Decarbonisation of islands: A multi-criteria decision analysis platform and application

.


Introduction
The EU has undertaken the task of reducing its greenhouse gas emissions by 80%-95% compared with 1990 by 2050, a decision compatible with the reductions that the group of industrialised countries must make if the global temperature increase is to be limited to two degrees Celsius above pre-industrial levels.Towards this end the EU has undertaken several initiatives aiming to assist the decarbonisation of its islands.Islands often have access to significant renewable sources of energy but regularly find those sources outside their ability to use, not only for economic reasons, but for social as well as local environmental concerns.This paper attempts to expand this understanding of geographic islands' positions and help local planners in transitioning their energy systems using an integrated and flexible approach.
To this end, this paper continues and completes the application of the novel energy planning decarbonisation platform, REACT-DECARB, developed within the context of the REACT 1 Horizon 2020 EU project, on eight islands within the EU.The platform provides an integrated mixed-integer linear programming model for energy scenarios creation coupled with life-cycle assessment and multi-criteria decision analysis modules.The overall scope of this methodology and its application is to establish a wider framework able to be used as a template to foster clean energy development and island decarbonisation.The data and results coming from the implementation of the various portions of the project, and the data from the theoretical analysis of the REACT-DECARB platform, can also serve as a benchmark for future studies.The platform seeks to fill the gap that exists when assessing energy systems on islands by guiding energy planners beyond solely techno-economic assessments to develop more comprehensive evaluations which include both social and environmental criteria.
For this application of the platform an environmental life-cycle assessment (LCA) and a social evaluation of a number of energy scenarios on the islands are integrated with an economic levelized cost of energy (LCOE) analysis and technological assessments for the same scenarios via multi-criteria decision analysis (MCDA).The rankings provided by the MCDA are then tested for their sensitivity and further examined in order to allow for an inter-island comparison which gives energy planners a comprehensive view of the relative importance of criteria and preferences that can shape their island's energy future.
The structure of the paper is as follows: Chapter 2 provides the literature review for the study, introduces the basic concepts of LCA and MCDA and highlights their relevance for integrated island energy planning.Chapter 3 fully describes the overall structure and flow of analysis of the REACT-DECARB platform for decarbonisation of islands, presents the selected tools and methods for energy scenario development, the LCA and MCDA analyses and identifies the sustainability criteria used.Chapter 4 comprises the case study design and details the specific portfolio of energy projects for each island.Chapter 5 presents the overall MCDA analyses' results for the islands' scenarios, an interisland comparison of the used criteria and an overall discussion of the paper's findings.Chapter 6 presents the final conclusions of the paper.

Literature review
Islands with their small to medium electricity grids, frequent dependence on imported fossil fuels, connection to larger mainland grids or autonomous operation and often times rich renewable energy resources present a unique challenge for a decarbonisation transition.They are often forced to deal with interruptions in electricity supply along with fluctuations in electricity characteristics, such as voltage and frequency, which cause clear issues when trying to meet inhabitant's electrical demands.Due to their distance from the mainland they may also face a lack of skilled personnel and an absence of complete local energy planning.All of these issues can lead to increased costs for island electricity generation and for the price for electricity being subsidised in order to align with mainland prices as was done, for example, on the islands of Majorca [1] and Lesvos [2,3].
To consider climate change and to begin a decarbonisation transition phase, islands must pull away from fossil fuel reliant energy systems in favour of systems based on renewable energies.This shift requires a new energy planning agenda as renewable energy sources (RES) exhibit different characteristics in their operation and local impacts [4].A focus on only greenhouse gas emissions may lead to other environmental impacts being missed and result in other, unwanted consequences [5].
Islands' transitions underline the importance of methods that can assess and comprehensively evaluate the different aspects of the various energy production technologies available [6], and the impacts associated with them [7].There have been recent attempts to develop such evaluation methods where environmental criteria fall into the categories of resources, climate change, impacts on ecosystems and waste [8].Apart from these main categories, other specific impacts may also affect the acceptability of renewable energy sources, e.g., noise, aesthetic impacts on scenery and impacts on recreational use.Furthermore, the integrated assessment of the sustainability of an energy system also requires the selection of indicators that encompass the resource, environmental, social dimensions as well as an economic evaluation [9,10].
