How to remove "almost holes" from polygons using R sf? - r

Question

I have a polygon with some holes and some "almost holes" that are long, skinny inlets into the polygon. (polygon files HERE)
field <- sf::read_sf("example_field.kml")
ggplot() +
geom_sf(data = field) +
theme_void()
I can remove the one "true" hole from the polygon using:
field_no_holes <- nngeo::st_remove_holes(field)
ggplot() +
geom_sf(data = field_no_holes) +
theme_void()
However, that still leaves several long, skinny "almost holes". Any ideas about how to efficiently remove these? For reference, here is the what I'm going for as an end goal (I created this by manually deleting vertices).
field_fixed <- sf::read_sf("example_field_fixed.kml") %>%
nngeo::st_remove_holes()
ggplot() +
geom_sf(data = field_fixed) +
theme_void()

Answer 1

A couple of ways to do this below. You'll have to figure out how to play around with things to figure out how to define "skinny".
First is based on the #dieghernan comment, with an added step of un-buffering to return the shape to approximately the same size it started as:
libarary(tidyverse)
library(sf)
x <- read_sf('example_field.kml')
x %>%
st_transform(3857) %>% # crs 3857 allows buffering in meters
st_buffer(dist = 20) %>% # make the polygon 20m bigger to get rid of 'skinny' holes
st_buffer(dist = -20) %>% # shrink it back down to approximately the right size
ggplot() + # plotting
geom_sf(fill = 'black') +
geom_sf(data = x, fill = NA, col = 'red') +
theme_void()
Original polygon in red outline, new polygon filled in black:
This could run into some problems with concave points that aren't quite the right kind of 'skinny'. See the dent in the bottom right.
Another more laborious solution that might not get stuck on the 'skinny-ness' of holes. Take a regular sample of the polygon as points, then remove those that are too far from a buffered and un-buffered polygon. Turn the points that are leftover into a polygon.
# regular sample of 1000 points of the polygon
points <- x %>%
st_cast('LINESTRING') %>%
st_sample(size = 1000, type = 'regular') %>%
st_zm() %>% # remove z
st_cast('POINT')
# similar to solution 1, buffer & un-buffer original polygon
buffered <- x %>%
st_transform(3857) %>%
st_buffer(20) %>%
st_buffer(-20) %>%
st_transform(st_crs(points)) %>%
st_cast('LINESTRING')
# select sampled points that are very near the buffer,
# then cast them to a polygon
poly_from_points <- points[st_is_within_distance(points, buffered, dist = 5, sparse = F)] %>%
st_combine() %>%
st_cast('LINESTRING') %>%
st_cast('POLYGON')
With the example file given, this method might not be necessary. There could be edge cases where this is less likely to fail though.
Image below is zoomed in on the notch at the bottom right (south-east) to show the differences in the two methods along with the original polygon. Red is buffering & un-buffering (method 1), blue is only points from original polygon (method 2), and near-black is the original polygon.