geom_sf() does not respect position argument of scale_y_continuous()

[sambweber]

I'm working with ggplot2 and geom_sf at the moment to create some maps, but I can't get the axis labels to re-position to the right using the position argument within scale_y_continuous as I would with other geometry types (e.g. geom_polygon in the example below). Is this intentional or just not supported yet for geom_sf?

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

nc = st_read(system.file("gpkg/nc.gpkg", package="sf"))
 
# axis labels still on left
 ggplot() + geom_sf(aes(fill = AREA),data=nc) + scale_y_continuous(position='right')
 
# Convert layer to sp and fortify
 nc2 = as(nc,'Spatial')
 nc2@data$id = rownames(nc2@data)
 nc2  = fortify(nc2) %>% left_join(nc2@data)
 
# axis labels correctly moved to the right
 ggplot() + geom_polygon(aes(x=long,y=lat,group=group,fill=AREA),data = nc2) + 
   scale_y_continuous(position='right') + coord_equal()

[batpigandme]

With reprex, below.

library(sf)
#> Linking to GEOS 3.6.1, GDAL 2.1.3, proj.4 4.9.3
library(ggplot2) 
library(dplyr)

nc = st_read(system.file("gpkg/nc.gpkg", package="sf"))
#> Reading layer `nc.gpkg' from data source `/Library/Frameworks/R.framework/Versions/3.5/Resources/library/sf/gpkg/nc.gpkg' using driver `GPKG'
#> Simple feature collection with 100 features and 14 fields
#> geometry type:  MULTIPOLYGON
#> dimension:      XY
#> bbox:           xmin: -84.32385 ymin: 33.88199 xmax: -75.45698 ymax: 36.58965
#> epsg (SRID):    4267
#> proj4string:    +proj=longlat +datum=NAD27 +no_defs

# axis labels still on left
ggplot() + geom_sf(aes(fill = AREA),data=nc) + scale_y_continuous(position='right')

# Convert layer to sp and fortify
nc2 = as(nc,'Spatial')
nc2@data$id = rownames(nc2@data)
nc2  = fortify(nc2) %>% left_join(nc2@data)
#> Regions defined for each Polygons
#> Joining, by = "id"

# axis labels correctly moved to the right
ggplot() + geom_polygon(aes(x=long,y=lat,group=group,fill=AREA),data = nc2) + 
  scale_y_continuous(position='right') + coord_equal()

Created on 2018-08-22 by the reprex package (v0.2.0.9000).

[clauswilke]

Indeed, the setting is currently ignored. The relevant line in the code is here:

ggplot2/R/sf.R

Line 408 in 5a37ae7

right = nullGrob()

(The right axis is simply hardcoded to nothing.)
For comparison, the relevant code for the other coords looks like this:

ggplot2/R/coord-.r

Line 72 in 5a37ae7

right = render_axis(panel_params, arrange[2], "y", "right", theme)

A similar problem exists for x axes placed on the top.

Not sure how hard it would be to fix this, but it would definitely be desirable to do so.

[clauswilke]

I guess the problem is that for coord_sf() we can't be certain that any tick labels calculated for one side of the plot will be appropriate for the other, because not every graticule that starts on the left ends on the right and vice versa. So this would require some additional work to figure out which graticules to label and with what.

[clauswilke]

We can highlight these issues by projecting into an unusual coordinate system (for the part of the world we're looking at).

library(sf)
#> Linking to GEOS 3.6.1, GDAL 2.1.3, proj.4 4.9.3
library(ggplot2) 

nc = st_read(system.file("gpkg/nc.gpkg", package="sf"))
#> Reading layer `nc.gpkg' from data source `/Library/Frameworks/R.framework/Versions/3.5/Resources/library/sf/gpkg/nc.gpkg' using driver `GPKG'
#> Simple feature collection with 100 features and 14 fields
#> geometry type:  MULTIPOLYGON
#> dimension:      XY
#> bbox:           xmin: -84.32385 ymin: 33.88199 xmax: -75.45698 ymax: 36.58965
#> epsg (SRID):    4267
#> proj4string:    +proj=longlat +datum=NAD27 +no_defs
p <- ggplot() + 
  geom_sf(aes(fill = AREA), data=nc) + coord_sf(crs = 3338)
p

