Converting points to polygons by group

Question

I have a dataframe with around 40 points. They are grouped by an ID variable (their name), which four points in each group. I'm trying to turn these points into polygons, but I've had a lot of trouble with different methods in both sf and sp — currently I have them as a large sf dataframe, but I can't seem to find any function that would use a grouping variable to create polygons.

I've tried using the st_cast method, the general idea being

polypoints <- polypoints %>% 
    group_by("ID") %>% 
    st_cast("MULTIPOINT") %>% 
    st_cast("MULTILINESTRING") %>% 
    st_cast("MULTIPOLYGON")

but the step from multipoint to multiline string leaves the geography empty — what are the steps for a process like this?

Answer 1

You need to summarise after your group_by statement then your approach works perfectly fine. Directly from POINTS --> POLYGON, and keeping the crs (if there is one).

set.seed(999)
xy = data.frame(x=runif(24), y=runif(24))
xy$ID = rep(1:6, each = 4)
head(xy)
xys = st_as_sf(xy, coords=c("x","y"))
polys = xys %>% 
  dplyr::group_by(ID) %>% 
  dplyr::summarise() %>%
  st_cast("POLYGON")
plot(polys)

If you want the outer polygon you can add st_convex_hull()

polys = polys %>% 
  st_convex_hull()
plot(polys)

Answer 2

Make a reproducible example:

> set.seed(999)
> xy = data.frame(x=runif(24), y=runif(24))
> xy$ID = rep(c(1:6), rep(4,6))
> head(xy)
           x         y ID
1 0.38907138 0.8260703  1
2 0.58306072 0.8195141  1
3 0.09466569 0.5684927  1
4 0.85263123 0.6196068  1
5 0.78674676 0.8308805  2
6 0.11934226 0.4588336  2

Make the data frame into an sf data frame:

> xys = st_as_sf(xy, coords=c("x","y"))

Then aggregate by ID, combine the points and cast to POLYGON, turn the whole thing into an sf data frame. In a one-liner:

> polys = st_sf(
  aggregate(
    xys$geometry,
    list(xys$ID),
    function(g){
       st_cast(st_combine(g),"POLYGON")
    }
   ))

> polys
Simple feature collection with 6 features and 1 field
geometry type:  POLYGON
dimension:      XY
bbox:           xmin: 0.03014558 ymin: 0.01902308 xmax: 0.9875201 ymax: 0.8690149
epsg (SRID):    NA
proj4string:    NA
  Group.1                       geometry
1       1 POLYGON ((0.3890714 0.82607...
2       2 POLYGON ((0.7867468 0.83088...
3       3 POLYGON ((0.3907724 0.52724...
4       4 POLYGON ((0.03014558 0.8162...
5       5 POLYGON ((0.1665847 0.10961...
6       6 POLYGON ((0.9074913 0.35951...

Looks pretty awful but that's random data for you...

Answer 3

The answer provided by @Spacedman doesn't work any in newer versions of sf. I think the issue is that aggregate automatically handles the union of the geometries and the FUN argument is intended to handle aggregating the non-geometry columns? I also think st_combine produces an sfc output, so the geometry column ends up as a list of sfc objects, not a single sfc object?

The following approach works in sf 1.0-1:

set.seed(999)
xy = data.frame(x=runif(24), y=runif(24))
xy$ID = rep(c(1:6), rep(4,6))
xys = st_as_sf(xy, coords=c("x","y"))
Aggregate - which unions the points into MULTIPOINTS (default do_union=TRUE)
and uses FUN to assign the first value in other fields
xymp  = st_sf(
  aggregate(
    xys,
    by=list(ID=xys$ID),
    FUN=function(vals){vals[1]}))
Cast to poly
xypoly = st_cast(xymp, 'POLYGON')
.or (more usefully?) convex hull
xychull <- xymp
st_geometry(xychull) <- st_convex_hull(xymp$geometry)
plot(xychull['ID'])

Answer 4

The code below shows that the solution proposed by David_O does not work properly (i.e. does not keep the order of the points provided in the dataframe):

library(sf)
xy = data.frame(x = c(0,1,1,0,0,1,2,2,1,1),
  y = c(0,0,1,1,0,0,0,1,1,0), 
  ID=c(rep(1,5),rep(2,5)))
xys = st_as_sf(xy, coords=c("x","y"))
Aggregate - which unions the points into MULTIPOINTS (default do_union=TRUE)
and uses FUN to assign the first value in other fields
xymp  = st_sf(
  aggregate(
    xys,
    by=list(ID=xys$ID),
    FUN=function(vals){vals[1]}))
Cast to poly
xypoly = st_cast(xymp, 'POLYGON')
plot(xypoly['ID'])

In order for this solution to work, one has to add the option do_union = FALSE in the call to aggregate:

library(sf)
xy = data.frame(x = c(0,1,1,0,0,1,2,2,1,1),
  y = c(0,0,1,1,0,0,0,1,1,0), 
  ID=c(rep(1,5),rep(2,5)))
xys = st_as_sf(xy, coords=c("x","y"))
Aggregate - which unions the points into MULTIPOINTS (default do_union=TRUE)
and uses FUN to assign the first value in other fields
xymp  = st_sf(
  aggregate(
    xys,
    by=list(ID=xys$ID),
    do_union=FALSE,
    FUN=function(vals){vals[1]}))
Cast to poly
xypoly = st_cast(xymp, 'POLYGON')
plot(xypoly['ID'])

It seems to be because the default option of aggregate (do_union = TRUE) calls st_union, and that st_union sorts its result by y.