Effect of the Type of Gas-Permeable Membrane in Ammonia Recovery from Air

Animal production is one of the largest contributors to ammonia emissions. A project, “Ammonia Trapping”, was designed to recover gaseous ammonia from animal barns in Spain. Laboratory experiments were conducted to select a type of membrane most suitable for gaseous ammonia trapping. Three types of gas-permeable membranes (GPM), all made of expanded polytetrafluoroethylene (ePTFE), but with different diameter (3.0 to 8.6 mm), polymer density (0.45 to 1.09), air permeability (2 to 40 L·min−1·cm2), and porosity (5.6 to 21.8%) were evaluated for their effectiveness to recover gas phase ammonia. The ammonia evolved from a synthetic solution (NH4Cl + NaHCO3 + allylthiourea), and an acidic solution (1 N H2SO4) was used as the ammonia trapping solution. Replicated tests were performed simultaneously during a period of 7 days with a constant flow of acidic solution circulating through the lumen of the tubular membrane. The ammonia recovery yields were higher with the use of membranes of greater diameter and corresponding surface area, but they were not affected by the large differences in material density, porosity, air permeability, and wall thickness in the range evaluated. A higher fluid velocity of the acidic solution significantly increased—approximately 3 times—the mass NH3–N recovered per unit of membrane surface area and time (N-flux), from 1.7 to 5.8 mg N·cm−2·d−1. Therefore, to optimize the effectiveness of GPM system to capture gaseous ammonia, the appropriate velocity of the circulating acidic solution should be an important design consideration.


Introduction
Animal production is one of the largest contributors to ammonia emissions (NH 3 ) [1] due to poor waste management. Ammonia is implicated in particulate formation (PM 2.5) with adverse effects on human health [2]. Ammonia also contributes to ecosystem degradation when it is deposited on land or water [3] with corresponding soil acidification and eutrophication of surface water bodies [4].
In 2016, the agricultural sector of the EU-28 was responsible for 92% of the total ammonia emissions in the region because of the volatilization of livestock excreta [5]. In Spain, according to the National Emissions Inventory , agricultural activities produced 96% of the ammonia emissions. In 2014 and 2015, the National Emission Ceilings for the NH 3 (353 kt·year −1 ) were exceeded by 7% [6]. According to EU Directive 2016/2284/EU [7], Spain must reduce the NH 3 emission ceiling by 3% during the period 2020-2029 and by 16% by the year 2030.
Mechanical ventilation is considered a basic control method to eliminate gaseous ammonia from the inside of livestock production barns [8] to ensure the health of workers and animals [9] and the animal production performance [10].
The application of gas-permeable membranes (GPM) for capturing ammonia has been tested, especially in liquid [11,12]. Methods designed for the capture and recovery of N as a resource are the most optimal [13][14][15]. Conservation and recovery of N are important in agriculture due to the high cost of commercial ammonia fertilizers [16]. In this way, it would contribute positively from both an environmental point of view (decreased ammonia emissions to the atmosphere) and an economic point of view (recovered ammonium to replace commercial fertilizers of nitrogen source). Furthermore, the advantages of gas-permeable membrane technology over other N recovery technologies are, among others, that it does not require the use of additives [17], and it has low energy consumption in relation to other methods of ammonia recovery [18].
The GPM process consists in the flux of ammonia gas through the microporous hydrophobic membrane by diffusion. This ammonia is captured in an acidic solution circulating inside the membrane. As shown in Equation (1), once in contact with the acidic solution, the NH 3 gas combines with the free protons of the acid to form non-volatile ammonium ions (NH 4 + ). When sulfuric acid is used in the process, the product is ammonium sulfate. Sulfuric acid is generally used as a source of acid to capture ammonia because of its lowest cost among inorganic acids. However, the process is also effective using other inorganic acids (nitric, phosphoric), organic acids (citric, lactic), and their precursors [19]. Additionally, ammonium sulfate (AS) may have some potential agronomic and environmental benefits compared with ammonium nitrate (AN) by creating a more acidic root rhizosphere that increases the availability of soil P, and by reducing denitrification in soil and N 2 O greenhouse gas emissions [20]. Therefore, it can be an adequate substitute of mineral fertilizers as a nitrogen source and valuable fertilizer.
