Cycling infrastructure and bikeability/cyclability indicators

[carlhiggs]

Cycling indicator working group meeting 1

20 March 2023
Attendees: Afshin, Marc, Xavi, Carl (minutes)

Context

Following feedback sessions on use of the software, several users' desire for incorporation of cycling indicators for benchmarking, monitoring and mapping healthy and sustainable city progress led to organisation of this working group to identify ways to do this. Initial suggestions raised in the context of the feedback sessions and related e-mails included a 'bikeability index', measures of accessibility, density of cycling infrastructure, as well as coverage and connectivity, and in addition the relevance of additional data on elevation and meterological conditions were noted as relevant.

Backgrounds

The attendees approached this from diverse backgrounds united by the working group's theme. Afshin, Marc and Xavi (respective backgrounds in transport modelling, geography and mobility studies) had all previously undertaken work related to provision of cycling infrastructure in urban settings, with a shared interest in active transport: Afshin's work has focused on modelling how people could use cycling to get to where they need/want to go; Marc and Xavi have collaborated on mapping and evaluating cycling infrastructure presence and quality, including undertaking surveys of resident perceptions and considering related complementary features. All three have experience using OpenStreetMap cycling data. My (Carl's) expertise is in the development of software to support indicators for use in research and planning; I'm interested in cycling indicators as a case study for how to incorporate analysis and validation for new indicators desired by users into the software.

Aims

The group articulated a shared interest in developing a spatial composite indicator of bikeability or cyclability, but recognised that: 1) existing measures for this exist (we should familiarise ourselves with them); 2) a focus on sub-indicators is needed in first instance (one thing at a time); 3) different settings have different needs (how do we capture this); 4) should be replicable; 5) need to be able to demonstrate/justify meaningfulness/relevance and evaluate quality of the measure.

Related to the last point, clarity is needed on

• what do we want to achieve?
  • for now, I think the above articulates this, but should aim for a formal agreed statement
• how do we measure success for this?
  • need to incoporate mechanisms and guidance for validation (e.g. using official gold standard data for comparison, and using travel survey data to evaluate relationship between indicators and the behaviours the bulit environment forms they aim to capture are theorised to predict or elicit
• how is this relevant for planning and/or research on healthy and sustainable cities?
  • we need to articulate this to justify inclusion as a formal indicator in the study for inclusion in the Global Observatory of Healthy and Sustainable Cities

While a starting assumption is that OpenStreetMap would be used as a general basis for this indicator, it is also recognised that some indicators will require supplementary data from other sources and different settings may require or prefer alternate data, and provision should be made for this (including for sensitivity analyses involving A/B comparison and evaluation of impact of different data choices).

Example indicators

• access to docking stations
• access to linear/zonal infrastructure
  • not just precence, but connectivity
    • need to consider how to evaluate this
  • access to linear/zonal infrastructure with connectivity meeting target
  • from this address, how far can you go using cycling amenable routing, given available data?
  • additional considerations:
    • terrain, bike racks (for locking bikes?), street trees/vegetation/shade
      • terrain and vegetation could potentially be incorporated through raster data

While the group agreed composite indices like walkability or bikeability are useful for communication and high level summaries, policy makers and researchers both require the specific underlying indicators and measures for the detail required in planning and modelling cycling infrastructure.

Limitations and challenges

linear vs zonal infrastructure/suitability

The group discussed how cycle lanes are just one lens for considering cyclability: a region may be quite pleasant for cycling but not have formal cycle lanes. In this case they would score poorly on an indicator specifically focused on seperated cycle lane provision. I would argue that's not necessarily bad (it can be useful to know what is not provided, or represented in data -- both are explanations for a poor outcome that should be considered), but highlights the need for multiple indicators to tell a fuller story (e.g. yes, there are not cycle lanes provided, but there are low speed zones that may also provide appropriate options for cycling).

It was agreed that low traffic zones in general can be desirable and pleasant for cycling, and an example was given of Gracia in Barcelona which has a 10km/h speed limit and is a shared space for different transport modes including cycling and walking. This is the argument fo looking at low stress infrastructure zones, as Afshin has done in his work: in Melbourne, many people don't havea bike lane near, so important to recognise a 20km/h shared street is also good.

Speed limits are not necessarily always provided or up to date in OpenStreetMap; Afshin pointed out that look up tables can be used to infer speed limits for missing data given infrastructure tags to indicate type of route (eg street width, or other zonal attributes), and shared a link for his work where examples for how this could be done had been published:
Jafari, Afshin, Alan Both, Dhirendra Singh, Lucy Gunn, and Billie Giles-Corti. ‘Building the Road Network for City-Scale Active Transport Simulation Models’. Simulation Modelling Practice and Theory 114 (1 January 2022): 102398. https://doi.org/10.1016/j.simpat.2021.102398.

• [Table 1. OSM road hierarchy and their default values for Australia.] (https://www.sciencedirect.com/science/article/pii/S1569190X21001003#tbl1)
• Table 2. OSM tag combinations used to identify different bikeway types.

(this is fantastic; thank you Afshin!)

We discussed that there are existing indicators, but some are 'terrible' - e.g. % of roads that have bike lanes; that doesn't say much (as per above points). This could also have challenges with 'noisy' derived network data which may be designed for mapping or routing rather than quantifying lengths of routes (paths may be replicated, and cleaning/consolidating network involves methodological choices and compromises). We discussed how we approached this in our measurement of local walkable neighbouhoods, that effectively creates a buffered zone for the walkable area --- so not distance that is measure, but area (which can encompass multiple paths, so has some robustness to this problem). We can do this when evaluating access/connectivity/coverage of cycling 'zones'.

In addition to the typologies shared by Afshin, I noted that OSMnx (which our software uses) also has defaults for deriving 'bike' networks, as well as aspects relating to speed and elevation (using a local raster).

