How can a scientific model be useful if it isn't a correct description of nature?

it must be internally consistent. can't contradict itself.

it must be consistent with past observations of reality, and offer some sort of explanation of them. it must be able to make testable predictions about future observations. it doesn't matter whether a scientific model is "true" or not, and in fact there's really no way of determining that.