Control of disinfection by-product derived from humic acid using MIEX process: optimization through response surface methodology

Abstract

This work investigated the removal of humic acid (HA) as the precursor of disinfection by-product (DBP) using a magnetic ion exchange (MIEX) resin. The Box–Behnken design of response surface methodology (RSM) was used to evaluate the effect of process variables (initial pH, MIEX dose and reaction time) and their interaction for HA removal. Specific ultraviolet absorbance (SUVA) removal was selected as the response value. The analysis of variance (ANOVA) of the quadratic model demonstrated that the model was significant, and the optimum conditions were obtained as pH 2.3, MIEX resin dosage 0.95 mL L⁻¹, and reaction time 59.18 min. At these applied optimum conditions, 39.74% SUVA removal from an initial SUVA value of 10.06 L mg⁻¹ m⁻¹ was achieved. Also the reduction in DBP (chloroform, dichloroacetic acid and trichloroacetic acid) formation potentials was about 76%, 73% and 80%, respectively. The MIEX process was effectively optimized by RSM with a desirability value of 1.0. After the process optimization, the MIEX treatment was proven to be a good option for controlling the DBP derived from HA.

---

1. Introduction

Natural organic matter (NOM) has a significant impact on drinking water quality either directly, by reacting with chlorine and disinfectants (to form disinfection by-products (DBPs)), or indirectly, by impacting water treatment processes (including fouling of membranes and reducing the effectiveness of activated carbon for contaminant removal).¹ NOM acts as a precursor to DBPs, amongst which the trihalomethanes (THMs) and haloacetic acids (HAAs), the main products of chlorination, are considered to be carcinogenic, mutagenic, or teratogenic. Since these DBPs present a health risk to humans, their levels in drinking water are regulated to limits of exposure: in the USA: THMs 80 μg L⁻¹ and HAAs 60 μg L⁻¹, in the UK, THMs 100 μg L⁻¹,² in China: THMs 60 μg L⁻¹, HAAs 50 μg L⁻¹ for DCAA (dichloroacetic acid) and 100 μg L⁻¹ for TCAA (trichloroacetic acid). In addition to serving as a DBP precursor, NOM can increase the ozone demand of water by hampering the effectiveness of ozone for disinfection. Meanwhile the formation of oxidation by-products, such as aldehydes, is increased in waters with elevated NOM concentrations because of higher amounts of ozone.^3,4 Further, the NOM (especially the dissolved organic matter (DOM)) in waters can favor biological regrowth in distribution network, which causes the biofilm problems. Also, NOM can significantly reduce the adsorption capacity and adsorption kinetics of activated carbon by competing with targeted contaminants.⁵

Humic acid (HA), which is primarily a product of the microbiological degradation of vegetation as well as animal decay, is generally considered as one of the ubiquitous NOMs. It enters surface and ground waters through rainwater run-off from the surrounding catchment.⁶ Usually HA is used as the model compounds to act as surrogates for the NOM removal tests.^7–9 In very recent years, a technology developed specifically for the removal of NOM is the patented MIEX process. This process utilizes a strong base anion-exchange resin, incorporating magnetic iron oxide particles within its core.¹ The strong base MIEX (magnetic ion exchange) resin is designed to be used as slurry in a completely mixed flow reactor or fluidized bed reactor.^10,11 It has been shown to remove NOM more effectively than enhanced coagulation^10,12–14 and to reduce the formation potential of DBPs.^4,15 Until now, to the best of our knowledge, most of the published literature on MIEX process is from the removal of NOM and the control of DBPs.^13–16 There is less published information on the optimization of HA (the model compound of NOM) removal by MIEX resin. In particular, the DBP formation potential (DBPFP) under optimal MIEX operating conditions has not been reported. Response surface methodology (RSM) is proven to be an effective means for the above-mentioned purpose. The use of RSM has been accentuated for developing, improving and optimizing the complex processes and to evaluate the magnitude of various influencing parameters.¹⁷ Box–Behnken design (BBD) is the most frequently used method of RSM due to more labor efficient.¹⁸

Accordingly, the objectives of this research were: (1) to apply Box–Behnken approach combined with RSM and optimization modeling for maximizing the adsorption capacity of MIEX resin for HA removal from aqueous solution; (2) to examine the effects of three independent variables (pH, MIEX resin dosage and reaction time) and their interactions on the removal of HA; (3) to verify the validity of the proposed model by means of analysis of variance (ANOVA) and confirmation experiments; (4) above all, to investigate the effects of optimal MIEX treatment process on formation potential of chlorinated DBPs.

