Microsimulation Model Design in Lower Manhattan: A Street Management Approach

Microsimulation Model Design in Lower Manhattan: A Street Management Approach

Varanesh Singh
Arup
155 Avenue of the Americas, New York, NY 10013
212-896-3115
Varanesh.Singh@arup.com

S. Brian Huey
Arup
212-896-3196
Brian.Huey@arup.com

Trent Lethco
Arup
212-896-3265
Trent.Lethco@arup.com

Peter Dunn
Arup
Level 17 1 Nicholson St, Melbourne Vic 3000
3-9668-5452
Peter.Dunn@arup.com.au

Suchi Sanagavarapu
New York City Department of Transportation
40 Worth St., Room 1012
New York, NY 10013
212-788-2128
ssangavarapu@dot.nyc.gov

Submitted for Presentation and Publication
88th Annual Meeting
Transportation Research Board
Submitted July 31st, 2008

WORD COUNT: 4,310 Words + 3 Figures + 5 Tables = 6,310 Total

Singh, Huey, Lethco, Dunn, Sangavarapu 1

ABSTRACT
Microsimulation models are an invaluable tool for transportation professionals who evaluate and
analyze network-level transportation impacts. Modeling dense urban street networks like central business
districts present significant challenges due to their size, density and complexity. In 2004, the Lower
Manhattan Development Corporation (LMDC) funded the New York City Economic Development
Corporation (NYCEDC) and the New York City Department of Transportation (NYCDOT) to contract
Arup to develop a microsimulation model of Lower Manhattan. This paper describes the design,
calibration and validation procedures of a Q-Paramics microsimulation traffic model of Lower Manhattan
in New York City. Lower Manhattan is the fourth largest central business district in the United States and
one of the oldest and densest areas in New York City. It contains some of the highest levels of pedestrian,
transit and automobile activity in America. As a result, the modeling process must account for a variety of
complex urban issues that are atypical in most microsimulation models.
An extensive, multi-modal data collection effort was conducted to create a detailed set of data,
which was then applied to the model design process. A street management framework was used to guide
the development of the network and address issues of vehicle assignment and route choice. The model
also addressed issues associated with vehicle interactions in high pedestrian flows intersections,
disparities in driver types, taxi maneuvers, delivery vehicles and other activities unique to central business
districts.


ACKNOWLEDGEMENTS
This project is made possible by a grant from the Lower Manhattan Development Corporation,
which is funded through Community Development Block Grants from the U.S. Department of Housing
and Urban Development.
The authors would like to thank the following firms and individuals who have provided extensive
support to the overall project. These include: Andrew Salkin, Joshua Kraus, Josh Rosenbloom, Luis
Sanchez, Meghann Rowley, Steven Weber (NYCDOT); Phil Plotch (LMDC); Venetia Lannon, Joan
McDonald, Michael Taylor (NYCEDC); Ken Hausman and Matt Jukes (StumpHausman); Umesh
Avadhani (B-A Engineering). We would like to thank our current and former colleagues who have been
extensively involved in the project: Andrew Wisdom, Daniel Peterson, Jonathan Drescher and Tim
Bryant.