For gas separation and recovery, organic hydrophobic gas-permeable membranes (GPM), especially expanded polytetrafluoroethylene (ePTFE), are preferred due to lower transference resistance, hydrophobic characteristics, organic resistance, and chemical stability with acidic solutions [21,22]. The final mass of NH 3 captured in the acidic solution depends on the concentration of NH 3 gas in the atmosphere, which depends on the pH and the TAN (total ammonia nitrogen, NH 3 + NH 4 + ) concentration of the emitting solution [23], pH of the acidic solution [24], and acidic solution flow rate [25]. In animal manures, NH 3 and NH 4 + are in equilibrium depending on the pH and the temperature.
The ammonia dissolves at the source and/or is emitted. At pH below 7, little of the ammonia is undissociated and present as dissolved gas in liquid mixtures [26], for example, only 0.36% at pH 6.8 and temperature 25 • C [27]. At higher pH, a higher concentration of the undissociated, free NH 3 is instantly produced (i.e., 26.4, 78.2, and 97.3% at pH 8.8, 9.8, and 10.8, respectively). These conditions favor NH 3 permeation through the membrane where an acidic solution circulates [28,29]. With a pH < 2, the acidic solution dissolves the NH 3 , transforming it into an ammonium salt [11]. Most ammonia capture applications with gas-permeable tubular membranes have been performed in the liquid state (effluents). An EU project, "Ammonia Trapping (AT)", was designed to recover gaseous ammonia from animal barns in Spain. The main objective of the AT project was to reduce NH 3 gas emissions from the atmosphere of swine and poultry farms by using gas-permeable ePTFE membranes and capturing the N directly from the air. Targets in AT project were a reduction in the NH 3 concentration > 70%, and flux rates of ammonia trapping of 1.3 g N m −2 d −1 . The goal of this study was to determine the efficiency of the different gas-permeable tubular membranes to capture ammonia from the air. The results of this laboratory study helped in selection of the materials before a larger on-farm pilot evaluation, especially given the difference in costs of these membranes. To avoid variations in ammonia emissions, among treatments each experiment used the same synthetic Environments 2019, 6, 70 3 of 12 N emitting solution. In addition, the pH of the acidic solution used for ammonia capture was kept below 2.

Experimental Design
Airtight chambers were used to recover gaseous ammonia using the method of Szogi et al. [19]. The experiment used three chambers (volume = 25 L), one for testing each type of gas-permeable membrane ( Figure 1). The lids of the chambers were sealed. A tank (volume = 11 L) containing 1 L of a synthetic N emitting solution was placed inside of each chamber. Tubular gas-permeable membranes were suspended in the air in the chambers. The membranes were connected to an acidic solution reservoir that contained 1 L of an acidic N capturing solution (1 N H 2 SO 4 ). Peristaltic pumps (Minipuls 2, Gylson, USA or Perimax 12, Spetec, Germany) recirculated the acidic solution in a closed loop [16] between the inside of the tubular membranes and the acidic solution reservoir. The membranes were made of expanded polytetrafluoroethylene (ePTFE), but with different characteristics: ZM, ZM4, FZM (ZEUS Industrial Products Inc., Orangeburg, SC, USA), and PM (PRODYSOL Company) ( Table 1).   Two experiments were conducted. In experiment 1, three conditions were evaluated. The same flow rate of the acidic solution (1.25 L h −1 ) was applied to the three treatments. Two treatments used membranes with contrasting characteristics (ZM and PM) and with equal surface area (282.3 cm 2 ). The third treatment used a membrane (FZM) with smaller diameter and lower surface area (66 cm 2 ) ( Table 1). Experiment 2 was conducted to verify the effect of the fluid velocity on ammonia flux. Four conditions were evaluated. The four treatments used the same membrane surface (125.7 cm 2 ).
Two membrane types with different diameter were tested (PM and ZM4), and each membrane type received two acid flow rates (0.83 and 1.5 L h −1 ).
