Umbrella Sampling for Surface Adsorption

GROMACS version: 2016.6
GROMACS modification: Yes

Hello experts,

I have a solvated crystalline surface with an adsorbed surfactant molecule. My goal was to determine the adsorption free energy via umbrella sampling. The system is oriented such that the z-direction is aligned with the crystal surface normal. I generated an initial trajectory by applying a harmonic bias in the z direction to pull the surfactant into the bulk water. This trajectory was then sliced into individual starting points for umbrella sampling windows. I then ran what I believe is a lengthy restrained equilibration for each window. So far so good, here is the sampling protocol that I used:

pull = yes
pull_ncoords = 1
pull_ngroups = 2
pull_group1_name = surfactant
pull_group2_name = surface
pull_coord1_type = umbrella
pull_coord1_geometry = distance
pull_coord1_dim = N N Y
pull_coord1_groups = 1 2
pull_coord1_start = yes
pull_coord1_rate = 0.0
pull_coord1_k = 1000

The individual trajectories look good and the PMF looks good near the surface. However, the PMF is constantly increasing as the surfactant enters the bulk water. I would have expected the PMF to diminish once the surfactant was sufficiently far enough away from the surface but this is not the case. I have attached a plot of the PMF. Does anyone have any experience with surface adsorption calculations? I have tried several modifications without much success. I do not understand if this is an issue with my simulation or post processing. I have used the built in wham utility.

Thank you for any and all possible wisdom.