traffic impact analysis v2
TRANSCRIPT
TRAFFIC IMPACT ANALYSIS OF AN URBAN OFF-HOUR DELIVERY PROGRAM 1 USING MACROSCOPIC AND MESOSCOPIC SIMULATION MODELS 2
Shrisan Iyer (corresponding author) 3 Research Associate, 4 Rutgers Intelligent Transportation Systems Laboratory 5 Rutgers, The State University of New Jersey 6 CoRE 736, 96 Frelinghuysen Road, Piscataway, NJ 08854 7 Tel: +1 732 445 6243 x274, Fax: +1 732 445 0577, Email: [email protected] 8
Kaan Ozbay, Ph.D 9 Professor & Director, 10 Rutgers Intelligent Transportation Systems Laboratory 11 Rutgers, The State University of New Jersey 12 Department of Civil and Environmental Engineering 13 623 Bowser Road, Piscataway, NJ 08854 14 Tel: +1 732 445 2792, Fax: +1 732 445 0577, Email: [email protected] 15
Satish Ukkusuri, Ph.D 16 Associate Professor, 17 School of Civil Engineering, Purdue University, 18 550 Stadium Mall Drive, West Lafayette, IN 47907-2051, USA 19 Tel: +1 765 494 2296, Fax: +1 765 494 0395, Email: [email protected] 20
Wilfredo F. Yushimito 21 Graduate Research Assistant, 22 Department of Civil and Environmental Engineering, Rensselaer Polytechnic Institute, 23 110 Eighth Street, 5107 Jonsson Engineering Center, Troy, NY 12180-3590, USA 24 Tel: +1 518 276 8306, Fax: +1 518 276 4833, Email: [email protected] 25
Ender Faruk Morgul 26 Graduate Research Assistant, 27 Rutgers Intelligent Transportation Systems Laboratory 28 Rutgers, The State University of New Jersey 29 Department of Civil and Environmental Engineering 30 623 Bowser Road, Piscataway, NJ 08854 Email: [email protected] 31
Jose Holguin-Veras, Ph.D 32 Professor, 33 Department of Civil and Environmental Engineering, Rensselaer Polytechnic Institute, 34 110 Eighth Street, 4030 Jonsson Engineering Center, Troy, NY 12180-3590, USA 35 Tel: +1 518 276 6221,Fax: +1 518 276 4833, Email: [email protected] 36
Abstract: 115 words 37 Word count: 4,694 Text + 9 Figures + 1 Table = 7,194 words 38 Submission Date: August 1st, 2010 39 Revised Submission Date: November 15th, 2010 40
41
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 1
ABSTRACT 42
This paper describes the usage of a regional travel demand model and an extracted mesoscopic 43
sub-simulation model in tandem to model and observe the traffic impacts of an off-hour delivery 44
program. The study is based on OHD-participation behavioral data for the New York City 45
borough of Manhattan, with traffic impacts measured throughout the New York metropolitan 46
region. Analysis is conducted to determine the effectiveness and impacts of the scenarios 47
modeled; focusing on the changes predicted by the traffic models, with link travel time and speed 48
changes as the key measure of output. The results from both models are compared and analyzed, 49
and a discussion on the usage of these models for this purpose is presented. 50
51
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 2
INTRODUCTION 52
Freight traffic constitutes a significant component of a region’s traffic system. Given its 53
considerable increase in recent years, commercial vehicles (CMVs) have become a primary 54
target for planners and policymakers in mitigating traffic congestion. Much work has been done 55
regarding the quantification of the traffic impacts of congestion mitigation programs but less has 56
been conducted concerning programs targeted specifically at trucks or commercial vehicles. 57
Programs addressing highway freight transportation problems are new and experimental, with 58
only a handful of cities implementing congestion control measures (1). The effects of trucks and 59
commercial vehicles on highway networks are known to be negative, specifically causing 60
congestion due to their nature as large vehicles generally traveling at a slower speed than 61
automobile traffic. Studies have shown that truck traffic negatively affects the flow rate of 62
highways and local roads, thereby causing congestion on roadways with high traffic volumes (2). 63
Holguin-Veras et al. (2006) determined that freight traffic generated by delivery vehicles to 64
city businesses not only contributed to congestion but caused added problems due to double-65
parking and blockages as a result of the lack of parking spaces during the day-time, peak, 66
delivery hours (3, 4). These claims are supported by research into the effects of illegal parking on 67
traffic congestion which show that illegal parking (primary conducted by CMVs) causes 68
significant capacity losses to roadways, which produce more severe effects during peak hours 69
than during off-peak hours (5). As a result, policy makers have sought to control truck and 70
commercial vehicle traffic, particularly within cities’ central business districts, with either value 71
pricing measures or by introducing off-peak delivery programs. Both ideas are gaining popularity 72
in the United States but their degrees of success are yet unknown. 73
An off-hour delivery (OHD) program is expected to provide benefits to the highway 74