A synthetic solution was used in both experiments as the source of NH 3 emission (instead of organic waste). In the first experiment, the N emitting solution contained 59.4 g L −1 NH 4 Cl, 108.5 g L −1 NaHCO 3 , and 10 mg·L −1 N-allylthiourea, and in the second experiment, the N emitting solution contained 24.6 g L −1 NH 4 Cl, 43.2 g L −1 NaHCO 3 , and 10 mg·L −1 N-allylthiourea. N-allylthiourea (98%) was added as a nitrification inhibitor, following the strategies presented in other assays [22,30].
Two repetitions were made with each treatment tested. Samples (7.5 mL) of the N emitting synthetic solution were collected every two days and samples (5 mL) of the N trapping acidic solution were collected every day. The room temperature was constant (20.0 ± 1.0 • C).

Methodology for Analyses
Temperature ( • C), pH, and TAN concentration (mg·L −1 ) were monitored in the acidic solutions and the N emitting solution. The weight variations were controlled in the acidic solution reservoir, taking into account the 5 mL of sample extracted.
The control of pH was realized in the N trapping acidic solution and the N emitting synthetic solutions: pH of the acidic solution was maintained at < 2 and the pH of the synthetic solution at > 8 [11,12,30]. pH modifications were not required because these conditions were not reached.
The pH was measured with a Crison GLP22 pH meter (Crison Instruments S.A., Barcelona, Spain). The ammonium analysis was performed with distillation (UDK 140 automatic steam distillation unit, Velp scientific), capture of distillate in borate buffer, and subsequent titration with 0.2 N HCl [31].
The internal surface morphology of the membranes ( Figure 2) was analyzed by scanning electron microscopy (SEM) in the Advanced Microscopy Unit of the University of Valladolid. The SEM images were obtained using a FEI QUANTA 200F device (FEI Co, USA). The pore size distribution (pores/m 2 ), porosity, and water and air permeability were measured using porosimetry equipment (Coulter Porometer II) [32]. The surface sizes of the FPM and ZM4 membranes were not suitable for porosimetry and permeability analyses, so these measurements were obtained for ZM and PM membranes only.
Data were analyzed by means and standard deviation. Linear regression analyses were used to quantify changes of weight of the acidic solution and N capture rates. Data related to N mass removed, N mass recovered, and N flux were subjected to ANOVA (SAS Institute, 2008) [33].

Mass Flow Calculation
Mass flow (J) of NH 3 -N or N flux (mg N·cm −2 ·d −1 ) from the air into the acidic solution was calculated based on the N mass captured per day and the surface area of the GPM tubing using Equation (2), where C is the concentration of NH 4 -N in the acidic solution (mg·L −1 ), V is the volume of the acidic solution (L), S the contact surface of the membrane (cm 2 ), and t the time (d).

Characterization of Membranes
Measurements of the number of pores, porosity, water and air permeability, and MFP (mean flow pore) for the membranes ZM and PM are shown in Table 2. Table 2. Values of number of pores, porosity, water permeability, and MFP (mean flow pore) of ZM and PM membranes in experiment 1. Compared with the PM membrane, the ZM membrane had a lower density (0.45), a higher number and size of pores, and a higher porosity and permeability ( Table 2). This result can also be verified by the SEM images in Figure 2, which indicate that the pore size was greater in ZM, followed by PM, and finally FZM. Measurements of the number of pores, porosity, water and air permeability, and MFP (mean flow porous) for the membranes ZM and PM are shown in Table 2. Table 2. Values of number of pores, porosity, water permeability, and MFP (mean flow porous) of ZM and PM membranes in experiment 1. Compared with the PM membrane, the ZM membrane had a lower density (0.45), a higher number and size of pores, and a higher porosity and permeability ( Table 2). This result can also be verified by the SEM images in Figure 2, which indicate that the pore size was greater in ZM, followed by PM, and finally FZM. The images correspond to the inner surface of the tubular membranes. Images A and B were taken with 5000× magnification and the scale bar is equivalent to 20 μm in length. Image C were taken with 10,000× magnification and the scale bar is equivalent to 10 μm in length.

