In train an argument comes either from caret::train or from the model used in method and for which argument are passed through ...
In you can't find it as a argument of train, (in help page or the book), you can assume it is from the model.
Here method = 'rf' means randomForest from the
randomForest and importance is an argument from randomForest::randomForest. See the help page.
importance Should importance of predictors be assessed?
It is FALSE by default.