This chapter will show that logistical elements found inadequate or no consideration in most national models until a few years ago. This hampered the accurate mapping of freight transport and logistics as an import influencing factor. An overview of developments in freight transport modelling concerning the integration of logistics will be given in the section below. Thus, the following introduction provides the general frame for the subsequent presentation of different models that consider logistics and transport logistics hubs in particular.
Early attempts towards integrating logistics aspects into models can be found in the field of disaggregated modelling dealing simultaneously with mode choice and logistics choices. In Chiang et al. [7] mode choice decisions are embedded in a larger inventory-theoretic and logistic framework. Winston [50] stated in his review that models in application were lacking logistics aspects since the time his article was published. The paper from Bergman [5], presented at the International Meeting on Freight, Logistics and Information Technology, can be recognized as the starting point of integrating transport logistics into modelling. He proposes a more detailed spatial representation of logistics processes in freight logistics models. Introducing elements of logistics decision-making in freight models took off in the Netherlands in the early 1990s. Furthermore, it has taken years before similar approaches started to be adopted elsewhere [39].
Broadly speaking, there are different models taking logistics into account. These models are currently operated in different countries and, to some extent, across borders. Although transport demand modelling concerning logistical matters has developed enormously in recent years, there are currently only a few models in use that incorporate logistical aspects concretely [27]. Some examples can be found in the British EUNET, the Dutch SMILE or in the Spatial Logistics Appended Module (SLAM) realized in the European model SCENES. The national transport model system implemented in Sweden and Norway (SAMGODS and NEMO) represent prime examples in this domain [43].
Even though there are different articles dealing excellently with integrating logistics into freight transport modelling (see. e.g., [14, 15]) these reviews do not focus on the integration of transport logistics hubs in particular. Almost all papers review international models in a more general way and address the integration of logistics in general. Differing to that, the following part focusses specifically on the integration of transport logistics hubs (see chapter 2) in models in application. Below we will present an overview of models in operation that consider logistics aspects and hubs. We chose to present the following models because until the end of the research those models were the only existing models in application that integrate logistics and logistics hubs to some extent. Therefore, they represent the most interesting models in use in respect to the topic of this paper. Due to the fact that the reviewed models differ in their characteristic (e.g., scale, depth of aggregation, resolution, etc.), the basic functionality as well as the integration of logistics will be explained, at first. Subsequently, the analysis will focus on the integration of transport logistics hubs.
The model applied to the area of Sweden, named SAMGODS, is a model of national resolution and macroscopic scale of analysis. From a certain point of view it can be seen as a mixed model (see next section) when referring to its depth of aggregation. The model is based on several sub-models that take into account developments of economy, trade as well as foreign trade etc. from which it derives traffic generation. The model considers 35 commodity groups and offers 86 predefined transport chains (with 34 possible means of transport) for transportation processes via a multimodal network. Decisions on shipment size, suitable routes and means of transportation are achieved by a logistics module [11, 39, 47, 49].
The logistics module consists of three steps and follows the ADA structure (aggregated-disaggregated-aggregated). Flows of goods between places of production and consumption are firstly provided on the aggregated level. In order to assign them to individual firms they are disaggregated. Consequently, firms’ logistics decisions (shipment size, utilization of collection and distribution centres, etc.) can be simulated in this disaggregated part of the model. To allocate the OD-flows to routes, the data are finally aggregated again [10].
Selecting one of the predefined transport chains, the logistics module sets modes of transport for each section and determines whether transport is accomplished directly or via the utilization of logistics hubs. Transport logistics hubs are also included in the model – defined as locations were goods are transhipped and possibly stored [11]. The logistics module consists of subroutines that develop decisions gradually. Therefore, available transport chains, including optimal transfer points between transport sections, are generated initially in a first subroutine (BUILD CHAIN). This one is followed by a second, which selects transport chains with regard to minimising total logistics costs (CHAIN CHOICE) (see Fig. 2).
