A GIS-MCDA Approach Addressing Economic-Social-Environmental Concerns for Selecting the Most Suitable Compressed Air Energy Storage Reservoirs

: This article presents an assessment of the most suitable compressed air energy storage (CAES) reservoirs and facilities to better integrate renewable energy into the electricity grid. The novelty of this study resides in selecting the best CAES reservoir sites through the application of a multi-criteria decision aid (MCDA) tool, speciﬁcally the simple additive weighting (SAW) method. Besides using geographic information systems (GIS) spatial representation of potential reservoir areas, for the MCDA method, several spatial criteria, environmental and social constraints, and positive incentives (e.g., the proximity to existing power generation facilities of renewable energy sources) were contemplated. As a result, sixty-two alternatives or potential reservoir sites were identiﬁed, and thirteen criteria (seven constraints and six incentives) were considered. The ﬁnal stage of this study consisted of conducting a sensitivity analysis to determine the robustness of the solutions obtained and giving insights regarding each criterion’s inﬂuence on the reservoir sites selected. The three best suitable reservoir sites obtained were the Monte Real salt dome, Sines Massif, and the Campina de Cima—Loul é salt mine. The results show that this GIS-MCDA methodological framework, integrating spatial and non-spatial information, proved to provide a multidimensional view of the potential reservoir CAES systems incorporating both constraints and incentives.


Introduction
Portugal has one of the highest shares of renewable energy production within the European Union (EU), with more than half of the electricity consumed in 2019 coming from renewable energy sources (RES). RES was responsible for the production of 27.3 TWh, contributing to 56.10% of the electricity mix [1]. With the increasing use of intermittent RES and their integration into the national electricity system, challenges are being constantly brought into the grid, and solutions must mitigate intermittency and load variation. Energy storage (ES) is one of the most interesting options since it increases the flexibility of generating, delivering, and consuming electricity. In addition, ES provides the ability to balance power supply and demand, making power networks more resilient, efficient, and Then, the selection of the most suitable reservoirs for CAES was obtained by apply the SAW methodology to these sixty-two potential geological sites (Figure 1).

Methodology
The SAW method, also called the weighted linear combination (WLC) method, widely known and often used MCDA technique [25,27,33], integrating criteria values weights into a single framework [34] due to its reliability and proven results. The SA method is based on a weighted average, calculating a score for each alternative by mu plying the scaled value given to the alternative of that attribute by the weights of rela importance directly assigned by the decision-makers [25].
This method was chosen because it is reliable and has the advantage of allowin proportional linear transformation of raw data, meaning that the relative order of mag tude of standardized scores remains equal [25]. The chosen method is based on the MC method selection tool [35] developed by Wątróbski et al. [7].

Methodology
The SAW method, also called the weighted linear combination (WLC) method, is a widely known and often used MCDA technique [25,27,33], integrating criteria values and weights into a single framework [34] due to its reliability and proven results. The SAW method is based on a weighted average, calculating a score for each alternative by multiplying the scaled value given to the alternative of that attribute by the weights of relative importance directly assigned by the decision-makers [25].
This method was chosen because it is reliable and has the advantage of allowing a proportional linear transformation of raw data, meaning that the relative order of magnitude of standardized scores remains equal [25]. The chosen method is based on the MCDA method selection tool [35] developed by Wątróbski et al. [7]. Figure 2 illustrates the different phases of this MCDA method.

Problem Definition and Alternatives
The approach followed herein aimed to identify the best and most suitable potential reservoir sites for the possible installation of a CAES facility to better integrate RES into the Portuguese electricity grid. In this case, the generated alternatives are the sixty-two potential geological reservoirs depicted in Figure 1, according to CAES suitability analysis for Portugal based on the criteria established by [32]. These alternatives are listed in Tables in Appendix A, namely: twenty igneous host rocks (Table A1), nine deep mines (Table  A2), eighteen salt formations and nine salt caverns (Table A3), and six saline aquifers (Table A4).

