A top-down parser builds the parse tree from the root downward, trying to match the input as it goes. Since the input is scanned left to right, the parser must always decide how to expand whichever non-terminal is furthest to the left in its current partial string, that's the part it needs to resolve next to keep matching.
Doing this repeatedly, step after step, produces exactly a leftmost derivation, in the same forward order the parser commits to its choices.
Bottom-up parsers are the ones associated with a rightmost derivation (built in reverse), not top-down parsers.
So the correct answer is leftmost derivation.