[Qgis-developer] Approximation and geoprocessing

[Andreas Neumann]

Hi,

I don't have a solution for the problem.

But I can confirm that the whole analysis chain is a bit problematic
with all these precision issues. As an example if you test intersections
of buildings against parcels (2 different layers) and the building
touches the parcel borders you may get very tiny intersects with parcels
(only a few square cms) and you wouldn't want to include these
intersections in your analysis.

So it would be cool if for geometry tests there could be a notion of
threshold values. If the area of an intersected polygon is below this
value it wouldn't be taken into account during the analysis.

These issues are really kind of show-stoppers and almost always require
intermediate steps to sort out these edge cases. It would be very handy
if the tools could handle this precision/uncertainty problems without
having to do a lot of intermediate steps.

Andreas

Am 10.01.2014 15:01, schrieb Paolo Cavallini:
> Hi all.
> I got into an interesting problem, before opening a few tickets I'd like
> to discuss a line of action.
> 1. Some sw (namely geomedia) produces shapefiles with very slight
> differences (same layer, modified and exported several times); I suppose
> this is due to truncation/approximation in coordinate precision.
> 2. As a result, the diff between these layers produces a number of
> virtually 0 width polygons, and some topological errors.
> 3. This in turn leads to 2 problems:
> 3.1. subsequent analyses fail because of the errors, and the user has no
> clue about the reasons
> 3.2. The microareas are tricky and time consuming to remove (if they are
> isolated, a possible solution is to select and remove those smaller than
> an arbitrary threshold, but if they are linked to some "real" polygons,
> this will not work).
> I would therefore suggest a few improvements over the toolchain, i.e.:
> * add a parameter in analyses, to either snap original data within a
> threshold before running the analysis, or to clean up the results afterwards
> * check & warn the user of topo errors in source layers, and of possible
> wrong results; maybe this should be optional, as this will slow down the
> analysis, and may be useless after first cleanup.
> 
> Sample data available for those interested.
> 
> All the best.
>