Global Positioning System (GPS) radio occultation (RO) currently represents the only self-calibrated SI traceable raw satellite measurements (Anthes et al., 2008; Anthes, 2011). The fundamental observation is time delay of the occulted signal’s phase traversing the atmosphere. The time delay is a function of several atmospheric physical state variables. Subsequent analysis converts the time delay to temperature and other parameters, which inevitably adds some degree of uncertainty to the derived temperature data. Intercomparisons of GPS-RO products show that differences are largest for derived geophysical parameters (including temperature), but are still small relative to other observing technologies (Ho et al., 2012). Comparisons to MSU and radiosondes (Kuo et al., 2005; Ho et al., 2007, 2009a, 2009b; He et al., 2009; Baringer et al., 2010; Sun et al., 2010; Ladstadter et al., 2011) show substantive agreement in interannual behaviour, but also some multi-year drifts that require further examination before this additional data source can usefully arbitrate between different MSU and radiosonde trend estimates.

Atmospheric winds are driven by thermal gradients. Radiosonde winds are far less affected by time-varying biases than their temperatures (Gruber and Haimberger, 2008; Sherwood et al., 2008; Section 2.7.3). Allen and Sherwood (2007) initially used radiosonde wind to infer temperatures within the Tropical West Pacific warm pool region, then extended this to a global analysis (Allen and Sherwood, 2008) yielding a distinct tropical upper tropospheric warming trend maximum within the vertical profile, but with large uncertainty. Winds can only quantify relative changes and require an initialization (location and trend at that location) (Allen and Sherwood, 2008). The large uncertainty range was predominantly driven by this initialization choice, a finding later confirmed by Christy et al. (2010), who in addition questioned the stability given the sparse geographical sampling, particularly in the tropics, and possible systematic sampling effects amongst other potential issues. Initializing closer to the tropics tended to reduce or remove the appearance of a tropical upper tropospheric warming trend maximum (Allen and Sherwood, 2008; Christy et al., 2010). There is only low confidence in trends inferred from ‘thermal winds’ given the relative immaturity of the analyses and their large uncertainties.

In summary, new technologies and approaches have emerged since AR4. However, these new technologies and approaches either constitute too short a record or are too immature to inform assessments of long-term trends at the present time.