If you have a CDC stream capability, you can use the APPLY CHANGES INTO API to perform SCD1, or SCD2 in a Delta Lake table in Databricks.  You can find more information here.  This is the best way to go if CDC is a possibility.

If you do not have a CDC source, using MERGE INTO can handle appends, updates, and deletes.  If the table size is several terabytes and the merge conditions required are very complex, other techniques may work.  Such as a partitioning scheme at the level of granularity the data can change by, change detection and logic for deletion of entire partitions (if that is a potential), and overwriting the changed or new partitions and deleting those that are no longer in the parquet directory.

I may not be covering all of the practical possibilities here.  But without more specifics about your data type, volume, complexity, rate of change, etc. it is difficult to provide prescriptive options if APPLY CHANGES INTO and MERGE INTO do not work for your use case.

I hope this is helpful.