plotting pie graphs on map in ggplot

Question

This may be a wish list thing, not sure (i.e. maybe there would need to be the creation of geom_pie for this to occur). I saw a map today (LINK) with pie graphs on it as seen here.

I don't want to debate the merits of a pie graph, this was more of an exercise of can I do this in ggplot?

I have provided a data set below (loaded from my drop box) that has the mapping data to make a New York State map and some purely fabricated data on racial percentages by county. I have given this racial make up as a merge with the main data set and as a separate data set called key. I also think Bryan Goodrich's response to me in another post (HERE) on centering county names will be helpful to this concept.

How can we make the map above with ggplot2?

A data set and the map without the pie graphs:

load(url("http://dl.dropbox.com/u/61803503/nycounty.RData"))
head(ny); head(key)  #view the data set from my drop box
library(ggplot2)
ggplot(ny, aes(long, lat, group=group)) +  geom_polygon(colour='black', fill=NA)

#  Now how can we plot a pie chart of race on each county 
#  (sizing of the pie would also be controllable via a size 
#  parameter like other `geom_` functions).

Thanks in advance for your ideas.

EDIT: I just saw another case at junkcharts that screams for this type of capability:

Answer 1

A slight variation on the OP's original requirements, but this seems like an appropriate answer/update.

If you want an interactive Google Map, as of googleway v2.6.0 you can add charts inside info_windows of map layers.

see ?googleway::google_charts for documentation and examples

library(googleway)

set_key("GOOGLE_MAP_KEY")

## create some dummy chart data
markerCharts <- data.frame(stop_id = rep(tram_stops$stop_id, each = 3))
markerCharts$variable <- c("yes", "no", "maybe")
markerCharts$value <- sample(1:10, size = nrow(markerCharts), replace = T)

chartList <- list(
  data = markerCharts
  , type = 'pie'
  , options = list(
    title = "my pie"
    , is3D = TRUE
    , height = 240
    , width = 240
    , colors = c('#440154', '#21908C', '#FDE725')
    )
  )

google_map() %>%
  add_markers(
    data = tram_stops
    , id = "stop_id"
    , info_window = chartList
  )

Answer 2

I stumbled upon what looks like a function to do this: "add.pie" in the "mapplots" package.

The example from the package is below.

plot(NA,NA, xlim=c(-1,1), ylim=c(-1,1) )
add.pie(z=rpois(6,10), x=-0.5, y=0.5, radius=0.5)
add.pie(z=rpois(4,10), x=0.5, y=-0.5, radius=0.3)

Answer 3

I've written some code to do this using grid graphics. There is an example here: https://qdrsite.wordpress.com/2016/06/26/pies-on-a-map/

The goal here was to associate the pie charts with specific points on the map, and not necessarily regions. For this particular solution, it is necessary to convert the map coordinates (latitude and longitude) to a (0,1) scale so they can be plotted in the proper locations on the map. The grid package is used to print to the viewport that contains the plot panel.

Code:

# Pies On A Map
# Demonstration script
# By QDR

# Uses NLCD land cover data for different sites in the National Ecological Observatory Network.
# Each site consists of a number of different plots, and each plot has its own land cover classification.
# On a US map, plot a pie chart at the location of each site with the proportion of plots at that site within each land cover class.

# For this demo script, I've hard coded in the color scale, and included the data as a CSV linked from dropbox.

# Custom color scale (taken from the official NLCD legend)
nlcdcolors <- structure(c("#7F7F7F", "#FFB3CC", "#00B200", "#00FFFF", "#006600", "#E5CC99", "#00B2B2", "#FFFF00", "#B2B200", "#80FFCC"), .Names = c("unknown", "cultivatedCrops", "deciduousForest", "emergentHerbaceousWetlands", "evergreenForest", "grasslandHerbaceous", "mixedForest", "pastureHay", "shrubScrub", "woodyWetlands"))

# NLCD data for the NEON plots
nlcdtable_long <- read.csv(file='https://www.dropbox.com/s/x95p4dvoegfspax/demo_nlcdneon.csv?raw=1', row.names=NULL, stringsAsFactors=FALSE)

library(ggplot2)
library(plyr)
library(grid)

# Create a blank state map. The geom_tile() is included because it allows a legend for all the pie charts to be printed, although it does not
statemap <- ggplot(nlcdtable_long, aes(decimalLongitude,decimalLatitude,fill=nlcdClass)) +
geom_tile() +
borders('state', fill='beige') + coord_map() +
scale_x_continuous(limits=c(-125,-65), expand=c(0,0), name = 'Longitude') +
scale_y_continuous(limits=c(25, 50), expand=c(0,0), name = 'Latitude') +
scale_fill_manual(values = nlcdcolors, name = 'NLCD Classification')

# Create a list of ggplot objects. Each one is the pie chart for each site with all labels removed.
pies <- dlply(nlcdtable_long, .(siteID), function(z)
ggplot(z, aes(x=factor(1), y=prop_plots, fill=nlcdClass)) +
geom_bar(stat='identity', width=1) +
coord_polar(theta='y') +
scale_fill_manual(values = nlcdcolors) +
theme(axis.line=element_blank(),
axis.text.x=element_blank(),
axis.text.y=element_blank(),
axis.ticks=element_blank(),
axis.title.x=element_blank(),
axis.title.y=element_blank(),
legend.position="none",
panel.background=element_blank(),
panel.border=element_blank(),
panel.grid.major=element_blank(),
panel.grid.minor=element_blank(),
plot.background=element_blank()))

