Abstract
In 1988, the algorithm for retrieving temperature soundings from radiances measured by NOAA's polar-orbiting satellites was changed from a statistical to a “physical” retrieval system. Because of this change, the National Meteorological Center (NMC) wanted to update the satellite temperature error statistics used in the NMC analyses. The authors, therefore, updated the estimates of observational error variances, horizontal covariances, and vertical correlations for layer mean temperatures retrieved from NOAA-10 satellite radiance data. The temperature error statistics have also been used to estimate analogous error statistics for isobaric height.
The computations used radiosonde data as a substitute for true temperatures. Each “matchup” in the dataset consisted of a satellite retrieval close in space and time to a radiosonde sounding. The matchups were stratified into clear, partly cloudy, and cloudy cases, depending on the amount of cloud contamination in the satellite radiance data. In each of the nine mandatory pressure layers considered, from 1000–850 to 50–100 mb, the clear and partly cloudy matchup cases have nearly equal temperature error variances, while the variances for cloudy cases are substantially larger. Vertical error correlations for all three stratifications are similar. Root-mean-square height errors computed from satellite temperature errors are comparable to those computed from radiosonde errors in the clear and partly cloudy matchup cases, but larger in cloudy cases.
