Very few systematic, quantitative and empirical methods exist which allow an impact assessment of urban freight solutions or policies. Especially the behavioral responses of logistic agents are difficult to include systematically in impact assessments. This is mainly due to limitations on urban freight data and absence of validated and proven simulation tools. We present an urban freight simulator: Multi-Agent Simulation System for Goods Transport (MASS-GT). MASS-GT is an agent-based model consisting of a framework of discrete choice and optimization models to describe the logistic choices of shippers, carriers, producers and consumers. The disaggregate level of detail allows the analysis of a wide variety of logistic developments and policies across all or specific logistic segments. The model was first implemented for a study area in the Netherlands and calibrated and validated using a variety of data sources.

Despite the importance of urban freight transportation for the accessibility and livability of cities, few systematic, quantitative and empirical methods exist which allow an impact assessment of urban freight transportation solutions or policies. There is a lack of transparent literature on the full specification and estimation of these models, which not only hampers continued research, but also the development of evidence-based urban freight transport policies. We present the urban freight simulator Multi-Agent Simulation System for Goods Transport (MASS-GT) with its full specification and empirical implementation for a study area in The Netherlands. It concerns an agent-based model based on a framework of discrete choice and optimization models, which describes logistic choices of shippers, carriers, producers and consumers. The disaggregate level of detail allows the analysis of a wide variety of logistic developments and policies across all or specific logistic segments. The model is estimated and validated using a variety of data sources: truck trip diaries, supply/use statistics, an e-commerce demand survey, traffic counts and other relevant statistics. The article presents the full specifications of the model and their empirical estimation, including the data sources used. Also, the validity of the model is evaluated using road freight traffic counts. Finally, examples of applications of the model to case studies are provided.

1. Overview and motivation
E-commerce is a rapidly developing segment within urban goods transportation. While currently making up only 5.5% of all kilometers driven by vans in the Netherlands (CBS, 2018), this segment is experiencing rapid growth. For example, in the Netherlands, the volume of parcel deliveries grew by 20% in 2018 (ACM, 2018). This growth is further accelerated by the current coronavirus pandemic. The increasing importance of this segment is not reflected yet in strategic models used in practice for transportation demand forecasting. Parcel deliveries are not modelled explicitly, which does not only prevent the model users from calculating effects of scenarios and policies related to e-commerce, but it also disregards the unique nature of this segment (e.g. specific depot locations, strong growth rates) and, therefore, leads to flawed reference forecasts of van traffic.
In the scientific literature there are some first examples of simulation models for parcel deliveries (Sakai et al, 2020; Hörl & Puchinger, 2021; Llorca & Moeckel, 2021; Mommens et al., 2021; Reiffer et al., 2021). However, each of these examples lacks an empirical disaggregate demand model, focusses on one (part of a) city rather than a whole region, or focusses on one product/service segment rather than all parcel deliveries in a region. Moreover, little experience is gathered in applying such urban freight simulators for traffic forecasting.

2. Methodology, results and main contributions
To combat these shortcomings of the state-of-practice freight models, we developed a module for last-mile parcel deliveries and used it to explore the impacts of different scenario assumptions regarding parcel demand and scheduling. The module consists of two sub-modules: the parcel demand sub-module and the parcel scheduling sub-module.
The parcel demand sub-module calculates parcel demand based on the households and businesses in each zone. For parcels to households (B2C), a logit model is estimated on the Mobility Panel Netherlands, which included several questions regarding parcel orders in 2017 (Hoogendoorn-Lanser et al., 2015). For parcels to businesses (B2B) an average factor is deduced from aggregate statistics reported by ACM (2018). Once the total number of parcels is determined, they are spread over the different courier companies, based on their respective market shares in terms of number of parcels. Finally, for each zone and courier, the nearest depot is determined; from this depot the parcels will be shipped to their end destinations.
The parcel scheduling sub-module forms round-tours to deliver all the parcels for each parcel courier in the study area. From each depot, simple heuristics (such as nearest-neighbor and 2-opt) are used to form efficient tours that respect the vehicle capacity (in terms of number of parcels) and maximum shift lengths for drivers. This, in turn, leads to trip matrices that can be assigned to the network to arrive at network statistics such as vehicle kilometers and emissions.
The conceptual architecture of this module has been applied in four modelling systems:
(1) the strategic freight model of Flanders (SVRM);
(2) the travel demand model of the Municipality of Amsterdam (VMA);
(3) the module for logistical decision-making of the national strategic freight model of the Netherlands (BasGoed);
(4) and the tactical freight simulator of the HARMONY project for the European Commission (test bed: Province of Zuid-Holland, the Netherlands).
In this research we will analyze the impacts of different scenarios and policies, for this purpose we will use the HARMONY implementation of the model in Zuid-Holland. The scenarios explore the impacts of different developments for:
• increased demand for parcels;
• horizontal collaboration between couriers, with shared use of depots;
• a zero-emission zone in Rotterdam, in combination with consolidation centers at the outskirts of the city.