So

• the software will need to use some configuration look up tables such as these to provide defaults where information is otherwise sparse to support deriving cycling indicators using OpenStreetMap data
• simultaneously should allow for switching out for custom/official data for evaluation or where otherwise required (ie. validate OpenStreetMap, and also not rely on it solely).
• Access to cycling infrastructure in itself isn't so much the point, but where you can get to (in a convenient/safe/pleasant way, so its practical/easy to use cycling as a mode of transport); so its also about weighting for co-location/access to destinations along routes through contiguous cycling-amenable zones (bikeable neighbourhoods)

Elevation (slope)

We discussed the relevance of slope, and the challenges accounting for this. While digital elevation model/terrain data may be available that can support inference around this, you really need a bare earth model (ie. not with buildings, as these could be averaged over and suggest hilliness where hilliness isn't), and at a reasonable resolution (values suggested as acceptable were 10m-25m).

If one were choosing a route from one location to another specific location, it may be possible to optimise for certain slope conditions (e.g. a preference for moderate to no slope, or negative slope perhaps). However, it isn't necessarily obvious how best to interpret slope when calculating some overall summary score for cycling amenability. Perhaps high variability and extreme values in slope are penalised to represent inconvenience/detraction (e.g. may be nice going down, not so nice going up --- unless you want to get your heart rate up perhaps?).

There may be interactions of slope with meteorological conditions too --- icy and wet slopes may be more dangerous and less desirable as routes for cycling.

Marc provided a link to a 25m digital elevation model that could be used for trialling indicators accounting for this in Spanish cities: https://centrodedescargas.cnig.es/CentroDescargas/index.jsp

I've previously trialled using the 30m resolution data from JAXA in use with OpenTripPlanner: https://www.eorc.jaxa.jp/ALOS/en/dataset/aw3d30/aw3d30_e.htm

Next Steps

• Create a github repository branch for cycling indicators
• Trial implementing an indicator access to docking stations
• Schedule series of cycling catch ups (monthly?)
• and continue to plan for next steps for evaluating zonal bikeability/cyclability-related indicators

[marcdmallafre]

Regarding the slope, maybe this repository can help. Basically there must be somewhere to get information associated with geometries (in this case from Openstreetmap i think):
https://github.com/gpxstudio/gpxstudio.github.io

[carlhiggs]

I reckon they must be using a digital elevation model API in there somewhere; thanks for sharing - will be interesting to check out how others implement this!

[carlhiggs]

Regarding the 'Next Steps' actions from meeting 1, these have all been done on my side:

• The new repository for prototyping cycling indicators is currently accessible here
• I implemented proof of concept indicators for bicycle rental and bicycle parking using OpenStreetMap data (view changes here)
• and we scheduled our next catch up :)

Regarding the proof of concept indicators, I modified the current report template to illustrate how the generated data can also be used to plot maps for these new indicators and they can be incorporated with translation into a report.

For example, some snippets of:

Including the prototype indicators in the radar chart (the reference bars are just for 25%, 50% and 75%; don't really mean anything for this)

Including spatial distribution maps translated to Spanish in the report:

So, not suprisingly percentage of population with access to bicycle renting and bicycle parking as represented using OpenStreetMap data for Las Palmas is low-ish (in absolute terms) and concentrated to the coastal, urban areas of the city. I'm not sure that these are meaningful indicators -- but they provide a proof of concept and illustration of how we can modify our OpenStreetMap destination definitions to identify new kinds of POIs and develop these into indicators.

You can download the code from the cycling branch here if you want to run this example here, and generate a geopackage for Las Palmas containing these proof of concept indicators and corresponding maps and report.

It was a good exercise defining the new indicators as it highlighted some areas that we could streamline to make the process more easy (eg as per issues #233 and #234).

[carlhiggs]

Cycling indicator working group meeting 2

19 April 2023
Attendees: Afshin, Ana, Marc, Tayebeh, Xavi, Carl (minutes)

Dr Tayebeh Saghapour joined us for this meeting, a research fellow in Healthy Liveable Cities Lab at RMIT. Tayebeh did work related to walkability and cycling in her PhD 5 years ago, and is now looking to create a national bikebaility index for Australian cities funded by a recently awards grant.

Ahead of our meeting we shared links to a couple of papers, one of Tayebeh's and another involving our colleague Anne Vernez Moudon. Citations with links and abstracts with indicators that may be of interest bolded copied below --- and this ties in with our main agreed 'to do' for this meeting

• Literature check: we'll try and report back on some relevant papers and add summary in comment below
• and... add this as a task on GitHub (Afshin request!)

Tayebeh Saghapour, Sara Moridpour & Russel Thompson (2019) Sustainable transport in neighbourhoods: effect of accessibility on walking and bicycling, Transportmetrica A: Transport Science, 15:2, 849-871, DOI: 10.1080/23249935.2018.1540502

A growing number of recent studies have focused on improving the sustainability of transportation systems by routinely converting motorised travel to walking and conventional bicycling. The importance of physical activity has recently attracted the attention of practitioners as well as planners and policy-makers. In order to identify effective strategies for increasing active transportation, planners need to identify how current levels of accessibility in neighbourhoods affect walking and cycling trips. Despite a substantial amount of research on modelling active transportation, there have been limited studies on the importance of accessibility considering availability of activities and travel distances for pedestrians and cyclists. Hence, this study employs new approaches for measuring cycling and walking accessibility against land-use features in separate models to examine how accessibility can affect participation in active transportation. Key findings indicate that more accessible neighbourhoods have more active trips, while models using accessibility measurements show better fit on data.

