Temporal and spatial variability of energy intensity for atmospheric water generators

Atmospheric water generators (AWGs) produce potable water from the moisture in the air, providing a potentially viable water source in austere locations or emergency response scenarios. In this study, the operating constraints of three existing commercially available AWG devices are investigated, compared to historical weather data from across the continental United States. Utilizing linear regression modeling and weather station data for the years of 1985–2019, the monthly and spatial trends of energy demand to produce water from these devices are estimated. Energy and water production efficiencies for the devices are highly dependent on environmental conditions with relative humidity (RH) and temperature as the two driving factors. Publicly available manufacturer specifications for each AWG system were modeled to predict yield and specific energy consumption (SEC). A spatial analysis depicts the change in SEC in kilowatt-hours per liter (kWh l−1) across the country at a monthly scale. SEC for refrigeration AWG ranged between 0.02 and 3.64 kWh l−1 and solar driven sorption was between 3.19 and 5.29 kWh l−1, significantly larger than conventional water treatment energy demands. Additionally, the results are synthesized based on the Köppen–Geiger climate classification system, to approximate projected water production and energy demand for each environment, with arid climates demanding larger energy consumption per unit volume of water produced. Excluding arid and cold climate classes, solar powered refrigeration devices have the potential to operate more efficiently than solar driven sorption due to advances in photovoltaic solar panel technology, but still require more energy than alternatives.


Introduction
Emergency, humanitarian, and military response operations have a need for rapid deployment of potable water systems, especially in locations impacted by infrastructure failure or lacking a readily available source. Across the world, approximately 4 billion people suffer from water scarcity at least one month of the year, which is projected to increase to 4.8-5.7 billion people living in a water-scarce area at least one month a year by 2050 [1,2]. With water availability increasingly scarce in many regions, there is a need for alternative water supply sources. Atmospheric water generators (AWGs) are devices that can produce potable water from the humidity naturally found in ambient air, offering a possible alternative freshwater source. Of the total volume of water found in the hydrosphere only 2.5% of it is freshwater, a significant portion of which is inaccessible due to being frozen or trapped in deep groundwater aquifers [3,4]. Of the accessible freshwater there is approximately 13 000 km 3 available in the atmosphere in the form of water vapor [3,4]. AWGs are a critical technology that enables access to this supply of atmospheric water at locations that lack access to either surface or groundwater resources.
As AWGs extract water from the ambient air, their energy consumption and water production efficiencies are highly dependent on environmental conditions with RH and temperature being the two largest factors. Historically, AWG devices have been plagued by high energy demand, and their dependence on weather conditions combined with low efficiency indicates that the technology requires further research [5]. The three main technologies utilized for atmospheric water harvesting are fog collection, direct cooling, and sorption-based systems. Fog harvesting collects freshwater using gravity through the condensation of water vapor from saturated air on a cold surface [6]. Refrigeration vapor compression cycle devices extract water from the atmosphere by directly cooling air to its dewpoint and then collecting the condensed water [7]. Solar driven sorption is a process through which water is absorbed through a desiccant and then released and desorbed later at high temperatures using solar energy [7].
There is significant interest in leveraging technologies to address water scarcity in arid environments, humanitarian operations, and for military applications. The scope of this study is evaluating the energy demands and performance of refrigeration vapor compression cycle and solar driven sorption AWG devices and their climate dependencies. Three commercially available AWG devices are assessed by geospatially mapping predicted mean monthly specific energy consumption (SEC), required photovoltaic solar panel array size to meet the maximum monthly energy demand, and estimated average annual daily water production. The results are then synthesized by Köppen-Geiger climate classification, to enable the findings to be used as objective metrics when considering the use of AWG technology based on climate region. The three devices analyzed are the AWG550 produced by AWG Contracting [8], the water generating systems WGS-900 [9], and the SOURCE Hydropanel [10]. The AWG550 and WGS-900 are both refrigeration-based systems while the SOURCE Hydropanel utilizes solar driven sorption. The Hydropanel is based on an adsorption desorption cycle in which both solar thermal and photovoltaic electricity are used to drive the process [10,11]. The final output depicts how SEC in kilowatt-hours per liter (kWh l −1 ) changes across the United States (U.S.) for the existing technology, addressing how the energy demand of AWGs changes across various climate conditions.