ggplot_build(p)$layout$panel_params[[1]]$graticule
#> Simple feature collection with 12 features and 10 fields
#> geometry type:  MULTILINESTRING
#> dimension:      XY
#> bbox:           xmin: 0 ymin: 0 xmax: 1 ymax: 1
#> epsg (SRID):    NA
#> proj4string:    NA
#> First 10 features:
#>    degree type    degree_label                       geometry    x_start
#> 1     -84    E 84 * degree * W MULTILINESTRING ((0.402943 ... 0.40294301
#> 2     -82    E              NA MULTILINESTRING ((1 0.09396... 1.00000000
#> 3     -80    E              NA MULTILINESTRING ((1 0.35286... 1.00000000
#> 4     -78    E              NA MULTILINESTRING ((1 0.60422... 1.00000000
#> 5     -76    E              NA MULTILINESTRING ((1 0.84879... 1.00000000
#> 6      32    N              NA MULTILINESTRING ((0.8528854... 0.85288544
#> 7      33    N              NA MULTILINESTRING ((0.587803 ... 0.58780298
#> 8      34    N              NA MULTILINESTRING ((0.3201988... 0.32019880
#> 9      35    N              NA MULTILINESTRING ((0.0499912... 0.04999124
#> 10     36    N 36 * degree * N MULTILINESTRING ((0 0.19757... 0.00000000
#>       y_start     x_end     y_end angle_start angle_end plot12
#> 1  0.00000000 0.0000000 0.1169753   150.39212 150.39212   TRUE
#> 2  0.09396438 0.0000000 0.3642504   152.11761 152.11761  FALSE
#> 3  0.35286919 0.0000000 0.6037669   153.84310 153.84310  FALSE
#> 4  0.60422472 0.0000000 0.8362994   155.56859 155.56859  FALSE
#> 5  0.84879350 0.2926339 1.0000000   157.29408 157.29408  FALSE
#> 6  0.00000000 1.0000000 0.1403366    61.22872  62.37905  FALSE
#> 7  0.00000000 1.0000000 0.4039505    60.70585  64.15683  FALSE
#> 8  0.00000000 1.0000000 0.6872251    60.18297  66.14375  FALSE
#> 9  0.00000000 1.0000000 0.9962199    59.66010  68.33983  FALSE
#> 10 0.19757586 0.7499034 1.0000000    60.91500  68.02611   TRUE

Created on 2018-08-22 by the reprex package (v0.2.0).

[sambweber]

Yes, I see the issue. It is more complicated than with other coords. I don't know much about how axes are rendered in ggplot, but in the case that position = 'right' is called would it be possible to place tick marks/labels at y_end where 0 < y_end < 1 in the ggplot_build() output above (i.e 32 - 35 N in this case), rather than at y_start? Is there a hack to do something along these lines by modifying the output of ggplot_build() while a more elegant solution can be found?

[clauswilke]

I've made a branch that can draw labels on all sides: https://github.com/clauswilke/ggplot2/tree/issue-2846-coord_sf It's not configurable yet, though. You can look at the changes and hack your own ggplot2 that draws labels where you need them.

@hadley There's a question about API design here: Should coord_sf() accept secondary axes? I assume not, in which case where labels should be drawn would be a coord setting rather than a scale setting. Making this a coord setting would be quite easy. Carrying all the info properly through from the scales would be more work. I'm also not clear how manual tick placement and labeling works with coord_sf() and primary/secondary axes. Should coord_sf() support the most general case where I can set up a secondary axis with its own ticks and labels? Right now, coord_sf() doesn't even allow manual setting of labels, only of tick positions.

library(sf)
#> Linking to GEOS 3.6.1, GDAL 2.1.3, proj.4 4.9.3
library(ggplot2) 

nc = st_read(system.file("gpkg/nc.gpkg", package="sf"))
#> Reading layer `nc.gpkg' from data source `/Library/Frameworks/R.framework/Versions/3.5/Resources/library/sf/gpkg/nc.gpkg' using driver `GPKG'
#> Simple feature collection with 100 features and 14 fields
#> geometry type:  MULTIPOLYGON
#> dimension:      XY
#> bbox:           xmin: -84.32385 ymin: 33.88199 xmax: -75.45698 ymax: 36.58965
#> epsg (SRID):    4267
#> proj4string:    +proj=longlat +datum=NAD27 +no_defs
ggplot() + 
  geom_sf(aes(fill = AREA), data=nc) + coord_sf(crs = 2264)

ggplot() + 
  geom_sf(aes(fill = AREA), data=nc) + coord_sf(crs = 3338)

Created on 2018-08-23 by the reprex package (v0.2.0).

[hadley]

It feels like it should be a coord setting to me (or maybe we should adopt a NSEW convention where you could show n/s latitude or e/w longitude labels?) (i.e. if the projection was rotated 45 degrees, the northern labels would occur on top and right)

I'm fine with providing little control over the labels until we work around the general parameters that people want to control.

I think the thing I believe most strongly is that the default labels should align with the graticules.

[clauswilke]

To clarify what kind of an API I'm thinking about: coord_sf() could take an argument, say, graticule_labeling that accepts a named vector or list specifying what types of graticules should be shown on which side of the plot. E.g., graticule_labeling = c(top = NA, right = NA, bottom = "E", left = "N") which would represent the default.

[clauswilke]

@sambweber I made a pull request (#2849). Would you mind installing that branch (with devtools::install_github("tidyverse/ggplot2#2849")) and trying it out?

[hadley]

I was thinking you could make the spec a bit more compact since I don't think you want to be able to say (e.g.) put the east labels on the bottom, so you could collapse to graticule_labels = "EN" or graticule_labels = "NESW" (in others the physical position of the labels is determined by which side of the projected rectangle you want to label)

[clauswilke]

I'm not sure that E labels on the bottom are never useful. In fact, we only have "N" to indicate latitude and "E" to indicate longitude, there is no "S" and "W" (but the sf package translates negative "E" values to positive "W" values when generating labels).

One option for a more compact specification would be to allow hyphens as well, e.g. graticule_labels = "--EN". The one problem with the four-letter string is that we'll have to remember the order in which it specifies sides. Everything else in ggplot2 goes top, right, bottom, left, so it might make sense to mirror that.