Merge sub-regions of multiple regions

Question

First, let's set up the data. Download adjoining Seoul, Incheon, Gyeonggi-do region shapefiles from this website (in Korean). I attach the files, as they are not accessible outside Korea. Then read into R, name, and combine (credit):

library(sf)

#Read, name and combine regions
sca_nsdi <- rbind(cbind(st_read("LARD_ADM_SECT_SGG_11.shp"), Name="Seoul"), 
                  cbind(st_read("LARD_ADM_SECT_SGG_28.shp"), Name="Incheon"), 
                  cbind(st_read("LARD_ADM_SECT_SGG_41.shp"), Name="Gyeonggi-do"))

#Plotting reveals sub-regions (the legend was natively clipped)
plot(sca_nsdi[, "Name"])

I wish to merge the sub-regions of these 3 regions. I have tried applying this solution, but without success:

library(dplyr)

sca_nsdi <- sca_nsdi %>% group_by(Name) %>% summarise(geometry=st_union(geometry)) %>% 
  ungroup()

plot(sca_nsdi[, "Name"])

The sub-regions are imperfectly merged. (For some reason, the bounding box is also shrunk to the size of Seoul.)

Answer 1

The first solution to dissolving problems is usually to find a set of pre-dissolved borders. Do they exist?

These borders have a lot of digitising errors if they are supposed to represent a division into contiguous areas.

Using QGIS its easy to zoom into your data and see what's happening. Here's an area where three features all overlap and cross with about a width of 20m:

and here's a massive gap of about 170 metres width:

and there are slivers everywhere:

My usual go-to tool for fixing these things is a standalone programme called pprepair https://github.com/tudelft3d/pprepair - there is some work on implementing this within R but I don't think its there yet.

Otherwise the trick is to buffer the polygons so they overlap, then do the union. Given your workflow I would buffer then union the three individual area objects first, then perhaps apply a negative buffer to shrink each one back, and only then combine them into a single object. This way you can tailor the buffer size for each of the three regions.

A quick test shows a 40m buffer works pretty well.

I don't understand why the bounding box is going wrong but I suspect its the grouped object only taking the extent of the first (or last?) group it processes... possible bug..

The pprepair process does a good job of fixing all this once you get the parameters correct. If you try this approach and struggle start a new question and ask.