Decarbonizing Freight Through Intermodal Transport: An Operations Research Perspective—Part I: Methodological Foundations and Model-Driven Insights

Abstract

Intermodal transportation (IMT) has long been recognized as a key strategy for decarbonizing freight transportation (FT), which is one of the most polluting sectors worldwide. While IMT has been extensively examined using operations research (OR) methods, the integration of decarbonization objectives has only recently gained momentum. Despite this growing interest, to the best of our knowledge, no prior comprehensive review has systematically synthesized OR methodologies specifically addressing IMT decarbonization. To address this gap, we conduct a systematic literature review of OR studies on IMT decarbonization and organize the survey into two complementary parts. Part I focuses on methodological foundations of OR applications in IMT decarbonization. We classify studies by problem type and OR technique, analyzing modeling characteristics, solution approaches, and uncertainty treatment. Our analysis reveals that exact methods dominate the literature (41% of studies), while meta-heuristics show rapid recent growth with 50% of studies published recently. Approximately 20% of studies incorporate uncertainty, and they are predominantly demand-focused. We identify critical research gaps including limited multistage stochastic frameworks to capture cascading uncertainties, insufficient attention to terminal operations and network reliability, and the underutilization of emerging technologies such as reinforcement learning and digital twins. This systematic synthesis establishes the current state of OR methodologies in IMT decarbonization and provides a foundation for future innovations in sustainable freight systems.

1. Introduction

Freight transportation (FT) is a vital component of the global supply chain (SC), facilitating commercial relationships among stakeholders (e.g., suppliers, distributors) across regions. However, its broad application leads to significant environmental consequences, including increased greenhouse gas (GHG) emissions [1]. The global FT industry is responsible for up to 11% of worldwide GHG emissions [2] with road freight generating over 100 times the CO₂ per unit of freight distance compared to maritime transport and accounting for nearly 80% of global diesel consumption growth. In response to this challenge, environmentally conscious solutions have prompted a paradigm shift in FT toward diverse transportation configurations. These endeavors are exemplified by intermodal, multimodal, and synchromodal transport (IMT, MMT, and SMT, respectively), which integrate multiple modes (e.g., road–rail, rail–waterways) of transportation [3]. Specifically, IMT involves multiple carriers and contracts for different segments; MMT uses a single contract with one provider managing multiple modes; and SMT dynamically adjusts modes in real time. Empirical evidence consistently shows that these configurations achieve substantially lower carbon intensity than truck-only transport with reductions of about 46% on average and up to 45–60% for road–rail or road–waterway routes [4,5]. Another detailed case study of an intermodal container journey between China and the US shows that using rail for inland segments in place of trucks substantially lowers total freight emissions [6].

Recognizing this potential, governments worldwide have expanded programs to minimize IMT emissions. For instance, the European Union (EU) established modal shift targets in its “White Paper on Transport”, aiming to move 30% of road freight over 300 km to rail or waterways by 2030 and 50% by 2050 [7]. The United States (US) Advanced Research Projects Agency-Energy (ARPA-E) has launched initiatives promoting low-carbon transport modes through technological and data-driven solutions [8], while China has accelerated sea–rail IMT development under its “dual carbon” strategy targeting carbon peak by 2030 and carbon neutrality by 2060 [9]. These policy developments have encouraged researchers to integrate sustainability and decarbonization strategies into IMT research, aligning the sector with evolving regulatory frameworks [10]. Over the past decade, these developments have generated substantial interest within the OR community, resulting in the widespread application of modeling and optimization approaches to support emission reduction goals.

Despite this growing research activity, existing reviews have not, to the best of our knowledge, comprehensively examined OR methodologies for IMT decarbonization. Early studies by Macharis and Bontekoning [11] and Merrina et al. [12] synthesized research on OR modeling techniques during the emergence of the IMT concept, emphasizing the diverse stakeholders involved and the system complexity arising from their differing objectives. These works recognized the potential of OR techniques to address coordination challenges by balancing stakeholder interests and enabling integrated decisions for system-wide efficiency. Furthermore, Crainic et al. [13] proposed a taxonomy for simulation modeling in IMT, while Archetti et al. [14] focused on long-haul transport optimization. However, these studies largely overlooked sustainability considerations and carbon emissions as explicit modeling objectives or constraints. Conversely, broader FT surveys adopted predominantly qualitative approaches [15,16], examining drivers and environmental impacts but providing limited analysis of operational modeling and quantitative OR methods. Consequently, a critical research gap persists, as no comprehensive synthesis assesses OR techniques in IMT decarbonization, their formulation and solution approaches, and key methodological trends in this growing field.

To address this gap, the present survey is organized into two complementary parts. Part I (the present paper) examines the methodological landscape by systematically reviewing OR techniques applied to IMT decarbonization. We provide a comprehensive taxonomy across key problem domains (e.g., allocation–location, network design), examining model formulations, solution techniques, and modeling components. Key contributions include identifying the need for multistage stochastic optimization frameworks to capture uncertainty and enhance network reliability as well as recognizing the potential of data-driven and machine learning approaches for decision making and coordination. A comprehensive summary table mapping problem types, modal configurations, objectives, and solution approaches is provided in the Supplementary Materials.

Part II [17] complements this methodological foundation by providing a synthesis focused on applications that examines chronological evolution, modal trends, and sustainability patterns in FT decarbonization. The analysis draws on empirical evidence to evaluate how OR techniques enhance emission mitigation and operational effectiveness across IMT, MMT, and SMT systems, categorizing studies by modal composition, decision contexts, and sustainability objectives to establish a framework connecting operational models with practical implementation. While Part I focuses on OR techniques and modeling structures, Part II focuses on application contexts and temporal trends. Combined, these contributions extend the existing literature beyond isolated surveys and qualitative assessments, providing a complete picture of the role of OR in sustainable FT systems. The analysis reveals accelerated growth and methodological advancement since 2018, identifying emerging trends, leading approaches, and persistent gaps.

The remainder of this Part I is organized as follows: Section 2 outlines our methodology, Section 3 provides an in-depth OR analysis, Section 4 discusses research gaps and future directions, and Section 5 summarizes our findings. For reference, a glossary of acronyms is provided at the end of this paper underneath the heading Abbreviations.

