1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
A Behavioral Freight Transportation Modeling System
Zahra Pourabdollahi
Ph.D. Student
Department of Civil and Materials Engineering
University of Illinois at Chicago
842 W. Taylor Street
Chicago, IL 60607-7023
Tel: (312) 996-0962
Email: zpoura2@uic.edu
Amir Samimi, Ph.D.
Assistant Professor
Department of Civil Engineering
Sharif University of Technology
Azadi Avenue,
Tehran, Iran
Tel: +98 (21) 6616-4219
Email: asamimi@sharif.edu
Abolfazl (Kouros) Mohammadian, Ph.D.
Associate Professor
Department of Civil and Materials Engineering
University of Illinois at Chicago
842 W. Taylor Street
Chicago, IL 60607-7023
Tel: (312) 996-9840
Email: kouros@uic.edu
Kazuya Kawamura, Ph.D.
Associate Professor
College of Urban Planning and Public Affairs
University of Illinois at Chicago
412 S. Peoria Street
Chicago, IL 60607-7064
Tel: (312) 413-1269
Email: kazuya@uic.edu
Word Count: 2,162 + (3*250) = 2,912
1
1
1. Introduction
2
Freight movements have significant impacts on transportation system, regional wellbeing and
3
economic growth. Because of the complexity of decision-making in freight system, sophisticated
4
modeling tools are used for forecasting, policy assessment, and counteracting the impacts of
5
freight movements. Unlike the passenger transportation subject, freight transportation is a
6
relatively less-researched field in terms of advanced modeling applications. However, this has
7
been recognized by researchers around the world and the field is rapidly developing in many
8
directions, including data collection, modeling frameworks, and operational strategies. There are
9
still significant gaps in terms of our understanding of the fundamentals and the nature of freight
10
movement systems and their behavioral decision making process.
11
Review of the literature revealed that many available freight transportation models have
12
aggregate nature. Moreover, many of the current models are short in term of logistics elements
13
such as, use of intermediate handling facilities, determining optimum shipment size, mode
14
choice, and multi-modal shipments. Many researchers have discussed the drawbacks of the
15
aggregate models in capturing the complexity of freight policy systems (1 and 2). In particular,
16
they cannot replicate the supply chains and logistics decisions made by individual actors in the
17
freight system. Recently, there have been some advances in developing logistics models at the
18
urban level with a primary focus on road transportation. The urban logistics models are effective
19
analysis tools at the local level. However, a more comprehensive framework and detailed
20
behavioral models are needed to assess policies at the national level and to examine the effects of
21
changes to the system-level elements (e.g., expanding ports capacity, or widening the Panama
22
Canal) on regional/national level freight movement.
23
This paper introduces a behavioral microsimulation framework for freight transportation
24
modeling named FAME (Freight Activity Micro-simulation Estimator). FAME has a wide
25
geographic coverage and to the best of the authors’ knowledge, it is the first comprehensive
26
nationwide freight microsimulation model in the U.S. The FAME model uses a fresh approach in
27
modeling freight demand by incorporating a modular structure and working at the disaggregate
28
level of firm as the basic decision-making unit in the freight market. The large-scale model
29
simulates commodity movements in the U.S. at the disaggregate level of firm-to-firm. Data
2
1
scarcity appears to be a crucial barrier in developing such a disaggregate logistics model. A set
2
of publicly available datasets are utilized to develop the model. In addition, more detailed
3
information concerning individual shipments is collected through an online nationwide
4
establishment survey, conducted by the research team (3).
5
6
2. Freight Activity Micro-simulation Estimator (FAME)
7
FAME was introduced as a freight activity-based modeling framework with five basic
8
modules(4). This paper provides very brief overview of this framework. Interested readers may
9
refer to Samimi et al. (4) for further elaboration of the FAME framework. FIGURE 1 illustrates
10
FAME framework. In the first module, all firms in the study area are synthesized and their basic
11
characteristics, including industry type, employee size, and location are identified. Based on each
12
firm's characteristics, types and amounts of incoming and outgoing goods are determined, and
13
trade relationships are formed for each pair of supplier-buyer firms in the second module. Third
14
and forth modules deal with logistics choices. In the third module, the shipment sizes are defined
15
based on the acquired information on the firms’ characteristics, and shipping modes are
16
determined in the fourth module. Finally, in the last module, the impact of the goods movements
17
on transportation network is investigated.
18
Four categories of data are required for developing FAME: information on business
19
establishments for the first module, aggregate freight flows (OD matrix) for the second module,
20
detailed information on a sample of individual shipments and supply chains for the third and
21
forth tasks, and specifications of the transportation networks for the last module.
22
23
24
3
Firm-type Generation
Identifying individual decision-makers
Supplier selection
Determining relationships between firm-types
Logistics decisions
Shipment size determining
Based on observed shipment size distribution
Mode choice
Select mode of transportation
Network analysis
Assigning commodity flows to the traffic
network and assess the impacts
1
2
FIGURE 1. FAME OPERATIONAL FRAMEWORK
3
4
Firm-types Generation
5
As discussed earlier, FAME simulates freight flows at the disaggregate level of firm-to-firm.
