How can a data analyst writing Presto queries in watsonx.data reduce the amount of data scanned from Parquet files?

Parquet stores data by column; when a query selects only a subset of columns; Presto reads only those columns from disk (column pruning) — dramatically reducing I/O compared to row-oriented formats that require reading full rows. The incorrect options ("Avoid using WHERE clauses and filter data after retrieval in Python", "Run separate queries for each column and combine results in the client application", "Convert the table to CSV format which is faster to scan for specific columns") are distractors that don't fully capture the concept described.