Final Report
(Deliverable D4 Final Release Version)
For ITEA
Department of Transport, Local Government and the Regions
Contract PPAD 9/65/71
(ME&P ref. P959/2)
ME&P
(Part of the WSP Group)
Cambridge
May 2002
(Corrections to minor typing errors made July 2003 for DfT web site publication)
The purpose of this report is to present the LASER Model Version 3.0. This is a MEPLAN-based land use/transport interaction model for a Study Area that covers Greater London and 2 UK Government Office regions around London, i.e. the South East and the East of England. The model includes two main components:
a) A land use component, which simulates some key relationships between industries, employed residents, households, and floorspace. This land use component forms the basis on which all residents' travel demand within the Study Area is estimated.
b) A strategic transport component which assigns the travel demand of the AM peak to modal networks and estimates the monetary and time-based generalised costs of travel subject to capacity restraint on road and rail-services.
These two-model components together simulate the interaction of land use activities and transport supply conditions. The model is intended for use to forecast the future location of land use activities and of transport demand under different policy scenarios, thus providing a consistent basis for assessing the packages of infrastructure schemes and regulatory measures under consideration.
Chapter 2 of this report provides an overview of the modelling approach, model structure, and zoning. Chapters 3 and 4 discuss respectively the implementation and calibration of the land use model and the strategic transport model. Chapter 5 describes the model runs of 1991 and 1997, and provides a validation of the model against the observed data that is currently available. Chapter 6 assesses the strengths and weaknesses of LASER3.0, and discusses its use for future policy applications and the scope for further development of some aspects of the model.
This LASER Enhancement Project for the ITEA Division of DTLR had the specific task of implementing a limited package of model enhancements designed to be the minimum requirement needed to improve the performance of the model. The work programme aimed to tackle the weaknesses identified by the Laser Model Validation Study (Atkins Wootton Jeffreys and David Simmonds Consultancy, 1997), particularly in those areas where good quality data has become available since the previous version of LASER (i.e. Version 2). The updated and enhanced version of the LASER Model is now labelled Version 3.0.
Since the commissioning of the LASER Enhancement Project there have been three further studies which have had a bearing on the design and implementation of LASER3.0 as follows.
First, LASER has been chosen as the strategic land use/transport model for the London Orbital Multi Modal Corridor Study (known as Orbit). LASER3.0 is being applied in the Orbit study, within which further extensions to the model have been requested to address the particular needs of that study. These model extensions have been developed in parallel to the LASER Enhancement project. They include a major improvement in the representation of rail and underground services, refinement of the road network characteristics, and a new residential floorspace model. Furthermore, in order to address the needs of Orbit the number of zones was substantially increased beyond that originally envisaged. Since all of these extensions now form an integral part of the LASER model, model extensions are reported here as part of LASER3.0, irrespective of which project originally commissioned them. The main usage of the LASER model in the Orbit study has been to test policy initiatives in the year 2016. This usage is not reported in this report which only presents results for the model runs in 1991 and 1998.
Secondly, LASER has been chosen as the strategic land use/transport model for the London-Ipswich Multi Modal Corridor Study (known as LOIS). Since the LOIS study is focused only on the north east sector of the LASER study area, the LASER3.0 zoning structure will have to be altered to suit the LOIS requirements. This has no direct impact on the LASER Enhancement Project.
Thirdly, the South East Regional Airports Study (SERAS) has expressed their interest in using LASER3.0. Special airport zones and an airport employment category have been included in LASER3.0, and the road and rail connections to airports have been coded. Furthermore, since LASER3.0 does not model air travel, SERAS has provided passenger travel and goods vehicle matrices to and from Heathrow, Gatwick, Stansted, London City and Luton airports. LASER3.0 makes use of these road and public transport matrices as an exogenous input in its network assignment procedures. No additional LASER model development is envisaged for SERAS.
Appendix 1 provides the details of how LASER3.0 is related to Orbit, SERAS and LOIS in terms of model development.
The LASER model is expected to evolve further as it is applied to practical projects such as those mentioned above, and further modifications and extensions of LASER3.0 may be identified as version 3.1 and so on.
The methodology that has been adopted in enhancing LASER can be summarised as follows:
(a) It aims to address the weaknesses of LASER 2 as identified by the LASER Model Validation Project.
(b) It is focused on updating those areas of the model which can significantly benefit from new land use and transport data which became available since the calibration of LASER 2. The data includes the SWS, LBS and SAR datasets from the population census, the AES and IDBR data for employment, the LATS and NTS data on travel choice, and the NAOMI and PLANET transport networks (see the Glossary for definitions of these various acronyms). The data was used not only to define a cross-sectional snapshot, but also, more importantly, to calibrate the patterns of residential location and modal choice behaviour of different socio-economic groups.
(c) A modular structure has been developed. Whilst retaining a coherent theoretical framework, it enables some of the key elements of land use/transport simulation to be set up and used quickly for practical purposes within the tight time limits of practical studies. Its zoning structure is flexible enough to suit the varied requirements of applications in different geographical areas covered by the LASER study area.
The information flow within the LASER Model can be broadly described as being between four simulation procedures, the first two of which are considered as land use and the remaining two are transport:
(a) Estimation of demand for employed persons, households, local services and floorspace; this is implemented through a generalised input-output model similar to a social accounting matrix (SAM).
(b) Estimation of how such demand is distributed amongst the geographical locations within the study area. Journeys to work, school and services between geographical locations are a by-product of this procedure. Single level multinomial discrete choice models were calibrated on the Census and NTS data for their concentration parameters and residual disutilities
(c) The journeys are attributed to different modes of transport that are available between each pair of locations, according to the modal choice behaviour of each socio-economic group. This is done by demand segment defined in terms of the SEGs, car availability and purpose of travel. Multi-level multinomial discrete choice models were calibrated on the London Area Transport Survey and National Travel Survey data for their concentration parameters and modal disutility functions.
(d) The journeys on each mode are then assigned to the road and rail (including London Underground) networks, using a discrete-choice-model based stochastic user equilibrium assignment procedure incorporating road and rail-service capacity restraints.
