A multi-criteria decision support framework for assessing seaport sustainability planning: the case of Piraeus

ABSTRACT Seaports will play a pivotal role in the low-carbon transition of maritime shipping, the policy landscape in which is currently being shaped. In this context, we introduce a multi-criteria decision support framework for seaport sustainability planning, to identify the competitiveness of interventions under uncertainty and evaluate the direction of the sector’s policy context in terms of required actions. The framework, based on the 2-tuple TOPSIS model, heterogeneous variables, and a Monte Carlo robustness analysis, is empirically applied to the port of Piraeus, Greece, to explore the most competitive interventions and their vulnerability to uncertainties. To inform port authorities and policymakers in the sector, we emphasise the added value of selected, inexpensive actions for energy efficiency and hybrid mobility. Furthermore, we find costly and seemingly obligatory actions under current European legislation, like cold ironing and LNG, to be robust and in the right direction if perception of non-financial risks is reduced.


Introduction
Climate change constitutes one of the most prominent global challenges for sustainability, and efforts to mitigate its impacts are evident in all economic sectors and industries.This is also the case for ports.First, it is a matter of sustainability: as with all transport facilities, seaports are largely affected by climate change-related challenges, like sea level rising and powerful storms (Jiang et al. 2020).Second, maritime ports are among the multitude of factors contributing to the climate crisis.Intertwined with international shipping, which accounts for approximately 3% of global greenhouse gas (GHG) emissions and is under significant pressure to reduce environmental and climatic impacts, seaports (both cargo and passengers, especially considering container ships are found to be among the largest emitters; Nunes et al. 2017) are major producers of CO 2 and other air emissions (Lindstad and Eskeland 2016).This is because port services-including safe mooring, passenger access to ships, cargo loading and unloading, handling and storage, as well as refuelling and waste collection and disposal-require great amounts of energy and are thus responsible for significant GHG emissions.At the same time, maritime shipping dominates international transportation of goods (Carpenter et al. 2018), making it an important area of economic development.Ports are thus deemed critical to the sector's transition to sustainability (Bjerkan and Seter 2019).
In the fight against climate change, the Paris Agreement is the first legally binding agreement on climate change, adopted in 2015.Nations joined the ambitious call to limit global warming to well below 2°C by the end of the century, committing to consistent efforts to reduce global temperature representation model, the variable homogenisation process, and the 2-tuple TOPSIS model.Section 4 provides and discusses the results of the evaluation analysis, while Section 5 discusses the conclusions and prospects of the study.

Context
Greece is considered a significant force in shipping.The country's shipowners control 53% of the European and 17% of the global freight fleet, in terms of capacity, with the industry being the second largest after tourism, covering almost 7% of the country's GDP (Deloitte 2020).The Greek shipping-sector accounts for 12% of CO 2 emissions in the national transport mix, six times the average European share (UNFCCC 2018).Therefore, the participation of Greece in initiatives and actions to support the transition of the shipping sector to greener technologies and alternative fuels is deemed crucial.
The port of Piraeus is the largest in the country, with more than 24 km of coastline and a total area of five million square meters, with the passenger part of the port being ranked as the largest European and among the largest in the world (Progiou et al. 2021).Its geographic location makes it a vital transport, commercial, logistics, tourist, and communication hub.Located close to international trade routes, connecting the three continents (Europe-Asia-Africa), the port is a centre of international trade and an important hub of the global supply chain.The Piraeus Port Authority (PPA) was formed in 1930, while COSCO shipping has owned two-thirds of its stocks since 2016.PPA is responsible for all port activities and services, including the container, car, cruise, and ferry terminals, the shipbuilding and repair zone, and the freight management and distribution centre.In 2019, the PPA Master Plan was approved, including strategic expansion projects of the existing port facilities to enhance both its own and the country's financial development, and to create new jobs.
Like most ports (Azarkamand, Ferré, and Darbra 2020), the port of Piraeus is responsible for direct (vehicles, cargo handling, buildings, facilities, etc.), indirect (power consumption in the port premises), and other indirect (cargo trains, moored ships, cranes, passenger vehicles, and non-port buildings) GHG emissions (IMO 2018).As far as energy and environmental management are concerned, PPA implements numerous monitoring and evaluation programs aligned with the national and EU legislation.The port is also a member of the Ecoports network, which consists of European ports that have assessed their environmental performance according to the network's Self Diagnosis Method (SDM) (Darbra et al. 2004).PPA, in collaboration with the University of Piraeus and the University of Cardiff (United Kingdom), implements on a yearly basis a quality monitoring program of the marine environment: in collaboration with the School of Chemical Engineering of the National Technical University of Athens, an air quality monitoring program is implemented through a dedicated monitoring system.Among the few low-carbon interventions, the Authority has also installed a solar power plant of 430KWp in its premises, aiming to reduce the use of conventional energy sources and promote renewables.The port is often found in urgent need of stringent emissions control (Tzannatos 2010).
After investigating strategies and targets that other European ports have announced as well as the effective measures that they implement, we select eleven actions for the port of Piraeus.In general, climate targets include CO 2 emission reductions or even climate neutrality, while broader energyrelated strategies include improvement of energy efficiency in infrastructure, exploitation of renewable energy sources (RES), adoption and promotion of alternative fuels, equipment electrification, and cold ironing.Moreover, all three types of emissions as well as the port's current state have been considered.A detailed description of the actions considered in the context of the Port of Piraeus is presented in Appendix A.
Table 1 summarises the actions considered in the study and key assumptions used as input into the analysis.Table 2 summarises key factors used to calculate energy consumed and emissions produced.We assume 150 port calls and an average lay time per ship of around 12 hours.Emissions reductions refers to the emissions produced at berth.Energy savings from the action are estimated between 30 and 40%.The cost of the action is estimated at €5.5 million with a lifetime of 30 years.LNG Refuelling infrastructure Installation of LNG STS supply infrastructure Similar assumptions with cold ironing.Energy Savings are estimated between 20 and 30%.The cost is estimated at €3.5 million with a lifetime of 40 years.

