Separation of cloud layers in multispectral imager data

In this paper, we introduce methodology for multispectral layer separation. Efficient alternating minimization and fast operator-splitting methods are used to solve minimization problems. Specifically, we apply our methodology to separate strongly stratified and optically thin upper (cirrus) clouds from optically thick lower convective (cumulus) clouds in atmospheric imagery approximated as additive contributions to the observed signal. After setting up synthetic “truth” scenarios, we evaluate the accuracy of the two-layer separation results while varying the effective opaqueness of each of two types of cloud. We show that multispectral cloud layer separation is consistently more accurate than channel-by-channel cloud layer separation.