Create Multilines from Points, grouped by ID with sf package

Question

I have a collection of Points, with a LINEID and an ID_SEQ. The LINEID determines the unique Line-IDS, while the ID_SEQ determines the order of the points in a Line-ID.

I want to transform Points to Lines, ordered by ID_SEQ and grouped by LINEID.

With the package sp, I am able to achieve the desired results, but I want to do it with the sf package. What am I missing here?

Here is some dummy data, the desired result illustrated with sp-functions and two attempts to do the same with sf, but the plots show different results.

library(sp)
library(sf)

sfpoints <- {
structure(list(LINEID = c(4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 
4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 10L, 10L, 
10L), ID_SEQ = c(10L, 11L, 12L, 13L, 16L, 30L, 31L, 32L, 33L, 
34L, 35L, 36L, 37L, 38L, 39L, 40L, 41L, 42L, 43L, 44L, 45L, 46L, 
1L, 2L, 3L), geometry = structure(list(structure(c(15.423568, 
47.06248), class = c("XY", "POINT", "sfg")), structure(c(15.423644, 
47.062523), class = c("XY", "POINT", "sfg")), structure(c(15.423691, 
47.062553), class = c("XY", "POINT", "sfg")), structure(c(15.423712, 
47.06257), class = c("XY", "POINT", "sfg")), structure(c(15.423716, 
47.062576), class = c("XY", "POINT", "sfg")), structure(c(15.423712, 
47.062588), class = c("XY", "POINT", "sfg")), structure(c(15.423731, 
47.062595), class = c("XY", "POINT", "sfg")), structure(c(15.423779, 
47.062626), class = c("XY", "POINT", "sfg")), structure(c(15.423835, 
47.062664), class = c("XY", "POINT", "sfg")), structure(c(15.423879, 
47.062714), class = c("XY", "POINT", "sfg")), structure(c(15.423897, 
47.062767), class = c("XY", "POINT", "sfg")), structure(c(15.423862, 
47.062828), class = c("XY", "POINT", "sfg")), structure(c(15.423783, 
47.062897), class = c("XY", "POINT", "sfg")), structure(c(15.423681, 
47.062973), class = c("XY", "POINT", "sfg")), structure(c(15.423564, 
47.06306), class = c("XY", "POINT", "sfg")), structure(c(15.423437, 
47.063164), class = c("XY", "POINT", "sfg")), structure(c(15.42331, 
47.06327), class = c("XY", "POINT", "sfg")), structure(c(15.423186, 
47.063385), class = c("XY", "POINT", "sfg")), structure(c(15.423062, 
47.063496), class = c("XY", "POINT", "sfg")), structure(c(15.422941, 
47.063602), class = c("XY", "POINT", "sfg")), structure(c(15.422821, 
47.063717), class = c("XY", "POINT", "sfg")), structure(c(15.422699, 
47.063824), class = c("XY", "POINT", "sfg")), structure(c(15.422518, 
47.061687), class = c("XY", "POINT", "sfg")), structure(c(15.422617, 
47.06179), class = c("XY", "POINT", "sfg")), structure(c(15.422717, 
47.061893), class = c("XY", "POINT", "sfg"))), class = c("sfc_POINT", 
"sfc"), precision = 0, bbox = structure(c(15.422518, 47.061687, 
15.423897, 47.063824), .Names = c("xmin", "ymin", "xmax", "ymax"
), class = "bbox"), crs = structure(list(epsg = NA_integer_, 
proj4string = NA_character_), .Names = c("epsg", "proj4string"
), class = "crs"), n_empty = 0L)), .Names = c("LINEID", "ID_SEQ", 
"geometry"), row.names = c(NA, -25L), class = c("sf", "data.frame"
), sf_column = "geometry", agr = structure(c(NA_integer_, NA_integer_
), .Names = c("LINEID", "ID_SEQ"), .Label = c("constant", "aggregate", 
"identity"), class = "factor"))}

par(mfrow=c(1,3))

## SP - way
tstssp <- as(sfpoints, "Spatial")
tstssp <- SpatialLines(lapply(split(tstssp, tstssp$LINEID), function(x) 
  Lines(list(Line(coordinates(x))), x$LINEID[1L])))
plot(tstssp, col=1:2, lwd=3, main="SP-Desired Result")


## SF - way ???
tst <- sfpoints %>% 
  group_by(LINEID) %>% 
  st_coordinates() %>%
  st_linestring()
plot(st_geometry(tst), col=1:2 , main="SF-Wrong Result")

tst <- sfpoints %>% 
  group_by(LINEID) %>%
  summarise() %>%
  st_cast("LINESTRING")
plot(st_geometry(tst), col=1:2, main="SF-Wrong Result")

Gong

I had a very similar problem a few days ago. There are a few issue reports on the sf github: github.com/r-spatial/sf/issues/325 and github.com/r-spatial/sf/issues/321. st_union doesn't seem to pay attention to the order in your data frame, instead sequencing based on which points are nearest each other. – Glowworm

In my case I ended up splitting my data frame by ID and made use of shapes_df_as_sfg from the gtfsr package, then bound it all back together and cast as a linestring. Unfortunately, I'm having trouble adapting that to your sf – Glowworm

Thanks for your reply. I saw those github issues already, but they didnt help in my case. The only thing I can do is to change to sp-structure, split the points to lines and then transform back to sf.. But I dont see st_union taking the nearest next points, as this would produce the correct result in plot 3, so I am still wondering how it unions the points together, that outputs this zigzag structure. – Gong

My mistake in saying it's the nearest points, I think it actually sorts by increasing longitude values, which is how you end up with the zigzag – Glowworm

That would explain the weird shape. Too bad i cannot pass the ID_SEQ to tell him the correct ordering.. – Gong

Answer 1

As Edzer explained in this issue, you have to supply the summarise function with the argument do_union = FALSE:

tst <- sfpoints %>% 
  group_by(LINEID) %>%
  summarise(do_union = FALSE) %>%
  st_cast("LINESTRING")
plot(st_geometry(tst), col=1:2)

Nihon

Beautiful!! I read that issue several times and was missing the crucial part do_union = FALSE. Thank you very much!!! – Gong