I want to add Dac's answer about $\ce{NH3}$ impropers here I think it will help you:
The very simple answer to your question, is that improper dihedrals are needed when they are necessary. (Not very helpful, huh :) ) Let me elaborate some. In my experience, there are times when improper dihedrals are not needed because proper dihedrals are sufficient for modeling the energetics and the conformations of the atoms in question. (Note that in Amber proper and improper dihedral angles are defined by the same functional form.) Other times the person who has done the parameterization deems it necessary to use impropers. So ultimately, to use or not use impropers often comes down to the individual tastes of the person who created the parameters. I personally feel that impropers are need in selected case to correctly model out-of-plane bending, and there are very few, if any, force fields that do not use them.
Some else who knows better can correct me if I am wrong, but I believe that parmchk reports provides any improper dihedrals if it recognizes that they are present in the molecule's topology. A value of 10.5 kcal/mol is reasonable for an improper dihedral. If you are uncertain, the best way to determine if you need the impropers, or not, is to run some test simulations with and without them - then choose the setup that gives the best results. (However, if the goal is to force the improper dihedrals to be 100% planar, then adding an additional force constant to do so is "probably" okay. What you will miss by doing this is the proper modeling of out-of-plane bending, but I am not sure how well Parm99 models this anyway.)
Yes, this is David's opinion from the amber forum.
Additionally, if you can not scan improper dihedrals using ab inito method, you can use genetic algorythm with sander.
To do this first use in your *.frcmod file:
I-J-K-L 1 180.0 0.0
and draw chart with CPPTRAJ, then change energy and repeat it until you get a reasonable result. Actually you can find for this some software, too.
Or you can use ATB server for gromacs. If your molecule doesn't have metal ion, upload pdb format to ATB server and then it will give you force all force constants with improper dihedrals. If you use it in amber, just change the energy from joule to kcal ($\pu{1 kcal} = \pu{4.18 J}$) and multiply it by 0.5 and then use it in your *.frcmod file.