2. Review Methodology

In this section, we discuss the details of the methodology we employ to identify studies related to IMT and decarbonization. We adopt a systematic literature review (SLR) methodology, as illustrated in Figure 1, that was previously utilized in this research domain (see Agamez-Arias and Moyano-Fuentes [18]). SLR practices are commonly adopted in OR, where transparency is ensured through the explicit reporting of databases, search strings, time windows, inclusion criteria, and screening steps, enabling the reproducibility of the study selection process.

In response to the gaps identified in the literature, we develop three key research questions.

1.

Which OR techniques, along with their characterization, have been applied in IMT decarbonization?

2.

How has decarbonization been addressed in the IMT OR literature over the years, and how has research on modality mix, decision levels, and emissions evolved?

3.

What are the promising future research directions in IMT decarbonization?

Next, we define our search strategy and databases. We establish the search string as shown below by combining relevant terms aligned with our research questions [19]: “intermodal transportation” OR “multimodal transportation” OR “freight transportation” AND (green OR decarboni* OR “environmental sustainab*” OR ((fuel OR “carbon emission” OR “carbon dioxide emission” OR emission OR “greenhouse gas” OR ghg)) AND (reduc* OR improve* OR efficiency)).

Our research primarily utilizes Web of Science, Scopus, IEEE, and ABI Inform (ProQuest) databases, which were selected for their extensive coverage of relevant literature in this field [15]. The search focuses on the titles, keywords, and abstracts of publications written in English from 2010 to early 2024. This period corresponds to the increasing application of OR modeling techniques in IMT systems and the growing emphasis on freight decarbonization to reduce environmental impacts.

Third, we establish inclusion criteria to filter publications that (1) apply OR techniques, (2) involve at least two transport modes, and (3) incorporate explicit decarbonization strategies. Studies that did not simultaneously satisfy all three criteria were excluded. Borderline cases were assessed conservatively and included only when explicit evidence of multimodal integration and decarbonization objectives was present. For OR techniques, studies must apply scientific and mathematical modeling, supported by computational methods, to analyze, design, and optimize FT systems for informed decision making. Additionally, each study must employ at least one widely recognized OR method, such as linear programming or simulation [20]. For decarbonization strategies, eligible studies must address at least one of the five pathways defined by McKinnon [21]: (1) reducing demand, (2) shifting to low-carbon modes, (3) optimizing asset utilization, (4) increasing vehicle energy efficiency, and (5) promoting new energy sources. This framework has been applied in other studies that examine specific OR techniques for FT decarbonization, such as system dynamics (SDs) [22]. It is important to note that while our search strategy primarily focused on IMT and MMT, we also included SMT studies identified during this filtering process, as SMT serves as a means to enhance IMT efficiency through dynamic planning and real-time information exchange [23]; therefore, we classified SMT studies as part of IMT.

We identify 89 publications meeting these criteria. Applying a backward snowballing technique [24], we examine references from these publications to identify additional relevant papers. This process results in a final set of 137 publications. Study screening and classifications were conducted by two researchers using consistent criteria, and assignments were cross-checked to ensure consistency. All publications were thoroughly reviewed and analyzed. Given the comprehensive nature of this review, we structure our analysis into two complementary parts as described below. This study, Part I, presents a methodological analysis of 127 studies from an OR perspective, examining problem categories, optimization techniques, and formulation methods.

For problem categorization, we defined five main categories: (1) Network Design Problem (NDP), (2) Service Network Design Problem (SNDP), (3) Supply Chain Network Design (SCND), (4) Policy Support (PS), and (5) Mixed. The first four categories are based on their established presence in the OR literature and relevance to FT, and they are well documented across foundational review studies [25,26,27], representing the most frequently appearing problem classifications in FT optimization research. The fifth category, Mixed, was created for studies addressing multiple problem types to preserve their integrated nature. A detailed description of each category is provided in Section 3.1. Few studies focusing primarily on emission estimation models were excluded from Part I but retained for Part II, as these estimation models play a key role in measuring CO₂ emissions across multiple studies.

Part II provides a practical analysis focused on applications and is organized into two subsections: chronological evolution and integrated sustainability assessment. The chronological analysis examines all 137 studies, including four studies from early 2024 to capture emerging research applications. The integrated sustainability assessment analyzes 133 studies, examining modal distribution, decision-level classifications, and CO₂ policy instrument integration.

3. OR Techniques Analysis

This section synthesizes OR models supporting IMT decarbonization. Our aim is to identify dominant modeling paradigms and the key operational aspects they address. We summarize the essential methodological elements in tables and draw insights that guide the discussions presented in our study. To substantiate the relevance and trends of the reviewed literature, we provide descriptive statistics and quantitative summaries throughout this section, including counts and percentages by problem type, modeling approach, objective formulation, and solution technique.

Before diving into an extensive analysis, we present an overview of the mathematical models discussed in the articles. Table S1 in Section A of the Online Supplement summarizes the mathematical models used in the reviewed studies, including their objectives, modeling characteristics, and solution techniques. The table indicates that economic objectives dominate the literature, with about 75% of studies addressing transportation costs, while emission and handling costs are considered in approximately 40%. In contrast, transportation time and emissions are objectives in only 17% and 33% of studies, respectively. Notably, there has been a clear increase over the last five years in studies incorporating transportation time as an objective.

3.1. Classification Based on Problem Addressed

To illustrate how OR techniques support IMT decarbonization, we categorize the studies by the problems addressed and the methods applied. Based on the problem classification framework established in the Methodology (Section 2), this section describes five problem categories: (1) NDP, (2) SNDP, (3) SCND, (4) PS, and (5) Mixed category representing combinations of these problems. Studies addressing more than one problem type were assigned to the Mixed category to preserve their integrated nature. A detailed analysis of the distribution of studies across these categories is presented in Section 3.3. We now provide a brief description of each primary category. NDP focuses on optimizing transportation infrastructure by construction or improvement of facilities as well as integrating decisions related to the allocation of resources to improve the route efficiency, cost, and capacity of the network [28]. Within this category, we identify three main subcategories: Location Problem (LP), Capacity Expansion Problem (CEP), and Hub Selection Problem (HSP). Next, SNDP involves planning the selection, routing, and scheduling of transportation services, as well as managing terminal operations and freight routing, ensuring the fulfillment of demand and profitability [25]. Four main subcategories are included in this regard [29]: Service Selection Problem (SSP), Traffic Distribution Problem (TDP), Asset Management Problem (AMP), and Revenue Management Problem (RMP). While SCND focuses on determining the optimal location and size of facilities and managing the flow of goods through these facilities [30], the PS category encompasses studies that support the decision-making processes of stakeholders by evaluating strategies across various decision levels (e.g., tactical, operational) [31]. These studies offer insights and evidence-based assessments for policy formulation, implementation, and optimization in transportation. The categories and subcategories for each article are specified in Table S1 in Section A of the Online Supplement.

3.2. Classification Based on OR Techniques

Accordingly, we group the OR techniques found in the reviewed studies into eight categories: (1) Exact (e.g., branch-and-bound, simplex), (2) Heuristic, (3) Meta-heuristic, (4) Simulation, (5) Hybrid (i.e., the combination of at least two techniques), (6) Game Theory (GT), (7) Decision Support System (DSS), and (8) Others. When multiple techniques were used within a single study, the paper was classified as Hybrid. Hybrid approaches include combinations such as exact methods with heuristics, simulation-based optimization, or simulation coupled with meta-heuristics. The “Others” category encompasses less commonly used methodologies, such as Markov Decision Processes (MDPs) [32], Lagrangian Relaxation (LR) [33], and Benders Decomposition (BD) [34]. Next, we offer a concise overview of OR categorization. Exact methods, like branch-and-bound and simplex [35], guarantee optimal solutions but can be time consuming, especially as the problem size increases. For simplicity, we separate exact solution algorithms such as BD and Column Generation from this category. Heuristics provide quick, problem-specific solutions without ensuring optimality [36]. Meta-heuristics guide and modify heuristics to explore search spaces more effectively and avoid local optima [37]. Simulation techniques, including discrete event simulation (DES) [38], agent-based modeling (ABM) [39] and SD [40], are used for evaluating and optimizing systems through various what-if scenarios [13]. Each method offers unique strengths: DES in event sequencing, SD in feedback understanding, and ABM in modeling intricate agent interactions. While hybrid methods either combine optimization and simulation to handle complex systems where social factors complicate precise modeling [41] or employ exact methods and heuristics together, GT models interactions among stakeholders and is often paired with heuristics for problems at large scales [42]. On the other hand, DSSs improve decision making by enhancing various performance metrics in transportation [43].

3.3. Distribution of OR Techniques Across Different Problem Categories

The distribution of studies across problem types and OR techniques is presented in Figure 2. Each study was classified exclusively within one problem type (Section 3.1) and one OR technique (Section 3.2) to ensure mutually exclusive categorization. These studies are further analyzed below.

Exact Methods: The decarbonization of FT through OR techniques has been driven primarily by exact methods. Most studies focus on SNDPs, particularly the SSP, emphasizing structural decisions on transportation services. Research has traditionally centered on deterministic formulations with greater attention to strategic decisions and a secondary focus on tactical levels. Since the first exact-method study incorporating uncertainty in 2016 [44], two major trends have emerged: a shift from generalized cost objectives (e.g., emissions or penalties) to broader factors like travel time and emissions, and a marked rise in research output, with 57% of studies published between 2019 and 2020. Recent works increasingly address three core aspects: sustainability trade-offs among stakeholders, nonlinearities (such as congestion and pricing), and uncertainty (mainly demand). Multi-objective formulations are commonly used to capture these trade-offs, with 27 of 54 exact-method studies addressing multiple objectives and 11 applying the epsilon-constraint method, confirming its dominance in handling multi-objectivity. For nonlinear formulations, researchers often employ auxiliary variables and linear constraints to enable solution through commercial solvers.

Meta-heuristics: Meta-heuristics are the second most widely used approach after exact methods.

Table 1. Summary of papers using meta-heuristic approaches.

Article	Problem	Decision(s)	OT	Algorithm	Key Characteristics
Sawadogo et al. [46]	SNDP	Determine sustainable path	MO	ACO	Defines heuristic information a priori and updates dynamically. Updates pheromone trails based on transportation mode.
Kim et al. [47]	NDP	Upgrade facility and assign traffic flow	BLO	GA-based	Performs feasibility checks with budget constraint and neighborhood improvement based on volume-to-capacity ratios.
Chen et al. [48]	PS	Define state subsidies and transport service design	BLO	GFWHA	Alternates selection methods. Includes carbon reduction in fitness function and shipping routes in chromosome coding.
Fahimnia et al. [49]	SCND	Choose production and distribution plans under carbon tax policies	MO	NICE	Considers adaptive updates for binary variables and iterative updates for feasible solutions until the ratio drops below a set value.
Ji and Luo [50]	SNDP	Optimize service selection	MO	HEDA	Uses heterogeneous marginal distribution law as probability model. Embeds local search methods.
Zhang et al. [51]	Mixed	Optimize service selection and node selection	SO	Hybrid GA	Considers two-part chromosome (node selection and mode selection). Applies two-point crossover, two-point mutation, and customized mutation.
Zhao et al. [52]	SNDP	Reposition empty containers under uncertainty	SO	Two-phase TS	Uses swap move sample, dynamic tabu duration and aspiration criterion for both relocation and the routing phase.
Zhang et al. [53]	PS	Optimize investment, subsidy value, and flow assignment	BLO	GFWHA	Considers two-part chromosome (infrastructure selection and subsidy values). Uses two-point crossover, customized crossover, two-point mutation, and customized mutation.
Maiyar and Thakkar [54]	SNDP	Choose hub terminal locations and flow assignment under disruptions	SO	PSO with DE	Conducts particle encoding with population size and decision variables. Uses a discretization scheme for boundary violations.
Lu et al. [45]	SNDP	Optimize service selection	SO	DE	Addresses the railway transport echelon.
				CW + LS	Addresses the highway transport echelon. Uses CW for initial solution, LS for optimal solution (with four operators).
Wei and Dong [55]	SNDP	Allocate freight flow in cross-border network	MO	awGA	Uses matrix integer encoding (five layers) and adaptive weight fitness function with elitism strategy for offspring generation.
Wang et al. [56]	SCND	Schedule production and transportation	SO	Improved GA	Uses a multi-heterogeneous coding method with heuristic rules (two-part chromosome: production and transportation).
Wu et al. [57]	SNDP	Optimize logistics network under 4PL scheme	SO	Improved PSO	Uses a constraint processing mechanism with three algorithms for flow allocation, capacity limit, and capacity adjustment.
Liang et al. [58]	SNDP	Optimize service selection with user satisfaction	MO	NSGA-III	Uses priority-based real-number chromosome encoding, a simulated binary crossover and polynomial mutation operator.
Li and Sun [59]	SNDP	Obtain sustainable path under uncertainty	SO	FAGSO	Considers Gaussian mutation sparks and modulo operation for mapping rules. Uses attractive force search operators.
Xie et al. [60]	SNDP	Locate air hubs in cross-border logistic	MO	NSGA-II	Uses integer encoding, crowding distance calculation, simulated binary crossover, and polynomial mutation.
Zhang et al. [61]	SNDP	Develop pick-up and delivery SMT plans	MO	ALNS	Customizes operators for node and route removal. Ensures synchronization procedures. Addresses constraint violations.
Zhang et al. [62]	SNDP	Develop routing plans for flexible SMT service	SO	ALNS	Predefines initial solutions. Customizes swap operator. Utilizes simulated annealing for acceptance criterion.
Zhang et al. [63]	SNDP	Develop collaborative SMT	MO	ALNS	Considers preference constraints (higher load assignment to trains and barges).
Zhang et al. [64]	SNDP	Optimize SMT planning	MO	ALNS	Prioritizes requests based on preference. Conducts synchronization checks for time and preferences.
Ke et al. [65]	PS	Adjust freight structure	MO	AwGA	Uses floating point for chromosome encoding, roulette selection, adaptive crossover and mutation rates.
Shoukat and Xiaoqiang [66]	SNDP	Optimize service selection	MO	GA	Uses random generator for initial population, binary chromosome coding. Addresses two objective functions in the fitness function simultaneously.
Liu [67]	SNDP	Optimize service selection under uncertainty	SO	HEGA	Uses insertion heuristic for initial population and queen-bee evolution method for crossover. Implements elite retention strategy.
Zhang and Chen [68]	SNDP	Optimize service selection	MO	Modified GA	Utilizes hierarchical encoding, adaptive crossover and mutation mechanism. Conducts fitness evaluation using TOPSIS.
Li and Wang [69]	NDP	Determine secondary hub locations	MO	Adaptive GA	Considers three-part chromosome coding for freight allocation with multi-point crossover and single-point mutation.
Yang et al. [70]	SNDP	Optimize service selection	MO	Improved fuzzy GA	Uses fuzzy controller to dynamically adjust GA parameters based on population variance and fitness values.
Guo et al. [71]	SNDP	Optimize service selection	SO	PSO	Considers particle swarm optimization with path selection ratio as well as inbound and outbound specification.

