I would argue that the natural selection occurs outside the hatchery. Man doesn't decide which smolt gets eaten by predators, which smolts learn to hide, etc. By taking a representative sample of an existing wild stock of a river, I don't see how this is manipulating natural selection. You could even rear the smolt in a pond complete with predators and cover, to help better emulate a natural senario. Don't continue to help the fish spawn after that, let them spawn naturally. And if they don't succeed, that is natural selection telling you something negative about the river, overfishing, strain etc.