These costs include costs resulting from different processes during the transport: costs for loading/unloading processes at the sender/receiver, costs for the transport itself, as well as costs from inventory management, for instance, at warehouses. Due to the fact that the model also includes transhipment processes, linking different legs of the transport chain, the corresponding costs incurred by using transport logistics hubs are also considered (transport logistics hubs like truck terminals, ports, intermodal terminals and airports). Included costs then vary per ton and vehicle type.
Existing information regarding terminal properties (e.g., access to different modes; spatial situation; feasibility of handling different types of goods etc.) is considered as well. Due to economies of scale, for instance, or differences in technologies operating at hubs, the derived costs vary at these nodal points. Therefore, the model distinguishes between different classes of terminals, which are characterized by a technology factor between zero and one. It is based on the assumption that, for example, ports which handle more goods use more advanced technologies [10, 11, 13].
Beside a general integration of hubs, the differentiation of diverse types of transport logistics hubs represents a further step in considering and distinguishing hubs in demand modelling – also with regard to node-specific characteristics within a category [11, 13].
NEMO
(Network Model for Freight Transport) is a national model applied to the area of Norway. Due to its evolution (parallel with SAMGODS), NEMO considers logistics hubs similarly to SAMGODS. Thus, the model represents an extension of the Swedish one to the spatial area of Norway and, therefore, will not be examined separately.
The Dutch model SMILE (Strategic Model for Integrated Logistics Evaluation), which predicts traffic flows at the national level, was one of the first models to consider logistical aspects.
SMILE simulates transport flows by taking economic developments into account and linking the economy, logistics and transportation. It was the first model especially developed to include distribution centres into the routing processes of commodity flows [39]. Land use (via production), trade (via sales, sourcing), logistics (via inventory) and transport are coupled across four stages [40].
The impact of logistics hubs, which is represented by distribution centres in this case, becomes noticeable by considering hub characteristics, and the attributes of goods and their requirements in terms of inventory, handling and transportation. With respect to this, 542 different types of products are clustered into 50 logistics families. The clustering process is based on certain characteristics of the product (e.g., value of goods, packing density, perishability, delivery time, shipment size etc.). The characteristics of these logistics families impact the potential and capability to handle certain types of goods. Therefore, hubs influence distribution chains and transport flows, and provide a spatial redistribution of the flows of goods through handling and stock rates, packaging density and volume to weight related to hub characteristics (see Fig. 3) [40].
The mapping of distribution centres with appropriate opportunities of consolidation and the resulting transport flows based on characteristics of goods and hubs is one way to include hubs in transport demand modelling. Similar to other models, trade and transport relations are linked with inventory and storage services in SMILE. However, transport logistics hubs are not considered in particular.
The Spatial Logistic Appended Module (SLAM) is integrated in the European model SCENES (trans-boundary macroscopic model for the EU). During the development, basic ideas and experiences from the Dutch model SMILE were consulted. SLAM is designed to evaluate the effects of changes in logistics and in the transport system across Europe. Therefore, a major application is the detection and location of distribution centres in Europe. Accordingly, SLAM ought to give a more accurate picture of transport flows involving logistical matters. The model considers changes in distribution structures (e.g., number and location of utilized intermediate warehouses for distribution) and incorporates them into the distribution flows [11, 16, 43].
SLAM receives production and consumption flows (e.g., from SCENES) and produces origin–destination-matrices (OD-matrices) that integrate alternative distribution chains. In this context a distribution chain is defined as the combination of distribution centres and transportation relations for trade flows between producer and consumer region. In this manner, a main function of the model is the consideration of alternative distribution chains (production – distribution centre – consumption) (see Fig. 4) [16, 39].
In order to determine alternative distribution chains, located hubs are listed in accordance to characteristics of products, markets and transport services. A location score module then calculates a score for each region related to its attraction as a possible location for distribution centres (based on economic activity, centrality and accessibility to infrastructure etc.). Afterwards, a chaining module selects the most attractive regions for distribution centres and constructs logistics chains via these centres. Furthermore, logistics costs are calculated for each single chain composed of transportation, inventory and other logistics costs. The construction of transport chains thus follows the approach of minimizing total costs [9, 26].
