Network Assumptions

Description of the Network

The road network used for modelling draws on local sources where possible, or OpenStreetMap where local data is not obtainable. For Queensland, Australia the network has been built from the State Digital Road Network (SDRN). TransPosition uses a fully automated process for constructing the road network. This has a number of advantages:

  • It is clear where data comes from – any changes to the base network are clearly recorded in their own map layers
  • All assumptions are explicit, and can easily be changed
  • It is possible to update the network as new base road data becomes available (the SDRN is updated quarterly)
  • It is easy to change the coverage or level of detail in the model

The SDRN contains street centrelines for every road in Queensland, and has a good coverage for off-road cycle and walking paths. It has limited data on each road so a number of assumptions must be used to construct the road network. These assumptions are outlined below.

Basic Network Assumptions

The attribute information available for the state-wide road network was fairly limited, so a number of assumptions had to be made to prepare a network suitable for modelling. Except when overridden, all roads are given assumed free-flow speeds and number of lanes based on their hierarchy and urban/rural status. The following table shows the basic assumptions.

Assumptions on roads based on hierarchy and urban/rural status
Hierarchy Road Type Speed Lanes Friction
1 Freeway/Motorway 80 3 Low
2 Highway 70 2 Medium
3 Secondary Road/Arterial 60 2 Medium
4 Local connector road 55 1 High
5 Local street 50 1 High

Note that the number of lanes is given for a single direction – an entry of 1 in the table means a standard 2 lane road. The table works with lanes in a single direction to make the coding of one-way roads more consistent. The measure that is here called friction is sometimes known as traffic impedance, and is an overall measure of the degree to which traffic is constrained by side friction, road design standards etc.

These speeds reflect posted mid-block speeds; the delays at intersections are calculated separately.

Updating Network Information

The basic assumptions described above do not always apply, so the network information in the model was extensively reviewed, based on physical inspection and detailed examination in Google Earth, with Globe for the particular state of interest overlayed, and StreetView. The review aimed to identify

  • Changes in the number of lanes
  • Changes in posted speed
  • Location and type of intersections
  • Location and type of railway crossings and signalized pedestrian crossings

These changes were coded as point based link transitions, which are a feature of the TransPosition modelling suite. They allow changes to network data to be specified with a point and a bearing. The points are snapped to the closest road that matches the bearing at that point, and then modify the data on that road in the direction specified by the bearing. This process makes it easy to see where basic network data in the SDRN has been overridden. It also means that the changes that have been made can be automatically applied to any updated version of the SDRN. As an example in Queensland, around 8,400 link transition overrides have been made to the network, and around 2,500 intersections or crossings have been coded.

Capacities

Once the number of lanes and degree of side friction are determined, a separate table is used to assign hourly one-way capacities, and the Davidson J congestion parameter. The capacity of a single lane is dependent on the hierarchy, side friction, number of lanes, and whether or not the road is median divided.

Lane capacities and the Davidson J congestion parameter
Hierarchy IsDivided Friction LaneCap1 LaneCap2 LaneCap3+ Speed Factor J
1 0 Low 1600 1700 1700 0.95 0.20
1 0 Medium 1500 1600 1600 0.90 0.20
1 0 High 1400 1500 1500 0.80 0.20
1 1 Low 1800 2100 2100 1.00 0.10
1 1 Medium 1700 2100 2100 0.95 0.10
1 1 High 1550 1700 1700 0.90 0.10
2 0 Low 1200 1300 1300 0.80 0.25
2 0 Medium 1100 1200 1080 0.64 0.30
2 0 High 850 900 840 0.45 0.35
2 1 Low 1250 1500 1500 0.80 0.20
2 1 Medium 1150 1250 1170 0.64 0.25
2 1 High 900 1000 980 0.45 0.30
3 0 Low 1200 1300 1200 0.73 0.35
3 0 Medium 1100 1200 1080 0.64 0.45
3 0 High 750 800 700 0.40 0.55
3 1 Low 1250 1400 1300 0.73 0.30
3 1 Medium 1150 1250 1170 0.64 0.40
3 1 High 850 900 840 0.40 0.50
4 0 Low 1000 1050 1050 0.73 0.90
4 0 Medium 850 900 900 0.64 1.00
4 0 High 850 900 840 0.40 1.00
4 1 Low 1100 1200 1200 0.73 0.80
4 1 Medium 900 950 950 0.64 0.90
4 1 High 900 1000 980 0.40 1.20
5 0 Low 600 600 600 0.70 1.30
5 0 Medium 600 600 600 0.60 1.30
5 0 High 600 600 600 0.40 1.30
5 1 Low 600 600 600 0.70 1.30
5 1 Medium 600 600 600 0.60 1.30
5 1 High 600 600 600 0.40 1.30

The Davidson-Akçelik speed flow relationship is used

t=t_0 [1+0.25 r_f (z+\sqrt{z^2+\frac{8J(z+1)}{r_f}})]

where,

t = congested travel time

t_0 = free flow travel time = length/uncongested speed

z = (volume/capacity) -1

r_f = ratio of flow period to minimum travel time = \frac{T_f}{t_0}

T_f = Duration of congestion – the model has assumed that T_f=1hr

Public Transport and Active Transport

The representation of the public transport network in the model draws also on local sources where possible. For Queensland the public transport network is built automatically from data provided by TransLink (part of the Queensland Department of Transport and Main Roads). All scheduled regular and school services are included, with full time tables used by the model. The Monte Carlo sampling used by the model includes a randomly selected arrival time (from the observed distribution for each travel purpose). When considering modes, the model will determine the best route and individual PT service to use. This means that the impact of travel time is implicitly incorporated – infrequent services will lead to long waiting times. To allow for the fact that people can plan their travel around Public Transport timetables, this first-stop waiting time is discounted. A random distribution is used for this to allow for the fact that people will have variation in their ability to reschedule their travel. The model also includes variation in people’s mode preferences (to account for the fact that some people are happy to use public transport and others are committed to cars) and weightings for different sub-modes (to reflect the general preference for ferries and trains over buses). The share of travel by each mode is an output of the model, and is influenced by the relative times and cost of travel by each mode, taking people’s mode preferences into account.

All major walking and cycling routes in the city are included in the model. The model also assumes that people can walk and cycle along all roads except for Freeway/motorways. Different weights are used for off-road vs. on-road walking and cycling. The model allows for variations due to on-road cycling lanes but no data on this has been coded. As well as incorporating variation in people’s preference for walking and cycling (including variation in end-of-trip times) the model includes random variation in walking and cycling speeds. There are also factors for specific travel purposes – for example people are less likely to walk for shopping trips.