Towards Continental Freight Transportation Planning Models
F Guo, L Aultman-Hall, University of Connecticut, US
The analytical methods for freight transportation are not fully developed and have not gained significant public attention until recently. One of the reasons is that freight flow historically accounted for a small proportion of the total traffic. At the same time, freight traffic is more complicated to predict than personal travel in that numerous factors that are difficult to forecast will influence freight traffic. Characteristics of household and trip makers are the common independent variables to estimate personal trip. Freight traffic, on the other hand, is more closely related to economic activities with employment by economic sector being the commonly used predictor of freight generation. Economic trends, growth, and international trade agreements can have significant impacts on the levels of freight shipments. The new operation management strategies, such as Supply Chain Management (SCM) and Just-In-Time (JIT) manufacturing, also influence the nature of freight transportation and are difficult to forecast.
While most of the current research has focused on the state or regional level, this paper is based on the premise that planning models are necessary on continental scale. The reason lies in the nature of the freight traffic. The freight trip length is usually longer than that of the personal trip. In the United States, for example, the 1997 Commodity Flow Survey (CFS) (US Census Bureau, 1997) found that the average freight trip length for all modes is 472 miles, the average trip length for rail, water, air, and multimodal trips is longer than truck-only trips (Rail: 769; Water, shallow draft: 177, great lakes: 204, Deep draft: 1024; Air: 1380). The long freight trip length leads to a significant proportion of the freight traffic having an out-of-state origin/destination. A region level planning model cannot sufficiently address these external freight trips. The trip length is also one of the key factors that deciding whether or not commodity will be shipped through intermodal or by trucks. The lack of a continental-level focus is particularly problematic for evaluating and deciding intermodal facility site and for studies of modal substitution. Furthermore, most freight highway corridors span several regions, states or countries that require multi regional models. However, with freight with nation-wide or global origins and destinations, it is different to clearly define a regional boundary. Continental models could complement smaller area models by covering national, international and intermodal freight trips and larger facility networks.
A major obstacle for continental freight planning models is the lack of freight flow data. Freight information is hard to collect partly because of the confidential issues. The existing freight data sources cannot satisfy freight analysis requirements due to deficiencies in coverage of commodity types, transportation modes, or lack of geographic details; and it is hard to merge different sources together to a common unit or to compare different data sources across regions or categories. Furthermore, international shipments are often well documented but difficult to aggregate. This paper considers the feasibility of continental freight planning models through development of trip generation models for the continental United States using commodity flow data. Specifically, we consider the appropriate predictor variables, zonal structure and modeling methodology for such large analysis zones.
Employment and population data have traditionally been used to estimate freight generation. The focus of this project was to evaluate the feasibility of producing large scale freight generation models based on the freight flow data and the national zone system from the most refined geography provided with the United States Commodity Flow Survey of 1997. The analysis and modeling conducted in this project indicate that these traditional variables are the best predictors of freight production and attraction of the variables tested. Although the most limiting aspect of this study was the small number of zones: even with a limited number of employment variables, the regression models for predicting tonnage and value of freight generated by zone had R-square measures over 0.8 and frequently over 0.9. The 106 CFS zones consisted of metropolitan areas as well as large state size zones and the appropriateness of modeling freight generation for these zones together might be questioned. However, given the limited number of disaggregate zones available publicly from the 1997 CFS, this assumption had to be made. Even given the differences of zone size and characteristics, the relationship between employment and freight generation was solid. Dummy variables for zone type (metropolitan area versus state) were sometimes but not always significant. Using freight per unit area as the dependent variable produced models with relatively inferior results. Use of Principle Component Analysis did not produce significantly better models from a predictive point of view.
A key objective of this study was to generate spatial explanatory variables for zones using GIS. Six single zone spatial variables were generated to measure highway and railway connectivity and density. The premise for this development was to attempt to produce explanatory variables relevant to transportation planning and transportation infrastructure. These variables were significantly correlated with freight generation and could be considered for nation-wide freight planning models.
Several multizone gravity-form spatial variables were also generated incorporating not only the characteristics of the population and transportation infrastructure of one zone, but also all other zones as a function of distance between the zones. These variables showed very poor correlations with freight generation and are not recommended for use in freight generation models.
Application of artificial neural network (ANN) models to the trip generation data was also conducted in this project. The results are similar to other previous studies of freight generation in that ANN models showed improved results over linear regression. However, in this case the improvement was only slight. We attribute this limited success to the small number of zones and recommend further pursuit of the ANN method when more refined geographic details become available.
Satisfactory freight generation models for the large scale freight generation models in North America are possible based on the existing structure and detail of the CFS database. However, the aggregate zonal structure might be sufficient for freight generation but would be inadequate when applying flows to the network in a traffic assignment model. Therefore the next stages of this work are examining techniques to estimate the optimal zone size and layout for the long haul transportation network.
Association for European Transport