Background and literature review 2.1. AWG technology
There are multiple classes of AWG technology, each categorized by their energy source [12]. Passive devices require no energy input and function based on present environmental conditions to extract water. However, active devices require external energy to be applied to produce water. Active AWGs are capable of greater production than passive, and, in terms of renewable energy sources, solar driven active AWGs have the greatest prospect for wide adaptability [7].
As a passive technology, fog harvesting requires no energy input. In this process water aerosols rather than vapor are captured, so RH is required to be greater than or equal to 100% for a passive fog collection system to operate [6]. The production efficiency is dependent on weather conditions, the net filament utilized, and the characteristics of the receiving surface [5]. This results in the devices being geographically constrained based on RH requirements and incompatible with universal application [6,13]. Due to this limitation, only active AWG systems were considered by this analysis.
Direct cooling systems utilize energy inputs to lower air to its dewpoint and condense water vapor into freshwater [7]. Lowering air temperatures to their dew point is accomplished either by refrigeration using the vapor compression cycle or thermoelectric cooling. Thermoelectric coolers, also known as Peltier devices, work by forcing heat transport by application of a voltage difference between a semiconductor assembly [14], resulting in a decrease in temperature on one side and heating on the opposite of the assembly. As air passes over the cool side, the air is lowered below its dewpoint resulting in condensation of water vapor [15]. The advantage of thermoelectric cooling is that it does not require a compressor or refrigerants, but these systems are significantly less energy efficient than vapor compression cycle systems [16].
The AWG550 and WGS-900 devices considered in this analysis are refrigeration devices. Systems based on refrigeration and the vapor compression cycle have the highest yield in locations with hot and humid environments [17][18][19]. Studies have found that refrigeration based AWG devices exhibit an 'exponential-like' increase in energy requirement as RH approaches zero, and energy requirements are four times lower when RH is greater than 75% compared to when it is less than 40% [16]. However, regardless of environmental conditions, energy consumption is greater than conventional desalination methods, leading AWG technology to lack commercial competitiveness when a source of brackish or seawater is available [17].
For sorption devices, water is adsorbed by a desiccant material and desorbed using high temperatures ideally sourced from solar energy [7]. Adsorption is a phenomenon in which molecules adhere to a surface through chemical or physical interactions [13]. Alternatively, absorption may be utilized in which molecules of a substance dissolve into the bulk volume of another substance [20]. Solid adsorbents retain their structure during adsorption, but absorbents such as hygroscopic salts dissolve into an aqueous solution at RH less than 100% when the vapor pressure of the solution is less than the partial pressure of vapor in air [20]. Sorption devices are either diurnal and constrained to a daily cycle, or continuous operating in which they may conduct multiple cycles a day [7,12,21]. When a diurnal cycle is utilized water is typically captured at night during periods of greater RH and released during the day from heat provided by solar energy [7]. This adsorption processes is driven by vapor pressure differences between ambient air and the desiccant surface, and extra energy is applied to increase the vapor pressure of the desiccant surface during the desorption process to produce greater yields [7,21].
The desiccant material utilized for the sorption process is a factor in the performance of the device and the amount of energy required for desorption. The main types of sorbents consist of classic adsorbents, metal organic framework, biomimetic materials, silica gels, and hygroscopic salts and composite adsorbents [7,13,20]. Based on currently available materials, sorption devices produce low yields in comparison to refrigeration vapor compression cycle systems making them less economic due to the requirement for large scale implementation for a sorption device to produce enough water to meet typical demands [6,16,17].
Across literature, the current major research focus for sorption devices is to improve their commercial competitiveness through material selection and system design to enable continuous multi cycle operation and increase yield [11,22]. Metal organic framework materials are able to produce a viable amount of water in environments with RH as low as 10%, enabling sorption devices to operate outside the operating range of direct cooling devices [13,22,23]. The demonstrated performance of sorption-based systems to operate in low humidity environments is the main advantage in comparison to vapor compression-based technologies, as most geographic locations facing water scarcity are in arid environments [11,20,23].

