Surrogate-Based Correlation Models in View of Real-Time Control

Subscriber access provided by UNIV LAVAL

Article

Surrogate-based correlation models in view of realtime control of ozonation of secondary treated municipal wastewater - model development and dynamic validation Michael Chys, Wim T.M. Audenaert, Emma Deniere, Severine Therese F. C. Mortier, Herman Van Langenhove, Ingmar Nopens, Kristof Demeestere, and Stijn W.H. Van Hulle Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.7b04905 • Publication Date (Web): 27 Nov 2017 Downloaded from http://pubs.acs.org on November 28, 2017

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Environmental Science & Technology is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 44

Environmental Science & Technology

1

Surrogate-based correlation models in view of real-

2

time control of ozonation of secondary treated

3

municipal wastewater - model development and

4

dynamic validation

5

Michael Chys1,*, Wim T.M. Audenaert1,‡, Emma Deniere1,2, Séverine Thérèse F.C. Mortier3,4,

6

Herman Van Langenhove2, Ingmar Nopens3, Kristof Demeestere2,†, Stijn W.H. Van Hulle1,†

7

1

8

Graaf Karel de Goedelaan 5, B-8500 Kortrijk, Belgium

9

2

LIWET, Department of Industrial Biological Sciences, Ghent University Campus Kortrijk,

EnVOC, Department of Sustainable Organic Chemistry and Technology, Ghent University,

10

Coupure Links 653, B-9000 Ghent, Belgium

11

3

12

University, Coupure Links 653, B-9000 Ghent, Belgium

13

4

14

Ghent, Belgium

15

‡

16

B-9112 Sint-Niklaas, Belgium

17

†

BIOMATH, Department of Mathematical Modelling, Statistics and Bioinformatics, Ghent

Department of Pharmaceutical Analysis, Ghent University, Ottergemsesteenweg 460, B-9000

present affiliation: AM-TEAM, Advanced Modelling for Process Optimisation, Hulstbaan 63,

These authors contributed equally to the work.

ACS Paragon Plus Environment

1


Page 2 of 44

18

ABSTRACT

19

New robust correlation models for real-time monitoring and control of trace organic contaminant

20

(TrOC) removal by ozonation are presented, based on UVA254 and fluorescence surrogates, and

21

developed considering kinetic information. The abatement patterns of TrOCs had inflected

22

shapes, controlled by the reactivity of TrOCs towards ozone and HO radicals. These novel and

23

generic correlation models will be of importance for WRRF operators to reduce operational costs

24

and minimize by-product formation. Both UVA254 and fluorescence surrogates could be used to

25

control ∆TrOC, although fluorescence measurements indicated a slightly better reproducibility

26

and an enlarged control range. The generic framework was validated for several WRRFs and

27

correlations for any compound with known kinetic information could be developed solely using

28

the 2nd order reaction rate constant with ozone (kO3). Two distinct reaction phases were defined

29

for which separate linear correlations were obtained. The first was mainly ozone controlled,

30

while the second phase was more related to HO reactions. Furthermore, parallel factor analysis

31

of the fluorescence spectra enabled monitoring of multiple types of organic matter with different

32

O3 and HO• reactivity. This knowledge is of value for kinetic modelling frameworks and for

33

achieving a better understanding of the occurring changes of organic matter during ozonation.

34 35


2

Page 3 of 44


36

INTRODUCTION

37

Water resource recovery facilities (WRRFs) have been identified as a major pathway through

38

which trace organic contaminants (TrOCs) enter the aquatic environment.1 TrOCs such as

39

pharmaceuticals may trigger unwanted ecological effects (e.g. bacterial resistance, chronic

40

toxicity, endocrine disruption and feminization of fish) or might expose human individuals due

41

to increased water reuse.2–5 Driven by pending (European) legislation6,7 and/or as a precaution to

42

protect the aquatic environment and drinking water sources, operators of WRRFs are preparing

43

for plant upgrades with advanced tertiary treatment. Ozonation, activated carbon filtration and

44

membrane filtration have been extensively tested in this respect.8 Ozonation is a major

45

technology currently being implemented in Switzerland (regulatory driven) but also in other

46

countries.6,9 At ozone dosages up to 1 g O3 g-1 DOC (dissolved organic carbon), the parent

47

TrOCs can be removed above 80 % or even below detection limits for components with kO3 > 10

48

M-1 s-1.10

49

Although WRRF effluent ozonation is rapidly growing, currently applied control strategies for

50

ozone dosage (e.g. flow-based) result in sub-optimal operation. Not only the effluent flow rate,

51

but also the ozone demand of secondary effluent are highly dynamic in time. Optimal ozone

52

dosing is required to 1) lower operational costs, 2) consistently and adequately remove target

53

TrOCs and 3) minimize by-product formation. With respect to the latter, bromate and N-

54

nitrosodimethylamine (NDMA), both examples of carcinogenic by-products, can be formed

55

when exceeding effluent-specific threshold values, mostly at higher ozone dosages.11–13 From a

56

monitoring perspective, frequent TrOC monitoring is costly and labor intensive.


3


Page 4 of 44

57

Therefore, readily available and real-time surrogate measurement techniques are essential for

58

both process monitoring and ozone dose control. Effluent organic matter (EfOM) is a main

59

influencer of the ozonation process,14 consuming significant amounts of ozone and indirectly

60

initiating hydroxyl radical (HO•) production.15,16 As EfOM is easily characterized using (online)

61

measurements, the monitoring of ozone and HO• induced changes of EfOM can be translated

62

into indirect monitoring of oxidant exposure, and hence TrOC removal. The following

63

measurements have shown to correlate well to ozone dose and TrOC removal: dissolved organic

64

carbon (DOC)10,17,18, UV absorption at 254 nm (UVA254)19–26 and total fluorescence (TF)22.

65

So far, mostly linear relationships were considered which only appear to produce a good

66

correlation for TrOCs highly susceptible to ozone (i.e. mainly O3 rather than HO• controls their

67

degradation). Often moderate or poor linear correlations are achieved between the decrease of

68

the surrogate (mostly ∆UVA254) and the removal of ozone-recalcitrant TrOCs.10,27 This might

69

lead to an over or under prediction of the actual TrOC removal, potentially resulting in

70

inefficient TrOC abatement or ozone overdosing, associated with higher operational costs and/or

71

a potential for by-product formation. Components with a low reactivity towards ozone (kO3 < 10

72

M-1 s-1) show a more convex curved behavior with only limited degradation in ∆UVA254 ranges

73

below 20%.24,27 In literature, it is insufficiently addressed how the TrOC-specific properties (e.g.

74

kO3) are influencing the linearity of these correlations. HO• production will clearly have a greater

75

influence on the removal of TrOCs with a lower affinity for direct ozone reactions. Kinetic

76

modelling efforts showed significant deviations between measured and predicted removal for the

77

more recalcitrant TrOCs.18 It became clear that HO• production is highly dependent on the

78

specific organic matrix. More recently, Chon et al.28 indicated a shift in electron donating

79

capacity of the EfOM when increasing the ozone dose, influencing the relation between ∆TrOCs


4

Page 5 of 44


80

and ∆UVA254. Especially ozone resistant species showed no significant removal within a first

81

phase of reaction. Ozone decrease follows pseudo first order kinetics after an initial, fast reaction

82

phase.29 Consequently, this shift in reaction pathways is likely to influence the used correlation

83

models. Further elaboration, with the aim of providing generic models independent of TrOCs

84

and effluent samples used during development is now needed.

85

Recently, the use of fluorescence spectroscopy has gained a lot of interest.22,25 Compared to a

86

single-wavelength UV absorption approach, a more detailed view of the entire (effluent) organic

87

matrix (e.g. different chemical moieties) can be extracted from these type of measurements.

88

Empirical correlations between differential TF (∆TF) and removal of a wide range of TrOCs

89

have been developed.22 Although much information is present, often (parts of) the fluorescence

90

Excitation Emission Matrices (EEMs) are used as such, without any statistical processing.