Figure 2.1 LASER3.0: model structure and data sources for c alibration
Whilst travel demand information is generated and fed top-down from (a) to (d), the travel costs, times and time based generalised costs are transmitted bottom-up from (d) to (a). This allows an extensive interaction between land use and transport. The model calibration procedure starts from component (d), and works its way back up to (a), following the order in which the costs and disutilities are estimated. LASER3.0 includes AM peak modal choice and assignment only. However, it can be extended to cover other time periods of travel. Figure 2.1 summarises the main simulation procedures of LASER3.0.
One way of approaching the structure of the LASER model is to compare it with a standard four-stage transportation model. In LASER, the generation and distribution stages are replaced by a Land Use Model, whilst the modal split and assignment stages remain similar to those of a four-stage model. The Land Use Model supplies demand matrices to the modal split and assignment stages. It also takes in return from them transport costs between zone pairs, which are used in estimating location choice and trip distribution.
Unlike in the four stage model, where the population and employment in each zone is taken as an input to the model for trip generation, in LASER3.0 the employed persons at zones of work and residence are estimated from demand for labour by industry. Households consisting wholly of unemployed, retired and other economically inactive adults are estimated externally and form an input to the model.
The details of the land use and transport models are presented in Chapters 3 and 4 respectively. Chapter 5 reports model validation. Section 2.2 below provides a summary of the main data sources used for model development (Also see Figure 2.1). Model zoning (i.e. subdividing the Study Area and the rest of Great Britain into suitable geographical units), which are identical for both land use and transport models, is described in Section 2.3.
Since LASER 2, a number of land use and transport data sources have become available. Some of the data sources that were available at the time of LASER 2 have also improved in terms of quality and availability. This has improved the conditions for model implementation and calibration. Table 2.1 provides a list of the data sources used in LASER3.0 model development. See Figure 2.1 for their input into the model calibration process.
It is appropriate here to note the empirical reasons why a large proportion of the data that is presented in Table 2.1 is ten years old, and they would still be valid to use for model calibration. Time series statistics show that the patterns of behaviour (such as perceptions of cost, travel time, the comfort of a mode, or the environmental quality of residential areas) generally remain stable over many years, provided that the demand groups are suitably segmented, as is done in LASER3.0. The demand models are established in such a way that the basic patterns of behaviour are parameterised separately from the external variables, such as the actual cost or time of travel, or the level of rent for dwellings. As the external variables change, the demand groups react in their choices of modes or residential location. Nevertheless, the basic patterns of behaviour represented in the model remain unchanged. Models calibrated this way are expected to remain valid for a substantial period of time. It goes without saying that their validity deteriorates over time, although how rapidly they do so will depend on the rates of change in the basic economic and cultural life styles of each household segment (i.e. how a high income car owner changes his/her preferences for residential locations and travel modes through the years, or how a low income non-car owner changes his/hers).
Table 2.1 Main data sources for LASER3.0
|
Data |
Purpose |
|
NAOMI highway networks: 1991, 1997 and 2016 |
Main source of the interzonal road network data for the model |
|
GIS based zoning map and networks |
Developing intra-zonal band based coding of intrazonal travel distances, costs and times |
|
Observed passenger demand data by mode from LATS91 |
For modal choice model calibration - using this detailed survey for London residents |
|
Other observed passenger demand data: NTS, SWS |
For model parameter estimation and validation for the LASER internal model area as a whole |
|
External demand matrices from NAOMI and SERAS |
For assigning non-modelled traffic on the model network to obtain appropriate network loading and speeds under congestion |
|
Travel costs and tariffs from LTS91, APRIL and own collection at ME&P |
For updating transport user costs |
|
LBS households, employed residents and dwellings for 1991 |
For setting up the land use model in 1991 |
|
SWS employment data for 1991 |
For setting up the land use model in 1991 |
|
AES and IDBR employment data |
For setting up and validating the land use model in 1997/98 |
|
NTEM 1998 and future year data |
For setting up and validating the land use model in 1997/98 and in the policy years |
|
SARs: demographic coefficients |
For setting up the land use model in 1991 |
|
FES |
For setting up the land use model in 1991 |
|
PLANET surface rail networks and services |
To improve the representation of rail services |
|
Underground and other auxiliary rail data from LASER 2 |
An improved dataset used to represent underground services, station access, and development of rail and LU overcrowding functions |
|
Dwellings data from Census and DTLR and other sources |
To improve the residential floorspace model |
|
Future land use planning policy scenarios |
For running the model into the future |
Since the 1991 Census represents the best available dataset for land use model calibration, and similarly the LATS91 data for an interzonal modal choice calibration by demand segment, LASER starts its runs at Year 1991. The calibrated 1991 model is then run forward to Year 1997 for validation purposes, before it is applied to policy simulation of a future year. The choice of 1997 is principally based on the fact that both the NAOMI road traffic model and the PLANET rail model can provide transport networks for LASER for that year; and also transport validation data is more easily available for that year. Furthermore, in Orbit and LOIS, LASER acts as an upper-tier strategic transport model, through an interface with NAOMI road and rail traffic models, the base year of which is 1997. The main land use data for LASER for this 1997 run in fact comes from the 1998 dataset within NTEM. It would have been more accurate to call this run 1997/98. However, for convenience it is named the 1997 run throughout this report.
The way in which LASER is run through time is outlined in Figure 2.2. The 1991 calibration of the land use and transport models provides the key model parameters for 1997 and later years, which are shown by the horizontal arrows pointing to the right. The 1991 transport model network, however, is derived by modifying those sections of the networks which have been changed between 1991 and 1997. This is because the definitive base NAOMI and PLANET networks are for 1997, and it is thus more convenient to identify the network changes between 1991 and 1997 and subtract all network improvements from the 1997 network in order to create a 1991 network. This is indicated by a horizontal arrow pointing to the left. In 1991, the land use and transport models are run iteratively until a stable solution is reached.
