This document provides an overview of the third phase of work for the WHO’s Urban Health Initiative, and the ongoing development of a web application to explore health benefits of different sustainable transport policies, with the working title of the Urban Planning and Transport Health Assessment Tool (Upthat; R Lovelace et al. 2018), which was previously named the Active Transport Toolkit (ATT). The main output is be a stable prototype, enabling interactive exploration of active transport scenarios and associated health impacts. The work builds on previous projects funded by the WHO (in grants 2017/773067 and 2017/773067-0). Upthat can be used for many purposes related to the interface between transport and human health, including assessment of exposure to air pollution, changes in the distribution of physical activity, and the spatial distribution of cycling and and walking.
The driving aim of this project across all stages was to create a
flexible framework for analyzing, visualizing and testing a range of
scenarios based on different input datasets and parameters, and to
ensure the framework is able to be updated and modified at any stage
within or beyond the project timetable as desired. We have delivered
that in the form of open source software, the R package upthat, which
users can install and adapt for their own use cases.
This builds on prior work, notably the Propensity to Cycle Tool (PCT)
project (Lovelace et al. 2017). The PCT has become the primary
government endorsed cycle network planning tool in the UK and is being
used by dozens of transport planning organizations to improve the
effectiveness of hundreds of km of cycleway schemes. The international
application of these methods can be supported by new high performace
software. This ‘software ecosystem’ includes dodgr, an R package for
shortest path calculation on spatial networks (Padgham 2019), stplanr,
an R package for processing origin-destination data (Robin Lovelace and
Ellison 2018), data access packages such as pct and stats19 and
interfaces to routing services such as rosm and opentripplanner.
The general applicability of such software raises the possibility of a
globally scalable tool for sustainable transport planning, if only a
user interface existed which did not require specialist programming
skills. Follow-up work will seek funding for such a global tool.
The prototype web application we developed for this project works at the city level, where city authorities have substantial power in many countries. The case study cities were Accra (Ghana), Kathmandu (Nepal) and Bristol (UK). Beyond the generation of an interactive web application, a key research outcome is exploration of methods to convert open data into estimates of transport behaviour down to the street level, and resulting health impacts. The framework enables comparison of multiple scenarios in health-economic terms building on health modelling projects such as the Health Economic Assessment Tool (HEAT Kahlmeier et al. 2014) using locally-provided data.
The previous two phases of this work established and calibrated methods to generate “flow layers” from a range of origins to trip attracting destinations, defined by the type of trip (work, education, etc). Each layer is calculated in two directions (origin (\rightarrow) destination; destination (\rightarrow) origin):
| origin | destination | mode |
|---|---|---|
| home | work | bicycle foot |
| home | education | bicycle foot |
| home | retail | bicycle foot |
| home | bus | foot |
| work | retail | bicycle foot |
| work | bus | foot |
| retail | bus | foot |
| retail | retail | foot bicycle bus |
In the second stage, relative density along each street segment was calculated as follows:
-
Home densities were estimated directly from population density layer (enabling subsequent finer distinctions between demographic groups)
-
Work densities were based on data on “activity centres” (centres of commerce, administration, education), scaled by estimated building sizes.
- Retail densities based on local densities and sizes of retail buildings.
The layers were generated in isolation, with associated levels of uncertainty, but can be combined converting relative flows into absolute flows and then combining the trip counts for each layer at a given level of temporal resolution (daily, on week days, in the first instance).
A spatial model of air pollution was explored with the aim of enabling the automatic creation of pollution heat maps, building on prior methods such as Vara-Vela et al. (2016). Most prior methods, including the Praise Hong Kong Air-Monitoring App, rely on coarse estimates of street network structure to provide crude estimates of vehicular densities. We will model air quality at relatively high resolution, including estimated rates of dispersion away from the street-based sources.
Estimates of exposure to air pollutants require data on both background (e.g. due to nearby industry) and spatially explicit sources (e.g. vehicles on streets). We aimed to combine city-wide background estimates, to generate estimates of aggregate exposure. Initially aimed a temporal snapshot of average pollution levels but the possibility of extending the air pollution component will be built-in. Due to time constraints and unexpected delays in developing and calibrating spatial network analysis models, we have not generated new methods of estimating air pollution heatmaps.
A parallel stream of work investigated the application of the methods used in the PCT project internationally. This involves the following main stages.
- The development of spatial interaction models (SIMs) to estimate flow. Bristol was used as a case study city to demonstrate a general methodology for estimating travel demand between discrete zones in the city (see Figure below).
- The use of routing services such as OpenTripperPlanner to identify plausible routes for each OD pair
- Estimation of mode share for each OD pair and route/mode combination currently and under ‘global’ scenarios of change, for example using the Dirichlet regression methods outlined in the adaptation manual.