91

Transformation rates may differ among different major groups of components (i.e. humic and

92

fulvic acid-like, soluble microbial products).30 Statistical methods, deriving the component

93

specific zones within the EEMs, are useful to study these transformations. For example, parallel-

94

factorial (PARAFAC) analysis has been applied to monitor natural water,31 and different stages

95

of wastewater treatment.32,33 Although, PARAFAC analysis is a multi-way method that requires

96

model development, a certain number of samples and several minutes of instrumental

97

measurement, it derives the wavelengths of interest within complex EEMs. In view of online

98

monitoring and control, the measurement time can be significantly reduced by only selecting the

99

EEM regions of interest.

100

Given the limitations of the current (linear) surrogate models and the possibility to use different

101

spectral measurements, the aim of this study was to develop improved correlation models based

102

on UVA254 and fluorescence for TrOCs having a wide range of ozone reactivity (kO3 from < 1 to


5


103

106 M-1 s-1) and for a broad range of ozone doses (up to ± 2 g O3 g-1 DOCeq). The shape of the

104

correlations was studied to gain in depth process understanding and to provide generic

105

knowledge that can be used to construct correlations for any compound for which kO3 and kHO•

106

are known. Consequently, a generic framework for an unlimited amount of TrOCs was put

107

forward and validated for several WRRFs. Fluorescence data was thoroughly processed using

108

PARAFAC to study the EfOM moieties and their reactivity towards ozone and HO•.

Page 6 of 44

109 110

MATERIALS AND METHODS

111

Standards and reagents

112

All chemicals used were of analytical grade with a purity of at least 98%. Nine TrOCs (all

113

pharmaceuticals) were selected based on their ozone reactivity, environmental relevance, and

114

occurrence in WRRF effluents to develop the model. Individual stock solutions of 1 g TrOC L-1,

115

stored at -18°C in the dark, were used to prepare diluted mixtures. The studied TrOCs can be

116

classified in three groups according to their reaction rates with ozone (Table S1). This

117

classification is rather arbitrary and – similar to other studies23,34,35 – mainly aimed to facilitate

118

the discussion. The individual second order reaction rate constants (kO3) of the TrOCs will be

119

most important in developing the correlation model framework.

120 121 122


6

Page 7 of 44


123

Experimental procedures

124

Secondary effluent samples were collected from a conventional activated sludge process at the

125

WRRF Harelbeke, Belgium (116,100 I.E., operated by Aquafin NV). Three sampling campaigns

126

were performed at different times (further denoted as effluent 1, 2 and 3 – effluent 3 was used to

127

verify conclusions drawn based on effluent 1 and 2). Clear differences in effluent characteristics

128

were noticed when rain fall occurred before sampling (24 and 72 hours), due to dilution effects.

129

Data on the effluent characteristics and rain fall before sampling can be found in Table S2.

130

Batch ozonation experiments at room temperature were conducted in 1 L glass reactors equipped

131

with a mechanical mixer at a speed of 200 rpm (Janke & Kunkel GmbH & Co.KG, Germany).

132

Effluent was spiked with a mixed stock solution of the selected TrOCs up to a concentration of

133

10 µg L-1 (effluent 1) and 1 µg L-1 (effluent 2 and 3) for each compound. These different

134

concentrations were applied to assess the impact of TrOC levels on the correlations. Freshly

135

prepared ozone stock solution was added to the spiked effluent with ozone doses varying

136

between 0 and 17.3 mg L-1 (or a specific dose up to ± 2 g O3 g-1 DOCeq). The stock solution had

137

an ozone concentration of around 90 mg O3 L-1 and was prepared by purging ice-cooled distilled

138

water with an ozone/oxygen gas mixture (300 mL min-1) using an oxygen-fed ozone generator

139

(up to 8 g O3 h-1, COM-AD-02, Anseros GmbH, Germany). After ozone addition and complete

140

reaction (30 min), each reactor was brought to 800 mL with distilled water. As such, equal

141

dilution was obtained, independent of the volume of ozone stock added (i.e. dilution < 20 %).

142

The samples were analyzed for surrogate parameters and TrOCs, after storage for maximum 3

143

weeks at 4°C and -20°C, respectively. Information on handling samples for TrOC analysis is

144

given in SI-Text 1.


7


Page 8 of 44

145

For validation of the proposed correlation model, additional effluent was collected from four

146

other municipal WRRFs across Belgium with capacities ranging from 500 to 80,000 I.E. The

147

plants were selected based on receiving water (i.e. influent), treatment train configuration (i.e.

148

size and type of treatment) and effluent characteristics. Details of those WRRFs are given in

149

Table S3. The diverse effluent water characteristics are displayed in Figure S1 and discussed in

150

SI-Text 2.

151

Six effluent points have been sampled twice except for the Waregem plant (only once before and

152

after a tertiary sand filtration step) resulting in a total of 10 different effluents. Ozonation

153

experiments were identical to those for model development although effluent was spiked with a

154

stock solution of pCBA (100 µg L-1) as HO• probe compound (kO3 < 0.1 M-1 s-1, kHO• = 5.0×109

155

M-1 s-1).36 It could be assumed that, if pCBA was removed to a certain level, other more ozone

156

susceptible TrOCs would at least be removed to the same extent. Six different ozone doses,

157

ranging between 0 and 15 mg O3 L-1, were added to each effluent.

158 159

Analytical methods

160

TrOCs were analyzed using a validated SPE-UHPLC-HRMS method, making use of a benchtop

161

Q-ExactiveTM Orbitrap mass spectrometer (Thermo-Fisher Scientific, USA). Details about the

162

analytical procedure and validation characteristics are provided by Vergeynst et al.37 and SI-Text

163

1. Alkalinity (mg L-1 CaCO3) was determined according to standard methods.38 Nitrite (NO2--N),

164

nitrate (NO3--N), ammonium (NH4+-N) and COD were determined spectrophotometrically

165

following the standard methods38 using Hach-Lange cuvettes and a DR2800 spectrophotometer

166

(Hach, Belgium). Conductivity (EC) and pH were registered by a HQ30D (Hach-Lange,


8

Page 9 of 44


167

Belgium). Turbidity was measured with a portable Hi 98703 (Hanna Instruments, USA). The

168

instantaneous ozone demand (IOD), defining the rapid ozone consumption within 5 seconds after

169

dosing, was determined based on Hoigné & Bader39 and Roustan et al.40 More information is

170

given in SI-Text 3. Ozone concentrations in the stock solution were measured based on the

171

indigo method of Bader & Hoigné.41 HO• exposure was determined (see SI-Text 4) using either

172

pCBA (during effluent characterization) or metronidazole (during model development) as probe

173

components due to their low kO3 (pCBA: < 0.1 M-1 s-1; metronidazole: < 1 M-1 s-1)34,36 and their

174

high kHO• (pCBA: 5.0×109 M-1 s-1; metronidazole: = 6.2×109 M-1 s-1)34,36. HPLC-DAD was used

175

to measure pCBA concentrations (see SI-Text 4).

176

UV-Visible (UV-VIS) absorption spectra between 200 and 800 nm with 0.5 nm increments were

177

obtained using a Shimadzu UV-1601 spectrophotometer and 1cm quartz cuvettes. Fluorescence

178

EEMs were obtained using 1 cm quartz-cuvettes and a Shimadzu RF-5301 Fluorimeter. No

179

dilution was performed as UVA254 < 0.3 cm-1 for all samples.31,42,43 Fluorescence intensities were

180

measured at excitation wavelengths of 220-450 nm in 5 nm increments, and emission

181

wavelengths of 280-600 nm in 1 nm increments. Excitation and emission slit widths were set at 5

182

nm, and a response time of 0.25 s was utilized. Raman scans of distilled water were obtained at

183

an excitation wavelength of 350 nm over an emission wavelength range of 365-450 nm in 0.2 nm

184

increments, for the calculation of the Raman peak. The area of the Raman peak was used to

185

normalize the fluorescence intensity of all spectra, finally expressed as RU (Raman units).44,45

186 187 188


9


Page 10 of 44

189

PARAFAC analysis

190

Fluorescence spectral corrections and PARAFAC analysis were performed by adapting the

191

drEEM® toolbox in Matlab.44 Correction of fluorescence EEMs was undertaken to minimize

192

bias due to sample and instrumental related variability. Raw spectral data were blank corrected

193

and sample specific matrices of correction factors for inner filter effects were applied, calculated

194

based on UV absorbance spectra. Normalization to RU was performed afterwards and Rayleigh-

195

Tyndell and Raman scatter lines were removed for quantitative analysis and to allow EEMs to be

196

displayed uniformly. All samples for model development were used for subsequent PARAFAC

197

analysis and for integration to determine the total fluorescence (TF) of each sample.

