Time to Replace PTAL with a better measure of PT Accessibility

PTAL was first developed by the London Borough of Hammersmith and Fulham and later adopted by Transport for London as the standard method for calculating public transport accessibility in London.  Through the London Plan it has become embedded in policy with density thresholds based on Public Transport accessibility.

PTAL was a breakthrough enabling for the first time scientific mapping of relative public transport accessibility.  I was once on the London PTAL working group and do not underestimate the sheer hard work that went into the development and the effort TfL has made in providing London wide maps.

The PTAL method however is showing its age and needs a replacement.  It simply adds up walking and waiting time to the public transport network.  It is therefore a measure of accessibility to the public transport network.  It does not take account of where the service is to.  Also reflecting the limited computational power of when it was developed it has arbitrary cut off points of walking distance so that PTAL levels can fall off from very high to none in a short distance.  More technically the PTAL index is calculated against a rectilinear grid of points which means that accessibility levels are distorted at a local scale as travel times are not equal in all compass directions. Finally TfL enormously simplified calculation by only calculating accessibility from population weighted centroids of super output area rather than every address, but that does average out levels across a district.  The bands 1-6 were chosen as a suitable range for H&F but the method has had to be hacked to add additional categories for central and outer London.

TfLs CAPITAL model shows travel times from a node of interest – such as a shopping centre, but does not show accessibility from home in a universal index.  Similarly the newer ATOS model shows access to the nearest schools, 10,000 jobs etc as a map but is essentially a measure of local rather than strategic accessibility.

Experience with GIS based accessibility modelling internationally has developed to the extent that a serious research effort is needed in the new London Plan to drive updated policies on the best areas for densification.

A particularly promising approach is offered by the Access Across America project by the Minnesota Dept of Transport/University.  This maps for all modes access to employment (number of jobs) within a certain travel time.  They have mapped all major US Metros.

The methodology is to take job numbers and population number from centroids of census blocks and to map travel times based on Google transit data.

to reflect the of transit service frequency on accessibility, travel times are calculated repeatedly for each origin-destination pair using each minute between 7:00 and 9:00 AM as the departure time, and an accessibility value is calculated using each travel time result. The accessibility results are averaged to represent the expected accessibility value that would be experienced by a traveler departing at a random time in this interval.

And then

accessibility is averaged across all blocks in a CBSA (census block), with each block’s contribution weighted by the number of workers in that block. The result is a single metric (for each travel time threshold) that represents the accessibility value experienced by an average worker in that CBSA.

In the UK we face more challenges in calculating employment locations, we no longer have a census of employment and travel to work data is taken on a 10% sample.  However various departments do have geocoadable employment data  which is available.

My suggestion is to base an index based

  1. on the number of jobs accessible within the 95th percentile commute time in England
  2. The boundaries between the zones be based on the natural breaks (Jenks) method rather than equal interval per category to produce a more uniform spacing of variations in the map more useful for policy purposes.
  3. The index is based off the population weighted centroids of output areas, however in areas where the number of commuters is too low to calculate a statistically meaningful result they should be weighted.
  4. The results need to be aerially interpolated across a uniform grid.  I have used a hexagonal grid – known as a planagon grid successfully on a number of projects developing a technique pioneered by the CSIR research institute in Pretoria.  On a hexagonal grid distances are equal in all compass directions to surrounding centroids.   A grid of 7 planagons to 1 sq km (about 212m accross) is sufficient.  Any smaller it becomes very data heavy.
  5. Extract non developed, roads and no parcel areas from the planagon polygons.
  6. For employment weight number of jobs in the output area by the building height in the planagon, the same for residential, using a linear regression technique pioneered by the Technology University of Riyadh to interpolate census data to buildings. Then add up the numbers for all buildings with centroids inside the plangon grid hexagon.
  7. Then calculate travel times using the OS ITN network and the TfL services database as per PTAL.  Rather than taking a equally weighted sample across the morning peak  the sample should be weighted by the likelihood of work related travel across the whole day.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s