• Land use measures
  • LUM entropy index
  • population density
• Design measures
  • roadway length relative to area (eg in walkable neighbourhood area)
  • intersections relative to intersections and dead-ends combined within area
• Accessibility measures
  • Cycling accessibility index
    • diversity of land uses (same as entropy index above?)
    • activity count in neighbourhood (statistical areas)
    • travel cost OD matrix
    • (a lot more to this! will be interesting to dive deeper when time)

Tayebeh summarised her previous work (apologies if my brief notes misrepresent!) involving area based, geographical measures --- with cycling service areas, consideration of accessibility, length of bicycle routes, and accounting for elevation.

Wang, L. et al. (2023) ‘Designing bike-friendly cities: Interactive effects of built environment factors on bike-sharing’, Transportation Research Part D: Transport and Environment, 117, p. 103670. Available at: https://doi.org/10.1016/j.trd.2023.103670.

Highlights

• Geographical detector models facilitate a comprehensive approach to urban Design.
• Interaction detector measures combined effects of the built environment on Cycling.
• Street network centrality has the largest explanatory power on cycling.
• A sense of enclosure defined by streetscape elements encourages cycling.

Designing bike-friendly cities could promote health and mitigate climate change. Most studies of the association between the built environment and cycling used the “5Ds” framework and linear modeling. However, the built environment exerts complex influences on travel behavior. To better inform urban design for cycling, this study employed geographical detector models that quantify the explanatory power of individual and interactively paired built environment factors on bike-sharing. Data came from 6.5 million bike-sharing orders in Shanghai. Expectedly, we found that street network centrality and important facilities like supermarkets and libraries have the greatest independent and interactive effects. More surprisingly, streetscape elements, including sky view and building frontage, offered significantly higher explanatory power when paired with each other or with street network centrality and important facilities. By identifying the overlooked interactive effects of urban environment factors, the study provides guidance for urban designers to consider combinations of factors that effectively promote cycling.

• we already have supermarket-like locations identified; we could allow for incorporation of community locations like libraries using OpenStreetMap and/or custom data
• we could calculate street network centrality measures (example code using OSMnx here this

Brief discussion on 'proof of concept' indicators

As mentioned, my primary interest in this group is from a technical perspective -- its a vehicle for me to scope out how to create mechanisms to facilitate including new indicators. I am interested in the enthusiasm some of you have expressed for cycling indicators, and helping to find ways to implement this technically: that is how I envision my main contribution, in addition to facilitating the group. I can help implement the indicators you are interested in, and in doing this expand the functionality of our software as well as provide indicators of interest.

So, in the spirit of this, to get things started I created a couple of indicators for access to amenities (described in the post above): docking stations, and bicycle parking, as represented on OpenStreetMap. Ana advised that more detail could be had by honing in with additional tags, such as

• bicycle_rental=docking_station
• bicycle_parking=stands

In both of these cases it seems that 99% of the time these key-value pairs co-occur with the terms already used (amenity=bicycle_rental) and (amenity=bicycle_parking) --- although only 32% of amenity=bicycle_rental is tagged as bicycle_rental=docking_station. We also discussed the value of using tags for exclusion. For example, 3.3% of amenity=bicycle_parking tags globally are also tagged as 'access=private'. In addition Ana suggested that bicycle rental from shops should be excluded. So there is a mechanism we can use for this in the OpenStreetMap definition spreadsheet we use, by defining a 'NOT' clause for key-value pairs under relevant categories where these are to be excluded.

• To do: demonstrate proof of concept exclusion of private bicycle parking

Relevant cycling measures

We had a pretty broad discussion around this, in terms of what might be of interest/value to planners/policy makers/and/or researchers, but also what could be feasible to calculate.

adapting existing walkability index to a cycling index

One approach from a feasibility perspective was to replicate the existing implementation of a walkability index (calculated for sample points, based on local catchment and accessibility metrics) and summarised for areas at a range of scales) but with modification of thresholds (e.g. drawing on targets from Tayebeh's paper) appropriate for cycling.

modifying network to reflect cycling

Related to the above is the use of a 'cycling' network, rather than a hybrid pedestrian/cycling network. While a basic implementation of this could be trivial - for example, these papers describe using OSMnx to model a 'bikeable' network using a preset typology:

• Alattar, M.A., Cottrill, C. and Beecroft, M. (2021) ‘Modelling cyclists’ route choice using Strava and OSMnx: A case study of the City of Glasgow’, Transportation Research Interdisciplinary Perspectives, 9, p. 100301. Available at: https://doi.org/10.1016/j.trip.2021.100301.
• Yen, Y., Zhao, P. and Sohail, M.T. (2021) ‘The morphology and circuity of walkable, bikeable, and drivable street networks in Phnom Penh, Cambodia’, Environment and Planning B: Urban Analytics and City Science, 48(1), pp. 169–185. Available at: https://doi.org/10.1177/2399808319857726.

However, we discussed the challenges of strict restriction to cycling specific infrastructure --- Afshin recently did this in an analysis for Melbourne, and 98% of people couldn't get around doing this. Xavi pointed out that this is an interesting/powerful finding from a policy-relevance perspective. (and I agree!) However, Afshin's point was that good infrastructure was in many instances close, and beyond the exceptions allowed for some good cycling routes overall. So --- that led me to think about some kind of route quality scoring (eg while distance is a cost when traversing the network, that cost could be modified by other attributes relevant to cycling, resulting in a more abstract, but potentially relevant estimate for amenability of a route --- but there are challenges of abstract measures, as noted below).

Tayebeh asked about what are the criteria for low stress; and Afshin, based on his work, suggested typical things include: traffic volume, provision of biking infrastructure, type of road (arterial etc), intersection density. Tayebeh was concerned that measuring 'stress' requires a lot of information - so which cities have this data? Afshin suggested that a basic typology for this can be constructed from OpenStreetMap based on traffic speed, number of lanes and road hierarchy to infer speed where not otherwise provided. The gap is traffic volume.

Origin-Destination analyses

Marc suggested it would be interesting to conduct origin-destination analyses for residential to work zones and evaluate connection.

