Spectra from the satellite instruments Iris (infrared interferometer spectrometer) were examined to find the number of independent variables needed to describe these broad band high spectral resolution data. The radiated power in the atmospheric window from 771 to 981 cm-1 was the first parameter chosen for fitting observed spectra. At succeeding levels of analysis the residual variability (observed spectrum minus best-fit spectrum) in an ensemble of observations was partitioned into spectral eigenvectors. The eigenvector describing the largest fraction of this variability was examined for a strong spectral signature; the power in the corresponding spectral band was then used as the next fitting parameter. The measured power in nine spectral intervals, when it was inserted in the spectral fitting functions, was adequate to describe most spectra to within the noise level of Iris. Considerations of relative signal strength and scales of atmospheric variability suggest a combination sounder (multichannel, broad field of view) scanner (window channel, small field of view) as an efficient observing instrument.