Micromobility
Purchase of 10 e-bikes and e-scooters for staff's transportation within the port.
We assume an average speed of 10 km, and an operation time of all e-bikes and e-scooters between 1 and 4 hours.The consumption required to cover the distance for a conventional diesel car is compared with the micro-EVs.Cost is assumed at €25,000 for a lifetime of 2 years.

Electric buses
Purchase of 3 electric buses for passengers.
Similar with micromobility but assuming an average speed of 25 km, and an operation time of all 3 buses between 10 and 13 hours.Costs for the 3 buses are estimated at €1.35 million with a lifespan of 12 years.

Electric forklifts
Purchase of 4 electric forklifts.
Assuming operation of the equipment between 2,000 and 2,500 hours.The consumption of the conventional diesel and the electric equipment is compared.A cost of €200,000 for a lifespan of 10 years is assumed.Hybrid vehicles for lifting empty containers (empty container handlers) Purchase of 2 hybrid vehicles for lifting empty containers.
Energy savings and emissions cuts from replacing conventional diesel with hybrid equipment is drawn from technical datasheets and adjusted for a range between 4,000 and 6,000 operating hours.
A cost of €300,000 for a lifespan of 15 years is assumed.Hybrid vehicles for stacking and transporting containers (straddle carriers) Purchase of 2 straddle carriers Similar with hybrid empty container handlers.A cost of €500,000 for a lifespan of 15 years is assumed.

Energy consumption monitoring system
Installation of monitoring systems in 3 buildings.
Electricity consumption and energy consumption for heating drawn from Piraeus Port Authority datasheets.We assume an energy reduction of 10 to 20% for both consumption types.Cost is assumed at €195,000 with a lifespan of 15 years.Outdoor LED lighting Installation of LED lighting and lighting control systems outdoors.
We assume that outside lighting-related electricity consumption corresponds to 12% of the total consumption of the port with estimated savings between 60 to 80% due to the addition of the control system.Cost is assumed at €360,000 with a lifespan of 8 years.

Energy interventions in buildings
Improving energy efficiency in 3 buildings with energy saving systems.
Similar with the energy monitoring system, but with the assumption that the interventions refer only to heating-related retrofits.Savings are estimated between 30 and 50%.Costs are estimated for the 3 buildings at around €1.5 million with a lifespan of 25 years.

PV installation
PV infrastructure with installed capacity of 1 MW.
The 1 MW plant is assumed to operate on a capacity factor between 16% and 25% corresponding to around 4 to 6 hours of daily operation at full capacity.Cost is assumed at €800,000 with a lifespan of 25 years.

A multi-criteria decision-making framework
MCDA techniques have long been used to provide solutions in energy and climate issues (Doukas and Nikas 2020) for, inter alia, the evaluation and prioritisation of low-carbon technologies.In transit-oriented development, MCDA and multi-objective optimisation are considered among the most dominant tools used for planning (Ibraeva et al. 2020).A key question arising is why an MCDA analysis that mostly produces semi-quantitative results is necessary and appropriate to analyse green port development instead of alternative multi-objective optimisation methodological approaches, like portfolio analysis, which can directly produce investment mixes for designing the ports' action plans.A key issue with such a direct approach would have been that most interventions ports are currently obliged to implement are high-cost investments with long implementation periods.This essentially means that expected benefits from such actions would feature significant delays, consequently making other alternatives with quicker environmental returns crowd out longer-to-implement actions from portfolios evaluating the cumulative benefits in a given time horizon.As such, performing a preliminary competitiveness analysis using MCDA can provide useful insights into each action, to answer the questions driving our study and to help reshape future policies and seaport action plans accordingly.This is especially the case considering the numerous uncertainties at play in a policy design and implementation domain that is currently being shaped, rendering the analysis of each action's behaviour under uncertainty a vital task.The methodological research process followed and analysed in the following sections is presented in Figure 1.

Data aggregation
The low-carbon port technologies and interventions identified in the previous section, based on best practices from ports in Europe and elaborated in the context of the port of Piraeus, are assessed in this study using a set of four evaluation criteria: (a) contribution to CO 2 emissions reduction; (b) economic viability; (c) investors' risk (referring to the perception of non-financial risks, for example, associated with limited infrastructural feasibility, envisaged bureaucratic burdens, little capacity to implement, etc.), and (d) time of implementation.To assess the economic viability, the levelised cost of energy (LCOE) is used, adapted to reflect energy savings when necessary, a metric found to be of relevance to the evaluation of low-carbon technologies (Sofia et al. 2018).To exploit literature information regarding costs of significantly different scales and time horizons for the selected technologies, it was deemed more meaningful to use LCOE as a cost measure rather than the amount of investment, depicting the cost of energy savings per 1 MWh per annum during the lifespan of the initial investment.
where i is the discount rate (here considered at 5%) and n the lifespan of the investment.Indirectly, LCOE also takes into consideration energy savings achieved by each action, supplementing the emissions reduction criterion in the environmental front.Although both represent the environmental impact of each action, one directly and the other indirectly as part of LCOE, both emissions reduction and energy savings are metrics individually targeted by the IMO (Rehmatulla, Calleya, and Smith 2017) and can therefore co-exist in a coherent family of criteria.The investors' risk, highly dependent on the context of the port of Piraeus, is directly evaluated by two stakeholders (high-ranking officials working at the port), while with the assistance of the experts and previous knowledge from similar actions (e.g., PV infrastructure) the implementation time is also estimated.Together, the four criteria represent a diverse set of environmental, economic, and infrastructural issues that are vital for the MCDA to deduce preference relationships and calculate a meaningful ranking among the alternatives.
Among the four criteria, emissions reduction and implementation time are handled as intervals to reflect the uncertainty and variance of data drawn from the literature (Akyuz and Celik 2018) (e.g., the environmental impact of micro-mobility varies according to public perception and acceptability, see Gössling 2020) or context-related uncertainties that could cause delays.LCOE is a numerical variable that was preferred to intervals, since costs usually refer to average investments.Risk is provided by the two officials in linguistic form based on their perception on how easy they consider each action to implement in the port.It is evident that the four criteria are non-homogeneous.As such, an aggregation process should unify all variables in the same form.In this respect, we use the 2-tuple linguistic representation model (Herrera and Martínez 2000), which has the capacity to deal with all three types of variables and allows the output of the analysis to be in a format closer to the natural language of the stakeholders (Doukas, Karakosta, and Psarras 2010;Gavalas, Syriopoulos, and Tsatsaronis 2021).Based on a 5-term scale of {Very Low, Low, Medium, High, Very High}, the 2-tuple model is directly applied to capture feasibility as a linguistic variable, while the input for the other variables must be converted.This process allows data to be properly handled without loss of information, while retaining comprehensibility of the output results (Nikas, Doukas, and Martínez López 2018).Details about the 2-tuple model can be found on Appendix B.1.
The data provided by the stakeholders to assess an alternative low-carbon action's feasibility at the port of Piraeus is elicited in a 5-term scale.Both intervals and numerical values are transformed in 2-tuples to unify all information in the same scale.For this purpose, the transformational methodology presented by Herrera, Martıńez, and Sánchez (2005) is used and adapted to convert the variables in the necessary linguistic form.Details about the transformation to 2-tuples and the deviation from Herrera, Martıńez, and Sánchez (2005) can be found on Appendix B.2.

