It's not possible to answer this without knowing which methods packages you use and which databases. In general, the impact methods that are directly integrated in the databases have to be used, and/or the openLCA methods package has to be imported which is compatible to more than one database like e.g. ecoinvent, agrifootprint, agribalyse, and not only compatible to one database.
What you describe often happens when datasets are taken from one database or a random source and impact methods are taken from a completely different source, so that the same elementary flows will not "know each other" by having the same unique identifier.
You can download the free openLCA methods package and have a look in which impact categories potassium has characterisation factors.