---
layout: post
title: "Higher-order geometry differences and intersections"
date: "`r format(Sys.time(), '%d %B, %Y')`"
comments: true
author: Edzer Pebesma
categories: r
---
TOC
[DOWNLOADHERE]
Suppose you have the following geometry, consisting of three overlapping square polygons:
```{r echo=TRUE, fig=TRUE, fig.path = "images/", label="geoms1"}
library(sf)
pol = st_polygon(list(rbind(c(0,0), c(1,0), c(1,1), c(0,1), c(0,0))))
b = st_sfc(pol, pol + c(.8, .2), pol + c(.2, .8))
par(mar = rep(0, 4))
plot(b, col = NA)
```
and you are interested in the area where all squares overlap (green),
or where exactly two squares overlap (orange):
```{r echo=TRUE, fig=TRUE, fig.path = "images/", label="geoms2"}
i = st_intersection(st_sf(b))
par(mar = rep(0, 4))
cl = sf.colors(3, categorical = TRUE)
plot(b)
plot(i[i$n.overlaps == 3,2], col = cl[1], add = TRUE)
plot(i[i$n.overlaps == 2,2], col = cl[2], add = TRUE)
```
So far, with package `sf` or `rgeos` you could only get
pairwise intersections, meaning you would have to go through
all pairwise intersections and see whether they are intersected
by others geometries or intersections. In [this StackOverflow
question](https://stackoverflow.com/questions/44631044/efficient-extraction-of-all-sub-polygons-generated-by-self-intersecting-features)
you can get an idea how ugly this can get.
## st\_intersection
Now, inspired by a meticulously prepared [pull
request](https://github.com/r-spatial/sf/pull/598) by [Jeffrey
Hanson](http://jeffrey-hanson.com/), it suffices to do
```{r}
(i = st_intersection(b))
```
to get all the unique pieces, for each unique piece the number
of contributing geometries, and a list-column with indexes of the
geometries that contribute (overlap) for a particular piece.
## st\_difference
The pull request Jeffrey wrote was to remove (erase) all overlapping
pieces, which you now get by
```{r echo=TRUE, fig=TRUE, fig.path = "images/", label="geoms3"}
d = st_difference(b)
plot(d, col = cl)
```
For this latter approach, obviously the input order matters: what is returned
are non-empty geometries with $x_1$, $x_2 - x_1$, $x_3 - x_2 - x_1$ etc.
To prove that these intersections or differences do not have any overlaps, we
can compute overlaps by
```{r}
st_overlaps(i)
st_overlaps(d)
```
## Further reading
Jeffrey's [pull request](https://github.com/r-spatial/sf/pull/598)
is worth reading; the sf [pkgdown
site](https://r-spatial.github.io/sf/reference/geos_binary_ops.html)
contains some further examples with squares.