Retrieval of Atmospheric Temperature and Moisture Profiles from Hyperspectral Sounding Data Using a Projected Principal Components Transform and a Neural Network

 

William J. Blackwell

Lincoln Laboratory, Massachusetts Institute of Technology

 

Abstract

A novel statistical method for the retrieval of atmospheric temperature and moisture (relative humidity) profiles has been developed and evaluated with simulated clear-air hyperspectral sounding data. The accuracies of the estimates produced by the algorithm meet or exceed (in some cases by a factor of two) the accuracies of the estimates from traditional iterated minimum variance retrieval techniques while requiring less computation. The algorithm is implemented in two stages. First, a projected principal components (PPC) transform is used to reduce the dimensionality of and optimally extract geophysical profile information from the spectral radiance data. Second, an artificial feedforward neural network (NN) is used to estimate the desired geophysical parameters from the projected principal components.  The performance of this method (henceforth referred to as the PPC/NN method) was evaluated using simulated clear-air observations from the 2378-channel Atmospheric InfraRed Sounder. Separate training and validation profile data were selected from the NOAA88b radiosonde set of approximately 7500 profiles. Surface, solar, and instrument effects were modeled. It was found that the PPC/NN method has a number of advantages over traditional statistical and physical/iterative hyperspectral profile retrieval techniques. Neural-network estimates based on the PPC transform were significantly more accurate than neural-network estimates obtained with conventional principal components techniques, including the Karhunen-Loeve transform and the noise-adjusted principal components transform. The retrieval accuracy of PPC/NN was also superior to that of a principal components regression technique. One particularly noteworthy result of the present work is a comparative study between the PPC/NN method and an iterated minimum-variance (IMV) method. The temperature profile retrieval accuracy of both methods is similar, but the relative humidity profile retrieval accuracy of PPC/NN was greater than that of the IMV method at all altitudes and substantially better near the surface.

 

Speaker’s Biographical Sketch

William J. Blackwell received the B.S. degree in electrical engineering from the Georgia Institute of Technology, Atlanta, in 1994, and the S.M. and Sc.D. degrees in electrical engineering from the Massachusetts Institute of Technology (MIT), Cambridge, in 1995 and 2002. He is currently a member of the technical staff at MIT Lincoln Laboratory. Dr. Blackwell held a National Science Foundation Graduate Research Fellowship from 1994 to 1997 and is a member of Tau Beta Pi, Eta Kappa Nu, Phi Kappa Phi, and Sigma Xi, and is a Senior Member of the IEEE. He is chapter chair of the Boston section of the IEEE Geoscience and Remote Sensing Society (GRSS), associate editor of the IEEE GRSS Newsletter, and a member of the organizing committee for IGARSS 2008 to be held in Boston, MA.  Dr. Blackwell serves on the science teams of several NASA and NOAA satellite programs and has authored over 40 journal articles, conference proceedings, and technical reports on atmospheric remote sensing. He is co-author of the book “Neural Networks in Atmospheric Remote Sensing,” to be published by Artech House in 2009.

 

DATE:  Wednesday, February 27, 2008 

TIME:  12:00-12:10 pm-refreshments; 12:10-1:00 pm-seminar

LOCATION:  424 Benedum Hall