Point simplification at low zoom levels?

[mapsam]

For very large point datasets, it'd be great to build a path for simplifying them at low zoom levels like z0, z1, z2 to avoid huge individual tiles. I'm not sure what the best route is, but perhaps we can take a clue from Tippecanoe?

cc @flippmoke @ericfischer

[e-n-f]

The Tippecanoe mechanism is pretty straightforward if you want to adapt it:

• calculate the inverse of the fraction of the features that should be kept in each zoom level based on the droprate (default 2.5) and basezoom (default same as maxzoom).
• sort the point features by quadkey
• step through the features in order, and for each of them increment the per-zoom counters and either give the feature a minzoom or let it through.

The thing that will be hard if you're only working one tile at a time is to keep the features that are preserved consistent from one zoom to the next. Even without that, I think dropping in quadkey order will still tend to look better than random dropping (which develops clumps).

[flippmoke]

It is possible to write a custom algorithm to simplify multipoints I believe, but I think we might struggle to do this with sets of features. The reasoning for this is the current processing pipeline doesn't know about other features when simplification takes place. https://github.com/mapbox/mapnik-vector-tile/blob/master/src/vector_tile_geometry_simplifier.hpp#L32

[e-n-f]

I don't do anything special in Tippecanoe with MultiPoints (it either includes or drops the entire feature) but I probably should. I got into the habit of thinking that nobody used MultiPoint geometries because Mapnik would only render the first point (not sure if that's still true).