Optimal site selection for floating photovoltaic systems based on Geographic Information Systems (GIS) and Multi-Criteria Decision Analysis (MCDA): a case study

ABSTRACT

Solar energy is growing rapidly thanks to reduced implementation costs; European development plans foresee a significant increase in installed capacity in the coming years. In this scenario, new identifiable areas on water surfaces are needed. This would lead to the development of floating photovoltaic (FPV) systems, which would mitigate the environmental impacts of terrestrial installations. Therefore, it is necessary to develop and propose a methodology to identify suitable sites for the installation of such plants. This study is useful to demonstrate the potential benefits that can be achieved using the methods of GIS for the optimal siting of FPVs in combination with the use of MCDA in areas such as Sicily, which are characterised by high summer temperatures. Seven watersheds were studied and the best site for FPV allocation selected using MCDA is San Giovanni Dam.

KEYWORDS:

• Renewable energy
• Multi-Criteria Decision Analysis
• floating photovoltaic system
• resource maximisation
• optimal site location
• GIS

1. Introduction

The world continues to energise, driven by the developing economy, at the same time there is an increasing need to produce new clean energy. Therefore, renewables play a crucial role in this continuous-changing scenario. Prospects for the coming decades explain that by 2040 renewable energies will overcome coal as the first global energy source (BP Energy Outlook: 2019 Edition 2019).

According to Italian energy policies, the electricity produced from renewable sources should reach approximately 132 TWh by 2030, thus covering 38.7% of the total electricity produced, against 34.1% of 2017. Focusing on the single sources, the significant residual potential technically and economically exploitable and the reduction in the costs of photovoltaic and wind power plants, for these technologies, also envisage growth in current policies. Still in the same time horizon, a limited growth in additional geothermal and hydroelectric power and a slight decline in bioenergy are considered. Looking ahead to 2040, the electricity from renewable sources will grow to 40.6% (Ministry of Economic Development, Ministry of the Enviroment and Protection of Natural Resources and the Sea, Ministry of Infrastructure and Transport. Integrated National Energy and Climate Plan 2019) (Table 1).

Table 1. Present and future European Renewable production.

	2020	2025	2030	2040
Renewable production (TWh)	118.5	120.5	132	142.9
Hydro (normalised) (TWh)	49.4	49.1	51	51.6
Wind (normalised) (TWh)	20.1	21.8	25.1	33.2
Geothermal (TWh)	6.7	6.9	7	8.3
Bioenergies (TWh)	16.3	14.7	14.2	12.3
Solar (TWh)	26	28	34.6	37.4
Denominator – Gross inland electricity consumption (TWh)	327.1	333.1	340.6	351.7
RES-E share (%)	36.30%	36.20%	38.70%	40.60%

The recent COVID-19 pandemic had a serious impact on global economy. The overall electric power and gas prices raised. This is a reminder that modern life needs abundant energy: without it, bill become unaffordable, as a consequence many communities that are still dependent on these technologies can suffer from these increases and businesses stall (CNBC 2021).

In Nižetić (2020), an analysis was conducted during the first four months of the pandemic COVID-19, focusing on the impact of the pandemic on mobility and thus on carbon dioxide emissions. Two airports in Croatia were considered as study cases and the results showed a decrease in flights, reaching a low point in April 2020 with 89% fewer flights in Europe. The pandemic had a strong impact on the national energy system and laid the foundation for the energy transition towards renewable and clean energy (Hoang et al. 2021).

Renewable energy systems such as photovoltaics (PV) can be used to achieve an energy transition: They allow urban and rural areas not to depend on traditional technologies, to achieve primary energy savings or even to sell electricity to the grid. Years ago, some studies (Gadsden et al. 2003) already examined the potential benefits of integrating photovoltaics, or PV, in urban areas and concluded that PV systems can be a powerful tool to achieve energy savings and reduce dependence on fossil fuels. Moreover, PV is competitive with other renewable energy sources. It is estimated (Handayani, Krozer, and Filatova 2019) that by 2050 it will be comparable to both other renewable energy sources and nuclear energy.

Floating photovoltaic panels are an interesting alternative to traditional ground-based photovoltaic panels, as water-cooling achieves higher efficiency (Sukarso and Kim 2020), which leads to a reduction in electricity costs.

Another study (Pimental Da Silva and Castelo Branco 2018) analysed a new type of PV technology that can be installed in rural areas, floating photovoltaic (FPV) systems, and concluded that these systems can generate much more electricity compared to traditional ground-based PV and are a useful tool for coupling with agriculture.

However, the installation of FPV systems requires many construction steps and challenges, as outlined in Pimental Da Silva and Castelo Branco (2018), so available sites for the installation of these systems need to be carefully selected. Various studies such as Kapoor and Garg (2021) suggest a geospatial approach based on GIS technology, not only to locate the available sites, but also to assess the average electrical power generated by the FPV modules by calculating the solar radiation, which can vary significantly in different geographical locations.

However, other studies (Nebey, Taye, and Workineh 2020; Izeiroski et al. 2018) pointed out that locating the available areas for the installation of these technologies can be a very difficult task due to the many requirements. Based on technologies that use Geographic Information Systems (GIS), they have proposed an analysis method to establish the sites available for the installation of floating photovoltaic systems: the required needs are entered into this software as input matrices and the tools of GIS output the geographical coordinates of the areas where the required needs are met.

It is necessary to find usable areas where there are no competing uses, the basins must in any case be suitable for providing floating power plants with solar capture. This is only possible if, thanks to the multi-criteria analysis, ZPS and SIC areas, basins where anthropic activities are carried out, areas where controlled recolonisation of flora and fauna takes place and unproductive areas that are not suitable for agricultural use are excluded from the outset.

This type of approach, based on GIS, has been applied in many studies (Sukarso and Kim 2020; Ayompe and Duffy 2014; Amjad and Shah 2020; Ates, Yilmaz, and Gulgen 2020), not only to locate suitable areas for the installation of FPV systems, but also to evaluate the energetic performance of these systems in different developing zones of the world, such as Cameroon, Indonesia, Pakistan or Turkey: they concluded that FPVS can be a very powerful tool for rural economies and community energy supply, no longer dependent on traditional oil and gas power plants.

Today, the global economy is creating new challenges for many local communities and has led to the introduction of a new circular economy. Renewable energy sources can be a very powerful tool to achieve this goal. The integration of such systems on Sicilian territory could be a great improvement for the economic development of the region.

For a local economy like the one in Sicily, the introduction of photovoltaic technology is a great improvement thanks to the high annual solar radiation. For example, electricity consumption for irrigation can be covered with electricity generated from this system, and the surplus electricity can be sold to the national grid: Residents are then no longer dependent on electricity from petroleum-based gas and steam turbine power plants, resulting in higher incomes and lower global environmental impacts.

This study follows a GIS methodological approach reinforced with the MCDA plugin on the Sicilian territory in order to locate the best performing sites where FPV equipment can be installed. The use of the software GIS makes it possible to quickly filter out basins that cannot be used for normative or environmental reasons. The MCDA, carried out using two methods, makes it possible to study the same issue and evaluate the results. Particular attention is paid on Levelised Cost of Electricity (LCOE): the criteria take into account the annual energetic profit, the CAPEX and OPEX, and the life cycle of the installation. These criteria cover some of the most important topics of current scientific research. A sensitive analysis must be carried out to evaluate the soundness of the results by the two different MCDA methods. Our study can be considered a novelty in the current scientific landscape, as to our knowledge there are no studies that use the GIS methodology together with MCDA to attribute energy harvesting systems from floating photovoltaic plants.

2. Background

The principle theoretical aspects of photovoltaic technology, deployments of FPVSs and the characterisation of MCDA are showed in this section.

