Geovalidation¶

Raster strategy¶

This allows constant-time, random-access lookup of locality.

1. expansion of shapefile boundaries so they partially overlap, to create fuzzing when a coordinate lies on or near a boundary
   • coordinates within 5 km of the boundary should be considered valid if they are in the locality on either side of the boundary
     • According to Brad, "In the past we have toyed with various buffers up to 10 km. But 5 km sounds good, let's go with that." _²
   • expansion performed by converting polygon vertices to polar coordinates, increasing the radius coordinate by 5 km, and converting back to rectangular
2. cylindrical projection of spherical Earth onto 2D lat/long rectangle¹, for each vector shapefile
   • *possible cylindrical projections*:
     • *Mercator projection*
     • *Lambert cylindrical equal-area projection* (uses a cylinder the same height as the sphere¹)
     • *equirectangular projection* (y-axis is latitude angle)
3. rasterization of projected shapefiles to 5 km pixels
   • when shapefiles overlap, pixel contains a locality value for both localities
4. constant-time, random-access lookup of locality using C++-style 2D array
   • lat/long mapped directly to x,y array index, allowing direct access without a log(n) sorted array/binary tree lookup
5. if exact point distance from boundary is needed, it can be determined separately using the appropriate vector shapefiles for the matched localit(ies)
   • According to Brad, "we would like to be able to determine error distance for points falling outside the boundary. However, quantifying error is much less important than simply knowing if a point is inside the buffer and can therefore be used in analyses. Perhaps quantifying error could be a background task, after all the observations have been classified as in or out." _²

¹ Lambert cylindrical equal-area projection from Wikipedia:

² e-mail from Brad on 2012-10-23