For the inclusion of the environmental dimension, in particular, lifecycle assessment (LCA) is a method to assess environmental impacts of various types of projects and is especially relevant from a sustainability perspective as it disregards local improvements that only result in the shifting of an environmental impact elsewhere [11,12].LCA can highlight which processes should be improved in priority order from an environmental standpoint.The method also attempts to link environmental performance to functionality, offers a quantification of multiple types of pollutant emissions and provides an accounting of the usage of raw materials based on the function of the product or system being assessed over its entire lifetime, from manufacturing to decommissioning and recycling [13,14].
Several software packages that assist in conducting an LCA, developed for both generic and specific applications, are available and a selection of some of these is presented in [15] and [16].Simapro, openLCA, GaBi, GEMIS and Umberto are examples of software designed to be able assist in conducting LCAs.It is important to also be aware that the choice of LCA software can impact the results due to differences in how the data is imported and then handled in the software [17] or from errors and differences in the databases they use [18,19].
At the same time, complex energy-environmental systems, like large wind and photovoltaic (PV) parks, which include both quantitative and qualitative data and engage competing groups of decision makers (DMs), usually need MCDA methods to support planning and decision processes with several alternatives and a multitude of criteria [20].MCDA methods provide a framework for collecting, storing, and processing all relevant information characterized by socio-political, environmental, and economic value judgements [21,22].They are used to clarify the planning process, to avoid various distortions, and to manage all the information, criteria, uncertainties and importance of the respective criteria [23].MCDA methods can alleviate problems caused by limited human computational power where intuitive or adaptive choices are replaced by a justified and jointly accepted model.In MCDA it is common to have DMs with conflicting preferences [21,24].
It should be noted, though, that MCDA methods can provide different results with the same data, and there is usually no means to objectively identify the best method for a particular case-study [25].This is compounded by the lack of consensus regarding which of the many approaches are best to use and under which circumstances they should be considered.While there have been attempts to provide a framework for MCDA method selection there is no universally accepted criteria for doing so [26,27].
Generally stated, the MCDA method to be used in public environmental problems must be well defined, the criteria adopted and their relevant weights must be easy to understand while the handling of inaccurate or uncertain data must be transparent [28].It must also cater to the amount of DMs involved and the number of alternatives and criteria present as well as the assigned weights that model the DMs' subjective preferences [29].The interpretation of the weights depends on the decision model and several techniques for eliciting weights are presented in literature [30].These vary from direct assessment to pairwise comparison methods.No correct weights exist that would allow comparisons between different procedures and the weights obtained depend on the technique used [31,32].
Both LCA and MCDA have been used to assess aspects of energy planning on both the mainland and islands, such as for energy system planning on Sri Lanka and Madagascar [33,34] as well as siting of offshore wind farms near Crete [35] and evaluation of wind-hydro projects on the island of Ikaria in Greece [36].Additionally, [37] provides a short review of studies on assessments of energy technology sustainability using MCDA methods as well as those that applied an LCA indicator in their assessments.
The particular issues that emerge with the use of LCA and MCDA techniques on islands relate to the nature of the islands and the specific system boundary adopted for the LCA analysis.Environmental load can exist both at the local level and at the country of origin for the RES technologies as well as those which occur during transport as noted in both [38] and [39].Additionally, islands can face a lack of local expertise with respect to energy planning, RES installation, operation and maintenance, acceptance issues and the need to cater to conflicting interests between local residents and visitors who only spend time there during holiday periods [40,41].
These concerns together mean that the LCA must be done carefully and meticulously in order to apportion costs and benefits appropriately, whereas MCDA techniques must be applied with caution to explicitly take into account conflicting preferences and the diverse criteria on islands as well as the difficulty in eliciting meaningful weights.