2.1. Photovoltaic solar collection systems overview

Photovoltaic renewable energy sources, which are among the most expanding renewable energy technologies (31% in 2018) (Tina, Scavo, and Gagliano 2020), are based on PV cells that essentially convert solar radiation into electricity (Tyagi et al. 2013). This type of technology has a higher energy density than other renewable energy sources, but lower than traditional ones (Van Zalk and Behrens 2018). PV is a well-established and defined technology, both from technical and economic aspects. There are many PV technology categories that are distinguished by the efficiency and cost of PV systems (Honrubia-Escribano et al. 2018). Depending on the material used, photovoltaic modules are categorised into crystalline, thin film, compound semiconductor and nanotechnologies. In 2018, technologies using crystalline silicon accounted for 90% of the photovoltaic module market (Honrubia-Escribano et al. 2018).

Since 1996, multicrystalline silicon has led the market and is the most widely used crystalline silicon (c-Si) technology (Close EurObservER 2016). The other technology types are still in the test phase (Honrubia-Escribano et al. 2018). The efficiency of crystalline silicon solar cells has improved; initially it was around 15% and then increased to 17%, reaching 20% in some cases (Erraissi et al. 2017).

The efficiency of a photovoltaic system can be reduced, differently from what was studied during the test phases, by various parameters (Yushchenko et al. 2018). The high temperature, during the operating phase, or the percentage of humidity in the air, as well as low solar radiation and the percentage of pollution in the air (Tyagi et al. 2013; Yushchenko et al. 2018). Another parameter to be taken into account when evaluating the potential of a photovoltaic system is the relationship between the theoretical efficiency/yield of a system and the actual efficiency/yield of the same (Yushchenko et al. 2018). This is called TREr or performance ratio and, for monocrystalline and polycrystalline silicon panels, it is between 70 and 85% (Sun et al. 2013). Of course, the relationship between theoretical and actual efficiency is highly dependent on site conditions, shading, panel cleaning conditions, mismatch and conversion losses, and values vary by site.

PV power plants are integrated into grid-connected or off-grid electrical systems (Parida, Iniyan, and Goic 2011). On the other hand, off-grid systems do not feed electricity into the grid and can be used by one or more consumers for self-consumption (Devabhaktuni et al. 2013).

2.2. Floating photovoltaic energy plants

Compared to other systems that generate electricity from renewable sources, such as wind, ground photovoltaics or hydroelectric, floating photovoltaics plays an important role in the world market due to its advantages (Ranjbaran et al. 2019); it is a new technology that is in constant growth, exploitation in the renewable energy market took place in 2016 (Gorjian et al. 2021).

These can be used in various places such as reservoirs of dams or hydroelectric power stations and on the surface of lakes and rivers. The solar energy captured by a photovoltaic panel that is transformed into electricity may vary according to the climatic conditions, the efficiency of the modules themselves and the time of year; usually this is a percentage that varies from 4% to 18% of solar energy. What is not converted can cause the temperature of the module to rise as it is converted into heat (Dubey, Sarvaiya, and Seshadri 2013; Azmi et al. 2013). As the temperature changes, the yield of the solar cells varies, thus varying the efficiency of the cells of the photovoltaic panels; this factor can be reduced if the systems are installed on a water surface thanks to the refreshing effect that the same can generate (Sukarso and Kim 2020; Sahu, Yadav, and Sudhakar 2016; El Hammoumi et al. 2021; Liu et al. 2017; Kjeldstad et al. 2021). Floating plants require a careful study of the site and of the design in order to make them feasible; this is necessary because the floating modules must be designed to be compatible with the geometries of the basins on which they are installed (Ferrer-Gisbert et al. 2013). The latest developments are stable, modular and scalable and designed to last at least 25 years (Transforming Unused Bodies of Water into Clean Energy Generators 2021). In Friel et al. (2019), an overview of the design, technical performance and feasibility of the structures for the modules of floating photovoltaic systems was given. The structural base is firmly anchored to the free surface of the water reservoir (Gorjian et al. 2021), which is not subjected to strong wave forces, and consists of a combination of multi-layered floating frame made of medium or high density polyethylene (M-HDPE), which ensures the stability and buoyancy of the network of units (Ferrer-Gisbert et al. 2013). The module is typically designed to accommodate standard solar panels, the access path behind the top of the panels, inverters, floating transformer stations and integrated DC cables certified for in-water installations (Transforming Unused Bodies of Water into Clean Energy Generators 2021). The structure can be anchored using different techniques depending on the characteristics of the basin (ground conditions, basin requirements and deviations from the water level). Anchoring concepts can choose between anchoring on land around the system, anchoring near the shore around the system, and anchoring on the ground below the system to achieve the best aesthetic integration with the landscape (Transforming Unused Bodies of Water into Clean Energy Generators 2021). To be able to withstand wind and wave loads, such as those acting on the edges, the modules positioned at the outer limit must be fixed to the basin by means of rigid anchors along the borders. The modules, positioned on the floating polyethylene monoliths and connected to each other by tensors and elastic fastening elements (bolt anchoring or metal tie rods), guarantee, thanks to the ability to adapt to varying water levels, good performance from the point of view structural. This minimises maintenance and maximises energy yield. Floating photovoltaic systems (FPVS), which have been widely reported in the literature, have several advantages over terrestrial applications: they reduce the effects of evaporation from reservoirs (Galdino and de Almeida Olivieri 2017), reduce algae growth (British Columbia Sustainable Energy Association 2021), reduce the effects of mutual shading of panels, and are also more efficient in terms of energy production than a ground-mounted system, aided by the natural cooling of the modules (Merlet 2018). On average, the solar panels used for floating photovoltaic systems have an efficiency of over 11% compared to the photovoltaic panels used for conventional systems (Ranjbaran et al. 2019). The methods studied to lower the temperature of the photovoltaic panels, so as to prevent their operation from being limited due to excessive irradiation (Ranjbaran et al. 2019), are various; these include the application of water (Grubišić, Nižetić, and Tina 2016; Tina et al. 2021).

2.2.1. FPV on closed basins

A possible location for floating photovoltaic systems is, as already mentioned, enclosed water basins. Photovoltaic systems in closed basins can reduce the evaporation of water in areas of the world where the climate is arid and in places such as India, Jordan, Australia, the United States and Spain where the evaporation of water has a high percentage and resources are poor (Padilha Campos Lopes et al. 2022). Numerous studies and experiments have been conducted on floating photovoltaics given the technical potential and the need to support the growing interest in energy from renewable sources. In Kim, Oh, and Park (2019), the distribution of FPV in 3401 catchments in Korea was evaluated assuming 10% water surface coverage. Using meteorological data, topographical information to assess irradiation and a database to study the depth of the basins, simulations were carried out which showed that the possible potential production would be 2932 GWh per year. A 10 kW floating photovoltaic system was installed in New Town in the West Bengal region of India. The plant is connected to the grid and 40 250 Wp polycrystalline modules have been installed, covering an area of 101.2 m². According to forecasts, the plant will have a life cycle of 25 years and generate at least 14 MWh per year (Floating solar power plant in West Bengal India 2017).

2.2.2. FPV coupled with HPP

Floating photovoltaic systems could generate electricity if coupled with hydropower plants. Hybrid systems are created by associating floating photovoltaic modules with already existing hydropower plants. This hybridisation optimises costs thanks to production guarantees that only take up a small part of the available area of the reservoirs (Perez et al. 2018). The use of hydropower plants in combination with floating photovoltaic systems has the advantage of reducing water evaporation, which increases the availability of water and reduces head losses. In addition, existing energy transmission infrastructures can be used (Perez et al. 2018; Rosa-Clot and Tina 2017). A case study was conducted in southern Brazil at a hydropower plant with a relatively large water area (Teixeira et al. 2015). Full coverage of this area would have required the use of a photovoltaic system with a capacity of just over 100 MW. Since the hydroelectric potential was much lower (about 60 kW), it was decided to use part of the area to obtain additional power similar to that of the hydroelectric plant. Hydroelectric plants thus have an unused potential that can be avoided by using floating photovoltaic modules, both operating in a hybrid hydroelectric-photovoltaic system (Teixeira et al. 2015). Another study was conducted in Rauf, Shuzub Gull, and Arshad (2019) for the allocation of floating photovoltaic panels in.