2. Materials and methods

2.1. Raw materials

The MIEX resin used in this study was virgin (unused) resin provided by the China Agent of Orica Watercare of Victoria. The resin was rinsed with Millipore de-ionized (DI) water to wash away the fines and stored in DI water before use. The following using details for MIEX resin were reported by Ding et al. previously.¹⁹ All the chemicals used were analytical grade and used without further purification. Hydrochloric acid (HCl) and sodium hydroxide (NaOH) having the purity of 95% were supplied by Sinopharm Chemical Reagent Co., Ltd., China. Distilled water was used to prepare a 0.1 or 1.0 M solutions of HCl and NaOH.

Humic acid (HA, Cat. No. H108498), consisting of greater than 90% fulvic acid (FA), was purchased from Aladdin Chemical Co., Ltd., Shanghai, China. HA stock solution was prepared by dissolving nominal amount of HA into ultrapure water, followed by filtering through 0.7 μm glass-fiber filter (Grade GF/F, Cat. No. 1825-055, Whatman, UK). Thereafter, certain amount of the stock solution was spiked into the 500 mL deionized water to achieve the nominal HA concentration (10 mg L⁻¹).

2.2. Methods

Batch experiments were carried out by method described by Zhang et al. under the various operating conditions such as MIEX resin dosage (0.4–1.2 mL L⁻¹), initial pH (2–12) and reaction time (20–60 min).²⁰ Solution pH was measured using an Accumet AB15 pH meter from Fisher Scientific. The pH meter was calibrated before each use. UV absorbance at 254 nm (UV₂₅₄) was measured on a Hitachi DR-6000 spectrophotometer using a 1 cm quartz cell. DOC concentration was determined using a Shimadzu TOC-V_CPH Total Organic Carbon Analyzer; DOC was analyzed according to Standard Method 5310B.²¹ Usually, the Specific ultraviolet absorbance (SUVA) is found to be a good surrogate for NOM fractions, such as humic and fulvic acids. In this study, the SUVA removal was used to characterize the changes of HA concentration. SUVA (L mg⁻¹ m⁻¹) and its removal percentage (%) were respectively calculated using the following formulae.where A₂₅₄ is the sample absorbance measured at 254 nm measured in a 1 cm cuvette, and DOC is measured in mg C per L. SUVA_i and SUVA_f are the initial and final specific ultraviolet absorbance of HA in samples.

In addition, the excitation–emission matrices (EEMs) of original HA and HA treated by MIEX resin were respectively acquired using a fluorescence spectrophotometer (model F-4500, Hitachi, Japan). Also molecular weight distributions of HA were determined by a high performance size-exclusion chromatography method. A high-performance liquid chromatography was used with an ultraviolet (UV) detector following size separation by a TSK-GEL G3000PWXL column (7.8 mm × 300 mm, TOSOH Corporation, Japan).

2.3. Response surface methodology

Response surface methodology is a set of mathematical techniques that describe the relation between several independent variables and one or more responses. Also the RSM method is based on the fit of mathematical models (linear, square polynomial functions and others) to the experimental results generated from the designed experiment and the verification of the model obtained by means of statistical techniques. At present, RSM often contains three steps:²² (1) design and experiments; (2) response surface modeling through regression; and (3) optimization.

2.3.1. Experimental design. Response surface methodology (RSM) was used to optimize the adsorption of HA on MIEX resin. A Box–Behnken design (BBD) was employed with three independent variables including initial pH (X₁), MIEX resin dosage (X₂) and reaction time (X₃). The ranges of the variables were selected based on the preliminary orthogonal tests and the related literature of similar experiments.²⁰ The matrix of design was shown in Table 1, including coded and actual values of the three variables in 15 runs. The relationship between coded and actual values could be described by the following equation:where x is the coded value of the independent variable, X_i is the actual value of the independent variable, X₀ is the actual value of the independent variable at the center point, and ΔX is the step change value of the variables.²³

Table 1 Variables, codes, low and high levels in Box–Behnken design matrix^a

Variables	Levels
	Low (−1)	Center point (0)	High (+1)
a (C): center point.
(X1) initial pH	2	7	12
(X2) MIEX resin dosage (mL L^−1)	0.4	0.8	1.2
(X3) reaction time (min)	20	40	60

