Improving aeration systems in saline water (part II): effect of different salts and diffuser type on oxygen transfer of fine-bubble aeration systems

The objective of the present study is to investigate the different effects on the oxygen transfer of finebubble aeration systems in saline water. Compared to tap water, oxygen transfer increases due to the inhibition of bubble coalescence. In Part I of the present study, we investigated in laboratoryscale experiments the effect of design of diffuser membrane. The objective of Part II is the assessment of effects of different salts, diffuser type and diffuser density. We measured the concentration of various salts (MgCl2; CaCl2; Na2SO4; NaCl; KCl) above which coalescence is fully inhibited and oxygen transfer reaches its maximum (referred to as the critical coalescence concentration; CCC). For this purpose, we developed a new analytical approach, which enables investigation of the coalescence behaviour of any aeration system and (mixed) salt solution quickly and easily by evaluating the results of oxygen transfer tests. To investigate the transferability to large scale and the effect of diffuser type and density, we repeated lab-scale experiments in a 17,100 L pilot-scale test tank and carried out additional tests with tube and plate diffusers at different diffuser densities. The results show that despite the higher pressure drop, diffusers with dense slit density and smaller slits are to be recommended in order to improve efficiency of aeration systems in saline water.


INTRODUCTION
Today, in biological wastewater treatment plants, mainly fine-bubble aeration systems are used to satisfy the oxygen demand of microorganisms in activated sludge (Wagner & Stenstrom ). Air is introduced via diffusers installed at the bottom of the aeration tank. From the rising air bubbles, the oxygen is transferred to the liquid phase. There are different diffuser types available (discs, plates, tubes) made out of various materials. Oxygen transfer from ascending air bubbles to the liquid phase is described by the volumetric mass transfer coefficient (k L a), which represents the product of the liquid-side mass transfer coefficient (k L ) and the liquid/gas interfacial area (a). The k L a mainly depends on water quality parameters (e.g. salt concentration, temperature) and diffuser design. Today, mainly some industrial wastewaters show increased salt concentration (Lefebvre & Moletta ; He et al. ). Nevertheless, the global trend towards saving fresh water, e.g. through (industrial)-wastewater reuse or by using seawater for toilet flushing (Sander ), will raise the volume of saline wastewater that needs to be (biologically) treated. Depending on the origin of the wastewater, the dissolved salt and salt mixture will change. Therefore, a deeper understanding of the effects of increased concentration of different salts and salt mixtures is crucial to enhance the (energy) efficiency of aerobic biological wastewater treatment.
For fine-bubble aeration systems, increased salt concentration (c Salt ) reduces k L but increases a due to the inhibition of bubble coalescence. The increase is enough to result in a net increase of k L a (Baz-Rodríguez et al. ). The increase of k L a in saline water (SW) is described by the f S value, which is the ratio between k L a in SW to k L a in tap water (TW). In Part I of the present work, we could show in laboratory-scale experiments with different conventional disc diffusers by combining bubble size measurement and oxygen transfer tests, that the length and density of the slits in the diffuser membrane influences f S (Behnisch et al. ). In this case, the smaller the bubbles detaching from the diffuser (primary bubbles) the higher is f S . In the early 1980s, Zlokarnik () investigated the coalescence behaviour of different aeration devices commonly used at that time. He also concluded that the more the coalescence is inhibited, the more advantageous it is to use aeration devices, which produce small primary bubbles. Sander et al. () carried out oxygen transfer tests at varying sea salt concentrations with disc diffusers in a pilot-scale test tank. They found the simple relationship f S ¼ k · c Salt þ 1, whereby k mainly depends on the airflow rate per disc diffuser. Nevertheless, f S does not increase infinitely. At a certain c Salt , the coalescence is completely inhibited and f S reaches its maximum ( f S,max ). This concentration is called the critical coalescence concentration (CCC) and is specific for each salt (solution) and bubble formation system (Cho & Laskowski ; Firouzi et al. ). Sovechles & Waters () and Quinn et al. () determined the CCC of varying salts in a lab-scale flotation cell by measuring the bubble size (given as Sauter mean diameter (d 32 )) via image analyses of the bubble swarm at different c Salt . Due to the inhibition of bubble coalescence with increasing c Salt , d 32 decreases linearly (Zone 1) up to CCC (Grau et al. ). Here (Zone 2), d 32 reached its minimum (d 32,min ), called the quasi-staticbubble diameter or initial bubble diameter (Marrucci & Nicodemo ; Sovechles & Waters ). According to Grau et al. (), the point at which the fitted lines for Zone 1 and Zone 2 meet defines the CCC. There are other methods to investigate the coalescence of a system than with image analyses. Craig et al. () measured the coalescence by detecting the change of light intensity from an expanded beam of light, which was passed through the bubble swarm produced by a sinter at the base of the transparent test column. Prince & Blanch () used a model-based approach to predict CCC for different salts using test results of Marrucci & Nicodemo () and Lessard & Zieminski (). While Marrucci & Nicodemo () also applied image analyses to determine the bubble size in a bubble swarm produced by a bronze porous plate, Lessard & Zieminski () observed bubble pairs detaching at the end of two adjacent capillaries. Similar to CCC, some researchers also use the transition concentration (TC X ) to quantify the grade of decrease of bubble size with increasing c Salt . The index X indicates the grade at which the bubble size reduces X% from that in pure water to the constant bubble size at high salt concentration. Accordingly, TC 100 and CCC are synonymous (Sovechles & Waters ). Currently, there is no definitive agreement explaining the inhibiting effect of salts on bubble coalescence. Firouzi et al. () gives a detailed review of all proposed theories, which include colloidal forces, gas solubility, Gibbs-Marangoni effect, surface rheology and ion specific effects.
Experiments show that bubble formation, as well as bubble size, affect CCC (Firouzi et al. ; Sander ). Since the bubbles in a flotation cell, as used by Sovechles & Waters (), are very small and are formed by cavitation and not by a diffuser, the validity of the results for finebubble diffusers is questionable. Furthermore, to investigate the influence of salt on oxygen transfer, previous studies used only disc diffusers (Sander et al. ; Behnisch et al. ) or porous plates and frits (Marrucci & Nicodemo ; Zlokarnik ), which are no longer used today for wastewater treatment. Additionally, only the influence of NaCl and sea salt has been investigated so far. Therefore, we determined for the first time CCC for different salts with conventional fine-bubble diffusers. Because measuring bubble size in a bubble swarm is laborious, we transferred the method of Grau et al. () for determining CCC to our results of oxygen transfer tests. To do so, we took advantage of the fact that f S behaves similarly to d 32 with increasing c Salt , only with reversed sign. The f S value increases linearly instead of a linear decrease in d 32 .
In order to investigate the transferability of the results to large scale as well as the influence of the diffuser type and diffuser density on f S , we also carried out oxygen transfer tests with disc, tube and plate diffusers in a 17,100 L fully glazed steel frame tank at c Salt between 0 and 20 g/L NaCl. Finally, we show the large improvement of aeration in SW by comparing specific standard oxygen transfer efficiency (SSOTE) and standard aeration efficiency (SAE) with TW conditions, and give some additional design considerations.