Ghazi Barotha Dam. The hydropower plant is capable of meeting 4 h of daily peak demand with typical storage. By covering 20% of the surface of the two basins of the plant with floating modules, it would be possible to produce 200 MW to meet the demand during the day with the FPV and at night with the HPP plant. The possible provision of a floating plant could cover the peak demand in the morning hours, which would lead to significant savings; these would be necessary for transmission and distribution in the case of new photovoltaic plants.

2.3. Multiple criteria decision analysis

Until the 1980s, the decision-making process commonly processed a single criterium approach to evaluate which were the best alternatives at the best possible cost. In the following years, it became clear that the environment plays an important role and this meant that the approach to decision-making processes changed. When it comes to sustainability, the decision-making process must take into account several factors, which are considered in the concept of triple bottom line (environmental, social and economic aspects). The correct way to make this happen is to use the MCDA because it evaluates multiple criteria that otherwise would not be directly comparable. The different methods can be classified according to three definitions: elementary methods, overcoming methods and methods with unique synthesis criteria. Elementary methods include the weighted sum method (WSM) and the weighted product method (WPM). Overcoming methods include ELECTRE (Elimination Et choice Translating Reality) and PROMETHEE (Method for Organising Ranking of Preferences for Enrichment). The AHP (Analytical Hierarchy Process), the TOPSIS (Technique for Order Preference by Similarity to Ideal Solutions) and the MAUT (Multi-Attribute Utility Theory) are the methods with unique synthesis criteria (Abraham, Biniyam, and Tewodros 2021).

Several authors (Abraham, Biniyam, and Tewodros 2021) applied Multi-Criteria Decision Analysis or MCDA methods to the renewable energy sector and pointed out the different types of criteria that influence the decision of which renewable energy system to install and where to install it (Shao et al. 2020). In the literature, to date, there are no other studies that examine, through the use of GIS software and MCDA methodologies, the position and allocation of floating photovoltaic systems. Only a few studies (Pimental Da Silva and Castelo Branco 2018) evaluated the positive impact that these plants have on local communities, both in terms of reduced costs for electricity consumption, as electricity is no longer imported from the grid but can even be exported to the grid, leading to higher economic profits, and in terms of new jobs related to the installation and maintenance of the plant.

In parallel, studies such as (Tercan, Mehmet, and Burak n.d.), have used GIS tools and the MCDA, through the AHP, to establish the best location for the design of a ground-based PV system, taking into account various criteria, which illustrate the great potential of GIS analysis.

This study adopts the following MCDA supported by the QGIS software:

1. Analytic Hierarchy Process or AHP

2. Technique for order preference by similarity to an ideal solution or TOPSIS

2.3.1. AHP

The Analytic Hierarchy Process (AHP) belongs to the MAUT (Multi Attribute Utility Theory) models; the main hypothesis in utility theory is that there is a real-valued function, called utility function, which associates a value (real number) that represents its degree of preferability to each eligible action. The preferences of the considered actions can be represented using this utility function: for each pair of actions the function of the ideal action is greater than the function of the other compared, while the function of two actions is equal if and only if they are indifferent.

From a formal point of view function U holds together the criteria belonging to set G, that is, $\mathrm{\forall }a\in$ A, with $a$ representing the single alternative belonging to set A (Di Grazia and Tina 2022): (1) $U\left(a\right)=V(g_1\left(a\right),g_2\left(a\right),\dots g_m\left(a\right))$(1) with V is a function of $m$ variables, incremental in its marginal utilities (all its arguments); that is, in other words, $\mathrm{\forall }$ a, b $\in$ A where a and b are the alternatives, (2) $g_{\left(a\right)}\ge g_{\left(b\right)},\mathrm{\forall }g\in G=>U_{\left(a\right)}\ge U_{\left(b\right)}$(2) One of the simplest forms of utility function U(a) is additive manifestation; in this the evaluation of an action is the sum of the products, between the weights of the criteria $w$ and the marginal utility $u$, of each alternative (Di Grazia and Tina 2022), that is: (3) $U\left(a\right)=\sum _{j=1}^m{w}_j{u}_j\left(g_j\left(a\right)\right)$(3)

where (Di Grazia and Tina 2022): $m$ is the number of criteria; $j$ is the number of the criterium; $w_j$ is the weight of the criterium; $u_j\left(g_j\left(a\right)\right)$ is the marginal utility of the criterium considered with respect to the alternative analysed.

To define the marginal utilities a method is used, which can be defined as the construction of a system of values, of direct comparison. In this way, the extremes of the system are defined. Considering a criterium $g_j$. The best performance compared to $g_j$ is assigned the highest score (utility) while the worst performance compared to $g_j$ is assigned the score equal to 0. All other performances are placed directly on the system to reflect their usefulness with respect to the two reference points (the one that has the best performance and the one that has the worst performance).

The AHP can be used, by comparing pairs of criteria or alternatives, to obtain a weighting of the criteria or to evaluate the performance of the alternatives with respect to the chosen criteria (Di Grazia and Tina 2022). A real value is attributed to each alternative contemplated through a weighted sum that represents the goodness of the alternative itself; very often it is considered a special case of the utility function. The real values of the criteria then make up the weighted sum (4) that expresses the function $U\left(a\right)$ (Di Grazia and Tina 2022): (4) $U\left(a\right)=WS\left(a\right)$(4) By filling in a matrix A (5) in which the elements $a_{ij}$ (where $i,j=$ $1,2,$ … , $m$ represent the number of criteria) stand for the preference entity of the criterium $g_i$ on criterium $g_j$ and, taking into account the preferences with respect to the pairs of criteria, the weight of the criteria can be determined. (5) $A=\left(\begin{array}{cccc}1& a_{ij}& \dots & a_{im}\\ 1/a_{ij}& 1& \dots & a_{jm}\\ ⋮& ⋮& \ddots & ⋮\\ 1/a_{im}& 1/a_{jm}& \dots & 1\end{array}\right)$(5)

Where $a_{ij}$ is the preferences of criterium $g_i$ over $g_j$; $a_{im}$ is the preferences of criterium $g_i$ over $g_m$; $a_{jm}$ is the preferences of criterium $g_j$ over $g_m$; the evaluations of the weights of criteria are expressed on a scale from 1 to 9, with the following interpretation: a_ij= 1 ⇔ $g_i$ has equal importance of the criterium $g_j$; a_ij= 3 ⇔ $g_i$ has a moderately higher importance than the criterium $g_j$; a_ij= 5 ⇔ $g_i$ has a much greater importance than the criterium $g_j$; a_ij= 7 ⇔ $g_i$ has a much higher importance than the criterium $g_j$; a_ij= 9 ⇔ $g_i$ has equal importance absolutely superior to that of the criterium $g_j$; The values 2, 4, 6, 8 are intermediate values used in case of indecision between the judgments just expressed. It arises that, for each $i,j=$ $1,2,$ … , $m$, a_ij =${\displaystyle \frac{1}{a_{ji}}}$ (that is the matrix is reciprocal) and a_ij = 1 for each $i,j=$ $1,2,$ … , $m$ (since each criterium is indifferent to itself).

Each value of the matrix A can therefore be interpreted as the ratio between the weight of two criteria $\left(g_{i,}{g}_j\right)$ that is a_ij =${\displaystyle \frac{w_i}{w_j}}$. If the information provided is consistent you will have to check that: (6) $a_{ij}=a_{im}\times a_{mj}\mathrm{\forall }i,j,m$(6) The matrix is defined as consistent, then check (6), if and only if λ_max = m. Where m is the order of the matrix and λ_max is the maximum eigenvalue of the same.

In order to draw up a ranking of the alternatives, the score, $S_a$, for each of them must be calculated with respect to the final goal established. This score is obtained using the formula (7) (Farhad, Bushby, and Williams 2019; Rios and Duarte 2021; Liu, Eckert, and Earl 2020): (7) $S_a=\sum _{j=1}^m{p}_a\left(j\right)\times wg\left(j\right)$(7) where $wg$ it is the main eigenvector of the matrix used for comparing the pairs for the objective with respect to the objective itself; (Di Grazia and Tina 2022); $p_a$ is the main eigenvector of the matrix used for the comparison of the pairs for each criterium (Di Grazia and Tina 2022).

2.3.2. TOPSIS

Once the best alternative has been determined, the TOPSIS method provides a useful approach in order to compare the chosen alternative with respect to an ideal solution: the comparison is made computing the Euclidean distance among the chosen alternative and the ideal solution, where the dimensions of the distance are the different criteria adopted for the site selection (Sánchez-Lozano et al. 2013).

The TOPSIS procedure consists of the following steps:

1. Calculate the normalised matrix;

2. Calculate the weighted normalised matrix;

3. Calculate the ideal best and ideal worst value;

4. Calculate the Euclidean distance from the ideal best (S_i⁺);

5. Calculate the Euclidean distance from the ideal worst (S_i⁻);

6. Calculate performance score;

7. Rank the preference order.

2.4. Sensitivity analysis

Very often it is assumed that the methods of Multi-Criteria Decision Analysis lack precision in their results. This consideration can very easily be applied to the assignment of weights to criteria. The uncertainty of the weightings lies in the planner’s or client’s subjective assessment of the importance of one criterium over another (Feizizadeh and Blaschke 2014). Sensitivity analysis is therefore an essential step to be carried out at the end of the analysis. It is a method used to verify the soundness of an assessment made; this is done by examining how and how much changes in methods, models, values or assumptions affected the final result (Luthra et al. 2016). This type of analysis consists of assessing the impact on the results provided by a model caused by changes in the values of the input variables, in our case the weighting of the chosen criteria. In general, we speak of scenario analysis, since a scenario represents one of the possible combinations of values assumed by independent variables, or ‘what-if’ analysis, since we evaluate what changes if the values assumed by the decision parameters change. The sensitivity analysis in this study was conducted using the results of AHP and TOPSIS.

3. Materials and methods

The methodological approach that has been adopted for the optimal allocation of floating photovoltaic systems in Sicily is organised according to the following steps:

• Identification of lakes in Sicily;

• Definition of criteria;

• Research for the optimal allocation of floating photovoltaic systems.

The next few paragraphs will consider these different aspects mentioned.

3.1. Identification of lakes in Sicily

To obtain the data of the geographical coordinates of 47 basins in Sicily (Table 2), a database compiled by ISPRA (SINAnet 2021) (Higher Institute for Environmental Protection and Research) was used. These were considered during the analysis phase.

Table 2. Water basins in Sicily.

Basin Identification	Coordinates (° ′ ″)	Extension [km^2]
Biviere	37°19′29″N, 14°57′03″E	9.312
Pozzillo	37°39′34″N, 14°35′32″E	6.275
Garcia	37°47′49″N, 13°07′17″E	3.496
Rosamarina	37°56′32″N, 13°38′21″E	3.206
Arancio	37°37′48″N, 13°3′36″E	2.687
Ogliastro	37°26′38″N, 14°33′57″E	2.535
Piana degli Albanesi	37°58′27″N, 13°17′49″E	2.264
Poma	37°59′24″N, 13°6′0″E	1.630
Trinità	37°42′02″N, 12°45′10″E	1.581
San Giovanni	37°18′40″N, 13°46′10″E	1.548
Castello	37°34′56″N, 13°25′05″E	1.488
Fanaco	37°39′33″N, 13°32′59″E	1.330
Rubino	37°53′34″N, 12°43′16″E	1.267
Longarini	36°42′41″N, 15°00′47″E	1.240
Santa Rosalia	36°58′33″N, 14°46′46″E	1.222
Pergusa	37°30′50″N, 14°18′20″E	1.126
Ancipa	37°50′12″N, 14°33′28″E	1.100
Baiata	37°58′26″N, 12°35′04″E	1.065
Roveto	36°47′24″N, 15°05′31″E	0.941
Nicoletti	37°36′36″N, 14°20′24″E	0.864
Prizzi	37°43′48″N, 13°24′25″E	0.816
Scanzano	37°55′11″N, 13°22′12″E	0.745
Dirillo	37°7′40″N, 14°41′52″E	0.730
Biviere	37°01′14″N, 14°20′27″E	0.690
Morello	37°35′04″N, 14°12′28″E	0.610
Disueri	37°12′0″N, 14°17′24″E	0.559
Olivo	37°24′18″N, 14°17′18″E	0.463
Pian del Leone	37°40′16″N, 13°28′30″E	0.431
Gorgo	37°24′31″N, 13°19′30″E	0.328
Ganzirri	38°15′40″N, 15°37′3″E	0.320
Comunelli	37°9′27″N, 14°9′19″E	0.277
Prèola	37°37′16″N, 12°38′20″E	0.263
Furore	37°15′42″N, 13°43′43″E	0.247
Biviere	37°57′6″N, 14°42′60″E	0.170
Trearie	37°57′8″N, 14°50′23″E	0.128
Gammàuta	37°41′08″N, 13°21′14″E	0.093
Gorghi tondi	37°36′40″N, 12°39′05″E	0.069
Soprano	37°27′36″N, 13°52′38″E	0.063
Cartolari	37°57′24″N, 14°49′9″E	0.052
Maulazzo	37°56′28″N, 14°40′18″E	0.050
Gurrida	37°51′24″N, 14°54′0″E	0.047
Gariffi	36°44′00″N, 14°56′19″E	0.020
Favara	37°35′52″N, 13°14′55″E	0.010
Pisciotto	37°58′45″N, 14°50′52″E	0.009
Marinello	38°08′12″N, 15°03′15″E	0.006
Urio Quattrocchi	37°54′5″N, 14°23′45″E	0.006
Spartà	38°01′50″N, 14°38′53″E	0.004

As shown in Table 2, the basin with the largest extent is Lake Biviere in Lentini, in the province of Syracuse, and the basin with the smallest extent is Lake Spartà in Sant’Agata di Militello, in the province of Messina. From the values listed in Table 2, it can be deduced that the average value of the Sicilian basins is 1.17 km². The total area of the basins analysed is about 55 km², which is about 0.2% of the total area of the Sicily region (25832.39 km²). The basins of Biviere and Pozzillo can be defined as large; with an area between 3.50 and 1.07 km², they are medium-sized basins, while the rest are small, considering that they are basins with an area of less than 1.00 km². Thus, the identified basins are mostly small and located mainly in the eastern part of Sicily, in the province of Messina, Catania, Syracuse and Ragusa.

3.2. Criteria definition

It is conceivable that in Sicily solar irradiation is optimal for the production of solar energy. The choice of criteria and their weights can be defined as subjective; the chosen criteria in order to evaluate the optimal siting on Sicilian lakes are:

• Costs of the plant: costs related to installation and maintenance greatly affect the construction of new systems, the minimisation of this criterium is a huge advantage;

• The distance of the plants from nearby medium voltage connection grid: proximity to connection grid is very important, thus the costs of capital expenditure will be reduced (Zoghi et al. 2017; Yushchenko et al. 2018);

• The annual energy yield by the plant: the more a plant is able to produce the greater its revenues, this is a criterium that depends on the surface of the basins (hypothesised coverage 1%) and acts to significantly reduce the payback time;

• LCOE (levelised cost of energy): it is a measure that quantifies the average cost at which to sell the electricity generated during the plant life. It is a criterium that depends on the investment costs, maintenance costs, interest, how long the life cycle of the plant lasts and the electricity it produces;

• CO₂ emissions: Measure the amount of carbon dioxide that can be avoided by generating electricity from renewable sources. This criterium depends on the energy produced and the area occupied by the plant, as it allows the measurement of emissions generated during the production cycle of the plant components, their transport and installation.

