Disinfection by-products (DBPs) in drinking water and predictive models for their occurrence: a...

Science of the Total Environment 321(2004) 21–46

0048-9697/04/$ - see front matter� 2003 Elsevier B.V. All rights reserved.doi:10.1016/j.scitotenv.2003.05.001

Disinfection by-products(DBPs) in drinking water and predictivemodels for their occurrence: a review

Rehan Sadiq , Manuel J. Rodriguez *a b,

Institute for Research in Construction, National Research Council, Ottawa, ON, Canada K1A 0R6a

Departement d’Amenagement, 1624 Pavillon Savard, Universite Laval Quebec City, QC, Canada G1K7Pb ´ ´ ´ ´

Received 17 November 2002; accepted 9 May 2003

Abstract

Disinfection for drinking water reduces the risk of pathogenic infection but may pose chemical threat to humanhealth due to disinfection residues and their by-products(DBPs) when the organic and inorganic precursors arepresent in water. More than 250 DBPs have been identified, but the behavioural profile of only approximately 20DBPs are adequately known. In the last 2 decades, many modelling attempts have been made to predict the occurrenceof DBPs in drinking water. Models have been developed based on data generated in laboratory-scaled and field-scaled investigations. The objective of this paper is to review DBPs predictive models, identify their advantages andlimitations, and examine their potential applications as decision-making tools for water treatment analysis, epidemio-logical studies and regulatory concerns. The paper concludes with a discussion about the future research needs in thisarea.� 2003 Elsevier B.V. All rights reserved.

Keywords: Drinking water; Disinfection; DBPs; Predictive models; Human health risk

1. Introduction

Since the 1970s, research in the drinking waterfield has significantly focused on documenting andunderstanding the occurrence of disinfection by-products(DBPs) in drinking water. In the recentyears, a particular interest has been grown on thedevelopment of models to estimate the formationand the fate of DBPs. Several predictive modelshave been reported in the scientific literature.

*Corresponding author. Tel.:q1-418-656-2131x8933; fax:q1-418-656-2018.

E-mail address:[email protected](M.J. Rodriguez).

These models used different types of explanatoryvariables for variety of applications. Because, DBPmodelling helps to guide decision-making in thedrinking water industry, it emphasises the need forexamining the state-of-the-art knowledge concern-ing this issue.

The main objective of this paper is to reviewDBP predictive models available in the scientificliterature. A brief overview of disinfection andrelated human health concerns of DBP in drinkingwater is presented, which is then followed by thedescription of these models. Models are reviewedaccording to the characteristics of data used forthe development, the methodology on which they

22 R. Sadiq, M.J. Rodriguez / Science of the Total Environment 321 (2004) 21–46

Table 1Percent of inactivation of various organisms under varying disinfection conditions

Disinfectants Conc. Organism(Group) Contact pH Temperature Percentage of Refs.(mgyl) time (min) (8C) inactivation

Chlorine 2.0 V. chloerae (B) 30 7.0 20 )99.99 Clark et al.(1994a,b)0.5 Coliform (B) 2 7.0 5 )99.99 Berman et al.(1988)1.1 E. coli (B) 2 7.0 5 )99.999 Rice et al.(1999a)0.1 C. jejuni (B) 1 6.0 4 )99.99 Blaser et al. (1986)0.61 A. butzleri (B) 1 7.1 5 )99.999 Rice et al.(1999b)0.6–2.5 Polio virus (V) 0.7–2.4 5 )99 Hoff (1986)0.5 Rota virus (V) 0.25 6.0 5 f99.99 Berman and Hoff(1984)0.5 Rota virus (V) 1.5 10.0 5 f992.0 MS2 coliphage (V) 1 7.0 5 f99.99 Berman et al.(1992)1.5 G. lamblia (P) 10 6–7 25 f99 Clark et al.(1989)1.5 G. lamblia (P) 10 6.0 15 f992.0 G. lamblia (P) 60 6–7 5 )992.0 E. intestinalis (P) 8–16 f99 Rice et al.(1999c)

Mono- 1.0 C. jejuni (B) 15 8.0 5 )99 Blaser et al. (1986)chloramines 10.0 Rota virus (V) )360 8.5 5 f99 Berman and Hoff(1984)

2.0 MS2 coliphage (V) 1 7.0 5 f99 Berman et al.(1992)80.0 C. parvum (P) 90 5 f90 Korich et al.(1990)

Chlorine- 0.5 Rota virus (V) -1 6.0 5 )99 Berman and Hoff(1984)dioxide 0.5 Rota virus (V) -0.25 10.0 5 )99

B: Bacteria, V: Viruses, P: Protozoa.

are supported and their predictive capacity. Themajor benefits and outcomes of the reported mod-els are also highlighted. In Section 4 the potentialapplications of the models are discussed with thehelp of three hypothetical case studies. In accor-dance with the models’ review, the critical researchneed for developing improved models to help inguide decision-making is identified.

2. Disinfection by-products (DBPs)

2.1. Disinfection for drinking water

The disinfection process has been routinely car-ried out since the dawn of the 20th century toeradicate and inactivate the pathogens from waterused for drinking purpose. Disinfectants in additionto removing pathogens from drinking water, serveas oxidants in water treatment. They are also usedfor (a) removing taste and colour;(b) oxidizingiron and manganese;(c) improving coagulationand filtration efficiency; (d) preventing algalgrowth in sedimentation basins and filters, and(e)preventing biological regrowth in the water distri-

bution system(US EPA, 1999a). Chlorine and itscompounds are the most commonly used disinfec-tants for water treatment. Chlorine’s popularity isnot only due to lower cost, but also to its higheroxidizing potential, which provides a minimumlevel of chlorine residual throughout the distribu-tion system and protects against microbialrecontamination.

