[GIS] Point overplotting on maps in R (e.g. with ggplot2) – how to push points to the side, mark original positions, combine close-by points …

data-frameggplot2mapsoverlapping-featuresr

I've got a database containing languages, their longitudes and latitudes and a feature value (either category 1, category 2 or both – in the plot these are marked red, blue and green respectively). There may be up to three points per language and naturally two language points may lie very close to each other.

    name            longitude   latitude    sp_sum
1   Modern Armenian 45          40          both
2   Modern Armenian 45          40          both
3   Modern Armenian 45          40          spatial
4   Dieri           138         -28.1667    both
5   Dieri           138         -28.1667    both
6   Finnish         25.5577     64.7628     non-spatial
7   Crimean Tatar   28.1418     43.8398     spatial
8   Ese Ejja        -67.515     -11.7268    non-spatial
9   Makhuwa         38.8052     -14.8509    non-spatial
...

I'm using the R package ggplot2 (that is the one I am most familiar with, so I'd be happy to keep using it – but other solutions are also welcome). Here's a crop from a previous attempt (code: see below ¹):

Crop from previous attempt

For every point, I'd like the (rough) position – as well as the value – still to be visible. (If there are multiple points for a single language, they may be combined.)

Is there a way either…

… to move points to the side just enough so that there is no overplotting (less randomly than, say, by using geom_jitter – there is a lot of that kind of dodging in the beeswarm package for example)?
… and/or to have some kind of "line" pointing to the original position of a point if it had to be moved?
… or to combine close-by points in a way that they are still clear (there is probably a working technique out there that uses binning, i.e. stat_bin* or something with a similar effect)?
… or to create an "interactive plot" like those seen on websites that still can be included into a pdf (I'm thinking also about the abilities of packages like animation and shiny here)? For example, it looks like this on wals.info:

From a previous post here, I know that the directlabels package can move labels, but I haven't found a way to make it move the points as well.

Feel free to ask for clarification!

Note: I am aware that there have been a number of questions on overplotting, but those that I have looked into all seemed to be have a different (i.e. statistic) purpose (I don't claim to have read it all, so I'd be happy to accept a link as well, of course). I'll try to list those posts that I know and that may well be relevant (- from what I've read, none of these exactly answers my question.)

¹ The following lines of code created the crop from above.

library(OpenStreetMap)
library(ggplot2)

data <- read.csv(header = T, sep = ",", dec = ".", quote= "'",
text = "'','name','longitude','latitude','sp_sum'
'1','Modern Armenian',45,40,'both'
'2','Modern Armenian',45,40,'both'
'3','Modern Armenian',45,40,'spatial'
'4','Dieri',138,-28.1667,'both'
'5','Dieri',138,-28.1667,'both'
'6','Finnish',25.5577,64.7628,'non-spatial'
'7','Crimean Tatar',28.1418,43.8398,'spatial'
'8','Sochiapam Chinantec',-96.6079,17.7985,'non-spatial'
'9','Ese Ejja',-67.515,-11.7268,'non-spatial'
'10','Makhuwa',38.8052,-14.8509,'non-spatial'
'11','Mualang',111.077,0.31083,'non-spatial'
'12','Martuthunira',116.607,-20.9294,'non-spatial'
'13','Evenki',108.626,53.85,'both'
'14','Afrikaans',30,-22,'both'
'15','Male (Ethiopia)',36.9892,5.91975,'both'
'16','Manchu',126.557,47.3122,'both'
'17','Dime',36.3329,6.20951,'non-spatial'
'18','Koorete',37.8679,5.80545,'non-spatial'
'19','Wolaytta',37.7537,6.32668,'both'
'20','Dizin',35.5763,6.1405,'both'")

map <- openproj(openmap(c(85, -179.9), c(-60, 179.9), zoom = 2, type = "nps"))
plot <- autoplot(map) + 
  geom_point(data = data, aes(x = longitude, y = latitude),
             color = "white", alpha = 0.8, size = 8) +
  geom_point(data = data, aes(x = longitude, y = latitude, color = sp_sum),
             alpha = 0.3, size = 4)
plot

Best Answer

So far I have found only one fairly decent looking workaround: The packcircles R package may have been designed for another purpose, but it does a nice job pushing the points away from each other (also see corresponding blog post). I might not understand all of the inner workings of this package, but luckily, as you will find, the example file from the website can be used almost directly - all one needs to change are the variable names, the distance between circles (or points, depending on the functions you use) and the "limits" of the graph (i.e. 180°).