Returning to the model structure, it can be stated that SLAM achieves a more accurate picture of transport flows by integrating distribution logistics hubs in the transportation system. Since the hubs considered represent distribution centres, the picture is not adequate regarding transport logistics hubs. In addition, SLAM does not go into detail concerning networking because flows act strictly cost-rationally and, therefore, take the most cost-effective way through the abstract distribution-consumption-net.
EUNET
is a regional model developed in the UK. It covers the goods transport within central Great Britain (Trans Pennine Corridor) as well as imports and exports from and into the region. Similar to SCENES, whose principles served as orientation in the development process, EUNET provides a link between regional economics and logistics. The aim of the model is to predict freight transport demand as a function of economic transactions and freight logistics [11, 14, 43].
Analogously to other models, EUNET contains a logistics module, which serves as a link between PC-matrices and OD-matrices [39]. Since logistics hubs (consolidation centres, national/regional distribution centres, major ports, local depots etc.) – here mostly defined as distribution logistics hubs or special generators – are included in the formation of OD-matrices, they play a crucial role. The distribution of commodity flows through logistics hubs is, comparable to other models, based on the properties of each hub (e.g., warehouse floor space data). Thus, logistics steps are explicitly considered by relating handling factors of products and characteristics of logistics hubs [22, 41].
In this way, multiple distribution channels arise via a plurality of possible hubs (see Fig. 5) [22]. The consideration of distribution logistics hubs is achieved by the inclusion of hub characteristics and goods attributes. Nevertheless, transport logistics hubs, however, lack consideration here.
The Los Angeles Freight Forecasting Model – LAMTA – is a multimodal transport demand model. Although the model for the Los Angeles Metropolitan Area (LAMA) mainly focusses on road freight transport (trucking), it also includes a multimodal framework to support freight transportation decisions and, hence, logistics hubs [3, 17].
Besides the sub-components for trip generation, trip distribution, and mode choice, LAMTA integrates a separate module explicitly aimed at modelling logistics hubs which are not included, for instance as special generators. The Transport Logistics Node Model (TLN) incorporates warehouses, distribution centres and intermodal terminals into its modelling. However, this affects long-distance freight transport – precisely defined as flows between zones situated inside the LAMA and external regions (outside LAMA). Commodity flows occurring entirely within the study area are not modelled by the TLN. Therefore, transport logistics hubs within the study area are defined as hubs where chains for long-haul transportation are formed (see Fig. 6). Some examples for these transport logistics hubs are intermodal freight yards, truck terminals and other loading facilities [3, 17].
The TLN module is based on two elements. First, characteristics of the logistics hubs are described by the TLN module. After that, OD-matrices are fed into the TLN in order to generate separate matrices for each combination of transport mode and commodity. For that purpose, the commodity flows are split into two segments. Long-haul traffic carried out by truck, rail or ship is distinguished for each direction (inbound/outbound). Thus, the percentage of shipments passing each hub is calculated for each product group and direction. In addition, a further allocation to short-distance transport, which is performed by trucks only, is carried out as well [3]. The final output of the TLN-module is four matrices per transport mode and product group (direct short-distance transport without TLN; direct long distance transport without TNL; long distance transport from/to TLN, short-distance transport from/to TLN).
Finally, transport logistics hubs are only considered in LAMTA when commodity flows cross the LAMA border.
Due to the appointment of the Chicago Metropolitan Agency for Planning (CMAP), a powerful and innovative prototype of meso-scale freight model was developed for Greater Chicago.
The model consists of a macro-scale model that generates commodity flow data on macroscopic level using economic modelling tools. Its’ output serves as input for the meso-scale model that breaks down the high-level commodity flows on shipments between individual companies by agent-based analysis and disaggregated choice modelling. Within this, the demand in the Chicago region is explained by several steps (generation of firms for production and consumption; creation of individual relations between firms; disaggregation of macroscopic flows of goods to annual shipments between buyer and seller; choice of transport and logistics paths). By means of the output of the meso-scale model, a dynamic multi-modal route assignment is made in a last step and detailed trips are generated in a micro-scale environment [45, 46].