The disinfection process is affected by differentphysico-chemical and biological factors and itsefficiency can be characterised by dose and inten-sity (Gates, 1998). The disinfection efficiency(Ct)is a product of residual disinfectant and the contacttime of chlorine in the water. This product is usedas a design parameter for the disinfection facility.Disinfectants have varying capacities to inactivateor kill pathogens. The types and nature of organ-isms as well as the process conditions, includingtemperature and pH, also affect disinfection. Table1 compares the disinfection efficiency of threedisinfectants under varying conditions of temper-ature, pH, contact time and dose. Generally, inac-tivation of organisms’ increases with increasingCt. The pH has different effects on different

23R. Sadiq, M.J. Rodriguez / Science of the Total Environment 321 (2004) 21–46

disinfectants but in general at lower pH, chlorineis more effective against organisms than in alkalineconditions. GenerallyCt required for inactivatingmicroorganism is lower in warm water than incold water. For a specific contact time, requiredchlorine doses for disinfection are consequentlyhigher in winter than in summer conditions. How-ever, in most drinking water utilities the applica-tion of a disinfectant(such as chlorine) in additionmaintains adequate residuals to avoid the reap-pearance of microorganisms in the water distribu-tion system. The disinfectant residuals depleterapidly when the water temperature is high, whichexplains the difficulty of maintaining minimumresidual level in the large distribution systemsduring summer. Also, microbial activity withindistribution systems is higher in warm than in coldwaters(Arora et al., 1997). To maintain an ade-quate level of residual disinfectant in the distribu-tion system, higher disinfectant doses are appliedduring the summer. Generally, the conditionsaffecting the disinfection efficiency and the requi-rements to maintain disinfectant residuals in thedistribution systems simultaneously affect the for-mation of DBPs.

2.2. Occurrence of DBPs in drinking water

The application of disinfection agents to drink-ing water reduces the microbial risk but poseschemical risk in the form of their by-products. TheDBPs are formed when the disinfectant reacts withnatural organic matter(NOM) andyor inorganicsubstances present in water. More than 250 differ-ent types of DBPs have already been identified.Table 2 lists major classes of DBPs formed due tovarious disinfectants. The DBP concentrations mayvary in orders of magnitude during different dis-infection processes. Some DBPs reported in Table2 have not been identified in field-scaled studies,however, they were observed in laboratory-scaledstudies(Richardson, 1998). The formation of chlo-rinated DBPs in drinking water like trihalometha-nes (THMs) has emphasized the need forexploring alternate disinfectants and new treatmenttechnologies. Because organicyinorganic sub-stances act as precursors for DBPs, their removalprior to disinfection has proven to be an effective

method for reducing chlorinated DBP formationpotential. The NOM can be partially removedusing a conventional treatment(coagulation, floc-culation, sedimentation and filtration) or by com-biningyreplacing its components with moreefficient processes such as granular activated car-bon (GAC) filtration, enhanced coagulation andmembrane filtration. Another effective method tocontrol chlorinated DBPs in drinking water is theuse of alternative disinfectants—ozone, chlora-mines, chlorine dioxide and more recently ultra-violet (UV) light—alone or in combination withchlorine. The use of various disinfectant alterna-tives to chlorination must be considered, however,they may form non-chlorinated DBPs. Finally, abetter control of operational factors(e.g. controlof pH or disinfection contact time) may contributeto a reduction in the formation of DBPs.

For chlorination, generally chlorine gas(Cl ) is2

bubbled into pure water and rapid hydrolysis tohydrochloric(HCl) and hypochlorous acid(HOCl)takes place(Sadiq et al., 2002). The HOCl under-goes subsequent reactions resulting in the forma-tion of THMs. HOCl oxidizes the bromide(Br )y

present in the water, which reacts readily withNOM to form brominated THMs(Stevens et al.,1976). Similar parameters that affect the disinfec-tion efficiency (Ct) and residual depletion in thedistribution system affect the rate and the degreeof THM formation. THM occurrence is influencedby chlorine dose, concentration and nature ofNOM (mainly humic substances), chlorine contacttime (water residence time in distribution system),pH, temperature of water, and bromide ion(Amyet al., 1987a). In general, higher THM concentra-tions are expected at higher levels of the above-mentioned parameters.

In temperate environments, THM levels indrinking water are significantly affected by season-al conditions(Singer et al., 1995; Health Canada,1996; Arora et al., 1997; Chen and Weisel, 1998;Rodriguez and Serodes, 2001; Sadiq et al., 2002).´In the winter months and in some cases where theice cover protects surface raw waters, the THMconcentrations are lower due to lower water tem-perature and NOM. In these conditions, the chlo-rine demand is lower, therefore, the chlorine doserequired to maintain adequate residual in the dis-


Table 2Important groups of DBPs produced using different types of disinfectants

Class of DBPs Common example Chlorine Ozone ClO2 Chloramines

Trihalomethanes (THM) Chloroform �a �b �Other haloalkanes �Haloalkenes �

Haloacetic acids (HAA) Chloroacetic acid � �Haloaromatic acids �Other halomonocarboxylic acids � �Unsaturated halocarboxylic acids � �Halodicarboxylic acids � �Halotricarboxylic acids �MX and analogues � � �Other halofuranones �Haloketones � � �

Haloacetonitrile (HAN) Chloroacetonitrile � �Other halonitrile Cyanogen chloride � �Haloaldehyde Chloral hydrate � �Haloalcohals � �Phenols 2-Chlorophenol � �Halonitromethane Chloropicrin �

Inorganic compounds Bromate, Hypobromite � �Chlorite and Chlorate etc.

Aliphatic aldehyde Formaldehyde � � �Other aldehydes � � �Ketones(aliphatic and aromatic) Acetone � � �Carboxylic acids Acetic acid � � �Aromatic acids Benzoic acid � � �Aldo and Ketoacids � �Hydroxy acids � �Others � � � �

NB: Major classes of DBPs are shown in bold.There are four regulated THM compounds, but if iodomethanes are included in THMs then there will be nine compounds.a

Bromoform is produced if bromide ion is present.b

tribution system is also less important. Moreover,higher DBPs concentrations have been observedparticularly in the extremities of water distributionsystems, especially in the summer months(HealthCanada, 1996; Sadiq et al., 2002). The type ofraw water also affects the THM levels. Generally,ground waters are naturally protected from runoffNOM, while the difference in occurrence of DBPprecursors in river and lakes depends on geologi-cal, physical and environmental factors(trophicstage, watershed soil characteristics and land use,lake size, river flow rate, etc.).

