Regression Technique for Determining Temperature Profiles in the Upper Stratosphere From Satellite-Measured Radiances

Abstract

This paper describes a statistical regression technique for specifying the vertical temperature profile above 10 mb from satellite radiances simulated for the Satellite Infrared Spectrometer B (SIRS B) instrument when the temperature profile up to 10 mb is known. Sensitivity of the radiance information to temperature at the high atmospheric levels is attained by subtracting, from the total radiance, that part of the radiance emanating from the known temperature profile below 10 mb. The remainder is that portion of the radiance representative of the temperature profile above 10 mb. Statistics are derived using a sample of 50 carefully selected temperature profiles representative of worldwide atmospheric conditions above 30 km during all times of the year. Regression equations are developed relating temperature at 10 pressure levels between 10 and 0.5 mb to a set of predictors [temperature at 50, 30, and 10 mb and radiance information derived from SIRS B channels 7 (679.8 cm?1) and 8 (668.7 cm?1)]. For an independent data set, root-mean-square errors in specification ranged from 2.1°C at 9 mb to 8.8°C at 0.5 mb, with the shapes of all profiles very well distinguished. Regression-specified temperatures above 10 mb are then used as first guess in simulated retrievals of complete atmospheric temperature profiles. These regression results are shown to significantly increase the accuracy of temperature retrievals at tropospheric as well as stratospheric levels over those retrievals derived using a climatological first guess.