In addition to data from the macro scale model, there is also input data coming from a network model. As a result of this, logistics facilities and hubs (intermodal terminals, rail terminals, container and loading terminals, distribution centres as well as other freight hubs, airports and water ports and also large firms) are included in the evaluation of logistics and transportation decisions [31, 45].
Several distribution channels via hubs were identified using data collected by a national survey (FAME project). Shipments that require, for example, intermodal loading facilities, warehouses or distribution centres are assigned to corresponding hubs. Costs that emerge at these hubs (e.g., intermediate handling, inventory, deterioration, damage and ordering or stock out costs) – based on hub characteristics – influence the total costs of the different shipment types [33]. Following the steps mentioned above, transport decisions are ultimately based on evaluation of total transport and logistics costs relating to available paths [31].
The approach of the model for the CMAP follows mainly the research of de Jong and Ben-Akiva and their comprehensive accounting of transport and logistics costs and is, therefore, similar to the aggregated-disaggregated-aggregated (ADA) approach used in SAMGODS and NEMO.
The concept for the freight transport model FAME (Freight Activity Micro simulation Estimator), developed at the University of Illinois (Chicago), is a fairly new conceptual modelling framework. It was designed for behavioural freight transportation modelling and considers logistical elements in detail. Therefore, it focusses iteratively on the benefit of different types of intermediate handling facilities, mode choice, and shipment size [36].
Initially, individual decision-makers with their specific characteristics and geographical distribution are introduced in the firm-type generation. Trade relations between firm-types are determined in the following module (supplier selection). The subsequent determination of shipment size and frequency is based on observed shipment size distributions. A probit model is used for behavioural mode choice before the commodity flows are finally assigned to the network [34].
Logistics aspects and hubs are taken into account via considering logistics transport chains. Thus, the number of stops and mode choice are determined for each of the corresponding shipments and chains. In addition, logistics hubs are defined and optimal shipment sizes are determined for each transport chain passing related hubs. There are different combinations and transport chains to connect suppliers and buyers. Apart from direct transportation, transport chains may pass intermodal terminals, distribution and consolidation centres with various combinations of these within the chain [34, 36].
Characteristics of commodities (e.g., shipment size) and costs are the crucial variables that influence the utilization of logistics hubs in the model.
Only few agent-based or rather actor-based models have been developed in order to incorporate behavioural elements of logistics actors [35]. The actor-based approach to commodity transport modelling developed by Liedtke (InterLog) is such an approach. The model involves both commodity and vehicle-related aspects and, therefore, combines two in modelling mostly parallel existing approaches.
The agent-based approach (integrates logistical elements into modelling road freight transport by considering decisions taken by individual actors. Senders and receivers are classified according to produced and consumed commodities in the model. They are equipped with appropriate behavioural parameters. Logistical strategies and decisions are included in transport-related decisions by emerging total logistical costs. These costs include ordering and communication costs, inventory costs, costs for loading and unloading as well as general transport costs. Furthermore, total logistics costs are influenced by goods’ characteristics and business relationships that are crucial to distribution and transport deals between actors.
Choices of agents (firms, freight forwarders) are modelled in a market interaction model in order to examine their interactions within the transport market and their ambitions to maximize their profits by minimizing total logistics costs. Minimizing cost also involves adjusting delivery frequencies and the contracting of suitable transport companies. In this way, the InterLog model represents one of the first models to integrate aspects of behaviour in terms of microscopic modelling. [26, 27].
Although the model illuminates transport behaviour of the transport company and, thus, related logistical aspects, the utilization of transport logistics hubs is, however, not regarded. There is no explicit consideration of this aspect because the model focuses primarily on the decisive logistical factors of shipment size and existing contracts between senders/receivers and transport operators. Complex logistical reaction patterns between actors are examined in this way. However, transport logistics hubs are not linked with actors and, therefore, not represented in the model.
