Pedestrian Demand Forecasting

This tool provides an implementation of the three pedestrian forecasting procedures described in the TMR Pedestrian demand forecasting guideline.

The guidance describes three forecasting procedures:

Comparison method: use counts for other comparable sites to estimate demand at the project site. This method is implemented here on the Database tab where a larger number of pedestrian counts in Queensland can be filtered based on location and facility type.
Factoring: apply an uplift factor to pre-construction counts obtained at or near the project site to account for likely additional walking activity that will be attracted from other routes, other modes or are all-new (induced) trips. This procedure is implemented on the Factoring tab.
Direct demand: regression model based on the pedestrian counts database linking pedestrian demand to land use (e.g. population, employment, schools) and network (e.g. signalised intersection, shared path) attributes. This procedure is implemented on the Direct demand tab.

Each of these procedures are subject to large uncertainties. It is recommended that practitioners use all three procedures to provide a range of forecasts.

This procedure applies expansion factors to observed demand prior to construction to estimate demand after construction.

Step 1: Observed demand

Demand should be observed from counts obtained at or near the site over at least three days.

Weekday:
Weekend:

Step 2: Purpose split

How much walking activity at the location is for recreational activities (e.g. exercise, walking the dog) as opposed to transport activities (e.g. walking to work, shops, to visit friends or relatives)? Enter 100% if all activity is for recreation, and 0% if all activity is for transport.

Weekday:
Weekend:

Step 3: Pre-existing walking

How much walking activity after construction would have occurred irrespective of the project? For example, a very modest project is unlikely to encourage much new walking so should have a value close to 100% (i.e. all walking activity was occurring anyway). A project that has a large impact on walking level of service (e.g. a crossing of a major arterial road where none previously existed, or a new footpath where none existed) may have pre-existing trip proportions closer to 60% or less (i.e. 40% of walking trips are new).

Recreation:
Transport:

Results

Drop site markers onto map using the

Draw marker button

Show 2 km buffer around sites

Download results

Download

Why don’t the three procedures give the same answer?

There is very large uncertainty in forecasting pedestrian demand. The three procedures in this tool provide different ways of approaching the problem and make very different assumptions. It is highly unlikely they will give the same forecasts, and in many cases will give wildly different forecasts. The practitioner is advised to use the three procedures together as a means of providing a range within which the true demand may fall. Moreover, the practitioner should apply their knowledge of the local context to make a judgement as to where within this range the most likely estimate may reside. Indeed, there will be situations where the practitioner can make the case that the most likely estimate falls outside this range altogether.

These forecasts are guidance to support and not replace expert judgement.

Why does’t the direct demand estimate match the database estimate?

The direct demand model is a regression fit of the pedestrian counts in the database. As with almost all regressions the model will not perfectly fit the data. This is particularly true in this case where there are many predictors of pedestrian demand (not all of which are included in the model, and those which are use imprecise data sources) and the demand counts themselves are often of modest quality (that is, only over one day).

There will be instances where the direct demand model predicts demand that is very different from the observed count. This reflects the difficulty of correlating noisy data (demand) to an unknown range of predictors of demand (e.g. population, employment, walking infrastructure types, proximity to trip attractors). An analysis of the model errors (residuals) is provided here.

Where can I learn more about the modelling approach?

This set of procedures were developed under contract for Queensland Department of Transport and Main Roads. Several background papers and technical notes were prepared as part of this work:

Research report: background review of current pedestrian demand forecasting procedures.
STREAMS validation: estimation of expansion factors to apply to STREAMS signal pedestrian activations
Direct demand models: estimation process for direct demand regression models.

Pedestrian counts

Counts were obtained from the Department of Transport and Main Roads and local authorities. The counts were obtained between 2009 and 2020 across Queensland and vary from single day counts to permanent automatic counts.

Counts from multiple days were averaged by weekday and weekend days. As data was only available from many sites between 6 am and 6 pm all daily counts are reported only for this 12-hour period.