INTRODUCTION
Microsimulation models are an invaluable tool for transportation professionals who evaluate and
analyze network-level transportation impacts. Modeling dense urban street networks like central business
districts present significant challenges due to their size, density and complexity. Effective management of
the transportation network is a critical element in the redevelopment and long-term viability of New
York’s Lower Manhattan Central Business District (FIGURE 1). As part of an effort to develop tools that
will allow the City to assess the transport impacts of development, street closures and changes to the road
network, the Lower Manhattan Development Corporation (LMDC) funded the New York City Economic
Development Corporation (NYCEDC) and the New York City Department of Transportation (NYCDOT)
to contract Arup to undertake a multi-year effort to develop a microsimulation model of Lower Manhattan
using Quadstone’s Paramics (Q-Paramics) microsimulation software.
Lower Manhattan is the fourth largest central business district in the United States, behind
Midtown Manhattan, Chicago and Washington D.C. It is New York’s fastest growing residential
neighborhood, seeing a 145% increase in residential population since 2001. Lower Manhattan had 8.1
million visitors in 2003 compared to 8.5 million in Midtown Manhattan and 1.1 million in Chicago
(1,2,3)
From a modeling perspective, Lower Manhattan presents a number of significant challenges. The
size of the network means that the number of potential route choices for any trip is high. The level of
demand and small block sizes mean that congestion develops quickly, making the model operation more
sensitive to small changes in demand. Compounding all of this is the need to model interactions between
vehicles and pedestrians, livery vehicles and goods delivery operations. As a result, the development of
the Lower Manhattan simulation model addressed a variety of urban issues that typically don’t exist in
freeway or corridor models.
This paper focuses on the practical solutions that were developed in order to achieve a validated
model. It begins with the multimodal data collection and literature review process. Network design issues
are presented, specifically focusing on issues germane to urban modeling. Lastly, validation criteria and
results are presented and commented on.

Previous Studies
While there are no standardized guidelines for microsimulation modeling in New York City, there
have been several recently produced guidelines for the design and calibration of microsimulation models
in America, Australia and the U.K. (4,5,6). These documents provide general guidance concerning
scoping, data collection, base development, error checking and calibration. They do not provide many
specific recommendations on issues pertaining to urban environments. Dowling and Skabardonis show
that a practical, top-down approach to the calibration stage can produce well calibrated models. This
approach was taken, with specific phases of the approach being elaborated on in this paper. However, the
process is also general, and does not address software specific issues with Paramics (7). Several
documents provide specific calibration and validation criteria, which informed the calibration and
validation criteria developed in this study (8 9,10,11).

DATA COLLECTION
An extensive data collection effort was conducted between 2003-2007, attempting to capture
seasonal differences, multiple modes, parking and curbside activity.

Counts
The major component of the data collection effort was the turning movement counts. Counts were
conducted at approximately eighty key intersections within the study area during the fall of 2006. The
counts were recorded in 15-minute intervals between the hours of 7:00-9:00 AM and 4:00-6:00 PM. The


counts were classified based on vehicle types. In addition, automatic traffic recorder counts were
conducted on highways where human observation was not possible.

Pedestrians
Pedestrian counts were collected for 22 intersections within the study area in order to capture
vehicle delay resulting from high pedestrian movements. Pedestrians were counted by direction at each
crosswalk for during the hours of 6:00-10:00 AM and 3:00-7:00 PM on a Tuesday, Wednesday or
Thursday in early November, 2006.

Parking
Off-street parking surveys were taken in various parking lots throughout the study area to gain a
better understanding of the temporal flows into and out of parking lots during the AM and PM peak hour.
A better understanding of parking lot flows was essential because they represent a major source/end of
trips within the internal study area.
The counts were recorded in 15-minute intervals between the hours of 6:00-10:00 AM and 3:00-
7:00 PM on a Tuesday, Wednesday or Thursday in January, 2007. The counts classified private
automobiles, for hire vehicles and commercial vehicles.

Livery Vehicles
The primary source of taxi demand information was traffic surveys. While traffic surveys provide
an indication of the level of taxi activity, no information was available regarding travel characteristics
through the network. A future goal is to collect more detailed information about taxi routes and activity
with the cooperation of the taxi industry.

Travel Time Surveys
Travel time surveys were taken along ten different routes within the study area. These routes were
selected to allow for comparison of observed and modeled travel times along specific corridors or districts
during the validation stage.
The travel time surveys were conducted using a floating car technique where a two-person team
of surveyors drove the routes at the prevailing speed of traffic while recording the elapsed time between
pre-determined control points such as the center of an intersection. The surveyors would also record the
reason and length of time for each stoppage along the route. The reasons for stoppage included
congestion, signal delay, curbside activity, incident or construction.
Travel time surveys were conducted between 7:00-9:00 AM and 4:00-6:00 PM on a Tuesday,
Wednesday or Thursday in November, 2006. The number of test runs captured in each survey session was
dependent on the route length and traffic conditions. On average, three to four runs were captured for
each travel time route for each time period.