Membrane diffusers
Six different conventional membrane diffusers (2 disc diffusers, 2 plate diffusers, 2 tube diffusers) with different slit densities (SD) and slit lengths (d S ) were used (Table 1). SD is defined as the number of slits relative to the active (perforated) membrane area (A A ). In the tube diffusers used here, perforation is only on the side of the diffuser and the top and bottom are not perforated (Figure 1). For plates and discs, A A is similar to the projected media surface area (A P ). Depending on the percentage of perforated membrane area, for tube diffusers A A ≫ A P applies. According to Behnisch et al. (), diffuser density (DD) is the total membrane area (including unperforated area) divided by area of the tank floor.

Measuring salt concentration (c Salt )
Determining the exact salt concentration during the tests, especially in the large pilot-scale test tank, faced us with some challenges. Large amounts of salt were needed to achieve the desired concentrations (up to 342 kg for a single test). Additionally, the tests for each salt or diffuser type required several days and a regular addition of TW was necessary to compensate for evaporation losses. To ensure constant experimental conditions, an on-line measurement of salinity was necessary. Therefore, we decided to apply a method commonly used in environmental technology and determined c Salt by measuring the conductivity (EC) on-site standardized at 20 C water temperature with standard conductivity cell (TetraCon ® 324, WTW, Germany).
The correlation between EC and c Salt is usually expressed by a simple linear equation: c Salt ¼ t · EC. However, the relationship between EC and c Salt is not always linear and depends on the activity of specific dissolved ions, ionic strength (IS) and the average activity of all ions in the liquid (Rusydi ). Therefore, before starting oxygen transfer tests, we measured EC of salt solutions with a known concentration in a 500 mL beaker. Our results show that the relationship in the chosen concentration ranges is always linear. This can be seen, for example, in the calibration line for MgCl 2 in Figure 2(a). Therefore, the calibration line is given by the following formula: EC ¼ b 1 · c Salt þ b 0 , b 0 being the EC of TW. In all our tests, the EC of TW was 0.7 mS/cm. With the known calibration lines, we were able to calculate the salt concentration by measuring EC on site during oxygen transfer tests. When measuring EC in aerated tanks, care must be taken since air bubbles can lead to incorrect measurements.

