Data Visualization

ggplot US state and China province heatmap

Yang Liu

Tags: Heatmap

(Updated on 2020-06-12, First posted on 2019-04-17)

It sounds easy but turned out not as straight-forward as I thought. I will show:

  1. 50-state (including Alaska and Hawaii) United States thematic map, with map scale and with state abbreviations

  2. China thematic map, with map scale, with names of provinces in either English or Chinese

1. US Map by state

The dt1 below can be used by both methods, but to use usmap only need two variables: state names and value to plot.

suppressPackageStartupMessages({
library(ggplot2)
library(maps)
library(usmap)
library(rgeos) # for rgeos::gCentroid
library(data.table)
library(ggsn) # for adding map scale bar `ggsn::scalebar`
library(shadowtext) # for adding label
library(ggrepel) # if need to repel labels
})
# A sample dataset of 77 states: 
dt1 <- as.data.table(copy(state.x77))
dt1$state <- tolower(rownames(state.x77))
dt1 <- dt1[,.(state, Population)]
setkey(dt1, state)

states <- setDT(ggplot2::map_data("state"))
setkey(states, region)
# join data to map: left join states to dt1
dt1 <- dt1[states]
# 
rmarkdown::paged_table(dt1[1:50,])

Method 1. Use usmap

The easiest way is to use the package usmap, in which adding abbraviation or not is an option in the usmap::plot_usmap function. Easy to get started but hard to further finetune details.

dt1_simple <- unique(dt1[,.(state, Population)])
usmap::plot_usmap(data = dt1_simple, values = "Population", labels = TRUE) +
  labs(fill = 'State Population (1975)') + 
  scale_fill_gradientn(colours = RColorBrewer::brewer.pal(9, "GnBu"),
                       na.value = "grey90",
                       guide = guide_colourbar(barwidth = 25, 
                                               barheight = 0.4,
                                               #put legend title on top of legend 
                                               title.position = "top")) +
  # put legend at the bottom, adjust legend title and text font sizes
  theme(legend.position = "bottom",
        legend.title=element_text(size=12), 
        legend.text=element_text(size=10))

The plot_usmap returns a ggplot object so it is possible to further revise by just adding more layers.

Method 2. Use map_data and build from shape files

Could fine-tune the location of states’label as I did in the China map later. Now it is just the center of the states (mean(lon), min(lat))

# create labels
# incl `Population` (the value to plot) in the label dataset, if want to fill with color. 
#' get centroids from a shp file dataset
get.centroids <- function(
  data = dt1, 
  long = "long", lat = "lat", 
  by_var = "state",  # the grouping variable, e.g. state: get centroid by state
  fill_var = NULL # the variable to plot 
  ){
  data <- data[!is.na(data[[by_var]]),]
  data[[by_var]] <- as.character(data[[by_var]]) # sometimes there is empty factor level
  dt1_df <- sp::SpatialPointsDataFrame(coords = data[, c(long, lat), with = FALSE], data = data)
  dt1_geo <- by(dt1_df, dt1_df[[by_var]], function(x) {sp::Polygon(x[c(long, lat)])@labpt})
  centroids <- stats::setNames(do.call("rbind.data.frame", dt1_geo), c(long, lat))
  centroids$name <- names(dt1_geo) 
  if(!is.null(fill_var)){ # if need to join fill value 
    setkeyv(setDT(centroids), "name")
    dt_var <- unique(data[,c(by_var, fill_var), with = FALSE])
    setkeyv(dt_var, by_var)
    centroids <- dt_var[centroids]
  }
  return(centroids)
}
centroids <- get.centroids(data = dt1, long = "long", lat = "lat", by_var = "state", fill_var = "Population")
# join short state names (e.g. VA, SC)
abb_names <- data.table(state = tolower(state.name), abb = state.abb) # these are R dataset
setkey(setDT(centroids), state)
setkey(abb_names, state)
centroids <- abb_names[centroids]
head(centroids, 3)
##       state abb Population       long      lat
## 1:  alabama  AL       3615  -86.83042 32.80316
## 2:  arizona  AZ       2212 -111.66786 34.30060
## 3: arkansas  AR       2110  -92.44013 34.90418
x_boundary = -77
ggplot(data = dt1, aes(x = long, y = lat, group = group))+
  geom_polygon(aes(fill = Population))+
  geom_path()+ 
  scale_fill_gradientn(colours = RColorBrewer::brewer.pal(9, "GnBu"), 
                       na.value = "grey90",
                       guide = guide_colourbar(barwidth = 25, barheight = 0.4,
                                               #put legend title on top of legend
                                               title.position = "top")) +
  # shadowtext::geom_shadowtext(data = centroids,
  #                             aes(x = long, y = lat, label = abb),
  #                             size = 3,
  #                             inherit.aes = FALSE) + 
  # if need to repel labels... could further finetune
  geom_label_repel(data = centroids[long >= x_boundary,],
                   aes(x = long, y = lat, label = abb, fill = Population),
    force = 1, size = 3,
    inherit.aes = F
  ) +
  # the normal labels:
  geom_text(data=centroids[long < x_boundary,],
            aes(x = long, y = lat, label = abb),
            size = 3, inherit.aes=F) +
  labs(fill = "Population", x = "Longitude", y = "Latitude") + 
  coord_map() + 
  # map scale
  ggsn::scalebar(data = dt1, dist = 500, dist_unit = "km",
           border.size = 0.4, st.size = 4,
           box.fill = c('black','white'),
           transform = TRUE, model = "WGS84") + 
  theme_void() + 
  # put legend at the bottom, adjust legend title and text font sizes
  theme(legend.position = "bottom",
        legend.title=element_text(size=12),  # font size of the legend 
        legend.text=element_text(size=10))

