The HST Space Telescope Imaging Spectrograph (STIS) FUV-MAMA detector has a “glow region” dark background that varies slowly in morphology over the years, as well as quickly over hours as the detector warms up. It is typical to see glow region dark rates of four times the detector average by the fifth orbit of a visit. Dark monitoring observations are taken for five consecutive orbits every six weeks, but the low count rate in any given observation limits the SNR while the variability limits the usefulness of stacked super-darks. Currently, the HST/STIS pipeline does not correct for this background structure beyond 1D background subtraction during spectral extraction. However, the asymmetry of the glow region with respect to the spectral extraction location often leads to under-subtraction of the background. Here we demonstrate initial work to fit morphological changes with high SNR super-darks by training a Principal Component Analysis (PCA) model of the detector dark structure. Different spatial and temporal binning of input training data are investigated. Furthermore, we provide an algorithm to iteratively apply the PCA eigendarks to spectroscopic observations while masking non-background sources. Python code demonstrating the PCA model and its application are made available in Jupyter Notebooks.