The proposed research will develop long-range terrain characterization technologies for autonomous excavation in planetary environments. This work will develop a machine learning framework for long-range prediction of both surface and subsurface terrain characteristics that: (1) indicate the excavation-value of the material and (2) describe how hazardous terrain is to a robotic excavator. Factors influencing importance include the mineral composition of the material and the presence and concentration of volatiles. Terrain hazards will include loose terrain that could cause wheels to sink or slip as well as the presence of surface and subsurface rocks that would inhibit excavation. This work will develop technologies for long-range terrain mechanical characterization and volatile prediction with high spatial coverage. Ground penetrating radars and neutron spectrometers provide reasonable accurate estimates of subsurface composition and volatile accumulation; however, they are limited in sampling range and area. Cameras and LIDAR will instead be used to measure reflected radiation, temperature, and geometry at long range with a wide field of view. From these measurements, the thermal properties and spectral reflectance curves of the terrain will be estimated, since both are correlated to terrain composition and traversability. These properties, along with geometry, will be fed into a machine learning framework for prediction of terrain characteristics. Priors will be generated based on data from orbital satellites. Measurements of material composition, volatile accumulation, and traversability will be generated from expert labeling, neutron spectrometers, and wheel slip measurements, respectively. These measurements will be used to train machine learning algorithms for long-range prediction of terrain mechanical characteristics and resource concentration.