---

Run order	Real (coded) values			Response (Y (%))
	X1	X2	X3	Observed	Predicted
1	2 (−1)	1.2 (1)	40 (0)	32.14	29.38
2	7 (0)	0.4 (−1)	60 (1)	13.57	11.74
3	7 (0)	1.2 (1)	60 (1)	22.18	23.74
4(C)	7 (0)	0.8 (0)	40 (0)	13.55	14.55
5(C)	7 (0)	0.8 (0)	40 (0)	17.22	14.55
6	2 (−1)	0.4 (−1)	40 (0)	17.74	18.37
7(C)	7 (0)	0.8 (0)	40 (0)	12.87	14.55
8	7 (0)	1.2 (1)	20 (−1)	10.32	12.16
9	12 (1)	1.2 (1)	40 (0)	30.34	29.70
10	2 (−1)	0.8 (0)	20 (−1)	21.64	22.56
11	12 (1)	0.4 (−1)	40 (0)	15.91	18.67
12	12 (1)	0.8 (0)	20 (−1)	29.43	28.23
13	2 (−1)	0.8 (0)	60 (1)	37.31	38.51
14	7 (0)	0.4 (−1)	20 (−1)	3.70	2.14
15	12 (1)	0.8 (0)	60 (1)	34.39	33.46

---

The quadratic polynomial regression model was used to express the effect of variables in terms of linear, quadratic and cross product terms.

where Y is the predicted response, x_i, x_j are the coded values of the factors and b₀, b_i, b_ii and b_ij are the constant, linear, quadratic and interaction coefficients respectively, n is the amount of variables, e is the error.

In this study, the software Design-Expert 8.0 (Stat Ease, USA) was used to design the experiments and analyze the experimental data. The regression coefficients were used to make statistical calculation to generate dimensional and contour maps from the regression models. Meanwhile the useful information about statistical significance and the characteristics of suitability of predicted quadratic polynomial regression model (eqn (4)) can be taken according to the analysis of variance (ANOVA), the coefficient of determination (R²), the probability p-value (95% confidence level) and Fisher's test.

2.4. DBP formation potential

A sodium hypochlorite solution (free chlorine >5%, Sinopharm Chemical Reagent Co., Ltd., China) was used to prepare free chlorine stock solutions. Chlorination was accomplished head-space free in 40 mL amber vials with Teflon-lined screw caps. The chlorinated HA samples were incubated at 24 ± 1 °C, in the dark, for a 72 h reaction period. Solution pH was buffered at 7 with phosphate solution (KH₂PO₄–NaOH). Preliminary chlorination experiments were performed to determine the Cl₂/DOC (mg as free chlorine/mg DOC) ratio required to provide a free chlorine residual of 3.5 ± 0.2 mg L⁻¹ in each sample. The required mass ratio of chlorine to DOC was 3 mg Cl₂ : 1 mg C. Prior to analysis, the residual chlorine was quenched by ascorbic acid with a double normality of the initial added chlorine normality. THMs were quantified by liquid/liquid extraction with methyl-tertiary-butyl-ether (MTBE) followed by gas chromatography and electron capture detection (GC/ECD), based on USEPA Method 551.1. HAAs were analyzed by liquid/liquid extraction with MTBE followed by derivatization with acidic methanol and by GC/ECD according to USEPA Method 552.2.

2.5. Statistical analysis

Data in figures presented as the mean value of the tested data, and error bars indicated one standard deviation. Data were analyzed through one-way analysis of variance (ANOVA) for comparison of the effect of optimum MIEX process on DBP formation potential using software SPSS 19.0. Comparisons were considered significantly different only if the probability (p) was less than 0.05 and 0.01.

3. Results and discussion

3.1. Selection of the response value

Several surrogate parameters (TOC, DOC, UV absorbance indexes) have been used to quantify NOM reactivity in DBPs formation. Among those parameters, absorbance of ultraviolet light at 254 nm (A₂₅₄) and its specific value (SUVA₂₅₄) are probably the most widely used.²⁴ Although not all DOM is sensitive to UV light and the SUVA₂₅₄ cannot be used to predict DBPFP successfully in waters with low SUVA₂₅₄ (<2 L per mg C per m),^24,25 the UV absorbance and SUVA were generally shown to correlate well with DBP formation, especially in waters with relatively high DOC concentration (>3 mg L⁻¹) and high SUVA₂₅₄ values (>2–3 L per mg C per m).²⁵ In this work, the SUVA₂₅₄ values and DOC level of tested HA solutions are respectively 10.06 L per mg C per m and 3.33 mg L⁻¹. The SUVA value of a water was found to be a good surrogate for hydrophobic, aromatic and high molecular weight (MW) NOM fractions, such as humic and fulvic acids.^26,27 More importantly, the aquatic humic substances within the DOC are thought to be the primary precursors for THMs and HAAs.²⁵ So the SUVA₂₅₄ removal was selected as the optimum response value.