The 2-tuple TOPSIS-Monte Carlo integration
After unifying the data, the computed 2-tuples are inserted as input to the 2-tuple TOPSIS multicriteria analysis model (Labella et al. 2020) to provide the final prioritisation.The original TOPSIS method is based on the concept that the best alternative is the closest to a positive ideal solution and the furthest to a negative ideal solution, and it is very often employed in support of climate policymaking, where uncertainty is present in various decision-making problems (Doukas and Nikas 2020).It is also found among the most popular multi-criteria decision aid methods for both environmental assessments and in the transportation sector (Broniewicz and Ogrodnik 2020).Applications in ports range from selection of port community systems (Munim et al. 2020) to the assessment of ports' competitiveness (Nguyen and Woo 2021).The 2-tuple TOPSIS version used in this study handles data as 2-tuples, providing results in a format that is easily comprehensible by the stakeholders.A detailed description of the 2-tuple TOPSIS model is presented in Appendix B.3.
The existence of intervals in the multi-criteria framework that are aggregated in a universal form allows taking into consideration various existing uncertainties and enables post-modelling analysis to increase robustness of the prioritisation of the actions evaluated.Based on the low computational requirements of TOPSIS (Zhu et al. 2013) and its ability to be easily combined with other techniques (Zhang and Siu Lee Lam 2019), as well as exploiting the heterogeneous variables, we accompany the multi-criteria analysis with a Monte Carlo simulation to quantify uncertainty.Following an integrated approach, for each iteration in the simulation, the 2-tuple TOPSIS model recalculates the ranking and evaluation of the alternatives.Monte Carlo simulations have been performed in literature, both as standalone in the shipping sector to predict the energy efficiency of the ship (Coraddu, Figari, and Savio 2014) and combined with multi-criteria methods to evaluate alternative energy carriers in ferry transportation (Aspen and Sparrevik 2020), ship internal layout designs (Jafaryeganeh, Ventura, and Guedes Soares 2020), and alternative marine fuels (Hansson et al. 2020).However, to the best of the authors' knowledge, such a combination has not been suggested for evaluating low-carbon alternatives in ports and not in models with heterogeneous variables.Table 3 summarises the key methodological tools used in the study and presented in Section 3 and Appendix B.

Input data
The eleven technologies and interventions selected are evaluated against four criteria: reduction of CO 2 emissions, LCOE, investors' risk, and implementation time, as described in Section 3.1.Assessments against these non-homogeneous criteria are unified in 2-tuples, based on a 5-term linguistic scale, which comprises the following terms: {Very Low (VL), Low (L), Medium (M), High (H), Very High (VH)}.It is noteworthy that, for the purposes of our analysis, the four criteria are unweighted.Aside from the risk assessments that were carried out with the help of two Piraeus port officials (first-hand resources), assessments against the remaining three criteria that feed into the initial decision matrix of the MCDA technique were based on second-hand resources.These include data retrieved from literature and technical datasheets from pieces of equipment, which were adapted to the Piraeus port context following the assumptions described in Tables 1 and 2, with the help of stakeholders.Detailed calculations and handling of the assumptions, as well as an extended list of these sources are presented and described in the Supplementary Material.The evaluation and adaptation of the input in close cooperation with port authorities ensures reliability of the data, as well as that results are relevant and exploitable from a managerial perspective in the real world.The accompanied sensitivity and robustness analysis, apart from envisaging possible futures under conditions of uncertainty, also reduces biases in the input data, through the methodology's ability to provide additional results after triggering changes in the values of the input, thereby enhancing result reliability.
The resulting input data for each technology and/or intervention, to be aggregated into 2-tuples and then used in the analysis, is presented in Table 4.

Application of the model
The input data presented in Table 4 is transformed in 2-tuples based on the framework described in the Methods section.The aggregated data, to be used as inputs into the 2-tuple TOPSIS method, are presented in Table 5.In the conversion to 2-tuples, 'LCOE' and 'implementation time' are reversed to reflect the behaviour of benefit-type criteria (i.e., in Table 3 the smaller the LCOE/implementation time, the better the evaluation of the alternative).
Based on the input data in Table 5 and the 2-tuple TOPSIS method, Table 6 and Figure 1 present the evaluation and prioritisation of each low-carbon technology/intervention in the same 5-term linguistic scale used for the aggregation.
Table 3. Methods used and integrated within the framework of the study.

Type
Brief explanation Relevant Study 2-tuple model The 2-tuple model is a tool that translates data and values of different forms in linguistic terms to enhance comprehensibility.

