Kant's distinction between the noumenal and the phenomenal in his *moral philosophy** is both difficult to follow and integral to his moral thought. Moreover, Kant's use of terms like "nature" and "understanding" have definitions that are not intuitive if we are working from the standard English words.
I think it's easiest to understand it if we start with what Kant wants in a moral theory. Roughly speaking, Kant is hoping for a theory where the moral self using universal reason is able to decide what it morally ought to do and then where the self has the freedom to will this action.
One simple dictum that captures this is "ought implies can." Kant thinks we can know what we ought to do. And then he thinks we can do it. So if we know murder is wrong, Kant thinks we can always in every circumstance do what is right.
But considering he is writing after Descartes, this is clearly going to be difficult to square with Newtonian mechanics -- where everything appears to be explained by the laws of physics. Kant's solution in his metaphysical works. Put simply, everything was explained in terms of nature and its subsequent determinism.
Kant's epistemology hinges on a distinction between knowledge and reason and then things, sensibles, and objects. Put simply, we do not understand things as they; instead we receive them as sensibles through the forms of sensibility (space and time) and then know them as objects through the categories of the understanding. Nature is roughly speaking synonymous with determinism for Kant, which is one of the categories to which our understanding is always subject.
But reason does not function through the categories or the forms. Instead, reason is free. In this way, it lies outside the world of determinism. But the freedom comes at a price of being in a different realm than the phenomenal, which Kant calls the noumenal (derived from the Greek word nous which means mind). The moral self for Kant is the self that is noumenal and that can act freely unbuffetted by the deterministic world.
This is the background apparatus for answering your question.
On Kant's view, the self as a moral self is not in the world of nature. Instead, the self is in a world where it is free. And the reason why we don't see that is that we cannot see it since it is not a phenomenon. In fact, on most interpretations of Kant's moral theory, we can never directly perceive than act is moral, because we can only see and comprehend our world in terms of the forms and categories and encounter others regularly under these.
But Kant asserts that we must have this form of noumenal self. He makes very different arguments for this depending on where in his moral works you are reading. In the Groundwork, he claims he can prove this. In the Critique of Practical Reason, it becomes more a fact of reason that this is true. And in later works, the picture gets complicated as he handles cases of habit.
There are certain semi-cartesian features in Kant's account:
- he shares a deterministic picture of the sensible universe.
- he shares a belief in the superiority of rationality over the deterministic natural animal
- he has a problem connecting this account to the physical domain.
Where he differs from Descartes is in what is called the "Copernican revolution" whereby he looks at our understanding in a different way that hints at a cognitive faculty that has power over the world that we perceive. (note that the domain of its power is the "world" which here refers to a concept in the mind). Basically, Kant turns to the mind in a different way than Descartes -- informed by and accepting Hume.
Kant evades in part the question of how the moral self can interact with the world by making claims about the moral self that take it out of the world of our perception -- i.e., what we see is phenomenal, what we are as free creatures is noumenal. How the two meet is also beyond our ability to know, because knowing is the application of forms and categories to things (and we cannot know the things directly).
Note that we are not required to agree with Kant. If you're a reductionist physicalist, then his theory probably won't work for you in the standard form. But there are contemporary appropriations of Kant that might, such as Christine Korsgaard's constructivism -- whereby she believes we are committed to Kantian principles of agency insofar as we are agents. She does some rather elaborate footwork to keep parts of his ethics while jettisoning his account of metaphysics across her texts Sources of Normativity and Creating the Kingdom of Ends.