In summary, to be able to adequately and comprehensively assess an energy system's sustainability the environmental, economic and social dimensions, among others, need to be considered.MCDA has been shown to be an effective tool for integrating and managing all of these to support DMs in their planning and assessments.Further, LCA is an effective means of determining a system's environmental impacts and can be effectively integrated in an assessment using MCDA.
A. Barney

The REACT-DECARB platform
The energy transition that must materialise and come into full force to lead to a decarbonised energy system within the next 30 years entails a new approach to energy planning.It requires a holistic methodology introducing a number of changes covering and extending sustainability issues, communal appreciation and participation in decision making, a rigorous energy engineering rethinking, a smart grid as well as an ICT basis and infrastructure.Renewable energy resources together with an increased potential for electricity storage as well as public awareness of the need to balance demand and supply, changing energy use habits and adjusting to a new energy regime are prerequisites for a successful operation of a modern electricity system.
In the REACT-DECARB platform of analysis, the underlying initial model that was used for the optimisation process is based on a mixedinteger linear programming model, solved by a variety of solver engines and program frameworks.Since all islands considered were simulated with existing and potential renewable energy generation the renewable energy production subservice was a mandatory inclusion, along with net metering and battery storage [42,43].A significant number of scenarios were generated seeking electricity independence, and/or increasing RES and energy storage in each island's energy mix.Several cases also took the interactions with mainland grids into account.From these generated scenarios a shortlist was extracted based on heuristic and optimization methodologies.
Fig. 1 below identifies the structure of the REACT-DECARB platform, originally introduced in [43] and [44].The items in red, the LCA, the Multi-Criteria Decomposition and Decision Analysis and the finalization of the portfolio of scenarios together with their dynamic interactions and iterations are the focus of this work.
Data regarding technologies, costs, transportation to site, construction, energy production and storage scenarios for each island are obtained from Horizon 2020 REACT project Deliverables 2.1 through 2.4, [45], [42], [44] and [46] respectively, as well as in [43] and [47].This data is used in the subsequent analysis.
The life-cycle analysis for the systems involved comprised the construction, transportation, and installation phases.An appropriate software tool was employed for the analysis and several modules were developed to cater for the specific characteristics of the technologies installed [48].
For the evaluation of scenarios a MCDA using representative energy, environment, economic and social criteria was employed.The outranking method, PROMETHEE II, was used for the preference ranking of scenarios; other methods could also be utilized by the REACT-DECARB platform, such as TOPSIS [49], AHP [35], VIKOR [50], MABAC [51], MARCOS [52] and MAIRCA [53].
A number of key performance indicators (KPIs) commonly used for the analysis of energy systems were examined and from these a smaller group, appropriate to the environmental and economic analysis provided by the REACT project, was selected to enable the MCDA and the ranking of alternatives.Energy yield and LCOE values were obtained from [43] and local employment was determined from the REACT project partners and data from literature [54,55] adapted to the particular scenarios considered on each island.
The key criteria selected for the current analysis are: • Energy yield per inhabitant, [kWh/inhabitant].
• Local employment in job-years per inhabitant.
The selected criteria represent those most commonly used to evaluate energy systems [56], i.e., energy, environment, economy, and society, while also representing the main dimensions of sustainable development [30].More information on the KPIs used for the analysis of energy systems can be found in [42].

Case study design
Renewable energy sources tend to be widely distributed and the projects that convert these resources to useful energy must be located at the source of each.Furthermore, extensive use of intermittent renewables such as wind and solar to generate electricity must consider the temporal variation in the availability of these resources.This variability entails a need for special attention to system integration, storage options and awareness of potential transmission issues.At the same time, island characteristics such as the distance from the main grid, any related transportation costs and the fluidity of the number of people living on the island during the touristic season must all be taken into account.
The technologies examined in the present work comprise the following: Fig. 1.The REACT-DECARB platform for decarbonisation of islands: overall structure and detailed flow of analysis.
A. Barney et al. • Wind turbines • Solar PV • Electrical storage to operate either in conjunction with renewable generation systems or/and within the main grid of the islands.• Heat pumps for heating and cooling of buildings • Thermal storage units that operate in tandem with heat pumps to synchronise demand and supply.