Alternative water supplies
Alternative water supply solutions include desalination, contaminated water treatment, sewage recycling, and water production from air [5]. Compared to AWG, it is far more efficient to utilize desalination if there is a saltwater source available. The most commonly implemented forms of desalination are thermal, membrane separation, and humidification dehumidification based systems [16]. The energy required for desalination is dependent on the salinity of the source water as well as the technology utilized, excluding thermal desalination in which energy consumption in independent of salinity [24]. In thermal systems, energy is applied to drive the evaporation of seawater that is later condensed into freshwater and collected. The advantage of thermal systems is that they have high system reliability and low pretreatment requirements, but the drawback is that the systems are both cost and energy intensive [16]. Solar thermal interface desalination is a type of thermal technology that utilizes solar driven floating photothermal materials to generate a hot region and drive low-energy consumption water evaporation [25]. Solar thermal interface desalination does not require the combustion of fossil fuels, but is limited in production ability, so further research is required to increase its potential for industrialization [25].
Membrane methods to include reverse osmosis, pass water pressurized above its osmotic pressure through a membrane that denies the flow of salt molecules [16]. Reverse osmosis typically has a lower SEC than thermal desalination techniques, but membranes may leak small amounts of salt into the produced freshwater supply and do not function well when seawater concentrations are greater than 3.5% [16]. Solar thermal membrane desalination utilizes photothermal conversion nanomaterials to drive desalination based on traditional member desalination technology [26,27]. The technology has demonstrated the ability to provide a high salinity removal rate without the requirement for high temperatures, but there are currently limited studies that have been conducted to asses performance [26].
Humidification and dehumidification technologies work by drawing seawater into a chamber where the RH is increased up to as high as 100% and an atmospheric water harvester is utilized to produce freshwater [16]. As a result, the performance of a humidification and dehumidification system is limited by the characteristics of the atmospheric water harvester device utilized [16]. This causes poor energy efficiency in comparison to other widespread desalination methods.
Kwan et al conducted a comparative analysis of desalination and available active atmospheric water harvesting technologies, and concluded that it would be practically impossible for atmospheric water harvesting to become comparable to desalination in terms of ideal energy optimality [16]. Energy optimality was determined by the Gibbs free energy principle derived theoretical minimum SEC divided by specific energy exergy [16]. Among the AWG devices analyzed, refrigeration vapor compression cycle systems were the best performing and had a deviation in energy optimality ranging from 1.8%-12%, indicating that a vapor compression cycle AWG is the most feasible solution [16]. Despite the limitations of AWG devices, the benefit is that they can provide potable water in a viable amount at locations threatened by freshwater scarcity and lacking access to a brackish or seawater source. Desalination, energy requirements vary with the salinity of seawater, temperature, and recovery rate, and are much smaller than the energy demand by AWG technology. For comparison, the approximate SEC of reverse osmosis seawater desalination has been reported in literature ranging from 2.5 to 8.5 kWh m −3 (0.0025-0.0085 kWh l −1 ), and new emerging technology may be able to reduce it to as low as 2 kWh m −3 (0.002 kWh l −1 ) [5,11,20,25,[28][29][30][31][32]. The energy required for brackish water is lower with SEC ranging from 0.66 to 2.5 kWh m −3 (0.000 66-0.0025 kWh l −1 ) [24,29,31]. Additionally, 44% of the cost related to reverse osmosis desalination is associated with energy consumption [29]. Based on these metrics, for an AWG to system to become competitive with desalination, it would have to be able to achieve a minimum of at least 89.9% energy optimality which is not currently possible [16].

Atmospheric water feasibility
Gido et al introduced the moisture harvesting index (MHI) as a means to assess the energy requirements of atmospheric water harvesting by means of direct cooling, and it has been utilized as a metric by multiple studies [19,33]. MHI represents the energy fraction that is associated with actual vapor condensation, and is the ratio of energy invested in the desired water condensation process to the total energy invested in the cooling of the condensable as well as in condensable gasses in the air bulk [33]. The purpose of MHI is to determine the feasibility and cost effectiveness of atmospheric water harvesting by direct cooling at any location. Gido et al conducted a multicity site specific MHI analysis [33] and LaPotin et al geospatially mapped average global MHI for the months of January and July of 2010 [20]. The results depicted significant temporal variability in MHI, SEC, and water yield [20]. A limitation of these studies is that they are theoretically based and do not quantify the SEC of fieldable technologies currently used by commercial systems.
Current research efforts to geospatially map the performance of AWGs have been focused on addressing water scarcity and determining the potential water production of such technology based on climate conditions. Mulchandani et al conducted an analysis with the goal of identifying the relationship between atmospheric parameters and solid desiccant material properties to quantify geospatial and temporal water production potentials [6]. Lord et al mapped the global potential of a theoretical solar driven continuous mode atmospheric water harvesting device to assess specific yield in liters per kilowatt hour (l kWh −1 ) [12], documenting global spatially continuous evaluation of atmospheric water harvester performance based on theoretical thermodynamic limits. The study determined that climate conditions may be sufficient for a continuous mode atmospheric water harvester to provide potable water to one billion people that lack safely managed drinking water at a global scale [12].