For the DBPs associated to alternative disinfec-tant to chlorine (chloramines, chlorine dioxide,

ozone), the similar operational factors influencethe formation of the associated by-products(dose,pH, temperature, reaction time). Chloramines pro-duce similar DBPs than chlorine but with muchlower concentrations(Health Canada, 1996; USEPA, 2001b). For disinfection with chlorine diox-ide (ClO ), there is no evidence of reactions with2

humic acids to form trihalomethanes(Lykins andGriese, 1986). However, the inorganic DBPs suchas chlorite and chlorate are formed and they alsohave human health risk implications(Bercz et al.,1982; US EPA, 2001b). Finally, in ozonation, themost important by-product formed is bromate,which depends on the presence of bromide and


Fig. 1. Microbial and chemical risk trade-off analysis for disinfection.

ammonia ion concentrations(von-Gunten et al.,1995).

Because of major benefits of water disinfectionand due to its outcomes associated with their DBPsa risk trade-off analysis between microbial andchemical risks becomes necessary(Fig. 1). How-ever in practice, this constitutes a major challengebecause often the conditions leading to betterdisinfection efficiency also lead to higher occur-rence of DBPs. The regulatory regime must estab-lish the acceptable levels of risk for both microbialand chemical agents.

2.3. DBPs associated with human health adverseeffects—guidelines

The Safe Drinking Water Act requires the USEPA to develop several new drinking water regu-lations. The regulations related to DBPs are thepart of the microbial-disinfection by-products(M-DBPs) rule (US EPA, 1999b). The DBP regula-tions are based on evidence of their adverse humanhealth effects, in particular cancer and reproductivedisorders(Cantor et al., 1988; Graves et al., 2002).A considerably richer literature reporting adversehealth effects through toxicological laboratorystudies is available. Adverse effects of some ofthe important DBPs are summarised in Table 3.

The World Health Organization(WHO, 1993)published drinking water guidelines for a fewDBPs including THMs, haloacetic acids(HAAs),haloacetonitriles(HANs), chlorite, chloral hydrate,

formaldehyde and cyanogen chloride. In additionto individual THM guidelines, the WHO has alsosuggested that the sum of the ratios of the THMlevels to the guideline values should not exceed 1(Table 4). Such guidelines have no official rec-ognition in the US or Canada. The US EPA(2001a) has established the maximum allowablecontaminant level of 0.08 mgyl for total THMsand of 0.06 mgyl for HAA5 (the sum of fiveHAAs, i.e. mono-, di-, and trichloroacetic acidsand mono- and dibromoacetic acids). Complianceof these by-products is based on an annual runningaverage of quarterly samples and since 2002 willbe based on a locational running average(Sharfen-aker, 2001). Bromate and chlorite are also regulat-ed by US EPA(2001a). Health Canada(2001)has set 0.10 mgyl for total THM as an interimmaximum acceptable concentration, which servesas a guideline for Provincial regulations. No Cana-dian drinking water quality guideline exists forother DBPs for the time being. The Aus–NZ(2000) and UK (2000) drinking water standardsare also summarized in Table 4 for comparison.

3. Predictive models for DBPs

Models for DBPs have been developed fordifferent purposes. In some cases, modelling isaimed at identifying the significance of diverseoperational and water quality parameters control-ling the formation of DBPs or at investigating thekinetics for their formation. In other cases, they


Table 3Toxicological information for DBPs(modified after US EPA, 1999b)

Class of DBPs Compound Ratinga Detrimental effects

Trihalomethanes Chloroform B2 Cancer, liver, kidney, and reproductive effects(THM) Dibromochloromethane C Nervous system, liver, kidney

and reproductive effectsBromodichloromethane B2 Cancer, liver, kidney, and reproductive effectsBromoform B2 Cancer, nervous system, liver and kidney effects

Haloacetonitrile(HAN) Trichloroacetonitrile C Cancer, mutagenic and clastogenic effectsHalogenated aldehydes Formaldehyde B1 Mutagenicb

and ketonesHalophenol 2-Chlorophenol D Cancer, tumour promoterHaloacetic acids(HAA) Dichloroacetic acid B2 Cancer, reproductive and developmental effects

Trichloroacetic acid C Liver, kidney, spleen and developmental effectsInorganic Bromate B2 Cancer

compounds Chlorite D Developmental and reproductive effects

A: Human carcinogen; B1: Probable human carcinogen(with some epidemiological evidence); B2: Probable human carcinogena

(sufficient laboratory evidence); C: Possible human carcinogen; D: Non classifiable.Inhalation exposure.b

Table 4Standardsyguidelines related to DBPs(mgyl) in various jurisdictions of the world

Compound Acronym WHO (1993) US EPA(2001a) Health Canada(2001) Aus–NZ (2000) UK (2000)

Trichloromethane TCM 0.200 0.000*(chloroform)

Bromodichloromethane BDCM 0.060 0.060*Dibromochloromethane DBCM 0.100 0.000*Tribromomethane TBM 0.100 0.000*(bromoform)

Total trihalomethanes TTHM

4 THM8 F1

WHOis1 0.080 0.100 0.250 0.100Chloroacetic acid 0.150Dichloroacetic acid DCAA 0.050 0.100Trichloroacetic acid TCAA 0.100 0.100Haloacetic acids HAA5 0.060 **Dichloroacetonitrile DCAN 0.090Trichloroacetonitrile TCAN 0.001Dibromoacetonitrile DBAN 0.100Haloacetonitrile HAN **Chloral hydrate CH 0.010 ** 0.020Formaldehyde 0.900 0.500Chlorite 0.200 1.000 **Cyanogen chloride 0.070 **Bromate 0.010 0.010***2-chlorophenol 0.3002,4-dichlorophenol 0.2002,4,6-trichlorophenol 0.020