network, since fewer trucks and commercial vehicles during congested hours would improve 75
highway speeds and decrease travel times. However the precise effects are unknown and their 76
estimation is difficult. Freight planning models or highway networks can be developed and run 77
in simulation software but these processes are both time-consuming and require extensive data 78
procurement. Instead many city transportation agencies or metropolitan planning organizations 79
already have developed traffic planning and simulation models. In order to assess the impact of a 80
freight targeted-policy program, it is important to consider the effects to all classes of vehicles 81
and the entire transportation system. While some domestic freight is moved by rail, a significant 82
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 3
percentage of goods are transported by trucks and delivery vans, which use the same highway 83
infrastructure as all other vehicles. This paper focuses solely on trucks and delivery vans, and 84
since any changes to their travel patterns will affect the entire highway network, this paper also 85
considers the effect on turn passenger cars. To fully assess traffic impacts throughout the region, 86
a model is required that considers both passenger and freight travel. Comprehensive travel 87
demand models developed by state and regional agencies consider many trip types and many 88
modes. This paper uses a regional travel demand model to estimate the effects of an off-hour 89
delivery (OHD) program. 90
Yannis, et al. (2006) used a transportation model to estimate the impacts of an OHD program 91
in the city of Athens. They presented a methodology to model modified commercial vehicle 92
origin-destination (OD) demands on a highway network when delivery operations within the city 93
of Athens were restricted (6). The study described modeling of the city’s roadways under 94
observed and modified commercial vehicle demands, accomplished by first observing existing 95
traffic conditions to collect sufficient data to build a comprehensive roadway network, 96
calibrating the collected data against actual conditions, and implementing the OHD program 97
changes into the model. Using SATURN, they were able to conduct traffic assignment based on 98
actual (base) traffic demands and again with modified demands, which were caused by 99
restricting delivery vehicles from entering the study area within certain times of day. It is 100
important to note that SATURN is strictly a traffic assignment model, not a large scale travel 101
demand model (7). 102
Yannis et al. created a network that consisted of 285 production and attraction zones, with 103
demand matrices for six separate time periods throughout the day. Additionally land uses and 104
average stop times were also studied to represent the effect that actual delivery activities have on 105
roadways, which manifest themselves in the forms of double parking or lane blockages. The 106
researchers were able to code these activities into the network and thus modify the capacity of 107
certain roads where deliveries were taking place. The simulations showed that by barring 108
delivery vehicles from the study area from 7:00am–10:30am, simulated average roadway speeds 109
increased by 4.7%, and a similar restriction from 2:00pm–4:30pm increased simulated average 110
speeds by 1%. Conversely, the average speed for the 10:30am–2:00pm period decreased by 5.8% 111
as the displaced delivery vehicles were assumed to use this period to enter the study area. 112
However the researchers noted that the increase in speeds during the morning and afternoon 113
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 4
periods had a greater benefit than the loss in the midday periods, due to higher traffic volumes in 114
the morning and afternoon. 115
This paper describes the usage of the New York Best Practice Model (NYBPM), a regional 116
planning model developed by New York Metropolitan Transportation Council (NYMTC), to 117
study the impacts of a proposed off-hour delivery program on the traffic network of the New 118
York City metropolitan region (8). The availability of NYBPM eliminates the need to create a 119
new network model for this study. Some of the shortcomings of a large-scale travel demand 120
model can be overcome by extracting a detailed localized sub-model which can be used for 121
further analysis in a mesoscopic simulation tool. In this case, a mesoscopic sub-network of 122
Manhattan, the most densely populated and commercial county in the city and region, is 123
extracted from the NYBPM for detailed simulation and analysis. Once the models’ results are 124
obtained analysis is conducted to determine the effectiveness and impacts of the off-hour 125
delivery scenarios modeled. The results from both models are compared and analyzed, and a 126
discussion on the usage of these models is presented. 127
The OHD program studied provides incentives for businesses within Manhattan to accept 128
