For data processed using the early weekly calibration table updates, the main limit on the absolute accuracy appears to be drift between the calibrations.
Gibson (1984) gives uncertainties of 
2K to 4K at the end of an inter-calibration interval, with the lower value in winter and the higher value in summer.
This effect is presumably related to the changing mean solar zenith angle at the satellite in the course of a year. A further 4K uncertainty can arise during
three weeks about each equinox when the satellite is eclipsed by the Earth during the night, and in these periods, the data may be unusable (Gibson 1984).
Data processed using the scan-by-scan calibration are much more reliable, having a relative accuracy of 0.5K, which may sometimes be limited by the digitization
truncation level.(a) Absolute accuracy is more difficult to specify but is believed to be about 1.0-1.5K (Personal communication with W. P. Menzel, 1991.),
after the removal of bias errors by daily comparisons with channel radiances calculated using the atmospheric profiles of the NMC weather forecast model. (Menzel,
W.P. 1990. Summary of the VAS calibration. Unpublished.)
A small number of case studies comparing VAS radiance measurements with those from the HIRS (High Resolution Infrared Radiation Sounder) on the NOAA-6 polar
orbiter produced "surprisingly good" (sic) results (Menzel et al. 1981) with the VAS 15-
m channels (2, 3, 4) showing a consistent negative bias of up to 2K
and the VAS 12.7-m and 7.3-m channels (7, 9) showing a positive bias of up to ~5K. These discrepancies include uncertainties that are due to the HIRS calibration,
slight differences between the VAS and HIRS channel transfer functions, and lack of absolute simultaneity (up to 90 min) and collocation (up to 20 km) of
the HIRS and VAS measurements.
A further comparison was made between the VAS radiances and those predicted by an atmos-pheric radiation transfer model using atmospheric profiles of temperature and humidity measured by radiosondes (Menzel et al. 1981). Eighty cases were studied and the differences in brightness temperatures were ~1.5K, except for the "noisy" Channels 1, 2, 9 and 11. Channels 2, 3, and 4 were consistently lower than the corresponding values predicted from the radiosonde data. Again, these discrepancies include contributions from uncertainties in the radiosonde measurements and possible lack of simultaneity and collocation.
Montgomery and Uccellini (1985) present a review of the utility of the VAS meteorological products and their validation by comparison with independently derived profiles and fields.