3.2. Establishment of second-order polynomial equation

The adsorption experiments were conducted according to the design matrix and corresponding results were listed in Table 1. The second-order polynomial equation for predicting the optimum point was obtained via the BBD model and input variables. Afterwards, the empirical relationship between the response and the independent variables in the coded units was presented on the basis of the experimental results as follows.

Y = 14.55 + 0.16x₁ + 5.51x₂ + 5.29x₃ + 5.537 × 10⁻³x₁x₂ − 2.68x₁x₃ + 0.5x₂x₃ + 13.87x₁² − 4.38x₂² + 2.28x₃² (5)

where Y is the response value denoted as the SUVA removal percentage and x₁, x₂ and x₃ are the coded term for the three independent variables denoted as initial pH, MIEX resin dosage, and reaction time, respectively. Positive coefficients mean that the corresponding terms affect the response positively and negative values affect it negatively. The predicted SUVA removal from different batches was obtained by analyzing the above second-order polynomial equation (eqn (5)).

3.3. Box–Behnken statistical analysis

3.3.1. Analysis of variance (ANOVA). The analysis of variance (ANOVA) of regression parameters of the predicted response surface quadratic model for HA removal are shown in Table 2. The model F-value (17.04) and lower probability value (p-value < 0.01) imply the regression model is statistically significant. The p-value of lack-of-fit (0.3548) implies the lack-of-fit is not significant and the model can appropriately explain the relationship between response and independent variables. The “Adequacy precision” measures the signal to noise ratio and compares the range of the predicted values at the design points to the mean prediction error. It has been reported that a ratio greater than 4 is necessary and confirms the applicability of the model for navigation of the design space.²⁸ As can be seen in Table 2, the adequate precision of 15.146, in the present case, shows that the model is acceptable. In addition, the R² value of 0.9684 is almost in reasonable agreement with the model adjusted R² value of 0.9116 and predicted R² value of 0.8348. The agreement is fundamental for a good fit of a model.²⁹ The R² value shows that the process can explain about 96% of the model output. The significant of the parameter term is determined by the p-value, and the smaller the value of p, the more significant. In the present investigation, x₂, x₃, x₁² and x₂² are identified as being significant parameter. Insignificant model terms, such as x₁, x₁x₂, x₁x₃, x₂x₃ and x₃², have limited influence on the model and are excluded because of the higher p value (>0.1).

Table 2 ANOVA results for response surface quadratic model^a

Source	Sum of squares	df	Mean square	F-value	p-value > F
a Standard deviation = 2.94, R^2 = 0.9684, R^2 adj = 0.9116, pred R^2 = 0.8348, adequacy precision = 15.146, press = 541.35.
Model	1326.56	9	147.40	17.04	0.0030	Significant
x1-initial pH	0.19	1	0.19	0.022	0.8870
x2-MIEX resin dose	242.73	1	242.73	28.06	0.0032
x3-reaction time	224.26	1	224.26	25.93	0.0038
x1x2	1.23 × 10^−4	1	1.23 × 10^−4	1.42 × 10^−5	0.9971
x1x3	28.67	1	28.67	3.31	0.1283
x2x3	0.98	1	0.98	0.11	0.7497
x1^2	709.90	1	709.90	82.07	0.0003
x2^2	70.89	1	70.89	8.20	0.0353
x3^2	19.16	1	19.16	2.22	0.1968
Residual	43.25	5	8.65
Lack of fit	32.29	3	10.76	1.96	0.3548	Not significant
Pure error	10.96	2	5.48
Cor total	1369.82	14

---

3.3.2. Adequacy of developed mathematical models. The suitability of the selected model to provide adequate approximation of the real system is also confirmed by the diagnostic plots. The plots including normal probability plots of the studentized residuals and the predicted versus actual value plot are usually used to judge the adequacy of a model. Fig. 1 shows the normal probability plot for the studentized residuals for HA removal. Points present along a straight line indicate a normal distribution of the residuals. Some scattering is expected even with normal data. According to Fig. 1, data are normally distributed without any need for response transformation; no apparent issues with normality are found.³⁰ It can be deduced from Fig. 1 that the data are evenly distributed. As given in Fig. 2, the predicted and actual values of response are in good agreement. The results also indicate that the selected quadratic model is effective in assuming the response variables for the experimental data.

Fig. 1 Normal probability plot of the studentized residuals (studentized residuals means the ratio of residuals to standard deviation, which can be used to assess the normal distribution of the residuals).

Fig. 2 Plot of the actual and predicted responses.

3.4. Effect of interactive variables and 3D response plot

3.4.1. Influence of reaction time and adsorbent dosage on HA removal by MIEX resin. The combined effects of MIEX resin dosage and reaction time on SUVA removal percentage are shown in Fig. 3a. Fig. 3a proved that both the independent variables have a little impact on the process of SUVA removal. The removal of SUVA increased as the reaction time increased. Also the SUVA removal increased with an increase in the MIEX resin dosage from 0.4 mL L⁻¹ to 1.2 mL L⁻¹. This may be attributed to the increased surface area and the more provided sorption sites.³¹ As shown in Fig. 3a, the maximal SUVA removal of 24.18% was obtained at a MIEX resin dosage of 1.07 mL L⁻¹ and reaction time of 60 min, with an initial solution pH at the middle value.

Fig. 3 Three dimensional surface plots of HA removal: (a) reaction time – MIEX resin dose; (b) MIEX resin dose – initial pH; (c) reaction time – initial pH.

3.4.2. Influence of initial solution pH and adsorbent dosage on HA removal by MIEX resin. Fig. 3b illustrates the interaction effects of initial solution pH and resin dosage in the response process. The initial solution pH showed a little effect, while a significant effect of MIEX resin dosage on the removal of SUVA is shown in Fig. 3b. Moreover, SUVA removal percentage decreased monotonously with increasing pH in pH range tested (pH 2–pH 6), which can be explained by the changes of physicochemical properties of adsorbent and HA molecules. In general, the adsorbent and adsorbate can be positively or negatively charged depending on its isoelectric point (IEP) and pK_a, respectively. The IEP of MIEX resin is about 6.05, indicating that at pH below 6.05, the MIEX resin may carry positive charges due to the protonation of the amine groups.³² HA is a macromolecule that can possess a negative charge, even at pH levels as low as pH 3 due to the deprotonation of carboxylic groups, although a pH above around neutral is required to enable phenol groups to dissociate.^33,34 The protonation and deprotonation phenomenon induce the electrostatic attraction between HA molecules and the adsorbent surface. With the increasing solution pH, the surface of the adsorbent was less protonated, which leads to the decreased SUVA removal. However, as shown in Fig. 3b, the SUVA removal gradually increases with the increasing solution pH when the pH ranges from 7 to 12, which is not consistent with the results reported by some researchers.^32,35,36 Wang et al. demonstrated the increased pH had an adverse influence on the removal of HA because of the increase in electrostatic repulsive force.^32,35 Similar finding was reported by Li et al.³⁶ Usually, the adsorbent surface could be charged negatively at high pH, and the electrostatic repulsion between HA molecules and the adsorbent may prevent HA molecules from migrating to the adsorbent surface, which may reduce the HA removal. However, the MIEX resin has a strong pH tolerance due to the strong base quaternary ammonium functional groups which give the resin a positive charge in alkaline solutions.³⁷ And the more phenol groups will dissociate besides the carboxylic groups, which will enhance the adsorption removal of HA. That is to say, the characteristic of adsorbent should be considered when discussing the adsorption behavior. Further study is required to verify the removal mechanism of HA by MIEX resin in alkaline condition. Fig. 3b also indicates that the maximum removal (30.30%) occurred under alkaline condition (pH 12) and a MIEX resin dosage of 1.05 mL L⁻¹, which is in accordance with the model.