198

Fluorescence intensities below an excitation wavelength of 225 nm and an emission wavelength

199

of 281 nm were excluded from all EEMs for PARAFAC analysis based on leverage plots.

200

Outlier samples were excluded based on outlier tests (i.e. examining the structure in the error

201

residuals) and leverage plots. A non-negativity constraint and random initialization were applied

202

for the models. A model with four components could be identified (Figure S2). All components

203

could be associated with different types of chemical groups with specific properties according to

204

Chen et al.30 Split-half analysis was performed for validation of this model using following

205

settings: four alternating determined splits, three runs per model of two combined splits and a 10-

206

10

207

‘overall result’ of the drEEM toolbox. Further details comparing the different developed models

208

(i.e. loadings) during split-half validation are given in Figure S3-6. Intensity values at peak

209

wavelengths of the defined PARAFAC components are expressed as Fmax1, 2, 3 and 4 (in RU).

210

The development requires significant computational power and is time consuming, making it

211

impossible to perform online for control applications. After successfully constructing the

convergence criteria. The model was ‘validated for all comparisons’ as stated by the given


10

Page 11 of 44


212

PARAFAC model, this model can be applied to other measurements without any additional

213

model construction or development necessary (as is done within this research).

214

Component one (Ex. 235 & 310 nm; Em. 385 nm) and two (Ex. 235 & 380 nm; Em. 487 nm)

215

both represent fulvic acid-like moieties based on prior studies defining the region of intensities

216

related to specific chemical moieties.30 Component two is expected to consist of hydrophobic

217

compounds whereas component one is probably less hydrophobic with a smaller molecular size.

218

Ishii & Boyer 32 also showed, indicated by the shorter excitation and emission peak wavelengths,

219

a positive association between compounds in the former region and a smaller molecular size.

220

Component three (Ex. 260 & 350 nm; Em. 440 nm) is affiliated with humic acid-like moieties,

221

hydrophobic in nature, and a rather large molecular size (indicated by higher peak wavelengths).

222

Tryptophan and protein-like compounds with a rather small molecular size exhibit fluorescence

223

intensity in the region of component four (Ex. 225 & 280 nm; Em. 345 nm).30 More information

224

on the location of the defined fluorescence components is given in Table S4. Although the main

225

goal of this manuscript was to develop a reliable framework of correlation models rather than to

226

define specific regions in EEM spectra, these peak allocations provided supporting information

227

for data interpretation.

228 229

Model validation

230

The correspondence between measured and predicted data was evaluated using (i) an unpaired t-

231

test of which the null-hypothesis (difference is not significant) was rejected by p < 0.05 and

232

which could also be used to compare different models with equal sample size, (ii) the Theil’s

233

inequality coefficient (TIC) of which a value below 0.3 is commonly seen as an indicator for a


11


Page 12 of 44

234

good agreement (see SI-Text 5)46 and (iii) a visual inspection to detect outliers or regions with

235

deviations supported with the determination of the MAE (mean averaged error). The predictive

236

power of the newly developed model was compared to single correlation models (1) developed

237

with the same data as for the new developed model and (2) earlier published models17,19,20,22

238

specifically developed for the prediction of pCBA removal based on ∆UVA254. The accuracy of

239

the new model in comparison with single correlation models was best exemplified when using

240

UVA254 as model input since EEM deduced PARAFAC components are not as widely used in

241

literature for describing correlations with ∆TrOCs. Several authors mention correlations for

242

selected TrOCs while using UVA254, but the number of TrOCs in these studies was limited and

243

the selection of TrOCs was different. The correlation of ∆pCBA with ∆UVA254, on the contrary,

244

is mentioned by multiple authors and forms a good basis for comparison. The equations and

245

slopes to establish the different single correlation models are given in SI-Text 6 and Table S5-S6.

246 247

RESULTS AND DISCUSSION

248

Surrogate correlation patterns of TrOCs

249

The abatement pattern of each TrOC was related to changes in UVA254, TF and Fmax1-4. Figure 1

250

shows the percentage of elimination of each TrOC in relation with the decrease of UVA254 (a)

251

and Fmax1 (b), respectively. The TrOC abatement in relation with the decrease of TF and other

252

PARAFAC components (Fmax2, 3 and 4) is given in Figure S7-S10. Group I compounds

253

(diclofenac, levofloxacin and trimethoprim) with the highest kO3 showed complete removal at

254

25% ∆UVA254, 52% ∆TF, 54% ∆Fmax1, 37% ∆Fmax2, 61-65% ∆Fmax3 and 51% ∆Fmax4. Group II

255

compounds (amitriptyline, ciprofloxacin and venlafaxine) only showed complete removal


12

Page 13 of 44


256

starting from 29-33% ∆UVA254, 56-66% ∆TF, 60-70% ∆Fmax1, 42-54% ∆Fmax2, 68-73% ∆Fmax3

257

and 57-70% ∆Fmax4. Amantadine, flumequine and metronidazole (Group III) were not

258

completely removed at the observed maximal ∆UVA254 (47%), ∆TF (81%), ∆Fmax1 (84%),

259

∆Fmax2 (70%), ∆Fmax3 (83%) and ∆Fmax4 (90%). It should, however, be stated that the data

260

related with Fmax4 were highly scattered (hence its 4th rank during PARAFAC analysis) which

261

makes it hard to draw strong conclusions from this surrogate parameter. Nevertheless, other Fmax

262

parameters showed clear abatement patterns with only limited scattering.

263 264

TrOC abatement related to ∆UVA254

265

A linear relationship, as assumed by most authors19–26, was not able to describe in a

266

representative way the observed trend over the full ∆UVA254 and ∆TrOC range in Figure 1a,

267

especially for the group III compounds. The shape of the curves is more complex and can be

268

related to the main occurring reaction mechanisms. At low ozone doses, and corresponding low

269

∆UVA254, removal of pollutants and EfOM is mainly due to direct ozone reactions. This is

270

supported by the relatively high values for rapid ozone consumption, often referred to as

271

“instantaneous” ozone demand, (7.7 and 10 mg O3 L-1, considering dilution during

272

experimentation and nitrite correction). Systematically increasing the ozone dose (and

273

consequently higher ∆UVA254) results in a significant increased removal of the pharmaceuticals

274

within group III, while the removal of group I compounds slightly levelled off as function of

275

∆UVA254. Overall, two main hypotheses are put forward to explain the observed phenomena: (i)

276

transformation of the organic matrix leads to products with higher HO• production yields, as

277

proposed by Audenaert et al.19 and Lee et al.10, and/or (ii) after an initial reaction phase with fast


13


Page 14 of 44

278

direct ozone reactions, the residual dissolved ozone can form HO• via autocatalytic

279

decomposition.16 It is clear that HO• production relies on a complex mechanism that is

280

significantly affected by oxidant induced changes of EfOM.29 During the initial phase, very high

281

transient HO• concentrations have been observed, playing an important role during oxidation.29

282

For example, HO• exposure of all effluents, measured during IOD determination, amounted on

283

average (1.37 ± 0.24)×10-10 M.s, i.e. 56% of the maximum obtained HO• exposure during all

284

experiments (see further, Figure 3). The degree of hydroxylation when ozonation proceeds might

285

increase and new electron rich moieties (e.g. phenolic compounds) may be formed which

286

accelerate HO• production.19,29,47 Prolonged ozonation results in dissolved ozone residual, and

287

the process can be assumed to be further controlled radical chain reactions and less by direct

288

reactions with EfOM moieties. This in contrast with the initial phase in which chain reactions or

289

autocatalytic decomposition seem not to play a key role.48 Such radical chain reactions are

290

extensively described elsewhere.16,49

291

Based on the reactions discussed above, the entire range of ∆UVA254 can be divided into two

292

distinct regions (Figure 1a). In phase 1, rapid ozone reactions dominate whereas in phase 2,

293

indirect (less selective) HO• radicals dominate the reaction pathways. For both phases and for

294

each TrOC, a distinct linear relationship could be obtained between ∆TrOC and ∆UVA254. No

295

intercept was considered for the first part as it was not statistically significant (p > 0.05 for all

296

nine TrOCs separately). The location of the inflection point (19% ∆UVA254) was determined

297

minimizing the standard error between the predicted values (based on the two part linear

298

function) and the experimental data. A more detailed approach compared to most already

299

published studies (e.g.20–23) was used, as up to 20 data points/measurements were taken during

300

each experiment.