*Maximum contaminant level goals(MCLG); **Under consideration; ***Interim maximum acceptable concentration(IMAC).

are developed with predictive purposes as an alter-native to monitoring in the field. In fact, measure-ment of DBP concentration in drinking water

usually requires gas chromatography(GC) analy-sis, which is a time consuming and relativelyexpensive technique. Predictive modelling for


DBPs consists of establishing empirical and mech-anistic relationships of water quality and opera-tional parameters with the prevailing levels ofDBPs at various stages after the water treatment.The research in the last 2 decades has been aimedprincipally at linking DBP concentrations(mainlyTHMs) with total or dissolved organic matter(TOC or DOC), UV-absorbance at 254 nm(UV-254), pH, water temperature(T), concentration ofbromide ion(Br ), chlorine dose(D) and reactiony

time of residual chlorine(t). TOC (or DOC), UV-254 and specific UV absorbance, i.e. SUVA(spe-cific ultraviolet absorbance, the ratio betweenUV-254 and TOC) are the common surrogates ofNOM. TOC and DOC are indicators of mass oforganic substance whereas UV-254 accounts forspecific structure and functional groups(Edzwaldet al., 1985; Croue et al., 1998; US EPA, 2001b).´The SUVA is an indicator of NOM reactivity.Other NOM indicators that have been used arechlorophyll-a and fluorescence. The majority ofchlorine demand is exhausted by the reaction withNOM but chlorine also reacts with various inor-ganic compounds(e.g. formation of chloraminesin the presence of ammonia, formation of bromi-nated compounds in the presence of Br , etc.).y

Studies have also shown that higher disinfectantdose increases the DBP formation potential inwater (Montgomery Watson, 1993; Rathbun,1996a). Longer reaction time generally leads tohigher consumption of residual disinfectant andresults in more formation of DBPs(Chen andWeisel, 1998; Rodriguez and Serodes, 2001). This´is one of the major reasons for the generally higherDBP concentrations observed in the extremities ofwater distribution systems compared to the finishedwater at treatment plants. However, recent researchsuggests that some chlorinated DBPs such asHAAs may degrade in extremities of distributionsystems(Chen and Weisel, 1998; Rossman et al.,2001). The pH effects on DBP formation vary fordifferent by-products(Singer et al., 1995). Forexample, in general THM formation increases withan increase in pH but the effects are reversed forHAA5. Temperature has a positive effect on DBPformation potential, and increases the rate of reac-tion (Amy et al., 1987a; US EPA, 2001b).

The modelling efforts in predicting DBP for-mation potential were started after the discoveryof chloroform in chlorinated drinking waters in1974 (Rook, 1974; Bellar et al., 1974). Consider-able research has been focused on ascertaining thevariables, which can significantly explain the DBPformation potential. Initial attempts included uni-variate analysis in which DBPs where correlatedwith TOC content in raw waters. For example,Singer and Chang(1989) developed linear rela-tionships between TOX, THM, UV-254 and TOC.Other researchers have also investigated the rela-tionships between precursor and operational indi-cators and DBPs as well as the relationshipsbetween different species of DBPs(e.g. Singer etal., 1995; Chen and Weisel, 1998; Arora et al.,1997; Gallard and von-Gunten, 2002; Gang et al.,2003).

Many other researchers have developed multi-variate models to relate DBP concentrations tovarious combinations of explanatory variables, i.e.water quality and operational parameters associat-ed with disinfection. The predictive models forDBPs are based on data obtained from field andlaboratory-scaled studies. The field data are col-lected at different sampling points including rawwater, finished water after disinfection and waterdistribution system. Laboratory-scaled studies havegenerally been based on batch sampling of rawand treated water samples. Laboratory studies havebeen found more reliable than field-scaled studiesfor developing empirical models because of con-trolled conditions. For example, in laboratory stud-ies, the effect of chlorine dose, pH and contacttime on DBP formation can be easily investigatedfor the desired concentration of NOM in water.The major drawback of laboratory studies is thatthe effects of the distribution system on residualdisinfectant concentration and DBP formation arenot accounted for. In contrast, for models basedon field data, human exposure can be measured orobserved. However, some parameters affectingDBPs are difficult to estimate in the field-scaledstudies. For example, estimating the actual contacttime of disinfectants within the distribution systemrequires tracer studies andyor hydraulic simulationmodels, which are time consuming and not alwaysvery accurate. Another major drawback of field-


Table 5A qualitative comparison of field and laboratory-scaled models

Attributes Laboratory-scaled Field-scaledmodels models

Statistical significance H MApplicability H (General) L (Site specific)Inclusion of effects of distribution system N H(biofilm, pipe material, etc.)

Controllability over explanatory variables, e.g. reaction time H LPredictability(actual human exposure) M HEase in model development H LCost and resources involved in model development L H

Hshigh; Msmedium; LsLow and NsNo.

scaled models is that they are generally site spe-cific (Rodriguez et al., 2000). Table 5 lists aqualitative comparison of laboratory and field-scaled studies in the development of predictivemodels for DBP formation.

Since chlorine is the most popular and tradition-al disinfectant, most modelling efforts have beenfocused on THMs. A comprehensive literaturereview of existing predictive models for DBPsrevealed that both laboratory-scaled and field-scaled studies are common practice. Most pub-lished models are empirically based, but somerecent attempts have also been made to combinemechanistic and empirically based approaches.Some models have also been developed for indi-vidual THMs, especially for chloroform. Predictivemodels for DBPs associated to alternative disinfec-tants (ozone, chloramines, and chlorine dioxide)have also recently been developed. However, thenumber of published models for non-chlorinatedDBPs is much lower than those for chlorinatedDBPs.