(In the end it all comes down to the circleLayout() function, which takes a data frame with lon, lat and radius (i.e. distance) columns and two 2-numeric xlim/ylim vectors - it returns the data frame with improved point positions.)

please compare this 'after' map with the 'before' map snippet from the question

Related Solutions

[GIS] Loading a png image as a map with ggplot2/ggmap

Although it is quite a long-winded approach, I came up with a solution which allows the user to display red-green-blue 'RasterStack' objects created from the initial 'ggmap' object using ggplot. But first things first, here is what I did.

First of all, I forked the read-only mirror of ggmap hosted on GitHub and, in order to retain the original source code, created a 'develop' branch which comprises all edits that were required to get this to work. No offense intended against the package developers! The modified package version can be directly installed from within R using

## install modified 'ggmap' package
try(detach("package:ggmap", unload = TRUE), silent = TRUE)

library(devtools)
install_github("fdetsch/ggmap", ref = "develop")

## load required packages
library(ggmap)
library(raster)
library(maptools)
library(plyr)

I am quite aware that there are still plenty of non-conformities involved in the package, but for now, the installation procedure should work just fine. Next, I amended get_map, which now takes the argument location = "world", in order to download a Google Maps image of the entire globe. The thus created 'ggmap' object can simple be converted to a red-green-blue 'RasterStack' object by running the newly introduced function ggmap2raster.

## retrieve and rasterize world map
map_world <- get_map(location = "world")
rst_world <- ggmap2raster(map_world)

In order to display the red-green-blue image via ggplot, I adopted and slightly modified the algorithm implementation of rggbplot presented on GURU-GIS. The function is currently not exported, and hence the usage of ::: is required to properly create the ggplot2-based figure.

## initialize figure
p <- ggmap:::rggbplot(rst_world, npix = ncell(rst_world))

Finally, let me add some sample data to demonstrate that not only the visualization of a ggmap-based world map works, but also the subsequent insertion of e.g. points or polygons.

## create and add sample points
spt <- data.frame(lon = c(-74.0059, -43.196389, 139.683333),
                  lat = c(40.7127, -22.908333, 35.683333),
                  city = c("New York City", "Rio de Janeiro", "Tokyo"))

coordinates(spt) <- ~ lon + lat
proj4string(spt) <- "+init=epsg:4326"
spt <- spTransform(spt, CRS = CRS("+init=epsg:3857"))

p <- p + 
  geom_point(aes(x = lon, y = lat), data = data.frame(spt)) + 
  geom_text(aes(x = lon, y = lat, label = city), data = data.frame(spt), 
            vjust = 1.6, size = 3)

## add polygons; see also hadley's tutorial at
## https://github.com/hadley/ggplot2/wiki/plotting-polygon-shapefiles
spy <- getData(country = "MNG", level = 0)
spy <- spTransform(spy, CRS = CRS("+init=epsg:3857"))

spy@data$id <- rownames(spy@data)
spy_points <- fortify(spy, region = "id")
spy_df <- join(spy_points, spy@data, by = "id")

p <- p + 
  geom_polygon(aes(x = long, y = lat, group = group), data = spy_df, 
               colour = "grey25", size = 2, fill = "transparent") + 
  geom_text(aes(x = x, y = y, label = "Mongolia"), 
            data = data.frame(coordinates(rgeos::gCentroid(spy))), 
            vjust = 0.5, size = 3, fontface = "bold")

## add google maps reference
p <- p + 
  geom_text(aes(x = 11500000, y = -18000000), label = "\uA9 Google Maps", 
            size = 3, fontface = "bold")

Et voilà, here is the resulting image. Please, feel free to browse the linked source code. Any suggestions on how to improve the scripts would be highly appreciated!

## save figure
png("~/worldmap.png", width = 12, height = 12, units = "cm", res = 500)
print(p)
dev.off()

Update:

Sorry, forgot to mention that ggmap2raster allows the user to specify an output CRS of the created 'RasterStack' object which is then passed on to projectRaster from the raster package. Hence, it is possible to display the data in EPSG:4326 (Lat/Long) rather than EPSG:3857 (Web Mercator), which would also render unnecessary the transformation of 'SpatialPoints', 'SpatialPolygons', etc. to Web Mercator. However, I strongly recommend that you stick with Google Maps native EPSG:3857 to avoid vertical stretching of the map labels (see here).

[GIS] way to get undistorted maps in R (when using the OpenStreetMap package)

If you reproject a raster with labels, you will obviously get squeezed labels.

The only way to avoid this is to render the raster from vector data directly into the desired projection. You might want to look into mapnik, tilemill or maperitive to do this from Openstreetmap raw data (which is vector data).

The R openstreetmap package only offers raster data from tiles, which is not suitable for this task. The osmar package might do a better job, but you will not get lucky with it on a worldwide scale. It's just too much data to process.

Best Answer

Related Solutions

[GIS] Loading a png image as a map with ggplot2/ggmap

[GIS] way to get undistorted maps in R (when using the OpenStreetMap package)

Related Question