their deliveries during the off-peak overnight hours instead of the daytime hours. Data from prior 129
behavioral studies is used to determine the percentage of business willing to accept incentives 130
based on a tax incentive amount, from which scenarios are constructed to measure traffic impacts 131
based on levels of receiver participation. Based on the receiver participation levels, CMV traffic 132
corresponding to the receivers’ deliveries is shifted from the daytime to the overnight hours 133
within the traffic models used. The models are then run with modified demands and the outputs 134
are compared to the base (existing) case to analyze the potential effects of the OHD program. 135
The paper is organized as follows: first an overview of each of the models and 136
methodologies used is provided, then the scenarios studied and modeled in both the NYBPM and 137
the mesoscopic simulation models are detailed, and the findings and results of each presented. 138
Finally a comparison of traffic impacts obtained from both models is conducted before 139
conclusions are discussed. 140
141
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 5
METHODOLOGY 142
Models Utilized 143
If commercial vehicle traffic is shifted from one time period of day to another – from peak 144
periods to off-peak periods – it is critical to observe whether there is a measurable improvement 145
to the remaining traffic conditions during the peak periods, and conversely whether off-peak 146
conditions are significantly disrupted. As mentioned, the New York Best Practice Model 147
(NYBPM), the travel demand model used by New York-area planners (9), is employed for 148
region-wide analysis. It covers 28 counties in the New York area, and allows for studying the full 149
effects of the OHD program to the complete highway network of New York City and its 150
surrounding areas. 151
NYBPM is particularly useful for analysis of the changes and redistributions of truck travel 152
patterns, since it utilizes TransCAD’s multi-modal, multi-class, assignment feature (9). The input 153
origin-destination (OD) matrices for highway assignment are six-fold, one for SOV, HOV2, 154
HOV3+, truck, and other commercial. In the multi-class assignment, each trip class is treated 155
separately and utilizes its own cost or volume-delay function, and classes prohibited on certain 156
links are accounted for. Cars and trucks are assigned separately, but still allowed to find the best 157
route to minimize their cost. However it should be emphasized that NYBPM was not designed 158
specifically for or with an emphasis on freight modeling. NYMTC is currently studying 159
alternative ways to study freight transportation and plan for future changes (10). 160
The assignment portion of the model is a collection of four models for four periods of the 161
day. The four periods, each with their own networks and origin-destination (OD) matrices, are 162
AM Peak Period (AM; 6–10am), Midday Period (MD; 10am–3pm), PM Peak Period (PM; 3–163
7pm), and Overnight Period (NT; 7pm–6am). The defined network periods and separate OD 164
matrices allow for modeling to be based on shifting demand from the daytime periods to the 165
overnight period, as described by the OHD program. Changes to CMV (trucks and other 166
commercials) behavior are represented by manipulating the number of CMV trips between each 167
origin-destination pair for each time period. Once the existing CMV OD matrices within 168
NYBPM are altered the models’ outputs are compared to the base-case. The NYBPM modeling 169
methodology is illustrated in Figure 1. 170
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 6
NYBPM
Sub-Network Extraction
Sub-Network Calibration
Hourly Time Dependent OD
Matrices Calibration
Simulation
Dyn
amic
Dat
a
Comparative Analysis of Traffic Impact
Scenario Hourly Time Dependent OD
Matrices
Simulation
Behavioral M
odel
Scenario Generation
Base Case Scenarios
Calibration
Calibrated NYBPM OD Matrices
Calibrated NYBPM
Shift Scenarios AssignmentNYBPM Base-Case Assignment
Comparative Analysis of Traffic Impact
BMS Data
Shift Scenario OD Matrices Shift Model
NYBPM
Mesoscopic Simulation 171
Figure 1: Research Methodology 172
173
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 7
The New York Best Practice Model, running in TransCAD, also allows for transfer of the 174
highway network to TransModeler, both software tools developed by Caliper. While TransCAD 175
is a static traffic assignment tool, TransModeler is a traffic simulator that is capable of 176
performing simulations at a microscopic and mesoscopic level. In order to conduct more detailed 177
analysis at the vehicle-level, the simulation is performed on an extracted sub-network of 178
Manhattan and its access facilities (bridges and tunnels). Figure 2 shows the NYBPM highway 179
network and the sub-network of Manhattan used for mesoscopic simulation. 180
The methodology for mesoscopic simulation consists of two simultaneous procedures 181
