Single-/triple-stage biotrickling filter treating a H₂S-rich biogas stream: Statistical analysis of the effect of empty bed retention time and liquid recirculation velocity

ABSTRACT

Biogas containing H₂S has limited use in electricity and heat production as H₂S can be corrosive to metal equipment. Bio-filtration has proved to be a suitable technology for biogas desulfurization because of economical and environmental benefits over physicochemical techniques. In the present study, a response surface methodology using 3² full factorial design was employed to determine the effects of two operating parameters, namely empty bed retention time (EBRT: 100–180 sec) and liquid recirculation velocity (LRV: 2.4–7.1 m³ m⁻² h⁻¹) on H₂S removal efficiency (%) in single-stage and triple-stage bio-trickling filters (SBTF and TBTF) treating an H₂S-rich biogas. Quadratic model was found to be the best predictive model for H₂S removal efficiency. The results indicated that H₂S removal efficiency was significantly influenced by the synergistic effect of linear terms of EBRT and LRV with a greater effect associated with EBRT. However, the quadratic term of LRV had an antagonistic effect. The quadratic term of EBRT and cross-product term between EBRT and LRV did not exhibit a significant effect on H₂S removal efficiency. The predicted values from the established models showed a close agreement with the experimental data with the coefficient of determination (R²) of 0.99 for H₂S removal efficiency in both SBTF and TBTF. Response analysis demonstrated that the performance of TBTF was superior compared to SBTF.

Implications: Bio-trickling filter technology has gained a lot of attention for biogas desulfurization because it is economically and environmentally superior over chemical methods. Empty bed retention time (EBRT) and liquid recirculation velocity (LRV) are crucial variables influencing the performance of bio-trickling filters. In this work, the authors established a model that can properly predict H₂S removal efficiency in a single/triple bio-trickling filter (SBTF and TBTF) treating H₂S-rich biogas with regard to the individual and interaction effects between EBRT and LRV. Analysis with the help of response surface methodology indicated that TBTF was more efficient compared to SBTF for H₂S removal.

Introduction

Hydrogen sulfide (H₂S) is a toxic and foul-smelling compound. It has a low odor threshold between 0.0001 and 0.03 parts per million (ppm) (Rava 2008). H₂S has been reported to cause a wide range of adverse impacts on human health including headaches and eye irritation (1–10 ppm) (Cheng et al. 2018), serious eye damage (50 ppm) (Ko, Xu, and Jang 2015) and instantaneous death at concentrations over 1000 ppm (Charnnok et al. 2013). H₂S not only causes life-threatening circumstances but it can also accelerate the corrosion of metal equipment where H₂S, oxygen, and moisture coexist. In addition, flaring biogas containing H₂S releases sulfur dioxide (SO₂) into the air which is one of the most pollutants contributing to acid deposition, more commonly known as acid rain, because it can react with water vapor and oxygen in the air to form sulfuric acid which is deposited to the earth when there is rain.

H₂S is released, as a by-product, from various industrial sectors including natural gas production sites, oil refineries (Vikrant et al. 2018), Kraft paper mills (Rava 2008) and geothermal power plants (Bassani et al. 2018; Gunnarsson et al. 2013). Landfill gas is also recognized as a major source of H₂S emission (Ko, Xu, and Jang 2015; Rasi 2009). The off-gas from either anaerobic digestion processes (Almenglo et al. 2016; Ziemiński and Kopycki 2016) or intensive livestock operations (Su et al. 2013) contain a considerable amount of H₂S. The biogas from an upflow anaerobic sludge blanket (UASB) reactor treating a concentrated rubber latex processing wastewater (Songkhla, Thailand) could contain H₂S at a concentration as high as 26000 ppm (Saelee et al. 2009).

To avoid the above-mentioned problems, H₂S must be removed from biogas prior to further utilization. Biotrickling filter (BTF) has become a widely accepted technology for biogas desulfurization and is a cost-efficient and environmentally friendly alternative to physico-chemical methods (Khoshnevisan et al. 2017). In BTFs treating a contaminated gas, the gas and recirculation liquid streams are passed through a packed bed on which a microbial consortium is immobilized. The contaminants are transported from the gas stream into the bio-film where they are biodegraded (Mudliar et al. 2010).

