--- title: 'Mapping population data' date: "`r Sys.Date()`" output: rmarkdown::html_vignette urlcolor: blue vignette: > %\VignetteIndexEntry{Mapping population data} %\VignetteEngine{knitr::rmarkdown} \usepackage[utf8]{inputenc} --- ```{r, include = FALSE} knitr::opts_chunk$set( collapse = TRUE, comment = "#>", eval = identical(tolower(Sys.getenv("NOT_CRAN")), "true"), out.width = "100%" ) ``` Spatial distribution of population ```{r, message = FALSE, eval = TRUE} # load libraries library(aopdata) library(data.table) library(ggplot2) library(sf) library(scales) ``` **Download population data** ```{r, message = FALSE, eval = TRUE} # download aop data df <- read_population(city='Fortaleza', year=2010, geometry = TRUE, showProgress = FALSE) ``` **Map Population** Spatial distribution of the population ```{r, message = FALSE, eval = TRUE} ggplot() + geom_sf(data=subset(df, P001>0), aes(fill=P001), color=NA, alpha=.8) + scale_fill_distiller(palette = "YlOrRd", direction = 1)+ labs(title='Population distribution', fill="Total population") + theme_void() ``` **Map income levels** Spatial distribution population by income decile ```{r, message = FALSE, eval = TRUE} ggplot() + geom_sf(data=subset(df, !is.na(R002)), aes(fill=factor(R003)), color=NA, alpha=.8) + scale_fill_brewer(palette = "RdBu") + labs(title='Average household income per capita', fill="Income decile") + theme_void() ``` **Map racial groups** Spatial distribution population black population ```{r, message = FALSE, eval = TRUE} df$prop_black <- df$P003 / df$P001 ggplot() + geom_sf(data=subset(df, P001 >0), aes(fill=prop_black), color=NA, alpha=.8) + scale_fill_distiller(palette = "RdPu", direction = 1, labels = percent)+ labs(title='Proportion of black population', fill="Black population") + theme_void() ```