The classification theorem for surfaces says that the complete set of homeomorphism classes of surfaces is

{ $S_g : g \geq 0$ } $ \cup$ { $N_k : k \geq 1$ },

where $S_g$ is a sphere with $g$ handles, and $N_k$ is a sphere with $k$ crosscaps. The first homology groups are easy to compute. They are $H_1 (S_g) = \mathbb{Z}^{2g}$, and $H_1 (N_k)=\mathbb{Z}^{k-1} \times \mathbb{Z} / 2\mathbb{Z}$. My question concerns how the homology groups change once we start cutting holes in our surface.

In the orientable case, it is easy to see what happens. The first hole that we cut out does not change the homology. Every additional hole then introduces another factor of $\mathbb{Z}$.

In the non-orientable case something peculiar happens. Consider the projective plane with homology group $\mathbb{Z} / 2 \mathbb{Z}$. If I cut out a hole, then I get the Mobius strip, which has homology group $\mathbb{Z}$ (it is homotopic to a circle). In general, if I cut a hole out of $N_k$, then in the homology group I lose a factor of $\mathbb{Z} / 2 \mathbb{Z}$, and introduce a factor of $\mathbb{Z}$. Each additional hole will then just introduce another factor of $\mathbb{Z}$.

**My question:** In the non-orientable case what happened to the factor of $\mathbb{Z} / 2 \mathbb{Z}$? Is there a nice geometric explanation of why it went away? I'm slightly disturbed because I had the intuition that torsion was supposed to record non-orientability, but I guess this doesn't work for surfaces with holes.