Curbside Parking
Sample on-street surveys were used to understand curbside activity (double parking, picking up,
and dropping off) on typical street blocks. These surveys were performed during peak hours and took into
account the type of vehicle that stopped along the curb, the arrival time and the departure time.

Other Data
The above data was augmented by a series of site visits to assess actual vehicle behavior and
operation. These qualitative assessments helped inform the visual audits of the model network. Other
traffic data made available by various authorities was utilized including BPM model and census datasets.


NETWORK DESIGN
Network design began with building the set of links and nodes in order to depict the physical
streets of Lower Manhattan. Once this was done, vehicles, and roadways were configured to ensure that
vehicle behavior reflected the observed data.

Vehicle types
Fifteen vehicle types were specified in the Lower Manhattan model. Each vehicle type has unique
characteristics including physical dimensions, performance parameters, driver behavior parameters and
demand characteristics that affect performance. TABLE 1 describes the vehicle types, their parameters
and typical route choice characteristics. The perturbation factor provides variability in route choice by
adding a stochastic element to the generalized cost (described further in the route assignment section) of
each possible route. The familiarity factor, expressed as a percentage, represents the proportion of drivers
assumed to have knowledge of the network. Familiar drivers makes a route choice based on minimizing
their generalized cost regardless of link type, while unfamiliar drivers minimize their generalized cost, but
are constrained to routes that are predominately over major road links. Light goods vehicles (delivery
vans) were given the same perturbation and familiarity factors as private cars because they were found to
exhibit similar behavior compared to large trucks.
The study area is unique because the density and frequency of bus services and the presence of
many different operators. Bus routes and stops were coded based on public timetables, route maps and
field visits. Bus routes were designed to run beyond their route termination point in order to represent
realistic conditions. Rather than buses disappearing from the network at the end of their route, bus routes
were coded to simulate deadheading to an appropriate exit point (like a layover area) in order to capture
the impact on other intersections.
The data used in coding the bus routes was gathered from various sources such as published bus
schedules, studies (12) and discussions with New York City Transit. Because there are many private bus
operators that have scheduled routes and stops in Lower Manhattan (coach and tour bus companies), not
all bus data was available from the aforementioned sources. When data was not available assumptions
were made based on local knowledge of bus depots and layover areas.

Road Hierarchy
Aside from coding the physical roadway, traffic behavioral and operational characteristics must
be taken into consideration. Adjacent land uses, traffic composition, pedestrians and transit activity all
impact traffic operations in Lower Manhattan. While many of these impacts cannot be explicitly modeled
in the software, there are a series of parameters that can be applied to reflect these impacts. Therefore it
was important to understand and define the functional road hierarchy so that parameters can be applied


consistently across the network.

FIGURE 2 depicts a road hierarchy developed in a previous study of Lower Manhattan streets
(10). In that study, the following street hierarchy is defined:
• Through streets – Major traffic and bus movements through the area (ex. FDR, Route 9A).
• Access streets – Major traffic and bus movements circulating within the area (ex. Broadway,
Church Street).
• Activity streets – Streets where land use consists of concentrations of retail and restaurants
(ex. South Street, Chambers Street).
• Support streets – Small streets serving delivery and pick-up, loading, entry to parking lots and
similar activities (ex. Albany Street and Pearl Street).


• Residential streets – Streets where land use is primarily housing.

This framework was shown to be a useful way to categorize streets in a systematic way, avoiding
ad hoc modifications to the link categories in the network. The framework was also advantageous in that
it considered important transportation characteristics beyond levels of traffic, such as land use, user
perception and urban form.
The number of link categories in the model was expanded to account for specific geometries and
effects from the major highways (FDR, Brooklyn Bridge, Route 9A) as well as narrow streets and alleys.
TABLE 2 shows the definition of key groups of categories based on the hierarchy defined in FIGURE 3.
Lane widths and speeds were input based on existing data, while category cost factors were based on a
combination of the street framework, site knowledge and observation.

