A shift-share based tool for assessing the contribution of a modal shift to the decarbonisation of inland freight transport

In 2004, Macharis and Bontekoning [23] and Bontekoning et al. [4] stated that intermodal freight transportation research was emerging as a new transportation research application field, arguing that it could and should be a research field in its own right. Bontekoning et al. [4] therefore proposed several research needs, including research into policy formulation and evaluation, and research into methods and techniques for addressing the problems in intermodal freight transport. This paper meets these two research needs by proposing a tool that can help policy makers in the field of freight transport to assess the background of a modal shift.

Since 2004, research of intermodal freight transport has matured. Given the scope of this study, a complete review of the intermodal freight transport literature is not appropriate. Instead, in our discussion of the literature we have only included studies that analyse both modal shift and the resulting change in CO₂ emissions. We have studied the relevant literature from a methodological perspective; that is, for each study we identified and categorized the method used to analyse a mode shift in freight transport. This facilitates a comparison of existing methods with the method on which our tool is based.

We conducted a literature search in Google Scholar and Scopus, using combinations of the key words ‘modal shift’, ‘climate change’, ‘CO₂ emission’, ‘models’, ‘tools’, ‘European Union’, and ‘inland freight transport’. Having found some relevant studies, we then used forward and backward snowballing to find more relevant work. We do not claim to have included all relevant studies in the literature overview. However, we are fairly certain that we have covered all existing methods for the analysis of a modal shift in freight transport.

One way to classify studies on mode shift is the distinction between macro and micro, as explained by Ruesch [38]. The distinction is based on the spatial level of the data and information used. A macro-approach for analyzing modal shift is based on analyses of aggregated freight flows on regional, national or international levels, using freight flows matrices and characteristics of the transport network. A micro-approach analyses freight flows and logistics/transport chains on the company level, using information on these chains, the behavior of individual companies, and the key factors for the decision making process such as cost, reliability and transport time. Another way for classifying methodologies on mode shift takes the various types of models as a starting point. We have used this latter way as the primary way to structure the discussion, as it better reveals the uniqueness of the methodology behind our tool. Table 2 presents the different methods. In general, four methods can be distinguished: choice modelling, Life Cycle Analysis (LCA), strategic freight transport network models, and decomposition analysis. Studies which cannot be classified in one of those four groups are listed in the category ‘Other methods’. Not all the studies in Table 2 are discussed below. Only those studies which are needed to illustrate a methodology are referred to.

We have included two examples of freight mode choice modelling in our overview. One is based on Stated Preference (SP) data, and the other on Revealed Preference (RP) data obtained in a survey among shippers. Because the data is gathered at the company level, this is clearly a micro-approach. The models include variables such as costs, transport time and transport reliability. The mode shares are generated by using the parameter values in combination with average values for the mentioned variables. Applying these shares to the total amount of ton-kilometers and CO₂ emission factors results in total CO₂ emissions per transport mode. In this way several modal splits, with corresponding CO₂ emissions, can be generated for different policy scenarios. Regmi and Hanaoka [37] explain these steps in detail. A disadvantage of this method is the need for substantial amounts of disaggregated data [44]; consequently, application at the European level is difficult to achieve. Marcucci and Gatta [25] recently proposed an innovative procedure for acquiring stakeholder specific data for discrete choice modeling that can reduce data acquisition time and costs. Although they applied this procedure in an urban freight transport context, it could be transferable to a larger spatial scale, thus rendering discrete choice modeling on the EU level more feasible.

Kim and van Wee [21, 22] proffered Semi-LCA modelling as a means of estimating CO₂ emissions from transport; they consider their LCA assessment as ‘Semi’ because they include emissions from exhaust and production of fuel, but exclude emissions from the construction of infrastructure and vehicle maintenance.² In short, as based on the input data for demand, distance, speed, load factor, weight, vessel type, engine type, and fuel type, the emissions for rail and inland waterways intermodal systems are estimated according to the drayage, long-hauling and terminal operation processes, and for long-hauling only when it involves an all-road solution. Nocera and Cavallaro [32] also use the LCA methodology, and, based on the ‘Well-to-Wheel’ principle, they estimated the future CO₂ emissions of transalpine corridors for several modal shift scenarios. They then used meta-regression to economically evaluate the CO₂ reduction. Given the nature of the inputs, LCA can be deemed a macro-approach.

Models moreover that fall into the category of strategic freight transport network models use a macro-approach. The main characteristic of these models is that they contain several or all steps of the four-step model, which comprises trip generation, trip distribution, mode choice and route assignment (see for example [34]). The first two steps are often supported by an economic model. The TRANSTOOLS model for example contains a spatial Computable General Equilibrium (CGE) model, while the modal split module is aggregate logit [16]. NODUS, which only performs the modal split and route assignment steps, needs OD-matrices, cost functions, and transport networks as inputs. BASGOED, a conventional four-step transport model, has a limited number of zones and commodity types; moreover, its distribution and model choice model coefficients are estimated on aggregate data, and it uses inputs (generation and attraction) from the economy model of another transport model called SMILE+. For more details, we refer to de Jong et al. [16]; they provide a complete overview and description of national and international transport models in Europe. Finally, LAMBIT (Location Analysis for Belgian Intermodal Terminals), a GIS-based location analysis model that allows for ex-ante and ex-post analyses of policy measures in favor of intermodal transport, is built on three main inputs: transportation networks, transport prices, and container flows from the municipalities to and from the port of Antwerp.

