Road Traffic Model Calibration and Validation Summary Report 4 2mb

Transport for South Hampshire Evidence Base Road Traffic Model Calibration and Validation Summary Report 4

Report for Transport for South Hampshire

August 2011

This report, and information or advice which it contains, is provided by MVA Consultancy Ltd solely for internal use and reliance by its Client in performance of MVA Consultancy Ltd’s duties and liabilities under its contract with the Client. Any advice, opinions, or recommendations within this report should be read and relied upon only in the context of the report as a whole. The advice and opinions in this report are based upon the information made available to MVA Consultancy Ltd at the date of this report and on current UK standards, codes, technology and construction practices as at the date of this report. Following final delivery of this report to the Client, MVA Consultancy Ltd will have no further obligations or duty to advise the Client on any matters, including development affecting the information or advice provided in this report. This report has been prepared by MVA Consultancy Ltd in their professional capacity as Consultants. The contents of the report do not, in any way, purport to include any manner of legal advice or opinion. This report is prepared in accordance with the terms and conditions of MVA Consultancy Ltd’s contract with the Client. Regard should be had to those terms and conditions when considering and/or placing any reliance on this report. Should the Client wish to release this report to a Third Party for that party's reliance, MVA Consultancy Ltd may, at its discretion, agree to such release provided that: (a) MVA Consultancy Ltd's written agreement is obtained prior to such release, and (b) by release of the report to the Third Party, that Third Party does not acquire any rights, contractual or otherwise, whatsoever against MVA Consultancy Ltd and MVA Consultancy Ltd, accordingly, assume no duties, liabilities or obligations to that Third Party, and (c) MVA Consultancy Ltd accepts no responsibility for any loss or damage incurred by the Client or for any conflict of MVA Consultancy Ltd's interests arising out of the Client's release of this report to the Third Party.

Document Control

Project Title: Transport for South Hampshire Evidence Base

MVA Project Number: C39344

Document Type: Road Traffic Model Calibration and Validation Summary

Directory & File Name: J:\C39344_Transport_For_SOUTH_HAMPSHIRE_Model_Suite\MVA_Docs\Re

ports\R4 RTM Calibration And Validation Report\V2\Tfsh_R4_RTM

Calibration And Validation Report_V2a.Doc

Document Approval

Primary Author: Nick Benbow

Other Author(s): Ian Wilkinson, Rehan Mian, Eddie Strankalis, David Carter

Reviewer(s): Ian Burden, David Carter

Formatted by: Sally Watts

Distribution

Issue Date Distribution Comments

0 28/04/11 Steve Williamson Structure for comment with DC revisions

1 10/06/11 Evidence Base Progress Group Draft for comments

2 25/08/11 Evidence Base Progress Group Final Incorporating client comments

Contents

Contents

Road Traffic Model Calibration and Validation Summary Report 4 2

Foreword i

1 Introduction 1.1 1.1 Background 1.1 1.2 Context and Scope 1.1

2 Model Dimensions 2.1 2.1 Introduction 2.1 2.2 Model Areas 2.1 2.3 Zoning 2.2 2.4 Dimensions 2.3

3 Data Sources 3.1 3.1 Introduction 3.1 3.2 Summary of Data Sources 3.1 3.3 Data Collection Locations and Routes 3.3

4 Network Development 4.1 4.1 Introduction 4.1 4.2 Network Coding 4.1 4.3 Quality Assurance 4.4 4.4 Network Calibration 4.4 4.5 Network Validation 4.5

5 Matrix Development 5.1 5.1 Introduction 5.1 5.2 Partial Matrices 5.1 5.3 Trip Ends 5.3 5.4 Synthetic Matrices 5.4 5.5 LGV and HGV Matrices 5.5

6 Calibration and Validation 6.1 6.1 Introduction 6.1 6.2 Matrix Estimation 6.1 6.3 Assignment Process and Convergence 6.2 6.4 Validation Criteria 6.2 6.5 Traffic Flow Validation 6.4 6.6 Journey Time Validation 6.5

7 Fitness for Purpose 7.1

Contents

Road Traffic Model Calibration and Validation Summary Report 4 3

Tables

Table 2.1 Network Structure by Model Area 2.2 Table 3.1 Data Sources 3.1 Table 4.1 Default Main Line Saturation Flows 4.3 Table 5.1 Trip Purpose Segmentations 5.1 Table 6.1 Validation Criterion and Acceptability Guideline 6.3 Table 6.2 Screenline Validation Results 6.4 Table 6.3 Screenline Validation Results 6.4 Table 6.4 Journey Time Validation Results 6.5 Table 6.5 Relaxed Criteria Screenline Validation Results 6.5 Table 6.6 Relaxed Criteria Link Validation Results 6.6 Table 6.7 Relaxed Criteria Journey Time Validation Results 6.6

Figures

Figure 1.1 TfSH Sub-Regional Transport Model 1.2 Figure 2.1 Study Area of the RTM 2.1 Figure 2.2 SRTM Zone system around the Study Area 2.3 Figure 3.1 Location of RSI Sites and Screenlines 3.3 Figure 3.2 Automatic Number Plate Recognition Sites 3.4 Figure 3.3 Location of Calibration Screenlines 3.4 Figure 3.4 Map of Journey Time Assessment Routes 3.5 Figure 3.5 Saturation Flow Survey Junction Locations 3.5 Figure 3.6 Map of Journey Time Assessment Routes 3.6 Figure 4.1 Network Building Process 4.1 Figure 4.2 RTM Network 4.2

TfSH Evidence Base: Sub‐Regional Transport

Model

Foreword The TfSH Steering Group was formed in July 2009 and has met every two months to oversee and direct the development of the TfSH Evidence Base. The Steering Group’s first objective is:

• To develop an Evidence Base consisting of WebTAG compliant analysis and forecasting tools to define current and future problems and develop practical solutions of improvement schemes and interventions to resolve them and achieve local, regional and national objectives;

The development of the Evidence Base is centred on the Sub‐Regional Transport Model (SRTM) and has covered to date:

• Tender specification and appointment of Consultants; • 2010 Data Collection and R1 Report of Surveys; • Development of the SRTM Suite consisting of:

o Main Demand Model, o Local Economic Impact Model, o Road Traffic Model, o Public Transport Model, and o Gateway Demand Model.