Curbside Activity
Curbside activity such as on street parking, livery pick up/drop off, goods delivery and
construction delivery frequently occurs on streets in Lower Manhattan. This activity is typically midblock
and creates small impediments to traffic flow in the network that can cumulatively create larger impacts.
A number of approaches were considered. A multi-stage plug-in, developed by a third-party, offered the
capability to model curbside activity but presented upgrade and usability issues in this case. Also
considered was placing zones on top of links, but this created problems with getting accurate link
measurements since vehicles would exit the network mid-link.
It was determined that the most appropriate solution was to develop an on-street zone system, and
locate them perpendicular to links, throughout the network as destinations for taxi and goods delivery
vehicles. There were 78 zones representing on-street parking, livery vehicles and commercial loading and
unloading. In addition there were 12 special zones representing security areas, loading docks and
construction sites.

Pedestrians
Based on data collection and field observation, there is a high level of pedestrian and vehicle
interaction in Lower Manhattan. Studies of pedestrian level of service in Lower Manhattan have
measured pedestrian volumes as high as 5,900 persons per hour in the AM period (14). Pedestrians
impact vehicular flow and vice versa, causing noticeable impacts on the network. Pedestrian movement
had to be represented in order to create an accurate model of Lower Manhattan. At the time of model
development, Paramics lacked the capability to explicitly model pedestrian movements in a network.
As a result, the modeling team applied “dummy” signal phases to represent the delay to turning
vehicles resulting from pedestrian movements. The dummy phase stopped traffic movements for a
specified period of time to account for conflicting pedestrian movement. The length of the phase was
based on the overall length of the master phase, pedestrian occupancy and volume. This method was
based on a standard method of calculating the percentage of time that pedestrians and vehicles are in
direct conflict (15) and determining the delay in excess of the programmed pedestrian signal phases at
each intersection. Because right-turn-on-red movements are not allowed on New York City streets, the
movement is prohibited in the model, and therefore interactions between vehicles and pedestrians are
assumed to only occur when vehicles are making right or left turns on green.
Shorter phases were shown to result in shorter dummy phases and longer phases resulted in
longer dummy phases. The approach adopted the following principles:
• Dummy phases were not applied in instances where there was an all pedestrian phase;
• Dummy phases were not applied to movements where pedestrians were prohibited from
crossing;
• If there was a leading pedestrian interval, the length of the interval was deducted from the
dummy phase due to pedestrians being allowed to clear the conflict zone prior to the start of the turning
vehicular phase.


It is not possible to model vehicle and pedestrian conflicts in the same manner at unsignalized
intersection. In addition, unsignalized intersections typically have low volumes of vehicles and
pedestrians. Therefore turns across crosswalks at unsignalized intersections were designated as “minor”
movements to create lower speeds. At unsignalized crossings where high pedestrian volumes were
observed the corresponding crosswalk link speed was reduced to simulate the slow speeds experienced by
drivers trying to negotiate that crossing.

Demand
Demand was estimated using Paramics Estimator, which develops origin-destination tables based
on collected data. The best available data was used to estimate demand by origin and destination pair as a
starting point for the estimation process; this is referred to as a seed matrix. The seed matrix was based on
the New York Metropolitan Transportation Commission’s Best Practices Model (BPM), which is used to
forecast regional travel patterns. Eight different origin destination matrices were estimated, representing
different vehicle types and purposes. This was done in order to provide modelers the freedom to adjust
individual demand on different vehicle types and trip purposes independent of other traffic.

Route Assignment
The choice of assignment methodology is important in a complex urban environment like Lower
Manhattan where congestion builds and dissipates quickly. Paramics provides three alternative route
assignment methodologies, all-or-nothing, perturbation and dynamic assignment. Alternative
methodologies were assessed. Given the complexity and scale of the modeled network the dynamic
feedback method was found to be essential to accurately replicating route choice and operations in Lower
Manhattan. Dynamic feedback functions by recalculating route costs at fixed intervals so that familiar
drivers may alter their route mid journey.