OT: Optimization Type, SO: Single-objective Optimization, MO: Multi-objective Optimization, BLO: Bi-level Optimization, ACO: Ant Colony Optimization, GA: Genetic Algorithm, GFWHA: Genetic and Frank–Wolfe Hybrid Algorithm, NICE: Nested Integrated Cross-Entropy, HEDA: Hybrid Estimation of Distribution Algorithm, TS: Tabu Search, PSO: Particle Swarm Optimization, DE: Differential Evolution, CW: Clarke–Wright Savings Algorithm, LS: Local Search, awGA: adaptive-weight GA, NSGA-III: Non-dominated Sorting GA III, FAGSO: Fireworks Algorithm with Gravitational Search Operator, NSGA-II: Non-dominated Sorting GA II, ALNS: Adaptive Large Neighborhood Search Algorithm, HEGA: Hummingbird Evolutionary GA, TOPSIS: Technique for Order Preference by Similarity to an Ideal Solution.

summarizes studies applying these techniques, detailing problem types, decision levels, optimization types (single, multi-objective, or bi-level), and algorithm-specific innovations. Their use has grown rapidly in recent years, representing nearly 50% of studies published between 2020 and 2023. This rise reflects their flexibility, adaptability, and effectiveness in handling complex, multi-objective problems. About 63% of studies employ meta-heuristics for multi-objective or bi-level optimization, leveraging their ability to explore large solution spaces and manage conflicting goals efficiently. A distinctive example is Lu et al. [45], who decomposed the model into two echelons, which are each solved with a different meta-heuristic algorithm. Most applications address NDPs, particularly SSPs and TDPs within SNDPs. Among 16 recent studies, 11 utilized genetic algorithm (GA), particle swarm optimization (PSO), or modified variants, highlighting their continued relevance. GAs are often applied to SSPs involving frequency determination and coordination across fixed and flexible schedules. Additionally, the adaptive large neighborhood search algorithm (ALNS) has been used exclusively for SNDPs in SMT contexts, where rapid convergence and collaborative decision making are essential for real-time problem solving.

Simulation Methods: Simulation techniques have gained considerable prominence with 90% of studies conducted in the past five years and 50% occurring in 2023 alone utilizing these methods.

Table 2. Summary of papers using simulation.

Article	Approach	Decision(s)	Decision Level	Transp. Modes	Findings
[72]	ABM + DES	Transport policy and infrastructure analysis	S	RW, RR, WW	Captures goals, interactions, and time aspects of transport chain actors.
[73]	DES	Estimate seaport vs. dry port usage effects	O	RW, RR	Dry ports reduce emissions and costs.
[74]	SD	Evaluate long-term effects of policies on emissions	S	RW, RR	Weight regulations and grade 1 railway construction reduce emissions.
[75]	SD	Containerization and tax effect on modal shift	S	RW, RR	Containerization leads to faster modal shifts than taxation.
[76]	DES	Evaluate multimodal transport route efficiency	O	RW, RR, WW	Reorganizing routes and setting up container centers cuts costs.
[77]	SD	Modal shift policies for freight decarbonization	S	RW, RR, WW	Stricter policies and investments accelerate shift to rail.
[78]	DES	Compare seaport/dry port and multimodal efficiency	S, O	RW, RR	Dry ports are cost-efficient for long-term storage.
[79]	SD	Carbon taxation impact on port-hinterland transport	S	RW, RR, WW	Medium-high taxes shift traffic to rail and waterways, reducing emissions.
[80]	SD + MC	Design policy mixes to enhance sustainability	S, T	RW, RR, WW	Cost-based pricing and better rail services are effective in mixes.
[81]	ABM + DES	Assess dry port strategies with different container substitution levels	S, T, O	RW, RR	Full ownership container substitution and extended free storage reduce empty container repositioning.