Desulfurizing BTFs are often operated under aerobic conditions. The final products can be either elemental sulfur (S⁰) or sulfuric acid depending on the availability of oxygen for the microorganisms in the BTF (Khoshnevisan et al. 2017; López et al. 2016). Qiu and Deshusses (2017) studied the effect of H₂S-to-O₂ molar ratio on the performance of a monolith desulfurizing BTF with an inlet H₂S concentration of around 1000 ppm corresponding to an H₂S-S loading rate of 127 g-S m⁻³ h⁻¹. The results indicated that H₂S removal efficiency at H₂S-to-O₂ molar ratio of 2:1 was 50%. Analysis of sulfur species in the liquid phase indicated that S⁰ accounted for about 40% of the total sulfur species recovered in the liquid. Tomàs et al. (2009) determined S⁰ in the biofilm formed on the surface of packing material of a full-scale BTF treating 80 m³ h⁻¹ biogas with an average H₂S concentration of 3000 ppm. They observed that 68–99% of total solids in the biofilm was S⁰ due to low oxygen availability. Since S⁰ formation in desulfurizing BTFs leads to bed clogging and a subsequent increase in pressure drop (Vikrant et al. 2018), oxygen appropriate load should be supplied to enhance H₂S oxidation to sulfuric acid (Tayar et al. 2019). Although sulfuric acid formation acidifies the biofilm, the successful application of acidic desulfurizing BTFs has been reported (Chaiprapat et al. 2015; Montebello et al. 2014; Tu et al. 2017).

Liquid recirculation velocity (LRV) and empty bed retention time (EBRT) are two important operating factors that influence the efficiency of desulfurizing BTFs (Chaiprapat et al. 2015; Charnnok et al. 2013). The use of recirculating liquid in BTFs allows supplying oxygen and regulating pH inside the reactor, as well as provides the nutrients required for microorganisms to grow and degrade the biogas contaminant. Good care should be taken to control LRV because high values of LRV could increase the thickness of the liquid layer on the biofilm surface that may reduce the contaminant mass transfer from the gas phase to the biofilm (López et al. 2016; Mirmohammadi, Sotoudeheian, and Bayat 2017). It is worth mentioning that a long travel path of recirculating liquid through BTF bed causes liquid channeling leading to partial wetting of the bed which negatively influences the BTF performance. Bed division with multiple-point recirculating liquid injection could help deliver liquid more evenly to the bed and avoid the flow channeling problems (Chaiprapat et al. 2015). EBRT, empty bed volume-to-influent biogas flow rate ratio, represents the time that an element of the biogas needs to travel from the empty bed inlet to the outlet. A number of studies have indicated that the longer EBRT of the desulfurizing BTF, the higher H₂S removal efficiency (Guerrero and Bevilaqua 2015; Vikromvarasiri et al. 2017). Although desulfurizing BTFs at longer EBRT could improve the treatment efficiency, excessively large BTF volume results in higher fabrication and material costs (Khoshnevisan et al. 2017). Most desulfurizing BTFs have been operated at an EBRT shorter than 7 min, for instance, 0.15–0.95 min (Solcia et al. 2014), 0.67–1.67 min (Zhuo et al. 2019), 1.67–3.0 min (Chaiprapat et al. 2015) and 5.5–6.3 min (Zhou et al. 2015).

The aim of this study was to develop a model for H₂S removal efficiency in a single/triple-stage BTF (SBTF and TBTF) in terms of LRV and EBRT using a 3² full factorial design (FFD) combined with response surface methodology (RSM), and to evaluate the adequacy and predictive capability of the model based on statistical criteria.