Weekday peak hours were calculated using a rolling 15-minute total. Where a site had multiple days of weekday data and the peak hour varied across days the average starting time is reported.

Counts at intersections are reported as the total pedestrian count across all intersection arms including slip lanes (where present). While this accurately reflects the pedestrian crossing events at the intersection the unique number of persons using the intersection will be lower given that some pedestrians will use multiple intersection legs. Counts by leg are provided in the popup window for individual sites on the map.

Land use data

Population

Population data is from the 2016 Census of Population and Housing using mesh block (MB) geographic areas. MBs typically contain between 30 and 60 dwellings.

Employment

Counts of employed persons (both fulltime and part-time) were obtained from the 2016 Census of Population and Housing for destination zones (DZ) using origin-destination trip tables.

Schools

Full-time equivalent student enrolments for public schools (both primary and secondary) were obtained from the Queensland Department of Education based on day 8 of the 2020 school year. Equivalent data for private schools was available only for the last day of February 2019.

Deprivation was estimated at the SA1 geographic level using the Index of Relative Socio-economic Disadvantage (IRSD) calculated by the Australian Bureau of Statistics from 2016 Census of Population and Housing data.

Residential, commercial and industrial land uses

Areas of residential, commercial and industrial land use were estimated from the Queensland Land Use Mapping Program (QLUMP) last updated in June 2019. It should be noted that the land may not necessarily be used for the purpose identified in QLUMP; construction may not have commenced or the land use may have been rezoned by the responsible local authority.

Waterways and coastlines

Inland waterways such as lakes and rivers were obtained from the Queensland Land Use Mapping Program (QLUMP) last updated in June 2019. The Queensland coastline was obtained from OpenStreetMap (downloaded July 2020).

Parks

Park locations were obtained from OpenStreetMap (downloaded July 2020).

Hospitals

Hospital locations were obtained from OpenStreetMap (downloaded July 2020).

Supermarkets

Supermarket locations were obtained from OpenStreetMap (downloaded July 2020).

Hotels and motels

Hotel and motel locations were obtained from OpenStreetMap (downloaded July 2020) using features tagged key = building and value = hotel or value = motel.

Transport networks

Method of travel to work

The method of travel to work was obtained from the 2016 Census of Population and Housing using Statistical Area 1 (SA1) as the origin (home) zone using place of enumeration. Walking mode shares are for walking as a sole mode only.

Signals

Signalised intersection locations were obtained from OpenStreetMap (downloaded July 2020) using features tagged key = highway and value = traffic_signals.

Pedestrian crossings

Three types of pedestrian crossings were obtained from OpenStreetMap (downloaded July 2020):

Crossing: feature = highway and value = crossing (these are mainly signalised pedestrian crossings but also include some zebra crossings)
Zebra crossing: feature = crossing and value = zebra
Pedestrian refuge: feature = crossing and value = island.

It is noted that the naming conventions are applied inconsistently in OpenStreetMap and that not all crossings are coded, this being especially true of pedestrian refuges (islands).

Public transport boardings

Train, bus and ferry boarding data was obtained from TransLink origin-destination data for 2019. The origin-destination matrix was aggregated to obtain annual boardings per stop and then divided by 365 to represent an average day.

Road network

Street centrelines were obtained from the State Digital Road Network. Comprehensive footpath data was only available for Gold Coast, Logan, Rockhampton and Ipswich City Council areas.

Pedestrian crashes

All severity crashes involving pedestrians across Queensland from 2014 to 2018, inclusive. The crash locations file from the Open Data Portal was used and severity was coded as the maximum severity of any involved person.

The procedures in this web tool were developed for the Queensland Department of Transport and Main Roads as part of work to develop pedestrian demand forecasting guidance. Feedback on the usability and accuracy of this tool are very welcome and should be sent to Walking@tmr.qld.gov.au.

Improving the accuracy of these forecasts is dependent on more and higher quality counts data. Jurisdictions with additional walking counts data are welcome to send data to Walking@tmr.qld.gov.au for incorporation as part of future tool updates.