ABM: Agent-Based Model, DES: Discrete Event Simulation, SD: System Dynamic, MC: Monte Carlo, RW: Road, RR: Railway, WW: Waterway, S: Strategic, T: Tactical, O: Operational.

presents a comprehensive overview of the diverse simulation methodologies utilized to analyze IMT systems and policies. The table is organized into six columns—article, approach, objective, decision level, transportation modes, and findings—outlining the simulation method, focus, decision level, modes considered, and key outcomes for each study. Based on the table, it is evident that the most commonly employed approach is SD with all studies utilizing this method focused on solving PS problems at the strategic decision level. Additionally, ABM has consistently been used in combination with DES. In terms of decision levels, SD primarily addresses strategic decisions, while DES focuses on operational decisions. Strategic decisions are more frequently explored through simulation compared to operational and tactical decisions. Interestingly, only one study considers all three decision levels, underscoring the need for further research in this area. Road and rail modes are included in every study, while only half of the studies consider waterways.

Game Theory: GT has been employed by researchers for mostly analyzing NDP and PS problems.

Table 3. Summary of papers using game theory approaches.

Article	Game Players (Leader, Follower)	Leader Actions	Follower Actions	Decisions
Wang et al. [82]	Government (leader), Firm (follower)	Impose carbon emission tax	Choose shipment transportation mode and selling price	Ec, S, En
Tsao and Linh [83]	Seaport (leader), Dry ports and shippers (followers)	Determine storage price	Determine service areas, prices, and delivery schedule	Ec, S, En
Xu et al. [84]	Port operators (both)	Strategy to locate dry ports	Strategy to locate dry ports	Ec, En
Chen et al. [85]	Government (leader), Carriers and shippers (followers)	Conduct pricing game	Select transportation mode	Ec, En
Shams et al. [86]	Government (leader), FT systems (followers)	Set emission reduction coefficient and penalties	Bargain for trading permits, determine equilibrium prices, penalize cap violation	Ec, S, En
Rahiminia et al. [87]	Railway operator (leader), Shipper (follower)	Set rail transportation price	Decide on freight volume to be transported	Ec, S, En
Wu and Zhang [88]	Government (leader), Shippers (followers)	Determine number, location, and capacity of dry ports	Choose seaports and paths for container export	Ec, En

Ec: Economic, S: Social, En: Environmental.

summarizes research on GT applications in these areas. A recurring theme in these studies is the interaction between a leader and followers, where the leader sets policies or makes strategic decisions, and the followers respond to these actions. The table highlights this interaction by identifying both the leaders and followers, outlining the actions of the leaders, and detailing the subsequent responses of the followers. It also indicates the impact of these decisions across economic, social, and environmental dimensions. In 50% of the studies, the government takes on the role of the leader, implementing policies or setting prices that influence the behavior of firms, ports, or shippers to adopt more sustainable practices. Another important insight is that nearly all studies examining the governmental role as leader were conducted toward the end of the review period, indicating a clear trend. We can see a strong emphasis on economic and environmental decisions, particularly regarding pricing strategies, transportation efficiency, emission reductions, and sustainability. Decisions related to social factors are addressed in only about half of the studies, being relevant primarily in the context of infrastructure planning.

Decision Support Systems: DSS applications concentrate primarily on SNDP (60%), PS (30%), and SCND (10%) problems, highlighting their operational and policy focus. In this context, DSS refers to optimization-based systems that integrate multiple criteria within software frameworks to support IMT and MMT decisions. Early pioneering work integrated geospatial tools [89,90] and spreadsheet applications [91] for routing and environmental analysis. A key characteristic of DSS applications is their strong reliance on multicriteria decision-making (MCDM) methods, including Analytic Hierarchy Process (AHP) [91], Data Envelopment Analysis (DEA) [92], fuzzy logic approaches [93], Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) [94] and combinations of these approaches [95]. Recent developments include web-based and interactive platforms to enhance accessibility. In particular, Laurent et al. [96] introduced CarbonRoadMap, which is a web-based DSS that generates Pareto-optimal multimodal routes, while Oudani [95] extended this framework by incorporating blockchain-related costs using multicriteria ranking techniques. Beyond individual decision tools, several studies embed DSSs within large-scale system models to evaluate modal shift policies and decarbonization pathways such as the TIMES energy models used to assess freight transitions between modes [97]. Within the policy development context, studies have also assessed the impact of infrastructure development through DSS models [92,98]. Notably, 70% of DSS studies have been published since 2019, reflecting growing methodological interest in integrating MCDM and multi-objective optimization for freight decarbonization. Overall, DSS applications emphasize trade-off analysis through Pareto frontiers and scenario-based optimization, enabling the evaluation of multiple solutions rather than identifying a single best solution.

Hybrid: These studies integrate multiple OR techniques within a unified framework to capture uncertainty, system dynamics, and operational complexity under realistic conditions. Early hybrid research primarily combined analytical optimization models with meta-heuristics to enhance solution quality. For instance, Zhang et al. [99] and Zhang et al. [100] paired traditional mathematical programming with meta-heuristic algorithms to address network design challenges, while Li and Zhang [101] combined gradient-based optimization with meta-heuristics to improve convergence and potentially identify nearly optimal railway scheduling and pricing strategies. A more recent and dominant trend involves integrating simulation methods with analytical models, representing approximately 56% of hybrid studies. Liotta et al. [102], Layeb et al. [103], and Gallardo et al. [104] combine DES with optimization frameworks to evaluate operational decisions under stochastic conditions. Some studies further extend this approach by integrating multiple simulation paradigms; Hrušovskỳ et al. [105] and Hrušovskỳ et al. [106] combine DES with ABMs to capture both overall system dynamics and individual stakeholder behavior. Recent methodological developments have also incorporated MC sampling with GA to explicitly handle uncertainty in decision variables and parameters [107]. The application focus has shifted from early NDP (11%) and SCND (11%) toward SNDP (67%) with one recent study integrating SNDP with NDP. Within SNDP, most studies address SSP, while one focuses on RMP [101]. This RMP study is particularly noteworthy, as revenue management remains critically underexplored across all 127 reviewed OR studies, highlighting the value of hybrid methods for pricing and capacity allocation in FT decarbonization.

Heuristic: A recurring theme in these studies is the use of methods based on search and approximate solution techniques to address optimization models of large scale and high complexity that are computationally intractable for exact algorithms. Most studies apply local search procedures [108] or decomposition approaches guided by heuristics to derive solutions that are nearly optimal within a reasonable computational time [109,110]. For instance, Pishvaee et al. [111] propose an interactive heuristic solution framework to balance economic and environmental objectives under uncertainty, while Yang et al. [112] develop an active set algorithm for local optimization to jointly determine shipping network design and toll policies for emission reduction. These methods have been applied primarily to NDP (40%), which is followed by PS (20%), SNDP (20%), and SCND (20%). As observed in hybrid approaches, the study by Wang et al. [110] is one of the few that address RMPs.