6
Thus, the decision-makers in this microsimulation are individual firms in the U.S. Since, there
7
are more than 239,000 firms with six digit North American Industry Classification System
8
(NAICS) industry type in the U.S. (5). Some form of aggregation is inevitable to keep the
9
computational burden at a reasonable level and diminish the need for highly disaggregated data.
10
We proposed to aggregate the firms based on firm-types. A firm-type is a collection of firms
11
with similar location, industry type, and establishment size. It is assumed that firms with the
12
same characteristics have the similar behavior in freight decision-making process. Number of
13
firm-types can differ based on the number of industry types, establishment size, and geographic
14
zones in the study area. Considering 123 domestic zones, 328 industry classes (NAICS), and
4
1
eight employee size groups in this simulation, a total of 45,206 firm-types were identified in the
2
U.S.
3
4
Supplier Selection
5
Trade relationships were replicated based on a fuzzy expert system that was developed by the
6
research team (6). This model scores the appropriateness of all the possible suppliers for a given
7
firm-type. Having the likelihood of partnership for any pair of supplier and buyer, annual
8
commodity flows is disaggregated from geographic zone level into firm-type level. For a given
9
origin, destination, and commodity type, the value of total annual tonnage was obtained from
10
Freight Analysis Framework data (FAF3.0) (7) and disaggregated between the top five percent of
11
supplier and buyer pairs with the highest appropriateness score.
12
Two-digit Standard Classification of Transported Goods (SCTG) used in FAF is also
13
utilized in this study to classify the commodities. All FAF industry sectors were considered in
14
FAME, but some of them were not present in specific zones in the simulation. This is due to the
15
limitations of business establishment data sources and also the crosswalks that were used in the
16
second module to connect industries to commodities. As a result, not all of the FAF commodity
17
flows between the zone pairs was allocated to firm-types. In total, freight flows for 40 classes of
18
commodity types is simulated in FAME.
19
20
Shipment Size Determination
21
A shipment size model, developed by the research team (6), provides a categorical output
22
variable with three clusters: small (less than 1,000 lb), medium (1,000-50,000 lb), and large
23
(more than 50,000 lb). Establishment size of the supplier and buyer, shipping distance and
24
commodity type, are the inputs of the model. This model was applied on the annual commodity
25
flow between each pair of supplier and buyer from the former section and determined share of
26
small, medium, and large shipments accordingly. However, knowing that a shipment is small is
27
not enough to run the modal split in the next module and a crisp value should be assigned as
28
well. In order to do so, a distribution of observed shipment sizes from the UIC National Freight
5
1
Survey (3) was obtained for each class of shipment size. Details of the shipment size model and
2
the shipment size distributions are elaborated elsewhere (6).
3
4
Modal Split
5
A binary probit mode choice model was deployed in this simulation to determine the share of
6
truck and rail (including truck-rail intermodal) for each shipment. This model has a proxy for
7
distance, weight, relative impedance between truck and rail, a dummy for containerized
8
shipments and commodity type as input variables all of which have to be determined for every
9
simulated shipment. In this simulation the overall probability of having containerized shipments
10
was assumed to be 11.8%, according to the UIC National Freight Survey (3). Although the
11
binary mode choice has a satisfactory goodness of fit, it underestimates the total number of rail
12
shipments. Therefore, the estimated probability of a rail shipment is increased by a 1.3 factor to
13
cover this underestimation in general.
14
Modal split of domestic freight movements in the U.S. was performed by calibrating and
15
simulating the first four modules of FAME. As mentioned, only truck and rail (including truck-
16
rail intermodal) are covered as the primary modes of freight transportation, and the relative
17
percentages of total tonnage, value, and ton-mile of commodities between the two modes are
18
estimated. Due to the random nature of the microsimulation, the simulation was repeated several
19
times. Although the tonnage of the shipments carried by each mode is obtained directly from the
20
model, the dollar value of the shipment is approximated using FAF3.0 dataset.
21
22
3. Simulation Results
23
Some of the major results of this microsimulation are presented in this section. Since there is no
24
disaggregate freight data available, the results are aggregated and validated with public aggregate
25
datasets. TABLE 1 shows results of the modal split model in FAME. It presents the relative
26
percentages of total tonnage, value, and ton-mile of commodities between two modes simulated
27
in FAME and compares them with FAF3.0 and Commodity Flow Survey (CFS) estimates. It
6
1
should be noted, that modal split information of none of these datasets has been used in model
2
calibration and thus can be an appropriate base line for validation.