Measuring CCC with oxygen transfer tests in laboratory-scale experiments
To measure CCC for different salts, oxygen transfer tests were performed with Disc 1 in a rectangular tank (0.5 m × 0.5 m) with a water depth of 1.0 m, using the pure oxygen desorption method according to EN - (). With this method, the oxygen concentration is increased 15-20 mg/L beyond the oxygen saturation concentration by aerating with pure oxygen gas or oxygen-enriched air (Wagner et al. ). By switching to aeration with ambient air, the oxygen concentration starts to decrease again until the saturation concentration is reached. From the curve of decreasing oxygen concentration, the k L a for a specific water temperature (T ) is calculated (k L a T ) by nonlinear regression. Four electrochemical oxygen probes (Oxymax COS51D, Endress þ Hauser, Germany) were installed to record dissolved oxygen concentration. Finally, the k L a was standardized to 20 C (k L a 20 ) according to the equation: k L a 20 ¼ k L a T ·1.024 (20-T) .
Tests were performed at different airflow rates (2.0, 3.5 and 4.5 m 3 /h) and salt concentrations. The airflow rate was measured at standard temperature and pressure (0 C; 101.3 kPa; 0% humidity) with a thermal flow sensor (TA16, Hoentzsch, Germany). The d 32,min was only measured during tests with NaCl by image analyses (Behnisch et al. ). Several experiments show that regardless of salt type, the minimum bubble size is always the same (Lessard & Zieminski ; Quinn et al. ). The first test was carried out in TW, then salt was added.  Figure 2(b) shows how CCC was determined using the example of test results with MgCl 2 . According to the method we adapted from Grau et al. (), the CCC results from intersection of the fitted lines of Zone 1 (linear increase of f S ) and Zone 2 (constant f S at f S,max ). Accordingly, CCC for MgCl 2 (CCC MgCl2 ) is 0.065 mol/L or 6.2 g/ L. To verify the reproducibility of the measurement, we repeated the experiment with a new disc diffuser of the same design. In the second run, we measured a CCC of 0.061 mol/L or 5.8 g/L, which represents a deviation of 6%. Therefore, the reproducibility of the results was good.

Oxygen transfer tests in pilot-scale test tank
The pilot-scale test tank is a fully glazed steel frame tank (L/W/H ¼ 3.0 m/1.5 m/4.0 m) with a water level of 3.8 m, yielding a water volume of 17.1 m 3 . Ten test series with different diffuser types ( Figure 1) and diffuser densities were carried out in the test tank. Therefore, 32 or 6 disc diffusers, 10 or 6 tube diffusers and 4 plate diffusers were installed in the glass tank, respectively. Diffuser depth of submergence was on average 3.65 m and varied negligibly between the individual diffusers.
The oxygen transfer was measured as before in the labscale experiments by using the desorption method according to EN - . As before during lab-scale experiments, dissolved oxygen in water is measured using four electrochemical oxygen probes (Oxymax COS51D, Endress þ Hauser, Germany) installed at different positions in the test tank. The compressed air is produced by means of a positive displacement blower (GMA 10.0, Aerzener Machine Factory, Germany). To measure the airflow rate precisely, a calibrated rotary-piston gas meter (Aerzener G 65, type ZB 039.0, Aerzener Machine Factory, Germany) with manometer and thermometer was installed downstream of the blower. The airflow rate (Q A ) was normalized at standard temperature and pressure conditions (0 C; 101.3 kPa; 0% humidity). Q A is indicated per aerated tank volume (Q A,VAT ¼ Q A · (tank volume) À1 ), per disc diffuser (q A,Disc ¼ Q A · (number of disc diffusers) À1 ), per plate diffuser (q A,plate ¼ Q A · (number of plate diffusers) À1 ), per length of tube diffuser (q A,tube ¼ Q A · (number of tube diffusers · length of single tube diffuser) À1 ) or per slit of diffuser membrane (q A,slit ¼ Q A · (number of diffusers · slits per diffuser) À1 ).
Tests were performed at different airflow rates and NaCl-concentration (c NaCl ). The airflow rate set during the tests depends on operational range of diffuser specified by the manufacturer and the capacity of the blower. As described before, c NaCl was calculated by measuring EC. Also, the standard oxygen transfer rate per aerated tank volume (SOTR VAT ¼ SOTR · (aerated tank volume) À1 ) was calculated, SOTR being the standard oxygen transfer rate normalized to 20 C water temperature and atmospheric pressure of 101.3 kPa (EN - ). Here it must be noted, that the solubility of oxygen (c S ) decreases with increasing c Salt . The oxygen saturation concentration for different temperatures and salt concentrations can be taken from tables, to be found e.g. in ISO  . According to ASCE/EWRI - , c S in SW (c S,SW ) can also be estimated with known c S in TW (c S,TW ) and c Salt (in g/L) as follows: c S,SW ¼ (1.0-0.01 · c Salt ) · c S,TW . However, when comparing c S,SW calculated according to the given formula and the tabulated values, we found an increasing deviation with increasing c Salt (see Annex 1). Therefore, we have adjusted the formula to calculate c S,SW as follows: c S,SW ¼ (1.0-0.0059 · c Salt ) · c S,TW.