3.4.3. Influence of reaction time and initial solution pH on HA removal by MIEX resin. In Fig. 3c, the 3D response surface plot was developed as a function of reaction time and initial solution pH at MIEX resin dosage 0.8 mL L⁻¹. From the plot, it is evident that the reaction time has a strong influence on the SUVA removal compared to the initial pH. Fig. 3c shows that when reaction time increased from 20 min to 60 min under constant pH 7 and resin dosage (0.8 mL⁻¹), the removal of SUVA increased from 12.87% to 17.22%. The plot also suggests the SUVA removal is beneficial in strong acid or alkaline medium. This means that higher values of SUVA removal can be obtained by simultaneous increase in reaction time and acid/alkaline conditions. As depicted in Fig. 3c, the maximal removal of 38.11% was achieved at an initial solution pH of 2.03 and reaction time of 59.69 min, while the dosage of MIEX resin was set at the middle value.

3.5. Optimization and validation

The desirability function was used to optimize numerical conditions for HA removal. The desired goal for each operational condition (initial pH, MIEX resin dosage and reaction time) was chosen within the range while the response (SUVA removal) was defined as maximum to achieve the highest performance. The program condition combines the individual desirability into a single number, and then searches to maximize this function.³⁸ Fig. 4 shows the ramp report, which summarizes the optimum predicted value for the independent variables leading to the maximum SUVA removal using the BBD model. The model predicted SUVA removal efficiency of 37.89% corresponding to initial pH 2.30, MIEX resin dosage 0.95 mL L⁻¹, and reaction time 59.18 min; whereas a removal efficiency of 39.74% was obtained from the experiment. The confirmation experimental results were shown in Table 3. Table 3 shows that the predicted value of SUVA removal percent is in agreement with the experimental value with a minor error. It is, therefore, evident that the model is adequate for prediction of HA removal using MIEX resin.

Fig. 4 Ramp report showing the optimum parameters at 1.0 desirability level.

Table 3 Measured and predicted values of HA removal for the confirmation experiments

Conditions			SUVA removal (%)
Initial pH	MIEX resin dose (mL L^−1)	Reaction time (min)	Measured	Predicted
2.30	0.95	59.18	39.74 ± 0.27	37.89

---

3.6. Effect of optimum MIEX process on the DBP formation potential

Fig. 5 illustrates the removal of DBP formation potential (DBPFP) with and without MIEX process for HA solution. The water quality parameter and DBPFP of the HA solution used in this experiment are presented in Table 4. Table 4 shows the DBPFP is respectively 204.78 μg L⁻¹ and 378.22 μg L⁻¹ for THM and HAAs, which is higher than that of commercially aquatic humic substances (177.4 μg L⁻¹ for Suwannee River Fulvic Acid (SRFA, reference no. 1S101F), 198.6 μg L⁻¹ for Suwannee River Humic Acid (SRHA, reference no. 1S101H) and 209.4 μg L⁻¹ for Nordic Lake Fulvic Acid (NLFA, reference no. 1R105F)).^39,40 As shown, MIEX resin treatment obviously lowered DBP formation potentials under the optimal conditions (initial pH 2.30, MIEX resin dosage 0.95 mL L⁻¹ and reaction time 59.18 min). The reduction in chloroform (CF), DCAA and TCAA formation potentials were about 76%, 73% and 80% respectively. The relatively higher removal percentage is because MIEX has great affinity to remove the UV absorbing components of DOC that contribute mostly to the formation of THMs and HAAs.¹⁵ Fig. 6 presents the effect of optimized MIEX process on the molecular weight distribution of HA. Before adsorption, the major molecular weight fractions of HA were found in the ranges of 0.1–0.3 kDa, and 1–18 kDa. After adsorption, a significant decrease in the both molecular weight ranges resulted in a decrease in DBP formation potential. This phenomenon can be explained by the fact that adsorption is effective for removing the higher MW fraction (1–18 kDa) and the larger molecular size DOM fractions are more reactive with respect to the formation of carbonaceous DBP (C-DBP).⁴¹ In addition, there is a highly linear relationship between the characteristic fluorescence intensity and the DBP formation potential of the sample.⁴² As shown in Fig. 7, an obvious decrease in the fluorescence intensity of HA was observed after MIEX treatment, which led to the decrease of DBP derived from HA. In this study, the DBP formation potentials from HA solution before and after optimal MIEX process treatment were also statistically compared using one-way ANOVA, as shown Fig. 5. For any particular DBP, formation potentials were significantly decreased following MIEX treatment (p < 0.05), thereby further suggested that the optimized MIEX process could effectively remove the DBP precursors. That is to say, the MIEX process is a good option for controlling the DBPs derived from HA.