Generalized Cost
Individual vehicles choose their route by evaluating the cost of all possible routes and choosing
the one with the lowest cost. Vehicle familiarity factors into the set of possible routes the vehicle can
choose from. For familiar vehicles, the available set of routes contains all possible routes to a destination;
for unfamiliar vehicles, the set of routes is restricted to routes composed of links designated as major.
Each link in the network is evaluated following a generalized cost formula:

Cost = a × T + b × D + c × P

Where Cost is in the user cost (minutes), T is time (minutes), D is route length (km), P is the price
of tolls (dollars). The units of coefficients a,b,c are unitless, minutes/km and minutes/dollar respectively.
Because cost is in minutes, the time coefficient a is equal to 1. The b coefficient was derived based on the
average travel speed of 25 mph which translates to 1.5 min/km. The c coefficient is zero because there are
no tolls on travel within Lower Manhattan.

Cost = 1× T + 1.5 × D + 0 × P

Feedback
Dynamic feedback is used to model the real-time assessment of travel times. This information is
made available to “familiar” drivers only, and is provided prior to and during their trip at the end of each
update period. This technique can be used in conjunction with stochastic assignment to provide a more
robust route-choice model.
A feedback period of five minutes was employed, meaning that “familiar” drivers (85% of the
total) calculate the cost of all available routes every five minutes. With the incorporation of perturbed


stochastic assignment, they might select a route that is not necessarily the shortest. A key objective was to
maximize the feedback period, given that drivers are generally not capable of making key route choice
decisions in short periods. However the longer feedback periods resulted in gridlock occurring during the
simulation. Different feedback periods were tested, with five minutes being the most appropriate in terms
of accurately modeling behavior while not making simulation runs computationally onerous.
Smoothing functions serve to dampen oscillations in travel time between update periods. There is
the possibility that dynamic feedback can induce large fluctuations in the traffic choosing alternative
routes after each update. A smoothing factor of 0.70 was used, which results in a weighted averaging of
70% of the latest values and 30% of the previously smoothed values.
A feedback decay factor keeps a link costs from going to zero immediately, should no car travel
along it during a time step. The default value of 0.995 was chosen resulting in an exceptionally slow rate
of decay in cost.

CALIBRATION
The calibration stage ensures that the model adequately reflects the observed traffic behavior,
traffic volume and travel times prior to a more robust and quantitative measure of performance in the
validation stage. Calibration involved a review of global and local model parameters that relate to
network and demand matrix definition and assignment. In addition, the calibration task involved a visual
review of the model operation during assignment using a variety of seeds to ensure the model replicated
traffic conditions that were observed on-site.

Visual Calibration
Visual examination of the network during simulation is important as a check on the quantitative
modeling described above. Although the effect on vehicle traffic is taken into account in the network
design stage, microsimulation models do not visually model the detailed maneuvers on the congested road
network. Parked vehicles, double-parked vehicles, bicyclists and pedestrians are not visually depicted in
the model. This results in an appearance that the street network may be less congested than it actually is.
To address this condition, a structured approach of applying link and node characteristics was taken to
replicate traffic impedances and ensure logical routing.
Furthermore, the model is designed to depict a typical day with recurrent congestion.
Nonrecurrent congestion such as incidents, break downs or other random events was not within the scope
of the model, although Paramics is capable of modeling these types of incidents if this were desired in the
future.

VALIDATION

Criteria
Validation criteria, shown in TABLE 4, focused on volumes, travel times and visual audits. The
criteria used several metrics: percent difference, R² and GEH statistics. Percent differences were used for
screenlines as they are the coarsest measure of traffic flow in the criteria. R² which measures goodness-of-
fit between an estimated and observed value is used for individual link counts and turning movements.
The GEH statistic is a standard traffic modeling measure used to evaluate the accuracy of flows given
wide ranges in observed flows across a network. The formula is:
2( M − C ) 2
GEH =
M +C
where
M = modeled volume
C = observed volume


The criteria were based on previous guidelines, literature (4,7,8,10,12) and available data.
Volume statistics were calculated at three levels of detail. This was done so that large flows could
be validated first, then slowly working toward validating the more detailed movements. Screenline flows
provide a coarse measurement across major inbound and outbound links in the network. Individual link
flows were measured for 120 intersections in the network. Next, turn movements were validated at critical
intersections where it was important to capture left turning behavior. Travel time measurements were
used to validate 10 different routes through Lower Manhattan.