Materials and methods

Data acquisition

Data were taken from the experiments carried out by Chaiprapat et al. (2015). Two lab-scale BTFs including SBTF and TBTF were operated under acidic condition (pH < 4.0) and a temperature of 31 ± 2.3°C, for desulfurization of H₂S-rich biogas containing an average H₂S concentration of 5522 ppm. Coconut husk used as bed media was submerged in the full-scale anaerobic digester effluent containing a sulfide concentration of about 31 mg/L (Songkhla, Thailand). Aeration was continuously applied and fresh effluent from the anaerobic digester was replaced every 2 days. After a 2-week inoculation, the media were packed into the BTFs and aerated anaerobic digester effluent was used as the recirculating liquid. Two variables, namely EBRT and LRV ranging from 100 to 180 sec, and 2.4 to 7.1 m³ m⁻² h⁻¹, respectively, were tested to evaluate H₂S removal efficiency (Chaiprapat et al. 2015).

Choice of statistical model

The RSM coupled with a 3² FFD was used to establish a functional relationship between EBRT and LRV (termed independent variables) and responses (H₂S removal efficiency in SBTF and TBTF). The 3² FFD matrix containing the independent variables in terms of actual and coded values (−1, 0 and +1 for low, medium and high level, respectively), and the actual values of the responses are outlined in Table 1 in which each row describes an experiment. The relationship between actual and coded values of independent variables is given by Equation (1).

(1) $X_i=\left(2x_i-x_{max}-x_{min}\right)/\left(x_{max}-x_{min}\right)$(1)

Table 1. 3² FFD matrix containing tested levels of the independent variables along with the values of responses used in this study (adapted from Chaiprapat et al. 2015).

	Independent variable		Response
Experimental run	X1	X2	Y1 (%)	Y2 (%)
1	180 (+1)	7.10 (+1)	93.0	96.7
2	180 (+1)	4.75 (0)	96.0	94.3
3	140 (0)	7.10 (+1)	86.6	92.2
4	100 (−1)	4.75 (0)	76.5	82.0
5	180 (+1)	2.40 (−1)	89.1	87.9
6	140 (0)	2.40 (−1)	79.1	83.3
7	100 (−1)	2.40 (−1)	62.0	69.0
8	100 (−1)	7.10 (+1)	70.0	84.5
9	140 (0)	4.75 (0)	88.8	91.9

Note. X₁: EBRT (sec); X₂: LRV (m³ m⁻² h⁻¹); Y₁: H₂S removal efficiency (%) in SBTF; Y₂: H₂S removal efficiency (%) in TBTF; Parenthesis in data indicates coded values of independent variables; 4.7 was taken as 4.75 (the centre point for LRV of 2.40 and 7.10 m³ m⁻² h⁻¹).

where X_i and x_i represent the coded and actual values of independent variable “i”, x_min and x_max are the actual values of the independent variable “i” at the minimum and maximum points, respectively (Singh et al. 2018).

The 3² FFD experimental data (Table 1) were fitted to linear, two-factor interaction (2FI) and quadratic models in order to obtain an adequate functional relationship between independent variables and responses. Since two independent variables (X₁ and X₂) were tested, the linear, 2FI and quadratic models are expressed as Equations (2), (3) and (4), respectively.

(2) $Y=b_0+b_1{X}_1+b_2{X}_2$(2)

(3) $Y=b_0+b_1{X}_1+b_2{X}_2+b_{12}{X}_1{X}_2$(3)

(4) $Y=b_0+b_1{X}_1+b_2{X}_2+b_{12}{X}_1{X}_2+b_{11}{X_1}^2+b_{22}{X_2}^2$(4)

where Y represents the predicted response; b₀, b₁-b₂, b₁₂, and b₁₁-b₂₂ are the model intercept, linear coefficients, interaction coefficient, and quadratic coefficients, respectively.

In this study, firstly, the best predictive model (for each response) was selected based on the sequential model sum of squares analysis and model summary statistics. Secondly, analysis of variance (ANOVA) was used to calculate the model coefficients, and to evaluate the significance of the model and the effect of each model term on the response. Thirdly, diagnostic plots were generated to assess the models' adequacy and clarify whether there is any violation against the ANOVA assumptions. Lastly, the selected models were graphically represented as three-dimension (3D) response surface and two-dimension (2D) contour plots to provide better insight about the individual and simultaneous influence of the independent variables on the responses, and then the optimal value of responses was determined.