Tables 6a and 6b summarize the predictivemodels for chlorinated and other DBPs, respec-tively. Tables 7a and 7b present brief descriptionof these different models. Mostly these models arebased on multivariate regression analyses in whichexplanatory variables were subjected to a logarith-mic transformation. In few models, first and sec-ond order kinetic models are proposed with thekinetic coefficients, which are also estimated sub-sequently by multivariate regression. As observed,most models are associated to DBP data generatedunder laboratory-scaled conditions where opera-

tional factors and some water quality characteris-tics were controlled. Generally, laboratory-scaledmodels consider more explanatoryypredictive var-iables than models based on field-scaled studies.The choice of explanatory variables in models isnot motivated by mechanistic(physical) or statis-tical relationships alone, but rather by availabilityof reliable data. Also, the models based on labor-atory-scaled data are generally developed with ahigher number of observations. In fact, the logisticsfor laboratory-scaled experiments is generally lesstime-consuming than for field-scaled measure-ments and allows the investigation of variablechlorination conditions and investigating the for-mation of DBPs according to different contacttimes. This features the fact that reaction time ofchlorine in water is precisely known in laboratory-scaled models, which makes the better statisticalperformance than the performance of field-scaledmodels.

Through this review, it appears difficult to makea precise judgement about the performance of theDBP models. In fact, models are often evaluatedby means of classical statistical criteria alone(coefficient of determination, correlation coeffi-cient, mean absolute errors between measured andpredicted levels, etc.). Authors often rely theirjudgement on model predictive ability accordingwith such criteria without specifying the specificconditions(boundary conditions for predictors) orcircumstances at with these models can be applied.In addition, most models are evaluated with thesame data that have been used for their calibrationand do not consider external databases for the

29R

.Sadiq,

M.J.

Rodriguez

/Science

ofthe

TotalE

nvironment

321(2004)

21–46

Table 6aA summary of THM, HAA and total organic halogenated(TOX) predictive models

AuthoryYear Data source N r 2 Output Units Predictive models for chlorinated by-products

1. Minear and Morrow, 1983 Laboratory 40 )0.90 TTHM mmolyl y3.91q(Br ) q0.23 (log(Cl )q0.24(pH)q10 q0.26 (NVTOC)y 0.15 0.009T( )2

2. Morrow and Minear, 1987*

3. Urano et al., 1983 Laboratory NR NR TTHM mgyl 0.00082(pHy2.8) (TOC) (D) (t)0.25 0.36

4. Amy et al., 1987a,b* Laboratory 995 0.90 TTHM mmolyl 0.0031(UV TOC) (D) (t) (T) (pHy2.6) (Brq1)0.440 0.409 0.265 1.06 0.715 0.036

5. Chowdhury et al., 1991*

6. Adin et al., 1991 Laboratory NR 0.90 TTHM mgyl (K )(K )(TOC)w(1y((K qK )(K q0.19)))1 2 1 3 2

q(1y(K qK yK y0.19))=(((1y(K qK ))exp )y K qK tc( )( )1 31 3 2 1 3

y((1y(K q0.19))exp ))x wherey K q0.19 tc( )( )22

K s4.38=10 (D); K s11.36=10 (D); K s7.14=10 (D)y8 y7 y13 21 2 3

7. Montgomery Watson, 1993 Laboratory 864 0.88 CHCl3 mgyl 0.064(TOC) (UV) (Br q0.01) (pH) (D) (t) (T)0.329 0.874 y 0.404 1.161 0.561 0.269 1.018

157 0.80 BDCM 0.0098(Br ) (pH) (D) (t) (T) (for DyBr -75)y 0.181 2.55 0.497 0.256 0.519 y

110 0.92 BDCM 1.325(TOC) (Br ) (D) (t) (T) (for DyBr )75)y0.725 y 0.794 0.632 0.204 1.441 y

116 0.82 DBCM 14.998(TOC) (Br ) (D) (t) (T) (for DyBr -50)y1.665 y 1.241 0.729 0.261 0.989 y

99 0.83 DBCM 0.028(UV) (TOC) (Br ) (pH) (D) (t) (T) (for Dy Br )50)y1.175 y1.078 y 1.573 1.956 1.072 0.2 0.596 y

106 0.86 CHBr3 6.533(TOC) (Br ) (pH) (D) (t)y2.031 y 1.388 1.603 1.057 0.136

81 0.82 MCAA mgyl 1.634(TOC) (Br q0.01) (pH) (D) (t)0.753 y y0.085 y1.124 0.509 0.300

172 0.97 DCAA 0.605(TOC) (UV) (Br q0.01) (D) (t) (T)0.291 0.726 y y0.568 0.48 0.239 0.665

172 0.98 TCAA 87.182(TOC) (UV) (Br q0.01) (pH) (D) (t)0.355 0.901 y 0.679 1.732 0.881 0.264

79 0.80 MBAA 0.176(TOC) (UV) (Br ) (pH) (t) (T)1.664 y0.624 y 0.795 y0.927 0.145 0.45

81 0.95 DBAA 84.945(TOC) (UV) (Br ) (D) (t) (T)y0.62 0.651 y 1.073 y0.2 0.12 0.657

8. Lou and Chiang, 1994 Field 16 NR TTHM mgyl TTHM q7.01(pHy2.3) (NVTOC) (t) (b)0.11 1.06 0.764O

9. Ibarluzea et al., 1994 Field 12 0.82 CHCl3 mgyl 10.8q0.04 (Flu)q1.16 (pH)q0.12 (T)q1.91 (C )O

10. Rathbun, 1996a Laboratory 669 0.98 TTHM mgyl 14.6 (pHy3.8) (D) (UV) (t)0.3061.01 0.206 0.849

Rathbun, 1996b NPOX 42.0(13.0ypH) (D) (Br q1) (UV) (t)1.07 0.21 y y2.75 0.847 0.142

