<- read_csv("drought-fips.csv")
drought_county <- read_csv("drought.csv") drought_state
Drought Conditions in the US
6/14/2022
Data
The data this week comes from the National Integrated Drought Information System.
This web page provides more information about the drought conditions data.
ggridges by state
h/t to @adambushman for the ggridges idea and @nrennie for the formatting of the facet_geo
.
<- drought_county %>% group_by(State) %>%
drought_county_state summarize(State = first(State))
%>%
drought_county #dplyr::filter(State == "CA") %>%
mutate(year = year(ymd(date))) %>%
ggplot(aes(x = DSCI, y = year, group = year, fill = stat(x))) +
::geom_density_ridges_gradient(rel_min_height = 0.01, color = "#ffffff") +
ggridgesfacet_geo(~State, grid = "us_state_contiguous_grid1") +
scale_fill_gradient(high = '#d0162e', low = '#fcc425') +
labs(title = 'Drought Severity by Year',
subtitle = 'Since Year 2000',
y = '',
x = 'DSCI Score [Drought Severity]',
fill = 'DSCI',
caption = "Tidy Tuesday Plot: @hardin47 | Data: National Integrated Drought Information System") +
theme_minimal() +
coord_cartesian(expand = FALSE) +
theme(
plot.background = element_rect(fill = '#ffffff'),
text = element_text(color = '#808080'),
axis.text = element_text(color = '#808080'),
panel.grid.major = element_line(color = '#efefef'),
panel.grid.minor = element_line(color = '#efefef'),
plot.title = element_text(size = 18, hjust = 0.5, face = 'bold'),
plot.subtitle = element_text(size = 12, hjust = 0.5),
legend.position = "bottom",
legend.title = element_blank(),
panel.grid = element_blank(),
axis.text.x = element_blank(),
axis.text.y = element_text(size = 6),
axis.ticks = element_blank(),
strip.background = element_blank(),
strip.text = element_text(hjust = 0, size = 10)
)