Statistical and graphical analysis

Design-Expert® software 10.0.7.0 (trial version, Stat-Ease Inc., Minneapolis, USA) was used for statistical analysis and graphical representation of the experimental data. ANOVA was performed with respect to a 95% confidence level. The overall model and the model terms were considered to be statistically significant when their corresponding “p-value” was less than 0.05.

Results and Discussion

The effect of two operational parameters (EBRT: 100–180 sec and LRV: 2.4–7.1 m³ m⁻² h⁻¹) on the performance of a desulfurizing SBTF and TBTF was investigated using a 3² FFD of RSM. The dependent variable (response) was H₂S removal efficiency (%).

Model fitting

Tables 2 and 3 represent the results of the sequential model sum of squares and model summary statistics for H₂S removal efficiency in SBTF and TBTF, respectively. As seen in Table 2, both linear and quadratic models were significant (p-value < 0.05). However, the quadratic model showed the higher adjusted coefficient of determination (R²_adj.) and predicted R² (R²_pred.) values, and lower residual error sum of squares (PRESS) (Table 3). Therefore, the quadratic model was chosen for further analysis.

Table 2. Sequential model sum of squares for H₂S removal efficiency in SBTF and TBTF.

Bio-trickling filter	Model	SS	df	MS	F-value	p-value	Remarks
SBTF	Linear	870	2	435	17.8	0.003	significant
	2FI	4.2	1	4.2	0.15	0.716
	Quadratic	130	2	64	15.6	0.026	significant
TBTF	Linear	498	2	249	21.2	0.002	significant
	2FI	11.2	1	11.2	0.95	0.375
	Quadratic	52	2	26	10.9	0.042	significant

Note. SS: sum of squares; df: degree of freedom; MS: mean square; F-value: Fisher’s function value; p-value: probability value (level of significance); 2FI: two-factor interaction; Significant at the 0.05 probability level.

Table 3. Model summary statistics for H₂S removal efficiency in SBTF and TBTF.

Bio-trickling filter	Model	R^2	R^2adj.	R^2pred.	PRESS	Remarks
SBTF	Linear	0.856	0.808	0.664	341
	2FI	0.860	0.776	0.222	790.9
	Quadratic	0.988	0.967	0.857	145.4	Selected
TBTF	Linear	0.876	0.835	0.709	165.4
	2FI	0.896	0.833	0.454	310.4
	Quadratic	0.987	0.966	0.847	86.8	Selected

Note. R²: coefficient of determination; Subscripts “adj.” and “pred.” stand for “adjusted” and “predicted”, respectively; PRESS: predicted residual error sum of squares; 2FI: two-factor interaction.

The ANOVA for the quadratic model was performed to develop an equation relating H₂S removal efficiency and the two aforementioned variables (LRV and EBRT) and to check the significance of the effect of each term of the model on the removal efficiency (Table 4). The larger the t-value and smaller the p-value, the more significant is the corresponding coefficient term.

Table 4. ANOVA analysis for the quadratic model of H₂S removal efficiency in SBTF and TBTF.

Response	Term	Effect	Coeff.	St. Error	t-value	p-value
Y1	Intercept		89.6	1.52	59.0	.000
	X1	23.2	11.6	0.83	13.9	.001
	X2	6.47	3.23	0.83	3.88	.030
	X1^2	−7.47	−3.73	1.44	−2.59	.081
	X2^2	−14.27	−7.13	1.44	−4.95	.016
	X1X2	−2.05	−1.03	1.02	−1.01	.389
Y2	Intercept		91.7	1.152	79.6	.000
	X1	14.47	7.23	0.631	11.47	.001
	X2	11.07	5.53	0.631	8.77	.003
	X1^2	−6.80	−3.40	1.093	−3.11	.053
	X2^2	−7.60	−3.80	1.093	−3.48	.040
	X1X2	−3.35	−1.68	0.773	−2.17	.119

Note. Y₁ and Y₂ represent H₂S removal efficiency (%) in SBTF and TBTF, respectively; X₁ and X₂ represent independent variables EBRT and LRV in coded values, respectively.