Data processing
Two datasets required for this analysis include historical weather data and performance metrics for the three AWG devices. Weather data was obtained from a dataset sourced from AccuWeather covering the period from 1985 to 2019 [34], based on a compilation of national weather station data. The dataset consisted of 1938 individual weather stations. For each of the three devices, publicly available manufacturer specification sheets were used to obtain performance metrics and operating ranges required to generate performance based linear regression models [8][9][10]35]. Operating limits for the devices were based on temperature in degrees Celsius, RH, and water yield [8][9][10]35]. The metrics for each device were provided in varied formats.
For the WGS-900 device performance information regarding water production (l d −1 ) and power consumption (kW h −1 ) were directly provided in tabular format [9,35]. Data for the AWG550 was obtained from a diagram depicting device harvesting rate (l h −1 ) and SEC (kWh l −1 ) for various temperature and RH conditions [36]. The provided metrics for the devices were utilized when generating linear regression equations for both the power demand (kW) and yield in either liters per day (l d −1 ) or liters per hour (l h −1 ).
For the SOURCE Hydropanel, production (l d −1 ) and daily solar energy demand (kWh m −2 ) were provided by a panel production contour plot within the device technical specifications [10]. The units of energy demand incorporate area because the device operates directly from solar energy based on the nature of its design. Within the weather dataset, the product of total global horizontal irradiance (W m −2 ) and minutes of sunshine with appropriate unit conversion was used as the location specific solar energy (kWh (m 2 d) −1 ). Global horizontal irradiance was selected as the solar energy input source because it quantifies irradiance from a horizontal reference plane, and involves the fewest assumptions [12]. The SOURCE Hydropanel specifications were used when generating linear regression equations for yield in liters per day (l d −1 ).

