Abstract

We use neural networks to perform temperature retrievals [1,2] from simulated clear-air radiances for the Atmospheric Infrared Sounder (AIRS) [3]. Neural networks allow us to make effective use of the large AIRS channel set, giving good performance with noisy input. Using 653 temperature and surface sensitive channels, RMS error on retrievals with 0.2K noise is below 1.2K. We briefly describe the sort of networks we are using, describe our datasets, and present a number of representative results.

© 1993 Optical Society of America