3. Manipulating Simple Feature Geometries

For a better version of the sf vignettes see https://r-spatial.github.io/sf/articles/

knitr::opts_chunk$set(fig.height = 4.5)
knitr::opts_chunk$set(fig.width = 6)
knitr::opts_chunk$set(collapse = TRUE)

This vignette describes how simple feature geometries can be manipulated, where manipulations include

Type transformations

This sections discusses how simple feature geometries of one type can be converted to another. For converting lines to polygons, see also st_polygonize below.

For single geometries

For single geometries, st_cast will

  2. convert from MULTIXX to XX if MULTIXX has length one (else, it will still convert but warn about loss of information)
  3. convert from MULTIXX to XX if MULTIXX does not have length one, but it will warn about the loss of information
  4. convert GEOMETRYCOLLECTION of length one to its component if

Examples of the first three types are:

st_point(c(1,1)) %>% st_cast("MULTIPOINT")
st_multipoint(rbind(c(1,1))) %>% st_cast("POINT")
st_multipoint(rbind(c(1,1),c(2,2))) %>% st_cast("POINT")

Examples of the fourth type are:

st_geometrycollection(list(st_point(c(1,1)))) %>% st_cast("POINT")

For collections of geometry (sfc) and simple feature collections (sf)

It should be noted here that when reading geometries using st_read, the type argument can be used to control the class of the returned geometry:

shp = system.file("shape/nc.shp", package="sf")
class(st_geometry(st_read(shp, quiet = TRUE)))
class(st_geometry(st_read(shp, quiet = TRUE, type = 3)))
class(st_geometry(st_read(shp, quiet = TRUE, type = 1)))

This option is handled by the GDAL library; in case of failure to convert to the target type, the original types are returned, which in this case is a mix of POLYGON and MULTIPOLYGON geometries, leading to a GEOMETRY as superclass. When we try to read multipolygons as polygons, all secondary rings of multipolygons get lost.

When functions return objects with mixed geometry type (GEOMETRY), downstream functions such as st_write may have difficulty handling them. For some of these cases, st_cast may help modify their type. For sets of geometry objects (sfc) and simple feature sets (sf),st_cast` can be used by specifying the target type, or without specifying it.

ls <- st_linestring(rbind(c(0,0),c(1,1),c(2,1)))
mls <- st_multilinestring(list(rbind(c(2,2),c(1,3)), rbind(c(0,0),c(1,1),c(2,1))))
(sfc <- st_sfc(ls,mls))
st_cast(sfc, "MULTILINESTRING")
sf <- st_sf(a = 5:4, geom = sfc)
st_cast(sf, "MULTILINESTRING")

When no target type is given, st_cast tries to be smart for two cases:

  1. if the class of the object is GEOMETRY, and all elements are of identical type, and
  2. if all elements are length-one GEOMETRYCOLLECTION objects, in which case GEOMETRYCOLLECTION objects are replaced by their content (which may be a GEOMETRY mix again)

Examples are:

ls <- st_linestring(rbind(c(0,0),c(1,1),c(2,1)))
mls1 <- st_multilinestring(list(rbind(c(2,2),c(1,3)), rbind(c(0,0),c(1,1),c(2,1))))
mls2 <- st_multilinestring(list(rbind(c(4,4),c(4,3)), rbind(c(2,2),c(2,1),c(3,1))))
(sfc <- st_sfc(ls,mls1,mls2))

gc1 <- st_geometrycollection(list(st_linestring(rbind(c(0,0),c(1,1),c(2,1)))))
gc2 <- st_geometrycollection(list(st_multilinestring(list(rbind(c(2,2),c(1,3)), rbind(c(0,0),c(1,1),c(2,1))))))
gc3 <- st_geometrycollection(list(st_multilinestring(list(rbind(c(4,4),c(4,3)), rbind(c(2,2),c(2,1),c(3,1))))))
(sfc <- st_sfc(gc1,gc2,gc3))
class(st_cast(st_cast(sfc), "MULTILINESTRING"))

Affine transformations

Affine transformations are transformations of the type $f(x) = xA + b$, where matrix $A$ is used to flatten, scale and/or rotate, and $b$ to translate $x$. Low-level examples are:

(p = st_point(c(0,2)))
p + 1
p + c(1,2)
p + p
p * p
rot = function(a) matrix(c(cos(a), sin(a), -sin(a), cos(a)), 2, 2)
p * rot(pi/4)
p * rot(pi/2)
p * rot(pi)

Just to make the point, we can for instance rotate the counties of North Carolina 90 degrees clockwise around their centroid, and shrink them to 75% of their original size:

nc = st_read(system.file("shape/nc.shp", package="sf"), quiet = TRUE)
ncg = st_geometry(nc)
plot(ncg, border = 'grey')
cntrd = st_centroid(ncg)
ncg2 = (ncg - cntrd) * rot(pi/2) * .75 + cntrd
plot(ncg2, add = TRUE)
plot(cntrd, col = 'red', add = TRUE, cex = .5)

Coordinate reference systems conversion and transformation

Getting and setting coordinate reference systems of sf objects

The coordinate reference system of objects of class sf or sfc is obtained by st_crs, and replaced by st_crs<-:

geom = st_sfc(st_point(c(0,1)), st_point(c(11,12)))
s = st_sf(a = 15:16, geometry = geom)
s1 = s
st_crs(s1) <- 4326
s2 = s
st_crs(s2) <- "+proj=longlat +datum=WGS84"
all.equal(s1, s2)

An alternative, more pipe-friendly version of st_crs<- is

s1 %>% st_set_crs(4326)

Coordinate reference system transformations

If we change the coordinate reference system from one non-missing value into another non-missing value, the CRS is is changed without modifying any coordinates, but a warning is issued that this did not reproject values:

s3 <- s1 %>% st_set_crs(4326) %>% st_set_crs(3857)

A cleaner way to do this that better expresses intention and does not generate this warning is to first wipe the CRS by assigning it a missing value, and then set it to the intended value.

s3 <- s1  %>% st_set_crs(NA) %>% st_set_crs(3857)

To carry out a coordinate conversion or transformation, we use st_transform

s3 <- s1 %>% st_transform(3857)

for which we see that coordinates are actually modified (projected).

Geometrical operations

All geometrical operations st_op(x) or st_op2(x,y) work both for sf objects and for sfc objects x and y; since the operations work on the geometries, the non-geometry parts of an sf object are simply discarded. Also, all binary operations st_op2(x,y) called with a single argument, as st_op2(x), are handled as st_op2(x,x).

We will illustrate the geometrical operations on a very simple dataset:

b0 = st_polygon(list(rbind(c(-1,-1), c(1,-1), c(1,1), c(-1,1), c(-1,-1))))
b1 = b0 + 2
b2 = b0 + c(-0.2, 2)
x = st_sfc(b0, b1, b2)
a0 = b0 * 0.8
a1 = a0 * 0.5 + c(2, 0.7)
a2 = a0 + 1
a3 = b0 * 0.5 + c(2, -0.5)
y = st_sfc(a0,a1,a2,a3)
plot(x, border = 'red')
plot(y, border = 'green', add = TRUE)

Unary operations

st_is_valid returns whether polygon geometries are topologically valid:

b0 = st_polygon(list(rbind(c(-1,-1), c(1,-1), c(1,1), c(-1,1), c(-1,-1))))
b1 = st_polygon(list(rbind(c(-1,-1), c(1,-1), c(1,1), c(0,-1), c(-1,-1))))

and st_is_simple whether line geometries are simple:

s = st_sfc(st_linestring(rbind(c(0,0), c(1,1))), 
    st_linestring(rbind(c(0,0), c(1,1),c(0,1),c(1,0))))

st_area returns the area of polygon geometries, st_length the length of line geometries:

st_length(st_sfc(st_linestring(rbind(c(0,0),c(1,1),c(1,2))), st_linestring(rbind(c(0,0),c(1,0)))))
st_length(st_sfc(st_multilinestring(list(rbind(c(0,0),c(1,1),c(1,2))),rbind(c(0,0),c(1,0))))) # ignores 2nd part!