Linear regression modeling
Regression modeling was the selected method for predicting device performance in terms of both nominal power demand and water harvesting rate. All data to support the regression models are found in the supporting information. Pokorny et al adopted a similar approach when utilizing a quadratic polynomial function to analyze a prototype mobile autonomous AWG designed for arid conditions [37]. However in this analysis, we found linear models to be an appropriate fit given the manufacturer's data. Linear models were generated for each device based on environmental inputs and were assessed for statistical significance and the required assumptions of a linear regression model to include normality and constant variance of the residuals. Statistical significance of the generated models was confirmed using an analysis of variance (ANOVA), normality of the residuals was evaluated via a Shapiro-Wilks test, and constant variance was checked by the Breusch-Pagan method. The alpha value was established as 0.05 for comparison to the calculated p-values. Additionally, multicollinearity of the independent variables was considered via the variance inflation factor (VIF) metric in comparison to an established threshold of 5. Models were validated via the k-fold method utilizing five folds. The original root mean square error for each model was compared to the root mean square error of the cross validated model, to ensure that there was not a significant change in model performance.
For the refrigeration devices to include the AWG550 and WGS-900, two linear models using the independent variables of temperature and RH were used to achieve statistical significance via the described ANOVA. Temperature and RH were selected as independent variables as they are the variables for which performance metrics were provided by manufacturer specifications. Additionally, the two parameters are the established industry standards for rating refrigeration AWGs. The first was established to predict nominal power demand (kW), and the second linear model projected production in either liters per day (l d −1 ) or liters per hour (l h −1 ). The ratio of predicted power demand and production was utilized to obtain the expected SEC.
For the sorption-based SOURCE Hydropanel, solar energy availability and RH were the independent variables utilized to generate a linear model to predict water production in liters per day (l d −1 ). These two independent variables were selected as they matched the technical specifications for the data provided by the Hydropanel manufacturer. In this case, temperature efficiencies were not reported and, instead, were replaced with solar energy availability. The SEC (kWh l −1 ) for each location was then able to be calculated by determining the ratio of available solar energy (kWh m −2 d −1 ) from the weather station dataset to the predicted water production (l d −1 ) and multiplying it by the surface area of the panel, equivalent to 2.88 square meters (m 2 ) [10].
Following validation, the models were paired with the datasets containing daily weather conditions between 1985 and 2019 to predict SEC and production, as well as solar panel array size requirements for the refrigeration devices. The time period was chosen as it coincides with an available dataset with minimal missing data points and was sufficiently long enough to demonstrate long-term averages. The data was consolidated for all locations, grouped by year, and then aggregated by both average calendar month and average year across the 35 year time frame. The results were constrained so that negative values were not generated, and metrics were not calculated during periods when environmental conditions were below minimum or above maximum operational limits. Additionally, constraints were placed to ensure device production did not exceed maximum device yield. Exclusive to the SOURCE Hydropanel, the operational range of the device was constrained by available solar energy between 4.58 and 6.66 kWh m −2 based on manufacturer specifications [10]. Values exceeding the maximum solar energy threshold were capped at a 6.66 kWh m −2 during calculations. Device specific operational limits are summarized in table 1. The data in table 1 are directly from the manufacturers specifications. There is a large difference in rated and maximum production for the WGS-900 device which is based on a theoretical maximum from the manufacturer. Additionally, the production rate of refrigeration systems is significantly greater than sorption devices, which is one of the major limitations of the latter technology.
When calculating mean performance, metrics calculated for days for which the devices did not reach operational conditions were represented as null values rather than zeros. The objective of the study was to determine performance trends when the devices are operational, so excluding nonoperational days prevented zero values from skewing calculated results when averaging them. Each calendar month across the 35 year period was considered in the analysis. If a device reached operational performance at least once in a calendar month for an individual year, it was included when the mean calendar month performance metrics across the full 35 year time frame were calculated. For example, the AWG550 device reached operational conditions in Teton County, Wyoming on a single day in January of 1990 and the calculated metrics were taken as representative of predictive performance of the device when reaching site specific operational capability during January.
To comparatively analyze solar panel array surface area requirements for the refrigeration devices, a photovoltaic conversion efficiency of 20% solar energy to electrical work input to the devices was utilized as representative of current commercial photovoltaic technology [38]. Lord et al utilized a similar approach when conducting an assessment of the potential for a solar powered sorption device to address global water scarcity [12]. For the investment in solar panels in relation to the quantity of water yielded to be understood, the average annual production was then paired with the solar panel array sizing, except for the SOURCE Hydropanel device.

Spatial analysis and visualization
The final dataset containing predicted SEC and yield for each device, as well as solar panel array requirements for the refrigeration devices was utilized to visualize the metrics across the continental United States using ArcGIS Pro. The calculated weather station location data for an average calendar month and year between 1985 and 2019 was spatially joined with the U.S. counties. In doing so the mean values for the metrics of interest were determined within an 80 km radius of each county. A similar geospatial analysis approach was utilized by Mulchandani and Westerhoff who selected a radius of 120 km when mapping the performance of a solar desiccant-based AWG [6].
To synthesize the results by climate region, the present day Köppen-Geiger climate classification map developed by Beck et al was utilized [39]. The map classifications were generated from threshold values and seasonal air temperature and precipitation data covering the time period of 1980-2016 with a 0.0083 • resolution [39]. The classification system consists of 5 main classes and 30 sub-types [39]. For this AWG analysis, the calculated mean annual device metrics for each individual weather station were spatially joined with the Köppen-Geiger climate classification classes. Next, for weather station locations where each device achieved year-round operational capability, summary statistics were calculated for both expected SEC and yield normalized by manufacturer rated production. Table 2 shows the results of the linear regression models that were generated to predict power demand and device production. The ratio of the results calculated by equations (1) and (2) were utilized to predict SEC for the AWG550 device, and the ratio of calculated results from equations (3) and (4) with appropriate unit conversion for the WGS-900. It is observable that for the two refrigeration devices the variation in temperature has a greater influence than RH on device performance due to the magnitude of the variable coefficients.