This Report documents the development, calibration and validation of one or more of these models. An important objective of the consultant’s specification and subsequent role of the Steering Group has been to achieve the successful development of the SRTM to time and budget, and in accordance with current best practice and guidance laid down by the Department for Transport in its Transport Appraisal Guidance (webTAG). In so far as it has been feasible, the Steering Group is satisfied that the development of the SRTM described in this report meets its first objective. Signatories

1

http://www.dft.gov.uk/

http://www.go-se.gov.uk/

Road Traffic Model Calibration and Validation Summary Report 4 1.1

1 Introduction

1.1 Background

1.1.1 MVA Consultancy and Hyder were commissioned, as part of a wider team, to support

Transport for South Hampshire (TfSH) with the development and application of a Sub-

Regional Transport Model Suite (SRTM) for this nationally important area.

1.1.2 The SRTM will be used to support a wide-ranging set of interventions across the TfSH sub-

region, and is specifically required to be capable of:

forecasting changes in travel demand, road traffic, public transport patronage and

active mode use over time as a result of changing economic conditions, land-use

policies and development, and transport improvement and interventions;

testing the impacts of land-use and transport policies and strategies within a

relatively short model run time; and

testing the impacts of individual transport interventions in the increased detail

necessary for preparing submissions for inclusion in funding programmes within

practical (but probably longer) run times.

1.1.3 This Report describes the development, calibration and validation of the Road Traffic Model

(RTM) within the SRTM.

1.2 Context and Scope

1.2.1 SRTM is a suite of linked models comprising the following components as shown in Figure

1.1:

the Main Demand Model (MDM) which predicts when (time of day), where

(destination choice) and how (choice of mode) journeys are made;

the Gateway Demand Model (GDM) which predicts demand for travel from ports and

airports;

the Road Traffic Model (RTM) which determines the routes taken by vehicles through

the road network and journey times, accounting for congestion;

the Public Transport Model (PTM) which determines routes and services chosen by

public transport passengers; and

an associated Local Economic Impact Model (LEIM) which uses inputs including

transport costs to forecast the quantum and location of households, populations and

jobs.

1 Introduction


Figure 1.1 TfSH Sub-Regional Transport Model

Main Demand ModelMDM

Road Traffic Model RTM

Public Transport Model PTM

HW

Dem

and

HW speeds

Bus Frequency

Gateway Demand Model

GDM

Sub-Regional Transport ModelSRTMLocal Economic

Impact Model LEIM

HW

Gen

Cos

t

PT G

en C

ost

PT D

eman

d

Costs

Port/Airport demand(HW & PT)

Costs

Population & Employment

Main Demand ModelMDM

Road Traffic Model RTM

Public Transport Model PTM

HW

Dem

and

HW speeds

Bus Frequency

Gateway Demand Model

GDM

Sub-Regional Transport ModelSRTMLocal Economic

Impact Model LEIM

HW

Gen

Cos

t

PT G

en C

ost

PT D

eman

d

Costs

Port/Airport demand(HW & PT)

Costs

Population & Employment

1.2.2 The RTM has been developed to represent the base year demand, route choices and costs on

the highway network. In terms of future scenarios, it will represent the network impacts of

different policy and infrastructure interventions.

1.2.3 It is important that the RTM includes the ability to model traffic behaviour at junctions,

including flow metering downstream from bottlenecks as well as blocking-back through

upstream junctions. As such SATURN was selected as the most appropriate software

package to use. SATURN is perhaps the most commonly used highway modelling software in

the UK, benefiting from a large user base, customer support and regular maintenance, and

has been used successfully for many applications since its first release in 1981.


2 Model Dimensions

2.1 Introduction

2.1.1 This chapter summarises the features of the RTM and includes the following sections:

Geographic scope;

Zoning system;

Time periods;

Modelled years; and

User classes.

2.2 Model Areas

2.2.1 The modelled area of the RTM is sub-divided into four regions, shown in Figure 2.1, which differ by zone

aggregation and modelling detail, as follows:

Core Fully Modelled Area (FMA) – the Transport for South Hampshire area (detailed

zoning);

Marginal Fully Modelled Area (detailed zoning);

Buffer Area (zones based on wards); and

External (zones based on districts).

Figure 2.1 Study Area of the RTM

2 Model Dimensions


2.2.2 These area are represented by three levels of network detail, as shown in Table 2.1.

Table 2.1 Network Structure by Model Area

Model

Area

Roads Modelled Network

Type

Modelling Description

Core

FMA

Mways, A roads, B

roads & other high

volume roads

Simulation

network

Junction capacity restraints are explicitly

modelled for priority junctions,

roundabouts, & signalised junctions

considering the interaction of different

movements

Marginal

FMA

Mways, A roads, &

strategic B roads

Speed

/flow

network

Capacity restraint is based on flow delay

curves, where increased flows on a

particular link result in increased travel

times along that link

Isle of

Wight

A roads, & strategic B

roads

Fixed

speed

Fixed speeds are modelled along each

link

Buffer

Area

Mways & strategic A

roads

Fixed

speed


link

External

Area

Skeletal network of

main Mways & A

roads into Buffer Area

Fixed

speed


link

2.3 Zoning

2.3.1 Travel in the model is aggregated into zones which therefore determine the spatial detail

available. The zone system, shown in Figure 2.2, has been defined following current

guidance1 taking account of:

natural barriers (rivers, railways, motorways or other major roads);

areas of similar land use that have clearly identifiable and unambiguous points of

access onto the road network included in the model;

existing zone boundaries, where an existing model is being used as the basis for the

new model (in this case the Solent Strategic Transport Model (SSTM) and Portsmouth

Western Corridor Study (PWCS));

administrative and planning data boundaries (for TfSH zones are aggregations of

Census Output Areas in the fully modelled area and wards elsewhere);

the location of the main parking areas, where town centres are included in the

model; and

the need for internal screenlines for trip matrix validation.