14

Page 15 of 44


301

Relating the slope of phase 1 (ap1, obtained from Figure 1a) to kO3 showed a logarithmic relation

302

with a good fit (R² = 0.77, n = 7, Figure 2a). Therefore, a linear regression was performed

303

according to eq. 1. Logically, ozone-based degradation of TrOCs in a near neutral pH

304

environment strongly depends on kO3.10 Zimmerman et al.12 already indicated that mostly a

305

correlation between the removal of the component and the kO3 occurs. Lee et al.34 observed

306

similar TrOC abatement in different WRRF effluents when normalizing the ozone dose to the

307

initial DOC. The relationship as presented in Figure 2a (and eq. 1) can be generically used to

308

construct correlations for every TrOC for which kO3 is known.

309

The correlation in phase 2 of Figure 1a shows no dependency on the kO3 (Figure S11a). In

310

addition, no statistical difference was noticed between the average slopes of each TrOC group,

311

with an overall averaged slope of 2.54 ± 0.60. This supports the hypothesis that, independent of

312

the affinity of each TrOC towards ozone, unselective reactions with HO• play a more prominent

313

role in phase 2. In contrary to the kO3 values of the nine TrOCs which span several orders of

314

magnitude, the kHO has a much smaller variation (see Table S1). Consequently, it is difficult to

315

make a strong correlation between the kHO and slope in phase 2.

316

The given approach is somewhat in line with recent available literature also including a two

317

phase relation between ∆TrOCs and ∆UVA254 or ∆TF. Park et al.50 used a kinetic approach by

318

including the Rct constant and empirically determined parameters, attributing the contribution of

319

ozone and HO during two reaction phases. Also Nanaboina and Korshin24 used a kinetic

320

approach including empirically determined parameters (i.e. water or TrOC specific) to correlate

321

∆TrOCs and ∆UVA254. However, the currently considered fixed empirical parameters are

322

depending on a varying water quality (or are TrOC specific). Therefore some additional offline

323

work (i.e. plant or effluent specific empirical parameters such as e.g. Rct) still needs to be done


15


Page 16 of 44

324

before applying the model in a full-scale installation. For online control, the empirical correlation

325

model of the current work seems advantageous at the current timing as no additional offline

326

work is necessary once the model is established, although both types of models (requiring

327

effluent dependent offline work or not) are of value for full-scale applications.

328

The above observations taken into account, the ∆TrOC can be predicted using eq. 2 (∆S ≤ I) and

329

eq. 3 (∆S ≥ I), with ∆S the decrease of the surrogate measurement (e.g. ∆UVA254) and I the

330

inflection point between both correlations. This prediction allows for online monitoring if

331

combined with eq. 1. Non-continuous correlations were used for curve fitting to handle both

332

phases separately and to be able to interpret the relations during each phase independently.

333

Correlations for the TrOCs (excluding amantadine) showed minor changes (< 7% for ∆UVA254)

334

near the intersection or inflection point, resulting in almost equality of both curves for most

335

TrOCs at the inflection point. The deviation of amantadine had to deal with a lot of scattering,

336

especially in the low ∆UVA254 region. This component was consequently not used further for the

337

regression between the slope of phase 1 and kO3.

338

The uncertainty on both ap1 and kO3 has been experimentally determined during previous curve

339

fitting (ap1) or has been based on a conventional order of error that is often seen for reaction rate

340

constant (i.e. 50% for kO3). These uncertainties might explain the outlier value of venlafaxine

341

(Figure 2a). Noteworthy is that kO3 values (at pH 7) should be used with caution and

342

uncertainties thereon should always be considered as actual values can be strongly influenced by

343

experimental errors or by even slight changes in conditions (such as pH).15 Also the presence of

344

large molecular structures, e.g. humic acid-like moieties, having low dissociation rates can affect

345

species speciation having pKa values close to near neutral pH (i.e. the natural effluent pH). These

346

uncertainties have been taken into account for the determination of both m and b (and the error


16

Page 17 of 44


347

thereon) in the linear regression (eq. 1) by using the technique proposed by York et al.51

348

Applying Sobol sampling52 with 10.000 samples, the uncertainty on the model predictions was

349

determined. The model parameters included in the analysis and their uncertainty are also listed in

350

Table S7. Finally, the 95% confidence intervals for ∆UVA254, ∆TF and ∆Fmax1-3 are given in

351

Figures S13-S17.

352 353

ap1 = m×ln(kO3) + b

eq. 1

354

∆TrOC0→inflection = ap1×∆S

eq. 2

355

∆TrOCinflection→… = ap1×I + ap2×( ∆S – I )

eq. 3

356 357


17


Page 18 of 44

358 359

Figure 1. Abatement pattern of TrOCs in relation to ∆UVA254 (a) and ∆Fmax1 (b), divided in two

360

phases related to the main occurring mechanisms (dotted line indicates the inflection point).


18

Page 19 of 44


361

362 363

Figure 2. Relationship between the TrOC kO3 (see Table S1) and the slope of the first phase for

364

(a) ∆TrOC vs ∆UVA254, and (b) ∆TrOC vs ∆Fmax1. Data points are given from left to right for

365

metronidazole, amitriptyline, venlafaxine, ciprofloxacin, levofloxacin, trimethoprim and

366

diclofenac. The error on ap1 and kO3 is shown by the circles around the data points.

367 368

TrOC abatement related to fluorescence based surrogates

369

The relationship between ∆TrOC and both ∆TF and ∆Fmax1-3 could be described in a similar

370

way as for ∆UVA254. Surrogates including the largest share of the EEMs, and therefore

371

containing the most spectral information (i.e. representing absorbance or intensities for a broader

372

range of chemical moieties) showed a higher maximal removal (∆TF = 81%, ∆Fmax1 = 84%,

373

∆Fmax2 = 70% and ∆Fmax3 = 83%; Figure 1 and S7-S10) than UVA254 (47%). Nevertheless, it

374

was clear that each fluorescent component exhibited a different behavior. A higher decrease of


19


Page 20 of 44

375

fluorescence intensity was observed before reaching the inflection point, compared to UVA254.

376

Inflection of the curves was only established at 55% ∆Fmax3 and at 47% ∆TF, 44% ∆Fmax1 and

377

26% ∆Fmax2. Fmax4 was not further considered as no significant correlations were found, most

378

likely due to the limited amount of effluent samples used for building the correlations and the

379

high variability (i.e. high degree of scattering) in Fmax4 data.

380

The relationship between the slope of the linear correlations in phase 1 (through the origin) and

381

the kO3 showed a logarithmic behavior with a good match (R² = 0.75 - 0.76, n = 7, Figure 2b and

382

S10a,c,e). Also here, the slope of the correlation in phase 2 showed no dependency on the kO3

383

(Figure S11b and S12b,d,f) and with similar slopes for each TrOC group (Table S10). Hence, for

384

fluorescence similar equations as for UVA254 (eq. 1-3) can be used.

385 386

Applicability of the correlation models

387

The models for both UVA254 and fluorescence were applied on an independent set of data

388

(effluent 3) and were able to describe ∆TrOC with good accuracy. This is confirmed by Theil’s

389

inequality coefficient (TIC)46 with all values below 0.3. Some variation was noticed on the

390

individual data although mostly near the 95% confidence interval of the model (Figure S13-S17),

391

especially using fluorescence measurements (TF and Fmax1-3), being a potential indication of

392

higher robustness of fluorescence models. Effluent 3 had a higher turbidity than effluents 1 and 2

393

which might have affected the ozonation process and/or the spectroscopic measurements (i.e. the

394

increased degree of scattering for effluent 3). Zucker et al.53 and Zimmerman et al. 12 already

395

pointed towards the presence of solids as potential reasons for over- or under-prediction.