In Bandaru et al. (2021), different hypotheses for the weighting of criteria are put forward; in most of the proposed scenarios, technical and financial criteria are given greater importance than ecological and social criteria. In Villacreses et al. (2022), the group of climatic criteria has a weight of 40%, that of socio-economic criteria 35% (social 14% and economic 21%) and that of environmental criteria 25%. In Yushchenko et al. (2018), the economic criteria are maximised in relation to the technical and social criteria; environmental criteria are not considered. In this study, two scenarios were considered, with the technical criteria having a weight between 38.1% and 50.6%, the economic criteria between 43.9% and 56.3% and the social criteria between 5.4% and 9.5%. In another case study (Sindhu, Nehra, and Luthra 2017), various criteria are considered, classified as social, technical, economic, environmental and political; these are assigned 9.9%, 18.4%, 24.2%, 15.6% and 31.9%, respectively. We can thus conclude that in the literature, the criteria of the technical domain are assigned between 18.4% and 50.6%, those of the economic domain between 21.0% and 56.3%, and those of the environmental domain between 15.6% and 25%. In this study, the capital and operating costs and the electricity production costs are the economic criteria, the environmental criterium is the CO₂ emission, and the technical criteria are the annual energy yield and the distance to the power line.

Therefore, it is considered that the focus should be on minimising LCOE, costs, both installation and maintenance, and maximising annual energy yield and CO₂ emissions. Relatively high and equal weight was given to costs and energy yield. The distance of the plants from the medium-voltage grid must be minimised, but this criterium was given a lower weight than the previous ones. A relatively high and important weight was assigned to CO₂ emissions. The levelling of energy costs was given the highest weight among the five criteria, as this is the parameter that makes clear what the economic performance of the plants should be in order to balance the costs.

The weights, like the criteria, were determined based on what was said before, with a relative weight assigned to each criterium. Minimising costs (installation and maintenance) and maximising CO₂ emissions and energy yield are given much greater importance in this analysis than minimising distance to the connection grid. As mentioned above, each criterium must be assigned a weight between 1 and 9, depending on how important it is for the case study. The interaction between weighting and criteria has to be done according to the ideas of the experts dealing with the case, the local regulations and the rules for awarding floating photovoltaic plants. Greater weight was given to the electricity production costs and CO₂ emissions, and somewhat less weight to the costs and energy yield, as these are criteria of fundamental importance. Minimising electricity production costs and costs and maximising CO₂ emissions and annual energy yield is a requirement that must be taken into account in the planning and construction of new floating photovoltaic systems. Keeping the distance to the connection grid as small as possible is important for capital expenditure and reducing grid losses. This criterium also depends on the size of the plant, as it determines the connection to a specific voltage line (low, medium, high and extra-high voltage). In addition, the cost of connection to the grid, to be borne by the owner who builds the plant, varies according to the voltage level and distance from the connection line; however, less weight is given to this criterium, as it has a lesser impact on the study carried out. The energy generated, which is relevant for the annual energy yield, and the associated grid losses, together with the capital costs and the distance from the power line, influence the LCOE criterium. Taking into account some ranges that can be derived in the literature and after performing the relevant assessments, the criteria listed in Table 3 were weighted for the present case study.

Table 3. MCDA’s criteria.

Parameter	Weight
Costs	16%
Distance from the connection grid	13%
Annual energy yield	16%
LCOE	30%
CO2 emissions	25%

The construction of floating photovoltaic systems involves various costs. In Tina and Rosa-Clot (2020) the capital costs (CAPEX) for floating plants of 1 MW, with different technologies, are divided for photovoltaic panels, electrical parts, inverters and cables, assembly costs, for the structure and for the rafts. The amount for the solution relating to a plant in Singapore is US$ 803,692, the Gable ‘Slender’ Solution has a cost of US$ 590,556, the capital amount relating to the construction of the Gable2 solution is US$ 630,106. In Oliveira-Pinto and Stokkermans (2020) CAPEX are assumed to be 1.09 US$/Wp and include photovoltaic modules, inverters, system balancing costs, engineering, procurement and construction costs (EPC) and other costs. In Devabhaktuni et al. (2013) a review was carried out on the capital costs of FPVs and it is stated that the costs, in 2018, varied in a range between 0.8 and 1.2 US$/Wp.

The maintenance costs are assumed to be constant during the plant life cycle, considered, in this study, to be 25 years. For floating photovoltaic systems, maintenance costs are limited as no maintenance is required at the site where the panels are located unlike the ground systems where soil cleaning is required. For the tracking systems, maintenance of the movement systems and panels is necessary, while in the case of panels with built-in cooling it is not necessary to clean the modules (Tina and Rosa-Clot 2020). For the evaluation of the costs, in this study, the basins investigated were divided into four categories based on the installed capacity; for the CAPEX the model used in (Oliveira-Pinto and Stokkermans 2020) was applied which takes into account the cost of the panels, inverters, balancing and construction costs, etc. For plants with size between 0 and 1 MWp, a capital expenditure of US$/Wp 1.24 and a maintenance cost of US$ 18,750 was considered, for plants between 1 and 2 MWp capital expenditure have been set at US$/Wp 1.09 and maintenance cost at US$ 37,500; CAPEX for the plants with size between 2 and 3 MWp are fixed at US$/Wp 0.98 and OPEX at US$ 56,250, for larger plants, 3–4 MWp, capital costs at US$/Wp 0.87 and operating cost at US$ 75,000. It should be noted that connection costs to the power line were not taken into account during this analysis. (8) $\mathrm{C}\mathrm{o}\mathrm{s}\mathrm{t}\mathrm{s}\left[\mathrm{U}\mathrm{S}\mathrm{\$}/\mathrm{W}\mathrm{p}\right]={\displaystyle \frac{\mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E}\mathrm{X}+\mathrm{O}\mathrm{P}\mathrm{E}\mathrm{X}}{\mathrm{P}\mathrm{o}\mathrm{w}\mathrm{e}\mathrm{r}\mathrm{u}\mathrm{n}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{l}\mathrm{e}\mathrm{d}}}$(8) (9) $\mathrm{P}\mathrm{o}\mathrm{w}\mathrm{e}\mathrm{r}\mathrm{u}\mathrm{n}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{l}\mathrm{e}\mathrm{d}=\mathrm{p}\mathrm{a}\mathrm{n}\mathrm{e}\mathrm{l}\mathrm{s}\mathrm{n}\mathrm{u}\mathrm{m}\mathrm{b}\mathrm{e}\mathrm{r}\times \mathrm{m}\mathrm{a}\mathrm{x}\mathrm{i}\mathrm{m}\mathrm{u}\mathrm{m}\mathrm{p}\mathrm{a}\mathrm{n}\mathrm{e}\mathrm{l}\mathrm{p}\mathrm{o}\mathrm{w}\mathrm{e}\mathrm{r}$(9)

where the number of panels depends on the available surface of the basin (1% of the total surface) and the surface of the panel; maximum panel power depends of the characteristics of the panel itself; in this case the power of the panel is 410 Wp (ENF Solar 2021).

The distances from the connection grid have been derived from QGIS using polyline layers and the annual energy yield has been calculated as: (10) $\mathrm{A}\mathrm{n}\mathrm{n}\mathrm{u}\mathrm{a}\mathrm{l}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{g}\mathrm{y}\mathrm{y}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}\left[{\displaystyle \frac{\mathrm{M}\mathrm{W}\mathrm{h}/\mathrm{y}}{\mathrm{M}\mathrm{W}\mathrm{p}}}\right]={\displaystyle \frac{\mathrm{E}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{g}\mathrm{y}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{d}}{\mathrm{P}\mathrm{o}\mathrm{w}\mathrm{e}\mathrm{r}\mathrm{u}\mathrm{n}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{l}\mathrm{e}\mathrm{d}}}$(10) (11) $\mathrm{E}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{g}\mathrm{y}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{d}=(\mathrm{s}\mathrm{o}\mathrm{l}\mathrm{a}\mathrm{r}\mathrm{r}\mathrm{a}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\times \mathrm{h}\mathrm{y}\mathrm{p}\mathrm{o}\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{s}\mathrm{i}\mathrm{z}\mathrm{e}\mathrm{d}\mathrm{c}\mathrm{o}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{g}\mathrm{e}\times \eta \times \mathrm{T}\mathrm{R}\mathrm{E}\mathrm{r})\text{–}\mathrm{g}\mathrm{r}\mathrm{i}\mathrm{d}\mathrm{l}\mathrm{o}\mathrm{s}\mathrm{s}\mathrm{e}\mathrm{s}$(11)

where GHI is evaluated in each month of the year 2016; hypothesised coverage is the 1% of the extension expressed in Table 2; η denotes the real efficiency of one module. According to ENF Solar (2021) the chosen efficiency has been stated to the yield of monocrystalline technologies, corresponding to a value of 20.38%; TREr is the Theoretical-Real Efficiency ratio (average ratio between the manufacturers’ declared efficiency and the one site’s real one for mono-polycrystalline cells) whose value, evaluated in (Sun et al. 2013), is equal to 78%; grid losses are established by the national authority of electricity and gas (ARERA Autorità di Regolazione per Energia, Reti e Ambienti 2020), on grids with third party connection obligations. These are conventional values set in percentages that depend on the system voltage levels, the installed power of the plant and on the grid energy injection point taken into consideration. The grid losses are 2.3% of the energy generated; the power unit installed is (9).