The land use model for 1997 does not directly use the transport cost matrices (which include both money costs and generalised costs) generated by the 1997 transport model. To do so would mean that land use activities would have reacted instantaneously to transport costs of the current year. In fact there is usually a time lag between a change in transport cost and a change in land use decisions, and only some proportion of the travellers can react to changes in transport at any one time. At present there is no suitable empirical study on how long the time lag is for travellers to react to transport cost changes; conceptually the time lag depends both on the nature and magnitude of the transport cost change and the nature of the land use activities. For simplicity it is assumed in this model that half of the travellers make their land use decisions on the current year transport costs (i.e. 1997), and the other half do on the historic transport costs of the previous model period (i.e. 1991). This is in practice achieved through creating a set of average transport cost matrices that take the 1991 and 1997 matrices on equal weight. This is indicated by the two merging arrows pointing from the 1991 and 1997 transport models towards the 1997 land use model.
For a future policy year, there are two ways in which the land use/transport interaction can be represented.
a) A future policy year can be run with inter-temporal cost averaging as in 1997, as shown for 'Future Year 1'; model outputs from such a run indicate the most likely land use pattern based on time-lagged land use decisions.
b) Alternatively, the land use and transport models can be run iteratively until a stable solution is reached for a future year, as shown for 'Future Year 2'; this is applicable, for example, where the long term equilibrium solution is of policy interest.
The model can be switched to run under either procedure with minimal effort. In practice, options (a) and (b) above may be selected on the grounds of policy interest. For a medium term forecast, it is important to include the time-lag effect; it is therefore appropriate to adopt (a). For a very long term forecast, however, (b) is perhaps a theoretically preferred choice.
Figure 2.2 Running LASER3.0 through time
A key variable in running the model over time concerns how the valuation of travel time changes. This has been a subject of intense analysis recently (See Wardman, 2001, DTLR, 2001). LASER3.0 adopts the latest DTLR advice and link the value of time changes to the growth of GDP, with an elasticity value of 0.75. This means the values of time increase at 75% of the rate of GDP growth.
A zoning scheme represents the way in which the Study Area and the rest of Great Britain are subdivided into suitable geographical units. It is worked out based on the criteria concerning the geography, transport, data availability, model run time, and compatibility with the National Trip End Model (NTEM):
In terms of the geography and transport supply conditions of the study area, there have been the following considerations:
(a) The Central Statistical Area of London is subdivided into its borough components, so that model data can be obtained using both boundary definitions. As the LASER3.0 zones strictly follow the boundaries of the 1991 Census wards, whilst those of the Central Statistical Area do not, the two respective sets of boundaries do not exactly match. However, they are close enough for meaningful comparisons to be made.
(b) In the rest of London, the boroughs are subdivided into zones according to key rail stations and trunk road access
(c) Around the M25, the Local Authority Districts are subdivided according to key rail stations and motorway junction/trunk road access
(d) Further away from London, dense urban areas are separated from surrounding countryside as zones
(e) The Thames Gateway area is identifiable as a group of zones. As the Thames Gateway area is officially defined using roads rather than Census wards as boundaries, the boundaries do not exactly match but are close enough to be meaningful.
(f) Gatwick, Stansted, Luton and the London City airports are defined as point zones (the Heathrow Airport is subdivided into 5 point zones), following the advice from SERAS
(g) The main ports such as Dover, Folkestone (for Channel Tunnel), Tilbury, Harwich, Felixstowe, and Southampton are defined as point zones for assigning overseas traffic coming and going through the ports
In terms of data availability for the land use model,
(a) All zones are established based on ONS fixed 1991 wards/Local Authority District boundaries, with the exception of ports and airports
(b) The LTS and NAOMI zone boundaries are largely respected unless their boundaries conflict with the ward or district boundaries; the port and airport zones also correspond to NAOMI zones or their amalgamations
(c) External zones (i.e. those areas of Great Britain outside the study area) follow the NAOMI zones or their amalgamation
A critical constraint on the total number of zones is model run time, and ease of pre- and post-run data processing. The land use and transport model use matrices whose dimensions are determined by the number of zones squared and the number of demand segments. In other words, the dimension of the model increases with the square of the number of zones. The larger the model, the more time it takes to run and to analyse.
(a) For acceptable model run time in calibration and policy testing using currently available personal computers, the maximum limit on total number of zones is 335. This zone number implies a single transport model run will take 15 hours (i.e. overnight) on the best currently available Personal Computer at 1.7 MHz processor speed and 2 GB of RAM. Any slower running time would seriously impinge on the turnaround of model calibration and model running.
(b) For easy interface and data analysis, the land use and transport models have identical zones.
As LASER3.0 uses the employment and demographic data from the National Trip End Model (NTEM) in 1998 and for future years, its zones have also been made compatible with NTEM.
(b) The broad zoning principles are similar in both models. For consistency and compatibility, the LASER zones are defined to nest with the NTEM zones. In London, the LASER zones are either the same size as the NTEM ones or smaller. In the South East and the Eastern regions the LASER zones are either the same size as the NTEM or larger. (There is one exception to this rule however, of a small ward in Buckinghamshire, Chenies, for which some Census data was suppressed - this zone was incorporated into Amersham, which is the corresponding Census LBS data importer).
For modelling exogenous employment in retail and education, an additional dummy zone is added as a modelling device.
Table 2.3 summarises the resulting zoning scheme. The detailed representation of these zones is presented in Figure 2.4 for London itself, in Figure 2.6 for the rest of the internal Study Area and finally in Figure 2.8 for the set of external zones covering the rest of Great Britain.
The extraction of the 1991 Census data of households, dwellings and employment was made by zone, which helped to verify the zone sizes (in terms of the number of households) and the degree of homogeneity of dwelling types. Some initially defined small zones (in terms of population size) were amalgamated into neighbouring zones. The zoning system differentiates well the areas of flats, terraces, and detached/semi-detached houses, and of different car ownership patterns in London and in the rest of the study area.