In the scientific literature, decomposition analyses was used only once to analyse a modal shift in freight transport (see [33]). The purpose of that study was to analyse changing patterns of intermodal competition. Decomposition analysis was never used to analyse how a modal shift impacts CO₂ emissions. This paper will be the first to use a decomposition analysis for that purpose. Given the use of aggregated data on the sector level, this is clearly a macro-approach.

Lastly, in the category ‘Other methods’, Islam and Zunder [15] apply a mix of methods for analysing modal shift and the resulting drop in CO₂ emissions in several case studies. The case studies focus on several companies and corridors in Europe, and a mix of macro- and micro-approaches are used.

A key observation from the literature overview in Table 2 is that strategic freight transport network models are most frequently used to analyse the CO₂ emission reduction resulting from a modal shift. The advantage of such models is that users usually need not start from scratch. OD matrices, Geographical Information System (GIS)-layers, cost- and choice information are often already available from previous applications. Once used for analysing a certain freight transport problem, the adaptability of such models to analyse other problems is thus high. In order to perform a systematic comparison we have assessed the different methods on several criteria. Table 3 shows this comparison on the basis of the criteria ‘adaptability’, ‘data requirements’, ‘computation complexity’, and ‘accuracy’.

While the adaptability of strategic freight transport network models is high, it can be considered low for the rest of the methods. For example, for decomposition analysis, for every research question the user has to start from scratch gathering and tuning the data inputs. On the other hand, the data requirements for decomposition analysis are also low. Freight transport data on quantities transported by several transport modes by cargo type for two different moments in time is already enough. Data requirements for the other methods are higher. For choice modelling the effort for gathering sufficient and good quality data is relatively large. The same applies to LCA and strategic freight transport network models because different types of data is needed: data on quantities transported, cost data, data on economic development, data on energy production (in case of LCA), etc. The third criterion, computation complexity, is positively correlated with the level of data requirements. Consequently, the computation complexity of decomposition analysis is lower than the other methods. Because choice models produce point estimates and confidence intervals this method is considered most accurate. Regarding the remaining three methods it is more difficult to judge on the accuracy. Often the data inputs as well as the model outputs are of an aggregate nature. We have therefore judged the accuracy of these methods ‘Moderate to Low’, depending on the level of detail of the model used. Overseeing Table 3, the advantage of decomposition analysis lies in its low data requirements and low computation complexity.

A last, but important remark which must be made is that except for decomposition analysis, the methods in Table 2 most often analyse what-if scenario’s based on policy interventions; this concerns the analysis of a potential modal shift for which the cause is known, and not the background details of a change in CO₂ emissions due to a realized modal shift. Nevertheless, this is important knowledge for policy makers in the field of freight transport, as they may want to know if the realized shift is stable, i.e. if the shift is likely to last over the long-term. It makes a difference if the shift is established due to the competitiveness of the less emitting modes or because these modes are - possibly coincidentally - present in the strongly growing freight markets and taking advantage of that fact. Our tool can reveal this background information of a realized modal shift, thereby also contributing to the existing literature on methods for analysing a mode shift in inland freight transport. The tool and its use are illustrated in section 3. The method behind the tool is a type of decomposition analysis, shift-share analysis and is explained in section 2.3. Because this method is used to explain a modal shift, we first describe this shift in the next section.

The modal shift analysis focuses on the Netherlands, using freight transport data in tons, as provided by Statistics Netherlands. More specifically, we use a customized freight transport Origin-Destination (OD)-matrix whose origins and destinations comprise the 12 Dutch provinces, Germany, Belgium, Italy and ‘other’. Because our study is from the Dutch perspective, only those freight transport flows relating to the Netherlands (as an origin, destination or both) are included in the OD matrix. Further, the dataset contains information about distances, cargo types, ports of loading and unloading, and the number of tons transported by road, rail and inland waterways. At this level of detail, the data is available for two years: 2005 and 2014.

The dataset distinguishes between five distance classes, 0–50 km, 50–100 km, 100–300 km, 300–500 km, and more than 500 km. Figure 1 provides a visualization of the modal shift based on tons in those distance classes, presenting the modal splits for the two designated years.

