Background

• We need to build dense walking/cycling/wheeling networks
• But where?

Source: Cycle Routing Uptake and Scenario Estimation (CRUSE) tool

Tools of the trade

Modelling framework and long-standing limitations

What is Jittering?

Source: Lovelace, R., Félix, R., & Carlino, D. (2022, January 13). Jittering: A computationally efficient method for generating realistic route networks from origin-destination data. Transport Findings, in Press https://doi.org/10.31219/osf.io/qux6g

Current default: centroid-based desire lines (+routes+rnets)

Jittering in action: minimal reproducible example

Jittering a larger dataset

Adding value and detail to existing OD data. Source: Lovelace, Félix and Carlino (2022)

Resulting route network

Validating the approach: MKI

Data from Edinburgh. Source: GISRUK 2022 conference paper.

Model experiments: jittering parameters

Results From Edinburgh

Changing jittering and routing params

The 'jittering only' approach is assumes perfect routing, not true

Model/data discrepancies may be more due to routing than jittering/OD parameters

Enter Lisbon!

Network level results

Summary of results

Using the results: biclar project

• Project funded by Lisbon Metro Region's Department for Transport (population: 2.8m)
• Need for evidence to support effective investment in cycle plans
• Current level of cycling regionally: 0.5%
• Targets: at least 4% by 2025 and 10% of trips by 2030 with shift coming from cars
• Comparison with UK: ~doubling from 2013 to 2025 in cycling
  • from "0.8 billion stages in 2013 to 1.6 billion stages in 2025" (CWIS, 2017)

Biclar: approach

• R package and open source code base building on PCT approach
• See https://u-shift.github.io/biclar/

Existing and simulated network

Which trips will shift?

Next steps

Exploring the parameter space: different origin and destination points + weights, routing 'engines', disaggregation.

References

Cyclability hackathon at 15:00 today: https://github.com/itsLeeds/cyclability

Biclar project and open source codebase: https://u-shift.github.io/biclar/

Carlino, Dustin, et al. A/B Street. Zenodo, 2022, https://doi.org/10.5281/zenodo.6331922.

Lovelace, Robin, et al. Exploring Jittering and Routing Options for Converting Origin-Destination Data into Route Networks: Towards Accurate Estimates of Movement at the Street Level. June 2022, https://doi.org/10.5194/isprs-archives-XLVIII-4-W1-2022-279-2022.

Lovelace, Robin, et al. “Jittering: A Computationally Efficient Method for Generating Realistic Route Networks from Origin-Destination Data.” Findings, Apr. 2022, p. 33873, https://doi.org/10.32866/001c.33873.

Appendix: Reproducible code I: Rust implementation

See reproducible repo + manuscript here: https://github.com/Robinlovelace/odnet

System command line implementation (compile Rust code):

cargo install --git https://github.com/dabreegster/odjitter
odjitter jitter --od-csv-path od_iz_ed.csv \
  --zones-path iz_zones11_ed.geojson \
  --subpoints-path road_network_ed.geojson \
  --max-per-od 10 --output-path output_max50.geojson

Reproducible code II: R implementation

(See code in slides.Rmd in robinlovelace/foss4g22 for code to get data)

remotes::install_github("dabreegster/odjitter", subdir = "r")

od = read_csv("od_iz_ed.csv")
zones = sf::read_sf("iz_zones11_ed.geojson")

od_jittered = odjitter::jitter(
  od = od,
  zones = zones,
  subpoints = sf::read_sf("road_network_ed.geojson")
  )
od_jittered2 = odjitter::jitter(
  od = od,
  zones = zones,
  subpoints = sf::read_sf("road_network_ed.geojson"),
  disaggregation_key = "all",
  disaggregation_threshold = 10
  )

Results of reprex 1

od_sf = od::od_to_sf(
  od,
  zones
)
nrow(od_sf)

## [1] 10394

nrow(od_jittered)

## [1] 10394

nrow(od_jittered2)

## [1] 22432

Plot of unjittered data

library(dplyr)
plot(od_sf$geometry,
     lwd = od_sf$all / 500)

Results of reprex 2

plot(od_jittered$geometry,
     lwd = od_jittered$all / 500)

Results of reprex 3

plot(od_jittered2$geometry,
     lwd = od_jittered2$all / 50)

Alternative validation datasets: OA-WPZ data

There are 17,848,366 OA to WPZ records, 170k OAs, 54k WPZ

For 5km buffer around London, 1.5 million OD pairs with destinations

Reproducible example

u = "https://github.com/ITSLeeds/od/releases/download/v0.3.1/od_intra_top_sf.geojson"
desire_lines_oa_wpz_1k = sf::read_sf(u)
oas_in_buffer = sf::read_sf("https://github.com/ITSLeeds/od/releases/download/v0.3.1/oas_in_buffer.geojson")
wpz_in_buffer = sf::read_sf("https://github.com/ITSLeeds/od/releases/download/v0.3.1/wpz_in_buffer.geojson")
library(tmap)
tmap_mode("view")

## tmap mode set to interactive viewing

m = tm_shape(desire_lines_oa_wpz_1k) +
  tm_lines() +
  tm_shape(oas_in_buffer) + tm_dots(col = "darkgreen") +
  tm_shape(wpz_in_buffer) + tm_dots(col = "darkred")