(Herrera and Martínez 2000) Heterogeneous variables
Each variable encompasses a different meaning, thus requiring the provision of input data in different forms.These forms usually include numerical values, intervals, and linguistic terms.Here, all this input is transformed and homogenised in the uniform 2-tuple linguistic format.(Herrera, Martıńez, and Sánchez 2005) 2-tuple TOPSIS The 2-tuple TOPSIS, is an extension of the TOPSIS MCDA technique that can draw from data in the 2-tuple form, and similarly provide the output in the 2-tuple form, while ensuring no loss of information in the linguistic terms.(Labella et al. 2020)

Monte Carlo Simulation
A Monte-Carlo simulation is a common technique that performs a large number of iterations with the aim to quantify the uncertainty in the input data.Here, 1,000,000 iterations are performed to increase the robustness of the calculated TOPSIS prioritisation (Fox, Axsen, and Jaccard 2017) As seen in Figure 2, replacing outdoor lighting with more efficient LED bulbs is prioritised as the best alternative, with an evaluation in the 'very high' term of the scale.Efficient lighting schemes are usually regarded as cost-effective, providing significant amounts of energy savings, especially in industrial installations, even when replacing the already efficient CFL bulbs (Paul, Kamath, and Mathew 2017).When coupled with automation systems, the environmental impact can be maximised (Ożadowicz and Grela 2017).However, such actions, albeit highly cost-effective and a necessary starting point for the port's action plan, fail to achieve the targets without additional measures.On the other hand, the rest of the 'energy efficiency' cluster performed relatively worse, with the thermal intervention on buildings being second to last and the monitoring system being above medium.Both these actions' lower performance is driven by the low emission cuts in both cases, especially compared to the non-negligible LCOE in the case of the thermal renovations.At the same time, the engaged stakeholders considered the monitoring system as slightly difficult to implement at scale.Furthermore, extensive thermal renovations are deemed to potentially create discomfort, especially considering the role of the port for the country's tourist sector.
Following the replacement of outdoor lighting, on the mobility front hybrid equipment seems to rank high, preferred over electric equipment, with the latter interventions being positioned around and below medium.From this, we can deduce that efficient equipment to be used at full capacity (e.g., vehicles for container stacking and lifting) is preferred to equipment highly uncertain in terms of usability, like vehicles for micromobility.The latter's efficiency lies in the acceptability of technologies like e-bikes and e-scooters or notions like shared micromobility, as discussed in Section 3.1, which should not be taken for granted, especially in workplaces (Shaheen et  2020).This is also evident inside the electrification cluster of mobility.Electric forklifts performed better than the rest, however still less competitive overall, considering smaller performance on emissions, as a result of less intensive use.
Moving from the actions evaluated as of high priority, the installation of a PV power plant also demonstrated potential as a possible action.It ranked in the upper-medium level, contributing significantly to the reduction of CO 2 emissions by replacing existing electricity supply in numerous high-demand areas.The port of Piraeus has already been operating a small PV park since 2016, which has been found efficient (Spyrou and Kontogiorgi 2019).Based on the assumptions in the input data, the installation of additional 1 MW of solar PV-that is, more than double the capacity of the existing installation-could be a competitive action.However, since the port draws electricity from the interconnected national system, the emissions reduction potential (and, hence, the evaluation of the action) is subject to the decarbonisation progress of the power generation system in the country.Still, as a major electricity consumer, the port can play an important role in decentralising generation in the system, moving production much closer to where electricity is consumed.
Finally, two high-cost actions yet with high amounts of absolute emissions reduction potential, LNG and cold ironing, are averagely placed.These actions constitute a mandatory prospect for ports, especially after 2023 when the IMO targets will come in effect.Here, we observe that, despite the high investment capital required, their benefits balance the costs.In fact, emissions cuts and energy savings drive LCOE values down, rendering the two actions averagely competitive compared to the rest.Although emissions cuts refer to those emitted by the ships during berth, and thus considered indirect for the planning of the port, these actions contribute to the broader sustainability of the sector by attracting and offering relevant services to more efficient ships.However, considering the obligatory nature of these actions for ports, efforts need to be intensified to increase competitiveness, by optimising implementation time and taking actions to mitigate stakeholders' perceptions that these actions feature significant risks.

Sensitivity analysis
Considering the uncertainty from actions expressed in the assumptions of Table 1 and intervals of Table 4, changes in constants of Table 2, as well as inherent uncertainty over the acceptability of interventions for micromobility, we perform a sensitivity analysis to study the effects caused by the change in different parameters, including additional funding to support the costlier but also mandatory interventions.

Variation of the grid's emissions factor
In their 2017 financial statement, which is the latest to calculate environmental impacts, port authorities assumed an electricity grid emission factor of 850 gr.CO 2 /kWh.However, drawing from the latest UNFCCC submissions of Greece, in our analysis we assumed a factor of 606.1 gr.CO 2 /kWh as shown in Table 2.The difference between the two close periods is attributed to the country's efforts towards decarbonising its power generation sector, pledging to phase-out lignite by 2025 and using the more efficient natural gas as a transitional fuel, while increasing the share of RES (Nikas et al. 2020).These rapid changes in the system increase uncertainty of electrification-related actions, such as switching fuel or increasing consumption efficiency.In Figure 3 we examine the changes induced in the evaluation of the interventions, based on the progress of decarbonisation, as expressed by changes in the emissions factor.
Interventions aiming to replace electricity consumption sources with more efficient equipment, most notably LED outdoor lighting (although the same applies to thermal interventions, monitoring system, and PV installation to different extents), are the ones with evident decreases in their evaluations.This is expected since no fuel switch takes place and, therefore, as the emission intensity of the system improves, the emission reduction margin decreases.On the other hand, in the case of cold ironing where conventional fuel is substituted by electricity, the improved emissions factor subsequently improves the performance of the intervention.This leads to cold ironing placing above all other medium-evaluated actions for an emissions factor lower than 450 gr.CO 2 /kWh.Although such improvement in the power generation system would require drastic changes outside the port's jurisdiction, our results indicate an additional reason for considering cold ironing as a very competitive action due to its large potential especially as decarbonisation progresses.Also, this should motivate national policymakers to enhance efforts in decarbonising the electricity system, as this would provide opportunities for sub-national and/or non-state actors-in our case the port of Piraeus-to proceed with electrification actions that can be much more efficient under such circumstances.This also applies for the port itself since, despite the reduced performance of PVs as decarbonisation progresses, there is a positive trade-off with the increased performance of cold ironing that could be taken into consideration in favour of PV installations.Finally, although the LNG supply network intervention is not directly affected by changes in electricity-triggered emissions, its evaluation reduces as a result of the improved performance of cold ironing.