3. Conclusion and future works

A disaggregate region-wide simulation model for parcel deliveries is necessary to evaluate the impacts of policies and developments in e-commerce. Using the developed model, we can show, for example, that vehicle kilometers do not increase linearly with parcel demand due to increased consolidation and that the impacts of zero-emission zones can be diffuse due to rerouting of van trips.

Future efforts may focus on modelling the whole transport chain of e-commerce, rather than only the last-mile deliveries. Furthermore, a demand model for parcels to businesses in line with the model for households is desired, this would require additional data collection.

1. Overview and motivation

The emergence of e-commerce in the past decade and the surging growth during the pandemic, partially at the cost of in-store shopping, have reinforced the need for a better representation of this type of consumer demand and its effects in urban transportation studies (Reiffer et al, 2021). Since this is a recent development, conventional passenger transport models only model the personal mobility for in-store shopping. Standard modelling tools for large-scale demand forecasts for online and in-store shopping are limited. A proper representation of this demand segment first of all requires an estimate of e-commerce demand, and second the simulation of the delivery of the orders. Jaller and Pahwa (2020) developed both an econometric MNL model for in-store and online shopping and applied it to a synthetic population to estimate externalities of the alternatives. The econometric model explains the preferences for type of shopping but not the total level of product consumption, and the delivery of online orders is estimated on aggregate statistics. Other disaggregate simulation studies only focus on e-commerce demand, without considering the trade-off between online versus in-store shopping, such as Cheng et al (2021). In effect, online ordering may reduce physical movements of people to stores, while increasing the delivery of orders to people’s home addresses. This shift is taking place for many consumer products and groceries as well. Weltevreden and Rotem-Minaldi (2007) show early evidence that e-commerce ordering in the Netherlands increases freight transport, while personal travel decreases marginally. On the side of e-commerce deliveries, the simulation of urban freight transport is a well-studied topic in recent literature (Mommens et al, 2021; Hörl and Puchinger, 2021; Reiffer et al, 2021). However, modelling the demand side of e-commerce is often still minimal.

2. Methodology, results and main contributions

We present an empirical e-commerce demand model that is implemented in an urban freight simulator developed in the H2020 project HARMONY (Kamargianni et al, 2020). This new demand model for e-commerce is now a part of the simulator’s parcel module, which generates delivery tours based on the parcel demand by households and businesses. We estimated an ordered logit model with the demand for e-commerce shipments to households as the dependent variable, based on the assumption that one online order equals one parcel, as a function of personal and household characteristics which are known within the simulator.
A second model, connecting e-commerce with the demand for traditional in-store shopping, is also presented here, albeit not yet implemented in the urban freight simulator. In this model we first estimated total consumer demand separately for groceries and non-groceries, and next an adoption model for e-commerce services. The model has the structure of a two-step logit model: an ordered logit model for the total consumer demand, and next a binary choice model for the choice between online and in-store shopping for each of the shopping occurrences that make up a person’s consumer demand.
The models are estimated on the Mobility Panel Netherlands, the MPN (Hoogendoorn-Lanser et al, 2015). The 2017 wave of the MPN contained additional questions regarding online and in-store shopping that can be used for the estimation of choice models. To make the models suitable for application in the freight simulator, we focused on explanatory variables that differ between locations (i.e., zones in a model). The most important variables in the choice models that explain the spatial pattern of e-commerce demand are household income, age of the respondent (in 10 categories) and urbanization level at household location. Other personal characteristics that do not vary spatially are included if they improve the explanatory power of the models (e.g., gender).

