Edinburgh City Centre : A Microsimulation Case Study

Introduction

This article is the third in a series on microsimulation and follows 'An Introduction to Microsimulation' and 'Some Real Applications of Microsimulation' published in Traffic Engineering and Control in late 1998. These suggested that microsimulation offers a unified approach to the analysis of road traffic because the technique models the components of traffic flow directly, and not by way of mathematical proxy. Examples of microsimulation applications were presented in outline, and demonstrated that the the technique is appropriate for modelling the full spectrum of road networks, ranging from individual junctions to wide area problems of rural and urban congestion, covering all conceivable loading and geometric circumstances. Here we concentrate on a microsimulation model for the Edinburgh City Centre, and discuss the rationale behind opting for wide area microsimulation, the issues associated with developing and applying the S-Paramics microsimulation modelling system, and the differences between this and a more conventional approach.


Background

The City of Edinburgh Council has adopted a proactive strategy towards transport accessibility within Edinburgh city centre. There is the usual current imbalance of demand and supply, where the 50% who travel by car take up 85% of available roadspace, while the 42% who travel by bus occupy only 3%. To redress this imbalance, Edinburgh has seen the implementation of a coordinated traffic restraint strategy that has included the implementation of widespread bus priority measures on arterial routes (known as Greenways) and the closure to general traffic in one direction of Princes Street, Edinburgh's principal shopping thoroughfare.

These schemes are also designed to make the city centre more attractive, safe and clean, by giving space back to pedestrians. Following the removal of eastbound general traffic from Princes Street in 1996, environmental improvements have been observed in terms of air quality and improved pedestrian and cycle facilities. This appears to have led to an increase in the number of pedestrians, to the consequent satisfaction of the business community.

As part of its "Moving Forward" strategy the City of Edinburgh Council wishes to continue its theme, and schemes currently under consideration include bus priority measures on Edinburgh's Royal Mile, a segregated guided busway from Edinburgh airport to the city centre, and the complete removal of cars from Princes Street, in addition to more radical proposals associated with the Transport White Paper. Scottish Devolution and the site of the new Parliament presents a new conundrum, where aspects of maximum access and minimum impact are to be reconciled. The combination of these proposals against the background of increasing congestion and pollution legislation requires the council to be exceptionally well informed about the consequences of its proposals, many of which are likely to be contentious at a local level. Transport Planning Consulants SIAS were appointed to develop the Edinburgh City Centre S-Paramics Model (ECCPM), with an objective to complement the council's existing models and assist the investigation of the traffic and pollution impacts of `Do-Nothing' and proposed traffic management measures. The model's visual presentation of information was to be utilised during public consultations, to enlighten and focus the debate.


Modelling Strategy

Although currently concerned only with the city centre, The ECCPM draws travel demand data from the City of Edinburgh Council's two existing strategic transport and traffic models, namely the Edinburgh Area Traffic Model (EATM) and the Joint Authorities Transportation and Environmental Study model Into Fife (JIF). The EATM is TRIPS based, and covers the Edinburgh conurbation to beyond the City Bypass into bordering local authority territories. JIF is a strategic multi-modal model with similar coverage, but with an extension across the Forth Estuary into Fife*.

The ECCPM provides some facilities not previously available, which are now enhancing the investigation of the traffic impacts of Edinburgh city centre. These include:

• dynamic modelling.


• junction interaction including the effect of an overcapacity junction affecting the operational performance of another junction upstream (i.e. blocking back).


• conflicts and interactions between road users, such as buses queuing to access bus stops, lane blocking by HGVs, or incident modelling.


• sensitivity to highway design issues, such as vehicle restrictions, road alignment, turning radii, location of bus stops.


• sensitivity to transport policy issues such as on and off street parking restrictions.


• a pollution model sensitive to vehicle acceleration, deceleration and queuing vehicles.

Additionally, whilst the S-Paramics system reports traditional measures such as traffic flows, percentage of heavy vehicles and journey times in a manner similar to other modelling suites, the ECCPM's visual presentation is easily understood by members of the public.