Table 2.2 Summary of LASER3.0 z ones
|
Area |
Number of zones |
|---|---|
|
Inner London |
45 |
|
Outer London |
75 |
|
South East and Eastern regions |
177 |
|
Airports |
9 |
|
Ports |
6 |
|
Total Internal Zones |
312 |
|
External |
22 |
|
Dummy zone (modelling device) |
1 |
|
All |
335 |
Figure 2.3 LASER3.0 model zones: London
Figure 2.4 LASER3.0 model zones: the South East and the Ea stern Regions
Figure 2.5 LASER3.0 model zones: e xternal zones in Great Britain
The land use model consists of two broad economic sectors (see Figure 3.1):
(a) the industrial, or productive, sector which represents all primary, manufacturing and service industries in the Study Area
(b) the household sector, which acts both as a final consumer and a supplier of labour to the industries
Within the model, each sector is composed by a number of land use factors (such as industries, the employed workers, households, floorspace, travel demand, etc.) which represent the various categories of producers, consumers, and activities of production and consumption.
Figure 3.1 The conceptual structure of the land use model
The industrial sector consists of a number of industries (classified using aggregations of Standard Industrial Classification or SIC codes), their demand for labour measured in employed and self-employed persons, the demand for commercial floorspace in which to locate and conduct the businesses, and the demand for travel on business. The left part of the diagram in Figure 3.1 depicts these relationships.
The household sector, on the other hand, is represented by employed and self-employed residents, their households, other unemployed and economically inactive households, and the demand generated by the households for housing, consumer goods and services, any delivery associated with the goods and services, and (through the individual members within the households) the demand for personal travel to schools, shopping/personal business, leisure, and others. This is shown on the right hand side of Figure 3.1.
The model links the industrial and the household sectors with
(a) residents going to work as the employed workers, which provides the basis for simulating commuting journeys to/from work as well as the demand and supply of labour.
(b) the households' demand for goods and services. LASER3.0 only models household demand for local retail services, and local education services. All other production activities are assumed exogenous to the model, i.e. the location of their employment is input to the model by industry by zone before the model is run (see Section 3.2.3 for details).
The basic units of measurements are: cost in pounds, time in hours (for the time-based measurement of generalised travel cost). Households, employed persons, jobs and dwellings are also units used for the respective land use factors.
Figure 3.1 implies a generalised input-output (I-O) structure (similar in concept to a partial social accounting matrix) as shown in Table 3.1. The columns of the table represent demand. The rows of the table generally represent supply, with the exception of those shown in italics, which are either land use model constraints used for parameter estimation, or for accounting. (The constraints are used for parameter estimation and re-adjustment within the model to ensure that it matches the observed data or externally determined control totals as far as possible. The derived parameters, rather than the constraints themselves, are then used in the main runs of the model.)
It is appropriate here to clarify the definitions of some land use factors used in the I-O structure, particularly concerning employment.
(a) Industries. Industries are represented as employed and self-employed persons. They are segmented by industry and thus represent persons working for each of the industries. In some planning datasets such as the NTEM, jobs rather than persons are used. Accounting in terms of jobs is not consistent with that in persons, and in this project a translation procedure is developed to convert jobs to an estimate of persons for the NTEM data (see Section 3.2 below)
Tab le 3.1 LASER land use model I-O s tructure
This Table is available separately for download in Word format from the foot of this page.
(b) The employed workers. Industries demand employed and self-employed persons. The employed and self-employed persons as one group are referred to as the employed workers at the place of work. In previous LASER reports terms such as 'employed persons' or 'jobs' were used; these terms were found to lead to confusion on various occasions
(c) Employed residents. Employed residents at place of residence travel to the place of work to become the employed workers.
(d) Employed households. Households in which at least one adult is employed or self-employed
(e) Unemployed households/Inactive households. Households in which no employed or self-employed person is present. Rather, an unemployed or otherwise economically inactive adult heads them.
This I-O framework is used to generate all domestic passenger travel made by the residents within the study area. In and out commuting are permitted, with journey to work trips crossing the study area boundaries. For journeys on other purposes, however, the study area is treated as a closed region, with no cross boundary travel allowed. If LASER is to be used for applications close to the periphery of the study area, further work will be required to implement in and out trips crossing the boundary, so that the travel patterns for the purposes other than commuting can also be represented there.
The above land use model structure is established based on the datasets, which have become available since LASER 2. The main items include the LBS data on households and dwellings for 1991. The SWS data on the employed workers and employed residents for 1991, the AES and IDBR data used for estimating the employment changes between 1991 and 1997, the Samples of Anonymised Records (SARs) which describe the relationships between industries. In addition, the employed workers, households, employed residents, and other household members, and the FES data which represents the household consumption patterns.
Many items of land use data require a standard data use licence or special permission. In all cases such permission has been obtained.
An issue that is specific to land use data is to define the geographical areas for the data extraction. Many zones in the centre of the study area consist of a ward or a group of wards. Where possible, for example with the Census data, a data extraction was made at ward level. Further away from the centre the model zones are larger and consequently data was extracted at the district level and for the external zones, at the county level.
The main calibration tasks for 1991 are:
(a) Define I-O coefficients between industries, the employed workers, households and individual household members
(b) Input 1991 calibration constraints for industries, and the employed workers
(c) Define inputs of exogenous industries and exogenous households for 1991
(d) Set up the dwellings model
(e) Calibrate spatial distribution models for journeys to work, other private travel and business travel
(f) Model verification and validation
For 1997, whilst the bulk of the model parameters will be inherited from 1991, the following tasks are necessary:
(a) Define inputs of exogenous industries and exogenous households for 1997 from the NTEM, AES and IDBR databases
(b) Modify I-O coefficients between industries, the employed workers, households and individual household members with estimated constraints from the NTEM database so that LASER for 1997 has the same study area wide totals of the employed workers, households and individual household members for all appropriate categories
(c) Model verification and validation
Model calibration details are discussed below with each of the land use factors.
Industries are represented in LASER3.0 in terms of the number of employed and self-employed persons working in each industry in the study area. Table 3.3 defines this segmentation via SIC80 codes (which are used to extract the 1991 Census information) and SIC92 codes (which are used for IDBR, AES and NTEM data).
