Accessibility

1. Introduction

Accessibility indicators measure the ease with which opportunities, such as jobs, can be reached by a traveler from a particular location (Levinson and et al. 2020). This vignette shows how to calculate and visualize accessibility in R using the r5r package using a reproducible example. In this example, we will be using a sample data set for the city of Porto Alegre (Brazil) included in r5r.

There are two ways to calculate / visualize accessibility using r5r. The quick and easy option is using the r5r::accessibility() function. The other alternative requires one to first calculate a travel time matrix, and then to use the {accessibility} package. This is a more flexible options because the {accessibility} package provides a wider range of options of accessibility metrics. We will cover both approaches in this vignette.

Before we start, we need to increase Java memory + load a few libraries, and to build routable transport network.

2. Build routable transport network with setup_r5()

Increase Java memory and load libraries

First, we need to increase the memory available to Java and load the packages used in this vignette. Please note we allocate RAM memory to Java before loading our libraries.

options(java.parameters = "-Xmx2G")

library(r5r)
library(accessibility)
library(sf)
library(data.table)
library(ggplot2)
library(interp)
library(h3jsr)
library(dplyr)

To build a routable transport network with r5r, the user needs to call setup_r5() with the path to the directory where OpenStreetMap and GTFS data are stored.

# system.file returns the directory with example data inside the r5r package
# set data path to directory containing your own data if not running this example
data_path <- system.file("extdata/poa", package = "r5r")

r5r_core <- setup_r5(data_path)

3. Accessibility: quick and easy approach

There are different types of accessibility metrics. One of the simplest ones is the cumulative-opportunity metric, which counts the number of opportunities accessible from each location considering a maximum travel time cutoff. This is what we’ll be calculating in this vignette using the parameter decay_function = "step".

In this example, we will be calculating the number of schools and public healthcare facilities accessible by public transport within a travel time of up to 20 minutes. The sample data provided contains information on the spatial distribution of schools in Porto Alegre in the points$schools column, and healthcare facilities in the points$healthcare column.

With the code below we compute the number of schools and healthcare accessible considering median of multiple travel time estimates departing every minute over a 60-minute time window, between 2pm and 3pm. The accessibility() function can calculate access to multiple opportunities in a single call, which is much more efficient and convenient than producing a travel time matrix of the study area and manually computing accessibility.

# read all points in the city
points <- fread(file.path(data_path, "poa_hexgrid.csv"))

# routing inputs
mode <- c("WALK", "TRANSIT")
max_walk_time <- 30      # in minutes
travel_time_cutoff <- 20 # in minutes
time_window <- 60        # in minutes
departure_datetime <- as.POSIXct("13-05-2019 14:00:00",
                                 format = "%d-%m-%Y %H:%M:%S")

# calculate accessibility
access1 <- r5r::accessibility(r5r_core,
                        origins = points,
                        destinations = points,
                        mode = mode,
                        opportunities_colnames = c("schools", "healthcare"),
                        decay_function = "step",
                        cutoffs = travel_time_cutoff,
                        departure_datetime = departure_datetime,
                        max_walk_time = max_walk_time,
                        time_window = time_window,
                        progress = FALSE)

head(access1)

Mind you that the r5r::accessibility() also allow users to calculate gravity-based accessibility metrics, which can be calculated by setting the decay_function to one of the following: "exponential" "fixed_exponential", "linear" or "logistic". Nonetheless, there are several other types of accessibility metrics not implemented in R5, including floating catchment area metrics, travel cost to closest N opportunities, time interval based cumulative opportunity, etc. This is where the {accessibility} package comes in.