Fig. 5 DBP formation potentials of raw and treated HA solution (error bars indicates one standard deviation. * and ** represent statistically significant differences of p < 0.05 and p < 0.01, respectively. OMIEXP: optimal MIEX process; DCAA: dichloroacetic acid; TCAA: trichloroacetic acid).

Table 4 Characteristics of the humic acid solution

Parameter	Value	Unit
a THMFP = trihalomethanes formation potential.b HAAFP = haloacetic acid formation potential.
Nominal concentration	10.00 ± 0.1	mg L^−1
pH	6.50 ± 0.1
DOC	3.33 ± 0.03	mg L^−1
UV254	0.335 ± 0.001	cm^−1
SUVA	10.06 ± 0.06	L mg^−1 m^−1
Conductivity	380.00 ± 10	μS cm^−1
THMFP^a	204.78 ± 8.72	μg L^−1
HAAFP^b	378.22 ± 3.22	μg L^−1

---

Fig. 6 Molecular weight distribution of HA before and after MIEX process treatment.

Fig. 7 EEM contour plots of the HA before (a) and after (b) MIEX process treatment.

4. Conclusions

Response surface methodology was successfully applied to optimize HA removal for controlling the DBP formation. SUVA was an appropriate surrogate parameter in terms of quantifying HA reactivity in DBPs formation. At optimal operating conditions of pH 2.30, MIEX resin dosage 0.95 mL L⁻¹ and reaction time 59.18 min, 39.74% removal of SUVA from HA was achieved. The effects of interactions of reaction time and MIEX resin dosage, pH and MIEX resin dosage, pH and reaction time on the removal of HA by MIEX resin were not significant. The optimized MIEX process was shown to effectively reduce the DBPs formation derived from HA. The reduction in CF, DCAA and TCAA formation potentials were about 76%, 73% and 80% respectively. Future research should evaluate the control of DBPs formation by MIEX process for waters having low SUVA values. Also, it would be worthwhile to consider the effect of MIEX process on algal organic matter (AOM) disinfection performance.

Acknowledgements

This research work was supported by the National Major Project of Science & Technology Ministry of China (2008ZX07421-002 and 2012ZX07403-001 and 2012ZX07403-002), National Natural Science Foundation of China (51178321), and the Specialized Research Fund for the Doctoral Program of Higher Education (20120072110050).