Micromobility acceptance
As already discussed, a key parameter influencing the performance of micromobility relies heavily on the acceptability of alternative vehicles like e-bikes and scooters by the users to cover their intraport transportation needs.As described in Table 1 and after discussions with the stakeholders over existing limitations-for example, workers carrying equipment being less willing or even capable of using alternative vehicles-it was assumed that the equipment could be used in total between 1 and 4 hours per day.In Figure 4 we examine changes triggered in the ranking of the alternatives based on two scenarios that assume the usage of micromobility to be in the endpoints of the interval, representing the lower (1 hour) and higher (4 hours) use of the range used in the baseline analysis (Figure 2).On the one hand, we observe that the high-use scenario renders micromobility slightly more competitive, surpassing all other medium-ranking alternatives and getting much closer to hybrid vehicles.However, it does not trigger significant fluctuations in the evaluation of the other actions.On the other hand, the low-use scenario triggers broader changes in both the evaluation of the alternatives and the ranking, with micromobility falling back in this case, causing significant improvements for actions of the same cluster (electric buses and forklifts).Interestingly, cold ironing and the LNG-related actions alike seem to gain from the worse performance of micromobility.This behaviour between the two scenarios, in terms of affecting both the evaluation of micromobility and-in the low-use case-the evaluation of other alternatives too, traces back to the definition of the positive and negative ideal solutions of the TOPSIS method but also offers key insights into the tendencies of micromobility.Specifically, since micromobility is not the worst alternative in the baseline scenario, positive changes avoid spillover effects to the other alternatives.On the contrary, the low-use scenario forces LCOE to become much larger due to the very low energy savings compared to the required budget.This changes the negative ideal solution of the method and affects the evaluation of the other alternatives, especially electric buses being the previously worst alternative, generally and in terms of LCOE.We can deduce that, in the bestcase scenario, micromobility is an average overall action that could be considered due to the low investment cost.However, it is vital to accompany this action with campaigns to inform users,  either citizens or port personnel, on the benefits of micromobility.Otherwise, moving to the worstcase scenario, micromobility becomes far less competitive to justify including in the port's action plan, despite the lower investment cost.

Cold Ironing and LNG funding
Cold Ironing and LNG Refuelling infrastructure are the two alternatives with the highest investment costs, yet with an obligation to construct Cold Ironing and LNG infrastructure until 2025 under the EU Directive 2014/94.In Figure 5, we examine the impact of reducing the contribution of the port by 30% in the initial investment cost through existing funding schemes like the Connecting Europe Facility (CEF) programme and established synergies with the private sector.We observe that, because the two actions were already competitive in terms of their LCOE even without additional funding, the effect of the funding scheme is not drastic; yet, it improves the evaluation of the two alternatives without affecting other actions.As such, available funding can contribute to the implementation of the interventions, increasing their investment attractiveness, and consequently accelerating their penetration.Especially considering that the LCOE of these high-cost investments was already better than most actions in the reference scenario, external investments could unlock the potential to bypass bureaucratic concerns related to the high capital required and reduce implementation time.This could make them more appealing to stakeholders and improve the performance of LNG and cold ironing in the two criteria (implementation time and risk) that pulled them down in the ranking.

Robustness analysis
After uncovering key tendencies regarding the evaluation changes based on key uncertainties, we accompany the sensitivity analysis with a Monte Carlo simulation to find the robustness of each alternative if all parameters fluctuate at the same time.As already explained in the methods section, the employed 2-tuple TOPSIS-based framework with heterogeneous variables is not computationally intensive, allowing us to perform a large number of iterations.Here, we perform 1,000,000 Monte Carlo simulation iterations, in the upper limit suggested by Tervonen and Lahdelma (2007), based on randomised parameters provided in Table 7.These parameters refer to uncertainties explained in Table 1 leading to the use of intervals, as well as those analysed in the sensitivity analysis section.Considering the divergent parameters of uncertainty, including but not limited to efficiency of equipment, hours of operation as well as acceptability levels, we assume a uniform distribution of the random numbers selected in the intervals.This is also the case in similar climate policy exercises based on Monte Carlo to handle uncertainty, when minimum-maximum values exist (Fox, Axsen, and Jaccard 2017).
The results of the evaluations of each alternative based on the 1,000,000 iterations are publicly available. 1Figure 6 presents an overview of these results to express the fluctuations and concentration areas of the evaluations for each alternative under the uncertainties described in Table 7.
This robustness analysis shows that replacement of outdoor lighting still ranks first in terms of average values, although the margin between the other alternatives is reduced, due to the high level of uncertainty of the action.Still, observations are fairly well distributed within a range of lowerhigh to very high evaluations, indicating that the action is among the most competitive even in an uncertain environment.Based on the results of the sensitivity analysis (Section 4.3), the most significant factor of the gap decrease is the decarbonisation progress of the system, with minor decreases being attributed to improvements in other alternatives.The two hybrid mobility actions also maintained their position in the second and third place, with an evaluation between medium and high without a large range.The main reason for this traces back to the lower uncertainty surrounding these actions.The equipment is expected to be used to its maximum potential Based on the assumptions of Table 1, random numbers for the anticipated share of energy savings are selected within the range of each action (both for the lower and upper endpoint).For the monitoring system, thermal and electricity savings are randomised independently and then aggregated.This choice affects the range of CO 2 emissions cuts providing a different interval within the one shown in Table 3.Also, LCOE is changed based on these values assuming the average of the interval in line with the explanation in Section 3. Following the sensitivity analysis in Section 4.3, we assume that the subsidisation level for these technologies ranges between 0-30% of the total investment cost.In each run, the two actions receive a different subsidy.
triggering adequate amounts of savings, which justifies the low fluctuation.As with the baseline, these three actions clearly stood out, indicating that they can be an appropriate starting point in the port's action plan.
Moving down to the six medium-ranking actions, although they are the same group as in the baseline, some evident changes should be noted.
When accounting for uncertainty, cold ironing surpasses all others in terms of average values.This was already hinted in the sensitivity analysis, since cold ironing showcased positive improvements under the different scenarios, but here it is further validated.In fact, cold ironing presents a smaller range of observations compared to the other average-ranking actions, while most evaluations are concentrated around the average/median and even tilt closer to the maximum than the negative value.On the other hand, the distribution of observations in the LNG-related intervention is more balanced, although it presents a higher range, indicating higher uncertainty than cold ironing.From the two high-cost actions of the alternative fuels cluster, a higher prioritisation of cold ironing is evident.As an electrification-type action it has a higher savings potential in the long run, especially as decarbonisation of the power system progresses.This, however, acts as both a requirement to unlock this potential and an opportunity to make an already competitive action even more competitive, especially considering that the differences between cold ironing and LNG are less obvious in the current state.The fact that LNG performs better in the baseline, which better reflects the current status, indicates that LNG may play a similar role as a transitional fuel to shipping and ports as natural gas in power generation, and avoid becoming a stranded asset in the future, considering that natural gas could turn out more than a 'transition' fuel under current EU policy (Nikas et al. 2021).Electric forklifts, the monitoring system, and the PV installation all present some similar characteristics, evaluated between the medium and upper-medium terms but also presenting small fluctuation, with a highlight on electric forklifts, which appear to distinguish from their cluster (mimicking to a smaller extent the behaviour of hybrid equipment).As such, and based on the available investment capital, these actions can accompany portfolios that emphasise the higher evaluated interventions.Also, the fact that PV installation remains fairly competitive, even under progress in the country's decarbonisation front, indicates that it can be a very useful supporting action to further enable other electrification-related actions.Although this characteristic was not evaluated in the model, the resilience presented by PVs as well as the underlying indirect benefits of 'greening' the electricity consumed by the port imply that it can be play a vital role as a complementary action.
The remaining two electric mobility actions are still evaluated at the medium-to-lower ends, with improvement signs in the evaluation being mostly attributed to the fluctuations in the assessment of micromobility.This is largely evident in the maximum value of electric buses, which is similar to the evaluation of this action under the scenario featuring low use of micromobility.This indicates that this higher value is only an outlier with a low share of appearances in the robustness simulation, triggered by the evaluation of another action, and does not necessarily indicate positive signs for electric buses.This is also reflected in the evaluation of thermal renovations, which rank last.In contrast, the evaluation of micromobility seems to reflect the range found in the sensitivity analysis, indicating that the uncertainty induced is mainly self-inflicted and traces back to the acceptance of the action in the public sphere of port users and workers.Generally, consideration of these lower evaluated actions can be less prioritised, and even for the low-cost micromobility action only if certain conditions are met, like campaigns to increase acceptability as discussed in Section 4.3.

