Error: Edge 46 crosses edge 48

[marianaosb]

Hi, I've used read_municipality(simplified = FALSE) to load the date by municipality. But when I try to make a Queen neighbours Matrix, the following erro occur:

Error in wk_handle.wk_wkb(wkb, s2_geography_writer(oriented = oriented, : Loop 0 is not valid: Edge 46 crosses edge 48

How can I fix that?

[marianaosb]

Is it correct to use the sf_use_s2(FALSE) command before running the neighborhood matrix?

[rafapereirabr]

could you please share the code with a reproducible example?

[rafapereirabr]

@marianaosb, could you please share the code with a reproducible example that returns this error ?

[gabrielmagno]

Hi @rafapereirabr !

I think I had a similar problem as @marianaosb when trying to use sf::st_union to "unify" shapes of a given group of municipalities.

My real groups are different, but to create an easy reproducible example let's say that I have all the municipalities and want to unify their shape by the region (I know this could be accomplished by using directly the states dataset, but as I mentioned this is just a toy example). This is the initial code:

library(dplyr)
library(ggplot2)
library(geobr)
library(sf)

cities <- read_municipality(code_muni="all", simplified=TRUE)

cities |> 
    group_by(abbrev_state) |> 
    summarise(
        geom = geom |> st_union()
    ) |> 
    ggplot() +
        geom_sf(fill="#2D3E50", color="#FEBF57", size=.15, show.legend = FALSE) +
        theme_minimal()

Which gives this error:

Error in `summarise()`:
ℹ In argument: `geom = st_union(geom)`.
ℹ In group 1: `abbrev_state = "AC"`.
Caused by error in `wk_handle.wk_wkb()`:
! Loop 0 is not valid: Edge 253 crosses edge 255

If I set simplified=FALSE the same error appears, but in a different group/state:

Error in `summarise()`:
ℹ In argument: `geom = st_union(geom)`.
ℹ In group 11: `abbrev_state = "MG"`.
Caused by error in `wk_handle.wk_wkb()`:
! Loop 0 is not valid: Edge 46 crosses edge 48

The way I was able to fix the error was by setting globally setting this sf flag:

sf_use_s2(FALSE)

This works both for simplified=FALSE and simplified=TRUE, but with different results (which is expected I think, considering that one is... simplified geometry, and the other is complete).

For simplified=TRUE this is the resulting map:

For simplified=FALSE this is the resulting map:

The error is gone, but the state shapes still have some internal polygons. I was able to get rid of these by using sfheaders::sf_remove_holes (or nngeo::st_remove_holes) after st_union.

When using sf_remove_holes with simplified=TRUE this is the resulting map: (notice some minor artifacts inside some states, like a straight line in Bahia, Rondônia and Roraima states)

When using sf_remove_holes with simplified=FALSE this is the resulting map: (this is the best version, with clean shapes for all states)

For reference, this is the full code of this last image map, which in my opinion is the best:

library(dplyr)
library(ggplot2)
library(geobr)
library(sf)
sf_use_s2(FALSE)

cities <- read_municipality(code_muni = "all", simplified=FALSE)

cities |> 
    group_by(abbrev_state) |> 
    summarise(
        geom = geom |> st_union() |> sfheaders::sf_remove_holes()
    ) |> 
    ggplot() +
    geom_sf(fill="#2D3E50", color="#FEBF57", size=.15, show.legend = FALSE) +
    theme_minimal()

@rafapereirabr I'm reporting all of these to (1) help other people that might be also trying to union some shapes and having issues, and (2) to help you debugging a potential bug in the shapes dataset or the library. I say potential because I'm not sure if there is really any problem with the library and datasets, but I hope that my reports could help understand better what is happening.

[rafapereirabr]

Hi @gabrielmagno , thanks a lot for sharing this. It's super helpful. When users need to do any operation of the geometry column (e.g. calculating areas, merging/dissolving borders, creating contiguity matrices etc), it is strongly recommended to download the data using simplified = FALSE.

This issue you both have reported suggests some issue with the geometry might still persist for some years even when setting simplified = FALSE. I will have a closer look at this in the next few weeks.