(Figure 1): developing a model for the base case and scenario generation. Since NYBPM has 182
been designed for modeling macro-scale static behavior, it cannot be directly used as an input for 183
meso-simulation, since only major intersections in the local network have intersection geometry 184
details. However, it is still appropriate for mesoscopic simulation provided that the NYBPM has 185
the necessary road class information, free flow speeds, capacities, number of lanes, etc. Hence 186
the network can be imported directly to TransModeler for the main features of the network. Once 187
imported, the network is calibrated to assure a minimum exit-flow and avoid queues and 188
spillovers. 189
190
Figure 2: NYBPM Highway Network with Manhattan Sub-network Highlighted 191
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 8
The mesoscopic simulation uses hourly OD matrices, with three classes being simulated: 192
passenger cars, trucks, and other commercial vehicles (pickups, vans). In all cases, simulations 193
are performed by the four general time periods with preloading the state of the previous 194
simulation period (e.g., end-state of AM is preloaded for MD). The route choice model uses the 195
default stochastic shortest path with response to high delay embedded in TransModeler (11). 196
Under this method, paths costs are randomized based on different parameters such as perception 197
of shortest path error and behavior parameters that have been calibrated according to the 198
available data. The simulation also reroutes vehicles when their delays (different with respect to 199
their free flow travel times) are too high. Thus the paths are initially fixed but are allowed to vary 200
when congestion is high. However since the simulation requires a good estimate of the travel 201
times, historical travel times are obtained from the base case using a sample of five runs and 202
aggregating the travel times following an averaging procedure akin to the method of successive 203
averages - MSA (11). 204
205
Shift Model 206
A model is developed to translate off-hour delivery participation levels into traffic demand 207
changes, to study the traffic effects of the OHD program. Based on previous studies (12, 13, 14), 208
there is a percentage of receivers (the data available is restricted to food and retail industries) 209
within Manhattan willing to accept annual tax incentives to shift their delivery operations to the 210
off-peak hours (assumed as 7pm–6am). In this section, we present the scenario generation 211
assuming that different tax incentives are used to implement OHD. The resulting scenarios 212
provide a shifting percentage of commercial vehicles that are shifted to the off-peak hours. In 213
order to understand which types of industries are more likely to be willing to implement off-214
hours deliveries, behavioral modeling results were obtained by applying discrete choice models 215
to scenarios discussed in detailed in Holguin-Veras et al. (13, 14). These scenarios were intended 216
to assist in determining the types of policies that would have the greatest impact upon the 217
willingness of carriers and receivers to participate in OHD by asking receivers how likely they 218
would be to accept a certain percentage of their deliveries during the off-hours in return for a tax 219
deduction. 220
The estimates are provided for every zip code in Manhattan, which for simplicity are 221
grouped into community board groupings. There are four main community board groupings in 222
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 9
Manhattan, each divided according to land use. The first two community board groupings (1-3 223
and 4-6) represent the two main central business districts of Manhattan, Downtown and Midtown, 224
where most commercial establishments and deliveries take place. Community board groupings 7-225
8 and 9-12 (Uptown) are more residential in nature and attract fewer CMV trips. This calculation 226
assumes that all CMV trips destined for Manhattan are deliveries. 227
For both models, CMV OD trips are stored in period-specific origin-destination (OD) 228
matrices, with each cell containing the number of trips between zones per period. Once the 229
existing commercial vehicle OD matrices are altered, the traffic assignment module of NYBPM 230
and the mesoscopic simulation network is re-run. Alterations to the CMV OD matrices are 231
accomplished with shift factors calculated from the behavioral data, which are applied to CMV 232
OD demands between all originating zones outside of Manhattan and destination zones within 233
Manhattan. The shift factors reduce AM Peak, Midday, and PM Peak period CMV OD trips by a 234
percentage (from the behavioral data), and the sum of the reduced trips is added to the Overnight 235
period, for each OD pair. This has been assumed to be uniformly distributed across the NT hours 236
in the mesoscopic model. Since only food- or retail-related truck traffic is modeled, the estimate 237