Discussion
Aggregating the remarks on the previous analyses, key takeaways can be finally deduced to inform future action plans in the port of Piraeus.Replacement of outdoor lighting with efficient LED bulbs can be a quick-win action, followed by the purchase of hybrid vehicles for cargo handling, which can complement the investment portfolio mixes, considering high robustness against uncertainties.However, to achieve further decarbonisation, actions should extend to deeper infrastructural changes.Cold ironing was found a robust alternative, showcasing improvements under uncertainty, thereby expanding our knowledge on the positive behaviour identified by Argyriou, Sifakis, and Tsoutsos (2021).It is also found that higher decarbonisation levels can improve the action's competitive advantage, which also indicates that the installation of PVs, despite the medium evaluation as a standalone intervention, can have positive indirect impacts and as such could be considered together as part of the action plans.LNG supply infrastructure proved more efficient in the current state of the system, highlighting that supporting LNG-fueled vessels with the necessary infrastructure and investments from the port's side can be of significant transitional importance, not unlike natural gas in the power generation sector.On the other hand, despite its low cost, micromobility appears to be heavily dependent on its acceptability-therefore, if implemented, it should be accompanied by targeted supporting campaigns.In Table 8 key policy implications are extracted based on the results of the model including the sensitivity and robustness analysis for each action.
Although these policy insights are tailored to the needs of the port of Piraeus, it is possible that some remarks can be extrapolated to other ports, like for example the positive reaction of cold ironing to uncertainty, especially in ports with similar characteristics with the one studied here.Still, the employed framework addresses the first research question posed in Section 1, managing to provide a prioritisation of actions with and without uncertainty that can be used by port officials.These results can then be a benchmark for other ports in exercises of similar methodological setups (or even the one described here) to understand similarities and differences in each case.
Table 8.Key policy implications per action for the Piraeus port extracted from the modelling analysis.

Suggested Actions Policy Implications Cold Ironing Infrastructure
The intervention receives an average evaluation, but appears to be very robust, improving under conditions of uncertainty.As such, it distinguishes from the rest of the higher cost interventions.The action can significantly benefit from the progress in decarbonising the country's electricity generation sector, as well as possible subsidisation, which can help reduce the risk stakeholders believe exists.LNG Refuelling Infrastructure Similar with cold ironing, this action is averagely evaluated, although it falls back compared to cold ironing under conditions of uncertainty, considering that it allows berth of ships fuelled by LNG instead of diesel, which is a fossil fuel even though significantly more efficient.Still, under current conditions, the action is prioritised higher than cold ironing, indicating that such interventions can be exploited in the short term, assisting the transition of the sector, acting as a transitional fuel (like natural gas in power generation).Micromobility A highly uncertain action, which can be of medium importance if acceptability levels increase, but with high risk of being significantly underused.The low cost required makes it an effective choice, which however needs to be accompanied by promotion campaigns to increase acceptability.

Electric Buses
The high cost to obtain the electric buses compared to their limited environmental impact (in comparison) makes the action less competitive.However, if there are strong indications that micromobility will be of no effect, a shift to this costlier alternative can be considered to shift intra-port road transportation to a lowcarbon alternative.

Electric Forklifts
In terms of goods handling in the port, the electric forklifts are less competitive compared to the hybrid vehicles, since the less intensive loads handled leave lower potential in terms of environmental impact.Hybrid Vehicles for Container Stacking and Transporting (VCST) The intensity of the loads handled by such types of vehicles indicate that the equipment will most probably be used to its highest potential, and as such showcase high robustness as an investment.With evaluations consistently above medium, hybrid equipment alternatives could be considered as complementary investments.

