Let’s explore county-level precipitation data using a choropleth map. Importantly, we’ll decide on a color palette / scale type and make any necessary adjustments.
What’s a choropleth?
Choropleths are maps that display the spatial distribution of a variable across divided geographical areas / regions, where variable is encoded by color.
Choosing the right color palette and scale type are critically important. Oftentimes, you’ll need to adjust the default mapping of colors to accurately tell your story.
Find public access to a massive inventory of climate data on their Climate Monitoring page. Today’s lesson will use the Climate at a Glance collection. Specifically, we’ll be exploring how precipitation across the continental US over the past 5 years compares to the 20th century average. To do so, we’ll work with county-level precipitation data, accessed via the County Mapping portal.
A shapefile is a vector data file format commonly used for geospatial analysis.
Shapefiles contain information for spatially describing features (e.g. points, lines, polygons), as well as any associated attribute information.
You can find / download shapefiles online (e.g. from the US Census Bureau), or depending on the tools available, access them via packages (like we’re doing today).
Simple features in R
Spatial data can take many forms – simple features is a standard that allows different types of software to specify spatial data in a common way.
Simple features are comprised of:
1. a geometry object (e.g. a point, line, polygon) that describes where on Earth the feature is located
2. attribute data associated with the geometry object (e.g. the precipitation across a county during the last 5 years)
Because of how simple feature (sf) objects are represented in R (they look like data frames!), simple features can be maniupulated and plotted by other well-known packages like {dplyr} and {ggplot2}. Packages like {sf} provide tools for working with simple features (sf objects), but we’ll only need to rely on {ggplot2}s built-in geom_sf() geometry to plot our data.
When we download our shapefile using {tigris}, it’ll be loaded as a simple features (sf) object with geometries that allow us to plot county lines. We’ll join our county-level precipitation data to our sf object so that we can color counties by precipitation.
Data Wrangling
Here, we’ll use the {tigris} package to import geometries for our US counties, then join it with our precipitation data:
##~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~## setup ----##~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~#..........................load packages.........................library(tidyverse)library(tigris)#.........................get shape data.........................county_geo <- tigris::counties(class ="sf", cb =TRUE) |># cb = TRUE to use cartographic boundary files# shift US to fit AK, HI, PR (we'll be filtering these out though) and transform CRS to USA Contiguous Albers Equal Area Conic (ESRI:102003) ----shift_geometry()#....................import precipitation data...................precip_data <-read_csv(here::here("week5", "data", "county-feb19-jan24-precip.csv"), skip =4)##~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~## data wrangling ----##~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~##~~~~~~~~~~~~~~~~~~~~~~~~~~~~## ~ wrangle geometries ----##~~~~~~~~~~~~~~~~~~~~~~~~~~~~county_geo_wrangled <- county_geo |># clean up col names ---- janitor::clean_names() |># rename county & state cols ----rename(county = namelsad, state = state_name) |># remove states / territories that we don't have precip data for ----filter(!state %in%c("Alaska", "Hawaii", "District of Columbia","United States Virgin Islands", "Puerto Rico", "American Samoa","Commonwealth of the Northern Mariana Islands", "Guam")) |># capitalize "city" (VA) ----mutate(county =str_replace(string = county, pattern =" city", replacement =" City"))##~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~## ~ wrangle precipitation data ----##~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~precip_wrangled <- precip_data |># clean up col names ---- janitor::clean_names() |># rename county col ----rename(county = name) |># filter out DC ----filter(!county %in%c("Washington, D.C.")) |># update name to match that in county_geo df ----mutate(county =str_replace(string = county, pattern ="Dona Ana County", replacement ="Doña Ana County")) |># coerce precip & 20th centruy avg from chr to numeric ----mutate(value =as.numeric(value),x1901_2000_mean =as.numeric(x1901_2000_mean)) |># calculate % change ----mutate(perc_change = ((value - x1901_2000_mean)/x1901_2000_mean)*100) |># select, rename, reorder cols ----select(id, state, county, mean_1901_2000 = x1901_2000_mean, precip = value, perc_change, anomaly_1901_2000_base_period)##~~~~~~~~~~~~~~~~~~## ~ join dfs ----##~~~~~~~~~~~~~~~~~~# join dfs (be sure to join precip TO sf object, not the other way around) -------joined_precip_geom <-full_join(county_geo_wrangled, precip_wrangled)
Start by creating a base map
base_map <-ggplot(joined_precip_geom) +geom_sf(aes(fill = perc_change), linewidth =0.1) +labs(title ="5-year precipitation compared with the 20th century average",subtitle ="February 2019 - January 2024",caption ="Source: National Centers for Environmental Information") +theme_void() +theme(legend.position ="bottom",legend.title =element_blank(),plot.caption =element_text(face ="italic",margin =margin(t =2, r =0.5, b =0, l =0, "lines")) )base_map
Because we want to map precipitation relative to the 20th century average (e.g. has precipitation for a given region over the last 5 years been above or below the average), a divering color palette makes a lot of sense.
Classed or unclassed color scale?
We’ve landed on a diverging color palette, but should we use a classed (aka binned) or unclassed (aka continuous) palette?
Use a classed color scale if you want to communicate statistical brackets:
the focus is on which data units fall into pre-defined classes, rather than overall pattern
best if you want you audience to read values (gets more difficult with more classes; easier with interactive visualizations)
the more classes you have, the more nuanced your map becomes
Use an unclassed color scale if you want to show general patterns:
the focus is on general patterns, rather than which statistical brackets regions fall into
best if you don’t want to interpret for your reader – it makes it easier to see outliers, transitions to and comparisons with neighboring regions
Start with an unclassed scale
“The unclassed choropleth is the most exact representation of the data model possible,”
“No matter if you decide for a classed map at the end, you should start your process by looking at an unclassed map. This will help you see subtle differences between regions and make a conscious decision if and how you should simplify them.”
We’ll heed this advice and start with an unclassed map!
Pick a color palette!
Recall from earlier that precipitation data is often encoded using a brown / blue color scheme (with drier conditions falling on the brown side and wetter conditions falling on the blue side).
Lucky for us, RColorBrewer has this exact palette. Let’s use all 11 hues for our unclassed map:
Preview the palette using display.brewer.pal() with our desired number of hues:
RColorBrewer::display.brewer.pal(n =11, name ='BrBG')
Save the HEX codes to a named object using brewer.pal() (we’ll call this in our plot later):
my_brew_palette11 <- RColorBrewer::brewer.pal(n =11, name ='BrBG')my_brew_palette11
Here, we leverage the awesome {scales} package to add %s to the colorbar labels and set our breaks. We also use guides() + guide_colorbar() to update label positioning and colorbar size:
0% (i.e. no change between 5-year precipitation and 20th century average) is currently on the bluer side of our color scale, rather than on the off-white color that’s at the center of our palette.
As a result, our map may be misleading – it would appear as if more counties received higher-than-average precipitation than in actuality.
By dropping the off-white hue, we can construct our scale so that 0% sits at the break point between brown and blue shades – any county that received more than the historical average will be a shade of blue, and any that received less will be a shade of brown.
By default, our resolution is pretty low
We only get 4 bins by default, which means we lose a lot of detail in our map:
