Comparison of neural networks and regression-based methods for temperature retrievals

Two methods for performing clear-air temperature retrievals from simulated radiances for the Atmospheric Infrared Sounder are investigated. Neural networks are compared with a well-known linear method in which regression is performed after a change of bases. With large channel sets, both methods can rapidly perform clear-air retrievals over a variety of climactic conditions with an overall RMS error of less than 1 K. The Jacobian of the neural network is compared with the Jacobian (the regression coefficients) of the linear method, revealing a more fine-scale variation than expected from the underlying physics, particularly for the neural net. Some pragmatic information concerning the application ofneural nets to retrieval problems is also included.