**1. Introduction**

Water scarcity [1,2], the energy crisis [3], and food scarcity [4,5] are the largest currently coupled challenges [6] facing the global community, where they most severely affect the arid and semiarid regions of the world [7]. There is a wide scientific consensus that combustion of fossil fuels for energy is increasing atmospheric carbon dioxide (CO2) concentrations and driving climate change [8]. This anthropogenic climate change is increasing globally averaged mean annual air temperatures and driving changes in precipitation, which are expected to continue and increase [9,10]. The IPCC (Intergovernmental Panel on Climate Change) warns that the climate change over the next century will affect rainfall, river flows and sea levels all over the world [11], which will negatively impact agricultural yield [12]; particularly in already-malnourished sub-Saharan Africa. de Wit and Stankiewicz [13] predict rainfall in sub-Saharan Africa could drop by 10% causing surface drainage to drop 30–50% by midcentury, which would cause major water shortages. It is widely agreed that to prevent the worst of climate change, humanity needs to rapidly convert fossil fuel-based energy systems to renewable energy systems [14]. Solar photovoltaic (PV) technology is the most widely accessible, sustainable, and clean renewable source of energy that can be scaled to meet humanity's energy needs [15,16]. To meet these needs, however, a substantial amount of land is still needed for PV to replace fossil fuels

and this creates competition for limited land resources between food and energy [17]. A utility-scale PV plant land occupation varies between 20 km2/GWh and 40 km2/GWh depending on the type of solar panels used [18]. Despite life cycle carbon emissions [19], PV is more land efficient than all carbon capture and sequestration plans for coal [20], but with nearly a billion people already living undernourished, further reductions in agricultural land are not acceptable during a world food crisis [21].

A potential solution to these coupled water–energy–food challenges is the concept of floating photovoltaics or floatovoltaics (FPV), which has been rapidly gaining a base in scientific literature [22–28]. FPV is growing fast and is expected to have an average growth rate of above 20% in the next five years due to extremely low costs (with an FPV bid recently coming in for a system in Thailand at under USD 0.50/Wp) [29]. FPV are easier to install and simpler to decommission than conventional PV systems and the racking costs are less, which lead to these overall cost savings [29]. As FPV are located near or immersed in water, the operational temperature is reduced, which raises the solar energy conversion efficiency [23,26,30–34]. In regions where water scarcity is an issue and particularly when this issue is likely to be aggravated by climate change, FPV can also be used to reduce water loss because it can reduce water evaporation by more than 70% [32,35–37]. The Penman–Monteith daily evaporation method indicates that FPV could even cut evaporation by as much as 90% [38]. Studies in China [39] and India [40,41] have all indicated massive potential water savings for both small and large FPV coverage areas. This is particularly important for preservation of water sources in arid and semi-arid regions, especially with water shortages in the region [42]. FPV, therefore, also holds substantial promise when coupling with existing hydro power to make dual use of the electrical infrastructure while improving the water resource itself [39,43]. Similar advantages are to be expected for hybrid systems with pumped storage [44]. Finally, there is also evidence that FPV deployment reduces the PV degradation rate below 0.5% per year [45], which improves the levelized cost of solar electricity further.

FPV research has focused on several system design strategies [46]:


The thin-film FPV design has the advantage of reducing racking costs even more so than pontoon style FPV, as it clearly stops more evaporation and gains an advantage by the operational temperature being lower. However, the temperature coefficients are better for amorphous silicon (a-Si:H) thin film materials than those of crystalline silicon (c-Si) so the benefits of the water cooling are muted for a-Si:H-based FPV.

In this study, a new approach is used with a flexible crystalline silicon module on a similar foam system to that described by Pierce et al. [54] for a-Si:H FPV. This approach enables a larger solar electric output gain (or FPV boost) and as solar is largely already profitable, there is an opportunity for the electricity production value of c-Si flexible foam-backed FPV to subsidize a means of water conservation by cutting water evaporation losses. To build on past FPV work and investigate the potential of FPV coupled to hydro power in the U.S., the water saving potential at Lake Mead using FPV is investigated in this study. Lake Mead is an artificial reservoir created by the United States government to run the Hoover Dam, which was built in 1935 [55,56]. This novel form of FPV is analyzed for water-saving using an evaporation calculation adapted from the Penman–Monteith daily evaporation model [57] that is approved by the Food and Agriculture Organization of the United Nations (FAO) [58]. An energy production analysis is performed and an open source spreadsheet was developed to simulate the evaporation and the energy yield of the flexible FPV [59], as well as to investigate the impact of using passive water-cooled FPV, where the cooling potential was measured experimentally for a foam-based FPV. The potential is determined for a case study based on the coverage of FPV ranging from 10% to

50% [60] of Lake Mead. The results are compared to "conventional" tilted pontoon-style FPV and are discussed in the context of the energy–water–food nexus.

