Input variable selection for hydrological predictions in ungauged catchments: with or without clustering