A technical question (related to Tayebeh's re data availability) is how do you identify these residential and work zones in diverse settings with variable data availability? Thinking out loud, i mused that given we have population estimates, and OpenStreetMap/custom data as possibilities there would be potential to relate population to clusters of mixed amenity -- where there are intensities of amenity, particularly mixed amenity, that could be considered a work location, and while that may intersect to some degree with population distribution (eg mixed use), some locations would have low amenity and would presumably be more or less residentially zoned. This is all a bit hypothetical --- but gist is, using clusters of amenities to identify some intense zones of activity ( 'activity centre' ) that could serve as proxies for employment (or otherwise be presumably attractive in some ways to go to). Marc suggested hospitals and other amenities could be aspects of interest.

The discussion touched on the concept of work at another point, and I think there was a sense that 'work' is hard to define in a general sense, but specific kinds of destinations are tangible.

Clarity of indicators

Afshin asked about what is the kind of thing that we are interested in calculating, and Xavi suggested:

*The percentage of residents having access to 10 amenities in a 500m threshold using bike-related infrastructure

This particular answer struck me with its clarity --- unlike an abstract 'bike score' or the result of some 'cycling amenability function', a planner or policy maker can understand immediately what this is measuring. So, I agree -- clear measures like this are the priority. This could use 'desirable distances' from literature (eg Tayebeh's paper; apparently this was from UK research, using 4 types of destination). We could make a start implementing an indicator like this using our existing destinations (eg supermarket, convenience, PT) --- although, is cycling to a tram or bus useful if you can't take your bike on it? and would you leave your bike there? (in Melb might get stolen?) -- that's just me thinking out loud about destination choice. I like the idea, but just need to be clear on which destinations.

Thanks all for the good discussion :)

[xavidelclos]

Great summary Carl!

One minor thing: regarding the bicycle_parking=stands tag, I believe there might be other relevant types of bicycle_parking available. At least in Barcelona there's others like 'anchors', 'buildings',etc., and others without a sub-tag.

[carlhiggs]

Cycling indicator working group meeting - 16 May 2023

Attendees: Afshin, Ana, Javi, Marc, Tayebeh, Xavi, Carl (minutes)

@global-healthy-liveable-cities/cycling-indicators-working-group

Proposal

Afshin and Tayebeh are both commencing research projects, respectively related to low stress cycling infrastructure and cycling accessibility to points of itnerest. Afshins work will involve modelling infrastructure using OpenStreetMap data, and Tayebeh mentions that her work will also consider aspects including distances to destinations, elevation and greenness.

It was proposed that we could collectively use 3 different cities -- Barcelona, Melbourne and Munich -- for case studies informing

• development of indicators using open data
• validation of indicators using travel surveys and/or other data

This selection of cities is based on familiarity and opportunity (a number of us are based in Barcelona and Melbourne, and so are most familiar with these cities; while we have colleagues in Munich, and Afshin plans to visit there).

We all agreed that there could be mutual interest in this approach (ie. in addition to the goals for Afshin and Tayebeh's proejcts, also inform development of indicators for the Global Healthy and Sustainable City Indicators Collaboration / Global Observatory of Healthy and Sustainable Cities).

One thought is, these are all cities in high income countries (Australia, Spain, Germany). We also have colleagues who may be interested in including Mexico City, which could offer a different kind of urban environment (climate/topography/cultural/urban context), for a middle income Latin American country. If you would like me to raise this proposal with colleagues let me know. Or we could start with the existing cities listed.

Literature

We discussed exploring the literature, and Javi kindly shared two references before and during the meeting:

• Porter, A. K., Kohl III, H. W., Pérez, A., Reininger, B., Pettee Gabriel, K., & Salvo, D. (2020). Bikeability: assessing the objectively measured environment in relation to recreation and transportation bicycling. Environment and Behavior, 52(8), 861-894.
• Molina-García, J., Campos, S., García-Massó, X., Herrador-Colmenero, M., Gálvez-Fernández, P., Molina-Soberanes, D., ... & Chillón, P. (2020). Different neighborhood walkability indexes for active commuting to school are necessary for urban and rural children and adolescents. International journal of behavioral nutrition and physical activity, 17, 1-11.

intersections and node degree

The second paper above related to a discussion that arose during the meeting around indicators relating to intersections. As Javi explains,

In relation to the number of four or more street intersections, we usually use it in studies on active commuting to school as an indicator of street connectivity ... Consecuently, it's considered as a positive attribute of the built environment. However, there are studies that use it as a negative indicator because it could be associated with a higher volume of traffic, air pollution or more dangerous crossings.

This is an interesting point to consider, as currently our software evaluates street connectivity using cleaned intersections intended to represent 3+ way intersections (currently we 'consolidate' intersections using a function in Geoff Boeing's OSMnx python package). This approach draws on the indicators used by Frank, Sallis et al. (and related IPEN study work) that described intersection density in this way:

Intersection density measured the connectivity of the street network, represented by the ratio between the number of true intersections (three or more legs) to the land area of the block group in acres. A higher density of intersections corresponds with a more direct path between destinations.
Frank, L.D. et al. (2010) ‘The development of a walkability index: application to the Neighborhood Quality of Life Study’, British Journal of Sports Medicine, 44(13), pp. 924–933. Available at: https://doi.org/10.1136/bjsm.2009.058701.

If it is considered useful we could look into seeing if we could also measure more detailed aspects relating to intersections (eg '3+ way intersections per sqkm' [which is what we do currently], '4+ way intersections per sqkm', or even just a layer of cleaned intersections that could be visualised, coloured by node degree). Would something like that be useful?

Its a good point that different indicators are useful for different purposes, and that while intersections may offer connectivity, as we discussed, increasing degree of connectivity also brings complexity and consequent risk of traffic incidents for cyclists. That's a nuance that our current indicators/measures don't really explore.