20

Page 21 of 44


396

Although not investigated, the model compounds had different log Kow values (Table S1)

397

potentially causing different sorption behavior.53

398

The different plots in Figure 1, S7-S9 and S13-S17 also indicate that a certain stable minimum

399

level of UVA254 or fluorescence intensity exists. Especially group III components were still

400

significantly reducing while further surrogate reduction was very limited. Comparable findings

401

by Nöthe et al.47 indicated that a maximal decrease of UVA254 and fluorescence should be

402

considered. In this context, van der Helm54 hypothesized a stable background UVA254 level in

403

drinking water ozonation. Potentially, the first stage of oxidation results in a significant reduction

404

of absorbance or fluorescence related to the breakdown of complex organic matter and

405

simultaneous reduction of aromaticity, as also observed during chlorination and ozonation of

406

natural organic matter.36,55,56 The EfOM reactions could also generate new components (e.g.

407

aldehydes or ketones) that continue to absorb.

408 409

HO• exposure

410

Measuring HO• exposure requires measurements of probe compounds such as pCBA.10 The HO•

411

exposure determined in three different effluents (used for model development) at various initial

412

ozone concentrations was plotted in Figure 3a and shows a maximum of 2.4×10-10 M.s. The HO•

413

exposure increased exponentially (R² = 0.75, n = 48) with an increasing ozone dose. The little

414

amount of HO• formed at low ozone doses was presumably due to the ozone consumption by

415

nitrite (between 0.2 and 0.7 mg O3 L-1, Table S2; associated with maximum 0.1 g O3 g-1 DOC),

416

and by direct ozonation of EfOM. Although Lee et al.10 described a linear relation between the

417

specific ozone dose (up to 1.75 g O3 g-1 DOC) and HO• exposure, they also determined a


21


Page 22 of 44

418

threshold value (varying between 0.06 – 0.24 g O3 g-1 DOC) below which no or little radical

419

production was noticed. The non-linear relationship between HO• exposure and specific ozone

420

dose as observed in Figure 3a was likely noticed due to the availability of data at very low ozone

421

doses (i.e. higher resolution) and the high IOD of the effluent samples. The abatement profile

422

group III TrOCs clearly shows similarities with the regression in Figure 3a. Similar to this

423

research, also Nöthe et al.47 observed only a significant removal of such components beyond ±

424

0.5 g O3 g-1 DOC.

425

The determined HO• exposures were exponentially correlated to the surrogate measurements (n

426

= 48) with good agreements for ∆UVA254 (R² = 0.82), ∆TF (R² = 0.89), ∆Fmax1 (R² = 0.90),

427

∆Fmax2 (R² = 0.86) and ∆Fmax3 (R² = 0.90). The correlations presented in Figure 3b-f show high

428

potential for online estimation of HO exposure. Although the curves were similar, ∆Fmax3

429

exhibited a more inflected curve with steeper incline going beyond exponential behavior (grey

430

shaded in Figure 3f). To quantify these differences, the reductions in surrogate signals

431

corresponding to HO• exposures of 25, 50, 75 and 100% of the maximal value was plotted in

432

Figure S18Error! Reference source not found.. It is clear that a HO• exposure increasing

433

above 25% of its maximum value has only limited impact on ∆Fmax3, ∆Fmax1 and – to a slightly

434

lesser extent – ∆TF, compared to ∆UVA254 and ∆Fmax2. A maximum additional decrease of 49

435

and 48% was noticed for respectively ∆UVA254 and ∆Fmax2, while this amounted to only 32 and

436

27% for ∆Fmax1 and ∆Fmax3. This exemplifies the difference in behavior of the different

437

surrogate variables. While UVA254 and Fmax2 seem to be associated with both direct and indirect

438

ozone reactions, Fmax1 and Fmax3 are most likely dominated by direct ozone reactions. A more in-

439

depth kinetic study is recommended to further elucidate these findings. Nevertheless, support can

440

be found also in considering the properties of the different chemical groups showing intensities


22

Page 23 of 44


441

within the Fmax1 and Fmax3 spectral regions. Both are described as fulvic and humic acid-like

442

matter, respectively, prone to molecular weight changes57 and probably with a high degree of

443

aromaticity.42 Swietlik et al.57 also concluded that especially hydrophobic acids consisting of C5-

444

C9 aliphatic carboxylic acids and humic acids, characterized with a high degree of aromaticity,

445

showed high ozone reactivity. Although UVA254 is known to be related to aromaticity, it is also

446

known that less aromatic (and less reactive) moieties absorb at this wavelength as well.54

447

Organic matter associated with Fmax2 (fulvic acid-like) indicates to be less reactive towards

448

ozone, and is therefore assumed to contain a lower degree of aromaticity.

449


23


Page 24 of 44

450 451

Figure 3. HO• exposure determined in three different effluent samples, correlated exponentially

452

with the initial O3:DOCeq dose (a), ∆UVA254 (b), ∆TF (c) and ∆Fmax1-3 (d-f). (dotted line

453

indicates the 25% level of the maximally obtained HO• exposure)

454


24

Page 25 of 44


455

Model validation using effluents of different plants

456

Predictability of the inflected correlation model

457

The newly developed and calibrated correlation model for UVA254, Fmax1-3 and TF adequately

458

predicted pCBA abatement in WRRF effluents. Model predictions are shown in Figure 4a and b

459

for ∆UVA254 and ∆Fmax1, respectively. Considering the total data set (n = 60), the good

460

performance of the model by using ∆UVA254 was exemplified by both a t-test (p-value of 0.41)

461

and a TIC value of 0.13. Good results were also obtained when using ∆Fmax1 as the surrogate (p

462

= 0.44; TIC = 0.14). Visually, a slightly better prediction was observed when using ∆Fmax1

463

instead of ∆UVA254, which was confirmed by the MAE being 8.3% (∆Fmax1) and 9.2%

464

(∆UVA254), respectively. The number of data points outside the model’s 95% confidence interval

465

was especially lower in the ∆pCBA range above 50% (4 compared to 9 on a total of 21). This

466

was also reflected in the slightly higher p-value of the t-test. Other surrogates such as ∆Fmax2-3

467

or ∆TF were not able to further increase the predictive power (i.e. enhance the results of the t-

468

test, TIC or MAE). For example, the TIC- and MAE-values obtained with ∆Fmax2 (0.31) and

469

∆Fmax3 (0.52) (slightly) exceeded the threshold value, while that of ∆TF (0.28) was just below.

470

Also the MAE (15.8-29.2%) was clearly higher, indicating an overall lower agreement in

471

comparison to ∆UVA254 and ∆Fmax1. This is also graphically clear from Figure S19.

472

All p- and TIC-values are summarized in Table S6, making a differentiation among the different

473

sampling locations. The model performance was the weakest for the WRRF effluent of

474

Waregem. ∆Fmax2 – although its overall TIC of 0.31 – showed to be a good predictor for all

475

effluents except for that of Waregem (p = 0.064; TIC = 0.75). That plant treated a significant

476

amount of wastewater originating from the textile industry. The dyes consisting of aromatic,

477

heterocyclic and nitrogen containing moieties such as azobenzene or triazines typically show a


25


Page 26 of 44

478

low reactivity towards ozone, while affecting the spectral signals significantly.36,58,59 ∆Fmax3 and

479

∆TF did not provide a sufficiently high predictive power for the effluents of Waregem and

480

Kruiseke for those models to be validated.

481

The highest model robustness is thus obtained with ∆UVA254, ∆Fmax1 and ∆Fmax2. This is logical

482

for Fmax1 and 2 since these components contain most of the spectral information of the EEMs (1st

483

and 2nd rank). The presence (i.e. signal strength) of Fmax1 and also Fmax4 (less ozone reactive

484

moieties) was clearly higher than that of Fmax2 or 3 (Figure S20) resulting in a better

485

correspondence after the point of inflection. UVA254 and Fmax2 were previously indicated to be

486

affiliated with both direct and indirect ozone reactions, giving a representation of all ongoing

487

reactions.

