Fix error in ReprojectRasterExtent

[jpolchlo]

Overview

While investigating a reprojection issue in geotrellis-contrib, it came to my attention that the raster extents produced by ReprojectRasterExtent are off. Consider the following: using https://github.com/geotrellis/geotrellis-contrib/blob/master/vlm/src/test/resources/img/aspect-tiled.tif as the base of this example, the following definitions apply:

val tiff = GeoTiffReader.readMultiband(getByteReader("aspect-tiled.tif"), streaming = true)
val targetRE = ReprojectRasterExtent(tiff.rasterExtent, tiff.crs, WebMercator)
val region = tiff.projectedExtent.reprojectAsPolygon(WebMercator, 0.001)

We then note that targetRE.extent and region.envelope diverge:

targetRE.extent: geotrellis.vector.Extent = Extent(-8769150.640916323, 4257706.9716441, -8750638.524926396, 4274455.441673627)
region.envelope: geotrellis.vector.Extent = Extent(-8769150.640916323, 4257707.61418559, -8750636.281752571, 4274455.441673627)

Note that ymin and xmax differ, and that the tolerance value of 0.001 has no effect, since the corners of the region are on the extent boundary.

I believe that the region is the authoritative source of truth. So, the reprojected raster extent should tightly bound this polygon.

Signed-off-by: jpolchlo [email protected]

Checklist

• docs/CHANGELOG.rst updated, if necessary
• Unit tests added for bug-fix or new feature

[jpolchlo]

As of now, some of the original tests for ReprojectRasterExtent are failing due to small variations introduced by this PR. I do believe that the old implementation was incorrect and needed to be fixed, but this is leading to a slight divergence between GDAL and GT, which is what the failing tests check against. We'll need some discussion as to whether the new results are precisely what we intend them to be and whether targeting strict agreement with GDAL is the right thing to do.

[pomadchin]

Holding this until geotrellis/geotrellis-contrib#46 question would be resolved.