Hybrid Vehicles for Empty Container Lifting Energy Consumption Monitoring Systems
For the buildings of the port, the energy monitoring system is an efficient alternative, and although not as impactful as the larger-scale port interventions (e.g., cold ironing), it is still more competitive than thermal renovations in buildings, despite stakeholders' concerns over the feasibility and efficacy of implementing such a system.Based on these concerns, if chosen as a solution for the buildings of the port, end users need to be informed on ways to optimise the use of such equipment to increase environmental gains.

Outdoor LED Lighting
Replacing outdoor lighting with efficient LED bulbs was deemed the overall best alternative.As a lower cost action, it is found to be a quick-win investment, triggering non-negligible savings without requiring large capital.Still, for this reason, it can only be considered a starting point for the port's action plan.

Energy Interventions in Buildings
Contrary to the monitoring system, large-scale thermal renovations are deemed less competitive, driven by the higher costs as well as the potential discomfort, especially in a port heavily relying on tourism.PV plant installation PV installations were found to perform averagely with a relatively stable robustness even under an increased decarbonisation of the electricity generation system scenario, which limits the environmental contribution potential for the action.However, the ability to improve the emissions factor on the electricity consumption of the port, based on self-generation as well as reduce energy costs, indirectly affects other electrification-based actions, most notably cold ironing, providing an extra motive on considering this investment.
From a broader sectoral policy perspective, responding to the second research question posed, it is evident from the previous prioritisation and uncertainty analysis that the sector is heading in the right direction in terms of the obligations imposed for specific actions like cold ironing and LNG supply infrastructure.These actions have significant environmental impact, while being financially competitive in terms of their LCOE.However, what pushed these actions lower in the ranking is the higher implementation times and most importantly the stakeholders' evaluation referring to the perceived investors' risk.Considering that Boile et al. (2016) also identified stakeholder engagement as a key pillar to energy performance in ports, IMO's roadmap and EU policies need to consider stakeholders' concerns.This includes infrastructural feasibility of implementing certain actions in each port, as well as expected delays in developing the infrastructure, which could delay any possible financial or other types of benefits post-2030.This is especially true for LNG-related investments, and to a lesser extent hybrid equipment.Further delays in publishing the roadmap and specific guidelines for the sector's low-carbon transition could disrupt the transitional role of this type of fuel, locking the system to a natural gas-based pathway.On the other hand, quick action could smoothen this path, while pushing for the decarbonisation of the power generation sector, which could enhance the dynamics of electricitybased solutions, most notably cold ironing.

Conclusions
Ports are expected to play a significant role in improving the efficiency of maritime shipping, by directly reducing their emissions and contributing to the sustainability of their local communities, while enabling the development of more efficient vessels that could be accommodated by their infrastructure.Under European Directives, seaports are already obliged to develop green infrastructure, pending guidelines from the publication of IMO's roadmap.This research contributes to the establishment of this roadmap and the relevant policy context, by introducing a multicriteria methodological framework that allows relevant policymakers and decision-makers operating in the shipping and port sectors to evaluate low-carbon interventions.The framework is based on the 2-tuple TOPSIS method, heterogeneous variables, and a fully integrated Monte Carlo simulation to account for the uncertainty accompanying each intervention, in a novel methodological setting for the sustainable planning of ports.The limited computational needs of the framework enable decision-makers to account for a diverse set of uncertainties and quantify their impacts in a format that is easy for them to comprehend and translate into action to reach the most robust policy mixes.The evaluation framework is used in a case study for the port of Piraeus.In particular, addressing the study's first research question, a concrete prioritisation of actions is reached with and without accounting for uncertainty to enable extracting policy insights that support authorities in evaluating green development projects, as part of the port's recent master plan.As part of the second research question, we highlight the right direction of the current policy context, while identifying topics that should be raised especially in relation to stakeholders' perceived risks.
Significant effort has been made to identify indicative assumptions in the context of the port of Piraeus and adapt literature values accordingly.However, future research can draw from further empirical evidence to increase the accuracy of the policy suggestions, or to account for additional uncertainties not covered by this research-that is, long-lasting strikes (Spyropoulou 2020) that could increase the implementation time of each action, or reduced passenger demand and opposition to external investments (Zhu et al. 2019).The framework can also be extended to cover more interventions, as well as additional criteria, such as the reduction of other pollutant emissions (e.g., SO x and PM).Finally, the framework can be enhanced to incorporate the preferences of multiple stakeholders, and in some cases of conflicting views, in a more structured group decision-making context, since the current results are based on the subjective opinion of stakeholders from the Port of Piraeus.However, acknowledging that both assessments against specific criteria and scales of implementation are adapted to reflect the port of Piraeus, as well as that investors' risks are highly dependent on the port's officials' perceptions, the framework can easily be adapted and exploited in other ports.Drawing from the conclusions reached in this study, better informed portfolio analysis that encompass MCDA-identified risks in the evaluation criteria (e.g., Forouli et al. 2019) can provide detailed investment mixes for the port, also accounting for the uncertainties and behaviours uncovered in this study, to optimally allocate a port's available investment capital.

Notes
1.The results of the Monte Carlo iterations are publicly available here: https://doi.org/10.5281/zenodo.51200422. https://www.elemedproject.eu/(Iris and Siu Lee Lam 2019).Examples in Europe include the smart energy-efficient and adaptive management system in Valencia, which monitors equipment and lighting energy consumption, as well as a similar system in Koper, which exploits digital meters and communication sensors installed in port transformers (Sdoukopoulos et al. 2019); and the port energy consumption monitoring tool in Jade Weser, which apart from the monitoring services provides approximate data for CO 2 emissions.
In our case study, real-time energy monitoring systems in three buildings (administration building and two buildings in cruise terminal) are selected for evaluation, for improved energy management.