Criteria Definition: Constraints and Factors
The second SAW phase selects and evaluates the criteria that directly influence the CAES facility site choice. In this study, thirteen criteria were adopted and subdivided into constraints and incentives. All the presented criteria are based on measures and legislation used for Portugal's natural gas (NG) storage safety [36]. Although compressed air does not have the same explosive potential as NG, assuming a conservative stance, it was decided to adopt the same criteria regarding distances to infrastructures since there is still subsidence risk due to potential underground caverns.
Constraints stand for the criteria that can limit or restrict the placement of a CAES reservoir at a particular location. For this study, seven constraints were identified (Table  1), overlaid individually with the identified reservoirs, and cross-checked with the defined criteria, resorting to basic GIS operations such as buffering and overlapping.

Constraints Description
Sensitive areas Environmental sensitive areas, including Natura 2000 areas, sites of community importance, and special protection areas. Groundwater Groundwater protection zones. Populated Areas Distance to populated areas of less than 200 m.

Roads
Distance to roadways or highways of less than 100 m. Land Slope Terrain slope of above 12%. Neotectonics Known active faults. Seismic risk High seismic risk.

Problem Definition and Alternatives
The approach followed herein aimed to identify the best and most suitable potential reservoir sites for the possible installation of a CAES facility to better integrate RES into the Portuguese electricity grid. In this case, the generated alternatives are the sixty-two potential geological reservoirs depicted in Figure 1, according to CAES suitability analysis for Portugal based on the criteria established by [32]. These alternatives are listed in Tables in Appendix A, namely: twenty igneous host rocks (Table A1), nine deep mines (Table A2), eighteen salt formations and nine salt caverns (Table A3), and six saline aquifers (Table A4).

Criteria Definition: Constraints and Factors
The second SAW phase selects and evaluates the criteria that directly influence the CAES facility site choice. In this study, thirteen criteria were adopted and subdivided into constraints and incentives. All the presented criteria are based on measures and legislation used for Portugal's natural gas (NG) storage safety [36]. Although compressed air does not have the same explosive potential as NG, assuming a conservative stance, it was decided to adopt the same criteria regarding distances to infrastructures since there is still subsidence risk due to potential underground caverns.
Constraints stand for the criteria that can limit or restrict the placement of a CAES reservoir at a particular location. For this study, seven constraints were identified (Table 1), overlaid individually with the identified reservoirs, and cross-checked with the defined criteria, resorting to basic GIS operations such as buffering and overlapping.
Incentives are the criteria that may be beneficial to the implementation of a CAES reservoir and facility. In this research, six incentives were identified (Table 2) and overlaid with the sixty-two reservoirs.  The thirteen criteria were divided by decision-makers into three classes (Table 3): (a) environmental, (b) social, and (c) economic. Constraints are non-beneficial criteria to be minimized, while incentives are beneficial criteria to be maximized, as depicted in Table 4. Although SAW may be used if all the criteria are being maximized [34], there are ways of converting minimizing into maximizing criteria, just by using a simple inversion of the scale for the minimizing criteria, as explained in the following sub-section. On the one hand, a CAES facility should be as far away as possible from sensitive areas, such as ecological and agricultural value, like special protection areas, Natura 2000 areas, and sites of community importance, to protect the environment and reduce any risk. On the other hand, proximity to energy sources (RES, HV networks, or even NG networks), proximity to roads, and land slope are important factors when considering the economic feasibility of any candidate site. Last but not least, social factors such as distance to populated areas should also be considered since a CAES plant can impact the population living within proximity to the chosen site due to noise, safety, or even a decrease in property value.
Some incentives are related to the proximity to energy sources. RES are used to store energy provided by renewable sources; transmission grid high-voltage (HV) networks are used for transmission and distribution purposes; and NG networks are used since natural gas is usually the fossil fuel used in the diabatic CAES expansion phase [30]. Other incentives are the availability of deep geological data, proven caverns for storage, and the technology's maturity depending on the type of geological reservoir. Table 4