Others: A small subset of studies adopts diverse OR techniques that fall outside the primary methodological categories yet still contribute substantively to freight decarbonization analysis. These works primarily focus on network evaluation, stochastic control, and integrated system modeling. A common theme is the use of decomposition and solution refinement techniques that increase the tractability of complex optimization problems by exploiting problem structure, iteratively refining solutions, or providing bounds to guide the search for optimal solutions. These include LR [113], cutting plane methods such as BD [114] and logic-based cuts [115], and augmented Lagrangian multipliers [88]. Recent methodological developments also incorporate reinforcement learning (RL) to balance costs and eco label compliance [116]. These studies primarily address SNDPs (60%), which is followed by SCND (20%) and PS (20%). Notably, within the SNDP category, Hoen et al. [113] represents one of the few works that also focus on RMPs.

Beyond the methodological classifications analyzed earlier (Figure 2), two additional formulation aspects emerge across the reviewed studies: bi-level decision structures and stochastic modeling of uncertainty. These aspects are essential for capturing hierarchical stakeholder interactions and dynamism in FT systems, carrying significant implications for both model realism and computational complexity. We assessed how uncertainty is addressed in the reviewed literature, particularly examining whether studies explicitly state uncertainty considerations, identify specific sources of uncertainty, employ stochastic or robust modeling methodologies, and adopt appropriate mathematical formulations. Similarly, bi-level formulations are analyzed to identify hierarchical decision-making structures. Rather than introducing new studies, the following analysis revisits the reviewed literature through these two perspectives.

Bi-Level Programming: The number of studies utilizing a bi-level programming approach has declined recently with only a 20% increase in the last two years. This trend reflects a diminishing interest in applying this method. The detailed overview of studies using bi-level programming is presented in

Table 4. Summary of papers using bi-level programming.

Article	Problem	Upper-Level Aim	Lower-Level Aim	Level Interactions	Solution Technique	Decision Level
[47]	NDP	Improve network by assessing facility capacity and mobility	Determine traffic volume of each facility	LL considers facility capacity set by UL	MH	T
[99]	NDP	Best terminal network setup and emission price	Assign multi-commodity flow in the network	LL follows UL to optimize cost and emission	H, MH	S, T
[48]	PS	Select optimal liner route to minimize state subsidies	Assign traffic flow, estimate emissions and profit using UEA model	LL returns profit to UL	MH	O
[100]	SNDP	Minimize total costs and CO2 emissions	Assign multi-commodity flow	LL returns total link costs and flows to UL	H, MH	S, T
[117]	NDP	Choose projects for railway network planning	Assign railway freight flow to maximize profit	UL assigns freight flow via LL model to compute objective	E	S
[53]	PS	Select infrastructure investments and subsidies	Describe selected service routes of logistics users using UEA model	LL returns carbon emission to UL for cost–benefit ratio	MH	S, T
[101]	SNDP	Subsidize rail shift while maximizing profit	Minimize total cost	LL returns equilibrium freight volume to UL	H	S, T, O
[118]	PS	Maximize total routing flow	Minimize transportation cost using flow assignment	LL returns assigned flows to UL	E	S, T
[110]	SNDP	Maximize revenue while minimizing emissions	Assign cargo flow using UEA model	LL returns freight volume and price to UL	H	T
[119]	Mixed	Optimize channel upgrades, liner size, and frequency	Determine optimal container routes using discrete choice theory	LL output feeds UL. Total cost from both LL and UL	H	S, T

LL: Lower Level, UL: Upper Level, E: Exact, H: Heuristics, MH: Meta-Heuristics, S: Strategic, O: Operational, T: Tactical, UEA: User Equilibrium Assignment.

. The table is structured into columns detailing key aspects of each study, including problem classification, the aim of upper-level (UL) and lower-level (LL) analyses, interactions between levels, techniques used, and the decision level addressed. This approach is particularly beneficial for IMT systems for its ability to address their hierarchical nature. Despite this, we believe that the disinterest in bi-level programming is likely due to a shift toward more flexible modeling approaches, such as simulation, and the emergence of new methodologies, like machine learning, which offer simpler and more effective solutions. From the table, we observe that 8 out of 10 studies employ heuristics or metaheuristics as solution techniques, while only two utilize exact methods. Additionally, decision making at the tactical level is the most explored, represented in eight studies, whereas only two analyze strategic decisions. In the earlier years, NDPs were mainly formulated as bi-level programming models, but recent years have seen a shifted focus on SNDPs.

Stochasticity in IMT: Approximately 20% of IMT studies incorporated uncertainty, with about half published between 2020 and 2023, indicating growing interest in stochastic modeling (

Table 5. Summary of papers addressing uncertainty.

Article	Decision Level	Uncertainty Source													Uncertainty Handling Technique
		CA	TRT	FC	TRC	PC	TRE	TT	CP	D	SU	ET	R	AT
Holmgren et al. [72]	S									✓					ABM
Pishvaee et al. [111]	S	✓		✓	✓	✓	✓			✓					Fuzzy CCP
Demir et al. [44]	T, O		✓							✓					SAA
Rezaee et al. [120]	S, T								✓	✓					Scenario generation
Sun et al. [121]	O	✓	✓												Fuzzy CCP
Zhao et al. [52]	T, O									✓	✓				CCP, SAA
Layeb et al. [103]	T		✓							✓					SBO
Hrušovskỳ et al. [105]	T		✓												ABM, DES
Haddadsisakht and Ryan [114]	S, T								✓	✓					Probabilistic scenario generation, RO + SP
Tsao and Thanh [122]	S, T						✓			✓					RO + fuzzy programming
Sun et al. [123]	S, T												✓		Fuzzy credibilistic CCP
Baykasoğlu and Subulan [124]	S, T, O	✓			✓			✓		✓					Hybrid CCP + fuzzy interactive resolution-based approach
Sun [125]	O	✓						✓							FST, Chance constraints
Wang et al. [126]	T	✓													FST, Fuzzy constraints
Wang et al. [56]	O													✓	Probability distribution
Jiang et al. [118]	S, T									✓					RO
Dai and Yang [127]	T, O									✓					Distributionally robust CCP + approximation technique
Mousavi Ahranjani et al. [128]	S, T			✓	✓	✓				✓					Robust stochastic possibilistic CCP, Fuzzy theory
Zhang et al. [107]	T		✓							✓					RO + SP
Li and Sun [59]	T								✓	✓					RO + SP
Sun et al. [129]	T		✓				✓								Interactive fuzzy CCP
Ko et al. [130]	S				✓										Scenario generation
Heinold et al. [116]	O		✓							✓					SBO, SP
Li et al. [131]	T		✓					✓		✓					Fuzzy expected value method, fuzzy CCP
Sun et al. [132]	T	✓													Fuzzy CCP
Guo et al. [80]	T											✓			Scenario generation

