Area-based interpolation

[asinghvi17]

See https://gistbok.ucgis.org/bok-topics/areal-interpolation and https://github.com/chris-prener/areal.

The most basic form is simple and can be done already. Consider a vector of polygons polys, associated values values, and a polygon that we want to interpolate, p1.

In this case, we can simply execute GO.area.(GO.intersection.(polys, (p1,), target = PolygonTrait()) ./ GO.area.(polys) .* values with some normalization factor. However, spatially aware versions of this also exist that could be integrated into Rasters.jl.

That means that area-based interpolation should probably be its own package, accepting GeoInterface geometries and potentially also RTree/STRtree objects to speed up intersection.

FeatureCollections would be decomposed to tables on input. The main entry point should probably return a GI feature. GO.reconstruct would be helpful here.

---

In order to get raster statistics, we could use something like Rasters.zonal(sum, provided_data; of = polygon). This provides a nice accumulator interface without many allocations. We could also use Stencils.jl to do some stencilling on a fully populated raster, if that proves necessary.

If we do end up using Rasters, then we must mention in many places a method by which you can create a Raster from a regular array.

[asinghvi17]

I just pushed a basic framework up. It still needs support for:

• intensive vs extensive variables
• how to reduce (could be sum or mean or something else)
• more settings/tuning