1 Design Manual for Roads and Bridges, Section 12.2.1

2 Model Dimensions


Figure 2.2 SRTM Zone system around the Study Area

2.3.2 With the exception of the LEIM, the same zone systems are used for all components of the

SRTM, and so for consistency with the PTM, catchment areas for rail stations and bus stops,

and fare boundaries are also considered. Additional zones are included for the ports and

airports.

2.4 Dimensions

2.4.1 In accordance with guidance three weekday periods are modelled in the RTM:

AM peak: busiest hour between 0700 and 1000, 38.2% of the 3 hours;

Inter peak: average of 10.00 to 1600; and

PM peak: busiest hour between 1600 and 1900, 35.8% of the 3 hours.

2.4.2 In line with the Main Demand Model the RTM has a base year of 2010, and forecast years of

2014, 2019, 2026 and 2036. In addition LEIM provides forecasts through to 2041.

2.4.3 Demand is split into the following four categories which have significantly different values

and time and vehicle operating costs influencing selection of routes::

Car - Employer’s Business;

Car - Other;

LGVs; and

OGVs.


3 Data Sources

3.1 Introduction

3.1.1 This chapter describes the data used to calibrate and validate the RTM.

3.2 Summary of Data Sources

3.2.1 Data used in the creation of the RTM is summarised in Table 3.1:

Table 3.1 Data Sources

Data Type Data Source

Travel

patterns

Road Side Interviews (RSIs see Figure 3.1):

99 sites surveyed in 2008/9

17 sites surveyed in 2010

self completion postcards issued where face-to-face interviews were

impractical

accompanying 2-way Manual Classified Counts (MCCs) for single day

0700-1900)

accompanying 2-way Automated Traffic Counts (ATCs) for 2 week

periods to allow for day-to-day variation

one-way (as is common best practice)

surveys capture journey purpose, origin and destination postcode, time

of day, vehicle type, and car ownership

Census Journey To Work data from 2001

National Trip End Mode trip rates

Trip end estimates from TEMPRO for Buffer and External Areas

Automatic Number Plate Recognition Survey at 5 sites for motorway traffic

where RSIs are not possible.0630 to 1930 on 6th July 2010. See Figure

3.2. Data collected:

Site location/number and lane description/number;

Direction of vehicle movement;

Registration number;

Vehicle type; and

Time of observation.

Outbound / return trip proportions from the National Travel Survey

3 Data Sources


Data Type Data Source

Counts 2-way single day MCCs on screenlines (see Figure 3.3) between 0700 and

0900). Recorded at 15 minute intervals and classified as:

car;

taxi;

van (car based);

van / Light Goods Vehicle;

HGV 2 axles;

HGV 3 axles;

HGV 4+ axles;

Public Service Bus;

Coach or Private Bus;

Motorcycle / Scooter;

Pedal Cycle; and

Other.

2-way 2 week ATCs at the same screenlines

Journey

times

Moving car observer surveys

12 routes in 2010 (see Figure3.4)

10 or more observations were made for each route in each time period

Also 5 routes from PWCM (2009 – see Figure 3.5)

GPS derived journey times from the TrafficMaster2 for the full modelled

area

Network

specification

Ordnance Survey ITN Geographic Information System

Web based satellite imagery

Traffic Regulation Orders

Local authority signal data (green and inter-green times and staging)

DfT recommended speed / flow relationship from COBA 10

Toll and ferry costs from local authorities

Bus routes and frequencies (for pre-loads) from Local Authorities

Saturation flow surveys in July 2010 to confirm standard signalised junction

capacity assumptions (from LINSIG)

2 Note that TrafficMaster / CJAMS data is now known as Strat-e-gis

3 Data Sources


3.3 Data Collection Locations and Routes

3.3.1 The following figures, referenced in Table 3.1, illustrate the location of interview and count data, and

journey time routes.

Figure 3.1 Location of RSI Sites and Screenlines

3 Data Sources


Figure 3.2 Automatic Number Plate Recognition Sites

Figure 3.3 Location of Calibration Screenlines

3 Data Sources


Figure 3.4 Map of Journey Time Assessment Routes

Figure 3.5 Saturation Flow Survey Junction Locations

3 Data Sources


Figure 3.6 Map of Journey Time Assessment Routes


4 Network Development

4.1 Introduction

4.1.1 This chapter summarises the processes used to construct the highway network, and steps

followed to ensure its fitness for purpose.

4.2 Network Coding

4.2.1 A systematic approach was adopted to develop to RTM network using the data sources

described in Table 3.1. The approach followed the sequential steps shown in Figure 4.1.