Laboratory-scale experiments
In Table 2 determined dependencies between EC and salt concentration are listed. The coefficients of determination (R 2 ) close to 1.0 prove the linear dependence between EC and c Salt within the chosen concentration range up to maximum selected salt concentration (c max ). As previously mentioned, the relationship is influenced by many factors, and therefore the equations shown here are only valid in TW for the specified concentration ranges. To calculate the salt concentration in solutions with other background contamination (i.e. dissolved substances other than the salt to be measured), the equation may have to be adapted. Results in Table 3  bubble size was not measured and is therefore not available (n.a.). However, since similar fine-bubble diffusers were used, we can assume that similar bubble sizes were achieved as in our tests.
A direct comparison of the CCC reported in the different papers is not possible due to the different measurement methods and test conditions. Nevertheless, some tendencies can be identified. For example, with the exception of Na 2 SO 4 , the CCCs reported by other authors are higher than our results. Only the CCC determined by Prince & Blanch () are lower, except for KCl. A possible correlation can be seen here with the observed bubble size, which is larger in the case of Prince & Blanch () and smaller in the case of the other papers than in our own experiments. Already, Firouzi et al. () and Sander () showed that the smaller the bubbles the higher the CCC of a salt. If only the CCC is considered as the molar concentration (i.e. in mol/L) of the salts investigated in all the different studies, the following order emerges for most  (Marrucci & Nicodemo (1967)) or 'limiting bubble size' (Sovechles & Waters (2015)). of the studies reported here: MgCl 2 < CaCl 2 < Na 2 SO 4 < NaCl < KCl Already, Sovechles & Waters () showed that CCC decreases for salts containing multivalent ions. While 1:1 (cation:anion charge) salts had the highest CCC; 1:2 and 2:1 salts had intermediate CCC; and 2:2, 3:1 and 3:2 salts had the smallest CCC. Therefore, they try to describe the inhibition of coalescence by salt (and multicomponent salt solutions) by calculating IS. In analogy to CCC, they postulated that coalescence is completely inhibited in their flotation cell when IS ¼ 0.28, which they called Critical-Coalescence Ionic Strength (CCIS). To verify this, all results in Table 3 were supplemented by the IS we calculated. ISvalues of salt solution by Prince & Blanch () varies between 0.13 and 0.62. Also for our test results, IS fluctuates between 0.18 and 0.27. This variation is very large, which is why we cannot confirm the approach postulated by Sovechles & Waters () to describe coalescence inhibition by determining the IS. Indeed, the CCC must be determined for each new (bubble formation) aeration system individually.
So even if the absolute values of the determined CCC do not match those from the literature, which is due to the different bubble size, it can still be said that they are in a specific order relative to each other, independent of the aeration system and the measurement method. The validity of the determined CCC and thus of the new measurement method is therefore given.

Experiments in pilot-scale test tank
In the following, first we will present the results of the tests in the pilot-scale test tank for each individual diffuser type. Then we try to consider all the results together and discuss the different effects on f S more in detail. At the end, we will discuss other parameters that have to be taken into account in design of aeration systems in SW in addition to k L a (or f S ).

Results from disc diffuser
In Figure 3  1 f S increases up to 10 g/L NaCl, which corresponds to the CCC of NaCl (CCC NaCl ) previously determined in the labscale experiments. Then f S reaches its peak value ( f S,max ). The same is true for Disc 2, although the transition point from Zone 1 (linear increase) to Zone 2 (constant value) is not as obvious as with Disc 1. The pilot-scale experiments show, that for both disc diffusers, the level of f S,max is strongly dependent on q A,Disc . The highest f S values could be reached in the middle of the operation range (3.0-4.0 Nm 3 /h per diffuser) given by the diffuser manufacturer (1.5-8.0 Nm 3 /h per diffuser) which corresponds to the results of Sander et al. (). In case of Disc 2 at very low q A,Disc of 0.2 m 3 /h/disc (Figure 3(d)), f S is only about 1.0. This means that no salt effect can be observed at such low q A,Disc values. The reason for this will discussed below in the present paper, when discussing the influence of airflow rate on f S .
Comparing the results of both disc diffusers shows that higher f s,max values were reached with Disc 1 than with Disc 2. Considering the results when the full range of airflow rate was tested (Figure 3(a) and 3(b)), mean f S,max of Disc 1 (2.0) is 10% higher than of Disc 2 (1.8). In comparison, k L a 20 -values in TW are in a comparable range for both diffusers (Annex 2). This confirms the results of previous lab-scale experiments with similar disc diffusers (Behnisch et al. ). Accordingly, higher f S values could be reached with smaller slit lengths (Disc 1 superior to Disc 2). Nevertheless, the effect of airflow rate on f S was not as obvious as in the experiments conducted in the pilot-scale test tank, because of the limited blower capacity and reactor volume during the lab-scale experiments. Consequently, a scale effect has to be taken into account in order to obtain adequate results during measurement of oxygen transfer in SW. A sufficient reactor volume as well as blower capacity is necessary to ensure that the full range of airflow rate of diffusers can be set during the tests. If only two diffusers have to be compared relative to each other, a small reactor is sufficient, whereby the test conditions must be identical.
If both the f S value and the k L a value rises, so does the SOTR VAT . While in TW there is a linear relationship between SOTR VAT and Q A,VAT (Behnisch et al. ), this is not the case in SW due to the dependence of f S (or k L a) on q A,Disc . At a salt concentration of 12 g/L NaCl and a Q A,VAT ¼ 1 Nm 3 /m 3 /h, almost similar SOTR VAT of 123 g/m 3 /h and 132 g/m 3 /h could be achieved with 6 and 32 Disc 1 diffusers respectively. This represents an increase of 24% (99 g/m 3 /h) and 45% (91 g/m 3 /h) compared to 6 and 32 Disc 2 diffusers respectively. Compared to SOTR VAT in TW it is an increase of 105% (6 diffusers: 60 g/m 3 /h) or 60% (32 diffusers: 84 g/m 3 /h). The results show that due to the influence of q A,Disc on f S , it is possible that in SW with the same Q A,VAT but a different number of diffusers, due to the higher f S -value (or k L a), similar or even more oxygen can be transferred at low than at high diffuser density.