A.6 Outdoor LED lighting and energy interventions in buildings
Buildings account for 40% of total energy consumption and 36% of GHG emissions in the European Union.Following the European target for increasing energy efficiency by 32.5% up to 2030, ports' authorities focus on their buildings' energy upgrade.The ports of Helsinki, Stockholm, Aalborg, Ghent, Rotterdam, and Copenhagen-Malmo have implemented a variety of such measures including LED lighting, solar panels, efficient heating systems, thermal insulation, efficient ventilation system, lighting sensors and heating pumps while remarkable results have been achieved (Tsai et al. 2018).
In the case of Piraeus, installation of LED lighting and lighting control systems in car parking, the cruise terminal, and the containers station, as well as building interventions regarding LED lighting, lighting control systems, solar panels, and heating and cooling systems with high-efficiency heat pumps in three buildings (administration building and two buildings in cruise terminal) are evaluated.

A.7 Installation of solar photovoltaic (PV) infrastructure
Although shift to renewables is sometimes seen as a challenge for ports, the development of RES infrastructure in ports has also become increasingly popular, partly addressing their energy needs (Iris and Siu Lee Lam 2019;Sdoukopoulos et al. 2019), often coupled with cold ironing projects (Innes and Monios 2018).Several European ports have recognised the importance of reducing CO 2 emissions through the reduction of conventional fuels and the exploitation of cleaner energy sources and participate in relevant projects, including large-scale RES infrastructure, such as wind turbines, offshore wind turbines, biomass units and PVs (Acciaro, Ghiara, and Inés Cusano 2014), and with plans for further expansions within the next few years.
In this study, the expansion of the existing PV infrastructure in the port of Piraeus is evaluated by increasing installed capacity by 1 MW towards generating own green energy onsite.

B.1 The 2-tuple model
The 2-tuple model (Herrera and Martínez 2000) consists of the 2-tuple linguistic representation ðs; aÞ, where S is the linguistic term and a is a numerical value representing a symbolic translation.Suppose S ¼ s 0 ; . . .; s g � � a linguistic term set and β 2 0; g ½ � the result of a symbolic aggregation, where g þ 1 is the number of the set's terms.Let S ¼ s 0 ; . . .; s g � � be a linguistic term set and β 2 0; g ½ � be the result of a symbolic aggregation operation, where g þ 1 is the cardinality of S. Let i ¼ round β ð Þ and α ¼ β À i be two values, such that i 2 À 0:5; 0; 5 ½ Þ; then α is called a symbolic translation.The symbolic translation of a linguistic term s i is a numerical value within À 0:5; 0; 5 ½ Þ indicating the difference of the information between the calculated value β 2 0; g ½ � and its closest element within S ¼ s 0 ; . . .; s g � � , indicating the content of the closest linguistic term Sði ¼ round β ð ÞÞ.

Transformation of heterogeneous variables to 2-tuples
In order to convert arithmetic values 0; 1 ½ � into F S T ð Þ the following process is utilised.Supposing that F S T ð Þ is a set of fuzzy linguistic terms S T ¼ s 0 ; . . .; s g � � , an arithmetic value θ 2 0; 1 ½ � is converted to a fuzzy set in S T by calculating the membership of θ's value to the fuzzy value linked to the linguistic terms of the S T scale.The transformation function τ NST is presented below: The functions μ si are called membership functions for the linguistic labels s i 2 S T , representing a parametric function a i ; b i ; c i ; d i ð Þ.
Supposing that is an interval assessed in 0; 1 ½ �, in order to transform it, it is assumed that this interval is represented by a membership function of fuzzy terms as follows: Afterwards, the transformation function τ IST converts the interval I to a fuzzy set in S T .
F(S T ) is the set of fuzzy linguistic terms defined in S T and μ I and μ Sk are the membership functions of the fuzzy sets linked to the interval I and s k , respectively.
The interval and numerical values are transformed in fuzzy sets, with a membership value corresponding to each label of the linguistic scale.For each alternative, Herrera, Martıńez, and Sánchez (2005) aggregate the membership values on each label in the different criteria, using average values, to receive a similar fuzzy set.The new membership values are then aggregated to from the 2-tuple evaluation of each alternative.Although this method is useful to provide a quick indication, the final prioritisation relies on average values, which is not optimal.To avoid this process, in this study we propose that, instead of aggregating and then transforming the new fuzzy sets in 2-tuples, the transformation be performed first and then feed the results in a more elaborate multi-criteria framework.The transformation of the fuzzy sets F S T ð Þ computed from the interval or numerical functions is described below: Note that, compared to Herrera, Martıńez, and Sánchez (2005) we apply the x function directly in the fuzzy sets instead of aggregating them, which provides a 2-tuple for each initial data point.

B.3 The 2-tuple TOPSIS model
The steps of the 2-tuple TOPSIS model, adapting from Labella et al. (2020) are presented below: (1).Setting a weights vector U ¼ u j ; 0 À � 1�n , in which u j 2 U is a linguistic preference for criterion c j and U is a set of linguistic terms, with U ¼ u 1 ; u 2 ; . . .; u p � � transformed into a 2-tuple linguistic decision table U ¼ u j ; 0 À � 1�n .As is the case here, equal weights can also be set.
(3).Defining the decision table , in which r ij 2 S is the linguistic value and β ij is the symbolic translation for the alternative a i based on the criterion c j , and S is the linguistic terms set, with S ¼ s 1 ; s 2 ; . . .; s t f g, based on the transformation of the initial input from non-homogenous variables to 2-tuples as described in the previous section.
(5).Calculating the positive and negative ideal solutions jc j 2 B 0 g, in which i ¼ 1; 2; . . .; m, j ¼ 1; 2; . . .; n, and B and B 0 are the sets of the benefit and cost criteria respectively.(6).Determining the distance of each alternative form the positive and negative ideal solutions as:

Figure 1 .
Figure 1.Methodological framework and research process.

Figure 5 .
Figure 5. Sensitivity analysis on the funding of Cold Ironing and LNG.

Figure 6 .
Figure 6.Alternatives' assessment range: The crosses mark the median values among the iterations, the lower and upper limits of the box indicate the first and third quantile, the bars represent the minimum and maximum values, while the dashed line showcases the average evaluation for each alternative.

Table 1 .
Low-carbon port technologies and interventions considered in the study.

Table 4 .
Input data, prior to aggregation.Detailed sources used for the assumptions are provided in the Supplementary Material

Table 7 .
Randomisation Strategy for the Monte Carlo simulation.