Figure 4.1 Network Building Process

Develop road network structure

Define saturation flows

Define link types

Define gap acceptance parameters

Define speed flow curves

Identify junction types

Local Authority traffic signal data

Define core, marginal, buffer & external areas

Identify areas of buffer to be modelled using speed / flow relationships

Derive fixed speeds for the external area

Calculate fixed speeds for links not covered by TrafficMaster data by

calculating average speed by link type

Combine buffer & simulation network information to create whole network

Code vehicle specific banned turns

Code bus route / bus lane information

Code toll value information

Ordnance Survey ITN data

SRTM zoning system

Code number of entry lanes and flaring

Code no entries and one-way linksWeb based portal satellite imagery

Traffic Road Orders

Saturation Flow Surveys

Define location of centroid connectors

Code signalised junctions

TrafficMaster data

Code centroid connectors

Code link types

Code speed flow curves

Code fixed speeds

Code saturation flows

Code gap acceptance parametersExisting Local Models

COBA 10 DMRB V13 S1

Local Authority data

Key Data Definition Coding

4.2.2 The structure of the road network used in the RTM is shown in Figure 4.2.



Figure 4.2 RTM Network

LegendMotorway

Slip Road

A Road - dual carriageway

A Road - single carriageway

B Road

Shopping St

Other

Buffer

Detailed Junction Coding

4.2.3 Within the Core FMA, with the exception of the Isle of Wight, the interaction of traffic at

junctions is modelled in detail. The layout, type of control (eg roundabout, priority or

signals), saturation flows and gap acceptance need to be defined for each junction modelled

in detail. Coding was undertaken in a spreadsheet environment to facilitate checking and

auditing. Junction layouts were determined using web based mapping and aerial

photography.

4.2.4 Traffic signal phasing and times were obtained from the local authorities in Southampton and

the HCC area. For junctions with variable signal times controlled by SCOOT, average green

and inter-green times were obtained for each modelled time period. In the Portsmouth area

signal data were taken from the existing PWCS SATURN model and then checked against

data provided by the local authority. Traffic signal data were not available for around 20% of

signalized junctions. In these cases initial assumptions were made by experienced coders

and SATURN’s SIGOPT facility was then used to optimize timings to flows.

4.2.5 Standard gap and saturation flow parameters for each junction type and movement (e.g. left

turn, straight on, etc) were defined and applied consistently for efficiency and accuracy.

Saturation flow assumptions, shown in Table 4.1, were confirmed as appropriate for the

modelled area by comparison with those obtained from surveys at a sample of 10 junctions.



Table 4.1 Default Main Line Saturation Flows

Turn Left Ahead Right

Main lane – nearside 1764 1940 1764

Main lane – non-nearside 1891 2080 1891

4.2.6 The following gap values have been used for the RTM simulation network;

1.50 seconds for priority junctions;

0.75 seconds for merges; and

1.25 seconds for roundabouts.

4.2.7 These values have been adopted based on practical experience of calibrating and validating

SATURN based sub regional models in the South of England, including the West London Sub

Regional Model and the M25 Highway Assignment Model.

Speeds and Speed / Flow Relationships

4.2.8 Cruise speeds between junctions in the Core FMA were derived from GPS based TrafficMaster

data. TrafficMaster data were associated with the links in the RTM network. For each link

category (defined by road type, number of mid-link lanes, number of stop line lanes, speed

limit and presence of bus lanes) average speeds were calculated from all TrafficMaster

observations for that category. The averages were calculated such that links with high

standard deviations for speeds received a lower weight (and had less influence on the

average) than links with low standard deviations. Analyses of the data indicated that the

cruise speeds did not vary significantly in different time periods. Therefore the same cruise

speeds were used in all three periods.

4.2.9 Within the Marginal FMA link speed/flow relationships were applied, i.e. the speed on the link

is a function only of the flow on the link. Ninety-nine curves were calculated using the

COBA10 formulae depending on speed limit, road type and number of lanes. Speed/flow

curve shapes are in theory weakly affected by the proportion of HGVs using the link but

SATURN does not include functionality to model this effect during the assignment process.

Default COBA10 HGV proportions were used in the calculation of speed/flow curves as

recommended by Highways Agency.

4.2.10 In the External area and on the Isle of Wight average speeds for each link were calculated

from the TrafficMaster data using the same approach adopted for cruise speeds in the FMA.

These speeds do not vary with flow within the assignment process as the model does not

include the full quantum of traffic in External area or Isle of Wight, but only traffic travelling

to or from the TfSH area (or to/from the Isle of Wight).

Centroid Connectors

4.2.11 Centroid connectors are the links used to load traffic to/from each zone onto the network.

These were defined using professional judgement with reference to aerial photography to



determine where traffic was likely to join the modelled road network. In most cases in the

Core FMA notional links (known as spigots) are added joining the network between real

junctions representing multiple loading points such as driveways, car parks and side roads.

In a small number of instances real junctions were added as loading points where they have

a significant influence on the performance of the network.

4.3 Quality Assurance

4.3.1 Use of standardised and automated processes and assumptions to code networks, as

described above, is a key aspect of the quality assurance processed used for RTM. Further

checks were carried out using the following checklist:

check for appropriate junction types;

check that the appropriate number of entry lanes have been coded and that flaring of

approaches, where appropriate, are accounted for;

check that turn restrictions have been correctly identified (these may vary by time

period);

check that one-way roads and no entries have been correctly specified;

check that saturation flows are appropriate (particularly if turn rates appear

excessively high or low compared to straight ahead);

check that link lengths, link types and cruise speeds for both directions of a link are

consistent, and that the link type and cruise speed coding does not vary unjustifiably

along a series of links;

compare crow-fly link lengths against actual lengths and check that the coded link

lengths in the core modelled area for links greater than 500m in length are not

greater than 1.3 times the crow-fly distance, and inspect links which fall outside this

range;

checking the connectivity of the network to ensure that there are no gaps or isolated

areas; and

investigation of errors and warnings output by SATURN.

4.4 Network Calibration

4.4.1 Network calibration refers to the process of adjusting the network to remove any coding

errors or to fine tune standard assumptions to better represent individual links and junctions.

This step is undertaken following matrix construction (see Chapter 5) and initial assignment

(Chapter 6). Checks were made of speeds and flows where those observations (discussed in

Chapter 3) were available. Comparison of modelled and observed speeds were used to

identify junctions with excessive delay which could be due to incorrect coding. Flows were

compared with capacities to identify junctions where the capacity is lower than the traffic

count.

