Atmospheric and surface parameter retrievals from multispectral thermal imagery via reduced-rank multivariate regression

This paper introduces the application of a multivariate analysis technique for optimal retrieval of atmospheric and surface parameters from multispectral or hyperspectral observations. The method is based on canonical correlation analysis (CCA). The advantage of CCA over conventional statistical methods is that it operates in an orthogonal subspace where correlations are maximized, thereby avoiding problems arising from rank-deficient or ill-posed data. The method was applied to simulated and real data from the MODerate resolution Imaging Spectrometer (MODIS) and MODIS Airborne Simulator (MAS). In general, the errors in retrieved temperature and water vapor profiles were 1-2° K and 2-4 mm RMS, respectively. Land surface temperature retrieval accuracy was within 0.4°K