of the probability that a commercial vehicle in the traffic stream would be making food or retail 238
deliveries is taken to calculate the percentage of trips in the NYBPM model corresponding to 239
food or retail deliveries. The percentage of commercial vehicle traffic shifting to the off-hours 240
can be, αJ, calculated by Equation 1: 241
∑=e
eJ
eJJ ωρα (1) 242
243 where J = destination zone where receivers are located 244
e = industry segment {retail, food} 245
ρ = percentage of deliveries from industry ‘e’ shifting to off-hours 246
ω = proportion of total deliveries associated with industry ‘e’ 247
248
Computationally, all OD trip shifts are done exogenously in MATLAB. To apply a shit 249
factor to a certain group of zones, all OD pairs with destination in the community board group of 250
J zones being considered receive the shift factor. The qualifying OD pairs are those with the 251
origin open to all zones in the network, Zones 1-4000, and the destination within Manhattan, 252
Zones 1-318, (i,j) = (1:4000,1:318). This signifies that even trips originating within Manhattan 253
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 10
are shifted, to account for chained trips, and to maintain the link between ‘deliveries’ from the 254
behavioral data and ‘trips’ in the model. For simplicity, shift factors are used corresponding to 255
the community board groupings described to account for the more than 1,200,000 OD pairs 256
receiving a shift factor. 257
258
Scenarios Modeled 259
For the scenarios simulated in NYBPM the shift factors shown in Table 1 are used. The shift 260
factors are calculated for four community board groupings in Manhattan, and by accounting for 261
both the number of food & retail receivers located in these groupings, and the participation levels 262
predicted by the behavioral studies (12). 263
264
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 11
Table 1: Shift Factors by Scenario 265
Scenario Tax Incentive
Community Boards, J
Retail Proportion, ρR
Food Proportion, ρF
Retail Percent, ωR
Food Percent, ωF
Shift Factor, αJ
Average Shift
F1, 2, 3 16.47% 12.83% 3.60%4, 5, 6 21.45% 7.15% 2.57%
7, 8 11.82% 10.11% 2.79%9, 10, 11, 12 14.34% 17.13% 4.46%
1, 2, 3 16.47% 12.83% 8.51%4, 5, 6 21.45% 7.15% 6.02%
7, 8 11.82% 10.11% 6.58%9, 10, 11, 12 14.34% 17.13% 10.56%
1, 2, 3 16.47% 12.83% 12.62%4, 5, 6 21.45% 7.15% 9.29%
7, 8 11.82% 10.11% 9.73%9, 10, 11, 12 14.34% 17.13% 15.44%
1, 2, 3 16.47% 12.83% 15.12%4, 5, 6 21.45% 7.15% 11.70%
7, 8 11.82% 10.11% 11.60%9, 10, 11, 12 14.34% 17.13% 18.14%
1, 2, 3 16.47% 12.83% 17.05%4, 5, 6 21.45% 7.15% 13.97%
7, 8 11.82% 10.11% 13.01%9, 10, 11, 12 14.34% 17.13% 19.98%
1, 2, 3 16.47% 12.83% 23.49%4, 5, 6 21.45% 7.15% 22.28%
7, 8 11.82% 10.11% 17.65%9, 10, 11, 12 14.34% 17.13% 25.65%
22.03%
2.93%
6.90%
10.42%
12.79%
14.83%$25,000 36.39% 86.17%
$50,000 74.80% 87.12%
$15,000 18.39% 74.75%
$20,000 26.71% 83.55%
$5,000 4.59% 22.21%
$10,000 10.44% 52.92%
6
1
2
3
4
5
266 267
The scenarios studied can be interpreted in terms of the average shift factor of each scenario, 268
which is the average of the shift factors for each community board weighted by the proportion of 269
deliveries to each community board grouping. It can be observed that as tax incentive amount 270
increases, the marginal increase in average shift factor decreases, due to a limit in receiver 271
participation. 272
The six scenarios shown are modeled in NYBPM, while the first three are simulated in the 273
extracted mesoscopic sub-network. Prior to this, both models undergo a substantive calibration 274
process using available up-to-date data. The truck OD matrices of NYBPM are inflated to current 275
levels, while the mesoscopic network of Manhattan is calibrated to existing conditions. Both 276
models are refined to represent an accurate base case, since the results of each scenario run are 277
compared with the base-case, and presented in the following section. 278
279
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 12
RESULTS 280
NYBPM Results 281
The macroscopic network model’s (NYBPM) assignment output contains information for all 282
53,000 links in the highway network, including vehicle flows by class, travel time, and average 283
speed. Two of the important parameters for measuring traffic effects can be calculated from this 284
output: Vehicle Miles Traveled (VMT) and Vehicle Hours Traveled (VHT). VMT gives an idea 285
on the total distance traveled by all vehicles in the region on a typical day, while VHT is a 286
convenient method of measuring travel times, and by extension, congestion. While changes to 287
VMT do not clearly indicate whether the network is more of less congested, this conclusion can 288
be reached from observing changes to VHT. For example vehicles may take longer paths to 289
avoid congested links, and in turn reducing their overall travel time, thus saving time and 290
reducing VHT while increasing VMT. 291
The results show the net differences between output parameters from the calibrated year 292