#### **2. Materials and Methods**

#### *2.1. Data Collection*

#### 2.1.1. Lake Evaporation Data

Most of the weather data used in this study were collected on Lake Mead through buoys installed by the United States National Oceanic and Atmospheric Administration's National Data Buoy Center (NOAA-NDBC) [61]. The rest of the data were obtained from open-access weather data made available by the McCarran International Airport's weather station in Las Vegas [62], and from SOLCAST, a solar data provider [63].

The main characteristics of the lake differ slightly from one study to another and depend on the year the study was conducted. In this study, the lake characteristics' values used for the evaporation calculation are taken from the National Park Service (NPS) website [64]. According to the NPS, as of 2010, the lake has a maximum surface area of 159,866 acres (647 million m2), and a maximum capacity of 29,686,054 acre-feet (36,617 million m3). The mean depth of the lake is estimated to be 55.5 m by the National Park Service [56]. The elevation of the lake is 328.574 m above sea water level. The weather buoy used to collect the data is located in the North Boulder Basin of the lake at a geographical position of latitude 36.087 N and longitude 114.728 W. The temperature sensor for air temperature collection is located at a height of 2 m above the lake surface while the anemometer is at 3 m above. Additionally, the atmospheric pressure sensor is located at 330.574 m above sea water level or 2 m above the lake surface, and the water temperature is measured at 0.5 m below the lake surface [65].

The buoy installed in Lake Mead's North Boulder basin by the NOAA-NDBC has been capturing different types of variables since 2016, which are stored in a historical database on the agency's website. Among the data required to conduct an evaporation calculation using the Penman–Monteith model, the wind speed (*ws*), the atmospheric pressure (*P*), the maximum (*Tw,max*), minimum (*Tw,min*), and daily mean (*Tw*) water temperature; and the air temperature were obtained from the NOAA-NDBC historical database. The rest of the data were not captured by the buoy; therefore, alternative methods have been used to gather the required data. According to Moreo and Swancar, when data are not available for the study location, nearby airport weather data can be used instead [55]. In this study, the nearest airport close to Lake Mead is the Las Vegas Airport. The relative humidity (*Rh*) data have thus been obtained from the Weather Underground website that has made data from the Las Vegas Airport available. The remaining variable is the daily incoming solar irradiation or global horizontal irradiation (*RS*) that has been obtained from SOLCAST's historical database [63]. This variable is also used for the solar energy production modeling.

The raw data from the NOAA-NDBC database were collected with an interval of 10 min starting at 00 h 00 min each day while the data from the Las Vegas Airport were measured with an 1 h interval starting at 00 h 56 min each day. Since daily data were required for the calculation, a mean daily value has been calculated for each variable. First, the data obtained from the NOAA-NDBC were cleaned by keeping only hourly data at the beginning of the hour (00 min) in order to match the data from Las Vegas Airport. A MATLAB code [66] was developed to perform this operation. Then, the same code was used to strip the missing data from the data table. A line of data was considered missing from the data table if one or more of the variables were not recorded by either the NOAA-NDBC station sensors or the Las Vegas Airport station sensors. After that, the data were reported in a spreadsheet that was used to calculate the mean daily value of the wind speed (*ws*), the atmospheric pressure (*P*), the water temperature (*Tw*) and the air temperature (*Ta*) by averaging the hourly data for each day. Another method used in the literature to calculate daily mean weather data is to calculate the average of the maximum and minimum value of the day [67]. However, studies have shown that if data are available, it is best to calculate the mean daily temperature by averaging the hourly values [68,69]. The spreadsheet was also used to retrieve the maximum (*Tw,max*), and minimum (*Tw,min*) daily temperatures as well as the maximum (*Rhmax*), and minimum (*Rhmin*) daily relative humidity. The number of missing data points was 246 hourly data. Instead of having total hourly data of 8760 points, 8514 data points were used for this study after the data cleaning process. There was no more than 3 missing data points for a single day except for 5 specific days that are the 4th, 60th, 97th, 318th, and 347th day of the year 2018. These 5 days were, respectively, missing 4, 4, 10, 5, and 16 data points. The days with the highest number of missing data were the 97th and 347th day of the year. Since there are only two such days among the 365 that populated the year 2018, it has been considered that it will not have a significant impact on the results. Therefore, the available data were representative in estimating the mean daily values of the variables for each day.
