Although satellite-based SAR is sensitive enough to pick up the modulation effects of internal waves, the signal-to-noise ratio (SNR) of this observable is believed to be too low to afford detection of enemy submarines. Adaptive Software proposes that with a new interpretation of singular value decomposition, the SNR can be improved enough to allow automated detection of submarine paths. While most SVD-based analyses try to isolate one or more singular planes that contain the strongly-correlatedsignals of interest, the proposed approach is to remove those planes that can be shown to contain little or no information of interest. Rissanen's MDL metric for model order determination will guide the removal of strong interference, while a derivative of Shannon's entropy metrics will guide the removal of uninformative noise planes. This preprocessing will allow a stochastic edge detector to find the path over which a submarine has altered the statistical character of the waves.