The LCOE formula was used in a previous study (Nižetić et al. 2017). In this case, the parameter relating to fuel consumption is not used as it concerns production plants from renewable sources. (12) $\mathrm{L}\mathrm{C}\mathrm{O}\mathrm{E}\left[\mathrm{U}\mathrm{S}\mathrm{\$}/\mathrm{M}\mathrm{W}\mathrm{h}\right]={\displaystyle \frac{\left(\mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E}\mathrm{X}\times \mathrm{C}\mathrm{R}\mathrm{F}\right)+\mathrm{O}\mathrm{P}\mathrm{E}\mathrm{X}}{\mathrm{E}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{g}\mathrm{y}\mathrm{G}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{d}}}$(12)

where CAPEX are the investment costs; OPEX are the maintenance costs; energy generated is the energy efficiency of the (11); CRF is the capital recovery factor equivalent to ${\displaystyle \frac{\left(\left({\left(1-p\right)}^n\right)p\right)}{{\left(1-p\right)}^n-1}}$, p is the interest rate (2%), n is the plant life cycle (25 years). (13) $\mathrm{C}{\mathrm{O}}_2\mathrm{e}\mathrm{m}\mathrm{i}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}[\mathrm{t}\mathrm{C}{\mathrm{O}}_2]=\mathrm{C}{\mathrm{O}}_2\mathrm{e}\mathrm{m}\mathrm{i}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{f}\mathrm{o}\mathrm{r}\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{i}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{o}\mathrm{f}\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{F}\mathrm{P}\mathrm{V}\mathrm{s}-{\mathrm{C}\mathrm{O}}_2\mathrm{a}\mathrm{v}\mathrm{o}\mathrm{i}\mathrm{d}\mathrm{e}\mathrm{d}\mathrm{e}\mathrm{m}\mathrm{i}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{s}$(13)

where

• The $\mathrm{C}{\mathrm{O}}_2$emissions for the installation of the FPV were calculated taking into account the square meters of surface occupied by the floating photovoltaic system and the emissions entered into the atmosphere for each square meter of system (137.73 kgCO₂/m²) (Redón Santafé et al. 2014), making the relationship between the two factors are obtained the CO2 emissions produced during the production, transport and installation phase of the system;

• for the avoided emissions, on the other hand, we took into account how many emissions a thermoelectric production plant from fossil sources would produce, considering the same power as those from renewable sources covered by this study. This value was calculated with the product of the energy produced by the plant and the $\mathrm{C}{\mathrm{O}}_2$ emissions for each kWh, (536.4 gCO₂/kWh) (ISPRA 2022), produced from fossil fuels. The result was multiplied by 25 years, that is the life cycle of the floating photovoltaic system.

The same data were used in a subsequent phase of this study, where was examined the choice of the best photovoltaic technology (Di Grazia and Tina 2022).

3.3. Topological determination of the best reservoir for allocation of floating photovoltaic systems

GIS The software has been widely used to determine the performance of ground-mounted photovoltaic systems, taking into account variations in tilt and latitude, and to evaluate the integration of these systems in urban areas. The software has been used to analyse FPV systems to determine the optimal geographical position. The area of interest of this study concerns the region of Sicily in the southern part of Italy. The geographical coordinates of the natural lakes in the region (Table 2) were imported into QGIS as a geodatabase downloaded from the official website of the Italian Institute for Environmental Protection, the ISPRA Institute. With the QuickMapService plug-in it was possible to identify the basins via a satellite view and with OpenCoordinator, another plug-in, the coordinates were derived from the ISPRA database. The official data on the geographical coordinates of the protected areas were downloaded from the government portal (Ministry of Ecologica Transition 2020) and the resulting geographical points were loaded into QGIS. Lakes at least two kilometres away from SIC, ZPS, archaeological areas, rivers, etc. are evaluated in this study. Next, solar irradiance values such as global horizontal irradiance was obtained.

The global horizontal irradiance was evaluated for the lakes analysed, for the geographical coordinates of each lake and for each month. The GHI values for each month of the year were derived by a spatial and temporal evolution of the GHI through the PVGIS-SARAH database.

3.4. AHP

Using the SuperDecisions software, developed by the Creative Decisions Foundation, a model was created to carry out multi criteria analysis according to the AHP method. Having set the goal, that is, the identification of the optimal allocation site, the criteria were loaded into the software. The usable alternatives, i.e. the basins obtained from the analysis carried out with the QGIS software, were then attributed. The software used then requests, comparing the criteria two at a time, to assign the weights. The priorities of the criteria, as mentioned, have been assigned subjectively.

The matrix (5) was built by the software, the terms within it represent the coefficients of relative importance of each criterium with respect to the objective, the eigenvalues instead represent the relative priorities of each criterium with respect to the alternatives. The weights of each alternative are determined as the eigenvalues of the matrices, the terms are the normalised quantities of each alternative with respect to each criterium. It is therefore the terms of the matrix that quantify the relative importance of the various alternatives with respect to the others for each of the four chosen criteria. In order to overcome the consistency check of the hierarchical analytical process, the software requires that, for each criterium, a comparison must be made between two alternatives at a time.

After carrying out the entire process relating to Multi-Criteria Decision Analysis, the software is able to deliver to the designer the result relating to the initially entered objective by assigning a score to each alternative. Thus it is possible to determine which is the best site for the allocation of a floating photovoltaic system.

3.5. TOPSIS

To obtain the best site for the allocation of a floating photovoltaic system, through the TOPSIS method, as in Di Grazia and Tina (2022), a spreadsheet was used where the formulas useful for achieving the desired result were set. It is necessary to calculate the weighted normalised matrix in order to obtain which is the ideal solution and which is the worst solution. All the coefficients of the matrix are first normalised: (14) ${\overline{a}}_{ij}={\displaystyle \frac{a_{ij}}{\sqrt{{\sum }_{i=1}^n{a}_{ij}^2}}}$(14) where ${\overline{a}}_{ij}$ is the normalised coefficient, of a specific criterium linked to a specific alternative, which must then be related to the weight of the criterium; $a_{ij}$ is the criterium, linked to a specific alternative, to be normalised; $\sum _{i=1}^n{a}_{ij}^2$ is the summation of the criteria, of the alternatives, to be inserted in the same column of matrix. The coefficients, now normalised, must be related to the weight of the criterium so that they can be inserted in the weighted normalised matrix: (15) $v_{ij}={\overline{a}}_{ij}\times w_j$(15)

where $v_{ij}$ is the weighted normalised coefficient, of a specific criterium linked to a specific alternative, to be inserted into the matrix; ${\overline{a}}_{ij}$ is the normalised coefficient of a specific criterium linked to a specific alternative, which must be related to the weight of the criterium; $w_j$is the weight of the criterium.