Regression analysis
The assessment results of the linear model assumptions of normality and constant variance of the residuals, as well as multicollinearity are provided in table 3. Each equation was found to be statistically significant and passed the evaluated tests with a p-value greater than the established alpha value of 0.05. Additionally, the VIF values for each equation indicated that multicollinearity among the independent variables was not present due to being less than the established threshold value of 5. Even with relatively small sample sizes, the regression models show similar root-mean-squared error between the original and k-folds models, indicating low sensitivity to new data and good predictive capabilities.

Spatial and seasonal analysis
The SEC for each device is shown in figures 1-3 for an average calendar month, where a lower value indicates greater energy efficiency and is preferred. Grouping the results by meteorological season allowed for observable trends to be identified. The AWG550 and WGS-900 devices are presented on the same scales to enable direct comparison of the results because both devices utilize refrigeration technology. The SOURCE Hydropanel is presented on an independent scale due to the magnitude of the results of the sorption technology in comparison to the refrigeration devices. The null value for the single county of Treasure County, Montana consistently reported across all figures does not indicate inoperability of the devices, rather it identifies a lack of weather station data within an 80 km radius.   1. AWG550 specific energy consumption shows the increased energy demand of the system in arid environments during throughout the year, but specifically in the spring and early summer seasons.
Of the refrigeration devices the AWG550 has the greatest range of environmental conditions for which it can produce water per the manufacturer specifications. This is favorable for the device and allows it to achieve some level of production during each average calendar month across the U.S. excluding a small northern region consisting of five counties in the states of Minnesota and North Dakota. The SEC of the device was found to range from 0.13-3.64 kWh l −1 and had a distinct trend of greater SEC in arid environments found in the American Southwest.
In comparison, the WGS-900 device has greater production capacity with a more restrictive range of operating conditions. This restriction resulted in the WGS-900 failing to achieve production for a significant portion of the U.S. for five calendar months out of the year. Nationwide, from May through October, the device achieved widespread production with low SEC and maintained greater performance consistency than the energy consumption of the AWG550 device. The calculated SEC value range for the WGS-900 device was 0.02-2.68 kWh l −1 . Therefore, when compared to the AWG550 device, it is important to balance production potential with the range of operational conditions when selecting a device. Common among the refrigeration devices, the peak SEC occurred in the month of April.
The SOURCE Hydropanel has the potential to achieve operational conditions year-round across the entire continental U.S., as it has the widest operational limits of the three devices evaluated. Peeters et al estimated that the specific yield for the SOURCE Hydropanel to range from 0.14-0.32 l kWh −1 under varied climate conditions [11]. This ratio equals a SEC range of approximately 3.13-7.14 kWh l −1 , which validates the range calculated in this study (figure 3) of 3.19-5.29 kWh l −1 . The results support findings across literature that sorption technology has a greater range of operational environments, but also has greater SEC in comparison to refrigeration devices in ideal environments [11,16]. As a result, figure 3 has a different scale for SEC than the two previous refrigeration devices.
Interestingly the performance trends of the SOURCE Hydropanel differ from those of the refrigeration devices, in that the device performance displays greater geospatial variability in SEC because it is dependent on not only RH, but available solar energy. Additionally, the peak SEC for the device occurred in the month of June rather than the month of April consistent among refrigeration devices. The trend of higher SEC in the American Southwest aligning with arid Köppen-Geiger climate classes was consistent across both technologies.

Feasibility of pairing solar power with refrigeration devices
The feasibility of pairing solar panels with refrigeration type devices to create independent off grid systems is shown in figure 4. The first step in powering a system by solar energy is determining the photovoltaic solar panel array surface area requirement. Only locations where the devices demonstrated the potential for year-round operation are depicted, and the geospatially representative values selected are those that would ensure that the maximum average calendar month energy demand for a single device would be sufficiently met by an appropriately sized solar array. This assumption prevents the need for seasonal variation in solar panel array requirements to meet the year-round energy demand of the devices. The surface area to power a single device for the AWG550 and WGS-900 technologies ranged from 31-1630 m 2 and 117-5290 m 2 , respectively. For comparison an official football pitch is equivalent to 7140 m 2 in surface area.
In addition to energy consumption, device yield varies with location specific climate conditions, so the mean annual daily production for each device is provided in liters per day (l d −1 ). This enables expected device yield to be considered when making decisions regarding infrastructure investments. Although year-round production and the ability to support device energy demand by solar panel technology may be achievable, it would be a poor allocation of resources should a selected system configuration be unable generate sustained water supply. From figure 4, the Gulf Coast and Mid-Atlantic regions are the most feasible for implementation of a solar powered refrigeration AWG device based both on land area requirements and production. A similar conclusion was reached in regard to the suitability of the gulf coast region for AWG water production by Mulchandani and Westerhoff [6].