Normalization Process
The next step is the normalization process since some criteria are qualitative, and others are quantitative. Normalization in MCDA is a transformation process to obtain numerical and comparable input data using a common scale [37]. Normalization (or transformation) of the initial data is generally used so that the best criterion value (the largest one for a maximizing criterion and the smallest one for a minimizing criterion) would obtain the largest value equal to unity [34]. There are several normalization methods, but given the subjectivity of the qualitative criteria, a simplification was done by the experts using a rating scale and attributing values. The chosen rating scale is comparable for all criteria and sets in the interval (0, 1) with intervals of 0.25, and a linear normalization method was used, where: (a) For non-beneficial criteria or constraints Constraints were normalized and transformed into maximizing criteria by inverting their scale through Equation (1). Hence, constraints were rated from 0 to 1 with intervals of 0.25, where 0 means the most favorable situation, and 1 depicts the most unfavorable situation. However, the rating scale was inverted, and 0 became the most unfavorable situation and 1 the most favorable (Table 5). Incentives (already maximizing criteria) were also rated from 0 to 1 (with intervals of 0.25) and normalized according to Equation (2), where 0 means the most unfavorable situation, and 1 represents the most favorable situation (Table 5). Table 5. Normalized rating scale (0,1) attributed to all the criteria (constraints and incentives).

J1
Absence of constraint Presence of constraint not limiting more than 25% area.
Presence of constraint not limiting more than 50% area.
Presence of constraint not limiting more than 75% area.
Presence of constraint limiting the area.

J2
Absence of constraint Presence of constraint not limiting more than 25% area.
Presence of constraint not limiting more than 50% area.
Presence of constraint not limiting more than 75% area.
Presence of constraint limiting the area.

J3
Absence of constraint Presence of constraint not limiting more than 25% area.
Presence of constraint not limiting more than 50% area.
Presence of constraint not limiting more than 75% area.
Presence of constraint limiting the area.

J5
Absence of constraint Presence of constraint not limiting more than 25% area.
Presence of constraint not limiting more than 50% area.
Presence of constraint not limiting more than 75% area.
Presence of constraint limiting the area.

J7
Roads not present Roads not crossing more than 25% of the area.
Roads not crossing more than 50% of the area.
Roads not crossing more than 50% of the area.
Roads crossing and limiting the use of the area.

J8
Presence of RES Proximity of RES of less than 5 km.
Proximity of RES of approximately 5 km.
Proximity of RES of more than 5 km. Absence of RES.

J9
Presence of HV network Proximity of HV network of less than 5 km.
Proximity of HV network of approximately 5 km.
Proximity of HV network of more than 5 km. Absence of HV network.

J10
Presence of HG network Proximity of NG network of less than 5 km.
Proximity of NG network of approximately 5 km.
Proximity of NG network of more than 5 km. Absence of NG network.

J11
Availability of deep geological data Availability of 75% deep geological data but without enough data.
Availability of 50% deep geological data but without enough data.
Availability of 25% of deep geological data but without enough data.
Absence of deep geological data. Despite the equal rating scale, there is always some arbitrariness in this conversion and normalization process. It depends on the analysis of the overlaying layers of reservoirs; each of the criteria; and the scale that GIS maps are analyzed with.

Assigning Weights to the Criteria
An essential step of the methodology is the assignment of weights to the criteria. A weight can be defined as a value assigned to an evaluation criterion that indicates its importance relative to other criteria under consideration [8]. Such assigned weights are based on experts' judgments and should provide a general priority set to evaluate and compare the alternatives.
Two research team members, experts on underground energy storage, were responsible for this decision-making process. First, the experts (i.e., decision-makers) individually assigned the weights according to their experience to identify Portugal's most suitable CAES sites. This methodology considered all the environmental, social, and economic criteria (Table 3) and weighted together all the constraints and incentives. Then, the two experts were engaged in a discussion to reach a consensus and assign the weights in Table 6. Table 6. Weights assigned to the criteria (constraints and incentives) for CAES potential reservoirs.

