A cross-spectral analysis method is developed to estimate the noise variance spectra of three or more independent measurement systems observing the same input. The noise is modeled using a first-order autoregressive process. Estimates of the two process parameters are used to determine confidence limits for system noise that can be placed on the observed data.
The method has been applied to dew-point data collected in aircraft intercomparison flights and a research flight in the National Hail Research Experiment. The six dew-point systems used were manufactured by the same company and operate by electrically cooling a metal mirror until a film of water vapor or frost is optically detected on the mirrored surface. The ratio of the signal to noise variance was found to vary between about 10:1 and 100:1 at the origin and decrease to zero between about 0.15 and 0.3 Hz, both properties dependent on atmospheric conditions and the structure of system noise. Since the data were collected once per second, appropriate filtering and decimation could be performed for archiving purposes with a 50% space savings and small loss of “information.” Ninety-five percent confidence limits on the filtered observed data with a cutoff at 0.2 Hz vary from ±0.5 to ±0.1°C, with the latter figure most representative of those computed.