4.4.2 SATURN outputs statistics describe the convergence of flow and delay at junctions between

assignment iterations. Poor convergence can be a symptom of unrealistic junction coding.



4.4.3 Remedial actions to address poor match to observed data included adjusting saturation flows,

flare lengths (where coded) and gap acceptance, and reviewing the location of centroid

connectors. Adjustments were only made when justified by the conditions at the problematic

junction or link.

4.5 Network Validation

4.5.1 Network validation consisted of checking modelled routes. No data was available against

which to compare model outputs, rather local knowledge and judgement was required. Pairs

of zones were selected using the criteria listed below for which routes were reviewed:

relate to significant numbers of trips;

are of significant length or cost (e.g. 20+ minutes);

pass through areas of interest (e.g. scheme impacted areas);

include both directions of travel (to sense check differences);

link different compass areas (e.g. north to south, east to west, etc.); and

coincide with journey time routes as appropriate.

4.5.2 Route checking included:

making a judgement on the plausibility of the modelled route;

comparing routes for uncongested networks with congested networks;

comparing routes for different time periods; and

comparing routes for the same journey in opposite directions.

4.5.3 Where the comparisons listed above indicated different routes choices, this could have been

due to legitimate congestion effects but could also have been symptomatic of coding errors,

which could then be corrected.

4.5.4 An automated process was developed to identify any instances where the path between a

zone pair used the same node more than once. This can be a legitimate result, e.g. where

there is a banned turn, but such instances were checked and a number of corrections made.


5 Matrix Development

5.1 Introduction

5.1.1 This section describes the methodology for the development of the base year trip matrices.

These matrices were later subjected to matrix estimation as part of the process of calibrating

the model; the matrix estimation process and results are reported in Chapter 6. The

matrices described in this section are referred to as ‘prior’ matrices.

5.1.2 The demand matrices contain estimates of all travel to, from and within the Core and

Marginal FMAs. The only trips included to and from the Buffer and External areas are those

which traverse the FMAs.

5.1.3 Demand segments used during the assignment were described in Chapter 2. These

segments were defined to group demand based on the factors that influence route choice.

Matrices were developed with a more disaggregate set of purposes than those used for use

in the MDM, as shown in Table 5.1.

Table 5.1 Trip Purpose Segmentations

Vehicle

Type Abbr. OD Demand Matrices

RTM Assignment

Matrices

Car HBB HB Employers Business Employers Business

Car NHB Non HB Employers Business

Car HBW HB Work

Commuting and Other Car HBE HB Education

Car HBO HB Other

Car NHO NHB Other

LGV LGV Light Goods Vehicles LGVs

HGV HGV Other Goods Vehicles OGVs

5.1.4 The remainder of this chapter is structured as follows:

processing of RSIs to produce partial observed matrices;

calculating trip ends of total demand to and from each zone;

producing synthetic matrices for the whole modelled area using gravity models,

which are then used for zone-pairs that are not captured by the RSIs; and

calculation of LGV and HGV matrices.

5.2 Partial Matrices

5.2.1 The partial matrices were developed from the RSI data described in Chapter 3. The following

steps were followed to develop the partial matrices:



data checking and cleaning;

expansion of surveys to traffic counts;

transposition of data, which was collected in one direction only, to estimate travel in

the reverse direction; and

adjustments to account for instances where movements between zone-pairs would

be observed at more than one survey location.

5.2.2 The first stage of the process was to check the RSI data to identify illogical movements where the

crow-fly distance via the survey site was significantly higher than the direct origin-destination

distance; and where address information was incomplete.

5.2.3 As shown in Figure 3.1 RSI sites were arranged into screenlines and cordons which enclosed

areas of the model. Any observations of movements that crossed an enclosure were

removed as we could not be sure whether some trips between the same zone-pairs would

have used other routes which circumvented the enclosure.

5.2.4 Secondly expansion factors were calculated for each RSI location. These factors are the ratio

of classified traffic count to the number of interviews, and were used to weight the observed

data to match the total observed traffic. For each site factors were calculated for every

combination of vehicle type and hour.

5.2.5 Estimates of trips made in the opposite direction of the surveys were made as follows:

origins and destinations of forward direction surveys were transposed;

profiles calculated from the National Travel Survey (NTS) were calculated and applied

to determine the proportion of from-home trips that returned in each subsequent

modelled period;

similarly NTS data were used to determine the proportion of to-home trips that left

home in each preceding modelled period; and

expansion factors were derived to control the total number of reverse direction trips

in each hour to the corresponding MCC.

5.2.6 For some zone-pairs journeys could be observed at multiple RSI sites and therefore could be

double counted. Arranging RSIs to form enclosed areas facilitated application of the following

rules to adjust for multiple observations:

Forward direction data only were used for journeys which started in one enclosed

area and finished in another – these forward direction data are considered more

reliable than those estimated by transposition;

Trips which started in one enclosure, passed through a second, and terminated

outside any enclosure were not double counted as they were removed as described

in paragraph 5.2.3; and

Where a trip could have been observed twice in the forward direction or twice in the

reverse direction the estimates were each halved if they were of a similar magnitude.

If they were not of a similar magnitude the data were inspected to determine the



most robust source, either based on the sample size or completeness and relevance

of the enclosure.

