1: \begin{abstract}

2: 

3: Given a surface $M$ and a Borel probability measure $\nu$ on the group of $C^2$-diffeomorphisms of $M$  we study $\nu$-stationary probability measures on $M$. 

4: We prove for hyperbolic stationary measures the following trichotomy: either the stable distributions are non-random, the measure is SRB, or the measure is supported on a finite set and is hence almost-surely invariant.  

5: In the proof of the above results, we study skew products with surface fibers over a measure-preserving transformation  equipped with a decreasing sub-$\sigma$-algebra $\hat {\mathcal F}$ and derive a related result.   

6: A number of applications of our main theorem are presented.  

7: 

8: %In the case that $\nu$-\ae diffeomorphism preserves a common smooth measure $m$, we show for any non-atomic stationary  measure $\mu$ that  either there exists a $\nu$-almost-surely invariant $\mu$-measurable line field (corresponding to the stable distributions for \ae random composition) or the measure $\mu$ is $\nu$-almost-surely  invariant and coincides with an ergodic component of $m$. 

9: 

10: 

11: %In the proof of the above results, we study skew products with surface fibers over a measure preserving transformations equipped with an increasing sub-$\sigma$-algebra $\hat \Fol$. 

12: % Given an invariant measure $\mu$ for the skew product whose fiber-wise conditional measures are non-atomic, we  assume the $\hat \Fol$-measurability of the `past dynamics' and the fiber-wise conditional measures and prove the following dichotomy: either the fiber-wise stable distributions are measurable with respect to a related increasing sub-$\sigma$-algebra, or the measure $\mu$ is fiber-wise SRB. 

13: 

14: 

15: %We also deduce a number of corollaries for groups of surface diffeomorphisms preserving a common measure.

16: \end{abstract}

17: