I should preface this by saying that I am a physicist. My question pertains to the paper https://arxiv.org/abs/1802.09978, where $r$-spin structures are defined (Def. 2.1).
To orient the discussion, let me give my basic understanding of usual spin structures, before moving on to the $r$-spin generalization. I'm happy to restrict the discussion to Riemann surfaces of genus $g$ with no boundary, $\Sigma_g$. Heuristically, I think of a spin structure on $\Sigma_g$ as an assignment of a $\mathbb{Z}_2$ holonomy to each cycle in $\Sigma_g$, which gives $2^{2g}$ distinct choices of spin structure.
Here are two facts about $r$-spin structures, both found in Prop. 2.21 of the reference:
i) A $r$-spin structure exists on $\Sigma_g$ iff $2-2g = 0$ modulo $r$.
ii) When i) is met, there are $r^{2g}$ distinct $r$-spin structures on $\Sigma_g$.
Now here come the questions: Fact ii) makes me want to think of $r$-spin structures as assignments of $\mathbb{Z}_r$ valued holonomies to each cycle in $\Sigma_g$, in analogy with usual spin structures. Naively, this works for any pair $(r,g)$, in tension with fact i). I'm wondering if there is any easy way to see what the obstruction is in a concrete example, say $(r,g) = (3,2)$.
I should also mention that I have heard various explanations (and the paper mentions this) involving conditions under which one can take roots of the canonical line bundle, but these are not intuitive to me. Above all (and I realize this may not be doable), I would love a pictorial explanation in a simple case.
Thank you in advance.