Table 3.2 Segmentation of i ndustries
|
Industry code |
LASER3.0 Industry Groups |
SIC80 Codes (as in Census 1991) |
SIC92 Codes (as used in some post-Census91 data) |
|
1 |
Primary industries |
0,11,13,16,17,21,23 in zones outside Central London |
01,02,05,10-14, 40, 41 in zones outside Central London |
|
2 |
Manufacturing and distribution |
12,14,15,22,24-26,3,4,61,62, 79 in zones outside Central London |
15-37 in zones outside Central London |
|
3 |
Construction and transport |
5, 71-74, 75-77 in zones outside Central London |
45, 60-63 in zones outside Central London |
|
4 |
Industrial sector office work |
All primary and manufacturing industries in Central London zones |
All primary and manufacturing industries in Central London zones |
|
5 |
Finance and business |
8 |
65-70 |
|
6/26 |
Retail, catering and repairs |
63-67 |
52,55 |
|
7/27 |
Education |
93 |
80 |
|
8 |
Government and compulsory public services |
91-92 |
75 |
|
9 |
Other services NES |
79,94-00, other not stated |
50,51,64,71-74,85-95,99,00 |
|
10 |
Airport services |
Reserved for SERAS; not currently in use |
Reserved for SERAS; not currently in use |
Note: It is not possible to match exactly SIC80 and SIC92 codes. The categories presented here are believed to be the closest possible match for the purposes of LASER.
Within the model, these categories are defined as land use factors 1-10 for exogenous industries, and 26-27 for endogenous retail and education.
The SWS data on industrial employment does not distinguish office workers from factory workers within a given sector of industry. Yet the two type of workers tend to have distinctly different behaviour in land use and transport choices. As no data is readily available, a simple assumption was made. As there is little scope for factory activities (with the exception of specialised workshops which are becoming more and more similar to offices) in Central London, all workers belonging to the primary and manufacturing sectors are assumed to be the office type workers in the zones that fall within Central London.
A major difference exists between the 1991 Census and the 1998 NTEM data in terms of the definition of employment by industry. Whilst the 1991 Census represents employment with the number of employed and self-employed persons, the 1998 NTEM data only gives the number of jobs. For the majority of zones in the study area, the difference between the two is negligible. Nevertheless, the difference can be significant in areas where many employed or self-employed persons are represented by more than one job per person. Data analysis shows that such differences are most pronounced in the zones of Westminster, particularly the West End. Since in the land use model, employment is the generator of labour demand, and hence is the basis of journeys to work modelling, an inconsistency in the employment data would lead to erroneous results for some important zones in Central London. It is therefore necessary to translate the 1998 NTEM jobs data into employed and self-employed persons before it can be used by the land use model for the 1997 run. This translation was done via the AES and IDBR data.
Both AES and IDBR provide data on employment in terms of number of jobs. Whilst the AES data provides a consistent annual estimation of the levels of employment (which is used for time series analysis of the trend in employment changes), it cannot be provided at a detailed level of sectoral or spatial disaggregation owing to data confidentiality rules. The IDBR data can be released at a high level of sectoral and spatial disaggregation for use in DTLR projects under strict data confidentiality conditions. However, the IDBR does not provide historic figures, and provides no indication of trend through time. After extensive discussion with the AES and IDBR teams at the ONS the following approach was adopted:
(a) to obtain the AES data from NOMIS at a more aggregate level than that used in the model. This will provide consistent employment totals for 1991 through to 1998
(b) to request via DTLR the current IDBR data at the level of LASER3.0 zones, and use it to distribute the AES totals.
In this way both datasets are used in a way that is appropriate to the nature of the data.
Once the 1991 jobs data is estimated by industry by LASER zone from the AES and IDBR, it is used with the 1991 Census employment data to derive jobs to employed persons ratios by industry by LASER zone. As no data is available otherwise, these ratios are assumed to be constant through time and applicable to self-employed persons for each industry in each zone. On this basis the ratios are used to translate the 1998 NTEM jobs (which are believed to include both employed and self-employed) into persons by industry by zone, for input to the land use model in 1997. The NTEM grouping of industries is not defined in the same way as the 1991 Census. However, a closest possible match between the two is made in terms of industrial classification.
In the land use model, the bulk of the employment by industry is treated as an a priori, exogenous input to the model (in total 84% of industrial employment is exogenous). The employment numbers are input by industry by zone. This model implementation is primarily due to the fact that little information is currently available on commercial floorspace that the industries use to conduct their businesses. Since in the study area many of the industries in the majority of zones are constrained by the availability of industrial, retail or office space, it would be impossible to implement a sensible model of industrial location without the floorspace data.
A second reason is that industrial employment data is by and large available by industry group by LASER zone:
In the absence of commercial floorspace data, it would seem preferable to make use of the existing NTEM projections of employment by industry by zone, rather than setting up industrial location models within LASER without adequate input information.
However, in LASER3.0 the demand for two local service sectors is being generated endogenously within the model. These are local retail services, and primary and secondary education services. The approach is as follows:
(a) the demand generated by households in goods and services excluding dwellings and domestic travel (i.e. land use factor 'Other goods and services') is measured in pounds sterling and each pound spent generates a fixed demand for employed/self-employed persons working in the retail sector.
(b) the trips generated by household members to education are converted into a demand for employed/self-employed persons working in the education sector at the destination of the trips.
The employed workers required per unit of consumer demand is worked out as the study area wide average. Detailed AES data at 4 digit SIC level were used to determine the proportion of retail and education workers employed for these local services. Those working in retail and education sectors for non-local services are treated as exogenous employment. At the 1991 land use calibration stage, the zonal distribution of exogenous and endogenous employment in these two sectors is estimated simultaneously. Zonal attractors are estimated on the basis of 1991 Census employment data to distribute exogenous employment to appropriate zones (mainly to zones where national retail centres and universities are located). These attractors remain in force for the model runs of 1997 and the later years.
