De-Trending UAH

The University of Alabama, Huntsville Microwave Sounding Unit (UAH MSU) atmospheric temperature data is supposedly the gold standard for global mean temperature. Since I am playing with Open Office trying to get it to do some time series stuff, I detrended the Northern Extent, Tropics and Southern Extent data sets.

The method I used is cludgy, but appears to work fairly well. I add the maximum trend or slope of the individual data series to the start of the series and incrementally decrease the the addition per data point to the end of the series. In this case, there are 399 data points, the first has the slope times 399/399 added to the first and the slope time 0/399 to the final point. This increases the mean of the series by the slope, which I subtract to return the plot to zero. Using the linear regression and the mean value functions in the charting program, the regression after de-trending is equal to the mean which is equal to zero as a check. Obviously, the complexity of the system dynamics would not allow perfect removal, but a good portion of the internal variability would be reduced by subtracting the "average" from each series.

Here is the results of subtracting the "average" from each series and removing the de-trending. The final slope of the Tropics and Southern extent is 0.007 degrees C per year or 0.07 degrees C per decade. This should be the slope of response to continuous forcing change. The slope of the Northern Extent in this case is greater, 0.026 C per year or .26 degrees C per decade. This should be due to water vapor enhancement in the high northern latitudes, amplification from the oceans thermohaline current and land use change amplification of CO2 or other continuous forcing.

In this plot I change the start of the comparison to 1995. There is a minor change in slope that is unlikely to be statistically significant. This is not a proof of the method, but tends to lend some credibility since there is a much larger change in slope using the raw data with the internal variability included. Since there is some speculation that cooling started circa 2000 here is that plot.

In this case there is a statistically significant change in the Tropics and Southern Extent with no significant change in the Northern extent. The length of the series is only 11 years, so there cannot be a great deal of confidence in the significance, but it is slightly over 50% likely to be a shift in the climate, if the method is valid. Solar forcing is the only likely cause, so there may be more solar impact than generally thought in the global oceans which make up the majority of the Tropical and Southern extent surface.

Just to be complete, this is from 2002. This is much too short to be of any significance, but it is comforting to see there is no drastic changes from the 2000 start as far as the methodology is concerned. I am not going to attempt to derive some validation of the method at this time. In the future I may use the same procedure for other regions comparing land to oceans which may help either discredit the method or help determine the degree of land use impact. What will happen, I don't know, but it should be a reasonable way to remove most of the internal variability without trying to specifically target the individual causes. Then again, it could be a waste of time. Initially though, it appears to agree rather well with the Douglass and Christy "Limits on CO2 Climate Forcing from Recent Temperature Data of Earth" results published in 2009 with the addition of a little hint of solar impact. Note: I am sure this method has probably been used before, but I just developed what I am doing here on my own. If anyone has a link to the original development, if it is indeed a valid method, let me know. UPDATE: http://phys.org/news/2012-05-satellite-global-climate-closer.html There will be some adjustments to UAH which will be interesting since while it should increase the overall slope is should also increase the lower slope from 2000. I will revise this when the new data is available. UPDATE 5/9: Since the death of UAH MSU seems to have been announced in the climate world, I thought I would check if the reports are accurate.

Using the same method on RSS, the results are seriously different. Since I had the spread sheet for the UAH detrend minus common signal, I plugged in the RSS tropics and extents using the UAH common trend. This should highlight where the two sets differ.

There is a swoop curve, so the differences are near the start and end. For the start, early in the satellite program, is not all that exciting to see less than stellar performance. It's a learning curve thing ya know. The end though is a bit unusual, by now things should have been improving. So what's up? Notice that the RSS detrended has very pronounced change in the orange, tropics plot. The blue, northern extent plot is subdued. Since these are plotted with the average of the three detrended, the most subdued produced the strongest common signal. So the Northern extent would be driving climate according to the RSS data. While that may sound nifty, the ocean heat content is no piker in the climate game. The tropics trend is negative for the RSS with detrended variation removed, which goes counter to what I would expect. So I may have screwed up or there may be some unintentional bias in the RSS data. So I will go back to check to see how bad I screwed up, but I really suspect the RSS calibration is not consistent with the satellite changes. UAH is also likely to have issues, but it looks like any errors they may have made were consistent, which would mean the data is still useful. Found a issue in the spread sheet so the chart above is updated. Since the Tropics are mainly the issue.

This chart shows the RSS tropics in green with the average of the tropics and extents, detrended, removed. There is no change to the slope, but there may some interesting things.

That is the same plot using the UAH data. In the UAH there is a difference in the slope between the raw and the raw minus the detrend average. There is virtually no difference in the RSS tropics data. Which one makes more sense? Hard to say, but the no extent and so extent have larger changes in RSS. As a note, the GISS surface data is very similar to the RSS when I use the same procedure. So similar, that RSS may have fudged their calibration a touch to more closely match GISS. Now that would generate a little buzz in the remote sensing community :) Still, UAH may be high early but RSS looks low late. Time will tell.