Introduction

This vignette supports workshops on advanced usage and development of the Propensity to Cycle Tool (PCT). Beginner and intermediate PCT events focus on using the PCT via the web application hosted at www.pct.bike and the data provided by the PCT in QGIS.

The focus here is on analysing cycling potential in the open source statistical programming language R, in which the majority of the PCT was built. It will show how the code underlying the PCT works, how the underlying data can be accessed for reproducible analysis, and how the methods can be used to generate new scenarios of cycling uptake.

If you are an intermediate user, it may be worth brushing-up on your R skills, e.g. by taking a free online course such as that provided by DataCamp or by working through Chapter 2 onwards of the open source book Geocomputation with R (see reading list below for more transport-specific resources).

Prerequisites

To ensure your computer is ready for the course, you should be able to run the following lines of R code on your computer:

install.packages("remotes")
pkgs = c(
  "cyclestreets",
  "mapview",
  "pct",
  "sf",
  "stats19",
  "stplanr",
  "tidyverse",
  "devtools"
)
remotes::install_cran(pkgs)
# remotes::install_github("ITSLeeds/pct")

To test your computer is ready to work with PCT data in R, try running the following command:

source("https://raw.githubusercontent.com/ITSLeeds/TDS/master/code-r/setup.R")

If the new method does not work or you would like to be more hands on, run the code below. It should result in the map below, showing the % of short trips in Isle of Wight made by active modes.

library(pct)
library(dplyr)   # in the tidyverse
library(tmap) # installed alongside mapview
region_name = "isle-of-wight"
max_distance = 7
zones_all = get_pct_zones(region_name)
lines_all = get_pct_lines(region_name)
# basic plot
plot(zones_all$geometry)
plot(lines_all$geometry[lines_all$all > 500], col = "red", add = TRUE)


# create 'active' desire lines (less than 5 km)
active = lines_all %>% 
  mutate(`Percent Active` = (bicycle + foot) / all * 100) %>% 
  filter(e_dist_km < max_distance)

# interactive plot
tmap_mode("view")
tm_shape(active) +
  tm_lines("Percent Active", palette = "RdYlBu", lwd = "all", scale = 9)

We can also use the data to explore entrenched car dependence, as follows:

# Create car dependent desire lines
car_dependent = lines_all %>% 
  mutate(`Percent Drive` = (car_driver) / all * 100) %>% 
  filter(e_dist_km < max_distance)
tm_shape(car_dependent) +
  tm_lines("Percent Drive", palette = "-RdYlBu", lwd = "all", scale = 9)
#> Legend for line widths not available in view mode.

Exercises

How the PCT works and what you can use it for

Take a look at the image below from (Lovelace et al. 2017). Which parts of the process are of most use for your work?

In groups of 2, identify 3 real world questions/problems that you could use the PCT to help answer/solve

The PCT provides data at 4 geographic levels:

Zones
Desire lines
Routes
Route networks

Which types of data are most appropriate to tackle each of the questions/problems you identified?

Name 3 limitations of the data currently provided by the PCT and discuss how you could overcome them

Getting and viewing PCT data

G1: Using the PCT’s online interface, hosted at www.pct.bike/m/?r=isle-of-wight, identify the MSOA zone that has the highest number of people who cycle to work.
G2: Using data downloaded with the command get_pct_zones(), identify the zone that has highest level of cycling with the function top_n() and save the result as an object called z_highest_cycling (hint: you may want to start by ‘cleaning’ the data you have downloaded to include only a few key columns with the function select(), as follows):

library(pct)
library(dplyr) # suggestion: use library(tidyverse)
z_original = get_pct_zones("isle-of-wight")
z = z_original %>% 
  select(geo_code, geo_name, all, bicycle, car_driver)

G3: Use the plot() command to visualise where on the Ilse of Wight this ‘high cycling’ zone is (hint: you will need to use the plot() function twice, once to plot z$geometry, and again with the argument add = TRUE and a col argument to add the layer on top of the base layer and give it a colour). The result should look something like something this:

G4: Using the online interface, identify the top 5 MSOA to MSOA desire lines that have the highest number of people who cycle to work.
G5: Using the function get_pct_lines(), identify the top 5 MSOA to MSOA desire lines that have the highest number of people who cycle to work (hint: you might want to start with the code shown below).
- Bonus: also find the 5 desire lines with the highest number of people driving to work. Plot them and find the straight line distance of these lines with the function st_length().