To calculate the value of the ideal solution and the worst solution, the corresponding value must be chosen, depending on the criterium being analysed. For costs, distance and LCOE the ideal solution will be the lower coefficient between the row of values obtained for the two criteria, for the installable potential the ideal solution will be the higher coefficient of the column of the examined criterium. The worst solution for the first three criteria will be the higher coefficient, for the technical criterium it will be the lower coefficient. Having calculated which are the ideal solutions and the worst solutions, we can proceed to the calculation of the Euclidean distances, respectively from the ideal solution (S_i⁺) and from the worst solution (S_i⁻) (Phanden et al. 2021). (16) $S_i^{+}=\sqrt{\left[\sum _{j=1}^m{\left(v_{ij}-v_j^{+}\right)}^2\right]}$(16) (17) $S_i^{-}=\sqrt{\left[\sum _{j=1}^m{\left(v_{ij}-v_j^{-}\right)}^2\right]}$(17) After having calculated the Euclidean distance from the ideal solution and from the worst solution, we can proceed to deduce the score obtained from each alternative (P_i). (18) $P_i={\displaystyle \frac{S_i^{-}}{S_i^{+}+S_i^{-}}}$(18)

Since the values used during the final phase are normalised, the result will be analysed considering the maximum value of 1. The optimal allocation will therefore be the one with the score closest to 1, while the worst will be the one with the score furthest from 1.

3.6. Sensitivity analysis

Evaluating how much a criterium influences the final result is the main objective of the sensitivity analysis, this can be done by manipulating the variables (Winebrake and Creswick 2003). Carrying out this analysis helps in evaluating the minimal changes in the weights of the criteria, primarily hypothesised, and the alteration of the positions obtained after an initial Multi-Criteria Decision Analysis (Luthra et al. 2015). This approach can be done using EXCEL or, as in this case study, Super Decisions for the AHP method.

3.6.1. Sensitivity AHP

The sensitivity analysis for the MCDA AHP method was conducted with the aid of the Super Decisions software, as in Di Grazia and Tina (2022). Using the same model previously developed for the decision analysis, it is possible to carry out the sensitivity analysis. With this software it is possible to conduct the sensitivity analysis in two different ways:

1. The AHP sensitivity shows how the best alternative changes when the weight of a criterium changes;

2. Dynamic sensitivity shows how the priority of all alternatives changes as the priority of a criterium changes.

The first modality can be done using the ‘Sensitivity’ command: by selecting the criterium whose weight you want to change, the best alternative is displayed as the weight varies.

The second modality is used with the ‘Node Sensitivity’ command; there are four different views of the variation that the alternatives prioritise. With the ‘Barchart’ option it is possible to observe through a three-dimensional histogram how the ranking of the alternatives varies.

The method chosen to carry out the sensitivity analysis, during this process, is the second. It is possible to have a result, by being able to manage the weight of the criterium analysed, by a command placed above the display of the results.

3.6.2. Sensitivity TOPSIS

The sensitivity analysis for the TOPSIS method was carried out by varying, all five at the same time, the weights of the criteria chosen to carry out the multi criteria approach. This procedure was carried out using an Excel sheet and structuring four different scenarios; initially it was assumed that the four criteria have the same weight, i.e. that each of them accounts for 25% in the final decision. Then one criterium was maximised at a time: the energy potential, the costs, the distance from the connection grid, the LCOE and finally the CO₂ emissions.

4. Results and discussion

The data on the geographical coordinates of the 47 identified catchments (Table 2) on the territory of the Sicily region were loaded into the QGIS software. They were identified on the map using the QuickMapService plug-in; the coordinates were obtained using the OpenCoordinator plug-in.

Data on the geographical coordinates of the protected areas (SIC, ZPS, archaeological areas, etc.) were downloaded from the Sicily Region website and uploaded into QGIS. At this stage, the available lakes (Table 4) could be selected for the subsequent step of multi-criteria analysis in order to evaluate which is an optimal location for the placement of floating photovoltaic plants. Data useful for evaluating the best location were also sought, such as annual rainfall, average and maximum summer temperatures, and average annual temperatures. The data were extrapolated from the annual ISPRA database and refer to 2016, the latest year available (ISPRA AMBIENTE 2021). These data are useful for a possible assessment of water evaporation in the catchments and for calculating the decline in water volume. Annual global horizontal irradiance data (GHI) for the seven lakes studied were also extrapolated from the PVGIS-Sarah database. The contribution of global horizontal irradiance is fundamental to the assessment of the energy generated in the basins. Specifically, the 2016 data were used for the analysis phase. The values of GHI, available in Table 4, do not differ much because the evaluation is based on geographical coordinates of a relatively small area such as Sicily; if the values were different, they would not be justifiable. The calculation for the evaluation of the annual energy yield was carried out considering the values in the database PVGIS-SARAH, as explained below.

Table 4. Available basins for FPV’s allocation.

Basin Identification	Use	Annual rainfall [mm]	Tsum,avg [°C]	Tsum,max [°C]	Tann,avg [°C]	Annual GHI [kWh/m^2]
Trinità	Irriguous	508.0	26.0	39.7	19.1	1780.93
San Giovanni	Irriguous, sport	548.2	24.5	35.7	17.5	1794.99
Baiata	Irriguous	331.8	25.0	36.5	19.9	1777.71
Nicoletti	Irriguous, industrial	439.4	25.3	39.1	17.4	1771.32
Dirillo	Irriguous, sport, industrial	605.3	25.3	37.7	17.6	1810.96
Morello	Irriguous, sport	471.9	25.3	39.1	17.4	1767.90
Comunelli	Irriguous	391.6	24.9	37.8	18.2	1812.09

The coverage hypothesis of the study is 1% of the total area of each catchment area listed in Table 2, whose characteristics are shown in Table 4, also used in (Di Grazia and Tina 2022).

4.1. Criteria

The criteria addressed to paragraph 3.3 have been calculated with the formulas previously mentioned: (8) $\mathrm{C}\mathrm{o}\mathrm{s}\mathrm{t}\mathrm{s}\left[\mathrm{U}\mathrm{S}\mathrm{\$}/\mathrm{W}\mathrm{p}\right]={\displaystyle \frac{\mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E}\mathrm{X}+\mathrm{O}\mathrm{P}\mathrm{E}\mathrm{X}}{\mathrm{P}\mathrm{o}\mathrm{w}\mathrm{e}\mathrm{r}\mathrm{u}\mathrm{n}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{l}\mathrm{e}\mathrm{d}}}$(8)

The distances from connection grid have been derived from QGIS using polyline layers and the annual energy yield has been calculated as: (10) $\mathrm{A}\mathrm{n}\mathrm{n}\mathrm{u}\mathrm{a}\mathrm{l}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{g}\mathrm{y}\mathrm{y}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}=\left[{\displaystyle \frac{\mathrm{M}\mathrm{W}\mathrm{h}/\mathrm{y}}{\mathrm{M}\mathrm{W}\mathrm{p}}}\right]={\displaystyle \frac{\mathrm{E}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{g}\mathrm{y}\mathrm{g}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{d}}{\mathrm{P}\mathrm{o}\mathrm{w}\mathrm{e}\mathrm{r}\mathrm{u}\mathrm{n}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{l}\mathrm{e}\mathrm{d}}}$(10) (12) $\mathrm{L}\mathrm{C}\mathrm{O}\mathrm{E}\left[\mathrm{U}\mathrm{S}\mathrm{\$}/\mathrm{M}\mathrm{W}\mathrm{h}\right]={\displaystyle \frac{\left(\mathrm{C}\mathrm{A}\mathrm{P}\mathrm{E}\mathrm{X}\times \mathrm{C}\mathrm{R}\mathrm{F}\right)+\mathrm{O}\mathrm{P}\mathrm{E}\mathrm{X}}{\mathrm{E}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{g}\mathrm{y}\mathrm{G}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{d}}}$(12) (13) $\mathrm{C}{\mathrm{O}}_2\mathrm{e}\mathrm{m}\mathrm{i}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{s}\left[\mathrm{t}\mathrm{C}{\mathrm{O}}_2\right]=\mathrm{C}{\mathrm{O}}_2\mathrm{e}\mathrm{m}\mathrm{i}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{f}\mathrm{o}\mathrm{r}\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{i}\mathrm{n}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{o}\mathrm{f}\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{F}\mathrm{P}\mathrm{V}\mathrm{s}-\mathrm{C}{\mathrm{O}}_2\mathrm{a}\mathrm{v}\mathrm{o}\mathrm{i}\mathrm{d}\mathrm{e}\mathrm{d}\mathrm{e}\mathrm{m}\mathrm{i}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{s}$(13)

For the purposes of calculating the previously mentioned criteria, it is necessary to explicit the installed power unit and the capital and maintenance costs (Table 5). Those, in the calculation of the multi criteria decision making process, are added together and expressed as a function of the installed power unit.