Results from plate diffuser
In Figure 4, f S -values from test series with plate diffusers are plotted against c NaCl for different airflow rates. Unexpectedly, there are big differences compared to the results of the disc diffusers. For Plate 1 (Figure 4(a)), f S increases up to a concentration of 5 g/L NaCl, which is quite lower than the measured CCC NaCl in lab-scale experiments. Furthermore, the measured f S,max for Plate 1 is lower than for Plate 2 (Figure 4(b)), even though Plate 1 has smaller slits. In comparison, k L a 20 -values in TW are in a comparable range for both diffusers (Annex 3). In addition, the clear dependency between airflow rate and f S observed for the disc diffusers cannot be observed for Plate 1, in contrast to Plate 2 (see Figure 4(b)). Here, the f S -value increases with increasing airflow rate without reaching a noticeable peak, which is consistent with the results of the tests with 32 disc diffusers. However, as before with the disc diffusers, the operational range for Plate 2 (25-35 m 3 /h/plate) specified by manufacturer was not fully reached due to the limited blower capacity. Therefore, it can be expected, similar to the tests with the 32 disc diffusers, that with increasing airflow rate f S would first increase further, reach a maximum and then decrease again. If the f S value is plotted against the airflow rate, this is confirmed, as we will show below.

Results from tube diffuser
In Figure 5, f S -values from test series with tube diffusers are plotted against c NaCl for different airflow rates. For both tube diffusers, f S increases up to 10 g/L NaCl as with the disc diffusers until f S,max is reached. Therefore, the results confirm the CCC NaCl determined in the lab-scale experiments once again. The present data show that the average f S,max of Tube 1 (2.1) is 10% higher than that of Tube 2 (1.9), regardless of the number of installed diffusers. No effect of the number of installed diffusers (or diffuser density) on f S , in THE case of tube diffusers, was detected. In comparison, k L a 20 -values in TW are comparable for both diffusers (Annex 4). Thus, the higher f S values could be reached with the diffusers with smaller slits, which corresponds to the results of disc diffusers.
At a salt concentration of 13 g/L NaCl and a Q A,VAT ¼ 1 Nm 3 /m 3 /h, a SOTR VAT of 156 and 182 g/m 3 /h could be achieved with 6 and 10 Tube 1 diffusers respectively. This corresponds to an increase of 7% (146 g/m 3 /h) and 9% (167 g/m 3 /h) using Tube 2 diffusers. Compared to SOTR VAT in TW it is an increase of 79% (6 tube diffusers: 87 g/m 3 /h) and 84% (10 tube diffusers: 99 g/m 3 /h). Therefore, for tube diffusers, SOTR VAT can always be increased by a higher number of installed diffusers independent of q A,tube . However, as in TW, it is expected that the increase of SOTR VAT is minimal when DD exceeds a specific value. Behnisch et al. () observed in TW only minor improvement in oxygen transfer for tube diffusers, when DD exceeds 35%.