2007 base model and the shift scenario model, and percentage changes of the output parameters. 293
First Figure 3a shows the change in vehicle miles traveled (VMT) by a shifting scenario’s 294
assignment on the network, then Figure 3b shows the changes to vehicle hours traveled (VHT). 295
The figures show the resulting output from the entire New York area network of all links in the 296
NYBPM. The total 24-hour day (sum of all four periods) change in vehicle miles traveled (VMT) 297
as a result of a specific scenario’s assignment is represented by the heavy black line, while the 298
gray line represents only the three daytime periods from when truck traffic is subtracted. 299
The full-day results show that as the average shift levels increase vehicle miles traveled and 300
vehicle hours traveled for all vehicles in the network both decrease. However they also show that 301
as the proportion of deliveries shifted increases the marginal benefits decrease. For example 302
beyond a 15% average shift the net benefits are only minimally increased. Figure 4 also shows 303
the change in volume to capacity ratio (v/c), a measure of congestion, for Manhattan roads based 304
on the scenarios of demand shift. Daytime congestion (volume/capacity ratio) in Manhattan is 305
estimated to reduce in excess of 1.5% for significant demand shifts. 306
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 13
307
308 Figure 3: NYBPM Network Change in (a) VMT, (b) VHT by Shift Scenario 309
310
(a) Vehicle Miles Traveled (VMT)
(b) Vehicle Hours Traveled (VHT)
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 14
The 24-hour net changes in VMT and VHT are aggregated and the net changes from the 311
base-case scenario can be seen in Figure 5a for all network links, and in Figure 5b for changes in 312
Manhattan links. Manhattan, being the target of the program, covers a large proportion of the 313
total network effects and Lower & Midtown Manhattan are the core central business districts of 314
the city. In many scenarios, half of the reduction in VHT is experienced in Manhattan. 315
While the expected benefits from network assignment are expected to resemble the general 316
relationship between tax incentive and average percentage of freight traffic shifted, the exact 317
relationship cannot be followed due to network assignment affects. Particularly, vehicle miles 318
traveled do not always decrease with decreased levels of traffic. Specifically this can be 319
explained as vehicles taking longer paths that are less congested which might still save them time, 320
as they seek to minimize their total trip costs. Vehicle hours traveled however do incrementally 321
decrease with increasing tax incentives and decreased freight traffic in most cases. For some 322
scenario-to-scenario comparisons, reducing the amount of CMVs using the network does not 323
always result in a decrease of vehicle hours traveled. Since NYBPM employs user-equilibrium 324
assignment instead of system-optimal, the effect to the entire system is not always desirable. 325
326 Figure 4: Change in Manhattan Links’ Volume/Capacity Ratios 327
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 15
328
329 Figure 5: Scenario Net Benefits: (a) All Network Links, (b) Manhattan Links 330
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 16
Mesoscopic Model Results 331
The NYBPM, being a regional macroscopic model, is not largely sensitive to small changes in 332
the number of CMVs assigned during periods. To measure the effects of the OHD program at a 333
more detailed level, a mesoscopic sub-network of Manhattan is extracted and the OHD program 334
modeled. The results are presented with statistics calculated at a path level, where travel times 335
and speeds are computed from the beginning of vehicle trips until vehicles reach their 336
destinations. The results of the focused Manhattan sub-network account for changes in the travel 337
times and speeds for both an individual average trip and for the aggregated effect in total travel 338
times. The results are also aggregated to find the overall effect per day and per period of time 339
(AM, MD, PM, NT), which accounts for network effects in a more consistent fashion. 340
As expected the results show an inverse relationship between percentage shifts and travel 341
times (Figure 6), as well as speeds (Figure 7). ‘Daytime’ aggregates the results for the three peak 342
periods (AM, MD, and PM) in which the truck and other commercial vehicle demands are 343
reduced by the OHD model. These periods show a decrease in total travel time under all 344
scenarios, with the AM period showing the highest decrease in total travel time. The results also 345
show a monotonic increase in travel times for all scenarios, up to 4.2% in Scenario 3 (which 346
represents an average shift of 10.42% of the daytime CMVs) in the NT period. However, this 347
increase in travel time is in all cases outweighed by the reductions in the other daytime periods. 348
