r - Convert output from sf::st_within to vector

Question

Im trying to use the sf package in R to see if sf object is within another sf object with the st_within function. My issue is with the output of this function which is sparse geometry binary predicate - sgbp and I need a vector as an output so that I can use the dplyr package afterwards for filtering. Here is a simplified example:

# object 1: I will test if it is inside object 2
df <- data.frame(lon = c(2.5, 3, 3.5), lat = c(2.5, 3, 3.5), var = 1) %>% 
st_as_sf(coords = c("lon", "lat"), dim = "XY") %>% st_set_crs(4326) %>%
  summarise(var = sum(var), do_union = F) %>% st_cast("LINESTRING")

# object 2: I will test if it contains object 1
box <- data.frame(lon = c(2, 4, 4, 2, 2), lat = c(2, 2, 4, 4,2), var = 1) %>%
  st_as_sf(coords = c("lon", "lat"), dim = "XY") %>% st_set_crs(4326) %>% 
  summarise(var = sum(var), do_union = F) %>% st_cast("POLYGON")

# test 1
df$indicator <- st_within(df$geometry, box$geometry) # gives geometric binary predicate on pairs of sf sets which cannot be used 
df <- df %>% filter(indicator == 1)

This gives Error: Column indicator must be a 1d atomic vector or a list.

I tried solving this problem below:

# test 2
df$indicator <- st_within(df$geometry, box$geometry, sparse = F) %>% 
  diag() # gives matrix that I convert with diag() into vector
df <- df %>% filter(indicator == FALSE)

This works, it removes the row that contains TRUE values but the process of making a matrix is very slow for my calculations since my real data contains many observations. Is there a way to make the output of st_within a character vector, or maybe a way to convert sgbp to a character vector compatible with dplyr without making a matrix?

Answer 1

Here is how you can get a logical vector from sparse geometry binary predicate:

df$indicator <- st_within(df, box) %>% lengths > 0

or to subset without creating a new variable:

df <- df[st_within(df, box) %>% lengths > 0,]

I cannot test on your large dataset unfortunately but please let me know if it is faster than matrix approach.

Answer 2

The result of is_within is in truth a list column, so you can work your way out of this by "unlisting" it. Something like this would work:

library(dplyr)
library(sf)

# object 1: I will test if it is inside object 2 - to make this more interesting
# I added a second not-contained line
df <- data.frame(lon = c(2.5, 3, 3.5), lat = c(2.5, 3, 3.5), var = 1) %>% 
  st_as_sf(coords = c("lon", "lat"), dim = "XY") %>% st_set_crs(4326) %>%
  summarise(var = sum(var), do_union = F) %>% st_cast("LINESTRING")

df2 <- data.frame(lon = c(4.5, 5, 6), lat = c(4.5, 5, 6), var = 2) %>% 
  st_as_sf(coords = c("lon", "lat"), dim = "XY") %>% st_set_crs(4326) %>%
  summarise(var = sum(var), do_union = F) %>% st_cast("LINESTRING")
df3 <- rbind(df, df2)

# object 2: I will test if it contains object 1
box <- data.frame(lon = c(2, 4, 4, 2, 2), lat = c(2, 2, 4, 4,2), var = 1) %>%
  st_as_sf(coords = c("lon", "lat"), dim = "XY") %>% st_set_crs(4326) %>% 
  summarise(var = sum(var), do_union = F) %>% st_cast("POLYGON")

plot(df3) 
plot(st_geometry(box), add = TRUE)

# see if the lines are within the box and build a data frame with results
is_within <- st_within(df3$geometry, box$geometry) %>% 
  lapply(FUN = function(x) data.frame(ind = length(x))) %>% 
  bind_rows()

# add the "indicator" to df3
df3 <- dplyr::mutate(df3, indicator = is_within$ind) 
df3
#> Simple feature collection with 2 features and 2 fields
#> geometry type:  LINESTRING
#> dimension:      XY
#> bbox:           xmin: 2.5 ymin: 2.5 xmax: 6 ymax: 6
#> epsg (SRID):    4326
#> proj4string:    +proj=longlat +datum=WGS84 +no_defs
#>   var indicator                       geometry
#> 1   3         1 LINESTRING (2.5 2.5, 3 3, 3...
#> 2   6         0 LINESTRING (4.5 4.5, 5 5, 6 6)

HTH

Created on 2018-03-15 by the reprex package (v0.2.0).

Answer 3

Instead of using the st_within function directly, try using a spatial join. Check out the following example how st_joins works

library(sf)
library(tidyverse)

lines <-
data.frame(id=gl(3,2), x=c(-3,2,6,11,7,10), y=c(-1,6,-5,-9,10,5)) %>%
  st_as_sf(coords=c("x","y"), remove=F) %>% 
  group_by(id) %>% 
  summarise() %>%
  st_cast("LINESTRING")

yta10 <-
    st_point(c(0, 0)) %>%
    st_buffer(dist = 10) %>%
    st_sfc() %>%
    st_sf(yta = "10m")

With a left join all lines are kept, but you can see which of them that are located inside the polygon

lines %>% st_join(yta10, left=TRUE)

An inner join (left = FALSE) only keeps the ones inside

lines %>% st_join(yta10, left=FALSE)

The latter can also be obtained by

lines[yta10,]