GoodTrip - the model applied to Groningen (Netherlands) - is an urban freight transport model. It builds logistics chains by linking the activities of consumers, supermarkets, hypermarkets, distribution centres and manufacturers and is fundamentally based on consumer demand [39].
The four component model (components: spatial organization of activities; freight flows; transportation; infrastructure) calculates the volume of goods per commodity group for each spatially defined zone [6]. Thus, the attraction of goods between consumers and producers is determined. Within this determination the flows of goods are influenced by the spatial distribution of activities and the market share of each activity group (consumers). Afterwards a classification of commodity groups is carried out and an OD-matrix is created. In a final step, vehicle trips are generated and assigned to a network [35, 39, 43].
Although the model considers logistics aspects, transport logistics hubs are not covered explicitly, for example as loading facilities or similar. Indeed concepts of urban distribution centres can be considered in different scenarios. Scenarios for urban logistic distribution centres are one example here (see Fig. 7) [6].
The impact of logistics hubs in general and transport logistics hubs in particular on transportation is, however, disregarded within the study area, from a large-scale perspective.
The urban/regional model developed by the IVU Traffic Technologies AG called WIVER found its application in several cities and their surrounding area, including greater Hamburg and Berlin. It offers the ability to consider transport effects of logistics hubs [38].
The model mainly focuses on road transport and calculates freight transport demand in four steps. Thus, originating traffic is identified and derived from structural data related to traffic areas (e.g., economic sectors, number of active driving employees of companies). The originating traffic is then determined by the average number of trips and destination. In a second step, the terminating traffic potentials are calculated based on the distribution of industries and recipient structure of these industries. These characteristics are used to weight originating trips to destination. Subsequently, origins and destinations are linked by their volume and potential as well as taking into account distances, for instance. In a final step, tours are generated and assigned to a network model [21].
Due to the incorporation of the enterprises’ mobile employees (conductors) into the model parameters, logistics hubs are considered in a special way. However, hubs are integrated in trip generation, only, so that they are not included in tour construction as intermediate stops within transport chains, for instance.
Table 3 provides a summary of all models presented here and gives an overview of the model specifications and the integration of different types of logistics hubs.
Further details of models can also be found in de Jong et al. [15], which provide a more general overview over some models (e.g., SMILE, SAMGODS etc.). In contrast to de Jong et al. [15], Table 3 focusses on the integration of logistics hubs and transport logistics hubs in particular. We can summarize that most models integrate distribution logistics hubs (DLH) but, however, transport logistics hubs (TLH) are integrated in few models, only. The combination of Table 1, 2 and 3 gives an insight in how transport logistics hubs are integrated in freight transport models in application.
In order to summarize the insights into the different models it can be stated that there are different ways to integrate transport logistics hubs in modelling.
A rather basic method is the integration of hubs as sources and sinks in models. Thus, logistics hubs are considered in a simple way as so-called special generators or singular traffic generators whose transport volume is supplied/defined externally.
The integration of transport logistics hubs via logistics modules, which select between several predefined transport chains, is a more sophisticated method. A decisive influence on the consideration of hubs within logistics modules is the properties of shipments (e.g., commodity, shipment size). Due to the characteristics of hubs there are often limits in handling certain shipments. If hubs are not suitable to handle a certain shipments, the likelihood of transportation via these hubs will be reduced. Different transport chains, processing only via certain hubs, are the result of that consideration. In this manner, the characteristics of hubs in combination with the characteristics of shipments determine the utilization and impact of hubs or transport demand respectively. Another aspect of integrating logistics hubs into modelling is the inclusion of further hub characteristics that excess the characteristics mentioned above (capability of handling certain shipments). The integration of information about differences in technologies used at hubs, for instance, can turn out to be very helpful if used technologies vary significantly. This way offers the possibility to a more nuanced consideration of hubs if they differ regarding their characteristics.
Considering the gained information, we will discuss the results and its broader meaning as well as consequences for transport demand modelling in the next section.