Why is the car of the worst mode of transportation in a dense urban city? Network and Zoning Systems • One of the most important early choices facing the transport modeller is that of the level of detail (resolution) to be adopted in a study. • This problem has many dimensions: it refers to the schemes to be tested, the type of behavioural variables to be included, the treatment of time, and so on. This section concentrates on design guidelines for two of these choices: zoning system and network definition. • We shall see that in these two cases, as in other key elements of transport modelling, the final choices reflect a compromise between two conflicting objectives: accuracy and cost. • In principle greater accuracy could be achieved by using a more detailed zoning and network system; in the limit, this would imply recognising each individual household, its location, distance to access points to the network, and so on. With a large enough sample (100% rate over several days) the representation of the current system could be made very accurate indeed. • However, the problem of stability over time weakens this vision of accuracy as one would need to forecast, at the same level of detail, changes at the individual household level that would affect transport demand. • This is a very difficult and mostly unnecessary task. Lesser levels of detail therefore, are not only warranted on the grounds of economy but also on those of accuracy whenever forecasting is involved Zoning Design • A zoning system is used to aggregate the individual households and premises into manageable chunks for modelling purposes. The main two dimensions of a zoning system are the number of zones and their size. • The two are, of course, related. The greater the number of zones, the smaller they can be to cover the same study area. • It has been common practice in the past to develop a zoning system specifically for each study and decision-making context. This is obviously wasteful if one performs several studies in related areas; moreover, the introduction of different zoning systems makes it difficult to use data from previous studies and to make comparisons of modelling results over time. • The first choice in establishing a zoning system is to distinguish the study area itself from the rest of the world. Some ideas may help in making this choice: • In choosing the study area one must consider the decision-making context, the schemes to be modelled, and the nature of the trips of interest: mandatory, optional, long or short distance, and so on. • For strategic studies one would like to define the study area so that the majority of the trips have their origin and destination inside it; however, this may not be possible for the analysis of transport problems in smaller urban areas where the majority of the trips of interest are through-trips and a bypass is to be considered. • Similar problems arise with traffic management studies in local areas where again, most of the trips will have their origin, destination or both, clearly outside the area of interest. What matters in these cases is whether it is possible to model changes to these trips arising as a result of new schemes. • The study area should be somewhat bigger than the specific area of interest covering the schemes to be considered. Opportunities for re- routeing, changes in destination and so on, must be allowed for; we would like to model their effects as part of the study area itself. • The region external to the study area is normally divided into a number of external zones. In some cases it might be enough to consider each external zone to represent ‘the rest of the world’ in a particular direction; the boundaries of these different slices of the rest of the world could represent the natural catchment areas of the transport links feeding into the study area. • In other cases, it may be advantageous to consider external zones of increasing size with the distance to the study area. This may help in the assessment of the impacts over different types of travellers (e.g. long- and short-distance). • The study area itself is also divided into smaller internal zones. Their number will depend on a compromise between a series of criteria discussed below. • For example, the analysis of traffic management schemes will generally call for smaller zones, often representing even car parks or major generators/attractors of trips. • Strategic studies, on the other hand, will often be carried out on the basis of much larger zones. • For example, strategic studies of London have been undertaken using fine zoning systems of about 1000 zones (for about 7.2 million inhabitants) and several levels of aggregation of them down to about 50 zones (at borough level). • As can be seen, there is a wide variety of models and number of zones per million inhabitants. These vary depending on the nature of the model (tactical and short term planning, strategic and long term), the resources available and the particular focus or set of problems addressed. • Note also that sometimes a more aggregated zoning system is used for part of the model system, for example trip generation and distribution • Then, a finer zoning system is often used for mode choice and assignment: these are often referred to as Traffic Analysis Zones or TAZs. • Zones are represented in the computer models as if all their attributes and properties were concentrated in a single point called the zone centroid. • This notional spot is best thought of as floating in space and not physically on any location on a map. • Centroids are attached to the network through centroid connectors representing the average costs (time, distance) of joining the transport system for trips with origin or destination in that zone. • Nearly as important as the cost associated with each centroid connector is the node in the network it connects to. These should be close to natural access/egress points for the zone itself. • Locating centroids automatically at the center of gravity of each zone and measuring their distance to key nodes to produce centroid connectors is a quick fix valid only for the simplest ‘first cut’ network runs. • Centroids and centroid connectors play a key role in the quality of the rest of the models, but their definition and coding does not follow a strict and objective approach; they rely a good deal on the experience of the modeller. • The