SEC by climate region
Köppen-Geiger climate classes are widely used and have been geospatially determined on a global scale [39]. This analysis determined projected performance in terms of SEC and mean annual production normalized to device rated production across 16 Köppen-Geiger climate class subsets provided in the supporting information, table S1. The WGS-900 has the most restrictive operational limits of the three devices, and as a result metrics were only able to be obtained for 12 of the 16-climate class subsets due to the device failing to reach operational conditions for 4 classes. Of the five main Köppen-Geiger climate classes the polar class was the only one not present in the continental U.S. Figure 5 depicts the median value and interquartile range for the calculated metric of SEC by climate class. As expected, for the refrigeration devices, there is an observable trend of greater median SEC and interquartile range in both arid environments with low RH and cold environments with lower temperatures. These devices operate most efficiently in warm humid environments. On an annual scale, the refrigeration-based systems performed similarly in both tropical and temperate climate classes with limited variability in performance as depicted by the interquartile range values being close to the median value.
The findings for the SOURCE Hydropanel supported the findings in literature that SEC of sorption-based systems is higher, and a similar trend of increased SEC is displayed in both arid and cold environments. The key distinction is that the SOURCE Hydropanel inherently relies on solar energy for production and requires no additional supporting infrastructure to operate. The specific energy values for the refrigeration devices are based on the work supplied directly to each unit, and if sourced from solar energy accounting for a 20% solar panel efficiency would increase by a magnitude of 5. Applying this normalization, the median SEC of the Source Hydropanel is only exceeded by the AWG550 device in the desert cold and no dry season cold summer climate classes. Figure S1 in the supporting information furthers the climate region analysis by describing expected yield as a function of rated production for each device. The SOURCE Hydropanel exhibited the advantage of sorption-based systems by having the most consistent performance, in which the median production value exceeded the devices rated performance in 14 of the 16 climate classes with tight interquartile ranges indicating consistent performance. The lowest production occurred in arid environments including the two classes for which it did not exceed its rated production, but its median production value did not drop below 88% of its rating. In comparison the median production value of the AWG550 refrigeration device failed to reach its rated performance in any of the 16 climate class subsets. It operated the best in tropical environments with median values ranging between 81% and 83% of rated production, and in the remaining classes it performed between 26% and 65% of its production. The wide range of operating environments for which the AWG550 can function as a refrigeration device is favorable, but it is limited by its efficiency in producing water.
The environmental conditions were not met for the WGS-900 to operate in 4 of the 16 climate classes and is geographically constrained. However, it exceeded its production rating in tropical as well as the temperate no dry season hot summer climate classes, and nearly reached it in the arid hot steppe class at 99% production. Production ranged from 21% to 88% in the remaining climate classes, and the WGS-900 device had the widest interquartile ranges of the three devices indicating greater variation in performance.

Discussion
This empirical study evaluated and quantified the performance of commercially available AWG devices in different environments. The resultant spatial analysis enables a deeper understanding of the characteristics of refrigeration vapor compression cycle systems powered by photovoltaic solar panels in comparison to solar driven sorption. It is imperative that tradeoffs are understood when making device specific decisions regarding when and how to implement the technology.

