Let’s start with tiles: If you’ve used interactive maps or ‘slippy maps’ as they are sometimes known, you’re probably aware of tiles. A slippy map (E.g. Google Maps) is comprised of thousands to millions of little images, a selection of which are served to you depending on the spatial viewport you have established on the map as you slip-slide around.
These solve the problem of how to serve a high resolution map image to the user. High resolution images of the Earth use a ton of data and bandwidth to transfer, but the typical user is only chasing an image of a very small section that contains whatever they are interested in.
So some clever person came up with the idea of cutting up the images of the earth into three dimensional dataset of x, y, and zoom. A tileset, as they are called, contains the whole Earth projected onto an x/y plane, rendered at progressively higher resolution or higher detail, with increasing zoom
As we ‘zoom-in’, or increase our position on the zoom dimension, we are sent only the tile images for our new viewport of the Earth, which contains roughly the same amount of tiles, but those tiles represent increasingly less area of the Earth’s surface. It follows that the number of tile images associated with each zoom level increases geometrically with increasing zoom (by a factor of 4).
This information is combined with a separately transmitted stylesheet to produce the tile image. Different stylesheets can use the same vector data to produce radically different looking maps that either highlight or omit data with certain attributes.
As mappers of data we can use the same trick of streaming tiled sections of our data in vector form to drastically boost the amount of information we can make available on interactive maps.
Mapbox vector tiles
A vector tile specfication created by the Mapbox company is currently the de-facto standard for vector tile files. Mabbox Vector Tiles, or ‘MVT’ as they are often abbreviated, are stored as a Google protocol buffer - a tightly packed binary format.
MBTiles databases
MBTiles is a second specification created by Mapbox that describes a method of storing an entire MVT tileset inside a single file. Internally .mtbtiles files are SQLlite databases containg two tables: metadata and tiles.
The tiles table is indexed by z,x,y and contains a tile_data column for the vector tile protocol buffers, which are stored compressed using gzip.
This single-file tileset is much easier to ship around than millions of tiny files, and the SQLite format and gzip compression help with efficient retrieval and transmission.
The guide later in this blog post will focus on generating, serving, and viewing MVT tilesets in .mbtiles format.
At present, as far as I am aware, there is only one package in the R ecosystem which has a map widget that can render Mapbox vector tiles: {rdeck}. {rdeck} is a deck.gl wrapper created by my colleague Anthony North, and has been my team’s main interactive mapping workhorse for the last couple of years.
For this example we’ll serve the tiles with {mvtview}, a package I put together that provides a basic MVT server and viewer powered by {plumber} and {rdeck} respectively. It’s meant as a development tool, to help you iterate on tippecanoe parameters, not as a production solution.
GIS