Variation of the Weight of the Acid Solution
The acidic solution decreased in weight in the three GPM systems (Figure 3). Total weight losses of the acidic solution at the end of the experiment were 11 ± 2% for ZM (R2 = 0.89), 10 ± 4% for PM (R2 = 0.95), and 5 ± 1% for FZM (R2 = 0.99). Weight losses in all cases were related with an evaporation process as leaks were not observed. The rate of water weight loss (g·d −1 ) for each type of membrane was: 16.1 ± 6.1 g·d −1 for PM, 17.8 ± 2.9 g·d −1 for ZM, and 7.5 ± 1.2 g·d −1 for FZM. The hydrophobic nature of the membrane prevents the penetration of the acidic solution into the membrane pores, creating a liquid/vapor interface at each pore entrance. If a vapor partial pressure difference across the membrane is established, vapor transport across the membrane takes place [34]. The images correspond to the inner surface of the tubular membranes. Images A and B were taken with 5000× magnification and the scale bar is equivalent to 20 µm in length. Image C were taken with 10,000× magnification and the scale bar is equivalent to 10 µm in length.

Variation of the Weight of the Acidic Solution
The acidic solution decreased in weight in the three GPM systems (Figure 3). Total weight losses of the acidic solution at the end of the experiment were 11 ± 2% for ZM (R2 = 0.89), 10 ± 4% for PM (R2 = 0.95), and 5 ± 1% for FZM (R2 = 0.99). Weight losses in all cases were related with an evaporation process as leaks were not observed. The rate of water weight loss (g·d −1 ) for each type of membrane was: 16.1 ± 6.1 g·d −1 for PM, 17.8 ± 2.9 g·d −1 for ZM, and 7.5 ± 1.2 g·d −1 for FZM. The hydrophobic nature of the membrane prevents the penetration of the acidic solution into the membrane pores, creating a liquid/vapor interface at each pore entrance. If a vapor partial pressure difference across the membrane is established, vapor transport across the membrane takes place [34].
Environments 2019, 6, 70 6 of 12 of the acidic solution at the end of the experiment were 11 ± 2% for ZM (R2 = 0.89), 10 ± 4% for PM (R2 = 0.95), and 5 ± 1% for FZM (R2 = 0.99). Weight losses in all cases were related with an evaporation process as leaks were not observed. The rate of water weight loss (g·d −1 ) for each type of membrane was: 16.1 ± 6.1 g·d −1 for PM, 17.8 ± 2.9 g·d −1 for ZM, and 7.5 ± 1.2 g·d −1 for FZM. The hydrophobic nature of the membrane prevents the penetration of the acidic solution into the membrane pores, creating a liquid/vapor interface at each pore entrance. If a vapor partial pressure difference across the membrane is established, vapor transport across the membrane takes place [34]. Majd et al. [23] also observed volume losses in acid traps due to evaporation, with values between 1 to 2 mL·d −1 in suspended systems, with an acid volume of 190 mL and a flow rate 3 times lower than that used in this experiment.
The rate of weight loss of the acid was not affected by membrane density, porosity, and permeability; however, it was affected by surface area. Weight loss was higher in the two membranes (ZM and PM) with the larger diameter and surface area, even though they had different porosity and density, and the weight loss was lower in the FZM membrane with smaller diameter and surface area. Therefore, the greater surface area resulted in higher vapor transport across the membranes and acid weight loss.

Process pH in the N Capturing Acidic Solution and N Emitting Synthetic Solution
The pH values reached in the acidic solution for each type of membrane were 0.4 ± 0.1 for PM, 0.4 ± 0.1 for ZM, and 0.4 ± 0.1 for FZM. In all cases, the pH values reached in the acidic solution remained below 2, indicating that enough H + ions were available to react continuously with NH 3 [25], forming an ammonium salt.
The initial pH values in the N-emitting synthetic solution for each type of membrane were 8.74 ± 0.06 for PM, 8.76 ± 0.01 for ZM, and 8.78 ± 0.13 for FZM. Corresponding final pH values at day 7 were 8.27 ± 0.04, 8.45 ± 0.01, and 8.63 ± 0.06. In all cases, the pH of the synthetic solution was maintained above 8, which favored the emission of free NH 3 [30].

