I thought it would be useful to again outline the basic reasons why different satellite global temperature datasets (say, UAH and RSS) produce somewhat different temperature trends.

They all stem from the fact that there is not a single satellite which has been operating continuously, in a stable orbit, measuring a constant layer of the atmosphere, at the same local time every day, with no instrumental calibration drifts.

Instead, what we have is multiple satellites (we use 14 of them for the UAH processing) with relatively short lifetimes (2 to 16+ years), most of which have decaying orbits which causes the local time of measurement to slowly change over the years, slightly different layers sampled by the earlier (pre-1998) MSU instruments compared to the later (post-1998) AMSU instruments, and some evidence of small calibration drifts in a few of the instruments.

An additional complication is that subsequent satellites are launched into alternating sun-synchronous orbit times, nominally 1:30 a.m. and p.m., then 7:30 a.m. and p.m., then back to 1:30 a.m. and p.m., etc. Furthermore, as the instruments scan across the Earth, the altitude in the atmosphere that is sampled changes as the Earth incidence angle of view changes.

All of these effects must be accounted for, and there is no demonstrably “best” method to handle any of them. For example, RSS uses a climate model to correct for the changing time of day the observations are made (the so-called diurnal drift problem), while we use an empirical approach. This correction is particularly difficult because it varies with geographic location, time of year, terrain altitude, etc. RSS does not use exactly the same satellites as we do, nor do they use the same formula for computing a lower tropospheric (“LT”) layer temperature from the different view angles of AMSU channel 5.

We have been working hard on producing our new Version 6 dataset, revamping virtually all of the processing steps, and it has taken much longer than expected. We have learned a lot over the years, but with only 2-3 people working part time with very little funding, progress is slow.

In just the last month, we have had what amounts to a paradigm shift on how to analyze the data. We are very hopeful that the resulting dataset will be demonstrably better than our current version. Only time will tell.