The grid of the output in align_epi_anat.py is controlled by the -master_xxx options. If you are using giant_move or ginormous_move, the grid is set to be on the base's grid but with the isotropic resolution of the minimum dimension of the input dataset by default. The master options override any of the defaults, and you get the grid on the output you want. I think in this case, the dataset moved forward a bit with the alignment but stayed in the same grid location (no giant_move), and that removed the anterior portion. If you use the other dataset as the master, then you would get the full coverage at the cost of empty voxels. If you know that the aligned dataset only needs a few extra slices, you can make one with 3dZeropad, and that can be your target grid. You may also apply the affine transformation with 3dWarp, which moves the dataset and automatically computes the grid that fits if no gridset option is used.