Constraints & Incentives Weights (%) J1
Sensitive areas 10% J2 Groundwater Renewable energy sources (RES) 12.5% J9 High-voltage (HV) network 12.5% J10 Natural gas (NG) network 5% J11 Deep geological data 7.5% J12 Maturity of technology 7.5% J13 Existence of proven caverns 5% Total 100% The weights of constraints and incentives (Table 6) were attributed according to the level of importance, limitation, or motivation for the CAES purposes that each criterion can impose on an area.
Environmental criteria such as sensitive areas and groundwater constraints have higher weights since they can completely limit a potential site if they are overlapped with the potential reservoir. Sensitive areas are fundamental constraints in respecting environmental, conservative, and protectionist policies (flora, fauna, heritage, and natural reserves). Groundwater resources are also a significant constraint since underground reservoirs should be placed in areas with the minimum risk of contamination for groundwater, including natural springs and geothermal resources.
The land slope is important because slopes greater than 12% can increase the instability for surface CAES facilities, and their correction can also increase the project's capital costs. So, areas with slopes from 0% to 12% are the most suitable for a CAES plant due to lower economic costs and minimum morphological problems.
Portugal is a country with significant seismic risk due to its location near the boundaries of the European and African tectonic plates. Thus, the seismic risk may be an essential constraint for selecting CAES potential reservoirs where the risk is lower in the north of the country and higher in the south (according to Portugal's seismic risk map).
Lastly, constraints such as neotectonic structures, populated areas, and roads should also be considered. However, their attributed weights are lower since they are not disabling factors. According to Costa [36], for safety reasons (mainly subsidence risk), the distance between populated areas and CAES facilities should be at least 200 m, and the distance between roads or highways should be at least 100 m. So, a buffer was used in ArcGIS to determine the safety area around these constraints and visualize the site free of constraint.
RES and HV have higher weights because they are the most important energy sources for a large-scale CAES facility. However, HV networks have a slightly bigger weight than RES because HV lines can work as sources supplying energy from the grid to feed the CAES plant in periods of electricity shortage from RES or high energy demand.
NG has a lower weight than the previous two since NG pipelines proximity only matters for diabatic CAES facilities, which need fossil fuels for the expansion phase. Although the only two CAES facilities in the world are diabatic systems (Huntorf-Germany, and McIntosh-USA) [4,6], this criterion is not disabling because it is possible to build a more efficient system with Adiabatic CAES technology.
Deep geological data and technology maturity have similar weights to those assigned to NG networks. They are important factors to consider since they both can increase the capital costs of a CAES project. Deep geological data are scarce in Portugal, and acquiring such information is extremely expensive, meaning that potential areas with deep data are favored. CAES technology maturity depends on the type of geological reservoir. For instance, salt is the lithology where CAES technology is already proven and mature.
Lastly, proven caverns for storage have the lowest weight of all the incentives, demonstrating the area can support that type of underground caverns and decreasing the initial cost of a project if those caverns could be reused.

Obtaining SAW Results
SAW results were obtained by analyzing local conditions of the different criteria at the alternative locations in the GIS database and applying Equation (3) to each alternative and each criterion (constraints and incentives individually): where a i is the alternative, S(a i ) is the suitability level of alternative i or the result of the weighted sum for alternative a i , w j is the weight of criterion j, and v j is the value of alternative a i in criterion j. Therefore, the last steps of this MCDA methodology consist of sorting and applying the evaluation method and selecting the best alternatives after classifying and ordering them. Thus, Equation (3) was applied directly to all of the criteria. Therefore, the higher the total score, the better the alternative for CAES purposes, meaning the highest results obtained indicate the best alternatives and chose the best potential CAES reservoir sites in mainland Portugal.