According to Table 4, the linear terms of EBRT (X₁) and LRV (X₂), as well as the quadratic term of X₂ (X₂²), showed a significant effect (p-value < 0.05) on H₂S removal efficiency whereas the quadratic term of X₁ (X₁²), and the cross-product term between X₁ and X₂ (X₁ X₂) were insignificant (p-value > 0.05). Among the significant model terms, X₁ and X₂ positively influenced H₂S removal efficiency, with a greater effect associated with X₁, whereas X₂² exhibited a negative effect.

The final equation for the responses in terms of the coded and actual values of the independent variables is given by Equations (5)–(6) and Equations (7)–(8), respectively.

(5) $\begin{array}{rl}& Y_1=89.6+11.60X_1+3.23X_2-1.03X_1{X}_2\\ & -3.73{X_1}^2-7.13{X_2}^2\end{array}$(5)

(6) $\begin{array}{rl}& Y_2=91.7+7.23X_1+5.53X_2-1.68X_1{X}_2\\ & -3.40{X_1}^2-3.80{X_2}^2\end{array}$(6)

(7) $\begin{array}{rl}& Y_1=-39.7+0.995x_1+15.17x_2-0.011x_1{x}_2\\ & -0.002{x_1}^2-1.292{x_2}^2\end{array}$(7)

(8) $\begin{array}{rl}& Y_2=-13.9+0.861x_1+11.39x_2-0.0178x_1{x}_2\\ & -0.0021{x_1}^2-0.6881{x_2}^2\end{array}$(8)

where Y₁ and Y₂ represent H₂S removal efficiency (%) in SBTF and TBTF, respectively; (X₁, X₂) and (x₁, x₂) are the coded and actual values of the independent variables EBRT and LRV, respectively.

Model adequacy

The adequacy of the selected models was checked using the parity plot of the predicted values against the actual values for the responses (Figure 1). The actual values of responses were obtained from the experimental design matrix (Table 1), and the corresponding predicted values were obtained by calculation from Equations (5)–(8).

Figure 1. Comparison between predicted and experimental (actual) values for the responses (a) H₂S removal efficiency in SBTF (Y₁) and (b) H₂S removal efficiency in TBTF (Y₂).

It is evident from Figure 1, that the data points on the plot were dispersed close to the 45-degree line (100% correlation line) with R² value of 0.988 and 0.987 for H₂S removal efficiency in SBTF and TBTF, respectively. This implies that the models could satisfactorily explain the variability in the responses. The models did not explain only about 1.3% (= 100 × (1-R²)) of the total variability. In addition, diagnostic plots including the normal probability plot (NPP) (Figure 2) and the plot of studentized residuals versus the predicted values for the responses (Figure 3) were constructed to clarify any problem in the experimental data. A studentized residual is the residual divided by its standard deviation where the residual is the difference between an actual value for the response and its corresponding predicted value.

Figure 2. Normal probability plot of residuals for the responses (a) H₂S removal efficiency in SBTF (Y₁) and (b) H₂S removal efficiency in TBTF (Y₂).

Figure 3. Internally studentized residuals versus predicted values of responses (Y₁: H₂S removal efficiency in SBTF; Y₂: H₂S removal efficiency in TBTF).

As displayed in Figure 2, it is clear that the NPPs of residuals showed a straight-line pattern signifying that there was no violation of normality assumption. From Figure 3, the data points for residuals randomly dispersed within the recommended limits −3 and +3 (Bustillo-Lecompte, Mehrvar, and Quiñones-Bolaños 2016; Safari et al. 2015), suggesting that the selected models are adequate and do not show any violation of the constant variance assumption.

Response surface analysis

Since residual analysis exhibited that the selected models had no statistical problems, the graphical representation of the responses such as 3D response surface and 2D contour plots were generated to provide better insight into the relationship between the independent variables and the responses. It was performed by substituting the values of independent variables within the experimental range (see Table 1) in the models using Design-Expert software. Figures 4 and 5 illustrate the interaction effect of EBRT and LRV on H₂S removal efficiency in SBTF and TBTF expressed as 3D response surface and corresponding 2D contour plots.