# 
# ggsave("us_example.png", width = 10, height = 6)

2. China map by province

Method 1. China map by province using downloaded shap files

The map of China is more complicated. The related articles are all from several years ago. This blog by Zhang, Zhen is among the best, it has referred some very good resources, expecially this one on Capital Of Statistics (CH)](https://cosx.org/2009/07/drawing-china-map-using-r). This is a blog from 10 years ago. So far I haven’t noticed any map package published on cran for China like usmap, but it is not too complicated after having the shape files.

The shape files can be downloaded from the “Capital of Statistics” webiste, the zip file contains three files: bou2_4p.dbf, bou2_4p.shp, and bou2_4p.shx, we will only need the bou2_4p.shp file later.

Update 2020/06 There is a chinamap package created by Guangchuang Yu (and his team), who is also the author of the package nCov2019.

Another very good source is the GADM website, which has admin1 and admin2 shp files for every country. (Notice that China and Taiwan are downloaded seperately.)

On character encoding Showing Chinese is indeed sometimes a problem. The key is to make sure in the datafile all the province names are coded in UTF-8. In the markdown file below you can see province names shown as UTF-8 code. But if in the map (plot) they are rendered as Chinese characters. And if your system can show Chinese, they read as Chinese in the console too.

In Chinese (一点额外说明): 在Windows系统上，只要RStudio中显示中文没有问题好像Markdown的时候就没有问题。关键是要统一数据中的中文都是UTF-8编码的，不要从网站上拷贝中文进Rstudio，会产生各种问题。不需要特别调整Rstudio或者Markdown的设置。 所以虽然我下面的数据中显示UTF-8，在Console和最后在地图上都是显示为中文。

# China -------------------------------------------------------------------
china_map <- rgdal::readOGR("D:/liuyanguu/China.shp/bou2_4p.shp") # read it locally
## OGR data source with driver: ESRI Shapefile 
## Source: "D:\liuyanguu\China.shp\bou2_4p.shp", layer: "bou2_4p"
## with 925 features
## It has 7 fields
## Integer64 fields read as strings:  BOU2_4M_ BOU2_4M_ID
# extract province information from shap file
china_map_data = data.table::setDT(china_map@data)
data.table::setnames(china_map_data, "NAME", "province")

