With the changing precipitation patterns and melting of mountain glaciers and permafrost that result from global warming, information on the distribution of groundwater in mountainous terrains is becoming increasingly important for developing prudent resource and hazard management strategies. Obtaining this information across topographically craggy and variably frozen ground in a cost-effective and nonintrusive manner is challenging. We introduce a modified 2D surface nuclear magnetic resonance (NMR) tomographic technique that allows us to account for substantial variations in surface topography in locating and quantifying groundwater occurrences in rugged mountains. Because contact with the ground is not necessary, it is a rare geophysical technique not affected by sensor-to-ground coupling problems common in high mountain environments. To demonstrate the efficacy of the tomographic imaging scheme, we invert a large multioffset surface NMR data set collected across a partially ice-cored proglacial terminal moraine in the Canadian Rocky Mountains. Our preferred model contains a 2- to 5-m-thick water layer, the top of which has practically the same elevation as the surface of a nearby lake and the bottom of which coincides with bedrock resolved in companion seismic and ground-penetrating radar studies.