Figure 4. (a) 3D response surface and (b) 2D contour plots illustrating the effect of EBRT and LRV on H₂S removal efficiency in SBTF (Y₁).

Figure 5. (a) 3D response surface and (b) 2D contour plots illustrating the effect of EBRT and LRV on H₂S removal efficiency in TBTF (Y₂).

As depicted in Figures 4 and 5, H₂S removal efficiency in both SBTF and TBTF changed nonlinearly with EBRT and LRV variables. Each variable showed a positive effect with the stronger effect belonged to EBRT. These results are in line with findings of Abdehagh et al. (2014). The removal efficiency increased as LRV increased from 2.4 m³ m⁻² h⁻¹ (lowest in this study) up to a critical LRV (LRV_C) value at which the removal efficiency reached an optimum value. Such enhancement in the efficiency of H₂S removal could be associated with the higher amount of oxygen (through recirculating liquid) entered the BTFs. Chaiprapat et al. (2015) reported that the LRV_C in SBTF and TBTF at each EBRT tested was around 4.7 and 7.1 m³ m⁻² h⁻¹, respectively, whereas the models presented in this study demonstrated that the LRV_C varied over a range of 5.1 to 5.5 m³ m⁻² h⁻¹ and 5.9 to 7.0 m³ m⁻² h⁻¹ in SBTF and TBTF, respectively (Table 5).

Table 5. Effect of EBRT and LRV on H₂S removal efficiency in SBTF and TBTF (data from Figures 4 and 5).

	LRV (m^3 m^−2 h^−1)
	LRVL		LRVC		LRVH
EBRT (sec)	Y1 (%)	Y2 (%)	Y1 (%)	Y2 (%)	Y1 (%)	Y2 (%)
100	62.86	70.02	74.89 (5.5)	84.45 (7.0)	71.38	84.44
140	79.22	82.33	89.96 (5.3)	93.68 (6.5)	85.69	93.40
180	88.11	87.84	97.63 (5.1)	96.48 (5.9)	92.53	95.56

Note. LRV_L and LRV_H represent the lowest and highest LRV tested in this study, 2.4 and 7.1 m³ m⁻² h⁻¹, respectively; LRV_C is the critical LRV at which the H₂S removal efficiency reached an optimum value; Y₁ and Y₂ are H₂S removal efficiency in SBTF and TBTF, respectively; Parenthesis in data indicates the LRV_C; EBRTs of 100, 140 and 180 sec correspond to H₂S-S loading rates of 304, 218 and 165 g-S m⁻³ h⁻¹, respectively.

Further, increase beyond LRV_C while EBRT held constant led to a negative influence on H₂S removal efficiency that was obvious particularly in SBTF. This implies that, although oxygen is essential for H₂S biofiltration, excessive amount of liquid in the bed could result in an increase in the liquid film thickness on the biofilm and, consequently, a reduction in H₂S mass transfer from gas phase to the biofilm. The comparison between the performance of TBTF and SBTF at EBRT of 100 sec and LRV of 7.1 m³ m⁻² h⁻¹ clearly demonstrated that H₂S removal efficiency in TBTF was 18.3% higher than that of SBTF. This is because the upper two stages in TBTF were not exposed to an excessive amount of liquid. Only the lower stage received entire LRV since one-third of LRV was fed to each stage. These findings suggest that TBTF could alleviate problems arising from excessive liquid in SBTF such as H₂S mass transfer limitation.

