An automated method to classify Arctic fog into distinct thermodynamic profiles using historic in-situ surface and upper-air observations is presented. This classification is applied to low-resolution Integrated Global Radiosonde Archive (IGRA) soundings and high-resolution Arctic Summer Cloud Ocean Study (ASCOS) soundings in low- and high-Arctic coastal and pack-ice environments. Results allow investigation of fog macrophysical properties and processes in coastal East Greenland during melt seasons 1980–2012. Integrated with fog observations from three synoptic weather stations, 422 IGRA soundings are classified into six fog thermodynamic types based on surface saturation ratio, type of temperature inversion, fog-top height relative to inversion-base height and stability using the virtual potential temperature gradient. Between 65–80% of fog observations occur with a low-level inversion, and statically neutral or unstable surface layers occur frequently. Thermodynamic classification is sensitive to the assigned dew-point depression threshold, but categorization is robust. Despite differences in the vertical resolution of radiosonde observations, IGRA and ASCOS soundings yield the same six fog classes, with fog-class distribution varying with latitude and environmental conditions. High-Arctic fog frequently resides within an elevated inversion layer, whereas low-Arctic fog is more often restricted to the mixed layer. Using supplementary time-lapse images, ASCOS microwave radiometer retrievals and airmass back-trajectories, we hypothesize that the thermodynamic classes represent different stages of advection fog formation, development, and dissipation, including stratus-base lowering and fog lifting. This automated extraction of thermodynamic boundary-layer and inversion structure can be applied to radiosonde observations worldwide to better evaluate fog conditions that affect transportation and lead to improvements in numerical models.