This is expected given that the NT period has much less volume than the daytime periods, and 349
the increase is not as significant because the vehicles can move at slightly decreased speed. In 350
particular the effects are largest during the AM Peak and Midday periods, as compared with the 351
reduction in the PM peak, which has a more compact distribution of trips. 352
Figure 8 shows the congestion pattern for the 24 hours of the simulation, and it can be 353
observed that the scenario with the lowest average shift (2.93%) has a similar pattern as the base 354
case (no shift). However, once the shifts are larger (6.90% or 10.42%) the congestion reduces 355
significantly between 7–10 am and between 12–3 pm. During the NT hours (7pm–6am) 356
congestion is increased slightly due to the new traffic added during these hours. The shifts have 357
significant effects during the peak hours of each period, when the traffic is reduced. The overall 358
reduction in total travel times within the Manhattan sub-network over a 24-hour period are 359
0.93% for Scenario 1 (2.93% average shift), 2.93% for Scenario 2 (6.90% average shift), and 360
4.2% for Scenario 3 (10.42% average shift). 361
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 17
362
‐10.00%
‐8.00%
‐6.00%
‐4.00%
‐2.00%
0.00%
2.00%
4.00%
6.00%
Scenario 1 (2 93%) Scenario 2 (6 90%) Scenar
Chan
ge from
Base
363
Figure 6: Sub-network Change in Total Travel Time by Period 364 365
‐2.00%
‐1.50%
‐1.00%
‐0.50%
0.00%
0.50%
1.00%
1.50%
2.00%
Daytime NT 24‐hour
Chan
ge from
Base
Scenario 1 (2.93%) Scenario 2 (6.90%) Scenario 3 (10.42%) 366
Figure 7: Sub-network Change in Average Speed per Vehicle for Trips Completed 367
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 18
368
Figure 8: Sub-network Travel Time Pattern by Shifting Scenario 369
370 Regional Model/Sub-network Traffic Impact Comparison 371
A comparison between NYBPM and the sub-network simulation results is difficult to perform 372
since the mesoscopic simulation accounts for results and effects within Manhattan, while the 373
NYBPM aggregates the results of the overall regional network. Moreover, there is some 374
accuracy lost due to the reduction of the area of scope in the mesoscopic model. Even if the 375
results of only the Manhattan links of NYBPM are compiled, the output will differ from the sub-376
network links’ results due to the interactions with the neighboring regions’ links in the larger 377
model. In addition the results obtained through the TransModeler simulation are path–based, 378
while the NYBPM provides results at a link level from the traffic assignment nature of the model. 379
In order to perform the comparison TransModeler path-based results obtained are converted into 380
link-based results. Using the information from paths, the total travel times in segments 381
corresponding to the links of the NYBPM are obtained. These results per link are aggregated by 382
time period and presented in Figure 9, compared with NYBPM results for the three scenarios 383
(average shifts of 2.93%, 6.90%, and 10.42%) run in both models. 384
It can be observed that the mesoscopic simulation shows far greater travel time savings 385
during the daytime than the NYBPM. These differences are also reflected in the 24-hour results 386
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 19
for all scenarios, where the mesoscopic simulation is observed to provide larger changes, 387
percentage-wise, compared with NYBPM. The results indicate that the mesoscopic sub-network 388
is more sensitive to the reduction in daytime truck traffic than the NYBPM. These differences are 389
due to how each model manages congestion and the traffic flow model used. For instance a 390
macro-model, such as NYBPM, uses a simplified traffic flow model, while the mesoscopic 391
simulator uses a more sophisticated traffic model and accounts for more realistic ways to 392
compute and aggregate delays. During the NT period, this difference is not significant because 393
the period does not have significant congestion. However the daytime periods have significant 394
congestion, which causes differences in the model output. For the sake of completeness the 395
results of the path-based results were included, and it can be observed that in general the link-396
based results of the simulations slightly overestimate the travel times saved. Overall the 397
mesoscopic model is seen to be more sensitive to the OHD program studied, while the NYBPM, 398
being a large-scale model, is not as sensitive. 399
400
401 Figure 9: NYBPM-Sub-network Comparison (Manhattan Links) 402
403
CONCLUSION 404
This paper presents the results of modeling an off-hour delivery program for New York City in 405
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 20
two traffic modeling tools. Both networks are calibrated to the needs of this study and behavioral 406
modules representing the behavior of freight carriers responding shifting their deliveries to 407
overnight hours were implemented following the methodology. Based on the response to certain 408