Sharing documents

It was agreed that we would create and circulate some shared documents for collaboratively recording,

• citations in the literature (e.g. a Word or Docs document; Carl will create and share)
• data/measures/indicators (e.g. an Excel or Sheets workbook; Tayebeh agreed to create and share), also including sheets for
  • destination typology (what are relevant destinations, and synonyms for identifying these using OpenStreetMap?)
    • different kinds of destinations may be relevant for different kinds of trip purposes
  • potential available validation data case study cities, with description of data and alignment with trip purposes
• a conceptual map (already shared!)

Conceptual map

We trialled out using Miro to draft an initial Cycling indicator conceptual map. I think this was a useful exercise, which we can refine, and may be useful to inform some of the possible measures/indicators listed in our shared spreadsheet document. I think this will be useful for teasing out basic informative measures (eg elevation) that we could implement and use to derive additional measures and indicators (eg. relating to slope, or slope exceeding some threshold that indicates something meaningful, or area-level / route-based variation in slope which could be relevant as a consideration for bikeability), but also considerations with the data used to estimate / represent these aspects (eg resolution of terrain model for elevation, whether this is a bare earth model, and other considerations).

Zoomed out, it looks a bit like this, currently:

We can iterate on this to simplify the model, and find ways of communicating the concepts contained within it. For now, if you want to see it, best to log in using the link I e-mailed earlier!

[carlhiggs]

Cycling indicator working group meeting - 21 June 2023

Attendees: Ana, Javi, Marc, Tayebeh, Xavi, Carl (minutes)

@global-healthy-liveable-cities/cycling-indicators-working-group

Software update

Carl briefly shared an update of the latest version of the GHSCI software: You can now use custom population data and boundaries (eg for demographic sub-group specific indicators, using vector or raster data), the possibility of using your own intersection data (i.e. as an alternative to cleaning intersections with OSMnx, and possibly as a pre-cursor to custom density indicators with user-supplied POI data), a graphical user interface and more detailed example Jupyter notebook. While these aren't specifically cycling indicator related things, I hope they will be moving towards capabilities that can support some of our intentions.

Indicators spreadsheet

Tayebeh shared the spreadsheet she developed consolidating the themes/concepts/indicators/measures/relations from our earlier mindmapping exercise (thanks @Tsaghapour !).

We added a worksheet for Literature that we can add to - and that's an action for all of us

• add to cycling-related indicator literature list

This included an alignment of indicators with cities (Melbourne, Barcelona, Munich) and a rating that our group members had completed for the priority of these indicators for the different settings. Indicators prioritised across all settings should present a priority for implementation.

Carl suggested that we should consider adding in Mexico City, pending contacting our collaborators working there; following up and seeking priorities for Mexico City will be an action to do for this month.

• follow up with collaborators regarding cycling indicators for Mexico City (Carl)

Another action is providing a summary for each indicator about the level of existing implementation or feasibility for implementation within the GHSCI software.

• align indicators with level of implementation and feasibility of this in the GHSCI software (Carl)

Ana suggested including more greenery-related indicators. Carl had a question about what specific form these should take as measures --- for example, for the existing NDVI indicator, should this be point estimates that are aggregated to various scales (this is kind of like our current indicators for access to a destination within 500m), or is the point estimate in itself about a green cyclable catchment with aggregation to larger scales like a moving window average (kind of like our local walkable neighbourhood statistics for population density and dwelling density)? Tayebeh suggested that route based indicators would be particularly helpful. Carl found this particularly interesting as its a scale of aggregation we haven't currently implemented, but could implement in our software -- but requires some thought on how/what distance the network should be split at regular intervals (if that isn't done, the segments will be variable length with longer segments having less accurate/more diffuse estimates than shorter segments with more focal estimates). Notwithstanding the details, I think this is very do-able to implement in the GHSCI software and will look into it.

• look into aggregating indicators for route segments

Marc suggested using additional variables from the Global Human Settlements Layer Urban Centres Database for summarising the local geography and weather patterns. While this data (published in 2019, targetting 2015) is historical, notwithstanding climate change and related variation, it should be fair to assume the weather and geography related variables provide a reasonable representation of the current context, moreso than other urban attributes that could have changed in the past decade. Good idea! And this should be fairly straightforward to incorporate these into the GHSCI software, as its already set up to pull a custom set of covariates from the GHSL UCDB where its available. I'll have a go at doing this.

• scope and incorporate GHSL UCDB geography/weather variables into the GHSCI software

Marc also reminded us of the comment regarding elevation data earlier in this thread. This provided a link to software that visualises GPX data, like a cyclist might record using a device. This data includes elevation along with coordinates, and I believe that is how the elevation is sourced for cycling routes --- so its more about mapping/visualising data that is self-recorded. There are sources we could use for elevation data --- e.g. USGS provides some that can be downloaded (30m resolution, for different epochs earlier than 2013), there's an interesting data catalog at https://portal.opentopography.org/dataCatalog that could be worth exploring, and there are some APIs -- and this one also seems to provide instructions for running it offline with a guide for installing various datasets: https://www.opentopodata.org/server/#adding-datasets . Could be worth exploring.

We also briefly discussed the challenges of estimating traffic volume: number of lanes of traffic is an aspect, but absolutely not the full picture; congestion data probably is the full picture (at least for a time point, or average across some window of space time), but isn't necessarily easy to get as open data. Any thoughts are welcome!

Low-hanging fruit indicators for the GHSCI software

Separately Carl and Xavi had chatted about the possibility of implementing some basic 'cyclability' indicator, along the lines of that discussed in our earliest meeting -- this would be like our walkability indicator, but with sub-indicators adapted for relevance to cycling. There are some technical challenges involved in implementing this that could help with some of the more ambitious work, so tangential to the above is a plan to work towards this.