5.3 Trip Ends

5.3.1 The home-based purpose origin/destination person trip ends for zones within the FMA were

produced using the following steps:

Home-based production trip ends were estimated for all FMA zones by applying the

NTEM production trip rates to the population data. These trip ends represent the

‘outbound’ trip only;

Home-based attraction trip ends within the FMA were estimated by applying the

NTEM trip attraction trip rates to the employment data, and scaling total attractions

to match total productions for each purpose, mode (including active modes), time

period and car availability across the FMA;

The Outbound/Return factors were used to calculate the ratio of from-home and to-

home trips in each time period; these ratios were used to generate return trip ends

from the NTEM-based outbound trip ends;

Census Journey to Work data by zone was used to adjust the mode shares (including

active modes) for the home-based production/attraction trip ends so that they reflect

the accessibility of particular zones to public transport services. This was necessary

because NTEM-derived trip ends were not deemed representative at the sub-NTEM-

zone level; and

Origin/Destination trip ends were then derived from the production/attraction trip

ends by re-applying the Outbound/Return factors.

5.3.2 The non-home-based purpose origin/destination trip ends for zones within the FMA were

developed using home-based to non-home based trip rate factors derived from National

Travel Survey (NTS) data.

5.3.3 DfT’s TEMPRO software (version 5.4) was used to output 2010 origin/destination trip ends by

purpose, mode and time period for all zones outside the FMA. TEMPRO was also used to

generate car availability splitting factors for these zones.

5.3.4 An analysis of the NTEM trip ends, which are supposed to represent the full demand, showed

that there were actually less non home based employer’s business trips in the NTEM trip ends

than in the in the partially observed matrices. The NTEM trip ends were therefore adjusted to

increase the number of NHB EB trips, firstly to match the level of trips in the partial matrices,

and secondly to account for unobserved (e.g. intra-sector) movements not included in the

partial matrices.

5.3.5 In addition the trip ends were further adjusted, at a sector level, to ensure that they were

compatible with the synthetic matrices at the various enclosures.



5.4 Synthetic Matrices

5.4.1 Comprehensive estimates of all travel demand were developed and subsequently data

overwritten for zone-pairs captured in the observed partial matrices. In this way a prior

matrix was developed using observations where available and synthetic estimates elsewhere.

Three different approaches were applied to develop the synthetic matrices depending on the

available data:

a logit destination choice model (DCM) was calibrated and applied to estimate trips

to, from and within the Core FMA;

2001 Census Journey-to-Work data were adjusted to trip ends obtained from

TEMPRO for the Marginal FMA; and

ANPR number plate matching data were used for demand which passed through the

FMA.

5.4.2 The DCM predicts the probability of a trip from an origin travelling to each destination as a

function of the relative generalised costs of travel to each destination. This parameter which

determines the sensitivity of the DCM to generalised cost differences was calibrated using

destination choice profiles from the partially observed matrices and preliminary generalised

cost estimates from the network model. Separate models were calibrated for each time

period and purpose combination. Destination specific constants, which adjust the generalised

cost of travel for all trips to a destination by the same amount, were included in the models

to ensure that the total of distributions matched the trip ends calculated as described above.

Generalised cost adjustments were also calibrated so that the DCM matched observed

demands between RSI enclosures for which there were at least 10 survey records.

5.4.3 A number of checks were made of the outputs of the DCM:

plot trip cost distributions (TCDs) for zone-pairs which were observed were checked

to ensure that the model closely replicated observed TCDs;

checks that the trip-end and enclosure-to-enclosure constraints were met; and

the difference between the full synthetic matrix TCDs was compared with the TCDs

for zone-pairs that were observed which showed that the synthetic matrices included

more short distance trips, e.g. those which would not cross an RSI enclosure.

5.4.4 The DCM was not used for trips to and from the Marginal FMA, except those which travelled

into or out of the Full FMA, as modelled generalised costs were considered to be less reliable.

Rather trip distributions were extracted from the 2001 Census Journey-to-Work dataset and

adjusted to match the trip ends derived as discussed above.

5.4.5 Observations from the ANPR sites were used to estimate matrices of travel though the FMA.

Number plate records were matched between ANPR sites to determine a site to site matrix of

travel. Where the travel time between sites was greater than 75 minutes the records were

excluded on the basis that the vehicle was likely to have stopped within the TfSH area.

Adjustments were made to control the number of number plate observations to counts from

the Highways Agency’s TRADS dataset. It was then necessary to allocate journeys observed

at each ANPR site between external zones. Assumptions were made about which external



zones would use each ANPR site and trips distributed between the zones based on

proportions calculated from trips to and from the FMA.

5.5 LGV and HGV Matrices

5.5.1 LGV and HGV partial matrices were developed using the same process applied for cars. At

some RSI sites the number of goods vehicle observations were very low and so these were

grouped prior to expansion to reduce sample bias. NTS does not include information on

outbound/return factors for goods vehicles and therefore we assumed that the return trip was

in the same period as the outbound trip. Trip rates were not available to develop goods

vehicle trips ends, and so no trip ends were available to apply a DCM to. Rather ratios of LGV

and HGV trips to home based work and other car trips for each time period were calculated

and applied to the prior car matrices.


6 Calibration and Validation

6.1 Introduction

6.1.1 This chapter describes:

trip matrix estimation, including checks of significance of differences between prior

and estimated trip matrices;

the assignment process applied using SATURN and model convergence;

validation criteria set out in DfT and HA guidance;

validation of traffic flows; and

validation of journey times.

6.2 Matrix Estimation

6.2.1 Matrix estimation is an automated process where the number of trips between each zone-pair

can be adjusted to improve the match between assigned flows and traffic counts. The

primary purpose of matrix estimation is to refine estimates of trips not intercepted in surveys

and which have therefore been synthesised. This is why counts on screenlines independent

of the roadside interview cordons and screenlines are required.

6.2.2 SATURN’s ME2 can be used to match flows on individual links or total flows on groups of links

(screenlines or mini-screenlines). For TfSH counts have been grouped into mini-screenlines

wherever possible to reduce scope for inappropriately modifying the matrix to compensate for

network coding simplifications or errors. Locations of count sites are shown in Figures 3.1 to

3.3. Counts used in matrix estimation were derived from two-week ATCs split by vehicle

class using proportions calculated from MCC data.