The eight EU islands included in the Horizon 2020 REACT project are of differing sizes and have varying local climates.Fig. 2 below provides a sense of the scale differences between the islands included in the REACT project.Note that the y-axis of the Figure is logarithmic, so the smallest island in population and area, La Graciosa, is 1% and 0.1%, respectively, the size of the largest island, Mallorca.
All islands were evaluated for their tangible exploitable renewable energy potentials based on their specific environmental and regulatory conditions.This information was used to conduct energy system performance simulations which were evaluated using different technical key performance indicators to develop renewable energy projects on the islands.Detailed descriptions of each of the islands can be found in [45].
A number of different technical scenarios were developed for the islands in the REACT project.These scenarios were all generated using the characteristics and details of the island being considered though not every scenario was generated with the same objective.In some cases, complete electricity independence of the island using renewables and energy storage is targeted while other scenarios sought more incremental growths of renewables in the specific island's energy mix.In total 21 different scenarios were selected for further analysis and the capacities of the different technologies to be installed in each of the different scenarios for each island is shown below in Table 1, as first presented in [43].Solar PV, wind power, heat pumps and both thermal and electrical battery storage are used in these scenarios.More information on how these scenarios were generated can be found in [42].
The scenario colours in this table correspond to the scenario colours used in the following figures to increase the ease in scenario identification.Scenarios can be categorized according to the degree of autonomy sought.These range from full autonomy, in the cases of Gotland Scenario 2 and Isle of Wight Scenario 1, both with significant battery capacities, to varying degrees of partial autonomy.These degrees of autonomy can also be considered to reflect differing time horizons for potential implementation based on a chronological unfolding of an everchanging energy future with new technologies being introduced, reductions in the cost of RES technologies and batteries and changes in people's behaviour.
To rank the scenarios, the PROMETHEE II method estimates the net flow for each scenario according to specific criterion weights provided by the decision makers involved.These weights should represent the individual choices and sensitivities of the decision makers and show the relative importance of each criterion in the ranking.In this work, which was developed according to the needs of the REACT Horizon project, the weights were used to establish the sensitivity of the rankings to the four representative criteria for energy, environment, economy and society presented above.This was done by first giving all four criteria an equal weight, 25%, and subsequently with each criterion being assigned a 70% weight and the remaining three criteria with a 10% weight.
The results are analysed and discussed in the following section.

Results and discussion
The rankings of the scenarios for all islands, except Aran Islands where only one scenario was examined, are shown in Figs. 3 to 9.These scenarios represent a long term planning horizon aiming to address the 2050 target of zero carbon emissions that EU islands should aim for.The whole exercise is indicative of the difficulty of the endeavour ahead, which must mobilise both the necessary resources and the social acceptance for such an energy system transformation to take place.It is expected that similar steps should be undertaken for islandic grids in general in order for a full decarbonisation of energy systems to materialize.
La Graciosa Scenario 2, which only has solar PV, achieves the highest ranking when the LCOE and jobs criteria are given the highest weights.Scenario 2 is especially dominate in the case where LCOE is emphasised due to the increased cost of batteries assumed in the other two scenarios.Scenario 3, a mix of wind and solar PV with some storage capacity, ranks first when the energy yield and avoided CO 2 are weighted more heavily as its mixed production edges out only solar PV production and the carbon footprint of its wind production is also smaller than an all solar PV system.These changes in rankings indicate Scenarios 2 and 3 are relatively closely matched.
San Pietro Scenario 4, with the largest solar PV park and greatest energy production of all the island's scenarios, is top ranked in all cases considered, excepting when the avoided CO 2 criterion is emphasised, at which point Scenario 3, composed entirely of wind power, becomes slightly preferred.The San Pietro scenarios do not include batteries, so the switched ranking is the result of the higher carbon footprint of solar PV compared to wind power.For the island of San Pietro, Scenario 4 is almost certainly the preferred scenario within most ranges of weightings that do not place extreme emphasis on the CO 2 criterion.