[carlhiggs]

Cycling indicator working group meeting - 18 July 2023

Attendees: Ana, Javi, Marc, Tayebeh, Xavi, Carl (minutes)

@global-healthy-liveable-cities/cycling-indicators-working-group

Software update

Carl mentioned the updated GHSCI software documentation website and videos:
https://global-healthy-liveable-cities.github.io/software/
https://www.youtube.com/playlist?list=PLOllH2OXS1IYpZj6TXlYfdmAcuGUln1Jh

Earlier in the week I sent an e-mail re an article on the tool BikeDNA designed to support network quality assessment:

Vierø, A. R., Vybornova, A., & Szell, M. (2023). BikeDNA: A tool for bicycle infrastructure data and network assessment. Environment and Planning B: Urban Analytics and City Science, 0(0). https://doi.org/10.1177/23998083231184471

There are two associated repositories; the second is optimised for large scale anlaysis:
https://github.com/anerv/BikeDNA
https://github.com/anerv/BikeDNA_BIG

I haven't looked too much into the tool, but it clearly has functionality of interest to this group. Its focus on quality assessment could be of use for users of our tool. It also appears to have a function for converting a GeoPandas GeoDataFrame of linestrings into an OSMNx graph. This is something I've been wanting to do but too busy to develop. We could use that to support people to optionally BYO network, or modify an existing network (even one produced with our tool) to represent a hypothetical intervention etc, and then use our compare function to evaluate the change in indicators resulting from that.

I haven't trialled this yet, but given the similar software ecosystem I expect it should install without too many issues if anyone wanted to experiment. It can be installed as a python module for our GHSCI software environment by entering the following at command line:

pip install -e git+https://github.com/anerv/BikeDNA.git#egg=src

Destinations discussion

Afshin and Tayabeh introduced our colleague Steve Pemberton who they have engaged for assistance with the bikeability-focused work they are conducting, and in association with this group. Steve, Afshin and Tayebeh have been considering how to model destination features of interest for their accessibility analyses, with a view to creating route-based indicators based on analysis of randomly selected origin points.

The main question was, what kind of destinations/categories? Steve did some preliminary investigation of a range of OpenStreetMap tags, and shared this with us. I mentioned that we had done some similar work to this (conceptually mapping high level categories to the series of key:value pair tag synonyms that might be used in practice for identifying features like these).

For the [2018 Australian National Liveability Study (ANLS)](https://doi.org/10.1038/s41597-023-02013-5], following a review of OpenStreetMap tagging guidelines, OSM TagInfo and our own spot checks for Australian cities, we developed this CSV file set of terms aligned with boolean operators (or, and, not) to guide how they would be combined.

We (myself and Julianna Rozek, who had a particular interest in public open space) developed a pretty elaborate method for identifying 'areas of open space' for the ANLS, as one of our tasks was to develop indicators for early child development for which measuring aspects related to schools was particularly relevant. So --- schools and public open spaces of various kinds (including those with sporting facilities --- were all considered areas of interest (in constrast to Points of Interest, but both being features of interest). To evaluate access to areas using OD analyses, we generated pseudo-entry points every 20m around linestring representations of the polygon, retaining those points within 30m of the pedestrian network. This was done to reduce the processing load, but also capture where areas of public open space could plausibly by accessed (eg not via park segments not near a road, which might be assumed to be fenced off. The pseudo-entry point method was used in earlier liveability work, following methods of our colleagues Koohsari, Mavoa et al. (2015), although restriction to points near accessible network segments was novel (at least to us) to our ANLS work.

Similar methods have been used adopted in the GHSCI software. To date we have used a shorter list of destination categories](https://github.com/global-healthy-liveable-cities/global-indicators/blob/main/process/configuration/templates/osm_destination_definitions.csv), with some modifications to tags used advised by international collaborators from our 25 city study; in addition we allowed for custom import of destinations by CSV (the implementation of this could be refined). The means of configuring the criteria for Areas of public Open Space has been refined, but in effect is similarly complex. I acknowledge, its a bit of an obscure typology but its one we've found (in our 25 city global study global study and Australian work seems to work adequately as a representation for analysis. While reapproaching it I was reminded of the Borges short story about the cosmic classification schema found in a Chinese encyclopedia (summarised in Spanish and English here) --- much writing around about that, but there are some parallels with our discussion around destinations, categories and classifications that I think this quote captures well:

"categories can map to practical real-world cases or they can be neat and logical, but rarely both at once." (quoted in the article from another article by Tim Harford)

Regarding categories, there was some agreement that more granular classifications could be pragmatic for comparative use across different contexts. Irene Gomez Varo w/ Xavi (https://doi.org/10.1016/j.cities.2022.103565) used various categories --- commercial/retail (w/ sub-classification of basic / non-basic), public facilities (like social infrastructure; so, could encompass aged care facilities and other things). We discussed 'Health' being a broad important category. There are some parallels here with things we did in ANLS --- e.g. a 'local living' index, as a more comprehensive 11-destination version of the daily living index index (supermarket, convenience, PT); we also created some specialty food destination indicators and social infrastructure mix. However the latter used some specialty data sets for identifying childcare, schools, health services/aged care facilities and community centres (although the latter were partially from OpenStreetMap). Xavier suggested that the mandatory/non-mandatory distinction could be problematic/not helpful, as may not be relevant as daily destinations (we also discussed changing patterns of work and consideration of what constitutes work that problematise this typology). We discussed challenges of sport destinations --- ie. watching/entertainment (stadiums) vs participating.

We've been dealing with this same question about a meaningful typology of destinations in our Contextual Walkability project too (where we are looking at measuring walkable neighbourhoods in diverse settings in a local context rich way, using imagery classification in addition to openstreetmap/custom data). In that project, one of the sub-scales was a 'function diversity' index -- perhaps similar conceptually to the one Irene and Xavi calculated in the JANE Index, but with different destinations (specific to Harbin, China). To help us generalise this to other cities, and informed by earlier work by our group, i proposed a series of categories ... which I'll e-mail about. There would be sub-categories to these (e.g. some might like to differentiate fast food chains in 'eating out'; different kinds of public transport; kinds of public open space etc), but these are broad concepts that I think could be conceptually mapped to the kinds of destination features of interest in most places.