The employment by industry, either exogenous or endogenous, generates demand for employed and self-employed persons by SEG, household size and car ownership. As industrial employment is measured in persons, the demand-supply is a one-to-one relationship for the employed workers taken as one group. However, because the employed workers are segmented by SEG, household size and car ownership, demand coefficients represent the proportional composition of persons from each SEG-household size-car ownership segments. In other words these coefficients can be interpreted as probabilities of the employed workers of an industry coming from a certain SEG-household size-car ownership segment. By definition the coefficients over all segments for an industry sums to 1.
The 1991 Census provides the basic information for estimating these coefficients. The procedure makes use of the land use model and it involves a number of steps as follows:
(a) 1991 Industrial employment is input into the model. At this stage, all industrial employment is treated as exogenous (as extracted from the 1991 Census by industry by zone), and the households' demand for endogenous employment is turned off in the model. In other words, there is no exogenous/endogenous employment split at this stage
(b) Input to the model the constraints of the employed workers by zone. These are based on work place employment data by SEG and by car ownership, extracted from the SWS dataset of the 1991 Census. The SEG data accounts for all employed and self-employed persons working in the study area. The data by car ownership include all employed and self-employed persons from households but exclude those living in communal establishments. To avoid data discrepancy the car ownership data is scaled up to the SEG total in each zone, on the assumption that those living in communal establishments have the same car ownership characteristics as the rest of the employed workers working in that zone.
(c) Input to the model derived study area wide constraints which control the household size characteristics of the employed workers. The Census does not provide direct information on the numbers of the employed workers coming from one adult and more-than-one-adult households. However, it does give the total number of one-adult employed households resident in the study area. This is used to derive the total number of employed coming from one adult and more than one-adult households. Since the total number of the employed workers is higher than that of employed residents in the study area, the derivation assumes that the percentage of the employed workers from either type of households working in the study area is the same as those resident there.
(d) Use Census SWS data on employed residents by SEG (4 categories), by car ownership (3 categories), and Census LBS data on employed households by household size (2 categories) to derive segmentation of employed residents required by the model (20 categories by SEG, car ownership and household size). A tri-proportional balancing algorithm is developed which use the Census data as constraints of the 3 dimensions, and the study area wide average segmentation (obtained from the Census SARs data) as the starting matrix. This procedure gives an estimated segmentation of employed residents by zone which is consistent with all Census data that are used
(e) Input initial demand coefficients. Demand coefficients for the employed workers by SEG are extracted from the SARs database for each industry. These demand coefficients are segmented by car ownership, and further by household size, based on the car ownership and household size characteristics of the zone. For Central London zones, its commuting area average segmentation (i.e. the study area excluding Norfolk and Suffolk) is applicable; for the rest of zones the segmentation of the employed residents in their own zone is used. These area wide and zonal segmentations are obtained from step d). The segmented coefficients are input to the model as initial demand coefficients.
(f) Estimate final demand coefficients for 1991. Data derived in all steps above is input to the model. The elastic demand functions developed as part of the MENTOR package are used which adjust the initial coefficients subject to zonal and study area wide calibration constraints. The model is run to convergence which ensures that all constraints on the employed workers and employed residents are met. This means that all known Census employment data (with respective SEG, car ownership and household size segmentation) is reproduced by the model via the demand coefficients. This last step is carried out simultaneously with the spatial distribution calibration of the employed residents; this is because the adjustment of demand coefficients must take account of the segmentation profile of employed residents within the labour market catchment areas (see 3.3.2 below).
The 'jobs' represented by the employed workers at the places of work may attract not only employed residents from the same zone, but also from other zones farther afield. Hence it is necessary to calibrate a spatial distribution model that represents the probabilities of employed residents being attracted to the jobs in each zone.
This spatial distribution model adopts the same logit form as in LASER 2. However, it benefits from more accurately estimated transport costs that are fed from the revised strategic transport model: the representation of both intrazonal and interzonal travel has been substantially improved. Also, new data sources have now become available which significantly improve the model calibration.
The principal data sources used for calibrating the employed residents' distribution model is the Census SWS journey to work matrix, and the NTS distance band based data of trip length distributions for work commuting. The journey to work matrix is that of the total matrix comprising all SEGs on all modes. Persons who have no fixed places of work or whose places of work are undefined are assumed to commute within the zone of residence. This assumption was consistent with the definition of the rest of the SWS data from the 1991 Census. The NTS data was extracted from its database which contained data for 1988-1996. By lumping together the data of these years it was possible to extract a data sample of a reasonable size for each of the 20 employed resident groups, from the records of the survey which were collected within the study area.
Estimation of the spatial distribution model parameters is then performed on the LASER land use model. It is done simultaneously with the segmentation of the employed workers as described in Section 3.3.1. The purpose of the estimation is to determine the concentration parameter of the logit model for each segment of employed residents, the zonal production attractors (which represents the residual influences on employee distribution that are not captured by cost of living and travel disutilities), and residual disutilities of distribution for those zone pairs where the volume of journeys to work is known in 1991. There are a number of steps in the procedure as follows.
(a) The land use model is run using: the initial demand coefficients between industries and the employed workers, all the calibration constraints relevant to the employed workers and the employed residents, and the travel disutilities from the transport model. The concentration parameters for the segments of the employed residents are set at a level which produce mean trip lengths that are close to the NTS observed values for each individual segment. The model generates zonal production attractors during the model run based on constraints on employed residents, but zone-pair residual disutilities are kept with zero values. The model is run to convergence. As the mean trip lengths produced by the model for the employed residents are close to the NTS observed, the labour catchment areas of industries in each zone are approximately correct. The labour catchment areas are only approximately correct because at this stage the Census journey to work matrix has not entered the calibration process and thus the model does not reflect any local particularities at the zone pair level.