In the scenarios examined for Gotland only Scenario 2, which aims for full electrical autonomy, includes batteries.All scenarios have a similar level of total RES capacity installed (wind or both wind and solar PV), though the island produces more per installed kW from wind than solar given its location.The ranking of scenarios using equal weights shows a slight preference for Scenario 1, the all wind scenario, though a clear preference is established for Scenario 1 when the energy and avoided CO 2 criteria are given priority.Conversely, when employment is given high priority Scenario 3 is top ranked due to the higher number of jobs expected from solar PV installations.Scenario 2 performs especially poorly when the avoided CO 2 and LCOE criteria are emphasised due to the high economic and carbon costs of the batteries it uses to reach autonomy.Scenario 1 is likely the preferred scenario for the island of Gotland within most ranges of weightings though a stronger preference towards job creation by local planners would increase the attractiveness of Scenario 3.
Unlike Gotland, all scenarios for Lesvos contained an equal amount of battery storage capacity.There is generally a clear preference for Scenario 2, with the highest total installed capacity of wind and solar PV and greatest production.Scenario 3 becomes first ranked when jobs are emphasised due to the greater job creation expected for solar PV parks.There is a very clear bottom ranking for Scenario 1, which had the Fig. 2. Area and population of the REACT islands.
A. Barney et al. lowest installed capacity and production, an indication that increased production to offset system costs, both environmental and economic, is key.Unless there is a strong local preference for job creation on Lesvos, Scenario 2 is likely preferred.
The two scenarios examined in the case of the Isle of Wight represent two different approaches, Scenario 1 is a fully autonomous energy scenario including significant solar and battery capacities, whereas Scenario 2 is a comparatively modest RES scenario with a limited amount of wind and solar PV for complementary support of the energy system.Scenario 1 is ranked highest except for when LCOE is given priority due to the increased cost for full energy autonomy, the result of the large battery capacity required.From these rankings it is apparent that for the Isle of Wight an autonomous scenario can be preferred so long as the main focus of planners is not on scenario economic cost.
Majorca's scenarios both incorporate a mixture of solar PV and wind power, with Scenario 1 slightly emphasising wind and Scenario 2 emphasising solar PV.Scenario 2 ′ s higher installed total capacity results in an approximately equivalent higher production and avoided CO 2 while the Scenario's greater emphasis on solar PV, with its higher job creation compared to wind, created a noticeable drop in Scenario 1 ′ s comparative performance when the jobs criterion was emphasised.Despite the relative similarity in scenario energy mixes Scenario 2 ′ s Table 1 Selected technical energy scenarios for the REACT islands. A. Barney et al. emphasis on solar PV is a better fit for the island for all the considered criteria.
For La Reunion the three scenarios examined consider the same combination of technologies, solar PV and batteries, with different scales of capacity installed -high, medium and low.Scenario 3, with the smallest capacities, is always bottom ranked except when LCOE is given priority.It appears that, similar to the case for the Isle of Wight, significant battery capacity can be preferred so long as the focus isn't on scenario costs.
Considering all scenarios for the islands examined it can be deduced that LCOE has a strong bearing on a scenario's overall ranking, as it is common that the rankings changed when LCOE was given priority.The   A. Barney et al. values of LCOE ultimately reflect the differing energy supply systems and their related levels of energy autonomy.The highest LCOE values in the scenarios examined are found for Gotland Scenario 2 and Isle of Wight Scenario 1 and reflect the significant cost for the full autonomy of those scenarios.More analysis of these LCOE values for the REACT islands can be found in [43].
As can be seen in Fig. below, the yields per inhabitant for Gotland's scenarios stick out, as do to a lesser extent the Aran Islands' and that of the Isle of Wight's Scenario 1.These differences are explained by that the Gotland, the Isle of Wight, Majorca and La Reunion's scenarios all having significant added production capacities but the populations of Gotland and the Isle of Wight are comparatively far smaller.Similarly,   A. Barney et al. while having a more modest added capacity, the Aran Islands also have a quite low population.