Table 5 shows that the optimal H₂S removal efficiency in SBTF and TBTF increased from 74.9% to 97.6% and from 84.5% to 96.5%, respectively, as EBRT increased from 100 to 180 sec. These results are in line with findings of Montebello et al. (2014) who observed an increase in H₂S removal efficiency from (50%-60%) to (90%-100%) when EBRT increased from 33 to 130 sec in a BTF treating an inlet H₂S concentration of 2000 ppm H₂S. Comparable results were also obtained by Guerrero and Bevilaqua (2015) who reported a rising trend in H₂S removal efficiency from about (30%-50%) to (90%-98%) when EBRT increased from 96 to 174 sec in a BTF fed with biogas from a full-scale UASB reactor of the wastewater treatment plant of a brewery containing up 478 ppm H₂S. Likewise, the findings of Vikromvarasiri et al. (2017) towards biogas desulfurization using a BTF inoculated with Halothiobacillus neapolitanus NTV01 revealed that an increase in EBRT from 40 to 120 sec could enhance H₂S removal efficiency from about 42% to 93% with an inlet H₂S concentration of 1000 ppm H₂S. A number of studies focused on BTF treating lower concentration of H₂S (25–55 ppm H₂S) indicated an improvement in efficiency of H₂S removal (from 52% to 100%) when EBRT raised from 9 to 45 sec (Abdehagh et al. 2011; Jin, Veiga, and Kennes 2005; Namini et al. 2008). The longer EBRT could provide a longer time for H₂S to transfer from gas phase to biofilm where it is oxidized by sulfide oxidizing bacteria.

At EBRT of 180 sec, the longest EBRT in this study, H₂S removal efficiency in TBTF obtained 96.5%, which was almost similar to that achieved in SBTF (97.6%). This could be due to both SBTF and TBTF receiving the lowest H₂S-S loading rate tested in this study (165 g-S m⁻³ h⁻¹), and thus oxygen mass transfer was not hindered under this condition. However, when BTFs were operated at a shorter EBRT, the efficiency of H₂S removal in TBTF was remarkably greater than that of SBTF. For instance, at EBRT of 100 sec (corresponding to an H₂S-S loading rate of 304 g-S m⁻³ h⁻¹), removal efficiency of 74.9% and 84.5% obtained in SBTF and TBTF, respectively. It can be concluded that dividing the BTF bed into multiple stages (each stage with its own recirculating liquid injection point) improves the treatment efficiency. The reason is that the bed staging allows more uniform distribution of the recirculating liquid through the BTF bed which can be translated into a more even oxygen distribution leading to a better bacterial growth and, consequently, an enhancement in H₂S oxidation.

Conclusion

In this study, a modeling exercise was carried out on the available published work of Chaiprapat et al. (2015). A statistical investigational technique (response surface methodology coupled with a 3² full factorial design) was applied to establish a predictive model in order to evaluate the individual and interaction effects of empty bed retention time (EBRT) and liquid recirculation velocity (LRV) on H₂S removal efficiency in single-stage and triple-stage bio-trickling filters (SBTF and TBTF) treating a biogas stream containing H₂S, and also to determine the optimal H₂S removal efficiency. The quadratic polynomial model was the best model in fitting the experimental data because it showed higher R²_adj. and R²_pred., and lower PRESS compared to that calculated for the linear and 2FI models. The adequacy of the model was proven using robust statistical analysis. This study allowed drawing the following conclusions:

• H₂S removal efficiency in SBTF and TBTF was satisfactorily predicted using the quadratic model with coefficient of determination (R²) of 0.99.

• Linear terms of each variable (EBRT and LRV) and the quadratic term of LRV significantly influenced H₂S removal efficiency with the greater effect due to the linear term of EBRT.

• Response analysis reiterated that both TBTF and SBTF showed almost similar H₂S removal efficiency under long EBRT. However, at shorter EBRT, H₂S removal efficiency in TBTF was superior compared to SBTF.

Additional information

Notes on contributors

Reza Salehi

Reza Salehi is currently an independent researcher who obtained his Ph.D. in Environmental Engineering from Polytechnique Montreal, Canada in November 2016. His research interest is mainly enhanced biological phosphorus removal (EBPR) and modelling of biological wastewater treatment systems.

Sumate Chaiprapat

Sumate Chaiprapat is an Associate Professor at Department of Civil Engineering, Prince of Songkla University, Thailand. He is also the director of Energy Systems Research Institute, Prince of Songkla University, Thailand. His major research interest is energy technology.