centroid connector influences the route followed to load trips onto both the road and public transport networks and therefore affects the total cost of travelling from Origin to Destination and all the models that include them. Compilation of Zoning Criteria from Experiences in Several Practical Studies • Zoning size must be such that the aggregation error caused by the assumption that all activities are concentrated at the centroid is not too large. It might be convenient to start postulating a system with many small zones, as this may be aggregated in various ways later depending on the nature of the projects to be evaluated. • The zoning system must be compatible with other administrative divisions, particularly with census zones; this is probably the fundamental criterion and the rest should only be followed if they do not lead to inconsistencies with it. • Zones should be as homogeneous as possible in their land use and/or population composition; census zones with clear differences in this respect (i.e. residential sectors with vastly different income levels) should not be aggregated, even if they are very small. • Zone boundaries must be compatible with cordons and screen lines and with those of previous zoning systems. However, it has been found in practice that the use of main roads as zone boundaries should be avoided, because this increases considerably the difficulty of assigning trips to zones, when these originate or end at the boundary between two or more zones. • The shape of the zones should allow an easy determination of their centroid connectors; this is particularly important for later estimation of intra-zonal characteristics. A zone should represent the natural catchment area of the transport networks and its centroid connector(s) identified so as to represent the average costs to access them. • Zones do not have to be of equal size; if anything, they could be of similar dimensions in travel time units, therefore generating smaller zones in congested than in uncongested areas. • It is advantageous to develop a hierarchical zoning system, as in the London Transportation Studies, where subzones are aggregated into zones which in turn are combined into districts, traffic boroughs and finally sectors. This facilitates the analysis of different types of decisions at the appropriate level of detail. Hierarchical zoning systems benefit from an appropriate zone- numbering scheme where the first digit indicates the broad area, the first two the traffic borough, the first three the district, and so on. Network Representation • The transportation network is deemed to represent a key component of the supply side of the modelling effort, i.e. what the transport system offers to satisfy the movement needs of trip makers in the study area. • The description of a transport network in a computer model can be undertaken at different levels of detail and requires the specification of its structure, its properties or attributes and the relationship between those properties and traffic flows. For an early general review of network representation issues, see Lamb and Havers (1970). Network Details • The transport network may be represented at different levels of aggregation in a model. • At one extreme one has models with no specific links at all; they are based on continuous representations of transport supply (Smeed 1968). These models may provide, for example, a continuous equation of the average traffic capacity per unit of area instead of discrete elements or links. At a slightly higher level of disaggregation one can consider individual roads but include speed-flow properties taken over a much larger area; see for example Wardrop (1968). • Normal practice, however, is to model the network as a directed graph, i.e. a system of nodes and links joining them (see Larson and Odoni 1981), where most nodes are taken to represent junctions and the links stand for homogeneous stretches of road between junctions. • Links are characterised by several attributes such as length, speed, number of lanes and so on, and are normally unidirectional; even if during input a single two-way link is specified for simplicity, it will be converted into two one-way links in the internal computer representation of the network. • A subset of the nodes is associated with zone centroids, and a subset of the links to centroid connectors. Currently, the principal source of network data would be one of the many digital maps available for most cities. One should not assume, however, that they are error free. They will need checking, updating, pruning (to focus on the network of interest) and complementing with observations on items like on-street parking, pedestrian friction, bus lanes and other features that may affect their performance. • A problem with this scheme is that ‘at-node’ connectivity is offered to each link joining it at no cost. In practice, some turning movements at junctions may be much more difficult to perform than others; indeed, some turning movements may not be allowed at all. • In order to represent these features of real road networks better, it is possible to penalise and/or ban some turning movements. This can be done manually by expanding the junction providing separate (sometimes called dummy) links for each turning movement and associating a different cost to each. • Alternatively, some commercial computer programs are capable of performing this expansion in a semi- automatic way, following simple instructions from the user about difficult or banned movements. • The level of disaggregation can be increased further when detailed traffic simulation models are used. In these cases additional links are used at complex junctions to account for the performance of reserved lanes, give- way lines, and so on. • Sometimes networks are subsets of larger systems; they may be cordoned off from them thus defining access or cordon points where the network of interest is connected to the rest of the world. • These points are sometimes called ‘gateways’ and dummy links may be used to connect them to external zones. • A key decision in setting up a network is how many levels to include in the road hierarchy. • If more roads are included, the representation of reality should be better; however, there is again a problem of economy versus realism which forces the modeller to select some links for exclusion. • Moreover, it does not make