Binary operations: distance and relate

st_distance computes the shortest distance matrix between geometries; this is a dense matrix:


st_relate returns a dense character matrix with the DE9-IM relationships between each pair of geometries:


element [i,j] of this matrix has nine characters, referring to relationship between x[i] and y[j], encoded as $I_xI_y,I_xB_y,I_xE_y,B_xI_y,B_xB_y,B_xE_y,E_xI_y,E_xB_y,E_xE_y$ where $I$ refers to interior, $B$ to boundary, and $E$ to exterior, and e.g. $B_xI_y$ the dimensionality of the intersection of the the boundary $B_x$ of x[i] and the interior $I_y$ of y[j], which is one of {0,1,2,F}, indicating zero-, one-, two-dimension intersection, and (F) no intersection, respectively.

Binary logical operations:

Binary logical operations return either a sparse matrix


or a dense matrix

st_intersects(x, x, sparse = FALSE)
st_intersects(x, y, sparse = FALSE)

where list element i of a sparse matrix contains the indices of the TRUE elements in row i of the the dense matrix. For large geometry sets, dense matrices take up a lot of memory and are mostly filled with FALSE values, hence the default is to return a sparse matrix.

st_intersects returns for every geometry pair whether they intersect (dense matrix), or which elements intersect (sparse). Note that the function st_intersection in this package returns a geometry for the intersection instead of logicals as in st_intersects (see the next section of this vignette).

Other binary predicates include (using sparse for readability):

st_disjoint(x, y, sparse = FALSE)
st_touches(x, y, sparse = FALSE)
st_crosses(s, s, sparse = FALSE)
st_within(x, y, sparse = FALSE)
st_contains(x, y, sparse = FALSE)
st_overlaps(x, y, sparse = FALSE)
st_equals(x, y, sparse = FALSE)
st_covers(x, y, sparse = FALSE)
st_covered_by(x, y, sparse = FALSE)
st_covered_by(y, y, sparse = FALSE)
st_equals_exact(x, y,0.001, sparse = FALSE)

Operations returning a geometry

u = st_union(x)
par(mfrow=c(1,2), mar = rep(0,4))
plot(st_buffer(u, 0.2))
plot(u, border = 'red', add = TRUE)
plot(st_buffer(u, 0.2), border = 'grey')
plot(u, border = 'red', add = TRUE)
plot(st_buffer(u, -0.2), add = TRUE)
par(mfrow = c(1:2))
par(mfrow = c(1,1))
plot(st_centroid(x), add = TRUE, col = 'red')
plot(st_centroid(u), add = TRUE, col = 'red')

The intersection of two geometries is the geometry covered by both; it is obtained by st_intersection:

plot(y, add = TRUE)
plot(st_intersection(st_union(x),st_union(y)), add = TRUE, col = 'red')

Note that the function st_intersects returns a logical matrix indicating whether each geometry pair intersects (see the previous section in this vignette).

To get everything but the intersection, use st_difference or st_sym_difference:

par(mfrow=c(2,2), mar = c(0,0,1,0))
plot(x, col = '#ff333388'); 
plot(y, add=TRUE, col='#33ff3388')
title("x: red, y: green")
plot(x, border = 'grey')
plot(st_difference(st_union(x),st_union(y)), col = 'lightblue', add = TRUE)
plot(x, border = 'grey')
plot(st_difference(st_union(y),st_union(x)), col = 'lightblue', add = TRUE)
plot(x, border = 'grey')
plot(st_sym_difference(st_union(y),st_union(x)), col = 'lightblue', add = TRUE)

Function st_segmentize adds points to straight line sections of a lines or polygon object:

pts = rbind(c(0,0),c(1,0),c(2,1),c(3,1))
ls = st_linestring(pts)
ls.seg = st_segmentize(ls, 0.3)
pts = ls.seg
pol = st_polygon(list(rbind(c(0,0),c(1,0),c(1,1),c(0,1),c(0,0))))
pol.seg = st_segmentize(pol, 0.3)
plot(pol.seg, col = 'grey')

Function st_polygonize polygonizes a multilinestring, as long as the points form a closed polygon:

mls = st_multilinestring(list(matrix(c(0,0,0,1,1,1,0,0),,2,byrow=TRUE)))
x = st_polygonize(mls)
plot(mls, col = 'grey')
plot(x, col = 'grey')

Try the sf package in your browser

Any scripts or data that you put into this service are public.

sf documentation built on July 14, 2022, 5:08 p.m.