To avoid notational clash, I'll use the notation $D_\alpha(X)$ to describe the hierarchy built on an RCF $X=(A;f_1,f_2)$ with underlying set $A$, addition function $f_1$, and multiplication function $f_2$, defined precisely as follows:
At the successor and limit steps we take definable powersets and unions respectively, as usual.
We start with $D_0(X)=A\cup A^2\cup\{f_1,f_2\}$.
Here are a couple quick comments to demonstrate that $D_0(X)$ really does have at least the "bare minimum" of expressive power we want for a set-theoretic implementation of an RCF:
We have $f_1, f_2\subseteq D_0(X)$ (and consequently $f_1,f_2$ are definable subsets of $D_0(X)$ since also $f_1,f_2\in D_0(X)$). This is because $A^2\subseteq D_0(X)$ and $f_1,f_2\subseteq A^2$.
We have that $A$ is a definable subset of $D_0(X)$ - e.g. as "The set of left coordinates of elements of $f_1$."
We can tell which of $f_1$ and $f_2$ is addition and which is multiplication, by asking which has an annihilator.
Now right away, we can make the following observation. As we go along the $D$-hierarchy, we "accidentally" wind up following the usual construction of $L$. In particular, we have $A^{<\omega}\subseteq D_\omega(X)$. This lets us implement the "natural" definition of $M$ in $D_{\omega+1}(X)$: "$M$ is the set of $m\in A$ such that there is some finite sequence of elements of $A$ whose first term is $1_X$, whose last term is $m$, and whose $(i+1)$th term is the $i$th term $+_X1_X$." This gives us the following:
$\alpha\le\omega+1.$
Can we do better? Well, at least for some presentations we can easily. Specifically, suppose that $$\mathcal{P}(A)\cap D_1(X)=Def(X),$$ where $Def(X)$ is the set of subsets of $A$ which are definable in the RCF $X$ in the model-theoretic sense. Then by o-minimality of RCF, we have that the following are equivalent for $U\in \mathcal{P}(A)\cap D_1(X)=Def(X)$:
$U$ is discrete, has $1_X$ as its least element, and for each $d\in U$ with $d\not=1_X$ we have $d-_X1_X\in U$.
$U=\{1\cdot 1_X, 2\cdot 1_X, ..., n\cdot 1_X\}$ for some $n\in\mathbb{N}_{\ge 1}$.
This gives us $M\in D_2(X)$: we have $m\in M$ iff there is some $U\in \mathcal{P}(A)\cap D_1(X)=Def(X)$ satisfying the above two bulletpoints with $m\in U$. Consequently, we have:
Restricted to "model-theoretically efficient" presentations of RCFs, that is, ones where $D_1(X)$ is "minimal," we have $\alpha=2$.
(It's easy to show $\alpha>1$.)
Moreover, we can get this unconditionally if $X$ is additionally Archimedean. This is because we can simply add the criterion that $U$ be bounded above and below; the only subsets of $A$ which are bounded above and below, contain $1_X$, and are closed under subtracting $1_X$ from ever non-$1_X$ element are the sets of the form $\{1\cdot 1_X, 2\cdot 1_X, ..., n\cdot 1_X\}$ for some $n\in\mathbb{N}$. That is:
If $X$ is Archimedean, then $\alpha=2$.
However, we run into a problem if $X$ is non-Archimedean and is presented in such a way that non-definable-in-$X$ subsets of $A$ show up in $D_1(X)$. In general I don't see a way to improve on the $\omega+1$ bound.
Conjecture: There is an RCF $X$ whose $\alpha$ is exactly $\omega+1$.