See the second and third examples. The second example indicates IGDIST generates water:water g(r) for O:O, O:H, and H:H pairings. The third example (using ISDIST) computes g(r) for water oxygens around the backbone amide H atoms.
The g(r) facility in COOR ANAL is tailored to evaluate atom pairings, so you could look at e.g. the urea O atom around the Lys and/or Arg, or urea N atoms around Asp or Glu. To avoid overcounting, it is advised to use a single atom for cases like carboxylate and guanidinium groups, such as the central C atom in each of these moieties.
An alternative is the COOR DIST approach, which allows far greater flexibility in defining the distance pair, but is slower as it involves a loop over frames in the CHARMM input script. You need to take more care with image atoms, performing an UPDATE command after reading each frame with the IMALL keyword, with CUTIM set to the same value as the CUT option of COOR DIST. The COOR DIST command must also include the IMAGE keyword. You can also preprocess the trajectory to center the protein, so that it is never too close to edge of the box, and not have to worry about image atoms.