The methodology adopted to test city centre traffic management proposals varies according to the scale of impact that the proposal may have. Three possible methods are outlined :

• METHOD A : If a city centre traffic management proposal is small, and expected to have either a negligible or small impact on the road network beyond the city centre, then the ECCPM is used in isolation.



• METHOD B : If a proposal is expected to generate significant city centre and wide area re-routeing effects then the EATM is used to forecast the wide area impacts of the scheme. The ECCPM is subsequently used to analyse the city centre impacts of the scheme using EATM based travel demand patterns. Sensitivity testing of critical city centre junctions may be undertaken in the ECCPM using the Paramics junction simulation facility.



• METHOD C : If a proposal is expected to generate significant city centre and wide area re-routeing effects and other behavioural responses such as trip re-timing and mode switching then Method B can be enhanced through sensitivity testing, which would include adjustments to the ECCPM and EATM trip matrices through the use of external factors. The external factors could be derived from either policy objectives or from JIF.

Network Development

The ECCPM has been developed for two time periods to reflect 1998 average weekday morning and evening peak hour traffic conditions. The model focuses on the potential effects that the westbound closure of Princes Street to general traffic may have on the immediate surrounds (fig 1) Almost all roads within the study area have been simulated, exclusions amounting to some minor cul de sacs and private accesses. Network data requirements are similar to those of conventional models, but with emphasis on geometric detail. Both microsimulation and conventional assignment models require a fine definition of link lengths, junction locations, signal timings and bus route information. Microsimulation requires additional detail on kerbline positions, the number of lanes at all points on a link, lane restrictions (including bus lanes and on-street parking), positions of give way and stop lines, the location of bus stops, and service schedules. The benefit of this is that speed/flow curves and junction stopline capacities are not required, because it is the vehicle following, lane changing and gap acceptance sub-models within S-Paramics which simulate the nature of vehicle flow along a link (eg. speed, acceleration, deceleration, weaving) and at junctions (eg. waiting time at a stopline, queue lengths). The ECCPM therefore uses the real determinants of driver behaviour to model traffic flow rather than mathematical proxies such as speed/flow curves and junction capacities.

Whilst the data requirements of the ECCPM are more refined than those of a conventional model, the S-Paramics visual interface (fig 2) ensures that building the network is no more onerous than for a conventional model. This allows OS mapping information and AutoCAD drawings to be imported as backdrops, to enable the accurate positioning of nodes, kerblines and stoplines. OSGR coordinates are subsequently calculated for all items of data, obviating the necessity to code link lengths.

Trip Matrix Development

The travel demand information describes vehicle movements between 28 internal zones and 19 cross-boundary route zones, the ECCPM internal zones being the same as those of the parent model, EATM. Zone connectors do not exist in S-Paramics models because vehicles are loaded onto links in much the same way as real traffic enters a road network. This makes the position of zone boundaries less critical, and eliminates the usual problem of calibration near zone centroid connectors. Travel demand from EATM has been disaggregated into sixteen user classes using data from JIF, traffic counts and car parks.

A significant difference between microsimulation and conventional assignment models is the manner in which the trip matrix is loaded onto the network. In both cases the matrix is taken to represent the mean average number of trips travelling between each origin-destination pair in the modelled period over the course of, for example, a year. As a consequence of this, the ECCPM matrix contains fractions of trips. In a conventional assignment model this does not create a difficulty, since fifteen hundredths of a trip can be assigned in the same manner as one trip. In the ECCPM it is unrealistic and impossible to simulate the driving behaviour of fractions of a car, so the S-Paramics model uses the data contained in the trip matrix to define the probability of a trip being generated in any particular time step (usually 0.5 seconds). Whilst each ECCPM simulation is repeatable, the use of release probabilities to determine whether a trip is generated or not implies that each simulation is unique. The variation between simulations can be taken to represent the variation in traffic flows between days of the week. A set of different simulations is generated to provide traffic model outputs which can be analysed and averaged for the presentation of results.

The ECCPM also differs from a conventional assignment model in that the travel demand is profiled to simulate the peaks within the peak period. The ECCPM demand profile has been derived from an analysis of traffic count data, while a conventional assignment model would assume an average demand throughout the modelled period.


