Ground covered by one grid cell of a patch at this level.
The patch's geometric error, and the thing the whole selection turns on, so
it is one function: terrainMesh sizes a patch's morph band with it and the
traversal decides to refine with it. A disagreement between those two puts a
patch's handover somewhere other than where its neighbor expects it, which is
a crack.
It is the nominal size for the level, not the region's measured span, and that is deliberate. The gnomonic map is not equal-area — a cell at the middle of a cube face covers up to 2.1× the ground of one at a corner — so measuring each region gives a strictly better description of how big a patch is, and correcting for it is what Zucker & Higashi (JCGT 2018) is about. It was measured here, and it breaks the property the whole phase is for.
The no-crack argument is: a patch is fully morphed onto its parent's grid wherever a coarser patch abuts it, because the coarser one did not refine and so is at least its own range away. That needs the finer patch's handover distance to be under the coarser patch's range — and with measured spans, those two are measured at different points on the face, where the scale differs by up to 22% at level 2. A patch would be 15% short of its neighbor's grid, which is a lit gap. The distortion is smooth in position and the level is not, and no per-node number can be both.
So the correction is not applied, and what it costs is stated rather than hidden: a cell near a cube corner is refined as though it were up to twice its real size, which is one extra level over the eight regions around the cube's corners. Measured across the zoo that is 6–11% more patches — against a hairline seam wherever two levels meet.
- radiusnumber
- Undocumented
- regionRegionAddress
- Undocumented
- resolutionnumber
- Undocumented