6.2.3 ME2 requires modelled paths between each zone pair to be input. These paths are a function

of travel times and delays which in turn are influenced by the scale and distribution of trips in

the matrix. For this reason we have applied an iterative process where an initial set of paths

is input to ME2; the resulting matrix is then assigned to produce refined paths; these paths

are then input to a further run of ME2, etc. A total of six iterations of assignment and

estimation have been run.

6.2.4 The XAMAX parameter has been set to 2.5 to limit change in trips for each zone-pair to a

factor of 2.5 or 0.4 (1/2.5).

6.2.5 A number of checks and analyses have been undertaken to assess the reasonableness of the

matrix estimation process:

for cars and LGVs matrix totals change by 7.5% or less;

for HGVs, which account for about 2% of total trips and for which there was relatively

little observed data, matrix totals change by up to 35.5%;

linear regression analysis, comparing pre- and post-ME2 trip end totals, show a

strong correlation with R2 values in excess of 0.92 in all cases, and gradients are all



between 1.02 and 1.09 indicating that there has generally been a modest increase in

trips;

plots of trip cost distributions were compared which showed that the distribution of

trips was not unduly distorted;

mean trip lengths changed by no more than 17%; and

sectored matrices were compared which confirmed that ME2 did not distort the

distribution of trips.

6.3 Assignment Process and Convergence

6.3.1 The deterministic user equilibrium method implemented in the SATURN software is used.

This assumes that users have perfect knowledge of the time taken to pass through the

network from their origin to destination.

6.3.2 SATURN measures the convergence, i.e. how speeds and flows change on links, between

iterations. For model outputs to be stable and reliable, in particular for use in social cost

benefit appraisal, a high level of convergence is required. For the RTM we have ensured

much tighter convergence than suggested by guidance (DMRB) because our experience tells

us that the DMRB standards are insufficient to ensure that model “noise” is smaller than the

benefits which are being measured.

6.3.3 RTM SATURN is configured to stop when the %Gap measure falls below 0.1%. For the base

year models this equates to flow and delays changing by 1% or less on 99% of links for four

consecutive iterations.

6.3.4 The value of time and vehicle operating cost coefficients used for assignment were derived

from TAG Unit 3.5.6.

6.4 Validation Criteria

6.4.1 Guidance on validation of highway assignment models is provided in the Highways Agency’s

Design Manual for Roads and Bridges and summarised in Table 6.1.



Table 6.1 Validation Criterion and Acceptability Guideline

Criteria DMRB Acceptability Guideline

1. Differences between modelled flows and counts on

screenlines should be less than 5% of the counts

All or nearly all screenlines

2. GEH statistic based on modelled flows and counts

should be less than 4

All or nearly all screenlines

3. Individual link flows within 15% of counts for flows

from 700-2700 veh/h

Individual link flows within 100 veh/h of counts for

flows less than 700veh/h

Individual link flows within 400 veh/h of counts for

flows more than 2700 veh/h

> 85% of cases

4. GEH < 5 for individual link flows > 85% of cases

5. Modelled times along routes should be within 15%

of surveyed times (or 1 minute, if higher)

> 85% of routes

6.4.2 The GEH statistic which is a form of the Chi-squared statistic that incorporates both relative

and absolute errors, and is defined as follows:

))(5.0()( 2

CMCMGEH+×

−=

where:

M is the modelled flow; and C is the observed flow.



6.5 Traffic Flow Validation

6.5.1 Screenline validation results summarised in Table 6.2 show a good fit between modelled and

observed traffic at over 80% of screenlines meeting criteria 1 in Table 6.1, and over 74%

meeting criteria 2.

Table 6.2 Screenline Validation Results

Time Period Screenlines within 5% GEH < 4

AM Peak 64 of 80 80% 59 of 80 74%

Inter Peak 70 of 80 88% 69 of 80 86%

PM Peak 64 of 80 80% 62 of 80 78%

6.5.2 Validation results for individual links are shown in Table 6.3.

Table 6.3 Individual Link Validation Results

Time Period Individual Link Flows within

% or veh/hr of criteria*

GEH < 5 for individual link

flows

AM Peak 54% 47%

Inter Peak 65% 56%

PM Peak 54% 45%

*Individual flows within 15% of counts for flows from 700 to 2700 veh/h

Individual flows within 100 veh/h of counts for flows less than 700 veh/h

Individual flows within 400 veh/h of counts for flows more than 2700 veh/h

6.5.3 The individual link flow validation is not as good as the screenline validation. However as

discussed in more detail in paragraph 6.7.2 it is more critical that the screenline validation

performs well, as it ensures overall demand is correct, whereas individual link flow validation

is less significant as routing can vary from day to day.



6.6 Journey Time Validation

6.6.1 Journey time validation results are summarised in Table 6.4.

Table 6.4 Journey Time Validation Results

Time Period Routes within 15% or 1 minute if higher

AM Peak 21 out of 34 62%

Inter Peak 23 out of 34 68%

PM Peak 17 out of 34 50%

6.6.2 Detailed investigation of journey time validation results by route showed that the slope of the

observed and modelled journey times are generally similar and that the model representation

of observed conditions on the surveyed network is relatively accurate. These profiles are

therefore a good indication that the model is capturing network delay accurately and that

there is no systematic bias towards either low or high speeds.

6.7 Relaxed Validation Criteria

6.7.1 It is often considered that the WebTAG thresholds of acceptability are more suited to less

complex models, and as such it is argued that a certain level of flexibility is acceptable given

the scale and complexity of the SRTM. Tables 6.5 to 6.7 therefore presents the equivalent

results as shown in Tables 6.2 to 6.4, with slightly relaxed criteria, more appropriate for a bi-

centric, sub-regional area. The results demonstrate that although the WebTAG criteria are

not fully met, the model performance is in general good.