Figure 1 shows that rail and inland waterways increased their shares of the modal split in the longer distance classes (transport over 100 km) between 2005 and 2014. Concurrently, these modes lost freight in the less than 100 km transport market. Considering all distances, the shift was − 3.0 for road, + 0.2 for rail, and + 2.8 for inland waterways, with the figures representing the percentage points change between 2005 and 2014.³ The most striking changes occurred in the 300–500 km distance market. In this segment road lost 28.6 percentage points of its 2005 share, while inland waterways gained 30.9 percentage points. It must be noted that this is relatively small segment, at around 8%, as shown in Table 4. A closer look at the data for this distance market reveals that the gain in modal split for inland waterways can be traced back to the growth of cargo markets 2 and 7, and the gain of market share in cargo markets 0, 2 and 10.⁴ See Appendix 1 for the used classification of numerical cargo markets.

The shift from road to inland waterways is likely related to developments in the energy industry. CCNR [6] reveals an increase in coal transports on the Rhine between 2009 and 2013, which was expected to continue in 2014, owing to the low price of coal. Moreover, several power plants in Germany are located along waterways within a 300–500 km radius of the Port of Rotterdam [12]. The combination of these two findings could be one explanation for the large increase in the share for inland waterways between 2005 and 2014. The share for rail is highest in the transport market of over 500 km, which indicates that rail is cheapest over the longest distances. Moreover, the fact that freight trains can cross natural barriers, like the Alps, and inland waterways cannot, likely contributes rail’s higher share in this distance market compared to the other distance markets.

It appears that at the disaggregate level, changes in modal split can significantly differ from those at the aggregate level. Distinguishing between market segments sheds some light on possible causes for a modal shift. Section 2.3 addresses this point in greater detail by presenting a methodology that decomposes the growth of the transported quantity for each transport mode.

The basis of the modal shift presented in the previous section is a change in the quantities transported by each individual transport mode. Table 5 shows the direction and the size of these changes.

The observed change is negative for road and positive for rail and inland waterways. Of interest is discovering what the possible causes are for these changes. To this end we apply a shift-share analysis to decompose the changes into several components.

The shift-share method [9] derives from the field of regional economics, where it is often used to decompose the regional growth of jobs or productivity (see Nazara and Hewings [29], for example). In our study the variable of interest is the growth in the number of tons transported, and the three transport modes - road, rail and inland waterways - are the objects to which the method is applied. Notteboom and Coeck [33] have taken a similar approach in examining intermodal competition in Belgian inland transportation. Freight transport data from 1980 to 1991 and a shift-share analysis were used to position the main inland transport modes - road, rail, and inland waterways - and assess and partially explain the changing patterns in intermodal competition.

The shift-share analysis decomposes absolute growth in freight transport between two years into a transport market effect, a cargo market effect, and a competition effect for each transport mode. The transport market effect is equal to the expected growth of each transport mode, as if it had developed like the total transport sector. The cargo market effect results from the specialization of a transport mode in growing and shrinking cargo markets. Finally, the competition effect is the result of an increase or decrease of a transport mode’s market share in the cargo markets where it is active. The competition effect is an indicator for transport mode’s competitiveness. The decomposition of each transport mode’s total growth can now be presented in eq. 1. We refer to Table 6 for a description of the symbols.

\( {Q}_i^{t+l}-{Q}_i^t \) is total growth in tons during the period under analysis. The three components are calculated as follows:

In order to conduct a shift-share analysis for each mode, a summation over i is needed, as shown in eq. 5.

In our application of eq. 5, t = 2005 and t + l = 2014. The results of the shift-share analysis is presented in Fig. 2. The horizontal axis shows the various components’ contributions to the total growth.

The transport market effect is, logically, equal for all three modes. This effect is represented by the blue bars in Fig. 2. The decomposition reveals that the overall positive growth for inland waterways was the result of a positive transport market effect, negative cargo market effect and strong positive competition effect. The negative cargo market effect resulted from a specialization in shrinking commodity markets 3 and 8 in the Standard Goods Classification for Transport Statistics 2007 (NST2007). Table 7 shows that these commodity markets experienced the largest absolute decrease. The positive competition effect was mainly due to gains in market share in cargo markets 3, 7, and 8. In total, the inland waterways’ competition effect was positive in eight of the eleven cargo markets, which implies that inland waterways possess unobserved factors that render this mode more competitive. These unobserved factors can be highly diverse. CCNR [6] notes that “the waterways still have spare capacity”, and this was apparent in 2012 when the inland waterways were able to compensate for the lost production of two oil refineries. Inland waterways may therefore be better suited than other modes to handle increases in demand for liquid bulk transport and consequently capturing a larger part of the pie. Another possible cause could be improvements to waterway infrastructure. In 2013 a new lock was opened in the Mittelweser waterway in Germany [41]. Desk research focusing on specific transport markets can thus identify likely causes for the competition effect. Rail’s positive transport market effect was accompanied by a strong positive cargo market effect. In examining the data it appears that rail is specialized in growing cargo markets 2, and a ‘residual’ group ‘0’ consisting of cargo markets 5 and 13–20.⁵ Further, rail benefitted from a small positive competition-effect, which was mainly the result of its strong gain in cargo market 2. Road’s negative overall growth was caused by a negative competition-effect, which was larger than the sum of the positive transport market effect and cargo market effect. Road especially lost freight in cargo markets 3, 7, and 8.⁶