✓: Indicates that the corresponding uncertainty source is explicitly considered in the study; S: Strategic, O: Operational, T: Tactical, CA: Capacity, TRT: Transportation Time, FC: Fixed Cost, TRC: Transportation Cost, PC: Production Cost, TRE: Transportation Emission, TT: Transshipment Time, CP: Carbon Price/Tax, D: Demand, SU: Supply, ET: Economic Transition, R: Risk, AT: Arrival Time, ABM: Agent-Based Modeling, CCP: Chance Constrained Programming, FST: Fuzzy Set Theory, SAA: Sample Average Approximation, RO: Robust Optimization, DES: Discrete Event Simulation, SBO: Simulation-Based Optimization, SP: Stochastic Programming.

). Among these, demand uncertainty is most prevalent, appearing in roughly 90% of studies due to its strong influence on SC performance. Transportation time uncertainty ranks second, reflecting its importance for scheduling and coordination. In contrast, factors such as fixed costs, transshipment costs, supply, and risk are rarely examined. Most research considers only one or two uncertainties with just four studies addressing more than two simultaneously. Uncertainty is predominantly analyzed at the tactical level, as observed in 19 of 26 studies, which is a pattern that has remained consistent over time. Transportation time uncertainties have not been studied strategically, while emissions and carbon pricing uncertainties are absent from analyses at the operational level. Regarding methods, around 40% of studies employed chance-constrained or fuzzy approaches across decision levels. Only four used scenario generation, and just one quantified scenario likelihoods using probability distributions. Notably, only one study incorporated stochasticity at all decision levels, underscoring a need for broader and more integrated uncertainty analysis in IMT research.

4. Future Research Directions and Identified Gaps

This section highlights research gaps and future directions in which OR methodologies can more effectively support freight decarbonization efforts. The proposed directions are directly derived from the methodological limitations identified in the preceding analysis, including a limited incorporation of uncertainty, simplified operational representations, and insufficient coordination among stakeholders, all of which highlight the need for more dynamic, data-driven, and integrated modeling frameworks. The following subsections explicitly connect these observed challenges to corresponding methodological opportunities and emerging technologies.

Network Reliability: As shown in our uncertainty analysis, most studies rely on deterministic or two-stage formulations, which limits their ability to capture cascading disruptions and temporal effects. Time is a critical factor for both MMT and IMT system reliability [133], enabling the smooth coordination of multiple transport services with varying schedules arranged in sequence and directly influencing the overall performance of the system [134]. However, the current literature mostly considers transportation time and related uncertainties with only few studies including transshipment or handling time, making the state-of-the-art distant from real-world scenarios. Additionally, there is a significant lack of studies examining the interplay between multiple uncertainties, such as demand and travel times at the strategic level, as well as the relationship between demand and carbon pricing or taxes at the operational level. The importance of addressing these gaps is underscored by emerging research incorporating uncertainties due disruptions [135] and carbon policies under demand uncertainties [136]. However, these studies remain limited and primarily employ two-stage approaches that, as acknowledged by recent authors, cannot adequately handle the dynamic cascading effects of multiple interacting uncertainties [137]. The broader challenge of capturing multiple, dynamic evolving uncertainties through multistage stochastic frameworks thus remain largely unaddressed. Such frameworks could enhance uncertainty management in IMT systems by enabling decision-makers to assess how current decisions impact dynamic factors and future system conditions. While MDPs are effective for modeling complex, multistage problems, their application in the current literature is limited [138], highlighting the need for further research. Partially observable MDPs show promise for managing uncertainties by operating under incomplete information or limited visibility of the state of the system [139].

The efficiency of terminal operations also plays a crucial role in service network reliability. This issue, commonly studied under AMP, requires further exploration in the context of IMT and MMT decarbonization. Real-world terminal operations face numerous challenges, including traffic flow delays, equipment congestion, and labor shortages. To overcome these challenges, a shift toward Precision Scheduled Railroading can be beneficial, emphasizing regular trips over maximum train lengths and underscoring the necessity for research into labor scheduling and operational efficiencies that can enhance throughput without compromising environmental goals. Explicitly modeling inventories for container routing management at terminals can also provide valuable insights into the trade-offs between emissions and inventory costs [114].

Disruptions due to hub failures, such as the non-functionality of transfer nodes, can significantly disrupt the entire IMT network, leading to heightened vulnerabilities in critical infrastructure [140]. Until now, these infrastructures have been analyzed to determine the interdependency of critical nodes in terms of cost or operational efficiency [141]. However, an unexplored research avenue still considers the carbon emissions generated through node interdependency.

Emerging Technologies: Because current models often abstract away real-time operational complexity and system feedback, simulation-based and data-driven technologies such as digital twins (DTs) and reinforcement learning (RL) offer practical mechanisms to enhance adaptability and decision support. A key operational challenge is terminal congestion at intermodal transfer nodes, which requires efficient queuing systems in routing decisions [132]. Holistic network visualization can address this challenge through the integration of technologies such as the Internet of Things [142], DTs, and network initiatives like Virtual Watch Tower [143] for real-time rerouting capabilities. Emerging technologies such as artificial intelligence, RL, and blockchain could further enhance the resilience of IMT systems. RL, in particular, has been increasingly adopted in OR for solving complex problems like vehicle routing and the traveling salesman problem, especially when traditional methods face limitations in scalability and adaptability [144]. In IMT contexts, the ability of RL to handle dynamic, multidimensional environments and learn from complex interactions makes it particularly valuable for adaptive decision making. Additionally, improved SC visibility through enhanced intermodal shipment tracking, GPS systems, telematics technologies, and cloud-based software platforms can facilitate partnerships among companies operating in the intermodal sector.