# Aim: get top 5 cycle routes
l_original_msoa = get_pct_lines("isle-of-wight")
l_msoa = l_original_msoa %>% 
  select(geo_code1, geo_code2, all, bicycle, car_driver, rf_avslope_perc, rf_dist_km)

Top 5 MSOA to MSOA desire lines with highest number of people cycling (left) and driving (right) in the Isle of Wight.

G6 (Bonus): Repeat the exercise but for LSOA to LSOA desire lines (by setting the argument geography = "lsoa", remember to change the names of the objects you create). The results should look something like this:

Top 5 LSOA-LSOA desire lines with highest number of people cycling (left) and driving (right) in the Isle of Wight.

G7: Why are the results different? What are the advantages and disadvantages of using smaller zones, as represented by the LSOA data above?
G8 (bonus): do the same analysis but with the top 300 routes cycled and driven. Hint: set the line width with lwd = l_top_cycling$bicycle / mean(l_top_cycling$bicycle) to portray the relative importance of each route.

Top 300 LSOA-LSOA desire lines with highest number of people cycling (left) and driving (right) in the Isle of Wight.

Modifying PCT data to identify routes/roads of interest

M1: Building on the example above, add a new column called pcycle to the object l_msoa that contains the % who cycle to work (hint: you might want to start this by typing l_msoa$pcycle = ...) and plot the results (shown in left hand panel in plot below).

l_msoa$pcycle = l_msoa$bicycle / l_msoa$all * 100
# plot(l_msoa["pcycle"], lwd = l_msoa$all / mean(l_msoa$all), breaks = c(0, 5, 10, 20, 50))

M2 (bonus): identify road segments with the highest estimated number of people cycling currently, and under the Go Dutch scenario (hint: you can download the route network with get_pct_rnet("isle-of-wight"))

Scenarios of change

S1: Generate a ‘Go Dutch’ scenario for the Isle of Wight using the function uptake_pct_godutch() (hint: the following code chunk will create a ‘Government Target’ scenario):

l_msoa$euclidean_distance = as.numeric(sf::st_length(l_msoa))
l_msoa$pcycle_govtarget = uptake_pct_govtarget(
  distance = l_msoa$rf_dist_km,
  gradient = l_msoa$rf_avslope_perc
  ) * 100 + l_msoa$pcycle

Percent cycling currently (left) and under a ‘Go Dutch’ scenario (right) in the Isle of Wight.

S2: Think of alternative scenarios that would be useful for your work
S3 (bonus): look inside the function pct_uptake_godutch() - how could it be modified?

Routing

R1: Using the function route_osrm() find the route associated with the most cycled desire line in the Isle of Wight. The result should look similar to that displayed in the map below (hint: you may want to start your answer with the following lines of code - warning: the function may need to run a few times before it works):

library(stplanr)
l_top = l_msoa %>% 
  top_n(n = 1, wt = bicycle)

R2: What are the problems associated with this route from a cycling perspective? Take a look at the help page opened by entering ?route_osrm to identify the reason why the route is not particularly useful from a cycling perspective.
R3: Regenerate the route using the function line2route(). What is the difference in the length between each route, and what other differences can you spot? Note: this exercise requires an API Key from CycleStreets.net.
R4 (bonus): what features of a routing service would be most useful for your work and why?

Route networks

RN1: Generate a ‘route network’ showing number of people walking in the top 30 routes in the Isle of Wight, allocated to the transport network (hint: use the overline2() function and begin the script as follows, the results should look similar to the results below):

route_data = sf::st_sf(wight_lines_30, geometry = wight_routes_30$geometry)

RN2: Download the travel to school route network and compare the results with the route network created for RN1.
- Which roads have greatest overlap between the two route networks?
- For more information on the travel to school layer, see Goodman et al. (2019). What other trip purposes would you like to see in tools for cycle planning?

Ideas for further work and a ‘minihack’

Create a route network reflecting where you would invest if the priority was reducing car trips of less than 5 km
Design interventions to replace short car trips across London (or another region of your choice) using the PCT methods/data to support your decisions
Identify quiet routes and design a quiet route network for city/region of your choice, e.g. Westminter
Import alternative origin-destination datasets and use those as the basis for propensity to cycle analysis for trip purposes other than single stage commutes, encapsulated in the commut layer in the PCT
Any other layers/scenarios/hacks: welcome! Comments in this repo’s issue tracker also welcome.

Propensity to Cycle Tool Advanced Workshop