What is a good way to handle nominal spatial data with a changing number of categories to use in prediction model?

For a project I'm going to be working with spatial data with a nominal attribute (land use). Every year the number of categories for this attribute changes because categories split or merge. I do have access to a chart that shows me how the categories are transformed into each other from one year to another. For the same spatial extent, I also have data for a bunch of other variables. I want to use these as explanatory variables for the nominal variable mentioned before. My question is now: How do I build a prediction model when the number of possible values for the reponse variable changes every year?