4. Accessibility: flexible approach

The {accessibility} package provides a much more flexible approach to calculate accessibility estimates. A key input here is a travel time matrix, which we calculate using r5r:

# calculate travel time matrix
ttm <- r5r::travel_time_matrix(r5r_core,
                        origins = points,
                        destinations = points,
                        mode = mode,
                        departure_datetime = departure_datetime,
                        max_walk_time = max_walk_time,
                        time_window = time_window,
                        progress = FALSE)
head(ttm)

Now to calculate a traditional cumulative opportunity metric like we did above, we just need to call the accessibility::cumulative_cutoff() function, and pass our travel time matrix and land use data as input:

# calculate accessibility
access_edu <- accessibility::cumulative_cutoff(travel_matrix = ttm, 
                                        land_use_data = points,
                                        opportunity = 'schools',
                                        travel_cost = 'travel_time_p50',
                                        cutoff = 20)

access_health <- accessibility::cumulative_cutoff(travel_matrix = ttm, 
                                        land_use_data = points,
                                        opportunity = 'healthcare',
                                        travel_cost = 'travel_time_p50',
                                        cutoff = 20)
head(access_edu)
head(access_health)

5. Map Accessibility

The final step is mapping the accessibility results calculated earlier. We can use at least two different approaches to map our accessibility estimates.

5.1 Choropleth maps

The first approach is to use choropleth maps. In our example, each point of reference is the centroid of a H3 hexagonal grid at a fine spatial resolution. In this case, we basically need to retrieve the polygons of the spatial grid, and merge it with our accessibility estimates.

# retrieve polygons of H3 spatial grid
grid <- h3jsr::cell_to_polygon(points$id, simple = FALSE)

# merge accessibility estimates
access_sf <- left_join(grid, access1, by = c('h3_address'='id'))

# plot
ggplot() +
  geom_sf(data = access_sf, aes(fill = accessibility), color= NA) +
  scale_fill_viridis_c(direction = -1, option = 'B') +
  labs(fill = "Number of\nfacilities within\n20 minutes") +
  theme_minimal() +
  theme(axis.title = element_blank()) +
  facet_wrap(~opportunity) +
  theme_void()

5.2 Spatial interpolation

An alternative approach is to use our accessibility estimates for each reference point and do some spatial interpolation so we can have a smoother spatial distribution. The code below demonstrates how to do that, producing a prettier map.

# interpolate estimates to get spatially smooth result
access_schools <- access1 %>% 
  filter(opportunity == "schools") %>%
  inner_join(points, by='id') %>%
  with(interp::interp(lon, lat, accessibility)) %>%
  with(cbind(acc=as.vector(z),  # Column-major order
             x=rep(x, times=length(y)),
             y=rep(y, each=length(x)))) %>% as.data.frame() %>% na.omit() %>%
  mutate(opportunity = "schools")

access_health <- access1 %>% 
  filter(opportunity == "healthcare") %>%
  inner_join(points, by='id') %>%
  with(interp::interp(lon, lat, accessibility)) %>%
  with(cbind(acc=as.vector(z),  # Column-major order
             x=rep(x, times=length(y)),
             y=rep(y, each=length(x)))) %>% as.data.frame() %>% na.omit() %>%
  mutate(opportunity = "healthcare")

access.interp <- rbind(access_schools, access_health)

# find results' bounding box to crop the map
bb_x <- c(min(access.interp$x), max(access.interp$x))
bb_y <- c(min(access.interp$y), max(access.interp$y))

# extract OSM network, to plot over map
street_net <- street_network_to_sf(r5r_core)

# plot
ggplot(na.omit(access.interp)) +
  geom_sf(data = street_net$edges, color = "gray55", size=0.01, alpha = 0.7) +
  geom_contour_filled(aes(x=x, y=y, z=acc), alpha=.7) +
  scale_fill_viridis_d(direction = -1, option = 'B') +
  scale_x_continuous(expand=c(0,0)) +
  scale_y_continuous(expand=c(0,0)) +
  coord_sf(xlim = bb_x, ylim = bb_y, datum = NA) + 
  labs(fill = "Number of\nfacilities within\n20 minutes") +
  theme_void() +
  facet_wrap(~opportunity)

Cleaning up after usage

r5r objects are still allocated to any amount of memory previously set after they are done with their calculations. In order to remove an existing r5r object and reallocate the memory it had been using, we use the stop_r5 function followed by a call to Java’s garbage collector, as follows:

r5r::stop_r5(r5r_core)
rJava::.jgc(R.gc = TRUE)

If you have any suggestions or want to report an error, please visit the package GitHub page.

References

Levinson, David, and et al. 2020. “Transport Access Manual: A Guide for Measuring Connection Between People and Places,” January. https://ses.library.usyd.edu.au/handle/2123/23733.