You have a problem with the direction of your main lobe in the top figure, which results in the big differences in the bottom figures, because the cuts don't go through even near the top of your beam. Most of the problem is with the direction of the beam in the top plot. This may be due to the feed network - is it a patch array? Small length differences can have a big effect on pointing direction, and its sensitivity to frequency.
The top traces are much more similar if you turn one of them round, so you may have a mirror image problem, or have your angles the other way round. Then just redefining where 0 is would help as well! So it runs through your middle peak. You can do this by tilting your array slightly.
You can either tilt the array mechanically - by sloping the circuit board, or electrically by changing the patch separation, or by changing the dielectric constant of the board, or by putting extra length in the connecting tracks, perhaps by wiggling them.