Refraction seems to be invoked when you guys find it convenient to do so.
Nope. We only "invoke" refraction when it impacts the observations.
You really need to read about the thorough debunking of the FE's favourite snake-oil salesman Samuel Rowbotham.
Anyway... refraction of light
could produce the results noted by Rowbotham... Because the density of air in the earth's atmosphere decreases with height above the earth's surface, all light rays travelling nearly horizontally bend downward. This phenomenon is
routinely allowed for in levelling and celestial navigation.
If the measurement is close enough to the surface, light rays can curve downward at a rate equal to the mean curvature of the earth's surface. In this case, the two effects of curvature and refraction cancel each other out and the earth will
appear flat in optical experiments.
This would have been aided, on each occasion for Rowbotham, by a temperature inversion in the atmosphere with temperature increasing with altitude above the canal... Temperature inversions like this are common. An increase in air temperature or lapse rate of 0.11 degrees Celsius per metre of altitude would create an illusion of a flat canal, and all optical measurements made near ground level would be consistent with a completely flat surface.
If the lapse rate were higher than this (temperature increasing with height at a greater rate), all optical observations would be consistent with a concave surface, a "bowl-shaped earth". Under average conditions, optical measurements are consistent with a spherical earth approximately
15% less curved than its true diameter. Repetition of the atmospheric conditions required for each of the many observations is not unlikely, and warm days over still water can produce favourable conditions.