i didn't say when you do according to second method your results are absolutly correct. and i didn't say the first method is absolutly wrong.
in fact, all methods have some error but which is more accurate?
when you select a face as a radiation boundary, the software supposes that the radiation from that face must happen whether it is correct or wrong. so it is better that you do a compromise between infinit ground plane approximation and approximation of negligable radiation.
you should draw a box with the the same cross section as aperture surface if you want to use infinit ground plane approximation, and should draw a more large box if you don't want to ignore radiations below ground plane.
i think infinit ground approximation is better.
regds.