Sensibility Analysis
Saltelli et al. [38,39] state that sensibility analysis aims to ascertain how much the uncertainty in input factors influences the uncertainty in a model's output. So, MCDA methods usually resort to sensibility analysis as the last step of evaluation in all decision problems [22] because the majority of data in MCDA problems are unstable and changeable [40], and model outcomes are open to multiple types of uncertainty [41]. That is why doing a sensibility analysis after problem-solving can effectively contribute to make robust decisions [42]. A "what if" sensibility analysis is recommended to check the stability of results against the subjectivity of the experts [11], explaining how much the decisionmakers judgements bias the assessment of an MCDA study [43]. The sensibility analysis helps in the validation of results and enables assessing its robustness [44]. The aim is to ensure that results are more reliable and to identify the criteria that can significantly influence them [22].
The most common sensibility analysis method for MDCA is to modify the weighting obtained from the experts' judgment [11,27,42]. Thus, in this study, sensibility analysis was done using an approach based on Memariani et al. [42], where the effect of change in the weight of one attribute or criteria on the weight of other criteria was evaluated and the change in the final score of alternatives when a change occurred in the weight of criteria was calculated.
Within the scope of this work, two different sensibility analyses were developed to ensure that the results of the SAW method were robust. The first was based on the variation of the weights of two defined main clusters of criteria: constraints and incentives. The second one was developed with four new criteria sub-clusters. Then, the results obtained in both sensibility analyses were evaluated and compared with the original SAW results.
The first step of sensibility analysis is to determine the assumptions for the changes in criteria weights. After that, the computation must be executed, and the results are checked and compared.
For the first sensibility analysis, a uniform distribution of weights was used with variations of 5%. Since thirteen criteria were distributed in two main clusters (constraints and incentives), the variation of weights was done by cluster. It starts and ends with extreme cases, such as 100% weight for constraints and 0% weight for incentives, applying variations of 5% until the opposite percentage of 0% weight for constraints and 100% weight for incentives were reached (Table A5, in Appendix A). The criteria variations' computation was executed in Excel for each of the percentages, evaluating the change in the final score of alternatives (in light of criteria weight changes) and observing the influence of the weights' variation on the results.
For this step, the weighted linear summation represented by Equation (3) was used. As a matter of sensibility analysis comparison, the previous clusters were subdivided into sub-clusters. Constraints were divided into (a) surface and (b) sub-surface constraints. Incentives were divided into (c) energy sources and (d) technology/reservoirs maturity and data. Then, a new sensibility analysis was executed with weight variations of 0%, 25%, 50%, 75%, and 100% distributed by the new sub-clusters, according to the assumptions depicted in Table A3 (Tables A6 and A7, in Appendix A).
All the sensibility analyses results were cross-checked with the obtained SAW results, and the changes in the final score of alternatives were observed.

Results of the MCDA
In this MCDA-SAW method, the results obtained did not rely only on selecting one alternative, usually classified as the best. However, since choosing the best case studies for CAES was desired, it was decided to select several best alternatives.
The ranking of the best ten results is depicted in Table 7. The complete final results are represented in Table A8 (Appendix A) with a color gradation from green to red (from the best to the less good). Table 7. MCDA-SAW final results, ranking the best ten alternatives and identifying them by their reservoir name, set, and type of reservoirs. The columns "score" and "ranking have a greenish color gradation from darkest greens to lighter tones representing the decreasing gradation of the alternatives scores and ranking. The blue colors in the column "set of reservoirs" represent the gradation of each set of reservoirs according to their ranking since several alternatives can correspond to the same set of reservoirs.  The chosen final results are the three best sets of potential reservoirs for CAES in Portugal (Table 7), depicted in Figure 1, and also in a higher detail, from north to south in   Other alternatives or potential CAES reservoirs with good SAW scores and a high potential for CAES could be considered: the Matacães salt mine and salt dome, or the Pinhal Novo, Loulé, and Bolhos salt domes. However, the Matacães salt mine is abandoned and has severe stability issues (according to Solvay Portugal), and the other mentioned salt domes lack deep geological data that are very sparse or inexistent.

