Since AR4 there have been a large number of intercomparisons between radiosonde and MSU data sets. Interpretation is complicated, as most studies considered data set versions that have since been superseded. Several studies compared UAH and RSS products to local, regional or global raw/homogenized radiosonde data (Christy and Norris, 2006, 2009; Christy et al., 2007, 2010, 2011; Randall and Herman, 2008; Mears et al., 2012; Po-Chedley and Fu, 2012). Early studies focussed on the time of transition from NOAA-11 to NOAA-12 (early 1990s) which indicated an apparent issue in RSS. Christy et al. (2007) noted that this coincided with the Mt Pinatubo eruption and that RSS was the only product, either surface or tropospheric, that exhibited tropical warming immediately after the eruption when cooling would be expected. Using reanalysis data Bengtsson and Hodges (2011) also found evidence of a potential jump in RSS in 1993 over the tropical oceans. Mears et al. (2012) cautioned that an El Niño event quasi-simultaneous with Pinatubo complicates interpretation. They also highlighted several other periods of disagreement between radiosonde records and MSU records. All MSU records were most uncertain when satellite orbits are drifting rapidly (Christy and Norris, 2006, 2009). Mears et al. (2011) found that trend differences between RSS and other data sets could not be explained in many cases by parametric uncertainties in RSS alone. It was repeatedly cautioned that there were potential common biases (of varying magnitude) between the different MSU records or between the different radiosonde records which complicate intercomparisons (Christy and Norris, 2006, 2009; Mears et al., 2012).
In summary, assessment of the large body of studies comparing various long-term radiosonde and MSU products since AR4 is hampered by data set version changes, and inherent data uncertainties. These factors substantially limit the ability to draw robust and consistent inferences from such studies about the true long-term trends or the value of different data products.