Effect of the Type of Membrane on Ammonia Capture
The total NH 3 -N mass emitted by the synthetic solution was similar in the three membrane systems: 5381 ± 451 mg N for PM membrane, 5260 ± 514 mg N for ZM, and 4764 ± 606 mg N for FZM (Table 3). Table 3. Mass of NH 3 -N removal, NH 3 -N recovered by gas-permeable membranes, and N-flux with varied polymer density, surface area, and acidic solution velocity (experiment 1). Corresponding percent N removals were 46 ± 4%, 45 ± 4%, and 41 ± 5%. Similarly, no differences were observed in the total mass of NH 3 -N present in the synthetic solutions at the end of the experiment (Figure 4). The ammonia emission rate of the synthetic solution varied with time. There was a higher emission rate on the first day and a lower and almost constant emission rate in later days. For example, rates of emission the first day were 4138 ± 47 mg NH 3 -N·d −1 for PM, 3555 ± 433 mg NH 3 -N·d −1 for ZM, and 3342 ± 463 mg NH 3 -N·d −1 for FZM, and afterwards the rates of emission were 207 ± 83 for PM, 284 ± 158 for ZM, and 237 ± 24 mg NH 3 -N·d −1 for FZM. This emission behavior was also observed by Rothrock et al. [23] who noted that in the first 7 days, the concentration of NH 4 -N present in the synthetic source solution decreased faster, from 500 mg to 300 mg approximately. In contrast, from days 7 to 21, the concentration only decreased from 300 mg to 200 mg. The high recovery observed on the first day could be due to the high concentration of ammonium in the synthetic solutions. This generates a high concentration of ammonia in the gas phase. After the first day, a significant percentage of ammonium had been eliminated and, therefore, the driving force for transport decreased.  The ammonia emission rate of the synthetic solution varied with time. There was a higher emission rate on the first day and a lower and almost constant emission rate in later days. For example, rates of emission the first day were 4138 ± 47 mg NH3-N·d −1 for PM, 3555 ± 433 mg NH3-N·d −1 for ZM, and 3342 ± 463 mg NH3-N·d −1 for FZM, and afterwards the rates of emission were 207 ± 83 for PM, 284 ± 158 for ZM, and 237 ± 24 mg NH3-N·d −1 for FZM. This emission behavior was also observed by Rothrock et al. [23] who noted that in the first 7 days, the concentration of NH4-N present in the synthetic source solution decreased faster, from 500 mg to 300 mg approximately. In contrast, from days 7 to 21, the concentration only decreased from 300 mg to 200 mg. The high recovery observed on the first day could be due to the high concentration of ammonium in the synthetic solutions. This generates a high concentration of ammonia in the gas phase. After the first day, a significant percentage of ammonium had been eliminated and, therefore, the driving force for transport decreased.

Type of Membrane
The masses of NH3-N recovered in the acidic N trapping solution were 3628 ± 27 mg, 3407 ± 49 mg, and 2661 ± 307 mg for ZM, PM, and FZM, respectively. At similar emission and capture conditions, the NH3-N mass recovered by FZM was lower due to a lower surface area compared to the membranes ZM and PM. The surface area was 4.2 times higher for PM and ZM compared to FZM. Surprisingly, the mass of NH3-N recovered was not affected by large differences in material density (0.45 to 0.95 g/cm 3 ) between PM and ZM (Table 3), or by differences in porosity (5.6 to 21.8%), air permeability (2 to 10 L min −1 cm −1 at 1 bar pressure), and wall thickness (0.8 to 1.2 mm). This was surprising because it is logical to think that higher NH3 capture should be obtained with higher membrane porosity and air permeability, and with smaller wall thickness [35]. However, in the range tested in the experiment, these characteristics did not affect mass of NH3 recovered by the membranes.
