Skyhunter said:
Last April is an interesting month.
The http://www.ncdc.noaa.gov/oa/climate/research/2009/apr/global.html#temp" ranks last April as the fifth warmest on record. The lower troposphere (<8 kilometers) however is ranked eleventh by RSS and fifteenth by UAH.
Why the large discrepancy?
Should we expect the surface temperature (< 1.5 meters) ranking to correlate somewhat with the lower troposphere?
You should expect a fairly strong correlation, but with some significant differences. It is, after all, a different thing being measured; not merely different ways of estimating the same thing.
Note that the rank order of data points is of very little use for looking at correlations.
One major difference between atmosphere and surface is that air temperatures respond much faster to any transient effect. The surface record is dominated by the ocean, where there is a powerful lag from the huge heat capacity of the oceans. Put simply, the surface record looks a lot like the lower atmosphere, but with additional damping. Another more subtle source of difference is changes to lapse rate as the Earth's climate is shifting state. This changes the gap between surface temperatures and atmospheric temperatures.
Finally, you need to keep in mind the nature of the data we are looking at. It's very common to present summaries of the data and the trends for a legitimately interested public, in which there is a single global "temperature" value, changing with time.
In fact, the underlying datasets are thousands of different temperatures, from all over the planet. There are two major steps in constructing a single combined value. First, you convert from "temperature" to "anomaly". Then you average up all the anomalies to a single global value. This is quite a robust procedure, but it does mean there are small differences obtained depending on how these steps are performed. There are several groups around the world involved in producing global anomaly datasets, working independently of each other but sharing much of the same underlying data. The differences between calculated results are generally fairly small, but it is enough to alter "rankings" of individual months.
As an added complication when dealing with the atmosphere, there are all kinds of instrument difficulties which lead to much larger uncertainties than a surface measurement. The main sources of data for the atmosphere are radiosondes (weather balloons) and satellite data. Each has their own unique sources of error.
Finally, the satellite record used by RSS and UAH is the focus of a long running and still unresolved question of how to combine values from different satellites into a single record. This is something to watch for the future. I suspect we may see progress on resolving discrepancies to show up in the literature within the next year or two.
Looking at the data
The NCDC pages have links for obtaining data. Unfortunately, the surface data from NCDC is from their ftp site that I have been unable to access for a few months now. I don't know why. But I can get the GISS and HadCRUT3 surface data. There are small differences between these surface records, and the NCDC surface dataset tends to lie somewhere between them. The differences are not significant in the sense that they are less than the uncertainty bounds with each dataset. There is a clear explanation of the reasons for differences in the
Frequently Asked Questions at Met Office Hadley Center (UK). It's near the bottom, as an answer to "
What are the differences between HadCRUT3, GISS and NCDC global temperature analyses??
There are significant differences between surface and lower troposphere anomaly data, since this is a genuinely different quantity. The overall correlation remains strong. The atmospheric data can be obtained using links given at another NCDC page: http://www.ncdc.noaa.gov/oa/climate/research/msu.html (bottom of page).
Here's a plot of the datasets. I've shifted the two surface datasets vertically by adding 1 to GISS, and 2 to HadCRUT3. This let's you see the overall shape.
Correlations are:
\begin{array}{l|rrrr}<br />
\mbox{all months} & \mbox{HadCRUT3}& \mbox{UAH-LT}& \mbox{RSS-LT}\\<br />
\hline<br />
\mbox{GISS}& 0.88& 0.82& 0.81\\<br />
\mbox{HadCRUT3}& & 0.81& 0.87\\<br />
\mbox{UAH-LT}& & & 0.95<br />
\end{array}
You are looking simply at April. Looking at a particular month can help get around issues of how seasonal variation is handled, but it also means you look at much smaller datasets. Anyhow, here is the data from April only.
Note that values for 2009 are still preliminary -- and that HadCRUT3 does not include April in their online dataset. Here are correlations for the April timeseries:
\begin{array}{l|rrrr}<br />
\mbox{April}& \mbox{HadCRUT3}& \mbox{UAH-LT}& \mbox{RSS-LT}\\<br />
\hline<br />
\mbox{GISS}& 0.86& 0.78& 0.79\\<br />
\mbox{HadCRUT3}& & 0.83& 0.9\\<br />
\mbox{UAH-LT}& & & 0.97<br />
\end{array}
Here are a few observations about this data.
- The RSS and UAH lower troposphere figures show the highest correlation. That is because they are working from the same underlying satellite data. The differences are due to methods for obtaining a lower troposphere temperature from the microwave sounding units, and especially to how they merge data from different satellites with different calibration errors. These differences are a focus of a lot of work in the various groups working with satellite measurements.
- The biggest difference between RSS and UAH that you can see in these plots is that after about 1992, there is a jump, with UAH returning lower values than RSS. It's not the only difference, but this is the main cause of different rankings for April using these two tropospheric datasets.
- The correlation of GISS and HadCRUT3 is strong, but not as strong as the troposphere correlations. This is mainly because of differences in how they handle parts of the globe with poor data coverage. HadCRUT3 tends to leave it out. GISS attempts an interpolation. Over long time spans this tends to even out, but it does mean some short term variation in the two datasets.
- The strongest correlation between surface and lower troposphere is seen with HadCRUT3 and the RSS data.
- The troposphere data does not get any input from the poles, due to the satellite orbits used. This is also the region with the lowest levels of coverage for surface temperature records. The HadCRUT3 dataset therefore tends to omit information from the poles, and this might contribute to a better correlation of troposphere temperatures. But we can't be sure until people sort out the reasons for the differences between RSS and UAH.
Cheers -- sylas
