Angle-of-arrival (AOA) payloads on CubeSats enable passive geolocation of emitters for search-and-rescue and interference monitoring. Atmospheric refractivity bends radio rays and biases the line of bearing; although routinely corrected in ground tracking, it is rarely quantified or compensated for the reverse geometry of satellite receivers. Here it is shown that the bias can be estimated and corrected onboard by casting ray propagation as a geodesic in a refractive-index field and integrating an initial-value system seeded by the measured AOA direction. The method outputs both a corrected emitter position and an equivalent angular offset that can be applied directly to the payload measurement. Two engineering workflows are provided: an offline simulator for design and sensitivity studies, and an onboard pipeline requiring only the satellite state, an Earth shape model, and a compact refractivity profile. Solver benchmarks identify variable-step, variable-order integration and medium-order Runge-Kutta pairs as the most efficient options, compatible with FPGA-based processors. Sensitivity analyses for circular LEO orbits show errors decreasing with orbital radius and increasing with geometry (right ascension of the ascending node and true anomaly), reaching microradian angular biases and metre-to-tens-of-metres position shifts; differences in seasonal changes resulted up to similar to 9% of difference. This refraction-aware correction layer can improve future real-time or quasi-real-time AOA geolocation missions.