Stakeholder Collaboration: Our review also shows that many studies optimize decisions for a single actor, which motivates the development of collaborative and multi-stakeholder modeling frameworks. Collaboration among diverse stakeholders (i.e., governments, logistics providers, infrastructure developers, and policymakers) can harness the potential of IMT despite their varying and often conflicting objectives. OR methods play a crucial role in addressing these coordination challenges [145]. Incorporating stakeholder collaboration into optimization models can improve resource utilization and reduce carbon emissions by optimizing asset allocation [146]. Investigating both collaboration and competition among similar stakeholders (e.g., competing port operators) presents important research opportunities. For instance, Xu et al. [84] applied GT to optimize dry port location and reduced fixed costs through strategic sharing, yet such problems could be further addressed through hybrid OR approaches, like combining GT with dynamic programming [147]. At broader scales, developing integrated global and regional models and collaboration strategies that address trade imbalances represents a promising research direction. Hybrid simulation approaches, integrating ABM with DES, for example, could capture both strategic stakeholder behavior and operational logistics dynamics [81].

Infrastructure governance poses additional coordination challenges. Railway privatization exemplifies these challenges, complicating national transportation planning, as companies must negotiate individually with specific railroads, restricting their access to critical infrastructure. While collaboration among multiple rail providers could enhance efficiency, with potential cost reductions demonstrated in other contexts [148], significant barriers exist in information sharing, equity, and real-time planning coordination. Future OR research can leverage SDs modeling to assess the impact of standardized protocols and government incentives [149] alongside advanced digital technologies, such as DTs and machine learning algorithms, that enable real-time information exchange and coordinated decision making across integrated MMT operations [150].

Integrating shipper heterogeneity into IMT models is crucial when considering stakeholder collaboration, as carriers often transport containers for multiple shippers with diverse preferences regarding cost, speed, sustainability, reliability, and risk. Future research could focus on IMT and SMT planning considering these varying preferences [88,151] with preference-based multi-objective optimization models presenting an interesting research avenue [62]. Recognizing that the stated preferences of shippers may differ from actual behaviors, future studies should analyze historical decisions to model preferences more accurately and learn from current choices in dynamic contexts [64]. Thorough behavior analysis is essential to understand and replicate the decision-making processes of both shippers and carriers, while a cost–benefit analysis of policy measures remains vital [77].

Beyond technical optimization, the successful implementation of decarbonization strategies also depends on social and institutional factors, including workforce impacts, regional disparities, regulatory constraints, and organizational coordination barriers. Although only a small subset of OR studies explicitly incorporates such dimensions, for instance Sawadogo et al. [46] and Kim et al. [47] integrate societal costs into routing and network design decisions, while Tsao and Thanh [122] jointly model economic, environmental, and social objectives under uncertainty, our review indicates that these considerations remain limited overall. Social impacts are primarily addressed within game-theoretic and policy-support contexts, while most optimization and simulation models assume homogeneous regional and institutional conditions. Integrating equity, employment, and regional development metrics into OR-based decision-support frameworks therefore represents an important interdisciplinary direction for future research.

Underutilized OR Techniques: Given the dominance of heuristic approaches observed in the literature, a further exploration of decomposition-based and exact methods remains warranted for scalability and solution quality. The use of some OR techniques, such as LR [113] and BD [114], in IMT literature remains limited. This could be due to the challenges posed by the consolidation and deconsolidation of goods, which disrupt the block structure essential for these traditional optimization algorithms and complicate large-scale linear programming formulations. Consequently, there has been a noticeable inclination in the literature toward meta-heuristics and hybrid solution techniques, primarily because these methods effectively address the complexities of IMT and provide greater flexibility. However, we argue that the further exploration and assessment of OR techniques within the context of decarbonizing the IMT sector is essential, particularly given the demonstrated benefits of these methods. This includes column generation for handling pricing mechanisms [152], BD for managing economies of scale [153], and partial BD for solving NDPs [154].

A persistent challenge in optimization is developing effective solution methods for large problem instances. While exact solution techniques can be applied to small theoretical cases in IMT problems, they often fall short for larger, more complex scenarios. Various meta-heuristic approaches have been introduced and are typically faster, but it is unclear which algorithms best suit specific real-world problems, as the literature lacks clarity on efficiency. The absence of standardized reference problems complicates comparisons with most studies focusing on unique variants. Establishing benchmark datasets could improve evaluation, but reliance on proprietary datasets hinders meaningful comparisons. Another challenge is the computational time and scalability associated with complex problems. To tackle this issue, researchers have turned to techniques like Neural Combinatorial Optimization, which has emerged as a recent trend [155,156] to enhance network reliability and resiliency. However, their application is scarcely explored in the decarbonization of IMT. Another important limitation observed across the reviewed literature is the heavy reliance on synthetic or hypothetical datasets for model testing and algorithm validation. While such instances facilitate controlled experimentation, they may not fully capture the operational complexity and variability of real freight systems. A greater use of empirical logistics data, large-scale freight flow records, and case studies from industry would improve external validity and help bridge the gap between methodological advances and practical implementation.

5. Conclusions

This paper synthesizes how OR has been used to support the decarbonization of intermodal freight systems and clarifies the current methodological maturity of the field. Rather than merely cataloging studies, our review reveals a clear structural pattern: research has largely emphasized deterministic cost-based optimization, while the dynamic, uncertain, and multi-stakeholder nature of real-world intermodal operations remains comparatively underrepresented. This mismatch suggests that many existing models capture operational efficiency but only partially address resilience, reliability, and system-wide sustainability. The findings indicate that future progress will depend less on incremental refinements of traditional formulations and more on methodological shifts toward adaptive, stochastic and data-driven approaches. Incorporating multistage uncertainty modeling, terminal-level operational realism, and collaborative decision-making frameworks is critical for translating decarbonization goals into deployable strategies. Emerging technologies such as RL, DTs, and integrated decision-support systems offer promising avenues to bridge this gap and enable adaptive, real-time planning.

Overall, this review provides a consolidated foundation that helps researchers position new contributions and identify high-impact directions for advancing sustainable intermodal freight systems. Importantly, many of the methodological insights identified in this review have direct implications for transport policy and implementation. Network and service design models can inform infrastructure investment and terminal expansion decisions; pricing and bi-level formulations support carbon taxation, subsidy allocation, and modal shift incentives; and stochastic and simulation-based approaches enable policymakers to evaluate resilience and operational trade-offs under uncertainty. Framing these OR tools as decision-support mechanisms helps bridge the gap between theoretical modeling and actionable policy design in real freight systems. Part II complements this methodological perspective by examining chronological and practical applications across different modal configurations and sustainability strategies to further contextualize the evolution of the field.