# Use the latitude and longitude maxima and minima from the map to calculate the coordinates of each site location on a scale of 0 to 1, within the map panel.
piecoords <- ddply(nlcdtable_long, .(siteID), function(x) with(x, data.frame(
siteID = siteID[1],
x = (decimalLongitude[1]+125)/60,
y = (decimalLatitude[1]-25)/25
)))

# Print the state map.
statemap

# Use a function from the grid package to move into the viewport that contains the plot panel, so that we can plot the individual pies in their correct locations on the map.
downViewport('panel.3-4-3-4')

# Here is the fun part: loop through the pies list. At each iteration, print the ggplot object at the correct location on the viewport. The y coordinate is shifted by half the height of the pie (set at 10% of the height of the map) so that the pie will be centered at the correct coordinate.
for (i in 1:length(pies)) 
  print(pies[[i]], vp=dataViewport(xData=c(-125,-65), yData=c(25,50), clip='off',xscale = c(-125,-65), yscale=c(25,50), x=piecoords$x[i], y=piecoords$y[i]-.06, height=.12, width=.12))

The result looks like this:

Answer 4

Three years later this is solved. I've put together a number of processes together and thanks to @Guangchuang Yu's excellent ggtree package this can be done fairly easily. Note that as of (9/3/2015) you need to have version 1.0.18 of ggtree installed but these will eventually trickle down to their respective repositories.

I've used the following resources to make this (the links will give greater detail):

1. ggtree blog
2. move ggplot legend
3. correct ggtree version
4. centering things in polygons

Here's the code:

load(url("http://dl.dropbox.com/u/61803503/nycounty.RData"))
head(ny); head(key)  #view the data set from my drop box

if (!require("pacman")) install.packages("pacman")
p_load(ggplot2, ggtree, dplyr, tidyr, sp, maps, pipeR, grid, XML, gtable)

getLabelPoint <- function(county) {Polygon(county[c('long', 'lat')])@labpt}

df <- map_data('county', 'new york')                 # NY region county data
centroids <- by(df, df$subregion, getLabelPoint)     # Returns list
centroids <- do.call("rbind.data.frame", centroids)  # Convert to Data Frame
names(centroids) <- c('long', 'lat')                 # Appropriate Header

pops <-  "http://data.newsday.com/long-island/data/census/county-population-estimates-2012/" %>%
     readHTMLTable(which=1) %>%
     tbl_df() %>%
     select(1:2) %>%
     setNames(c("region", "population")) %>%
     mutate(
         population = {as.numeric(gsub("\\D", "", population))},
         region = tolower(gsub("\\s+[Cc]ounty|\\.", "", region)),
         #weight = ((1 - (1/(1 + exp(population/sum(population)))))/11) 
         weight = exp(population/sum(population)),
         weight = sqrt(weight/sum(weight))/3
     )


race_data_long <- add_rownames(centroids, "region") %>>%
    left_join({distinct(select(ny, region:other))}) %>>%
    left_join(pops) %>>%
    (~ race_data) %>>%
    gather(race, prop, white:other) %>%
    split(., .$region)

pies <- setNames(lapply(1:length(race_data_long), function(i){
    ggplot(race_data_long[[i]], aes(x=1, prop, fill=race)) +
        geom_bar(stat="identity", width=1) + 
        coord_polar(theta="y") + 
        theme_tree() + 
        xlab(NULL) + 
        ylab(NULL) + 
        theme_transparent() +
        theme(plot.margin=unit(c(0,0,0,0),"mm"))
}), names(race_data_long))


e1 <- ggplot(race_data_long[[1]], aes(x=1, prop, fill=race)) +
        geom_bar(stat="identity", width=1) + 
        coord_polar(theta="y") 

leg1 <- gtable_filter(ggplot_gtable(ggplot_build(e1)), "guide-box") 


p <- ggplot(ny, aes(long, lat, group=group)) +  
    geom_polygon(colour='black', fill=NA) +
    theme_bw() +
    annotation_custom(grob = leg1, xmin = -77.5, xmax = -78.5, ymin = 44, ymax = 45) 



n <- length(pies)

for (i in 1:n) {

    nms <- names(pies)[i]
    dat <- race_data[which(race_data$region == nms)[1], ]
    p <- subview(p, pies[[i]], x=unlist(dat[["long"]])[1], y=unlist(dat[["lat"]])[1], dat[["weight"]], dat[["weight"]])

}

print(p)

Answer 5

This functionality should be in ggplot, I think it is coming to ggplot soonish, but it is currently available in base plots. I thought I would post this just for comparison's sake.

load(url("http://dl.dropbox.com/u/61803503/nycounty.RData"))

library(plotrix)
e=10^-5
myglyff=function(gi) {
floating.pie(mean(gi$long),
             mean(gi$lat),
             x=c(gi[1,"white"]+e,
                 gi[1,"black"]+e,
                 gi[1,"hispanic"]+e,
                 gi[1,"asian"]+e,
                 gi[1,"other"]+e),
              radius=.1) #insert size variable here
}

g1=ny[which(ny$group==1),]
plot(g1$long,
     g1$lat,
     type='l',
     xlim=c(-80,-71.5),
     ylim=c(40.5,45.1))

myglyff(g1)

for(i in 2:62)
  {gi=ny[which(ny$group==i),]
    lines(gi$long,gi$lat)
    myglyff(gi)
  }

Also, there may be (probably are) more elegant ways of doing this in the base graphics.

As, you can see, there are quite a few problems with this that need to be solved. A fill color for the counties. The pie charts tend to be too small or overlap. The lat and long do not take a projection so sizes of counties are distorted.

In any event, I am interested in what others can come up with.