Hello Thierry
Le 06/01/2020 à 23:02, Thierry Danard a écrit :
We have one more below that gives a bigger error (still in Alaska!)
than we expected. Do you have any idea?
Just tested. That transformation uses the NTv2_0.gsb datum shift grid
file. The NTv2 format can contain many grids inside the same file.
Sometime the first grid covers the whole country with a coarse
resolution, and other grids cover smaller areas with finer resolution.
The NTv2 support in Apache SIS is currently built on this assumption: it
only reads the first grid contained in an NTv2 file assuming that it is
the one covering the larger area. In the case of NTv2_0.gsb file, this
assumption is not valid: that file seems to contain 114 grids, and I'm
not sure it has one covering the whole country. Apache SIS pickup the
first grid, which appear to be further on the East side of Canada, and
the tested point is outside that grid.
To fix this issue we need to read all sub-grids in NTv2 files. The NTv2
reader of SIS is actually already capable of that, but to finish that
work cleanly we need an R-Tree or QuadTree. I may try to implement a
"raw" approach this weekend without R-Tree/QuadTree for now, at the
sacrifice of performance, and revisit later when we have better index. I
will see if I can do that Saturday or Sunday and keep you informed.
Regards,
Martin
