You are viewing a plain text version of this content. The canonical link for it is here.

Posted to dev@sis.apache.org by Charles Karney <ch...@karney.com> on 2018/01/14 22:14:52 UTC

Bugs in DefaultEllipsoid.java

The file

core/sis-referencing/src/main/java/org/apache/sis/referencing/datum/DefaultEllipsoid.java

contains a routine orthodromicDistance.  A more standard (and more
euphonious) name would be geodesicDistance.  In addition the statement
"The orthodromic path is always on a great circle" is wrong, since, in
general, there are no great circles on an ellipsoid.

orthodromicDistance (or geodesicDistance) uses Vincenty's algorithm
which is known to fail for nearly antipodal points.  My package
net.sf.geographiclib.GeographicLib-Java includes an algorithm which is
more accurate than Vincenty's and (more importantly) always converges.

In the case where algorithm hasn't converged, the routine attempts a
lame patch-up in the case where the two points lie on the equator.  This
computes the distance as

    abs(lon1-lon2) * getSemiMajorAxis();

This is doubly wrong:

(a) lon1-lon2 needs to be reduced to the range [-pi,pi].  The tests in
DefaultEllipsoidTest.java don't catch this blunder.

(b) the formula assumes that the geodesic runs along the equator;
however this only holds if the longitude difference is less than or
equal to (1-f) * pi.  The tests in DefaultEllipsoidTest.java bake in
this incorrect assumption.

  --Charles

Re: Bugs in DefaultEllipsoid.java

Posted by Martin Desruisseaux <ma...@geomatys.com>.

Thanks for pointing the errors in DefaultEllipsoid.orthodromicDistance
too. I filled another JIRA task for this one:

    https://issues.apache.org/jira/browse/SIS-386

In both cases (DistanceUtils and DefaultEllipsoid.orthodromicDistance),
the action could be to deprecate them and replace them by some
GeodeticCalculator (or whatever name is best suited) class.

    Martin


Le 14/01/2018 à 23:14, Charles Karney a écrit :

> The file
>
> core/sis-referencing/src/main/java/org/apache/sis/referencing/datum/DefaultEllipsoid.java
>
> contains a routine orthodromicDistance.  A more standard (and more
> euphonious) name would be geodesicDistance.  In addition the statement
> "The orthodromic path is always on a great circle" is wrong, since, in
> general, there are no great circles on an ellipsoid.
>
> orthodromicDistance (or geodesicDistance) uses Vincenty's algorithm
> which is known to fail for nearly antipodal points.  My package
> net.sf.geographiclib.GeographicLib-Java includes an algorithm which is
> more accurate than Vincenty's and (more importantly) always converges.
>
> In the case where algorithm hasn't converged, the routine attempts a
> lame patch-up in the case where the two points lie on the equator.  This
> computes the distance as
>
>     abs(lon1-lon2) * getSemiMajorAxis();
>
> This is doubly wrong:
>
> (a) lon1-lon2 needs to be reduced to the range [-pi,pi].  The tests in
> DefaultEllipsoidTest.java don't catch this blunder.
>
> (b) the formula assumes that the geodesic runs along the equator;
> however this only holds if the longitude difference is less than or
> equal to (1-f) * pi.  The tests in DefaultEllipsoidTest.java bake in
> this incorrect assumption.
>
>   --Charles