(b) The readjusted demand coefficients are output by the model run in a). These coefficients are fed back to the model as second-round coefficients to replace those initial ones. A new land use model run is set up with these second-round coefficients and the concentration parameters are reduced to the level estimated by the NTS parameter exploration mentioned above. During this run the demand coefficients remain fixed. The model is run to convergence
(c) Based on the land use model run in b), a zone-pair level residual disutility estimation program (called DERFR in the MEPLAN package) is run, which estimates a matrix of residual disutilities for each employed residents segment. These ensure that the synthetic matrix entries in the trip distribution matrix match the observed commuting pattern. The inputs to this program are (i) the Census SWS journey to work matrix, and (ii) the production and consumption totals of each segment of employed residents by zone, as produced by run in b).
(d) The residual disutility matrices are fed back into the model to produce a third series of land use model runs. In this run the second-round demand coefficients are made elastic and all calibration constraints are applied to the employed workers and the employed residents. The purpose of this series of runs is to refine the concentration parameters of the spatial distribution model so that the mean trip lengths of the employed residents, and those making employer's business, education, and other private trips match those observed in the NTS. Zonal attractors are estimated during the model runs. The model is run to convergence
(e) The demand coefficients output from d) are extracted and used as fixed demand coefficients for the 1991 run. The complete set of calibrated model parameters include concentration parameters, zonal attractors, and zone-pair residual disutilities, which form part of the inputs to the 1991 and 1997 land use model runs.
As the estimation of travel disutilities is dependent on having a reasonable distribution of trips, the actual calibration process is an iterative one. Initially, an artificial set of link traffic was loaded onto the transport network which approximated the expected level of network congestion on urban, suburban and rural links by link category. Once a preliminary set of distribution parameters was obtained, they were fed back to the land use model to generate the distribution of trips. The level of congestion generated was verified, and a new round of land use run began, and so on. This process was repeated for 3 cycles from step a) to step e). More such cycles may be desirable but because this is extremely running time intensive, resources and time limits have not allowed more to be carried out.
Model verification shows that the model satisfies all calibration constraints, which suggests that it reproduces well the levels of employment and of households by zone by their respective categories.
Table 3.3 Spatial distribution model: e mployed residents by category
|
Code |
SEG |
Household size |
Car owning status |
Concentration parameter |
NTS average (km) |
Modelled average trip length (km) |
|
50 |
SEG1 |
1 Adult |
No car |
0.0420 |
14.70 |
14.99 |
|
51 |
1 Adult |
1 or >1 car |
0.0600 |
18.40 |
19.55 |
|
|
52 |
>1 Adult |
No car |
0.0340 |
20.00 |
18.88 |
|
|
53 |
>1 Adult |
1 Car |
0.0340 |
21.80 |
20.84 |
|
|
54 |
>1 Adult |
>1 Car |
0.0310 |
26.10 |
24.61 |
|
|
55 |
SEG2 |
1 Adult |
No car |
0.0664 |
9.40 |
10.09 |
|
56 |
1 Adult |
1 or >1 car |
0.0600 |
15.40 |
16.05 |
|
|
57 |
>1 Adult |
No car |
0.0460 |
11.20 |
11.23 |
|
|
58 |
>1 Adult |
1 Car |
0.0450 |
14.80 |
13.94 |
|
|
59 |
>1 Adult |
>1 Car |
0.0440 |
15.80 |
15.59 |
|
|
60 |
SEG3 |
1 Adult |
No car |
0.0580 |
8.40 |
8.83 |
|
61 |
1 Adult |
1 or >1 car |
0.0570 |
17.50 |
18.44 |
|
|
62 |
>1 Adult |
No car |
0.0380 |
15.20 |
12.33 |
|
|
63 |
>1 Adult |
1 Car |
0.0380 |
16.10 |
15.61 |
|
|
64 |
>1 Adult |
>1 Car |
0.0350 |
18.00 |
18.44 |
|
|
65 |
SEG4 |
1 Adult |
No car |
0.0507 |
6.90 |
6.84 |
|
66 |
1 Adult |
1 or >1 car |
0.0690 |
11.90 |
12.73 |
|
|
67 |
>1 Adult |
No car |
0.0407 |
8.40 |
8.48 |
|
|
68 |
>1 Adult |
1 Car |
0.0450 |
8.40 |
8.71 |
|
|
69 |
>1 Adult |
>1 Car |
0.0410 |
11.70 |
11.80 |
Table 3.5 reports the concentration parameters used for the spatial distribution models of the employed residents, and the NTS and modelled average distances of travel. It shows that the spatial distribution models perform well in most cases, matching the modelled trips lengths closely against the observed .
The resulting pattern of commuting trips by distance band is shown in Figure 3.3 for the main household segments. The observed data is that from the NTS, and is averaged over the years 1988 to 1996 to minimise the sampling errors. In general the fit of the modelled to the observed is good across the distance bands. It can also be seen that there are more long trips made by the high income group and by those with greatest access to car.
Figure 3.2 Distribution of trip volume by distance band in km - modelled 1991 vs NTS observed: commuting
(The six frames refer to: 1 - SEG1-3 employed residents with no car; 2 - SEG1-3 part car availability; 3 - SEG1-3 full car availability; 4 - SEG4 with no car; 5 - SEG4 part car availability; 6 - SEG4 full car availability)
Table 3.4 compares the total Census journey to work matrix and the modelled movements of the employed residents between the broad regions in the Study Area. Since the Census journeys to work data is defined in terms of a production-attraction matrix in persons whilst the modelled data is output in trips, the modelled data was factored to the same overall total as the Census to facilitate the comparison. The ratios presented in the cells of the table are calculated after the modelled matrix is factored to the same total as the Census table. As the modelled matrix is for the AM peak, the cells to the upper right of the diagonal are affected by the incompatibility between the production-attraction matrix (which does not include any return trips) and a trip matrix (where a small number of return trips are included). This particularly affects Central London, as the model applies an average ratio of return trips to a large inbound trip volume. The cells to the lower left of the diagonal are not much affected by this incompatibility. Out of 18 such cells, only 5 seriously deviate from 1.0. These 5 cells account for only 1.3% of the trip volume out of the 18.
