Something doesn't sit right with me. I would expect that the semicircle would be reflected nonlinearly. That would imply the reflection is not a true circle. Am I wrong?
I'm also a little skeptical with this approach, as it's not taking into account the curvature of light through the atmosphere. In reality, when the photo was taken, the sun was already fully below the horizon. It just looks like it isn't because the atmosphere is bending the light a little ways around the earth. (As a rule of thumb, the amount of bending is coincidentally nearly equal to the apparent size of the sun's disc, so as soon as the bottom edge appears to touch the horizon, that's the moment it has actually fully dropped below the horizon, and would be totally obscured if the light were travelling in the straight line paths assumed by this article.)
Atmospheric refraction does not matter for this method as it is not based on the time of sunset. Yes, the actual sunset occurs later than the observed sunset, but the relationship between the reflection and the observed position of the Sun remains the same. So, the only difference is you're observing the relationship at a slightly different time than when it actually occurred geometrically.