Ranking of the Best Ten Alternatives
The chosen alternatives for CAES potential reservoirs are generally located in the western and southern part of the country (Figures 3-5), having the most favorable locations with fewer constraints and more incentives.         (Figure 3), held by REN Armazenagem in Carriço (Pombal). On the one hand, these salt caverns are being used to store NG in that geological formation, meaning that the Monte Real salt dome has already proved its suitability for storing energy underground. On the other hand, REN storage facilities have infrastructure like HV lines and NG networks available on-site, decreasing the costs of a possible CAES project. Thus, joining the absence of limiting constraints, deep geological data availability, and the proximity to the sea, Monte Real/Carriço would be a great suitable location for settling new salt caverns to a CAES system in Portugal.
Alternative 19 corresponds to the Sines LPG reservoir, an engineered cavern to store LPG built in Sines' sub-volcanic massif (Figure 4). This potential underground reservoir has deep geological data and a proven storage capacity, both a plus when considering a CAES geological reservoir. It is located in the coastal line and has special wind conditions for installing wind parks. Sines is one of the most important Portuguese seaports and is the country's principal port of energy supply (oil and by-products, coal, and natural gas) [45]. So, it already has energy surface infrastructures such as HV lines and gas pipelines (necessary in case of potential diabatic CAES facilities), and it still has the potential to grow.
Alternative 29 corresponds to the Campina de Cima-Loulé salt mine ( Figure 5). This mine is settled in Loulé diapir and has several salt excavated galleries, which could host CAES underground reservoirs, decreasing the initial costs of a possible CAES project.

Results of the Sensitivity Analysis
The sensibility analysis provides information about the influence that criteria and clusters may have on the final score of alternatives and how the variation in weights of criteria may change the final results in terms of the chosen reservoirs for CAES purposes, contributing to accurate decisions.
The first sensibility analysis was done considering cluster weights with variations of 5%, analyzing twenty-one scenarios. The summary depicting only the main results (with intervals of 25%) is shown in Table A9 (Appendix A).
The results comparison did not show significant differences between scenarios, even in the extreme and improbable ones where 100% of the weight was attributed to one cluster. Thus, the possible reservoirs with the best scores remain the same throughout the various analyses: (a) the Monte Real/Carriço NG storages and salt dome, (b) the LPG Sines in the Sines Sub-Volcanic Complex, and (c) the Campina de Cima-Loulé salt mine.
The second sensibility analysis dividing each main cluster into sub-clusters evaluated seven scenarios (even the most extreme and improbable ones) to determine which subcluster had the most influence on the results. Those results are shown in Table A10 (in Appendix A). The results of scenarios one to three did not significantly change the previous GIS-SAW results. Thus, the case studies selected for CAES purposes were the same as before. However, this selection varied when extreme cases were contemplated. The best results for scenario four (placing 100% of the weight on the sub-cluster of surface constraints) were Sines LPG, Ervideira, the Loulé salt mine, the Carriço salt caverns, and the S. Pedro de Moel and Várzea da Rainha salt domes. Scenario five's (with 100% of the weight on the sub-cluster of subsurface constraints) best results were four host rocks (Vila Verde de Raia, Vila Nova de Covelo, Celorico da Beira, and Capinha) and five deep mines (Jales, Borralha, Pejão-Germunde, S. Pedro da Cova, and Panasqueira) followed by Soure, Ervideira, the S. Pedro de Moel saline domes, the Carriço salt caverns, and also the Lusitanian On_A3 aquifer. Scenario six's (with 100% of the weight in the subcluster of energy sources) best results were the Lusitanian On_J1 and Lusitanian On_A1 saline aquifers. Finally, scenario seven's (placing 100% of the weight in the sub-cluster of technology/reservoir maturity and data) best results were the Carriço salt caverns, the Monte Real and Matacães salt domes, and the Loulé salt mine.
Both sensibility analyses were done with different weights for clusters, sub-clusters, and criteria. In the first SA, there were minor variations in potential reservoirs with better scores. Still, there were no significant changes in the results with the highest scores, which gives robustness to the initial combination of GIS and SAW results and suggests they are correct. It also indicates that weight variation influence was not significant and did not drastically alter the outcome of the chosen case studies. Despite the first three scenarios maintaining the same highest score reservoirs in the second sensibility analysis, the last four scenarios changed the highest-score potential reservoirs. However, those four scenarios were based on extreme, unlikely, and unreal assumptions, where the entire weight of the criteria was placed only in one cluster or sub-cluster. They serve to understand the types of criteria that value certain reservoirs at the expense of others and the possible influence that each sub-cluster may have on the final decision of case studies for CAES.
Therefore, according to the analysis carried out through GIS-MDCA and corroborated by the sensibility analysis, the criteria that seem to have greater weight and influence in the three chosen case studies were: (a) For Monte Real/Carriço, the maturity and data availability on the reservoirs were predominant factors, but other criteria such as a lower absence of constraints and proximity of energy sources were also important; (b) For LPG-Sines, the lower absence of surface constraints; (c) For Campina de Cima-Loulé, the less lower of constraints and the reservoir's maturity and data availability.
However, it is mandatory to mention that these choices result from evaluating all the criteria, clusters, and sub-clusters together since, in reality, they are essential and take a significant part in the final decision.