- Route network generation, to identify the parts of transport networks with the highest potential for change, which represent promising places for investment in sustainable transport infrastructure and other interventions to enable mode shift away from cars and towards public transport and active modes.
Additional optional stages, highlighting the flexibility of the approach, could include the following.
- The generation of zoning systems for cities where appropriate zoning
systems are not provided by the local authority in question, this
stage of the process has be tackled by the development of a new R
package,
zonebuilder. - The modification of the geographic data from OpenStreetMap used to estimate routes and the re-running the mode choice component to estimate the mode shift response to ‘local’ interventions.
Results of observed data (left), unconstrained SIM (middle) and ‘production constrained’ SIM (right)
Each stage was delivered with reproducible code resulting in open data, for future research, transparency and validation. The emphasis of this third phase was the development of a stable prototype application, building on the experimental software development and proof-of-concept production focus of phases 1 and 2. As outlined in the next section, the ‘stable application’ will have a dedicated domain and be available for at least 2 months after completion of the project. The main stages of the work were:
-
Validation of estimated travel behaviour and flow estimates based on case study cities. The results of this phase are reported in a document hosted at https://github.com/ATFutures/nyped
-
Health integration of changes in mobility behaviour (densities of movement along street segments for a range of journey purposes, modes of transport, and demographic factors) into health-economic measures, extending from the calibration procedure underlying HEAT. The results are reported in the ‘health tab’ of the public facing tool hosted at https://atfutures.github.io/uta/ and described in the document hosted at https://github.com/ATFutures/who3/tree/master/health-econ
-
Scenario development, the definition of policy-facing high-level scenarios and their implementation on open datasets on cities (possibly supplemented with accessible data on travel behaviour, e.g. the percentage of trips made by different modes). The description of the high level scenarios is provided in the adaptation manual’s Scenarios section: https://atfutures.github.io/upthat/articles/adaptation.html#scenarios. Due to issues with the representation of scenarios of change in route network analysis, new methods were developed to estimate levels of mode shift in response to such scenarios, although data limitations prevented results being generated for the full range of scenarios, as outlined in the ‘Modelling mode shift’ section of the adaptation manual.
-
Prototype app Concurrent with the preceding two stages, the prototype was set-up and served from a stable web location, and will be maintained for the duration of the project: https://atfutures.github.io/uta/
-
Health impacts The output of the previous stages will be combined to enable comparison of scenarios in terms of their impact on health-economic measures (see the health impacts tab)
-
User manual, based loosely on the HEAT tool manual, to be used by non experts in case study cities. The manual will allow local stakeholders to understand, utilize, and provide feedback on the tool (see the Upthat user manual hosted at atfutures.github.io/upthat )
-
Adaptation manual, which serves the dual purpose of describing
- How Upthat may be adapted and applied to other, additional locations; and in doing so,
- How Upthat as presented to each location may be adapted and modified following feedback from local stakeholders.
The status of each deliverable is outlined below, with wording from the contract in bullet points and comments under each.
- Deliverable 1 (linked to output 1). Health integration: This stage will involve converting metrics of mobility (densities of movement along street segments for a range of journey purposes, modes of transport, and demographic factors) into health-economic measures, for example by building on HEAT, with input from UHI and HEAT development team.
This has been delivered, described in a report in the health-econ folder in the who3 repo.
- Deliverable 2 (linked to output 1). Scenario development: This stage will involve: (1) setting out high level policy scenarios of active transport uptake; (2) converting these changes into estimates of rates of shift towards walking and cycling down to route network levels; and (3) simulating the impacts of these scenarios on walking and cycling levels citywide. Scenario development will also be strongly informed by the transport scenarios assessed in Accra and Kathmandu as part of UHI project activities.
This has been done, with additional unexpected work done to develop new methods for converting scenarios of change into estimates of mode shift, with explicit consideration of uncertainty (see the Scenarios section of the Adaptation Manual).
- Deliverable 3 (linked to output 1). Health impacts: the output of the previous stages will be combined to enable comparison of scenarios, extending from the calibration procedure underlying HEAT and other methods used in similar tools, such as the Propensity to Cycle Tool.
This has been delivered in the Health Impacts tab on the tool. As discussed in issue 38, we plan to make this interactive in follow-on work. The spatial network modelling approach did not respond as expected to changes in scenarios, hence static results. Furthermore, we plan to integrate outputs from the model with existing open source components in published health models such as HEAT and ITHIM (Woodcock, Givoni, and Morgan 2013).
- Deliverable 4 (linked to output 1). Draft prototype webtool: concurrent with the preceding two stages, the prototype webtool will be set-up and served from a stable web location and will be maintained for the duration of the project. This stage includes the search and update of relevant datasets, coding and thorough documentation of the development process.