Rathbun, 1996c

11. Chang et al., 1996 Laboratory 120 0.94 TTHM mgyl 12.7 (TOC) (t) (D)0.291 0.271 y0.072

12. Garcia-Villanova et al., 1997a Field 66 0.86 CHCl3 mgyl exp(0.348q0.00059(T) y0.000023(T) q0.0237(pH) qaq´)3 4 2

Garcia-Villanova et al., 1997b exp(0.81Yy0.16(NF)q0.00047(T) y0.00002(T) q0.0034(pH) q´)3 4 2

13. Huixian et al., 1997** Laboratory NR 0.94 POX mgyl 7.20 t TOC D (pHq8.6) e0.08 0.49 0.41 y468.5 T( )

0.92 NPOX 28.7t TOC D (20.9ypH)e0.02 0.53 0.44 y632.4T( )

14. Clark, 1998*** Laboratory 42 As0.71 TTHM mgylwhere

B B EEC 1yKŽ .1C C FFA C y1 yutD D GG1yKe

15. Clark and Sivaganesan, 1998*** Ks0.78 usM(1yK); andMs0.42 As4.44 C TOC pH Ty0.44 0.63 y0.29 0.14

1

Ks1.38 C TOC pH Ty0.48 0.18 y0.96 0.281

Mse y2.46y0.19TOCy0.14pHy0.07Tq0.01T pH( )

16. Golfinopoulos et al., 1998 Field 88 0.98 TTHM mgyl 13.5 Ln (Ch-a)y14.5(pH)q230(Br )y140(Br ) y25.3(S)q110.6(Sp)y y 2

30R

.Sadiq,

M.J.

Rodriguez

/Science

ofthe

TotalE

nvironment

321(2004)

21–46

Table 6a(Continued)

AuthoryYear Data source N r 2 Output Units Predictive models for chlorinated by-products

y6.6 (T Sp)q1.48(T D)

17. Amy et al., 1998* Laboratory NR NR TTHM mgyl 0.00412(DOC) (D) (Br) (T) (pH) (t)1.10 0.152 0.068 0.61 1.60 0.260

18. Nokes et al., 1999 Field THMs mgyl Function of various reaction coefficients of intermediate products

19. Rodriguez et al., 2000 Field 174 0.34 TTHM mgyl 1.392(DOC) (pH) (T)1.092 0.531 0.255

Laboratory 1800 0.90 TTHM mgyl 0.044(DOC) (t) (pH) (D) (T)1.030 0.262 1.149 0.277 0.968

20. Milot et al., 2000 Field TTHM

a treatmentqb regionqcseasonqdsourceqeŽ . Ž . Ž . Ž .ePs

a treatmentqb regionqcseasonqdsourceqeŽ . Ž . Ž . Ž .1qewhereP denotes the probability of exceedence from established(regulatory) TTHM levels; a, b, c, d, e denote the logistic regressioncoefficients.

21. Clark et al., 2001* Laboratory 17–20Ms0.70 13 DBPs mmolylwhere,

B B EEC 1yKŽ .1C C FFA C yi 1 yutD D GG1yKe

Ks0.95 including usM(1yK) andTHMs and Ms52.46 e e Cl e ey0.325 Bry 0.0145 Cl1 pH y2.32 8.46 P y2.31 pH( ) ( ( ) ( ) ( ) ( )

1

HAAs Ks6.62 (pH) (Br q1) (Cl )y0.13 y 0.10 y0.751

AisyGqe (pH) (Cl ) (PqG) eai bi ci di ei Bry qfi Bry 2qgi Bry 3( ( ) ( )1

;is1, 2,«,13ymØBr

PsymØBr qmØCl1

where(m Br )smoles of Br ion and(m Cl )smoles of initial chloriney y1

22. Golfinopoulos and Arhonditsis, 2002 Field 126 0.52 TTHM mgyl y0.26 (Ch-a)q1.57 (pH)q28.74wBrxy66.72wBrx y43.63(S)q1.13 (Sp)2

q2.62 (T S)y0.72 (T D)126 0.51 CHCl3 mgyl y0.32 (Ch-a)q0.68 (pH)q2.51 (D)q1.93 (Sp)y22.1 (S)q1.38 (T S)y0.12 (T D)126 0.62 BDCM mgyl y0.37 (Ch-a)q0.32 (pH)q16.16wBrxy29.82wBrx q1.88 (D)q5.17 (S)2

y0.37 (T S)y0.12 (T D)

23. Villanueva et al., 2003 Field 18 0.57–0.97 HAAs mgyl Linear regression in function of various THM species

24. Serodes et al., 2003 Laboratory 51–53 0.56–0.92 TTHM mgyl Single linear and non-linear regression models for water of each utility(inHAAs function of water temperature, TOC, chlorine dose and contact time)

25. Gang et al., 2003 Laboratory NR NR TTHM a D{1yfe Ø y(1yf) e }ykr t yks t( ) ( )1

*More than one model presented; ** Temperature(T) is in degreesK; *** Time in minutes.Nomenclature: TTHMstotal trihalomethanes; TTHMsinitial TTHM concentration; CHClschloroform; BDCMsBromodichloromethane; DBCMsDibromochloromethane; CHBrsO 3 3

bromoform; MCAAsmonochloroacetic acid; DCAAsdichloroaceic acid; TCAAstrichloroacetic acid; MBAAsmonobromoacetic acid; DBAAsdibromoacetc acid; UVsUV absorbanceat 254 nm(cm ); TOCstotal organic carbon(mgyl); NVTOCsnon-volatile organic carbon(mgyl); DOCsdissolved organic carbon(mgyl); POXspurgeable organic halide(mgyl);y1

NPOXsnon-purgeable organic halide(mgyl); Tswater temperature(8C); Flusfluorescence(%); Dschlorine dose(mgyl); fsfraction of the chlorine demand attributed to rapid reactions;C sresidual chlorine at the treatment plant after chlorination(mgyl); Cl sinitial chlorine concentration; Csinitial residual chlorine(mgyl); asparameter depending on location at whichO 1 1

chloroform is predicted;a sTTHM yield coefficient;´srandom error;bswater dispersion parameter in the water distribution system;kr andkssthe first order rate constants for rapid1

and slow reactions, respectively; Brsbromide ion(mgyl); tsreaction time(h); Ssdummy variable(summer); Spsdummy variable(spring); Gs1 for chlorinated compounds andGsy

0.0001 otherwise; Ch-aschlorophyll-a (mgym ); ai, bi, ci, di, ei, fi, gisconstants depending on type of DBP(see Clark et al., 2001); NFsdummy variable near or far;YsYear of3

sampling expressed by binary numbers; NRsNot reported.