488


26

Page 27 of 44


489 490

Figure 4. Abatement patterns of predicted (full lines) versus measured (dots) ∆pCBA in relation

491

to ∆UVA254 or ∆Fmax1 by applying the inflected model (a and b), a single correlation model

492

based on own data (c and d), and the single correlation models reported by Audenaert et al.19 (e)

493

and Sharif et al.17 (f). The shaded grey areas indicate the 95% confidence interval of the models.

494

The abbreviations of the WRRF are explained in Table S1.


27


Page 28 of 44

495 496

Comparison of the inflected and single correlation models

497

A single correlation model for ∆UVA254 constructed with the same data as used for the

498

development of the inflected model showed a lower predictive power, based on the t-test (p =

499

0.17 versus 0.41), and to a lesser extent based on TIC-value (0.14 versus 0.13) and MAE (9.8%

500

versus 9.2%). Additionally, applying four previously reported correlations17,19,20,22 for ∆pCBA to

501

our experimental data shows TIC-values between 0.11-0.22, thus all below 0.3. The t-test

502

indicates a good predictive power for the correlation from Audenaert et al.19 (p-value of 0.36),

503

Gerrity et al.22 (p-value of 0.56) and Sharif et al.17 (p-value of 0.91), but not for that from Wert et

504

al.20 (p-value of 0.006). In accordance, the MAE was consistently in line with those observed for

505

the inflected model (7.6-9.2%) with the exception of that from Wert et al.20 (14.0%). When

506

looking at the data graphically (Figure 4c,e,f and S21), it becomes clear that that ∆UVA254 based

507

single correlations do not optimally describe the data trend in certain ∆UVA254 regions. The

508

models by Audenaert et al.19 (Figure 4e) and Gerrity et al.22 (Figure S21b) are showing a slight

509

under-prediction at high ∆UVA254 (> 30%), similar to the own single correlation model (Figure

510

4c). This is also supported by the MAE being higher (between 12.3-14.1%) compared to the

511

inflected model (11.4%) at these high ∆UVA254. The model of Sharif et al.17 (Figure 4f) resulted

512

in a slightly lower MAE above 30% ∆UVA254 (10.3%). Additionally, the model by Gerrity et

513

al.22 showed an over-prediction of ∆pCBA for low ∆UVA254. Although MAE values were

514

relatively close to each other, it is visually noticed that all single correlation models are not

515

following the trend of the data, i.e. situating the measured data more below the model at low

516

∆UVA254 and above at high ∆UVA254. The model by Wert et al.20 (Figure S21a) showed an


28

Page 29 of 44


517

under-prediction for the full range of ∆UVA254. Such outcomes might lead to additional

518

operational costs or by-product formation due to overdosing of ozone.

519

For the single correlation model of Fmax1 in Figure 4d, the under-prediction is even more

520

pronounced at higher ∆pCBA values (MAE = 11.0% compared to 6.8% if ∆Fmax1 > 50%). This

521

is the important range to reduce pCBA levels (or other TrOCs that react slowly with O3). The

522

newly developed inflected model (Figure 3b) was superior compared to that model. Whereas

523

85% of all samples (n = 60) were within a maximum absolute deviation of 20% for the single

524

model (MAE = 9.1%), this increased up to 92% for the inflected model (MAE = 8.3%). The

525

better performance of the latter was also exemplified by the p-value (0.44 versus 0.19). More

526

details concerning the different correlation models and their predictive power are given in Table

527

S8-S10.

528 529

Considerations for full-scale applications

530

The correlation models will determine the zones for ozone dose control in practice. The inflected

531

correlation models have shown to predict ∆TrOC in an adequate manner, solely based on

532

reported kO3 values. This is an important feature given the large amount of TrOCs potentially

533

present in effluent. The kO3 of the targeted TrOCs can be determined experimentally, based on

534

literature or by QSAR, reducing the need for extensive experimentation.60,61 The exact value of

535

kO3 might vary based on the method of determination but the established model allows a

536

straightforward prediction of the abatement pattern within the 95% confidence interval.

537

Both UVA254 or fluorescence surrogates seem adequate candidates. Real-time measurement of

538

∆UVA254 has already been established and sensors have been applied extensively. Stable and

539

reliable fluorescence sensors are still under development, although commercial sensors are


29


Page 30 of 44

540

starting to become available. Therefore, the use of UVA254 might be preferable at the current

541

timing for online ∆TrOC monitoring. Nevertheless, the more information rich EEMs might pose

542

advantages compared to the less selective measurement of UVA254 or TF. PARAFAC analysis

543

showed large potential with regard to enhancement of process understanding.

544 545

ASSOCIATED CONTENT

546

Supporting Information.

547

SI-Texts 1-6, Tables S1-S10 and Figures S1-S21 provide further information addressing

548

experimental procedures, data and discussion on TrOC analyses, determination of IOD and HO•

549

exposure, abatement profiles of TrOCs in relation to UVA254 and fluorescence measurements,

550

and statistical evidence supporting statements on the model validity. This information is

551

available free of charge via the Internet at https://pubs.acs.org

552 553

AUTHOR INFORMATION

554

Corresponding Author

555

*E-mail: [email protected], Phone: 00 32 56 24 12 06, Fax: 00 32 56 24 12 24

556

Notes

557

The authors declare no competing financial interest.

558

ACKNOWLEDGEMENTS


30

Page 31 of 44


559

The financial support (AUGE/11/016) from the Hercules Foundation of the Flemish Government

560

is acknowledge for the UHPLC-Q-ExactiveTM mass spectrometry equipment. Ghent University

561

is acknowledged for the PhD grant of Michael Chys and The Special Research Fund (Ghent

562

University)for funding the automated SPE equipment (01B07512). The authors further like to

563

thank the staff of Aquafin NV (Belgium) for their help during sampling. This project was

564

initiated within the LED H2O project which belongs to the LED network (www.lednetwerk.be)

565

and is financially supported by the Flemish Knowledge Center Water (Vlakwa vzw).

566 567

REFERENCES

568

(1)

VMM. Pesticiden in oppervlaktewater en RWZI’s in 2014; Aalst, Belgium, 2015.

569

(2)

Gerrity, D.; Snyder, S. Review of Ozone for Water Reuse Applications: Toxicity,

570

Regulations, and Trace Organic Contaminant Oxidation. Ozone Sci. Eng. 2011, 33 (4),

571

253–266.

572

(3)

Kidd, K. A.; Blanchfield, P. J.; Mills, K. H.; Palace, V. P.; Evans, R. E.; Lazorchak, J. M.;

573

Flick, R. W. Collapse of a fish population after exposure to a synthetic estrogen. Proc.

574

Natl. Acad. Sci. 2007, 104 (21), 8897–8901.

575

(4)

Paltiel, O.; Fedorova, G.; Tadmor, G.; Kleinstern, G.; Maor, Y.; Chefetz, B. Human

576

exposure to wastewater-derived pharmaceuticals in fresh produce: A randomized

577

controlled trial focusing on carbamazepine. Environ. Sci. Technol. 2016, 50 (8), 4476–

578

4482.

579

(5)

Franklin, A. M.; Williams, C. F.; Andrews, D. M.; Woodward, E. E.; Watson, J. E. Uptake

580

of Three Antibiotics and an Antiepileptic Drug by Wheat Crops Spray Irrigated with

581

Wastewater Treatment Plant Effluent. J. Environ. Qual. 2016, 45 (2), 546.


31


582

(6)

Page 32 of 44

Audenaert, W. T. M.; Chys, M.; Auvinen, H.; Dumoulin, A.; Rousseau, D.; Van Hulle, S.

583

W. H. (Future) Regulation of Trace Organic Compounds in WWTP Effluents as a Driver

584

for Advanced Wastewater Treatment. Ozone News 2014, 42 (6), 17–22.

585

(7)

Barbosa, M. O.; Moreira, N. F. F.; Ribeiro, A. R.; Pereira, M. F. R.; Silva, A. M. T.

