Title: Uncertainty propagation for NIST visible spectral standards
Abstract: Radiometric calibrations of spectral responsivity or spectral irradiance at particular wavelengths are required for a number of applications.Standards for these quantities are provided to external customers and used internally in deriving reference values for such quantities as color, correlated color temperature and distribution temperature.Most of the national laboratories provide such calibrations in spectral ranges from the ultra-violet to infrared regions.Primary standards in radiometry are increasingly detector-based, traced ultimately to power or irradiance measurements made with a cryogenic radiometer.While some laboratories operate such devices at the exit of a monochromator, or with laser-based sources tunable over a wide wavelength range, it is both time-consuming and expensive to cover all the wavelengths at which reference measurements may be required; most national metrology institutes provide measurements at a limited number of wavelengths and then apply either interpolation or fitting techniques to cover the complete range.New values generated from the limited set are then correlated through dependence on the common input set.A thorough propagation of uncertainties requires this correlation to be taken into account when spectral values are combined, a common practice in radiometry.Accurate estimate of the uncertainties resulting from the assumptions made in modeling the measurement and transfer process, rather than an over-estimate, increases the chance of detecting systematic components that may have been overlooked.The National Institute of Standards and Technology (NIST) methods and procedures in radiometry and photometry are described in detail in the Special Publication SP250 series [1][2][3].Some of the descriptions of the traceability paths are now dated, and none provide an estimate of the partial correlations in the reported spectral quantities required for a modern uncertainty calculation when combining values in spectral integrals.