Constraining the "quadtree" mapping algorithm with a shapefile?

[olivierhardy]

I wonder whether the "adaptive resolution" algorithm could be improved to take into account a shapefile that imposes limits between "cells". For example, a shapefile of elevation belts with the contrain that any cell must be embedded in a single elevation belt. This could help create cells that are more homogeous in terms of environmental conditions, instead of squares or combinations of connected squares that could still include a lot of environmental heterogeneity.