In all membrane systems, NH3-N accumulation in the acidic solution during the 7-day experimental period was linear ( Figure 5). Capture rates (mg NH3-N d −1 ) were calculated based on the slope of the linear regressions; they were higher with PM and ZM (487 ± 71 and 518 ± 4 mg NH3-N d −1 , respectively) with larger diameter and surface area, compared to FZM (380 ± 44 mg NH3-N d −1 ) with smallest diameter and surface area. The acidic solution had a similar composition of 0.3 ± 0.1% of nitrogen and 0.4 ± 0.1% of sulphur.  The masses of NH 3 -N recovered in the acidic N trapping solution were 3628 ± 27 mg, 3407 ± 49 mg, and 2661 ± 307 mg for ZM, PM, and FZM, respectively. At similar emission and capture conditions, the NH 3 -N mass recovered by FZM was lower due to a lower surface area compared to the membranes ZM and PM. The surface area was 4.2 times higher for PM and ZM compared to FZM. Surprisingly, the mass of NH 3 -N recovered was not affected by large differences in material density (0.45 to 0.95 g/cm 3 ) between PM and ZM (Table 3), or by differences in porosity (5.6 to 21.8%), air permeability (2 to 10 L·min −1 cm −1 at 1 bar pressure), and wall thickness (0.8 to 1.2 mm). This was surprising because it is logical to think that higher NH 3 capture should be obtained with higher membrane porosity and air permeability, and with smaller wall thickness [35]. However, in the range tested in the experiment, these characteristics did not affect mass of NH 3 recovered by the membranes.
In all membrane systems, NH 3 -N accumulation in the acidic solution during the 7-day experimental period was linear ( Figure 5). Capture rates (mg NH 3 -N d −1 ) were calculated based on the slope of the linear regressions; they were higher with PM and ZM (487 ± 71 and 518 ± 4 mg NH 3 -N d −1 , respectively) with larger diameter and surface area, compared to FZM (380 ± 44 mg NH 3 -N d −1 ) with smallest diameter and surface area. The acidic solution had a similar composition of 0.3 ± 0.1% of nitrogen and 0.4 ± 0.1% of sulphur. The NH3 recovery (%) for each type of membrane was calculated based on the relationship between the NH3-N mass recovered (final content of NH3-N in the trapping solution) and the NH3-N mass removed (difference between the initial and final content of NH3-N in synthetic solution). Percent recoveries were not different (p ≤ 0.05): PM = 63%, ZM membrane = 69%, and FZM = 57%. The percent recoveries were not quantitative (100%) probably because the rapid release of NH3 in the first day of the experiment exceeded the capacity of the membrane. Other authors such as Rothrock et al. [23] obtained similar results than under conditions of an NH3 emission flush. They achieved recoveries of NH3-N of 67.7%, 73.6%, and 76.2% with hydrated lime addition treatments of 0.4 w/v, 2 w/v, and 4 w/v to 300 g of poultry litter. Therefore, design of the membrane manifolds should consider possible situations of rapid release that may occur in filed situations such as disinfection of manure with alkali compounds.
On the other hand, when the NH3-N capture is expressed on a surface area basis (N-flux, Table  3), the results show additional insight on the best operating conditions for the membranes. The Nflux obtained in the membrane FZM with the smaller diameter (5.8 ± 0.7 mg N·cm −2 ·d −1 ) was significantly higher-approximately 3 times higher-compared with the N-flux obtained with the larger diameter membranes (1.8 ± 0.0 and 1.7 ± 0.2 mg N·cm −2 ·d −1 with ZM and PM, respectively). Rothrock et al. [23] also observed higher N-fluxes in membranes with a smaller diameter (1.37 g·m −2 ·d −1 N-flux for a membrane i.d. of 4.0 mm and acid flow 70-80 mL d -1 and 0.7 g·m −2 ·d −1 N-flux 0.70 for a membrane i.d. of 8.8 mm and same flow). Majd and Mukhtar [25] observed an N-flux of 0.2 g·m −2 ·d −1 in a suspended membrane system. However, higher ammonia fluxes have been obtained when the membranes were directly submerged in the liquid (liquid-liquid) instead of being suspended in the air (air-liquid). For example, Daguerre et al. [36] obtained N-fluxes of 7.1 to 8.9 g·m −2 ·d −1 placing the membrane manifold in liquid swine manure (4940 mg NH4-N L −1 ), and Fillingham et al. [37] obtained N-fluxes up to 51.0 g·m −2 ·d −1 using synthetic wastewaters containing 6130 NH4-N L −1 and NaOH to pH 8.5.
