Ground-penetrating radar systems are useful for a variety scientific studies, including monitoring changes to the polar ice sheets that may give clues to climate change. These systems produce vast amounts of radar image data that is typically processed by hand, because the echograms are noisy and difficult to interpret. We are investigating and developing computer vision techniques for analyzing this data, using probabilistic graphical models that can explicitly model the uncertainty and noise in the data. Our techniques are fast (typically milliseconds to seconds per image), and can incorporate a variety of diverse evidence including feedback from a human operator, or measurements from ice cores. We quantitatively evaluate our approaches on large-scale test data, typically including hundreds of echograms and comparing to human-labeled ground truth.