In datagistips/geosparklines: Create Geographic Sparklines

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(geosparklines)
library(RcppRoll)
library(readr)
library(sf)

John Burn-Murdoch is a DataViz Designer at Financial Times. He covered the covid-19 crisis with awesome wisualizations, one of which transforms input data in a way that captures better the dynamics.

See his interview on Medium

His transformations consists in :

taking a 7-day rolling average of cases
log transforming this 7-day rolling average

I integrated his approach in the geosparklines package with a dedicated geosparkBM function

Prepare Data

f <- read_csv("https://raw.githubusercontent.com/CSSEGISandData/COVID-19/master/csse_covid_19_data/csse_covid_19_time_series/time_series_covid19_confirmed_global.csv")

Extract necessary informations for sparklines :

coords <- f[, c("Long", "Lat")]
col_dates <- grep("[0-9]*/[0-9]*/[0-9]*", names(f)) # select all column dates
my_stats <- data.frame(f)[, col_dates]

Generate the sparklines

Then generate the 7-day mean-rolled sparkline based upon Burn Murdocn Method

## Whole world
geosparks <- geosparkBM(coords, my_stats, width=4, height=2, direction = "ne") %>% st_set_crs(4326)

## Add geometries to data
st_geometry(f) <- geosparks

Let's do some simple plotting :

library(rnaturalearth)
library(rnaturalearthdata)
library(ggplot2)
library(rgeos)
library(rgeos)
world <- ne_countries(scale = "medium", returnclass = "sf")

last_day <- names(f)[col_dates %>% tail(1)]

ggplot() +
    geom_sf(data=world) +
    geom_sf(data=f, col="red") +
    coord_sf(xlim = c(-20, 30), ylim = c(25, 60), expand = FALSE) +
    ggtitle(sprintf("Covid-19 cases sparklines (Burn-Murdoch's method)\n%s", last_day))

Pilot the rendering of individual lines

You can do it on one line if you want. This way, you can control the rendering of individual lines : width, height, direction

i <- which(f$`Country/Region`=="France" & is.na(f$`Province/State`))
my_values <- my_stats[i, ] %>% as.numeric
my_coords <- coords[i, ] %>% as.matrix
sp_line <- singleGeosparkBM(my_coords, my_values, width = 20, height = 20, n = 7)
plot(st_geometry(sp_line))

Or, maybe, try a smoother line with a 3-week moving average :

sp_line <- singleGeosparkBM(my_coords, my_values, width=20, height=20, n = 21)
plot(st_geometry(sp_line))

Let's show how to generate multiple directional lines (just to show how it works)

Generate the lines with random widths and heights and all directions, just to show you how you can pilot the rendering precisely

out <- lapply(c("ne", "nw", "se", "sw"), function(x) singleGeosparkBM(my_coords, my_values, width=runif(1,4,6), height=runif(1,2,4), n = 21, direction = x))

geoms <- st_sfc(out) %>% st_set_crs(4326)

You could also change coordinates not to be on the centroid.

Let's plot it :

library(rnaturalearth)
library(rnaturalearthdata)
library(ggplot2)
library(rgeos)
library(rgeos)

fr <- ne_countries(country = "france", scale = "medium", returnclass = "sf")

ggplot() +
    geom_sf(data=fr) +
    geom_sf(data=geoms, col=c("red", "blue", "green", "orange"), size = 1.2) +
    coord_sf(xlim = c(-7, 12), ylim = c(40, 52), expand = FALSE) +
    ggtitle(sprintf("Covid-19 cases sparklines (Burn-Murdoch's method)\n4-directions with random widths and heights\n%s", last_day))