In this study (experiment 1), the same recirculation flow of the acidic solution (1.25 L h −1 ) was used with the three membranes with outside diameters ranging from 3 to 8.6 mm (inner diameters 1 to 7). As a result, the smaller diameter resulted in a higher fluid velocity inside the membrane (2653.9 cm min −1 ) compared to the higher diameter membranes (54.2 to 69.0 cm min −1 ) (Table 3) and a more frequent renovation of the acidic solution in the submerged membrane manifold.
The Reynolds number (Re) is used in fluid dynamics to describe the character of the flow (flow is laminar when Re < 2300 and viscous forces are dominant characterized by smooth fluid motion, and flow is turbulent when Re > 3000 and it is dominated by inertial forces and vortices). Although the fluid flow was laminar in all three cases (Re 64 to 415, Table 3), the higher fluid velocity and Re in FZM resulted in a higher N-flux. Therefore, to optimize the effectiveness of the ePTFE membranes to capture gaseous ammonia, the fluid velocity should be an important design consideration because this study showed that the efficiency can be increased 3 times with changes in acidic solution velocity. The NH 3 recovery (%) for each type of membrane was calculated based on the relationship between the NH 3 -N mass recovered (final content of NH 3 -N in the trapping solution) and the NH 3 -N mass removed (difference between the initial and final content of NH 3 -N in synthetic solution). Percent recoveries were not different (p ≤ 0.05): PM = 63%, ZM membrane = 69%, and FZM = 57%. The percent recoveries were not quantitative (100%) probably because the rapid release of NH 3 in the first day of the experiment exceeded the capacity of the membrane. Other authors such as Rothrock et al. [23] obtained similar results than under conditions of an NH 3 emission flush. They achieved recoveries of NH 3 -N of 67.7%, 73.6%, and 76.2% with hydrated lime addition treatments of 0.4 w/v, 2 w/v, and 4 w/v to 300 g of poultry litter. Therefore, design of the membrane manifolds should consider possible situations of rapid release that may occur in filed situations such as disinfection of manure with alkali compounds.
On the other hand, when the NH 3 -N capture is expressed on a surface area basis (N-flux, Table 3), the results show additional insight on the best operating conditions for the membranes. The N-flux obtained in the membrane FZM with the smaller diameter (5.8 ± 0.7 mg N·cm −2 ·d −1 ) was significantly higher-approximately 3 times higher-compared with the N-flux obtained with the larger diameter membranes (1.8 ± 0.0 and 1.7 ± 0.2 mg N·cm −2 ·d −1 with ZM and PM, respectively). Rothrock et al. [23] also observed higher N-fluxes in membranes with a smaller diameter (1.37 g·m −2 ·d −1 N-flux for a membrane i.d. of 4.0 mm and acid flow 70-80 mL d −1 and 0.7 g·m −2 ·d −1 N-flux 0.70 for a membrane i.d. of 8.8 mm and same flow). Majd and Mukhtar [25] observed an N-flux of 0.2 g·m −2 ·d −1 in a suspended membrane system. However, higher ammonia fluxes have been obtained when the membranes were directly submerged in the liquid (liquid-liquid) instead of being suspended in the air (air-liquid). For example, Daguerre et al. [36] obtained N-fluxes of 7.1 to 8.9 g·m −2 ·d −1 placing the membrane manifold in liquid swine manure (4940 mg NH 4 -N L −1 ), and Fillingham et al. [37] obtained N-fluxes up to 51.0 g·m −2 ·d −1 using synthetic wastewaters containing 6130 NH 4 -N L −1 and NaOH to pH 8.5.
In this study (experiment 1), the same recirculation flow of the acidic solution (1.25 L h −1 ) was used with the three membranes with outside diameters ranging from 3 to 8.6 mm (inner diameters 1 to 7). As a result, the smaller diameter resulted in a higher fluid velocity inside the membrane (2653.9 cm min −1 ) compared to the higher diameter membranes (54.2 to 69.0 cm min −1 ) (Table 3) and a more frequent renovation of the acidic solution in the submerged membrane manifold.
