This is what you can typically do: For example, if you want to use GFDL data for initial conditions, and GFS data for boundary conditions, you first run GFS data for the entire time period, all the way through real.exe (later you will use the wrfbdy_d01 file and discard the wrfinput_d01 file). You then run GFDL data for only the initial time. Since the GFDL data does not have complete soil data, you can supplement it with GFS surface data only (this is only needed if you would like to use the LSM option in the model later). To get only GFS surface data, you should create a Vtable that contains only those surface variables. You can ingest GFDL and GFS surface data in metgrid to obtain a full dataset for the real.exe program. Once you run real (for only the first time period), you obtain wrfinput_d01 with GFDL data. You can then run the model.