586

Occurrence and removal of organic micropollutants: An overview of the watch list of EU

587

Decision 2015/495. Water Res. 2016, 94, 257–279.

588

(8)

Margot, J.; Kienle, C.; Magnet, A.; Weil, M.; Rossi, L.; de Alencastro, L. F.; Abegglen,

589

C.; Thonney, D.; Chèvre, N.; Schärer, M.; et al. Treatment of micropollutants in municipal

590

wastewater: Ozone or powdered activated carbon? Sci. Total Environ. 2013, 461–462,

591

480–498.

592

(9)

Eggen, R. I. L.; Hollender, J.; Joss, A.; Schärer, M.; Stamm, C. Reducing the discharge of

593

micropollutants in the aquatic environment: The benefits of upgrading wastewater

594

treatment plants. Environ. Sci. Technol. 2014, 48 (14), 7683–7689.

595

(10)

Lee, Y.; Gerrity, D.; Lee, M.; Bogeat, A. E.; Salhi, E.; Gamage, S.; Trenholm, R. A.;

596

Wert, E. C.; Snyder, S. A.; von Gunten, U. Prediction of Micropollutant Elimination

597

during Ozonation of Municipal Wastewater Effluents: Use of Kinetic and Water Specific

598

Information. Environ. Sci. Technol. 2013, 47 (11), 5872–5881.

599

(11)

600 601

Von Gunten, U. Ozonation of drinking water: Part II. Disinfection and by-product formation in presence of bromide, iodide or chlorine. Water Res. 2003, 37 (7), 1469–1487.

(12)

Zimmermann, S. G.; Wittenwiler, M.; Hollender, J.; Krauss, M.; Ort, C.; Siegrist, H.; von

602

Gunten, U. Kinetic assessment and modeling of an ozonation step for full-scale municipal

603

wastewater treatment: micropollutant oxidation, by-product formation and disinfection.

604

Water Res. 2011, 45 (2), 605–617.


32

Page 33 of 44

605


(13)

Lim, S.; Lee, W.; Na, S.; Shin, J.; Lee, Y. N-nitrosodimethylamine (NDMA) formation

606

during ozonation of N,N-dimethylhydrazine compounds: Reaction kinetics, mechanisms,

607

and implications for NDMA formation control. Water Res. 2016, 105, 119–128.

608

(14)

Chys, M.; Oloibiri, V. A.; Audenaert, W. T. M.; Demeestere, K.; Van Hulle, S. W. H.

609

Ozonation of biologically treated landfill leachate: Efficiency and insights in organic

610

conversions. Chem. Eng. J. 2015, 277, 104–111.

611

(15)

Dodd, M. C.; Buffle, M. O.; Von Gunten, U. Oxidation of antibacterial molecules by

612

aqueous ozone: Moiety-specific reaction kinetics and application to ozone-based

613

wastewater treatment. Environ. Sci. Technol. 2006, 40 (6), 1969–1977.

614

(16)

Beltran, F. J. Ozone Reaction Kinetics for Water and Wastewater Systems; 2004.

615

(17)

Sharif, F.; Wang, J.; Westerhoff, P. Transformation in Bulk and Trace Organics during

616 617

Ozonation of Wastewater. Ozone Sci. Eng. 2012, 34 (February), 26–31. (18)

Hollender, J.; Zimmermann, S. G.; Koepke, S.; Krauss, M.; Mcardell, C. S.; Ort, C.;

618

Singer, H.; Von Gunten, U.; Siegrist, H. Elimination of organic micropollutants in a

619

municipal wastewater treatment plant upgraded with a full-scale post-ozonation followed

620

by sand filtration. Environ. Sci. Technol. 2009, 43 (20), 7862–7869.

621

(19)

Audenaert, W. T. M.; Vandierendonck, D.; Van Hulle, S. W. H.; Nopens, I. Comparison

622

of ozone and HO· induced conversion of effluent organic matter (EfOM) using ozonation

623

and UV/H2O2 treatment. Water Res. 2013, 47 (7), 2387–2398.

624

(20)

Wert, E. C.; Rosario-Ortiz, F. L.; Snyder, S. A. Using Ultraviolet Absorbance and color to

625

asses pharmaceutical oxidation during ozonation of wastewater. Environ. Sci. Technol.

626

2009, 4858–4863.

627

(21)

Bahr, C.; Schumacher, J.; Ernst, M.; Luck, F.; Heinzmann, B.; Jekel, M. SUVA as control


33


Page 34 of 44

628

parameter for the effective ozonation of organic pollutants in secondary effluent. Water

629

Sci. Technol. 2007, 55 (12), 267–274.

630

(22)

Gerrity, D.; Gamage, S.; Jones, D.; Korshin, G. V; Lee, Y.; Pisarenko, A.; Trenholm, R.

631

A.; von Gunten, U.; Wert, E. C.; Snyder, S. A. Development of surrogate correlation

632

models to predict trace organic contaminant oxidation and microbial inactivation during

633

ozonation. Water Res. 2012, 46 (19), 6257–6272.

634

(23)

Dickenson, E. R. V. V; Drewes, J. E. J. E.; Sedlak, D. L.; Wert, E. C.; Snyder, S. A.

635

Applying surrogates and indicators to assess removal efficiency of Trace Organic

636

Chemicals during Chemical Oxidation of wastewaters. Environ. Sci. Technol. 2009, 43

637

(16), 6242–6247.

638

(24)

Nanaboina, V.; Korshin, G. V. Evolution of absorbance spectra of ozonated wastewater

639

and its relationship with the degradation of trace-level organic species. Environ. Sci.

640

Technol. 2010, 44 (16), 6130–6137.

641

(25)

Pisarenko, A. N.; Stanford, B. D.; Yan, D.; Gerrity, D.; Snyder, S. A. Effects of ozone and

642

ozone/peroxide on trace organic contaminants and NDMA in drinking water and water

643

reuse applications. Water Res. 2012, 46 (2), 316–326.

644

(26)

Wittmer, A.; Heisele, A.; McArdell, C. S.; Böhler, M.; Longree, P.; Siegrist, H. Decreased

645

UV absorbance as an indicator of micropollutant removal efficiency in wastewater treated

646

with ozone. Water Sci. Technol. 2015, 71 (7), 980–985.

647

(27)

Antoniou, M. G.; Hey, G.; Rodríguez Vega, S.; Spiliotopoulou, A.; Fick, J.; Tysklind, M.;

648

la Cour Jansen, J.; Andersen, H. R. Required ozone doses for removing pharmaceuticals

649

from wastewater effluents. Sci. Total Environ. 2013, 456–457, 42–49.

650

(28)

Chon, K.; Salhi, E.; von Gunten, U. Combination of UV absorbance and electron donating


34

Page 35 of 44


651

capacity to assess degradation of micropollutants and formation of bromate during

652

ozonation of wastewater effluents. Water Res. 2015, 81, 388–397.

653

(29)

654 655

Buffle, M. O.; Von Gunten, U. Phenols and amine induced HO. generation during the initial phase of natural water ozonation. Environ. Sci. Technol. 2006, 40 (9), 3057–3063.

(30)

Chen, W.; Westerhoff, P.; Leenheer, J. A.; Booksh, K. Fluorescence Excitation−Emission

656

Matrix Regional Integration to Quantify Spectra for Dissolved Organic Matter. Environ.

657

Sci. Technol. 2003, 37 (24), 5701–5710.

658

(31)

Murphy, K. R.; Hambly, A.; Singh, S.; Henderson, R. K.; Baker, A.; Stuetz, R.; Khan, S.

659

J. Organic matter fluorescence in municipal water recycling schemes: Toward a unified

660

PARAFAC model. Environ. Sci. Technol. 2011, 45 (7), 2909–2916.

661

(32)

Ishii, S. K. L.; Boyer, T. H. Behavior of reoccurring parafac components in fluorescent

662

dissolved organic matter in natural and engineered systems: A critical review. Environ.

663

Sci. Technol. 2012, 46 (4), 2006–2017.

