make_ijk_from_grid method does not work as expected

[oreiiikin]

I was trying to match measurements from wells (stored in a dataframe) with grid cells in grdecl format. However, the make_ijk_from_grid method does not seem to work quite correctly, for some reason it does not find the cell, although the measurement from the well gets there absolutely accurately, indexing simply stops at some point for some unknown reason. I'm attaching an example and a code snippet.

for well in tqdm(pd.unique(wells_df['well'])):
    wells_dict[well]["df"] = copy(wells_df[wells_df['well'] == well])
    wells_dict[well]["df"] = wells_dict[well]["df"][["X", "Y", "TVDSS"]]
    wells_dict[well]["df"].columns = ["X_UTME", "Y_UTMN", "Z_TVDSS"]
    wells_dict[well]["xtgeo_well"] = xtgeo.Well(df=wells_dict[well]["df"])
    wells_dict[well]["xtgeo_well"].make_ijk_from_grid(
        grid=grid,
        activeonly=False
    )

[mferrera]

Thanks very much for reporting, we'll take a look into this. In the mean time, you could give

wells_dict[well]["xtgeo_well"].make_ijk_from_grid(
    grid=grid,
    activeonly=False,
    algorithm=2
)

a try, but it is possible the bug exists there too :-)

[oreiiikin]

Thank you very much for a quick answer! "algorithm=2" is default parameter, so it probably won't give any positive result. It seems like the problem is occurred in C routine function "grd3d_points_ijk_cells.c".

I also get warning "splits or overlaps between vertical neighbouring cells is detected." when reading grid with "grid_from_file", likely because it has faults and so on.

[jcrivenaes]

I also get warning "splits or overlaps between vertical neighbouring cells is detected." when reading grid with "grid_from_file", likely because it has faults and so on.

Without knowing the data-set, I don't think this is related to faults. The grdecl format (Eclipse corner point format) specifies ZCORN "double", i.e. both for top and base cell, and although they in most cases are identical, they can easily have holes or overlaps.

The data model in xtgeo uses as simplified corner point approach, where inconsistent ZCORN's from grdecl are averaged somehow. I would think this potentially could explain this, but more investigations are needed.