One way to estimate the ELF maximum within the lone pair is by brute force: estimate the coordinates of the "empty" coordination site (e.g. for a square pyramid, find the vector of the central atom to the axial position. Negate all three components to approximate the "empty" axial position that would make an octahedron). Create a dx structure file with a hydrogen atom at those coordinates. Plot the structure file in dx over the ELF data. Raise the isosurface value to find the maximum ELF within the lone pair. See how far of the H atom is to the ELF max and make adjustments to the coordinates accordingly. Repeat the process with another, corrected structure file until the H atom occupies the same space as the ELF max.
A more elegant approach:
Run JS's program for integrating charge densities (perl integrate_charge.pl). This requires the files ELF and RHO (cannot be renamed). The first resulting lines will show the grid size, the coordinates of the ELF max within that grid, and the ELF max value. To visualize where this ELF max lies, create at dx structure file with a hydrogen atom at the ELF max coordinates. For example, if integrate_charge.pl gave an ELF max at 20 25 50 in a 100 100 100 grid, then the coordinates of the ELF max within the unit cell are 20/100 25/100 50/100 = 0.2 0.25 0.5. Plot the structure file over the ELF data in DataExplorer and set the isosurface value to be near the ELF max determined by integrate_charge.pl. The H atom should occupy the same space (or close to) that of the ELF isosurface.
NOTE the ELF maximum may not lie within the lone pair. If this is the case, integrate_charge.pl cannot be used to determine the lone pair center and the "brute force" approach must be used.