Assignment Methodology and Model Calibration

In a similar manner to the calibration of a conventional assignment model, the ECCPM has been calibrated through adjustments to the traffic assignment methodology and the network description. As for a conventional assignment model, each vehicle chooses a route to its destination that minimises a time and distance generalised cost function. The difference is that in the ECCPM each vehicle has a unique perception of travel costs, and re-evaluates its route choice at every node in the network according to the prevailing traffic conditions and its local familiarity. Thus a traffic assignment in the ECCPM is fully dynamic.

To reflect the fact that congestion delays can alter route choices, travel time delay costs are fed into the routeing algorithm every five minutes during the simulation. A decay function is used to ensure that route choices are most heavily influenced by travel costs derived over the previous fifteen to twenty minutes. Within the ECCPM the assignment method was assessed and adjusted by monitoring routes between critical origin-destination pairs at points throughout the simulation period. Since the ECCPM is a congested dynamic assignment model, the set of routes that may be chosen at any moment will depend on the congested state of the network leading up to that moment.

In a conventional model the assignment method would be monitored using link flow stability indicators (link flows are expected to be stable) and the proximity of the model to Wardrop's Users' Equilibium*. The link flow stability measure is inappropriate in the case of the ECCPM because it relates to the iterative mathematical procedures used in a conventional assignment model. Equilibrium proximity indicators are also largely irrelevant for the ECCPM because travel costs change every 0.5 seconds throughout the simulation period, and the number of vehicles completing the same O-D movement under similar traffic conditions is too small to make such a calculation meaningful.

In a fully dynamic congested assignment model such as the ECCPM, small network coding errors, such as those leading to incorrect lane utilisation at a roundabout, can have large effects, and may gridlock the model. Within a dynamic congested assignment environment such errors are very difficult to identify, and, to overcome this, a junction simulation utility within S-Paramics enables groups of key junctions to be isolated and simulated(fig 3). Fine corrections to the geometry, signal settings and lane descriptions within these groups constitutes the principal element of the calibration process. By ensuring that vehicle behaviour and the operational performance of links and junctions within each junction group is correct, the overall calibration effort for the ECCPM has been minimised. During model calibration, further adjustments to the network have been made to reflect the effects of signposted routes for drivers unfamiliar with the network, and the way in which some competing routes are perceived by drivers.


Model Validation

The ECCPM is validated to 1998 link and junction turning counts, journey times and queue lengths. Traffic flow output has been compared to observed data in a manner consistent with current best practice* but reflecting the microscopic nature of the model. Traditional link flow and journey time comparisons meet the required criteria, while additional comparisons regarding daily link flow variation, variation in journey times and queue length comparisons have been undertaken. These comparisons, which are unique to a microsimulation model, are described below.

Since each ECCPM simulation is unique, standard link flow comparisons undertaken for validation purposes consist of an average of five model runs for a particular time period. The variation between simulations is consistent with the observed daily variation at an ATC site on the boundary of the model. Vehicle flows across screenlines may vary by up to 6%, whilst links carrying in excess of 700vph may vary by up to 10%. Substantial variation (eg. +/-25%) can be found on links with flows of less than 200vph. Journey time comparisons in the ECCPM are similar to those undertaken in a conventional assignment model, although, as in the case of reality, modelled vehicles experience unique journey times. In the ECCPM the variation in journey times has been analysed and compared with observed variation, and Figure 4 and Table 1 demonstrate the close match obtained for the simulated median and the simulated spread of journey times when compared with observed values.
The microsimulation aspect of the ECCPM means that the nature of the validation differs from that of a conventional assignment model because maximum and minimum queue lengths have been compared at the key areas of the model. Figure 5 illustrates the areas where traffic queues form within the ECCPM during a particular 15-minute time slice. Since no DMRB guidelines exist regarding criteria that define an acceptable queue length comparison, the following categories were derived :

• Category 1: Low level of fit; simulated queue less than half observed.


• Category 2: Acceptable level of fit; simulated queue greater than half but less than observed.
• Category 3: Good level of fit; simulated queue equal to observed survey queue.