3. Conclusion and future works

Age and household income are important predictors for the adoption of e-commerce and the number of parcels ordered. The age-classes 18-39 have the highest preference for e-commerce ordering. Above 40, the preference for e-commerce steadily declines. Persons living in households in the highest income classes (more than 67,000€ per year) are the most likely adopters of online ordering for both groceries and non-groceries. The urbanization level does not affect the adoption of e-commerce services for non-groceries, but strongly for groceries. This can be explained by the limited availability of e-groceries in less urbanized areas, especially at the time of data collection in 2017.
The presented e-commerce demand model has been implemented in the HARMONY Tactical Freight Simulator where it is used to calculate the number of parcels delivered in an area, the subsequent delivery tours and their effects on traffic and emissions. As the explanatory variables differ between zones, we obtain spatially distinct effects. In a next step, the presented model can be linked to a passenger simulator to jointly model and assess the generation of shopping trips and parcel deliveries. Another important research topic is the formulation of representative growth scenarios for e-commerce demand. As online ordering adoption rates evolve over the coming decade, socio-economic developments alone will likely not be sufficient to explain them. Adequately representing the evolution of these adoption rates in transport models requires a tailored calibration approach.

As part of a broader vision for emission-free city logistics, the city of Rotterdam plans to introduce a zero-emission zone in combination with consolidation hubs at the outskirt of the city to generate a shift to zero-emission vehicles. For the design of this zero-emission zone many research questions arise that require a systematic analysis of the impacts of the transition scenarios on the freight demand patterns, the use and market shares of new (zero-emission) vehicles, and the impacts of truck flow and emissions. As a case study we implemented heterogenous transition scenarios for each logistic segment into the Tactical Freight Simulator from the project HARMONY and analyses the system wide impacts. This model is multi-agent, empirical and shipment based and simulates long term tactical choices (distribution channel choice, shipment size and vehicle type choice, sourcing) and short-term tactical choices (tour formation, delivery times).
Results shows that the impact of UCCs is not trivial: we can see a small increase in vehicle kilometers travelled (VKT) overall: +0.25% which can be attributed to the rerouting of shipments through the UCCs. Calculations confirm that emissions are reduced dramatically, by 90%, inside the ZEZ. At the city scale this corresponds to a reduction of almost 10%, as most freight related traffic is generated by the port and involves long haul HGV transport that do not enter the city center. At regional level the reduction of impacts is very small. More measures are needed of more ambitious reductions in emissions are to be achieved.

As part of a broader vision for emission-free city logistics, the city of Rotterdam plans to introduce a zero-emission zone in combination with urban consolidation centers (UCCs) on the outskirts of the city to generate a shift to zero-emission vehicles. For the design of this zero-emission zone, several research questions arise that require a systematic analysis of the impacts of the transition scenarios on freight demand patterns, the use and market shares of new (zero-emission) vehicles, and the impacts of truck flow and emissions. As a case study, we implemented heterogenous transition scenarios for each logistic segment into the Tactical Freight Simulator from the HARMONY project and analyzed the systemwide impacts. This model is multiagent, empirical, and shipment based and simulates long-term tactical choices (distribution channel choice, shipment size and vehicle type choice, sourcing) and short-term tactical choices (tour formation, delivery times). Results showed that the impact of UCCs is not trivial: we observed a small increase in vehicle kilometers traveled overall of +0.25%, which can be attributed to the rerouting of shipments through the UCCs. Calculations confirmed that emissions reduced dramatically, by 90%, inside the zero-emission zone. At the city scale this corresponds to a reduction of almost 10%, as most freightrelated traffic is generated by the port and involves long-haul heavy goods vehicle transport that does not enter the city center. At a regional level, impact reduction was very small. More measures are needed if more ambitious reductions in emissions are to be achieved.