The rapid development of aerospace technology, such as GPS satellite navigation system, represented by high precision and sensitivity, is gradually gaining wide attention from researchers and replacing traditional radio navigation systems, playing an important role in military defense, space exploration, engineering surveying, air-to-air combat and other fields. However, due to the limitations of traditional electromagnetic theory, satellite navigation technology has relatively weak anti-electronic deception and electromagnetic jamming capabilities. In order to enhance the autonomy and reliability of the navigation system, a passive and strong counter-jamming navigation method, which is named as starlight navigation, has been proposed. In the 1950s, the advent of star sensors has greatly improved the accuracy of starlight navigation. Star sensors are high-precision attitude-sensitive measuring instruments that measure the star vector component in the star sensor coordinate system by conducting the stellar observation, and determine the three-axis attitude of the carrier relative to the inertial coordinate system using known precise star positions. The high accuracy, strong counter-jamming ability, and independence from other systems of star sensor navigation technology have a wide range of applications and important military value on various airborne, shipborne, and vehicle-mounted platforms in near-earth space. However, as the development of observation platforms and the decrease in the observation height of star sensors in the atmosphere, a star sensor operating in the terrestrial space will inevitably be affected by sky background radiation, atmospheric turbulence, and atmospheric refraction during the observation. This three-part paper aims to extensively reveal these atmospheric effects on stellar observation. In Part Ⅲ, we select an optimal atmospheric refraction model for autonomous satellite navigation and study the effect of atmospheric refraction on star imaging. We introduce different models of refractive index and refraction calculation, and employ the parameter profile data of the U.S. standard atmosphere to calculate the refraction distribution characteristics of the plane-parallel atmosphere, spherical whole-layer atmosphere and spherical multiple-layer atmosphere. We compare and analyze the advantages and disadvantages of different refraction calculation models in terms of calculation accuracy, iteration number and algorithm speed, and theoretically and numerically select the most accurate and fast refraction calculation model for our specific calculation. Based on these models and the profile data of atmospheric parameters measured at different times in typical regions of China, we calculate the distribution of refraction angle, dispersion angle, lateral shift, and path elongation effect caused by atmospheric refraction under different observation conditions and values of wavelength, and evaluate the effect of the uncertainty of input parameters on the calculation of refraction. Lastly, we provide an illustrative example of a polynomial fitting function for the path elongation ratio at a particular observation height, which can be utilized for the fast computation of refractive effects in the real-time scenario. We find that we can most accurately calculate the refraction angle by the use of the Cassini model or the equivalent refraction index ray tracing method. Moreover, we reveal that the effect of atmospheric refraction on star imaging can be greatly reduced by increasing the observation height of the star sensor or decreasing the observation zenith angle of the star sensor compared to changing the value of wavelength. Finally, we found that when the input parameters exist noise and uncertainty, improving the accuracy of temperature measurement is more effective in reducing the error of refraction calculation than suppressing the noise of other parameters. This research sheds light on the atmospheric effects on star imaging and offers insights into improving the accuracy and reliability of autonomous satellite navigation. Moreover, the findings of this research could aid in the development of more reliable and accurate navigation systems in the future and the performance improvement of star sensors operating in terrestrial space.