The Reynolds number (Re) is used in fluid dynamics to describe the character of the flow (flow is laminar when Re < 2300 and viscous forces are dominant characterized by smooth fluid motion, and flow is turbulent when Re > 3000 and it is dominated by inertial forces and vortices). Although the fluid flow was laminar in all three cases (Re 64 to 415, Table 3), the higher fluid velocity and Re in FZM resulted in a higher N-flux. Therefore, to optimize the effectiveness of the ePTFE membranes to capture gaseous ammonia, the fluid velocity should be an important design consideration because this study showed that the efficiency can be increased 3 times with changes in acidic solution velocity. The study used two recirculation rates (0.83 and 1.5 L h −1 ), and two membrane types with different diameters (id 2.9 and 6.2 mm) but the same surface area (125.5 cm 2 ) ( Table 4). As a result, the fluid velocity inside the membranes gradually increased in the range of 49 to 315 cm min −1 and Re varied from 49 to 155. These modest differences in fluid velocity and Re (within laminar flow) significantly affected both the mass of NH 3 -N recovered in the acidic N trapping solution, and the N-flux (recovery per surface area). Figure 6 shows the relationship between N flux vs Re obtained using combined data from all seven treatments in experiments 1 and 2. It confirms that velocity of the circulating acidic solution should be an important design consideration to optimize the effectiveness of GPM system to capture gaseous ammonia.  Environments 2019, 6, x FOR PEER REVIEW 9 of 12 A second experiment was done to further evaluate the positive effect of fluid velocity on N-flux. The study used two recirculation rates (0.83 and 1.5 L h −1 ), and two membrane types with different diameters (id 2.9 and 6.2 mm) but the same surface area (125.5 cm 2 ) ( Table 4). As a result, the fluid velocity inside the membranes gradually increased in the range of 49 to 315 cm min −1 and Re varied from 49 to 155. These modest differences in fluid velocity and Re (within laminar flow) significantly affected both the mass of NH3-N recovered in the acidic N trapping solution, and the N-flux (recovery per surface area). Figure 6 shows the relationship between N flux vs Re obtained using combined data from all seven treatments in experiments 1 and 2. It confirms that velocity of the circulating acidic solution should be an important design consideration to optimize the effectiveness of GPM system to capture gaseous ammonia.

Conclusions
Gas-permeable membranes (GPM) made of ePTFE were effective for the recovery of gaseous NH3 using a closed-loop system. A pH < 2 in the circulating acidic solution and a pH > 8 of the synthetic emitting solution were favorable for the process. At similar emission and capture conditions, the mass of NH3-N recovered by tubular GPMs was significantly increased by surface area, which was related to differences in the membrane diameter tested. However, the mass of NH3-N recovered was not affected by large differences in GPM material density (0.45 to 0.95 g/cm 3 ), porosity (5.6 to 21.8%), air permeability (2 to 10 L min -1 cm -1 at 1 bar pressure), and wall thickness

Conclusions
Gas-permeable membranes (GPM) made of ePTFE were effective for the recovery of gaseous NH 3 using a closed-loop system. A pH < 2 in the circulating acidic solution and a pH > 8 of the synthetic emitting solution were favorable for the process. At similar emission and capture conditions, the mass of NH 3 -N recovered by tubular GPMs was significantly increased by surface area, which was related to differences in the membrane diameter tested. However, the mass of NH 3 -N recovered was not affected by large differences in GPM material density (0.45 to 0.95 g/cm 3 ), porosity (5.6 to 21.8%), air permeability (2 to 10 L·min −1 cm −1 at 1 bar pressure), and wall thickness (0.8 to 1.2 mm). A higher fluid velocity of the acidic solution significantly increased (approximately 3 times) the N-flux (mass N recovered per unit of surface area and time). Therefore, to optimize the effectiveness of the GPM system to capture gaseous ammonia, the fluid velocity is an important design consideration because this study showed that the efficiency can be increased 3 times with changes in acidic solution velocity. Funding: The authors gratefully acknowledge funding by the European Union under the Project Life + "Ammonia Trapping" (LIFE15-ENV/ES/000284) "Development of membrane devices to reduce ammonia emissions generated by manure in poultry and pig farms". Mention of trade names or commercial products in this article is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the USDA.

Conflicts of Interest:
The authors declare no conflict of interest.