How does Schwartz's paradox of surface area effect modelling of 3D objects? Question
I just became aware of Schwartz's paradox of surface area (explanation below for the unfamiliar). How does this effect mathematical modelling of real-life surfaces? For example, suppose I wanted to measure the surface area of a mountain and had the elevation data. I've found approaches that produce a polyhedral approximation (here), but how do we know this polyhedral approximation is actually approaching the surface area of the mountain? Thanks!
PS Maybe this is a better physics question? Also, Mandlebrot's first fractal paper comes to mind as a similar problem.
Schwartz's Paradox Explanation
If I understand correctly, Schwartz's Paradox shows that simply because a polyhedral approximation, $P_n$, of a curved surface $S$ approaches the curved surface as $n \to \infty$, the surface area of the polyhedral approximation, $A(P_n)$, does not approach the geometrically intuitive surface area of the surface, $A(S)$. In summary,
$$\lim_{n\to\infty} P_n = S \not\Rightarrow \lim_{n\to\infty} A(P_n) = A(S) $$
I surmised this from the following paper.
 A: When approximating the cylinder, for example, such an approximation does work, so long as we also have convergence of the normals. 
The approximation fails for Schwarz lanterns because the normals of the triangles in the Schwarz lanterns don't converge at all, much less to the normals of the cylinder. 
See for example here of how we can use triangulations to approximate surface area: http://arxiv.org/pdf/1404.1823v1.pdf
Also look at the last paragraph of the paper linked to in the original question -- Lebesgue already solved this problem in 1902 by demanding uniform convergence instead of just pointwise convergence. Clearly Schwarz's lanterns converge pointwise but not uniformly -- the reason why the surface area blows up is because the triangular approximations can become arbitrarily jagged (which is also why the normals fail to converge). With uniform convergence, this doesn't happen.
See this Wolfram demo for how it works: http://demonstrations.wolfram.com/CylinderAreaParadox/
If it is reasonable to demand uniform convergence of functions instead of just pointwise convergence, then it should also be reasonable for surface areas. 
A: In short, it doesn't. 
I don't have a general proof, but let's take a look at the original paradox. For a cylinder, the area is given by:
$$A(n,m) = 2nr\sin (\pi/n)\sqrt{h^2+r^2m^2(1-\cos^2 (\pi/n))^2}$$
Now, it's true that if you let $m$ go to infinity independently of $n$, you'll end up with nonsense. But in all real life scenarios, you only have finite sampling, and therefore, the ratio $c=m/n$ is bound by some constant. You can then show that if $m=cn$, the limit is indeed $2\pi hr$ irrespective of $c$.
