Multipolygon producing unexpected sum - antimeridian crossing scenario about geoblaze HOT 5 CLOSED

Got tests fixed on my local. Solution was to add a few more logical steps:

• filtering out geometries that don't cross the bounds of the raster (e.g., the right-hand geometry in the test case technically is just outside the bounds of the raster, so is filtered out)
• union/combine sampling areas expressed in pixel coordinates (avoiding double counting pixels)
• recompute bboxes of sampling areas in spatial reference system coordinates after previous step

Still will probably take a week or so to clean up and republish a new version of geoblaze. Just wanted to keep you up to date :-)

from geoblaze.

Hello. Appreciate you catching this. I can confirm the issue. The problem is here: https://github.com/GeoTIFF/geoblaze/blob/master/src/intersect-polygon/intersect-polygon.module.js#L54

Because geotiffs that cross the antimeridian generally use less than 1% of pixels in the overall extent, this triggers geoblaze to disaggregate the extent and sample around each sub-geometry. This is done to preserve memory.

There currently is no process for checking for overlap between sample areas because I have naively assumed the sub-geometries+samples would be far away from each other and not covering the same pixel.

Put simply, this wasn't a test case I had thought about and I'm very appreciative for you bringing this to my attention :-)

This can definitely be fixed by adding some union/de-confliction logic for sample areas, but may take a week or two.

Let me know if you need a fix on a quicker time table and I can push a patch that allows you to over-ride the fancy sub-sampling memory-safe logic.

from geoblaze.

technically got the union of some of the sub-geometry bboxes, but issue is that if the geometries are really small (not covering half a pixel), the size of the bbox will be 0-width or 0-height. I think we can get around the issue with this particular step by padding the snapping to the image space by 1 pixel on each side.

from geoblaze.

A week or two should be fine, thanks Daniel

from geoblaze.

Confirmed fixed in 2.5.0

from geoblaze.