References

1. M. Drikas, M. Dixon and J. Morran, Water Res., 2011, 45, 1539–1548.
2. T. Bond, E. H. Goslan, S. A. Parsons and B. Jefferson, Water Res., 2010, 44, 1645–1653.
3. I. N. Najm and S. W. Krasner, Am. Water Works Assoc., J., 1995, 87, 106–115.
4. C. J. Johnson and P. C. Singer, Water Res., 2004, 38, 3738–3750.
5. H. Humbert, H. Gallard, H. Suty and J. P. Croue, Water Res., 2008, 42, 1635–1643.
6. S. W. Krasner, H. S. Weinberg, S. D. Richardson, S. J. Pastor, R. Chinn, M. J. Sclimenti, G. D. Onstad and A. D. Thruston, Environ. Sci. Technol., 2006, 40, 7175–7185.
7. A. Mehrparvar, A. Rahimpour and M. Jahanshahi, J. Taiwan Inst. Chem. Eng., 2014, 45, 275–282.
8. N. A. A. Hamid, A. F. Ismail, T. Matsuura, A. W. Zularisam, W. J. Lau, E. Yuliwati and M. S. Abdullah, Desalination, 2011, 273, 85–92.
9. W. W. Tang, G. M. Zeng, J. L. Gong, J. Liang, P. Xu, C. Zhang and B. B. Huang, Sci. Total Environ., 2014, 468, 1014–1027.
10. T. H. Boyer and P. C. Singer, Water Res., 2006, 40, 2865–2876.
11. P. C. Singer, T. Boyer, A. Holmquist, J. Morran and M. Bourke, Am. Water Works Assoc., J., 2009, 101, 65–73.
12. P. C. Singer and K. Bilyk, Water Res., 2002, 36, 4009–4022.
13. D. A. Fearing, J. Banks, S. Guyetand, C. M. Eroles, B. Jefferson, D. Wilson, P. Hillis, A. T. Campbell and S. A. Parsons, Water Res., 2004, 38, 2551–2558.
14. T. H. Boyer and P. C. Singer, Water Res., 2005, 39, 1265–1276.
15. X. J. Gan, T. Karanfil, S. S. K. Bekaroglu and J. H. Shan, Water Res., 2013, 47, 1344–1352.
16. R. S. Kingsbury and P. C. Singer, Water Res., 2013, 47, 1060–1072.
17. U. K. Garg, M. P. Kaur, V. K. Garg and D. Sud, Bioresour. Technol., 2008, 99, 1325–1331.
18. A. Witek-Krowiak, K. Chojnacka, D. Podstawczyk, A. Dawiec and K. Pokomeda, Bioresour. Technol., 2014, 160, 150–160.
19. L. Ding, X. Lu, H. P. Deng and X. X. Zhang, Ind. Eng. Chem. Res., 2012, 51, 11226–11235.
20. X. X. Zhang, X. Lu, S. Li, M. Y. Zhong, X. D. Shi, G. Luo and L. Ding, J. Taiwan Inst. Chem. Eng., 2014, 45, 1835–1841.
21. S. C. Lenore, E. G. Arnold and D. E. Andrew, American Public Health Association, American Water Works Association and World Environment Federation, 20th edn, Washington DC, 1998.
22. M. Jain, V. K. Garg and K. Kadirvelu, Bioresour. Technol., 2011, 102, 600–605.
23. M. A. Bezerra, R. E. Santelli, E. P. Oliveira, L. S. Villar and L. A. Escaleira, Talanta, 2008, 76, 965–977.
24. M. Kitis, T. Karanfil, J. E. Kilduff and A. Wigton, Water Sci. Technol., 2001, 43, 9–16.
25. N. Ates, M. Kitis and U. Yetis, Water Res., 2007, 41, 4139–4148.
26. J. K. Edzwald and J. E. Van Benschoten, in Chemical water and wastewater treatment, Springer, 1990, pp. 341–359.
27. M. C. White, J. D. Thompson, G. W. Harrington and P. C. Singer, Am. Water Works Assoc., J., 1997, 89, 64–77.
28. A. A. L. Zinatizadeh, A. R. Mohamed, M. D. Mashitah, A. Z. Abdullah and M. H. Isa, Biochem. Eng. J., 2007, 35, 226–237.
29. L. Mohajeri, H. A. Aziz, M. H. Isa and M. A. Zahed, Bioresour. Technol., 2010, 101, 893–900.
30. A. Ozer, G. Gurbuz, A. Calmli and B. K. Korbahti, J. Hazard. Mater., 2008, 152, 778–788.
31. Z. Q. Liu, X. M. Yan, M. Drikas, D. N. Zhou, D. S. Wang, M. Yang and J. H. Qu, J. Environ. Sci., 2011, 23, 381–387.
32. J. H. Wang, X. J. Han, H. R. Ma, Y. F. Ji and L. J. Bi, Chem. Eng. J., 2011, 173, 171–177.
33. G. Abate and J. C. Masini, Colloids Surf., A, 2003, 226, 25–34.
34. L. P. Weng, W. H. Van Riemsdijk and T. Hiemstra, J. Colloid Interface Sci., 2007, 314, 107–118.
35. J. H. Wang, L. J. Bi, Y. F. Ji, H. R. Ma and X. L. Yin, J. Colloid Interface Sci., 2014, 430, 140–146.
36. C. J. Li, Y. Dong, D. Y. Wu, L. C. Peng and H. A. Kong, Appl. Clay Sci., 2011, 52, 353–357.
37. P. A. Neale, M. Mastrup, T. Borgmann and A. I. Schafer, J. Environ. Monit., 2010, 12, 311–317.
38. M. H. Isa, E. H. Ezechi, Z. Ahmed, S. F. Magram and S. R. M. Kutty, Water Res., 2014, 51, 113–123.
39. F. J. Rodríguez, P. Schlenger and M. García-Valverde, Sci. Total Environ., 2016, 541, 623–637.
40. M. Y. Z. Abouleish and M. J. M. Wells, Environ. Chem., 2012, 9, 450–461.
41. C. Xia, D. Ma, B. Gao, X. Hu, Q. Yue, Y. Meng, S. Kang, B. Zhang and Y. Qi, Bioresour. Technol., 2016, 211, 183–189.
42. R. Hao, H. Ren, J. Li, Z. Ma, H. Wan, X. Zheng and S. Cheng, Water Res., 2012, 46, 5765–5776.

---