Table 6.5 Relaxed Criteria Screenline Validation Results

Time Period Screenlines within 10%

AM Peak 70 of 80 88%

Inter Peak 74 of 80 93%

PM Peak 69 of 80 86%



Table 6.6 Relaxed Criteria Link Validation Results

Time Period Individual Link Flows within % or veh/hr of criteria*

AM Peak 75%

Inter Peak 86%

PM Peak 75%

*Individual flows within 20% of counts for flows from 700 to 2700 veh/h

Individual flows within 200 veh/h of counts for flows less than 700 veh/h

Individual flows within 400 veh/h of counts for flows more than 2700 veh/h

Table 6.7 Relaxed Criteria Journey Time Validation Results

Time Period Routes within 25% or 1 minute if higher

AM Peak 28 out of 34 82%

Inter Peak 28 out of 34 82%

PM Peak 31 out of 34 91%

6.7.2 Tables 6.5 to 6.7 demonstrate that the model performance is in general good, and that the

screenline validation performs particularly well. This is critical, as of the three validation

measures the matrix validation screenlines are of particular importance, as discussed below:

Matrix Validation – Highly important, as it ensures the demand in the model is

correct for assessing interventions and future changes;

Link Flow Validation – Less significant at an individual link level, because the

dense network provides alternative routes of similar generalised cost, and routing

can vary from day to day; and

Journey Times Validation – Also less crucial because journey times can vary, and

it is more important that changes can be represented in the model both within mode

and relatively between modes.


7 Fitness for Purpose

7.1.1 The SRTM model system covers a wide geographic area and contains a significant number of

strategic motorways, primary routes and complex urban road networks. An unusual feature

of the model is that it includes two main conurbations, Southampton and Portsmouth,

significant district centres such as Fareham and Gosport, a number of peninsulas, and a third

geographically distinct centre on the Isle of Wight. More typically traffic models are

developed for either single corridors, free-standing cities or conurbations. The strategic

validation of the Road Traffic Model needs to be considered in this context, i.e. a model of

multiple, often parallel, corridors and multiple centres that generate urban and inter-urban

trips combined with strategic road access routes using the Motorway and trunk road network.

7.1.2 The model has been constructed according to WebTAG recommendations, and validated

against DMRB guidelines. The calibration process did not reveal any significant problems or

shortcomings in the base year model. The quality of validation of the model is in general

good, with the screenline validation performing particularly well. This is critical, as it ensures

the demand in the model is correct for assessing multi-modal interventions and future

changes.

7.1.3 The journey time validation and the patterns of junction delay appear consistent and

plausible, although the link flow and journey time validation do not meet the WebTAG

criteria. However, these recommended criteria mask a good model performance that is close

to the meeting the acceptability guidelines.

7.1.4 The calibration and validation suggest that the model is fit for the purpose of representing

the highway traffic patterns in the base year, as part of the SRTM.

7.1.5 The model encompasses a large geographic area at different levels of detail and is expected

to be used to consider a range of strategic and specific interventions, e.g. representing the

main highway movements, the impact of major highway and public transport interventions on

those movements, and providing controlled and consistent inputs to local or more detailed

models.

7.1.6 It is acknowledged that whilst fit for general purpose, depending on the nature and scope of

the intervention being tested, additional local validation checks may be beneficial for model

application for specific interventions at a local level.

MVA Consultancy provides advice on transport, to central, regional and local government, agencies, developers, operators and financiers. A diverse group of results-oriented people, we are part of a strong team of professionals worldwide. Through client business planning, customer research and strategy development we create solutions that work for real people in the real world. For more information visit www.mvaconsultancy.com

Email: [email protected]

Offices also in

Bangkok, Beijing, Hong Kong, Shenzhen and Singapore

Abu Dhabi

AS Business Centre, Suite 201, Al Ain Road, Umm al

Nar, P.O. Box 129865, Abu Dhabi, UAE

T: +971 2 510 2402 F: +971 2 510 2403

Birmingham

Second Floor, 37a Waterloo Street

Birmingham B2 5TJ United Kingdom

T: +44 (0)121 233 7680 F: +44 (0)121 233 7681

Dublin

First Floor, 12/13 Exchange Place

Custom House Docks, IFSC, Dublin 1, Ireland

T: +353 (0)1 542 6000 F: +353 (0)1 542 6001

Edinburgh

Second Floor, Prospect House, 5 Thistle Street,

Edinburgh EH2 1DF United Kingdom

T: +44 (0)131 220 6966 F: +44 (0)131 220 6087

Glasgow

Seventh Floor, 78 St Vincent Street

Glasgow G2 5UB United Kingdom

T: +44 (0)141 225 4400 F: +44 (0)141 225 4401

London

Second Floor, 17 Hanover Square

London W1S 1HU United Kingdom

T: +44 (0)20 7529 6500 F: +44 (0)20 7529 6556

Lyon

11, rue de la République, 69001 Lyon, France

T: +33 (0)4 72 10 29 29 F: +33 (0)4 72 10 29 28

Manchester

25th Floor, City Tower, Piccadilly Plaza

Manchester M1 4BT United Kingdom

T: +44 (0)161 236 0282 F: +44 (0)161 236 0095

Marseille

76, rue de la République, 13002 Marseille, France

T: +33 (0)4 91 37 35 15 F: +33 (0)4 91 91 90 14

Paris

12-14, rue Jules César, 75012 Paris, France

T: +33 (0)1 53 17 36 00 F: +33 (0)1 53 17 36 01

Woking

Dukes Court, Duke Street, Woking

Surrey GU21 5BH United Kingdom

T: +44 (0)1483 728051 F: +44 (0)1483 755207

Documents

Road Traffic Model Calibration and Validation Summary Report 4 2mb