Machine-learning regression tree models were used to extrapolate airborne electromagnetic resistivity data collected along flight lines in the Yukon Flats Ecoregion, central Alaska, for regional mapping of permafrost. This method of extrapolation (r = 0.86) used subsurface resistivity, Landsat Thematic Mapper (TM) at-sensor reflectance, thermal, TM-derived spectral indices, digital elevation models and other relevant spatial data to estimate near-surface (0–2.6-m depth) resistivity at 30-m resolution. A piecewise regression model (r = 0.82) and a presence/absence decision tree classification (accuracy of 87%) were used to estimate active-layer thickness (ALT) (< 101 cm) and the probability of near-surface (up to 123-cm depth) permafrost occurrence from field data, modelled near-surface (0–2.6 m) resistivity, and other relevant remote sensing and map data. At site scale, the predicted ALTs were similar to those previously observed for different vegetation types. At the landscape scale, the predicted ALTs tended to be thinner on higher-elevation loess deposits than on low-lying alluvial and sand sheet deposits of the Yukon Flats. The ALT and permafrost maps provide a baseline for future permafrost monitoring, serve as inputs for modelling hydrological and carbon cycles at local to regional scales, and offer insight into the ALT response to fire and thaw processes.