much sense to include a large number of roads in the network and then make coarse assumptions about turning movements and delays at junctions. • It is not sensible either to use a very detailed network with a coarse zoning system as then spatial aggregation errors (i.e. in terms of centroid connections to the network) will reduce the value of the modelling process. • What matters is to make route choice and flows as realistic as possible within the limitations of the study. • Jansen and Bovy (1982) investigated the influence of network definition and detail over road assignment accuracy. Their conclusion was that the largest errors were obtained at the lower levels in the hierarchy of roads. • Therefore, one should include in the network at least one level below the links of interest: for example, in a study of A (trunk) roads one should also include B (secondary) roads. • In the case of public-transport networks an additional level of detail is required. The modeller must specify the network structure corresponding to the services offered. These will be coded as a sequence of nodes visited by the service (bus, rail), normally with each node representing a suitable stop or station. • Junctions without bus stops can, therefore, be excluded from the public-transport network. • Two types of extra links are often added to public-transport networks. These are walk links, representing the parts of a journey using public transport made on foot, and links to model the additional costs associated with transferring from one service (or mode) to another. Link Properties • The level of detail provided about the attributes of links depends on the general resolution of the network and on the type of model used. At the very minimum the data for each link should include its length, its travel speeds (either free- flow speeds or an observed value for a given flow level) and the capacity of the link, usually in passenger car equivalent units (pcu) per hour. • In addition to this a cost-flow relationship is associated with each link as discussed below. In some cases, more elaborate models are used to relate delay to traffic flow, but these require additional information about links. Link Properties • Type of road (e.g. expressway, trunk road, local street). • Road width or number of lanes, or both. • An indication of the presence or otherwise of bus lanes, or prohibitions of use by certain vehicles (e.g. lorries). • Banned turns, or turns to be undertaken only when suitable gaps in the opposing traffic become available, and so on. • Type of junction and junction details including signal timings. • Storage capacity for queues and their presence at the start of a modelling period. • Some research results have identified other attributes of routes as important to drivers, for example tolls, signposting and fuel consumption (see for example Outram and Thompson 1978 and Wootton et al. 1981). • Work in the Netherlands has shown that (weighted) time and distance explains only about 70% of the routes actually chosen. The category of the road (motorway, A road, B road), the predictability of the time taken, scenic quality, traffic signals and capacity help to explain additional routes. • As our understanding of how these attributes influence route choice improves, we will be able to develop more accurate assignment models. The counterpart of this improvement will be the need to include other features of roads, like their scenic quality, gradient, and so on. Network Costs • Most current assignment techniques assume that drivers seek to minimise a linear combination of time, distance (associated to fuel costs) and tolls (if any), sometimes referred to as generalised cost for route choice. • This is known to be a simplifying assumption as there may be differences not only in the perception of time, but also about its relative importance compared with other route features. However, the majority of network models in use today deal only with travel time, distance and tolls. • When modelling travel time as a function of flow one must distinguish two different cases. • The first case is when the assumption can be made that delay on a link depends only on the flow on the link itself; this is typical of long links away from junctions and therefore it has been used in most inter- urban assignment models so far. The second case is encountered in urban areas where the delay on a link depends in an important way on flows on other links, for example for non-priority traffic at a give-way or roundabout junction. • The introduction of very general flow-delay formulations is not difficult until one faces the next issue, equilibration of demand and supply. • There, the mathematical treatment of the first case (often called the separable cost function case) is simpler than the second; however, there are now techniques for balancing demand and supply in the case of link-delay models depending on flows on several links, i.e. When the effect of each link flow cannot be separated. Public Transport Networks • Public Transport networks are more complex than road networks. They require an identification of the route taken by each service as a unique sequence of links. It is also necessary to identify the locations where stops are possible and also those where interchange with other services is permissible. • The frequency of the service and in some cases the actual timetable and the fare, must also be specified and included in the network description. • Access to stops may be on foot or by another mode and this can be represented by centroid connectors in the simpler models and by one or more auxiliary networks of access modes in more realistic undertakings. This is why the centroid connectors for public transport are always different from those used for the road network. • The public transport network could be entirely independent of the road network as in the case of most rail and metro services; • Alternatively, the speed of the service may be affected by road traffic as in the case of buses and on-street running of trams, even when priority measures to support them help to reduce the impact of road congestion. • In addition to the effect of road congestion, it is sometimes necessary to account for the issue of passenger congestion: crowding on buses and trains leading to discomfort and even having to miss a service because it was full and impossible to board.