Table 3.4 Ratio comparison of modelled journey to work matrix over 1991 C ensus values
|
Central |
Inner |
Outer |
M25 Area |
Rest |
External |
All Origins |
|
|
Central |
0.98 |
2.13 |
6.45 |
9.45 |
12.62 |
2.12 |
1.53 |
|
Inner |
0.94 |
1.01 |
0.96 |
1.03 |
1.63 |
0.09 |
0.98 |
|
Outer |
1.01 |
0.93 |
1.03 |
0.96 |
1.82 |
0.13 |
1.01 |
|
M25 Area |
0.95 |
0.94 |
0.97 |
1.02 |
1.14 |
0.41 |
1.00 |
|
Rest |
0.96 |
1.04 |
1.25 |
0.99 |
0.98 |
1.24 |
0.99 |
|
External |
2.74 |
0.58 |
0.35 |
0.39 |
0.94 |
- |
0.95 |
|
All Destinations |
1.01 |
0.99 |
1.02 |
1.00 |
0.99 |
0.87 |
1.00 |
Once the employed residents are generated at the zones of residence, the next causal relationship represented in the model is that of household formation by these employed residents. Each employed resident generates either a one adult household (if belonging to a category whose household size is one adult), or part of a >1 adult household (if belonging to a category whose household has more than one adult).
The Census SARs data provides the basis for defining these household formation coefficients. The study area (South East, East of England and London) forms a substantial sized data sample in the SARs. Thus it was possible to use the SARs data directly for the study area, rather than using the national average.
Note in the case of >1 adult households, an employed resident does not always generate households of the same SEG category. This is because households are classified on the basis of the SEG of the head of household (through the Census First Person of the Household criterion), and the constituent adults within such households are not necessarily of the same SEG. The household formation coefficients for these >1 adult households are presented in Table 3.5. Employed residents from one adult households, by definition, always generate only one household that automatically is of the same SEG classification as that adult.
Table 3.5 Household formation coefficients: >1 a dult households
|
Employed Residents |
|||||||||||||
|
52 |
53 |
54 |
57 |
58 |
59 |
62 |
63 |
64 |
67 |
68 |
69 |
||
|
House hold Code |
SEG1 No car |
SEG1 1 car |
SEG1 >1 car |
SEG2 No car |
SEG2 1 car |
SEG2 >1 car |
SEG3 No car |
SEG3 1 car |
SEG3 >1 car |
SEG4 No car |
SEG4 1 car |
SEG4 >1 car |
|
|
72 |
SEG1 No car |
0.498 |
0.045 |
0.012 |
0.018 |
||||||||
|
73 |
SEG1 1 car |
0.495 |
0.096 |
0.019 |
0.049 |
||||||||
|
74 |
SEG1 >1 car |
0.446 |
0.166 |
0.053 |
0.109 |
||||||||
|
77 |
SEG2 No car |
0.083 |
0.388 |
0.041 |
0.047 |
||||||||
|
78 |
SEG2 1 car |
0.046 |
0.303 |
0.032 |
0.053 |
||||||||
|
79 |
SEG2 >1 car |
0.038 |
0.218 |
0.041 |
0.059 |
||||||||
|
82 |
SEG3 No car |
0.052 |
0.137 |
0.586 |
0.122 |
||||||||
|
83 |
SEG3 1 car |
0.033 |
0.119 |
0.464 |
0.134 |
||||||||
|
84 |
SEG3 >1 car |
0.024 |
0.095 |
0.362 |
0.129 |
||||||||
|
87 |
SEG4 No car |
0.076 |
0.164 |
0.103 |
0.563 |
||||||||
|
88 |
SEG4 1 car |
0.024 |
0.069 |
0.037 |
0.326 |
||||||||
|
89 |
SEG4 >1 car |
0.011 |
0.033 |
0.032 |
0.190 |
||||||||
In LASER3.0 the unemployed, retired and other inactive households are treated exogenously. That is, they are input to the model by zone in 1991 using the Census information, and in 1997 and the later policy years using the data from the NTEM database.
The Family Expenditure Survey (FES) is used to derive household consumption coefficients for LASER3.0. Specially commissioned FES tables were obtained which differentiate the patterns of consumption by SEG, household size and the car ownership for the LASER study area. By lumping together data samples for 1992-1999 it was possible to obtain a reasonable sample for each household category. Thus we have directly available the weekly expenditures in terms of
(a) housing
(b) travel expenditure excluding overseas holidays
(c) expenditure on other goods and services
In the model, in 1991 the households are assumed to spend their housing expenditure and expenditure on other goods and services exactly as depicted by FES for each household category. The travel expenditures are replaced by the monetary costs generated by the strategic transport model.
In 1997 the households spend their 1991 housing expenditure plus a congestion rent term which is generated by the dwellings model (see Section 3.5). The travel expenditures of the households are directly estimated from the transport cost matrices generated by the transport model.
At present for simplicity the households do not incur any cost through their demand for endogenous employment. The expenditure on other goods and services in the model completes the simple representation of the cost of living.
The following data on housing has been collected:
(a) The number of dwellings by zone in 1991 (from the Population Census).
(b) The number of dwellings completed for each year 1992-1998 for the majority of local authority districts in the study area; however the demolition and change of use data is not available. This means the total change of dwellings cannot be straightforwardly estimated from this data source. A number of additional estimates at the county and regional levels are collected. A study estimate of the 1998 dwellings by zone was done based on the county and regional estimates as control totals, and the 1991 Census and the housing completion statistics as zonal information.
(c) Average housing expenditure by LASER household category from FES. An extensive review of the literature suggests that housing expenditure may be a more stable and informative data source than the year-on-year house prices, as the observed house prices often include an unstable component of equity in the case of house owners.
Table 3.6 shows the average monthly housing expenditures derived from the FES. This is the average figure for the period 1993-1998, as the data samples are otherwise not large enough to obtain the figures at this level of household segmentation.
Table 3.6 Average housing expenditures by household category
|
Household Code |
SEG |
Household size |
Car owning status |
Monthly expenditure |
|---|---|---|---|---|
|
70 |
SEG1 |
1 Adult |
No car |
£ 546.20 |
|
71 |
|