Dependency of f S on airflow rate
To illustrate the dependency of f S,max and airflow rate for the different diffuser types, f S as a function of the airflow rate is plotted in Figure 6. For a simplified presentation, only results of measurements are shown, where c NaCl is higher than CCC NaCl and f S is constant. The airflow rate is given as the airflow rate per slit and slit length (q A,dS ¼ q A,Slit · (d S ) À1 ). This enables a comparison between the different diffusers, which have different membrane design properties (i.e. d S , SD, number of slits). It should be noted, however, that this does not take into account the elasticity of the slits, which expand depending on different material properties of the membrane (e.g. deflection, flexibility, thickness) with increasing q A,Slit (Loubière & Hébrard ). However, the expansion is very small despite high airflow rates as shown in experiments with multi-orifice diffuser membranes (Painmanakul et al. ). Therefore, the expansion of the slits with rising airflow rate is ignored here. Figure 6 illustrates the difference between the individual diffusers. While for tube diffusers q A,dS is very low (0.1-0.5 dm 3 /h/mm) and for disc diffusers and Plate 2 moderate (up to 2 dm 3 /h/mm), q A,dS for Plate 1 is very high (1.7-9 dm 3 /h/mm). The reason for this is the combination of the very low number of slits of Plate 1 (low SD) with relatively high q A,Plate values. Therefore, for a better illustration of the results of the tube and disc diffusers as well as Plate 2 diffuser, in Figure 6 the range q A,dS from 0 to 2 dm 3 /h/mm is shown enlarged.
As mentioned before, Figure 6 shows that there is no dependency between f S and airflow rate for tube diffusers within the considered range of airflow rate. A slight effect on f S can be seen for Plate 1. In contrast, the effect is very strong for disc diffusers and Plate 2. For both disc diffusers as well as for Plate 2, the highest f S values were reached at ∼1 dm 3 /h/mm. If the airflow rate changes, f S decreases again. Sander et al. () attribute the dependency between f S -value and airflow rate to the fact that at low airflow rates, coalescence also occurs less frequently in TW because of the isolated bubble rise behaviour. Baz-Rodríguez et al. () also observed no dependence between k L a and c Salt when gas hold-up or airflow rate is very low and the bubbles are relatively far from each other. The decrease of f S at very high airflow rates is explained by an increase in initial bubble size (Sander et al. ). Several experiments show that the initial bubble size increases with rising airflow rate ( When looking at the results of Plate 1, it has to be taken into account that the SD of Plate 1 is very low (0.88 slits per cm 2 ) compared to Plate 2 (5.5 slits per cm 2 ) and Disc 1 and Disc 2 (15.5 and 10 slits per cm 2 ). This results in distances between the slits (P R /d S and P S /d S ; see Table 1) being up to nine times larger than for the other diffuser types. To prevent coalescence close to the diffuser (in TW), Painmanakul et al. () recommend a distance between the slits in relation to the bubble size of 1.0 (i.e. P R /d 32,min and P S / d 32,min > 1.0). The d 32,min value of Plate 1 was not measured. If we assume a d 32,min between 1 and 2 mm, this ratio will be exceeded by more than four to nine times in the case of Plate 1. Therefore, no or less coalescence occurs also in TW. In addition, q A,dS increases very rapidly with increasing q A,Plate and so does the initial bubble size. Both effects lead to the fact that the coalescence behaviour as well as the bubble size distribution might differ fundamentally from the other diffusers. However, the slight influence of airflow rate on f S for Plate 1 and the lower CCC cannot be explained finally on the basis of the present results. Therefore, further experiments are necessary.
For the missing dependency of f S on airflow rate for tube diffusers, there are in our opinion three possible explanations, which we will discuss in more detail below.
• The material: Unlike the plate and disc diffusers made of EPDM, the tube diffusers are made of silicon. However, the material properties only influence the initial bubble size (Loubière & Hébrard ) but should not affect the coalescence behaviour of the rising bubble. Hence, the missing dependency of f S values on airflow rate cannot be explained by the membrane material.
• Low q A,dS values: The number of slits per tube diffuser is much higher than for the other two diffuser types. Therefore, maximum q A,dS for tube diffusers (0.49 dm 3 /h/mm) is very low compared to disc diffusers (1.9 dm 3 /h/mm) and Plate 2 diffuser (1.6 dm 3 /h/mm). Initial bubble size should therefore change only slightly. The decrease of f S observed for disc diffusers and Plate 2 with increasing airflow rate is therefore missing. In contrast to the maximum q A,dS , the achieved minimum q A,dS compares well with values for disc and tube diffusers (∼0.14 dm 3 /h/mm). The missing decrease of f S with decreasing airflow rate can therefore not be explained by different q A,dS values.
• The coalescence behaviour: as described previously, for disc diffusers f S decreases at low airflow rates because coalescence also occurs less frequently in TW due to the isolated bubble rise behaviour. The slits of tube diffusers are on the side of the diffuser and therefore the bubble formation occurs tangentially to the diffuser membrane and not vertically, as for disc and plate diffusers. Furthermore, this arrangement of the slits makes the cross-section of the assumed flow channel along the rising bubble swarm much thinner than with disc and plate diffusers (Figure 1). While the cross section of the flow channel for discs and plates roughly corresponds to the perforated membrane area, this is not the case for tubes. Here, the cross section of the flow channel corresponds to the projected surface area of the tube diffuser (A P ¼ 665 cm 2 per diffuser), which is much smaller than the active (perforated) membrane area (A A ¼ 1,450 cm 2 per diffuser). If the number of slits is related to A P of the tube diffuser, the resulting quasi-slit-densities (26 and 33 slits/cm 2 ) are more than twice as high as the SD of disc and plate diffusers. Therefore, it is obvious that for tube diffusers, even at relatively low airflow rates, the bubbles are not ascent isolated and coalesce continuously in TW. This results in high f S values even at low airflow rates.

Dependency of f S on diffuser density
After discussing the effect of airflow rate on f S , we checked if there was any influence of the number of installed diffusers (or DD) on f S . According to results from Figure 5(a) and 5(c) as well as Figure 5(b) and 5(d), there is no effect of the number of installed diffusers on f S in the case of tube diffusers. Regardless of the number of installed diffusers, the same f S,max was achieved as CCC NaCl was reached. In the case of disc diffusers, a comparison between the different diffuser densities is more difficult, because of the overlapping influence of the airflow rate. Considering individual results of the disc diffusers in Figure 6, the same f S could be achieved with both 6 and 32 disc diffusers for the same disc diffuser and the same q A,dS (or q A,Disc ¼ (q A,dS ) · d S · slits per diffuser). Hence, for 32 disc diffusers, the increase of f S with increasing q A,dS is identical to that for 6 disc diffusers. Due to the high number of diffusers and limited blower capacity, the range of tested airflow rates with 32 disc diffusers is significantly below the tests with 6 disc diffusers or the given specified operating range from manufacturers. If q A,dS and q A,Disc continue to increase, a similar dependency between f S and q A,dS (or q A,Disc ) will most likely result as was shown for the tests with 6 disc diffusers. Therefore, no influence of diffuser density on f S for disc diffusers as well as for tube diffusers was detectable. For plate diffusers only one diffuser density was tested. A comparison is therefore not possible. The result that diffuser density has no effect on f S promotes the notion that bubbles only coalesce with bubbles from the same diffuser or adjacent slits and not with bubbles from other diffusers. This confirms the observations of Hasanen et al. () and Behnisch et al. (), that coalescence occurs mainly close to the diffuser.
Summarize effects on f S The results shown above correspond to the results from Part I of this paper: Both c Salt as well as the design of the diffuser membrane (see Table 1) have an influence on k L a or f S in SW (Behnisch et al. ). Hence, at high c Salt diffusers with smaller detaching bubbles show higher f S and k L a values. Additionally, we could show that the type of the diffuser also influences the oxygen transfer in SW. For tube diffusers, f S only depends on c Salt , while for disc and plate diffusers f S also depends on q A,Disc and q A,Plate respectively. Except Plate 1, for all diffusers f S,max was reached when c Salt-! CCC. Furthermore, diffuser density does not seem to affect f S .