Done, a stable and publicly accessible tool is available at https://atfutures.github.io/uta/
- Deliverable 5 (linked to outputs 2 and 3). Draft background text and supporting documentation for users, with information on the tool, including methods used, key references and sources of data and information, and guidance on how to use and interpret results. The manual will presume as little computer expertise as possible, and should be intelligible to a general audience. Users’ manual annex will include detailed text on the methodology used for each of the tools components, which will serve as the basis for the academic paper.
Done, the calibration manuscript has been submitted: https://github.com/ATFutures/who3/tree/master/calibration
- Deliverable 6 (linked to outputs 2 and 3). Draft background text and supporting documentation for adaptation of the tool to other cities, with information on adaptations on the data and datasets used, code and scenarios, and guidance on key elements to consider when adapting the tool to other contexts. This adaptation guidance will inform the development of the academic paper and serve the dual purpose of describing (i) How the tool may be adapted and applied to other, additional locations; and in doing so, (ii) how the tool as presented to each location may be adapted and modified following feedback from local stakeholders.
Done, see these two manuals: https://atfutures.github.io/upthat/articles/upthat.html https://atfutures.github.io/upthat/articles/adaptation.html plus the calibration manuscript.
In the initial plan the work would be conducted over 4 months, with each stage taking between 2 weeks to 2 months, as illustrated below.
Due to delays in contracts and unexpected delays in the route network analysis software development, the contract ran into December 2019.
During the project limitations with the spatial network analysis method were identified, which have resulted static estimates of health outcomes at the city level, rather than the geographically disaggregated estimates of health impacts that were initially envisaged.
To overcome this limitation, a parallel method using origin-destination was developed in the final weeks of the project. Due to time constraints, the method has not been fully implemented for the case study cities. An advantage of this method is that it allows estimates of mode shift at the origin-destination level based on realistic data of trips made. This would allow geographically disaggregated estimates of health impacts resulting from global and local scenarios of change. Implementing this OD method is a priority next step.
An interesting development that has arisen from research into this ‘OD approach’ is an automated zoning system that could support city planners who lack access to appropriate official zones, as the basis of spatial interaction models. An illustration of this is shown in the figure below. This method was developed in the final week of the project so has not been deployed, but we plan explore using such automated zone generation methods as the basis of future work at the OD level.
Illustration of automated zoning systems applied to London and Bristol, a next step could be to apply this to case study cities as the basis of OD analysis.
An additional limitation of the work was that the air pollution exposure has not been deployed as a continuous ‘heatmap’ as expected. Instead, air pollution estimates based on estimates of vehicular flow have been developed, but these have not been verified for the case study cities. Using alternative methods to estimate current and projected future levels of air pollution, perhaps using machine learning methods demonstrated in work by Meng Lu of Utrecht University.
We will continue to work on these discrete components of the project in liaison with the WHO informally.
Kahlmeier, S, P Kelly, C Foster, T Götschi, N Cavill, H Dinsdale, J Woodcock, C Schweizer, H Rutter, and C Lieb. 2014. “Health Economic Assessment Tools (HEAT) for Walking and for Cycling, Methods and User Guide.” World Health Organization Regional Office for Europe, Copenhagen, Denmark, 2014.
Lovelace, R, N Groot, M Adepeju, and M Padgham. 2018. “Estimating Cycling Potential on Route Networks in Accra and Kathmandu.” World Health Organization.
Lovelace, Robin, and Richard Ellison. 2018. “Stplanr: A Package for Transport Planning.” The R Journal 10 (2): 7–23. https://doi.org/10.32614/RJ-2018-053.
Lovelace, Robin, Anna Goodman, Rachel Aldred, Nikolai Berkoff, Ali Abbas, and James Woodcock. 2017. “The Propensity to Cycle Tool: An Open Source Online System for Sustainable Transport Planning.” Journal of Transport and Land Use 10 (1). https://doi.org/10.5198/jtlu.2016.862.
Padgham, Mark. 2019. Dodgr: An R Package for Network Flow Aggregation. Vol. 2. Transport Findings. Network Design Lab. https://doi.org/10.32866/6945.
Vara-Vela, A., M. F. Andrade, P. Kumar, R. Y. Ynoue, and A. G. Muñoz. 2016. “Impact of Vehicular Emissions on the Formation of Fine Particles in the Sao Paulo Metropolitan Area: A Numerical Study with the WRF-Chem Model.” Atmospheric Chemistry and Physics 16 (2): 777–97. https://doi.org/https://doi.org/10.5194/acp-16-777-2016.
Woodcock, James, Moshe Givoni, and Andrei Scott Morgan. 2013. “Health Impact Modelling of Active Travel Visions for England and Wales Using an Integrated Transport and Health Impact Modelling Tool (ITHIM).” PLoS ONE 8 (1). https://doi.org/10.1371/journal.pone.0051462.