Fig. 11 below shows the avoided CO 2 per inhabitant for each island scenario over its life time.Gotland's autonomous energy generating scenarios, with and without batteries, show that they would create more carbon over a LCA basis than they avoid and they are discussed more below.The Aran Islands high performance is the result of a relatively modest capacity increase coupled with a low population.
The employment criterion is dependent on the type and size of systems installed.As general rule, the larger the system the more jobs created, but solar PV plants result in more local job creation than wind power and, as a result, similar yields between scenarios don't necessarily mean equal job creation.As can be seen in Fig. 12 below, the local employment criterion is impacted by seeking autonomy.
The preceding analysis provides some new insights into the development of energy systems on islands and, in particular, grid connected islands in countries with relatively low GHG emission intensity and for island's seeking electrical autonomy.Of the islands' countries in the review, Sweden has a significantly lower carbon intensity of electricity generation.This results in comparatively low avoided CO 2 emissions for the large amounts of electricity generated in the scenarios reviewed.In fact, the avoided CO 2 equivalents are so low that Gotland's autonomous Scenario 2, and its production equivalent without batteries Scenario 3, result in more carbon being generated than avoided.This is due in great part to the extreme quantity of batteries required and to a lesser degree on the relatively low carbon to yield effectiveness of the PV systems installed in the Scandinavia solar conditions.
A related insight can be drawn from the case of the other scenario seeking full electrical autonomy through renewable energy, the Isle of Wight's Scenario 1. Unlike Sweden, the country's emission intensity of electricity generation was high enough to more than offset the CO 2 equivalent cost of the scenario's batteries and the relatively lacklustre English solar conditions.Taken together, these scenarios indicate that, in countries with low carbon intensity of electricity generation, it may be environmentally damaging to seek full autonomy using batteries.A further point of consideration to be taken from the Isle of Wight's scenarios is that, when the cost criterion is not emphasised significantly more than the other criteria, an autonomous scenario can be an attractive option.
Overall, the integrated analysis platform presented in this paper provides a path for island planners to follow when considering new renewable energy projects.The different stages of the platform provide clear steps to be taken and how each step connects to the following.Once location specific scenarios fitting the needs of the planner have been generated, they are able to systematically evaluate the scenarios using the criteria most suited for their situations to ultimately determine the most appropriate scenario or portfolio of scenarios, as described above.The clarity of the steps in the process have the added benefit of allowing planners to more easily explain the processes they used when reaching their decisions.

Conclusions
In this work the REACT-DECARB integrated energy planning platform for the decarbonisation of the electricity systems of islands is fully presented and applied to eight disparate islands spread throughout the EU.Several energy scenarios on these islands were thoroughly examined and comparatively evaluated via an outranking MCDA method on a LCA basis.In this process, the most crucial sustainability criteria were established and the sensitivity of the results to each of these criteria was examined through the variation of their weights, providing valuable insights into the particulars of different islandic conditions and the impacts of decision maker preferences.The results allowed some clear and valuable conclusions to be drawn for future planning activities in regard to the impacts differing levels of autonomy can have on the sustainability criteria depending on both their cost and location.It was found that the amount of autonomy considered by each scenario had a substantial impact on the rankings, most evidently through the increased cost per unit of production.Additionally, it was determined that seeking   A. Barney et al. island electrical autonomy in countries with low carbon costs for electricity production may be not be environmentally beneficial.Moreover, if cost is not the highest priority then energy autonomy on islands can be attractive, particularly in countries with a high carbon cost of electricity production.The integrated analysis platform is shown to be able to effectively and clearly guide energy planners on islands towards decarbonisation of their energy systems but additional research where the tools and methods used by the platform are varied for different islands should be conducted to ensure its overall reliability and accuracy.

Fig. 7 .
Fig. 7. Isle of Wight.Ranking of scenarios for different criteria weights.

Fig. 11 .
Fig. 11.Avoided CO 2 equivalents per inhabitant on a LCA basis for the islands' scenarios.
et al.