Take a connected component \mathcal{Z}\subset X_P^S. It is clearly B^S-stable and hence, by Borel Fixed Point Theorem, contains a B^S-fixed point vP/P (for some v\in W^P) since a B^S-fixed point, in particular, is a T-fixed point. We claim that the map
\hat{\phi}_v:~(G^S)^o\rightarrow X_P^S, \quad g\mapsto g\cdot vP/P
induces an isomorphism
\phi_v:(G^S)^o/\left((G^S)^o\cap(vPv^{-1})\right)\overset{\sim}{\longrightarrow} \mathcal{Z}:
By the Bruhat decomposition, we get
X_P=\sqcup_{w\in W^P}BwP/P,
and hence
X_P^S =\sqcup_{w\in W^P}B^SwP/P.
Since vP/P\in \mathcal{Z}, we get
(G^S)^o\cdot vP/P\subset \mathcal{Z}.
Moreover, the isotropy subgroup of v in (G^S)^o is clearly (G^S)^o\cap (vPv^{-1})\supset B^S, thus we get a (G^S)^o-equivariant embedding
\phi_v:(G^S)^o/\left((G^S)^o\cap(vPv^{-1})\right) \hookrightarrow \mathcal{Z}.
We now prove that any (G^S)^o-orbit in X_P^S is closed: By identity (?), any (G^S)^o-orbit A in X_P^S is a certain union \sqcup_{w\in\theta(A)}B^S wP/P, for some subset \theta(A)\subset W^P. Take v_o\in \theta(A) of smallest length. Then, we prove that
B^S\cdot v_oP/P =v_o P/P.
For if (?) were false, take \mathcal{U}_{\alpha}\subset B^S \subset (G^S)^o such that
\mathcal{U}_\alpha\cdot v_o P/P \supsetneqq v_o P/P,
where \mathcal{U}_\alpha is the one-parameter additive group corresponding to a positive root \alpha. In particular
v_o^{-1}\mathcal{U}_\alpha v_o\nsubset P \mbox{ and hence } v_o^{-1}\mathcal{U}_{-\alpha} v_o\subset \mathcal{U}^P,
where \mathcal{U}^P is the unipotent radical of P, which gives
\mathcal{U}_{-\alpha}v_oP/P=v_oP/P.
By (?), v_o^{-1}\alpha is a negative root and hence
s_\alpha v_o< v_o\quad \mbox{ by [K, Lemma 1.3.13]},
where s_\alpha \in W is the reflection through \alpha. Since \mathcal{U}_\alpha\subset B^S\subset (G^{S})^o, then so is \mathcal{U}_{-\alpha}\subset(G^S)^o and hence the subgroup \langle \mathcal{U}_\alpha, \mathcal{U}_{-\alpha}\rangle\subset G generated by \mathcal{U}_\alpha and \mathcal{U}_{-\alpha} is contained in (G^S)^o. Thus, s_\alpha v_oP/P \in (G^S)^o\cdot v_oP/P. This is a contradiction to the minimal choice of v_o in the (G^S)^o-orbit. Thus (?) is true, proving that any (G^S)^o-orbit in X_P^S is closed. Further, by (?), there are only finitely many (G^S)^o-orbits in X_P^S.
We return to the embedding \phi_v as in (?). We now show that \phi_v is surjective: Since \mathcal{Z} is stable under the action of (G^S)^o, it is a (finite) union of (G^S)^o-orbits. But, each (G^S)^o-orbit is closed in X_P^S and hence \mathcal{Z} is a disjoint union of finitely many closed (G^S)^o-orbits. But, \mathcal{Z} being connected, it is a single (G^S)^o-orbit, proving that \phi_v is a bijective isomorphism.
Consider the map
\psi:Y_S\rightarrow \left\{\mbox{set of the connected components of } X_P^S\right\},
which takes
v\in Y_S \mapsto (G^S) \cdot vP/P.
By the above proof (G^S)^o\cdot vP/P is indeed a connected component of X_P^S and hence \psi is well defined. Moreover, \psi is injective since any compact homogeneous variety of a reductive connected algebraic group over \mathbbm{C} has a unique fixed point under any Borel subgroup, which follows from the Bruhat decomposition.
Further, as proved above, \psi is surjective. This proves the proposition.