Conclusions
The grouping of GIS-MCDA and sensibility analysis methods is a powerful tool for selecting sites for different installations, representing a promising research line in largescale ES, especially for selecting the best location of facilities.
This study is not comparable to others since the combined techniques of GIS and MCDA were never used, as far as we know, to select the most suitable CAES potential reservoirs in Portugal. Thus, it represents an innovation since, apart from the ESTMAP European project (which had a different scope), no exclusively CAES studies in Portugal could select and determine the three best reservoir case studies to store the excess RES.
Some uncertainties can be held since this method yields a certain degree of arbitrariness, where the most significant one is the decision-makers' subjective choices. Specifically, the criteria evaluation, the process of normalization, or attributing weights to the criteria are subjective, having a considerable effect on the entire evaluation process. However, most of the selection processes commonly used in the literature also present arbitrariness and are mainly dependent on the decision-makers' choices, turning them subjective. Thus, this well-known MCDA-SAW method was chosen since it can be straightforward and efficient to serve the defined purpose and provide the expected results.
In total, for sixty-two potential reservoirs for CAES represented in a GIS environment, thirteen criteria (seven constraints and six incentives) were identified. First, criteria and potential reservoir sites (the alternatives) were cross-checked using GIS techniques and the MCDA-SAW method, and the best results were chosen. Then, two sensibility analyses were conducted to check the robustness of previous results.
The most suitable reservoir sites for a possible CAES facility were Monte Real-Carriço Sines LPG and Campina de Cima-Loulé. The Monte Real salt dome holds NG reserves for the country in REN Armazenagem salt caverns, and Sines has an LPG engineered cavern. So, these two suitable sites have the advantage of being already proven capacity. Furthermore, Campina de Cima in Loulé salt dome is an out-of-labor salt mine with several salt galleries that could be reused for storage. Thus, these three sites have the highest potential and best location for a CAES system regarding lower constraints and proximity/overlapping positive incentives.
These results are important for the Portuguese electricity grid because they show the best potential CAES sites for large-scale ES of RES, adding flexibility to the grid and an alternative to the country's weather and topography-dependent PHES.
The results also show that this GIS-based and MCDA-SAW method integrating spatial and non-spatial information provided a multidimensional view of the potential reservoir CAES systems.
Techno-economic studies need to be done for further work, including more detailed studies about these three selected reservoirs. acknowledges that this work has been partially supported by FCT project grant: UID/MULTI/00308/ 2020 and the Energy for Sustainability Initiative of the University of Coimbra. The authors Catarina R. Matos and Júlio F. Carneiro acknowledge that this work has been partially supported by the Institute of Earth Sciences (ICT), under contract with FCT (The Portuguese Foundation for Science and Technology), with projects UID/GEO/04683/2019 and POCI/01/0145/FEDER/007690, funded by Portugal 2020 through the Operational Programme for Competitiveness Factors (COMPETE2020).

Conflicts of Interest:
The authors declare no conflict of interest.       (3) to constraints and incentives and the final score results with a color gradation (from green until red) and the chosen case studies highlighted in dark green.