664

(33)

Sgroi, M.; Roccaro, P.; Korshin, G. V.; Greco, V.; Sciuto, S.; Anumol, T.; Snyder, S. A.;

665

Vagliasindi, F. G. A. Use of fluorescence EEM to monitor the removal of emerging

666

contaminants in full scale wastewater treatment plants. J. Hazard. Mater. 2016.

667

(34)

Lee, Y.; Kovalova, L.; McArdell, C. S.; von Gunten, U. Prediction of micropollutant

668

elimination during ozonation of a hospital wastewater effluent. Water Res. 2014, 64, 134–

669

148.

670

(35)

Anumol, T.; Sgroi, M.; Park, M.; Roccaro, P.; Snyder, S. A. Predicting trace organic

671

compound breakthrough in granular activated carbon using fluorescence and UV

672

absorbance as surrogates. Water Res. 2015, 76, 76–87.

673

(36)

Von Sonntag, C.; von Gunten, U. Chemistry of Ozone in Water and Wastewater


35


674 675

Page 36 of 44

Treatment; IWA Publishing: London, 2012. (37)

Vergeynst, L.; K’oreje, K.; De Wispelaere, P.; Harinck, L.; Van Langenhove, H.;

676

Demeestere, K. Statistical procedures for the determination of linearity, detection limits

677

and measurement uncertainty: A deeper look into SPE-LC-Orbitrap mass spectrometry of

678

pharmaceuticals in wastewater. J. Hazard. Mater. 2017, 323 (part A), 2–10.

679

(38)

Eaton, A. D.; Clesceri, L. S.; Rice, E. W.; Greenberg, A. E. Standard Methods for the

680

Examination of Water and Wastewater, 21st Edition; Franson, M. A. H., Ed.; APHA

681

American Public Health Association: Washington, DC, 2005.

682

(39)

683 684

Hoigné, J.; Bader, H. Characterization Of Water Quality Criteria for Ozonation Processes . Part II : Lifetime of Added Ozone. Ozone Sci. Eng. 1994, 16 (2), 121–134.

(40)

Roustan, M.; Debellefontaine, H.; Do-Quang, Z.; Duguet, J.-P. Development of a Method

685

for the Determination of Ozone Demand of a Water. Ozone Sci. Eng. 1998, 20 (6), 513–

686

520.

687

(41)

688 689

Bader, H.; Hoigné, J. Determination of ozone in water by the indigo method. Water Res. 1981, 15 (4), 449–456.

(42)

Li, W.-T.; Chen, S.-Y.; Xu, Z.-X.; Li, Y.; Shuang, C.-D.; Li, A.-M. Characterization of

690

dissolved organic matter in municipal wastewater using fluorescence PARAFAC analysis

691

and chromatography multi-excitation/emission scan: a comparative study. Environ. Sci.

692

Technol. 2014, 48 (5), 2603–2609.

693

(43)

694 695 696

Ohno T. Fluorescence Inner - Filtering Correction for Determining the Humification Index of Dissolved Organic Matter. Environ. Sci. Technol. 2002, 36 (4), 742–746.

(44)

Murphy, K. R.; Stedmon, C. A.; Graeber, D.; Bro, R. Fluorescence spectroscopy and multi-way techniques. PARAFAC. Anal. Methods 2013, 5 (23), 6557.


36

Page 37 of 44

697


(45)

Shutova, Y.; Baker, A.; Bridgeman, J.; Henderson, R. K. Spectroscopic characterisation of

698

dissolved organic matter changes in drinking water treatment: From PARAFAC analysis

699

to online monitoring wavelengths. Water Res. 2014, 54, 159–169.

700

(46)

Audenaert, W. T. M.; Callewaert, M.; Nopens, I.; Cromphout, J.; Vanhoucke, R.;

701

Dumoulin, A.; Dejans, P.; Van Hulle, S. W. H. Full-scale modelling of an ozone reactor

702

for drinking water treatment. Chem. Eng. J. 2010, 157 (2–3), 551–557.

703

(47)

704 705

Nöthe, T.; Fahlenkamp, H.; Von Sonntag, C. Ozonation of wastewater: Rate of ozone consumption and hydroxyl radical yield. Environ. Sci. Technol. 2009, 43 (15), 5990–5995.

(48)

Buffle, M.-O.; Schumacher, J.; Meylan, S.; Jekel, M.; von Gunten, U. Ozonation and

706

Advanced Oxidation of Wastewater: Effect of O 3 Dose, pH, DOM and HO • -Scavengers

707

on Ozone Decomposition and HO • Generation. Ozone Sci. Eng. 2006, 28 (4), 247–259.

708

(49)

709 710

Von Gunten, U. Ozonation of drinking water: Part I. Oxidation kinetics and product formation. Water Res. 2003, 37 (7), 1443–1467.

(50)

Park, M.; Anumol, T.; Daniels, K. D.; Wu, S.; Ziska, A. D.; Snyder, S. A. Predicting trace

711

organic compound attenuation by ozone oxidation: Development of indicator and

712

surrogate models. Water Res. 2017, 119, 21–32.

713

(51)

York, D.; Evensen, N. M.; Martı́nez, M. L.; De Basabe Delgado, J. Unified equations for

714

the slope, intercept, and standard errors of the best straight line. Am. J. Phys. 2004, 72 (3),

715

367–375.

716

(52)

Mortier, S. T. F. C.; Van Hoey, S.; Cierkens, K.; Gernaey, K. V.; Seuntjens, P.; De Baets,

717

B.; De Beer, T.; Nopens, I. A GLUE uncertainty analysis of a drying model of

718

pharmaceutical granules. Eur. J. Pharm. Biopharm. 2013, 85 (3 PART B), 984–995.

719

(53)

Zucker, I.; Lester, Y.; Avisar, D.; Hübner, U.; Jekel, M.; Weinberger, Y.; Mamane, H.


37


Page 38 of 44

720

Influence of wastewater particles on ozone degradation of trace organic contaminants.

721

Environ. Sci. Technol. 2015, 49 (1), 301–308.

722

(54)

van der Helm, A. W. C. Integrated modeling of ozonation for optimization of drinking

723

water treatment (PhD thesis); Water Management Academic press: Delft; The

724

Netherlands, 2007.

725

(55)

Korshin, G. V; Kumke, M. U.; Li, C.-W.; Frimmel, F. H. Influence of chlorination on

726

chromophoress and fluorophores in humic substances. Environ. Sci. Technol. 1999, 33,

727

1207–1212.

728

(56)

Korshin, G. V.; Benjamin, M. M.; Chang, H. S.; Gallard, H. Examination of NOM

729

chlorination reactions by conventional and stop-flow differential absorbance spectroscopy.

730

Environ. Sci. Technol. 2007, 41 (8), 2776–2781.

731

(57)

Swietlik, J.; Dabrowska, A.; Raczyk-Stanisławiak, U.; Nawrocki, J. Reactivity of natural

732

organic matter fractions with chlorine dioxide and ozone. Water Res. 2004, 38 (3), 547–

733

558.

734

(58)

735 736

Chu, W.; Ma, C. W. Quantitative prediction of direct and indirect dye ozonation kinetics. Water Res. 2000, 34 (12), 3153–3160.

(59)

Wu, J.; Ma, L.; Chen, Y.; Cheng, Y.; Liu, Y.; Zha, X. Catalytic ozonation of organic

737

pollutants from bio-treated dyeing and finishing wastewater using recycled waste iron

738

shavings as a catalyst: Removal and pathways. Water Res. 2016, 92, 140–148.

739

(60)

Sudhakaran, S.; Amy, G. L. QSAR models for oxidation of organic micropollutants in

740

water based on ozone and hydroxyl radical rate constants and their chemical classification.

741

Water Res. 2013, 47 (3), 1111–1122.

742

(61)

Lee, Y.; von Gunten, U. Quantitative structure-activity relationships (QSARs) for the


38

Page 39 of 44


743

transformation of organic micropollutants during oxidative water treatment. Water Res.

744

2012, 46 (19), 6177–6195.

745 746


39

Environmental Science & Technology Page 40 of 44


Page 41 of 44





Page 42 of 44

Page 43 of 44





Page 44 of 44

Surrogate-Based Correlation Models in View of Real-Time Control

Recommend Documents