Our Contextual walkability group is proposing a sub-scale based on these that we will implement in software and we'll look to validate it (eg using our High Life study colleagues); I'm sure you could participate in this research if interested in applying/validating it elsewhere.

We also talked a little about distance thresholds --- these are tricky, hard to find one that fits all. We mentioned that hard vs 'soft' thresholds can be useful at different times: if aggregating, binary indicators yield proportions that are easier to explain and converge equivalently; but otherwise, continuous scores contain richer exposure information (eg for individual linkage analysis). Xavi mentiond distance decay as another well used approach yielding score rather than binary.

Literature

Tayebeh requested we add any additional literature to shared spreadsheet in coming week; she then plans to summarise this, looking at what indicators were used and methods.

• Add literature!

[carlhiggs]

Hi @global-healthy-liveable-cities/cycling-indicators-working-group -- I think I've missed the deadline for adding new literature, but I just came across the following recently published paper that those interested in cycling and gender equity and arguments for dedicated cycling infrastructure might be interested in:
Battiston, A., Napoli, L., Bajardi, P. et al. Revealing the determinants of gender inequality in urban cycling with large-scale data. EPJ Data Sci. 12, 9 (2023). https://doi.org/10.1140/epjds/s13688-023-00385-7

[Tsaghapour]

RMIT Classification: Trusted Hi Carl, Thank you for sharing the paper, it's really useful! And don't worry about timing - I'm still in the early stages of compiling the literature due to my involvement in other projects. Your contribution is right on time:)! Kind regards, Tayebeh Saghapour Research Fellow, PhD Healthy Liveable Cities Lab Centre for Urban Research RMIT University, City Campus Building 8, Level 11 Melbourne, Victoria 3001 Australia From: Carl Higgs ***@***.***> Sent: Thursday, 3 August 2023 1:19 PM To: global-healthy-liveable-cities/global-indicators ***@***.***> Cc: Tayebeh Saghapour ***@***.***>; Mention ***@***.***> Subject: Re: [global-healthy-liveable-cities/global-indicators] Cycling infrastructure and bikeability/cyclability indicators (Discussion #226) Hi @global-healthy-liveable-cities/cycling-indicators-working-group<https://github.com/orgs/global-healthy-liveable-cities/teams/cycling-indicators-working-group> -- I think I've missed the deadline for adding new literature, but I just came across the following recently published paper that those interested in cycling and gender equity and arguments for dedicated cycling infrastructure might be interested in: Battiston, A., Napoli, L., Bajardi, P. et al. Revealing the determinants of gender inequality in urban cycling with large-scale data. EPJ Data Sci. 12, 9 (2023). https://doi.org/10.1140/epjds/s13688-023-00385-7 - Reply to this email directly, view it on GitHub<https://github.com/global-healthy-liveable-cities/global-indicators/discussions/226#discussioncomment-6623054>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AVLYZNTKHBLDV7DMIZYNIVLXTMKDZANCNFSM6AAAAAAWBX4STY>. You are receiving this because you were mentioned.Message ID: ***@***.******@***.***>>

[carlhiggs]

Cycling indicator working group meeting - Australia and Americas - 22 August 2023

Attendees: Shirley, Cesar, Julio, Giovani, Cristina, Ricardo, Henrique, Afshin, Steve, Carl

@global-healthy-liveable-cities/cycling-indicators-working-group

We had our first cycling group catch up in an Americas-friendly timezone, motivated to hear from Shirley about her research as part of the Accessibility Observatory team at University of Minessota.

Introductions

• Shirley's research is focused on accessibility, most recently, bike access and level of traffic stress in US cities; Shirley is also an executive member of the Global Healthy and Sustainable City Indicators Collaboration / Global Observatory of Healthy and Sustainable Cities
• Afshin's background is in simulation and travel behaviour modelling and built environment indicators and accessibility are considerations for inputs to these models. He is collaborating with Tayebeh Saghapour (who was unable to make the meeting today), who also has a bike-ability focus in her research.
• Steve has a background in corporate law, and an interest in urban planning and spatial/network analyses; he is assisting Tayebeh and Afshin in their research, related to cycling accessibility and modelling
• Cesar works for the national institute of health in Mexico, amd will be conducting research on Mexican cities for the 1000 Cities Challenge as part of his postdoc; he has a built environment and health epidemiology-related background, with a dissertation related to access to parks and mental health/quality of life
• Julio and students Giovani, Cristina, Ricardo and Henrique are based in Porto Alegre and share an interest in modelling cycling accessibility and urban characteristics for active lifestyles
• Carl (me), doesn't have particular expertise in modelling cycling accessibility (although I have conducted multi-modal analyses using r5r and other related tools); I am interested in how we can expand the software we are develop to support the 1000 City Challenge to include cycling-related indicators (and other indicators)

Level of traffic stress

Shirley and colleagues' work focused on US cities

Shirley shared work she has been doing in conjunction with her colleagues, in particular based on an article:

Murphy, B. and Owen, A. (2019) ‘Implementing Low-Stress Bicycle Routing in National Accessibility Evaluation’, Transportation Research Record, 2673(5), pp. 240–249. Available at: https://doi.org/10.1177/0361198119837179.

This article contains a detailed set of classification rules for deriving level of traffic stress (graded 1 for separated bike lanes, low stress; to 4, for no bike lane on a busy street). In reverse order, this represents a gradient of increasing 'comfort, safety and interest in bicycling for transport' (quote from Shirley's presentation).