31R

.Sadiq,

M.J.

Rodriguez

/Science

ofthe

TotalE

nvironment

321(2004)

21–46

Table 6bA summary of predictive models for other DBPs

AuthoryYear Data source n r 2 Output Units Predictive models

1. Ozekin, 1994 Laboratory NR NR Bromate yBrOŽ .3 mgyl 1.55=10 (DOC) (pH) (O ) (Br ) (t)y6 y26 5.82 5.82 y 0.73 0.283

Ozekin et al., 1998 for temperatures other than 208C, the bromateconcentration can be modified by followingrelationship

Ty20y yw x w xBrO s BrO 1.035Ž .3 T 3 20C

2. Siddiqui et al., 1994 Laboratory 70 0.94 CHBr3 mgyl 7.3 (DOC) (pH) (O ) (Br ) (T)1.33 y1.25 0.771 y 1.56 0.9093

30 0.78 CHBr3 (24 h predictions)70 0.95 TOBr 2.68(DOC) (pH) (O ) (Br ) (T) (t)1.28 y1.31 0.742 y 1.55 0.956 0.353

3

54 0.88 Bromate 5.1(DOC) (pH) (O ) (Br ) (T)1.07 1.05 0.766 y 1.53 1.083

22 0.64 Bromate (24 h predictions)173 0.68 Bromate 1.5=10 (DOC) (pH) (O ) (Br ) (u)y3 y0.74 y2.26 0.64 y 0.61 2.03

3

1.5 (DOC) (pH) (pq1) (Br )y0.75 y2.25 1.31 y 0.60

0.26 (DOC) (pH) (DO ) (Br ) (tq1)086 3.27 0.22 y 0.67 0.253

(for 0-t-1 h)

3. Song et al., 1996 Field 119–239 0.87–0.97 Bromate yBrOŽ .3 mgyl 13 different linear regression models(one per fractionand per water source) for bromate in function ofbromide, DOC, nitrogen ammonia, ozone dose,inorganic carbon and reaction time.

4. Korn et al., 2002 Laboratory 112 0.95 Chlorite mgyl exp(y0.346y0.07 log(pH)y0.025 log(T)y0.597 log(Cq1)y0.136 log(tq1)y0.0038 log(NPOCØUV254)q0.293 log(T)Ølog(Cq1)q0.393 log(pH)Ølog(Cq1)q0.67 log(NPOCØUV254)Ølog(Cq1)y0.161 log(NPOCØUV254)Ølog(tq1))

112 0.95 Chlorate mgyl exp(y1.99q0.62 log(pH)y0.09(T)q0.698 log(Cq1)y0.104 log(tq1)q0.046 log(NPOCØUV254)q0.389log(T)Ølog(Cq1)q0.346 log(Cq1)Ølog(tq1)q0.486 log(NPOCØUV254)Ølog(Cq1)y0.119 log(NPOCØUV254)Ølog(tq1))

Nomenclature: DOCsdissolved organic carbon(mgyl); O stransferred ozone doses(mgyl); pHsozonation pH value; Brsbromide ion concentration(mgyl);y3

tsreaction time(mints); UV-254sUV absorbance(cm ); Cschlorine dioxide concentration(mgyl); NPOCsnon-purgeable organic carbon(mgyl); Tstemperaturey1

(8C); DO sdissolved ozone concentration(mgyl); TOBrsTotal organic bromine(mgyl); psperoxone ratio(H O yO ); andusozonation temperature(8C).3 2 2 3

32R

.Sadiq,

M.J.

Rodriguez

/Science

ofthe

TotalE

nvironment

321(2004)

21–46

Table 7aA description of THM, HAA and total organic halogenated(TOX) predictive models(numbers refer to models presented in Table 6a)

Modelystudy description Main advantages Main limitations

1. Untreated water from the Holsten River(Knoxville, Tennessee) was used in Good performance of model despite the Models do not containthis study. A series of experiments was carried out under controlled conditions number of observations. reaction time as anof pH, temperature, bromide-ion concentration and applied chlorine dosage. explanatory variable.Concentrations of non-volatile total organic carbon(NVTOC) were also variedusing commercially available humic acid. A constant reaction time of 96 h wasused for generating TTHM data in the model development. The model wascalibrated using different iterative methods. A preliminary verification of the model wasperformed using data derived from a field-sampling program.

2. With same data set, the authors used iterative modelling methods to Present alternative modelling These models do not containcalibrate the model, which included Gaussian, Dud, Maruardt, and Gradient methodologies(most authors have used reaction time as an explanatorymethods. The models provided acceptable fits and more than regression models). variable.74% of the predicted values were within"15% of the measured values overall.

3. Modelling carried out with a database from chlorination experiments with Considered different organic Model considered neitherwater samples collected from the Sagami River(Japan) which were combined matter content in water. temperature nor bromide ionwith a solution of humic acid. They applied several sensitivity analysis concentration as explanatorytechniques to study the effects of the three main experimental variables variables.chlorine dose, water temperature and pH.