Additional design considerations
Other factors also have to be taken into account for design of aeration systems in SW. An important parameter is c S , which decreases with increasing c Salt (Annex 1). Due to the lower concentration gradient between gas and liquid phase and reduced molecular diffusivity, k L is expected to decrease, resulting in a lower f S and k L a. When looking at the results of plate and tube diffusers (Figures 4 and 5), we did not observe a significant decrease in f S after reaching its peak value in the considered concentration range. Because of the lower range of tested salt concentration, a clear conclusion is not possible for disc diffusers. Nevertheless, we do not expect a significant decrease of f S for disc diffusers with further increasing c Salt . This is confirmed by the results of Baz-Rodríguez et al. (), who found no dependence of k L a, a and k L on c Salt when coalescence is completely inhibited. They show that the reduction of the molecular diffusivity with increasing c Salt and the effect on k L is marginal. According to these authors, the main effect on k L is the slip velocity, which is substantially influenced by the bubble size (i.e. d 32 or a). However, the reduced c S also decrease SOTR VAT and SSOTE, as shown in Figure 7(a). Similar to f S , SSOTE increases with increasing c NaCl until a maximum is reached. SSOTE of up to 15%/m could be reached. Diffusers with the smaller slits (Tube 1, Disc 1) always exhibit higher values than diffusers with larger slits (Tube 2, Disc 2). In TW, SSOTE increases with increasing diffuser density (Behnisch et al. ). When comparing the results of 6 and 10 Tube 1 diffusers, the same can be observed in SW. When reaching the peak value, SSOTE slightly decreases as a result of the reduced c S . The decrease averages about 0.7%/m for c NaCl ¼ 20 g/L compared to the peak value at c NaCl ¼ CCC NaCl . Nevertheless, SW SSOTE still outperforms SSOTE values that can be achieved in TW with conventional fine-bubble diffusers (8.5-9.8%/m), even at very high diffuser densities (Behnisch et al. ).
Another factor that must be considered in the design of aeration systems is the total air supply pressure (p T ), which includes the hydrostatic pressure resulting from depth of submergence (p h ), pressure drop of pipes and valves (p s ) and pressure drop of diffusers (p d ) (Krampe ). The p h increases with increasing c Salt due to the increment of water density. Nevertheless, in practice this increase can be neglected. The density of water rises marginally by 1.6% with an increase in c NaCl from 0 g/L (997 kg/L at 20 C) to 20 g/L (1.013 kg/L at 20 C). The p S is not affected by c Salt . Also no dependence against c Salt was found for p d (Annex 5). Nevertheless, when improving the oxygen transfer in SW by installing diffusers with smaller slits, p d increases. In our tests, for tube and disc diffusers the difference in p d between the diffuser with the smaller slits and those with the larger slits was on average 2.2 kPa (at the same q A , see Annex 5). Thus, in SW, a higher f S tends to go hand in hand with a higher p d and therefore p T . The increased p T results in a higher power requirement (P) for the blowers.
When calculating P, we must differentiate between isochoric compression and adiabatic compression. Positive displacement blowers (as used here) are widely used in WWTP. Since they compress a fixed volume of air in an enclosed space to a higher pressure, they operate using the isochoric compression principle (Mueller et al. ). P can be calculated according to the following formula: P ¼ Q A · p T /η; η being the overall efficiency of the blower. For our calculation, we assumed a typical η of 0.60 (Bell & Abel ). Then we calculated the standard aeration efficiency (SAE ¼ SOTR/P).
With SAE, we are able to illustrate the interaction between increased energy demand due to increased p T through the smaller slits and the improved oxygen transfer. The average SAE is shown in Figure 7(b) as a function of c NaCl . Since identical Q A were set for each diffuser type, its influence on SAE is eliminated and the average values are sufficient for a comparison. Figure 7(b) shows that SAE in TW ranges between 3.0 and 4.4 kg/kWh and thus within an expected range for such a kind of operation conditions (Behnisch et al. ). With increasing c Salt SAE increases and reaches its peak value, similar to SSOTE. SAE for the diffusers with smaller slits (Tube 1, Disc 1) are comparable or higher than for the diffusers with larger slits (Tube 2, Disc 2). Hence, with Tube 1 (smaller slits) SAE of up to 8.4 kg/kWh could be achieved.
Contrary to positive displacement blowers, with turbo blowers internal air compression takes place (¼ adiabatic compression). Turbo blowers are generally designed for large airflow rates and are therefore used especially on large WWTP. P depends upon the inlet pressure (p 0 ), air density (ρ) and inlet temperature (T 0 ) (Mueller et al. ). Depending on design and operation conditions, overall efficiencies for turbo blowers of up to 0.85 are possible (Bell & Abel ). In order to consider these types of blowers as well, we calculated SAE again using the adiabatic power formula (EPA ). All results and the formula with the assumed operating conditions are shown in Figure 8. Due to the higher overall efficiency (using a constant η of 0.80) and the chosen operation conditions (T 0 ¼ 20 C; p 0 ¼ 101.3 kPa; ρ ¼ 1.2 kg/m 3 ), on average 52% higher SAE values of up to 12.8 kg/kWh were achieved compared to positive displacement blowers. However, it has to be taken into account that the presented power requirement values are only based on theoretical calculations. Since overall efficiency and the power requirement will change under varying operation conditions, lower SAE values are likely to be achieved in practice. Apart from that, the previous conclusions remain valid. Higher SAE values were achieved with diffusers with smaller slits. In summary, the SAE results show that, independent of the blower type used and despite the higher pressure drop, in SW higher efficiency of the aeration system is achieved using the diffusers with smaller slits. However, it is questionable whether this can be applied to all diffusers. Whether a diffuser with smaller slits in SW ensures a more efficient aeration depends largely on the difference in oxygen transfer as well as on pressure drop compared to the diffuser with larger slits. Additionally, further measurements need to show whether a higher SAE value can also be achieved in activated sludge. The activated sludge could have an influence on the coalescence behaviour in the aerated tank. Diffusers with smaller slits could also lead to increased fouling. Therefore, further experiments are already planned to answer these questions.