Table 5. Cost and power installed value.

Lakes	CAPEX [USD]	OPEX [USD]	Overall costs [USD]	Power unit installed [Wp]
Trinità	3205018	75000	3280018	3683929
San Giovanni	3136723	75000	3211723	3605428
Baiata	2432079	56250	2488329	2481714
Nicoletti	1972420	56250	2028670	2012674
Dirillo	1845209	37500	1882709	1692853
Morello	1550002	37500	1587502	1422020
Comunelli	800565	18750	819315	645617

The following results for the four criteria have been obtained for each lake (Table 6).

Table 6. Criteria’s value.

Lakes	Costs [USD/Wp]	Distance from connection grid [m]	Annual energy yield [MWh/y/MWp]	LCOE [USD/MWh]	Avoided emissions in 25 years [tCO2]
Trinità	0.89	3433	1188.59	54.68	56477.55
San Giovanni	0.89	1716	1195.18	54.62	55727.14
Baiata	1.00	2060	1186.00	61.48	37975.16
Nicoletti	1.01	4882	1182.49	66.17	30682.93
Dirillo	1.11	2939	1209.38	64.59	26407.30
Morello	1.12	2428	1180.31	69.74	21634.95
Comunelli	1.27	4124	1199.95	76.61	10077.69

4.2. AHP

The AHP and its methodology (par 3.4), through the use of the SuperDecisions software, were the means by which the first of the two multi-criteria analysis was conducted; the five criteria, with their weights, related to the seven alternatives have given very important indications regarding the optimal allocation of a FPVs in Sicily.

The following scores, in Table 7, have been obtained.

Table 7. AHP’s score.

Lake	Score
San Giovanni	1.00
Trinità	0.69
Baiata	0.59
Dirillo	0.58
Comunelli	0.37
Nicoletti	0.33
Villarosa	0.32

The obtained results suggest San Giovanni dam as the best alternative to be chosen.

4.3. TOPSIS

The TOPSIS method pursues the idea of establishing ideal solutions and worst solutions to carry out the study; the alternative that comes closest to the ideal solution turns out to be the best of those analysed and consequently is the one to be taken into consideration. An Excel sheet was used to perform Multi-Criteria Decision Analysis which, assuming ideal solutions and worst solutions for each criterium, made it possible to calculate the performance score of each solution. These are normalised values ⁣⁣also in this case, having first normalised the weighted matrix.

The resulting, normalised, is reported in Table 8.

Table 8. TOPSIS’ score.

Lake	Score
San Giovanni	0.985
Trinità	0.830
Baiata	0.643
Dirillo	0.407
Nicoletti	0.406
Morello	0.347
Comunelli	0.081

According to the results, the minimum value of the distance from the ideal solution is found for San Giovanni dam.

4.4. Sensitivity analysis

For both MCDA methods, a sensitivity analysis was carried out to evaluate how the results vary as the weight of the criteria, inputted during the analysis, varies. The results obtained how much the subjective choice of the decision maker affects the final result and the choices that this could entail.

4.4.1. Sensitivity AHP

The sensitivity analysis was carried out with the SuperDecisions software, setting it dynamically, changing the weight of the criteria from time to time and displaying how the results vary.

The variation of the result, with the variation of the weight of each criterium, and therefore the variation of the best alternative, gave a result for each situation. For costs, the best alternative, when the weight varies from 0% to 100%, remains the San Giovanni Dam; the best alternative, also for the criterium relating to the distance from the national transmission grid, is the Dam. Up to 68% of the weight attributed to the energy efficiency criterium, the best alternative is the San Giovanni Dam, for a heavier weight it is Lake Dirillo. For the last two criteria, LCOE and CO₂ emissions, the best alternative is always the San Giovanni Dam. These results show that, except in the case in which the annual energy yield is attributed a weight higher than 69%, the San Giovanni dam represents the best alternative among those proposed.

4.4.2. Sensitivity TOPSIS

The scenarios assumed in this sensitivity analysis are six. The first scenario assessed is the one considered during the study phase explained earlier. In the second scenario, the five criteria were equally weighted, i.e. 20% of the total weight. The third scenario maximises the criterium of annual energy yield by weighting it with 30% of the total weight; 25% was attributed to the distance of the connection to the transmission line, 11% to the electricity production costs and 17% to the costs and CO₂ emissions. In the fourth scenario, 30% of the weight was attributed to costs, 25% to energy efficiency, 11% to distance and 17% to electricity generation costs and CO₂ emissions. In the fifth scenario, the distance-from-connection criterium was maximised at 30% of the weight; in the last scenario, the CO₂ criterium was maximised at 30%, and 17% of the weight was attributed to LCOE, 25% to costs and 11% to annual energy performance.

The analysis was carried out by varying the weighting of the criteria from time to time. Again, in all scenarios, the San Giovanni Dam basin was shown, in Figure 1, to be the best choice for the hypothetical deployment of a floating photovoltaic system.

Figure 1. Radar chart: TOPSIS’ sensitivity results.

5. Conclusions

The technology of floating photovoltaic systems is constantly spreading around the globe and has gave opportunities for the renewable energy sector, leading to great advantages over non-renewable energy. Floating photovoltaic systems can play an important role in meeting the energy needs in Italy and around the world. This study aims to improve the potential of the joint use of MCDA and GIS software for the optimal allocation of floating photovoltaic plants. The case study, applied to the region of Sicily (in Italy), was carried out by first selecting the usable basins. The seven alternatives analysed are in fact geographically far from protected areas, archaeological areas or from sites where it is not possible to allocate plants. Two different multi-criteria decision-making approaches (AHP and TOPSIS) were used; using five environmental, technical and economic criteria, the LCOE analysis was carried out, the results of which were used as criteria for the decision analysis. The Multi-Criteria Decision Analysis gave the same result for both methods. The San Giovanni dam was indeed the best location for the allocation of floating photovoltaic plants. The above results illustrate how floating photovoltaic systems can bring about a positive development in renewable energy production, focusing on environmental friendliness, costs comparable to conventional photovoltaic systems and high-energy performance. It should be noted that the study was conducted at a regional level and that a detailed analysis should be carried out, perhaps based on sites with different implementations of the method, to be sure of the robustness of the method. One of the limitations of this method may be the regional level analysis used to conduct the study: for example, the solar radiation data does not differ between the different methods. Another limitation of the study is the sparse literature on maintenance costs during the life cycle of the plants. This has limited the data collection and its comparison in the analysis. This approach could be used by those interested in identifying areas for power generation. In addition, the current state of the literature for this type of study is constantly evolving, especially for ground-based photovoltaic systems. Finally, a sensitivity analysis was carried out after obtaining the results of the multi-criteria decision-making approach. This shows the soundness of the methods used, as when the criteria selected and weighted by the decision maker were varied, the result did not change in either case. The observation that emerges from the sensitivity analysis is therefore that the use of multi-criteria decision-making methods is appropriate for the planning of floating photovoltaic systems.

A multi-criteria scenario should be carried out in the design phase as part of measures to consider several aspects simultaneously, but it should be emphasised that different results are obtained depending on the criterium to be maximised. The methodology used in this study can be further developed and applied, and the results obtained can serve as a starting point for a more in-depth study of the topic.

Future scenarios of this current article can be a study of the San Giovanni dam to select the best floating photovoltaic system with fixed and tracked panels using the MCDA system.