Results

Screenlines
The screenline totals are reported for each direction (i.e. eastbound and westbound or northbound
and southbound) for all seven screenlines. As shown in TABLE 5, the total screenline flows were well
within the acceptable range of 5 to 10% – no screenline total had a percent difference greater than 7%.

Individual Link Flows
Individual link flows within each screenline were compared with the results shown in TABLE 5.
This included over 120 individual link flow counts. In both the AM and PM peak periods, the individual
link flow results generally met or exceeded the validation targets. For the R² correlation, both periods
produced results above the targeted range of 0.85 to 0.95. This shows that variability between the
modeled volumes and observed volumes is very low and that statistically, the model is in line with the
observed volumes. The percentage of GEH values below 5 for individual links exceeded the 75-80%
target in the PM, but was slightly short of the target in the AM. The percentage of GEH values below 10
for individual links exceeded the target ranges for the AM and PM periods.

Turning Movements
Turning movements at key locations on strategic routes were selected for validation. There were
approximately 320 turning movement counts considered fit for validation, compared to 120 link flows. It
was important that the turning counts used for validation were consistent with observed and historic data
because they were often used in the matrix estimation process. Because of observed inconsistencies in the
data collection process, several checks were applied to the turning movement counts in order to assure
they had been correctly collected. First intersection counts were compared with adjacent intersections for
consistency. If the count was not consistent with adjacent intersections the count was then compared with
historic data at the intersection. Counts that were inconsistent with both sources were not used in the
demand estimation or validation processes. The validation results are summarized in TABLE 5.
The modeled turning movement volumes are typically difficult to validate against observed
counts because they require large sample sizes in order to reduce variability. As a result, lower validation
targets were set.
TABLE 5 shows that both the AM and PM models meet the R² targets for turning movements.
In the case of the GEH targets, the AM model results just fall below the target range for GEH less than 5,
but the results meet the criteria set for GEH less than 10. The PM model validation meets both GEH
targets. The slightly lower validation results for the AM compared to the PM is most likely due to
discrepancies and flow variability over the modeled period in localized areas. Overall both the AM and
PM models provide a good correlation to observed turning movement volumes in the study area,

Travel Times
The travel times along major corridors were validated based on probe vehicle runs collected
during the data collection phase. Neither the AM or PM models met the travel time guideline targets,
although the AM model produced better results than the PM. In general, the travel time validation may


suggest that vehicles in the model travel faster through the network when compared to observations
during the survey. However the limited sample size and the difficulty in measuring single travel time runs
in Paramics (travel time results were recorded by creating a public transit route to act as a probe,
resulting in an underestimation of modeled travel times) are possible explanations for the disparity.
Future model work and data collection will focus on strengthening the travel time validation by
collecting a larger dataset and utilizing the capability in the new version of Paramics to measure modeled
travel times.

CONCLUSION
The experience developing the Lower Manhattan microsimulation model illustrates a practical,
planning-based approach to modeling a complex urban transportation network. Beginning at the data
collection phase, detailed information was collected regarding vehicles, transit, pedestrian, parking
behavior and land use. This data informed the design of the model, as did prior studies of the area, so that
an understanding of streets and neighborhoods informed the network construction. Assignment
parameters were developed based on understanding of the types of vehicles in Lower Manhattan, as well
as the time period being modeled.
Validation was found to be a time consuming and complex process. The microsimulation
guidelines and standards that were reviewed tended to focus on modeling highways or corridors – not
central business districts. Because central business districts are unique in regard to network size and users,
developing standards and guidelines around these needs would help improve the practice of modeling
urban areas. The most difficult validation issue was the inability to validate travel time in the model. The
issue was complicated by the high number of possible routes, high volumes and nonrecurrent congestion.
Because of all of these issues, it became apparent that a robust sample of travel times is necessary to
better understand the variability.
The model was designed with the intention of estimating the route changing and travel demand
resulting from changes in development and the street network. With these intentions in mind, the model is
considered valid and accurate.

