Areas dropped in creation process

[SK53]

The creation process drops areas which have no adjacent areas in the second stage of merging areas. Mostly this weeds out false positives, but it also eliminates some places which should be included. For instance in the Southern Cone of South America: Caleta Olivia, Puerto Natales and El Calafate.

In the absence of a more rigorous way to remove false positives from the intial dataset a fairly simple minded approach is just to included all areas over a certain size which were isolated in stage 1. For instance for South America the total number of initial candidate areas is 45k, of which 25k are isolated, but only 1k are over 2 sq km in size.

Other parameters which could be looked at are number of children, or the ratio of short and long axes of an enclosing ellipse (some candidates will be nothing more than a single short road segment).

[SK53]

An example area from the Pampas west of Buenos Aires: strings of towns (originally along railway lines) some included and some not. Dark coloured areas were in first processing, those with an orange halo added from the larger areas (> 2 sq km in area) missed, light green ones still not incorporated. One or two false positives are visible, but most are valid urban areas.