared to the other 3 metrics, and that the recent temperature from RSS is higher when compared to the other metrics.
The table below shows a quantitative measure of the differences between the time series.
|
These values were calculated by taking the mean value of the difference time series before the change point and subtracting the mean value after the change point. For instance, the upper right value in the table (0.0425) means that the mean value in the UAH-GISS temperature difference time series was 0.0425 Celsius greater before 1992 than after.
I’ve used the average differences to compute “fixed” values for both RSS and UAH. Note that this amounts to changing the trends so they are equal to the average trend of the other 3 temperature metrics.
|
(No Ratings Yet)
Loading ...Related Posts:
5 Responses to “Comparing the time series of the major climate metrics”
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.
Nice post.
One point about trend calculations with the satellite time series: the residuals from the trend fit are highly autocorrelated. At a minimum, I think that such calculations should use a GLS fit allowing for autocorrelation in the residuals. The impact on the trend itself is fairly slight, but the impact on the standard error of the slope estimate is substantial. Check it out.
Agreed nice work Atmoz.
does this mean discusions get to switch from PCA to change point analysis?
I tried to visually compare the UAH-RSS chart against the charts you showed on Lucia’s blog for land and ocean differences. The step doesn’t show up or doesn’t appear as obvious as the land/ocean chart step does. Could the difference be in the way they merge land/ocean data?
[Reply: It could be. Although commenter 'Robb' on Lucia's site suggested that it's because of the way they merge data from different satellites, which makes sense too.]
Why, oh why do people insist on drawing straight lines through this data?
It is obvious looking at the chart that something unusual happened in 1998. A huge amount of heat was put into the atmosphere.
A perfectly good analysis could be that a straight line could be drawn to the “Big Bump” and another straight line drawn after roughly 0.3 Deg C higher. The atmosphere does not react to change in a linear straight manner. ie I agree with Bob Carter.
I came across the TLS RSS data the other day. I don’t know how to post a graph of it, but it is obvious that the reaction of the stratosphere to the two volcanoes results in step like changes. On the RSS webpage they draw a straight line through the TLS chart - ridiculous!
Here are the charts and data from RSS.
http://www.remss.com/pub/msu/monthly_time_series/
http://www.remss.com/msu/msu_data_description.html#msu_amsu_trend_map_tls
Also you might want to check out what William Briggs had to say about the RSS data.
http://wmbriggs.com/blog/
It will be interesting to see in the next year or so if the Atmosphere loses the heat of 1998. Hey, that’s my hypothesis and this is climate science!
[Reply: There's so much wrong with this post it's hard to know where to begin. But if you don't think the atmosphere has "lost" the heat of 1998 yet, you are either intentionally misinterpreting the graph, or are not qualified to interpret graphs. The temperature in the 80s were lower than in the 00s. The linear trend gives an estimate of how much it has increased.]
Atmoz,
Note that the comment regarding the heat loss of the atmosphere was tongue in cheek. Although somewhat sardonic.
However, I stand by my opinion regarding the veracity of using Excel Trendlines with satellite temperature anomaly data.
In this, I believe I am supported by two, I guess equally chart challenged, PhDs.
1. Bob Carter, in his video on You Tube (http://www.youtube.com/watch?v=vN06JSi-SW8 - 166 sec mark), showing the satellite data with the stepwise interpretation.
and
2. William - “The gist is that the ordinary regression line is inadequate and we have to search for something better” - Briggs at the above mentioned website. Search under RSS. If you read this excerpt, I think you will find it enlightening.
So, if we should perhaps look at the data somewhat differently ( and that is all I am advocating), how do we interpret this seemingly stepwise changes indicated by Carter’s interpretation and Brigg’s wavelet analysis? Including the TLS data. Seems there is a lot more going on here than meets a simple regression.