Next Steps
The Lower Manhattan model has been used to analyze the impacts resulting from a variety of
proposed street management and development scenarios. Going forward, the Lower Manhattan model
will be used for a variety of planning tasks including testing the traffic impacts of street changes related to
development, pedestrianization and reconfiguration. In addition the Lower Manhattan Construction
Command Center intends to integrate the model in to the Lower Manhattan construction scheduling
system in order to assess traffic impacts of various detour plans.
The next phase of the model development will focus on developing a more robust pedestrian and
transit component, with pedestrian agents interacting with vehicles and transit. In addition, the model will
expand further north to encompass Chinatown and the Holland Tunnel areas of Lower Manhattan. The
expanded model development will involve additional data collection, calibration and validation processes.

References

1. Downtown Alliance. Lower Manhattan Fact Sheet 2008 (Q2), 2008.

2. “Manhattan: City Report Record Number of Visitors” January 14, 2008. URL:
http://www.nytimes.com/2008/01/14/nyregion/14mbrfs-visitors.html?fta=y , Accessed: 10-22-08.

3. Chicago Office of Tourism. 2006 Statistical Information, 2006.

4. FHWA. Guidelines for Applying Traffic Microsimulation Modeling Software. Prepared by Dowling
Associates. August 2003.

5. SIAS. Microsimulation Consultancy Good Practice Guide

6. Austroads. The Use and Application of Microsimulation Models, Prepared by ARRB Group. 2006

7. Dowling, R., Skabardonis, A. et al. Guidelines for Calibration of Microsimulation Models: Framework
and Applications.Transportation Research Record: Journal of the Transportation Research Board No
1876, TRB, National Research Council, Washington D.C. 2004, pp. 1-9.

8. Traffic Appraisal in Urban Areas: Highways Agency, Manual for Roads & Bridges, Vol. 12.
Department for Transportation, London, May 1996.

9. FHWA. Model Validation and Reasonableness Checking Manual.

10. Land Transport New Zealand Project Evaluation Manual

11. Freeway System Operational Assessment. Technical Report I-33: Paramics Calibration & Validation
Guidelines (Draft). Wisconsin Department of Transportation, District 2, Milwaukee, June 2002.

12. Lower Manhattan Development Corporation. Lower Manhattan Bus Study. 2006.

13. New York City Department of Transportation. Lower Manhattan Street Management Framework.
Prepared by Ove Arup & Partners Consulting Engineers. September 2004.

14. New York City Department of City Planning. Pedestrian Level of Service Study, Phase I – Chapter 5.
April 2006.

15. Highway Capacity Manual. TRB, National Research Council, Washington, D.C., 2000.

Tables and Figures

FIGURE 1 The Lower Manhattan simulation study area.

TABLE 1 Vehicle Type Parameters
Type ID Comment

Matrix that type

Proportion of
Perturbation

is applied to
Familiarity

matrix
Car 1 5% 55% 1 38% Car external to external
Car 2 5% 85% 2 100% Car Brooklyn Bridge related
Car 3 5% 85% 3 88% Car - other zones
Car 4 5% 85% 4 88% Car - on-street zones
Car 5 5% 55% 1 50% Cars assigned to HOV lane out of
BBT
Taxis 9 5% 85% 6 100% Taxis
FHV 10 5% 85% 7 100% Black Cars/Limos
Minibus 11 - - Fixed - Fixed route vehicle released to
Route collect travel time data
LGV 12 5% 85% 1 12% Light commercial vehicle
external to external
LGV 13 5% 85% 3 12% Light commercial vehicle - other
zones
LGV 14 5% 85% 4 12% Commercial vehicle - on-street
zones
Bus 16 - - Fixed - Fixed route bus services assigned
Route according to published timetables
and surveys
Coach 17 5% 25% 8 10% Part of heavy truck matrix
OGV 18 5% 25% 8 40% Part of heavy truck matrix
LGV 19 5% 25% 8 50% Part of heavy truck matrix

