Differences between UAH and RSS TLT temperature trends:
Several comments on the difference between UAH and RSS have come to light. The difference in the 36-year global TLT trends are small with UAH about +0.017 C/decade warmer than RSS. The main reason for this is our different approaches in correcting for east-west satellite drift (i.e. diurnal drift.) UAH uses an empirical method based on actual satellite readings to correct of this while RSS uses a climate model estimate. In the period from 1979 to about 2000 the main effect of diurnal drift led to a spurious cooling of temperatures, so a correction needs to be applied to warm them back up. After 2000, the opposite occurs in which the main satellite (NOAA-15) drifts to warmer temperatures, so these must be cooled back down. The UAH corrections are smaller than RSS, so relative to UAH, RSS warms up the data in the period 1979-2000 more than UAH does. After 2000, UAH does not apply a diurnal correction because the main satellites we use were non-drifters, however, there is still a slight spurious warming in UAH due to our necessary use of NOAA-15 which we have not cooled back down (v6.0 will have this fixed). Since RSS has a relatively large correction to cool off the post-2000 data, their time series drops quite a bit relative to ours and to radiosondes. This was examined in several places, e.g. Christy et al. 2011, Int. J. Remote Sens. Since UAH has a bit too much warming after 2000 and RSS too much cooling, I’ve advised folks to simply take the average of our datasets for the best estimate. I’m the author of the upper air temperature section for the annual BAMS report on climate. I write that the 1979-2014 global TLT trend is +0.13 C/decade +/- 0.02 C/decade where this error range represents measurement error as it encompasses all of the estimates from radiosondes and satellites and even ERA-I. There is also statistical uncertainty which I calculate as +/- 0.06 C/decade using a reduction in degrees of freedom due to autocorrelation of the annual anomalies (N = 36, but Neff ~ 32). As an added note, the mean global TLT trend of 102 CMIP-5 RCP4.5 models for 1979-2014 is +0.27 C/decade with a standard deviation of +0.05 C. While this is significant, it becomes highly significant when considering TMT where the average model trend is over three times that of the observations from 1979-2014.