CONCLUSION
In the present work, the experiments presented in Part I of this paper (Behnisch et al. ) were continued to investigate the oxygen transfer of fine-bubble aeration systems and their influencing factors in saline water. In Part I, data from bubble size distribution were combined with results from oxygen transfer tests to investigate the impact of the design of the diffuser membrane. Within Part II, the effect of different salts as well as the effects of diffuser type and density on oxygen transfer of fine-bubble aeration systems in saline water is presented.
First, we determined the critical coalescence concentration (CCC) for various salts for the first time using conventional fine-bubble diffusers, for which we developed a new analytical approach. When CCC is reached, coalescence is completely inhibited and k L a 20 and f S remain constant. Regardless of salt type, the f S,max values reached were the same. With the new analytical approach, the CCC is determined by evaluating the oxygen transfer rate at different salt concentrations. The new method is much faster and easier to use than bubble size measurement and provides valid results. In the future, this will make it possible to investigate the coalescence behaviour for any aeration system and (mixed) salt solution quickly and easily. Furthermore, since salt and salt mixture in wastewater change and depend on the origin of the used water, we recommend that in the future not only the total salt concentration should be considered in wastewater analytics but also its composition from different salts (e.g. by ion analysis).
Second, we carried out oxygen transfer tests with three different types of diffusers at different diffuser densities in a pilot-scale test tank at a water depth of 3.8 m and at different NaCl concentrations. The effect of the design of the diffuser membrane observed in Part I was confirmed. Except for Plate 1, diffuser membranes with smaller detaching bubbles show higher f S values. Furthermore, our results show that there is a clear influence of the diffuser type. While f S for tube diffusers only depends on the salt concentration, the other diffuser types show a clear influence of the airflow rate per slit and slit length (q A,dS ). For disc diffusers and Plate 2, the highest f S values were achieved at q A,dS ∼ 1.0 dm 3 /h/mm. If q A,dS changes, f S and k L a will decrease again. This complicates the operation of an aeration system. When using tube diffusers, this problem can be avoided. To achieve a high oxygen transfer in saline water, the slit density of the diffuser membrane can be further increased. In tap water, the slit density is usually up to 15 slits per cm 2 to prevent coalescence close to the diffuser. Under saline conditions (c Salt > CCC), the coalescence is inhibited and the slit density can be increased further. This also prevents high q A,dS and thus an increase in initial bubble size.
Third, despite the reduced oxygen saturation concentration in saline water, extremely high SSOTE values compared to tap water of up to 15%/m were achieved during the tests in the pilot-scale test tank. SSOTE are much higher than is possible with conventional fine-bubble diffusers in tap water. Using isochoric power formula (valid for positive displacement blowers), calculated SAE values of up to 8.4 kg/kWh show that by using diffusers with smaller slits, the energy efficiency of the aeration system will improve in saline water despite the increased pressure drop. If even more efficient blowers (e.g. turbo blowers) were used, theoretically (using adiabatic power formula) even higher SAE of up to 12.8 kg/kWh values could be achieved. However, it is not yet known whether the use of diffusers with smaller slits also improves aeration in activated sludge tanks. The coalescence behaviour in the aerated tank as well as the fouling of the diffusers could be influenced by the activated sludge. Therefore, further investigations are already planned.