Water-energy nexus
AWG SEC ranged from 0.02-3.64 kWh l −1 for the refrigeration systems analyzed, and 3.19-5.29 kWh l −1 for the solar driven sorption SOURCE Hydropanel. AWGs lack commercial competitiveness in comparison to other water sources and treatment methods such as conventional water treatment, ultraviolet disinfection, and reverse osmosis desalination with SEC values of 1.0 × 10 −5 -4.8 × 10 −4 kWh l −1 , 1.0 × 10 −5 -5 × 10 −5 kWh l −1 , and 2.25 × 10 −3 -8.5 × 10 −3 kWh l −1 respectively [24,32,40]. The appeal of AWG devices is their ability to produce potable water in environments that lack a viable water source. However, due to their poor energy efficiency, if other water sources are available, even desalination is more energy efficient with current technology. Therefore, it is only practical to leverage AWGs at locations facing water scarcity or to temporarily provide relief to a small population.
The voltage requirement of refrigeration systems capable of producing suitable quantities of water is notable. The smaller AWG550 system can operate at 240 V 50-60 Hz or 460 V 50-60 Hz per customer request, however the larger WGS-900 system requires a 480 V 60 Hz, or 415 V 50 Hz system that is typically considered an industrial voltage and exceeds the capacity of mobile systems such as solar fabrics. There is limited availability of commercial off the shelf solar panel systems capable of operating at industrial voltages, driving the requirement for a professionally customized system design. The first step of designing such a system is quantifying the location specific energy requirements and analyzing the environmental conditions to determine solar panel array sizing requirements as done in this study. The selected system components including the rated power generation and efficiency of the selected solar panels are also important considerations during system design.
The infrastructure investment required to support the necessary solar panel array as well as the opportunity cost associated with land use tradeoffs must be considered. The return of investment must be competitive with powering an AWG by traditional electrical grid or diesel generator through either energy cost savings or the advantage gained by having an off grid self-reliant system. Another tradeoff is that for high production AWG units, powering a unit by solar panels may require an otherwise mobile system to become fixed.

Implications and recommendations
The results of this analysis identify significant variation in AWG performance across time and space, even among devices of the same technology type. This drives a requirement for the development of effective benchmarking and analysis to govern decision making in terms of device selection and strategy regarding when and how to employ the devices. These benchmarks aid in determining their appropriate use to maximize the potential benefit and assess the feasibility of powering refrigeration devices with conventional energy sources, in comparison to photovoltaic solar panels or solar driven sorption. When done appropriately, AWGs can be used to enhance a community's water resilience by providing an alternative drinking source, especially in a contingency environment. While the findings provided in the study focus on the United States, the underlying framework for assessment would work in any region where the relevant climate parameters were available.
In policy development, a two-approach system is recommended. The first is for mobile systems that are to be deployed and operated at any location for humanitarian, emergency response, military, or similar operations where mobility and scale are the priority. The second is to address the implementation of AWGs to sustain a small, isolated population by providing infrastructure for a decentralized water source. Each application has different unique factors that must be considered to optimize performance.
The objective of the methodology of this study was to observe geospatial trends associated with device performance based on climatic conditions. As such the results are not intended to be utilized to make location specific infrastructure decisions as they were normalized across the entire land mass of the continental U.S. Its strength is that it produces a result of predicted device metrics by climate classification. Multiple device packages consisting of different devices and solar panel configurations could effectively be pre-established, and then selectively deployed based on the intended region of use and results of a device specific assessment for disaster recovery or remote water supply. In the analysis, we utilize 35 year climate averages for assessing system performance, ignoring the impact of climate extremes on system performance as it falls outside the scope of the study. Additionally, considering our approach investigating long-term averages and fluctuations on the intra-annual scale, we ignore sudden and dynamic changes in environmental factors. However, the same framework of analysis could theoretically still be applied to capture efficiencies of the system with sudden changes provided more data were available. In this case, it might be expected that there would be some non-linear functioning of the system, requiring a different modeling approach for prediction.

Conclusion
Across the continental U.S. refrigeration devices can produce viable amounts of water with consistently lower energy efficiency from late spring through early fall. It has been demonstrated that in ideal environments, it is feasible to power an AWG by photovoltaic panels based on energy consumption. Tradeoffs must be considered in terms of land area, solar infrastructure investment costs, and the loss of mobility. Refrigeration devices are less suited for arid and cold climates, where they perform less efficiently and produce less water. Solar driven sorption can consistently produce water across a greater range of environments to include arid environments albeit with lower efficiency and reduced productivity in comparison to non-arid climates. Additionally, the modeled sorption device have increased SEC from the refrigeration devices, adding additional complications to the technology.
From a policy perspective this indicates the potential for widespread use of the devices from late spring to early fall. Refrigeration should be used in locations where it is able to be operated more efficiently than sorption and utilize solar energy if the required financial and real estate resources to support the development of an array are available. Regionalized climate-based studies are appropriate for devices intended to be used as a mobile asset, but it is highly recommended to conduct a location specific assessment to verify performance if the intended use of the system is as a fixed infrastructure asset.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary information files).