TABLE 2 Category Definition

Category Speed Lane Cost
Description Type
Numbers (mph) Width Factor
Highway
10-19 45 12 1
Through - FDR MAJOR
Highway
20-29 30 12 0.75
Through - BB MAJOR
Through - West 30-39 35 Urban MAJOR 12 0.75
Through - other 40-49 30 Urban MAJOR 12 1
Access 50-59 30 Urban MAJOR 11 2
Activity 60-69 30 Urban minor 11 1.8
Support 70-79 25 Urban minor 11 2.3
Support (Narrow
80-89 20 10 6
Sts) Urban minor
Residential & Alleys 90-99 15 Urban minor 10 8

TABLE 3 Vehicle Familiarity and Proportion
Vehicle Vehicle Vehicle Familiarity Proportion
Group Type Name (%) (%)
1 Car1 55 37.6 (Matrix 1)
2 Car2 85 100 (Matrix 2)
3 Car3 85 87.6 (Matrix 3)
4 Car4 85 87.6 (Matrix 4)
Car5-
Cars 5 55 50 (Matrix 1)
HOV
6 Car6 65 100 (Matrix 5)
12 UPS 85 12.4 (Matrix 1)
13 FedEx 85 12.4 (Matrix 3)
14 FedEx 85 12.4 (Matrix 4)
Taxis 9 Taxi 85 100
FHV 10 Black Car 85 100
17 Coach 25 10
Trucks 18 OGV2 25 40
19 LGV 25 50
11 Minibus Fixed
Buses
16 Bus Fixed

TABLE 4 Validation Criteria
Criteria Targets Comments

Screenline Flows

Percentage difference 5 - 10% Outliers may be accepted depending on
confidence of counts and other validation
criteria.

Individual link flows

R2 0.85 – 0.95 Correlation of all measured to modeled link
flows. Should tend toward 0.9.

GEH<5 75% - 80% of counts Small difference between modeled and
observed for most links

GEH<10 95% of counts No significant outliers, unless justification
provided.

Turn Flows

R2 0.85 – 0.95 Correlation of all measured to modeled turn
flows. Probably tend toward 0.85.

GEH<5 65% - 75% of counts Small difference between modeled and
observed for most turns

GEH<10 90% of counts A small number of significant outliers allowed,
that are shown not to significantly impact on
the models fitness for purpose.

Travel time

Mean difference <15% 85% of routes Difficult to achieve due to the lack of observed
travel time information along each route
compared to modeled

Average modeled travel 95% of routes Difficult to achieve given travel time variability
time within range of in network
observed times

TABLE 5 Summary of Validation Results
Criteria Targets Achieved Achieved Comments
AM PM

Screenline Flows

Percentage 5 – 10% All <6% All <7% Acceptable
difference

Individual link flows

R2 0.85 – 0.95 0.99 0.99 Acceptable

GEH<5 75% - 80% of counts 74% 84% Acceptable – AM
slightly low

GEH<10 95% of counts 96% 98% Acceptable

Turn Flows

R2 0.85 – 0.95 0.95 0.98 Acceptable

GEH<5 65% - 75% of counts 63% 70% Acceptable – AM
slightly low

GEH<10 90% of counts 91% 94% Acceptable

Travel time

Mean difference 85% of routes 50% 11% Doesn’t achieve
<15% targets

Average modeled 95% of routes 22% 6% Doesn’t achieve
travel time within targets
range of observed
times

Microsimulation Model Design in Lower Manhattan: A Street Management Approach

More Related Content

Similar to Microsimulation Model Design in Lower Manhattan: A Street Management Approach (20)

Recently uploaded (20)

Microsimulation Model Design in Lower Manhattan: A Street Management Approach