The two global radiosonde datasets used widely are RSS and UAH. While they use essentially the same raw data, they apply different statistical techniques to find the global average temperature. Because of this, RSS shows a greater global temperature tend (.17 C/decade) than UAH (.13 C/decade). This difference is mainly due to different tropical temperature records.
Out of curiosity, I graphed RSS minus UAH from 1979 to 2008, and I got something verryyy strange. I was expecting a linearly upward trend (RSS would become warmer and warmer each year than UAH). Instead:
There are some things to note.
1. The difference between RSS and UAH swings wildly back and forth from 1979 to 1990. This may be because the two datasets were still being tinkered with and adjusted.
2. There appears to be a step sometime during 1992 during the course of two months. Perhaps RSS adjusted upward or UAH adjusted downward. Or perhaps there was tinkering at all.
3. Around 1995, we begin to see an annual cycle in the difference between the two datasets, and that cycle becomes very well defined post-2002. This means that for some reason, seasons are impacting the temperature records in one or both datasets, which seems like it wouldn’t be a good thing.
4. Secondly, from 2003 to the present, the two datasets have been converging in a very linear manner aside from the strange season-related cycle. In fact, the convergance looks so linear and perfect that it seems as if RSS and UAH have reconciled their differences (besides the season-related cycle) and that an adjustment is being done each year on one of the datasets – either warming the UAH record or cooling the RSS record.
This simple analysis of subtracting one dataset from another turned out to be very informative. When you have two datasets intended to describe the same phenomenon, it is often helpful to graph the difference between the two over time, because most of the actual signal is removed, and the raw differences between the datasets over time are shown. Then we may be able to attribute causes to the differences, possibly reconciling the datasets and shedding light on possible problems with datasets.
I’ve graphed the difference over time between RSS and UAH for the NH, SH, and Tropics now. The seasonal-cycle is evident in the NH and less so in the SH. The tropics also contain the seasonal cycle, which is especially confusing. The recent convergence of the two datasets seems to be mostly due to changes in SH data from one or both of the datasets. Tropics data has contributed less to the convergance and the NH contributed barely at all. The clear RSS bias toward warming due to a difference in numerical analysis in the tropics is shown. This RSS bias toward more warming than UAH, the validity of which I can’t comment on, seems to have stopped since 2003.
It appears that the convergence of UAH and RSS might be due to series of corrections made by the UAH team (Spencer and Christy), adjusting the UAH record upward. Or maybe I’m reading the readme wrong. Or maybe I don’t understand diurnal drift. But this may be an explanation for the convergence.