I finally got to read the ISO 19107 standard that is referenced in the
OGC standard to define polygons. Apparently, the GM_Polygon contains GM_SurfaceBoundary
and GM_Surface (as exterior and interior rings). In turn, GM_Surface is a GM_Ring,
that in turn is a GM_Curve with other properties, that is in turn a
sequence of GM_CurveSegments.
And we finally got it. A GM_CurveSegment is an abstract type that has GM_LineString
and GM_GeodesicString as successors. GM_LineString is a sequence of GM_LineSegments
and GM_GeodesicString is a sequence of GM_Geodesics.
A GM_LineSegment has two points and must be interpolated linearly.
A GM_Geodesic also has two points but must be interpolated with geodesic curves.
In our case, this means that, as far as the standard specifies, if a polygon contain a
a GM_LineString as its boundary, the interpolation must be done linearly in
the lat-lon plane, specially if no interpolation attribute is present.
On the other hand, if the Polygon contains a GM_GeodesicString, the interpolation
must be done by using the geodesic curves of the associated spacial referenced
(WPS84 ellipsoid in our case).
What is strange to me is that, as far as my little experience in this field
can tell, Polygon exteriors are always defined in terms of a LinearRing, and
nothing is said about the interpolation method. This means that according to the
standard, linear interpolation must be used. I guess that this is an extended
practice, and when someone needs to interpret the polygon with geodesic interpolation,
this is done in the implementation, not complying with the standard.
And with KML, things get more confusing, since
a polygon can only contain LineStrings (there is not such thing as GeodesicString).
In this link, there is even a note that tells:
"Note: In Google Earth, a Polygon with an of clampToGround follows
lines of constant bearing; however, a LinearRing (by itself) with an
of clampToGround follows great circle lines."
That is, Google Earth shows great circle lines to join polygon points, and there is
no explicit way to specify the semantics of the polygon lines. I think this is
a not minor issue of KML as a geographical data standard.