Shirley has been implementing a python-based workflow using PostgreSQL with PostGIS to derive LTS metrics from OpenStreetMap data according to the above classification schema.

Aside: I did a quick search for level of traffic stress approaches on github to see if this had been published (perhaps as 'BicycleLTSAccessibility'?); I am not sure but found this other 'bike-lts' tool that apparently presents story maps designed for Delaware in the US a couple of years ago - might be of interest to some:

• https://github.com/dvrpc/bike-lts

Shirley is now using a python implementation of R5 to evaluate travel times for low stress cycling scenarios - e.g. distribution of low or high LTS streets, accessibility using only low stress routes, socio-demographic sub-analyses to consider equity dimensions, and evaluating impact of network interventions.

Afshin, Tayebeh and Steve's work

Afshin was aware of a UK team using a 0 to 1 grading schema for LTS, however his work with Tayebeh and Steve I believe is based on similar 1-4 US guidelines. While Afshin's approach uses simulation models to derive projections for average traffic volumes, this relies on traffic data that might be broadly/consistently available globally; for scaling, he recommended the approach of using OpenStreetMap road hierarchies.

Afshin and Steve shared their approach that aims to account for 'deviation percentage'. This isn't a concept I'm familiar with -- but sounds like something similar may be described in this paper:

Wang, H., Moylan, E. and Levinson, D.M. (2022) ‘Prediction of the Deviation between Alternative Routes and Actual Trajectories for Bicyclists’, Findings [Preprint]. Available at: https://doi.org/10.32866/001c.35701.
ie. quantifying the disjunct between an ideal route, and how far out of the way a cyclist might have to go to achieve a practical route that accounts for their comfort/level of stress needs. (I may be misunderstanding; but sounds interesting!). Steve described how their work adds progressive penalties for increasing traffic stress, ultimately allowing deviation of up to 15% to find a better or shorter route.

Afshin also discussed the disjunct between government interest in modelling, which might favour LTS1+2, and advocacy groups that favour minimising distance; there is an open question, what is an acceptable degree of deviation for low stress and comfort? (I may have distorted/missed the points here a little, as this is new for me!)

Julio and students' work in Brazil

The Porto Alegre team have been using r5r (developed in by the Institute for Applied Economic Research in Brazil, including Kaue Braga as a co-author who I think Julio mentioned; I have met Kaue before through his work developing a healthy cities tool for C40), and other tools for modelling routing and accessibility, for walking and more recently cycling. Julio mentioned his background was particularly in using choice, logit and multinomial models to examine the factors effecting walking behaviour.

The group have been collecting empirical behaviour data to better understand how traffic stress and other urban characteristics influence travel behaviours in the context of cities in under-developed countries. For example, it may be that other aspects related to personal safety, including avoidance of urban violence, may be prioritised over traffic stress. Julio shared a link to a pre-print article looking at how inequities in access to street markets intersect with those related to race and income in the context of Porto Alegre:

Borges, D. et al. (2023) ‘Social and ethnic-racial inequities of physical accessibility to street markets in Porto Alegre/RS, Brazil’. SciELO Preprints. Available at: https://doi.org/10.1590/SciELOPreprints.6326.

In general, great to know about this collection of behavioural data as it will be important to consider construct validity of new indicators we construct, in terms of their associations with/capacity to predict behavioural outcomes --- particularly if there is longitudinal data.

Capturing traffic stress and congestion using other data sources

We briefly discussed other approaches for characterising traffic stress and congestion for use in route modelling. Afshin mentioned that truck traffic in particular is important to capture, but challenging. One approach is using traffic sensors that capture passage of vehicles --- it is theoretically possible to derive vehicle types by considering road speed adn distance between first and second tyres to estimate length of vehicle; cities with smart traffic signals have these kinds of sensors, and the data may be available. While traffic lights were raised as a possibility, it was also pointed out that these are too sparse. Julio instead recommended aggregate mobile phone tower pinging aggregate count data -- although this may need to be purchased from telephone companies. In the US context, Shirley and colleagues use commercial TomTom datasets to capture congestion estimates.

A distinct approach to urban characterisation, raised by Julio, was semantic segmentation through deep learning using street view imagery. Julio mentioned that Ana Luiza Favarão Leão, a PhD student studying at Washington University at St Luis (who I have also met through her participation in our 1000 Cities Challenge software and reporting feedback sessions) is using this approach in her research. I located this article:
Leão, A.L.F. et al. (2020) ‘Identificação de Composições da Paisagem Urbana: Uma abordagem de Deep Learning’, Revista de Morfologia Urbana, 8(1), pp. e00140–e00140. Available at: https://doi.org/10.47235/rmu.v8i1.140.
which has an English PDF download for those interested. This is of interest to myself and colleagues (including Yang Ye, a researcher from Harbin, China, who has done similar work there, still under reivew I believe), as we have been exploring similar approaches as an alternate way of contextualising neighbourhood walkability for diverse global contexts, as a supplement to more traditional walkability indices (another possible indicator extension for the 1000 Cities Challenge). I'll have to get in touch with Ana, as it seems this is a shared interest! Finally, both Afshin and Julio mentioned footpath.ai, who I also was not aware of -- a start up with global ambitions based in Sydney associated with UNSW and AWS using deep learning from street view imagery, that I believe they collect. Good to know!

[shiqin-liu]

Thanks Carl, this is great summary! The 'BicycleLTSAccessibility' library is hosted on our enterprise github account so that's not public, there is a public version AOBikeLTS, it has not been updated as it should, but some of the logistics are there. We are currently working on upgrading that tool so if anything interesting comes out I can share in the future.

The LTS for cycling certainly worth a continuing discussion, and with OSM roadway attributes, the approach can be scalable both locally and globally. It is great to learn that different metrics are used in other countries, especially for global south, the context for biking and stress levels can be very different from developed countries. This is both valuable and challenging aspect to the global-indicators framework, more research can/should be done there!