• Category 4: Acceptable level of fit; simulated queue less than double but greater than observed. Category 5: Low level of fit; simulated queue more than double observed.

Queue length comparisons at 65 sites for each 15 minutes simulated indicated an acceptable level of fit, similar to that achieved for the link flow comparisons. For example, in the AM peak a median acceptable queue length category 2 was achieved and in the PM peak the level of fit was category 3.

Whilst the model validation exercise does not directly compare the simulated behavioural mechanisms (eg. route choice, lane weaving, conflicts between cars and buses or lane utilisation), the comparisons are considered to validate them nevertheless. This is because if the simulated mechanisms were incorrect then the observed characteristics of the Edinburgh city centre road network (ie. traffic flows, journey times and queue lengths) could not be reproduced. This argument is similar to that used in conventional assignment models, whereby link flow and journey time comparisons are understood to indicate that the mathematical proxies and procedures used to determine route choice, link speeds and junction delays are correct.

In summary, the validation exercise demonstrates that the ECCPM base year model is a suitable basis for the testing and assessment of detailed traffic management measures, including bus priority schemes and detailed radical or non-standard junction designs.


Option Testing

The first application of the ECCPM has been to test the traffic management proposals associated with the removal of general traffic from the westbound carriageway of Princes Street. Given the wide area re-routeing implications of such proposals, which extend beyond the ECCPM's model boundary, updated travel demand patterns were extracted from the EATM for testing purposes (see Method B above). The proposals have such a dramatic impact on travel patterns within the city centre that, with one exception, all ECCPM traffic signal settings had to be adjusted during the testing procedures.

The ECCPM is helping to inform The City of Edinburgh Council and local interest groups about the impacts of road proposals. At a technical level the ECCPM has assisted council officials in the evaluation of the operational performance of the city centre network under the proposed measures. In particular the assessment of three junctions critical to the operational success of the proposals has been enlightened through the use of the ECCPM. The features assessed at the three junctions are :

• The effect of a set of bus pre-signals at a junction that is currently already at capacity.


• The performance of an at-capacity city centre roundabout, currently experiencing blocked back traffic queueing from a downstream junction, under the increased demand conditions that the proposed traffic management measures would bring; and


• The performance of a proposed complex signalised controlled junction with short weaving sections at an existing at-capacity signalised junction.

As well as illustrating the technical effects of proposals, such as the result of removing on-street car parking, the ECCPM provides pollution emissions output sensitive to vehicle speeds and acceleration, enabling council officials to be particularly well informed in their recommendations to elected members.

The graphical nature of the ECCPM, particularly the fact that it can visualise at driver's eye view in 3D , has proved valuable in assessing the sometimes controvertial traffic management measures. Interactive demonstrations of the model to elected council members, the police and the bus operators has opened the modelling procedures up to public scrutiny, lending further confidence and credence to the council officials' recommendations. The result, for a subject matter that has historically provided protracted and lengthy debates, has been an unusually smooth passage for difficult proposals through council procedures.

Public consultations continue with local interest groups, including resident groups, business interests and taxi drivers. The provision of ECCPM visual real-time model runs on CD for council officials to demonstrate during the consultation process has been a welcome experience for all parties, and greatly aided the understanding of issues.


Further Work

Building upon the work undertaken during Phase 1 of the ECCPM, the City of Edinburgh Council has commissioned an extension to the model area, to encompass the whole of the city centre. Upon completion, the ECCPM will include all of Edinburgh's Old Town area, including the proposed site for the new Scottish Parliament.


ACKNOWLEDGEMENTS

The authors are grateful for material in compiling this article to the City of Edinburgh Council.



Footnotes

* BATES ,J. et al. Building a strategic model for Edinburgh. Paper presented at 19th PTRC Summer Annual Meeting, September 1991.
* DEPARTMENT OF THE ENVIRONMENT, TRANSPORT AND THE REGIONS.Design Manual for Roads and Bridges (Volume 12, Section2, Part 1); Traffic appraisal of road schemes. The Stationery Office, London, 1998.