# transform to UTF-8 coding format
china_map_data[, province:=iconv(province, from = "GBK", to = "UTF-8")] 
# create id to join province back to lat and long, id = 0 ~ 924
china_map_data[, id:= .I-1] # id = 0, 1, 2, ... , used to match to `dt_china`
# there are more shapes for one province due to small islands
china_map_data[, province:= as.factor(province)]
china_map_data <- china_map_data[!is.na(province)]
china_map_data <- china_map_data[AREA > 0.1]
head(china_map_data, 3)
##       AREA PERIMETER BOU2_4M_ BOU2_4M_ID ADCODE93 ADCODE99
## 1:  54.447    68.489        2         23   230000   230000
## 2: 129.113   129.933        3         15   150000   150000
## 3: 175.591    84.905        4         65   650000   650000
##                                                            province id
## 1:                                 <U+9ED1><U+9F99><U+6C5F><U+7701>  0
## 2:                 <U+5185><U+8499><U+53E4><U+81EA><U+6CBB><U+533A>  1
## 3: <U+65B0><U+7586><U+7EF4><U+543E><U+5C14><U+81EA><U+6CBB><U+533A>  2
# 
dt_china = setDT(fortify(china_map))
head(dt_china, 3)
##        long      lat order  hole piece id group
## 1: 121.4884 53.33265     1 FALSE     1  0   0.1
## 2: 121.4995 53.33601     2 FALSE     1  0   0.1
## 3: 121.5184 53.33919     3 FALSE     1  0   0.1
dt_china[, id:= as.numeric(id)]
setkey(china_map_data, id); setkey(dt_china, id)
dt_china <- china_map_data[dt_china]

# add the province EN, CH label file
province_CH <- china_map_data[, levels(province)] # the CH are in UTF-8 code
province_EN <- c("Shanghai", "Yunnan", "Inner Mongolia", "Beijing", "Taiwan",
"Jilin", "Sichuan", "Tianjin City", "Ningxia", "Anhui",
"Shandong", "Shanxi", "Guangdong", "Guangxi ", "Xinjiang",
"Jiangsu", "Jiangxi", "Hebei", "Henan", "Zhejiang",
"Hainan", "Hubei", "Hunan", "Gansu", "Fujian",
"Tibet", "Guizhou", "Liaoning", "Chongqing", "Shaanxi",
"Qinghai", "Hong Kong", "Heilongjiang"
)
# some value to plot (from years ago too)
value <- c(8893483, 12695396,  8470472,  7355291, 23193638,  9162183, 26383458,  3963604,  1945064, 19322432, 30794664, 10654162, 32222752, 13467663,  6902850, 25635291, 11847841, 20813492, 26404973, 20060115, 2451819, 17253385, 19029894,  7113833, 11971873, 689521, 10745630, 15334912, 10272559, 11084516, 1586635,  7026400, 13192935)
input_data <- data.table(province_CH, province_EN, value)
setkey(input_data, province_CH)
setkey(dt_china, province)

dt_china <- input_data[dt_china, nomatch = 0]

# create label file of province `label_dt`
centroids_cn <- get.centroids(data = dt_china, by_var = "province_CH")
centroids_en <- get.centroids(data = dt_china, by_var = "province_EN")

# plot
gg_cn <- ggplot(dt_china, aes(x = long, y = lat, group = group, fill = value)) +
  labs(fill = "Population")+
  geom_polygon()+
  scale_fill_gradientn(colours = RColorBrewer::brewer.pal(9, "GnBu"), 
                       na.value = "grey90",
                       guide = guide_colourbar(barwidth = 25, barheight = 0.4,
                                               #put legend title on top of legend
                                               title.position = "top")) + 
  labs(fill = "Population", x = "Longitude", y = "Latitude") + 
  coord_map() + 
  # map scale
  ggsn::scalebar(data = dt_china, dist = 500, dist_unit = "km",
           border.size = 0.4, st.size = 4,
           box.fill = c('black','white'),
           transform = TRUE, model = "WGS84") + 
  theme_void() +
  theme(legend.position = "bottom",
        legend.title=element_text(size=12),  # font size of the legend 
        legend.text=element_text(size=10))

# In Chinese
gg_cn +  shadowtext::geom_shadowtext(data = centroids_cn, aes(x = long, y = lat, label = name), inherit.aes = FALSE)

# In English
gg_cn +  shadowtext::geom_shadowtext(data = centroids_en, aes(x = long, y = lat, label = name), inherit.aes = FALSE)

Method 2. Using geojsonMap (leaflet)

Just saw it on cosx.org blog. It is an interactive map built on leaflet.

library(leafletCN)
dt_china <- as.data.frame(dt_china)
geojsonMap(dat = dt_china, mapName = "china",
           # namevar = ~ province_EN, 
           valuevar = ~ value, 
           popup =  paste0(dt_china$province_EN),
           palette = RColorBrewer::brewer.pal(9, "GnBu"), legendTitle = "Population")

About the Author

Yang Liu (刘杨) is a Statistician working in New York City.