tax incentives by businesses within Manhattan, commercial vehicles providing deliveries to them 409
shift their trips to the overnight period. Both a macroscopic regional travel demand model and a 410
mesoscopic sub-simulation network show a measurable impact to congestion and network 411
conditions. During the daytime hours, as trips are shifted away, travel times improve and link 412
speeds increase; while in the overnight period conditions deteriorate due to additional 413
commercial vehicle trips. However the benefits (in terms of travel time) observed during the 414
daytime hours in both modeling tools are greater than the losses to network efficiency observed 415
in the night hours. Therefore the OHD program is projected to have a net positive effect on the 416
traffic network. 417
Since both tools provide different type of results, the path based data obtained from the 418
mesoscopic simulation model is converted to link based data in order to compare the results of 419
both models. In general terms, the overnight period, where traffic is accommodated after the shift, 420
provides similar results. Differences arise in the daytime period, which are periods with 421
significant congestion. In these cases the mesoscopic model is more sensitive to the reduction in 422
the traffic congestion. This difference is expected because static models, such as NYBPM, do not 423
completely accounts for congestion effects in traffic. The mesoscopic simulation is more 424
appropriate for studying the detailed effects of traffic changes on vehicles, but extension to the 425
full regional network is impractical due to the simulation time. The usage of both models allows 426
for studying the OHD program’s effects on a large area while still understand the maximum 427
beneficial effects at a detailed, mesoscopic scale. 428
429
ACKNOWLEDGEMENTS 430
The authors would like to acknowledge USDOT for funding this innovative research study, as 431
well as New York/New Jersey-area transportation agencies such as NYMTC, NYSDOT, 432
NYCDOT, NJDOT, NJTA, and PANYNJ for providing data and models used in this study. Their 433
support is appreciated. 434
435
436
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 21
REFERENCES 437
1. Crainic, T. G., Ricciardi, N., & Storchi, G. (2004). Advanced Freight Transportation Systems 438
for Congested Urban Areas. Transportation Research Part C, 12, 119-137. 439
2. Ioannou, P., Anastasios, C., Bose, A., Jula, H., Kanaris, A., Pourhammadi, H., et al. (2001). 440
Modeling and Route Guidance of Trucks in Metropolitan Areas. Metrans Transportation 441
Center. 442
3. Holguin-Veras, J., Perez, N., Cruz, B., & Polimeni, J. (2006). Effectiveness of Financial 443
Incentives for Off-Peak Deliveries to Restaurants in Manhattan, New York. Transportation 444
Research Record: Journal of the Transportation Research Board, 1966, 51-59. 445
4. Holguin-Veras, J., List, G. F., Meyburg, A. H., Ozbay, K., Paaswell, R. E., Teng, H., et al. 446
(2001). An Assessment of Methodological Alternatives for a Regional Freight Model in the 447
NYMTC Region. New York: Region II University Transportation Research Center. 448
5. Han, L. D., Chin, S.-M., Franzese, O., & Hwang, H. (2005). Estimating the Impact of 449
Pickup- and Delivery-Related Illegal Parking Activities on Traffic. Transportation Research 450
Record: Journal of the Transportation Research Board , 1906, 49-55. 451
6. Yannis, G., Golias, J., & Antoniou, C. (2006). Effects of Urban Delivery Restrictions on 452
Traffic Movements. Transportation Planning and Technology , 29 (4), 295-311. 453
7. “Traffic Analysis Toolbox” (2004). Federal Highway Administration. 454
8. Parsons Brinckerhoff Quade & Douglas, Inc. (2005). Transportation Models and Data 455
Initiative: New York Best Practice Model (NYBPM). New York Metropolitan Council. 456
9. “New York Best Practice Model Final Report” (2005). New York Metropolitan Council 457
10. “2005-2030 Regional Transportation Plan” (2005). New York Metropolitan Transportation 458
Council. 459
11. “TransModeler User’s Guide” (2009). Caliper Corportation 460
12. Silas, M.A. & Holguin-Veras, J. (2009). A Behavioral Micro-Simulaion for the Analysis of 461
Off-Hour Delivery Prolicies. Transportation Research Board Annual Meeting, Washignton , 462
DC. 463
13. Holguin-Veras, J,R., Silas, M.A, et al. (2006). An Investigation on the Effectiveness of Joint 464
Receiver-Carrier Policies to Increase Truck Traffci in the Off-Peak Hours: Part I: The 465
Behavior of Receivers.” Networks and Spatial Economics. 466
TRB 2011 Annual Meeting Paper revised from original submittal.
Iyer, Ozbay, Ukkusuri, Yushimito, Morgul, and Holguin-Veras 22
14. Holguin-Veras, J,R., Silas, M.A, et al. (2006). An Investigation on the Effectiveness of Joint 467
Receiver-Carrier Policies to Increase Truck Traffci in the Off-Peak Hours: Part II: The 468
Behavior of Carriers.” Networks and Spatial Economics. 469
TRB 2011 Annual Meeting Paper revised from original submittal.