3
4
TABLE 1. MODAL SPLIT VALIDATION IN FAME
Item
Tonnage
Value
Ton-mile
Rail
Truck
Rail
Truck
Rail
Truck
CFS 2007
FAF3.0
FAME
19%
81%
7%
93%
53%
47%
16%
84%
18%
82%
48%
52%
16%
84%
18%
82%
48%
52%
5
6
Moreover, FIGURE 2 compares simulated commodity flows in FAME with FAF estimates
7
by commodity type. Based on FAF dataset, a total of 14,845 million tons of commodities valued
8
at around 12,276 billion dollars is transported between the domestic origin and destinations on
9
truck, rail, or truck-rail intermodal (7). In total, 14,202 million tons of commodities valued at
10
around 10,916 billion dollars is simulated in FAME. Thus, more than 95% of FAF domestic
11
tonnage and around 89% of commodity values is simulated in this study. FIGURE 2 shows the
12
result of second module (supplier selection) where trade relationships are formed and commodity
13
flows are simulated between firm-types.
14
Cereal grains
Other ag prods.
Animal feed
Meat/seafood
Milled grain prods.
Other foodstuffs
Alcoholic beverages
Tobacco prods.
Building stone
Natural sands
Gravel
Nonmetallic minerals
Metallic ores
Coal
Crude petroleum
Gasoline
Fuel oils
Coal-n.e.c.
Basic chemicals
Pharmaceuticals
Fertilizers
Chemical prods.
Plastics/rubber
Logs
Wood prods.
Newsprint/paper
Paper articles
Printed prods.
Textiles/leather
Nonmetal min. prods.
Base metals
Articles-base metal
Machinery
Electronics
Motorized vehicles
Transport equip.
Precision instruments
Furniture
Misc. mfg. prods.
Waste/scrap
2,500,000
1
2
2,250,000
FAF (KTon)
2,000,000
FAME (KTon)
1,750,000
1,500,000
1,250,000
1,000,000
750,000
500,000
250,000
0
FIGURE 2: COMMODITY FLOWS (KTON) (COMPARISON BETWEEN FAME AND FAF)
1
1
4. CONCLUSION
2
The primary motivation for this research was to develop a behavioral freight transportation
3
model in the U.S. Therefore, a nationwide freight activity microsimulation was conducted. A
4
major drawback of many previous studies of this kind is the aggregate nature which prevents the
5
development of an actor-based microsimulation. A large-scale behavioral microsimulation
6
framework, named FAME was developed in this study. This framework incorporates firms’
7
characteristics in replicating shipping behaviors, and aimed at paving the way for future
8
behavioral freight microsimulation efforts. Final results of simulation showed a close match with
9
CFS 2007, and FAF3.0 data. This study aimed at facilitating a sound microsimulation freight
10
model as a valid forecast tool that eventually could contribute to more reliable policy
11
assessments compared to currently available decision tools.
12
FAME is largely based on public freight data in the U.S. and therefore data collection costs
13
are substantially mitigated. Furthermore, this study covered a wide range of industry classes and
14
commodity types. FAME has a unique geographic coverage, as well, and to the best of the
15
authors’ knowledge, it is the first comprehensive nationwide freight microsimulation in the U.S.
16
This study presented the first steps toward a comprehensive activity-based freight
17
microsimulation. There are still several components that need further exploration.
18
19
ACKNOWLEDGMENTS
20
This work was (partially) funded by the National Center for Freight and Infrastructure Research
21
and Education (CFIRE) based at the University of Wisconsin. CFIRE is a national university
22
transportation center supported by the US Department of Transportation. Partial funding was
23
also provided by Illinois Department of Transportation (IDOT). All responsibility for the
24
contents of the paper lies with the authors.
25
26
2
1
REFERENCES
2
1. Chow, J.Y.J, C.H. Yang, and A.C. Regan. State-of-the Art of Freight Forecast Modeling:
3
Lessons Learned and The Road Ahead, Transportation 37(6), pp. 1011–1030, 2010.
4
2. Regan, A.C., and R.A. Garrido Modeling freight demand and shipper behavior: state of the
5
art, future directions. In: D. Hensher, Editor, Travel Behavior Research: The Leading Edge,
6
Pergamon, Amsterdam, pp. 1–30, 2001.
7
3. Samimi, A., A. Mohammadian, K. Kawamura, “An Online Freight Shipment Survey in the
8
United States: Lessons Learned and Nonresponse Bias Analysis”, Proceedings of the 89th
9
Annual Meeting of the Transportation Research Board (DVD), Washington, D.C., January
10
11-15, 2010.
11
4. Samimi, A., A. Mohammadian, and K. Kawamura. A behavioral freight movement
12
microsimulation model: method and data. Transportation Letters: The International Journal
13
of Transportation Research, pp. 53-62, 2010.
14
5. U.S. Census Bureau: Economic Planning and Coordination Division. County Business
15
Patterns. Available at: http://www.census.gov/econ/cbp/download/index.htm, accessed
16
February 2011.
17
6. Samimi, A., A Behavioral Mode Choice Microsimulation Model for Freight Transportation
18
in the U.S. PhD Dissertation, College of Engineering, University of Illinois at Chicago, 2010.
19
7. Federal Highway Administration. The Freight Analysis Framework, Version 3: Overview of
20
FAF3 National Freight Flow Tables, 2010.