4. This work is the single most extensively cited reference in THM modelling. Models based on a very robust database Water quality and chlorinationThe authors developed a comprehensive database from laboratory chlorination and on several waters with variable doses do not representexperiments on raw waters with variable qualities selected from nine US characteristics. characteristics of treated water andrivers. Linear and non-linear regression models were developed to predict operations in real waterTHM formation potential and kinetics. Boundary conditions of explanatory utilities.variables were defined. Different models were developed for short term(t-8 h)and long term reaction times(t)24 h).

5. The authors used the database developed by Amy et al.(1987a) for building Models for the different THM species Models do not representpredictive models for specific THMs including chloroform, BDCM, DBCM and bromoform. are available. conditions of real water utilities.

6. Authors developed a mechanistic model for DBPs based on their formation Modelling approach allows at It is hard to judge about modelkinetics. Humic acids were isolated from Lake Kinneret(Israel) in order to examining the kinetics of THM capabilities since the number ofundertake chlorination at laboratory scale. Models were developed to predict formation. data on which it is based is notTTHMs as a function of concentration of humic acids, chlorine dose and contact time. known.

7. This work resulted in the development of models for the four THMs and five The study use independent databases for The water quality and theHAAs. This study is one of the first that considered chlorinated DBPs other than model validation. chlorination conditionsTHMs. The project combined three laboratory scale databases developed in four do not representdifferent studies across the US, including the work by Amy et al.(1987a). situations encountered inDifferent models were developed according to the ratio between chlorine dose real water utilities.and bromide content in water. The explanatory variables for HAAs appeared tobe comparable to those for THMs. Good predictive abilities of these modelswere observed.

8. This work was based on a field scaled study in which samples from Taipei’s The model included the water Models are based on very(Taiwan) water utility were collected twice from 18 locations during a period of dispersion in the distribution system as few data points.6 months. Using this data, the authors developed a predictive regression model an explanatory variable for THM

33R

.Sadiq,

M.J.

Rodriguez

/Science

ofthe

TotalE

nvironment

321(2004)

21–46

Table 7a(Continued)


for TTHMs for the distribution system as a function of TTHMs in the finished occurrence.drinking water(after treatment).

9. Authors developed a multiple regression model for chloroform using monthly Very simple model with relatively good It has very small database. Thesamples(during 1 year) from the water treatment plant and the finished water results. model does not contain reactionof the city of San Sebastian(Spain). In addition to the normal water quality and time as an explanatory variable.operational parameters, they used fluorescence as an indicator of NOM insteadof more common indicators mentioned in the literature such as TOC and UV-254.

10. In this work, one of the most comprehensive laboratory scaled databases for Models based on a very robust database River waters do not represent rawdeveloping models for THMs and NPOX were established. THM and NPOX representing very variable water quality. waters used to be treated byformation potentials were determined from chlorination of water samples from utilities and chlorine doses arethe Mississippi, Ohio and Missouri Rivers. Regression coefficients for higher than those practiced in theexplanatory variables were comparable to those presented in previous literature. drinking water industry.With the same data generated from the three US rivers above, models were alsodeveloped for predicting THM speciation and bromine incorporation factorswith accurate results(Rathbun, 1996a,b,c).The author found that the predictive capacity of the model did not improve with both UV-254and DOC incorporated in the regression analysis. A similar conclusion was obtained byOssenbruggen et al., 1988 who used the ridge regression technique to eliminateone or more explanatory variables that are correlated.

11. In this work, three water samples were collected at the intake of the water Simple model with very good predictive Model does not consider waterutility in Taiwan. Chlorination experiments were undertaken with variable ability. temperature variations.chlorine dosage at a constant temperature. For each sample, they developed asingle regression model representing THM formation. A simple THM modelwas subsequently developed combining the information of three samples.

12. In this work, samples were collected at six different locations in the water The predictive model incorporates The brominated THM species aredistribution system of the city of Salamanca(Spain). At each location, 11 spatial and temporal variations of not considered.samples were taken to represent seasonal variations of chlorinated DBPs and chloroform.other water quality parameters. These models used dummy variables to considerthe location within the distribution system and the effect of time of sampling.The model contained third and higher order polynomials of temperature whichmade it highly sensitive.

13. Models for the formation of POX and NPOX using chlorination of fulvic Model appeared with very accurate It is not possible to adequatelyacid present in water. This study was planned to investigate TOX formation, predictive abilities. judging the model capabilitiesbecause THM accounts for only 5–20% of the TOX compounds formed during since the number of data onchlorination. The sensitivity analysis results revealed that mainly chlorine which it is based is not known.dosage(35%), TOC (48%), and reaction time(12%) contributed to thevariability of POX. Similar results were obtained for NPOX.

14 and 15. In this work, a second order kinetic model is developed to describe The model allows characterizing THM It is not proved that the secondthe formation of TTHMs. In the model, the rate of reactions is considered to be formation as a function of chlorine order model would give betterproportional to the first power of the concentration product of hypochlorous acid demand. predictions than first-orderand the substances responsible for chlorine demand. The model is validated kinetics.using two different laboratory-scaled databases generated in previous work byClark et al.(1994a,b,c) and AWWARFyEPA study(Vasconcelos et al., 1996).

16. Authors developed multiple regression models for TTHM using the data The model represents the seasonal Not common NOM surrogates are

34R

.Sadiq,

M.J.

Rodriguez

/Science

ofthe

TotalE

nvironment

321(2004)

21–46

Table 7a(Continued)


generated from samples collected at four locations at the treatment plant of Athens(Greece). variations of THMs and is very considered(TOC and UVMultivariate regression models were developed for TTHM in accurate. absorbance).the finished drinking water leaving the plant. A particular feature of this modelis that chlorophyll-a was introduced in the model as an indicator of NOM.

17. TTHM models developed throughout this work were based on data Empirical based models were developed Water quality does not representgenerated from untreated US River waters. The model used the similar for haloacetic acids, chloral hydrate, and characteristics of treated water.explanatory variables in their original work(Amy et al., 1987a), except that their bromate.latest model uses DOC instead of both TOC and UV as indicators of NOM.

18. In this work, authors developed a kinetic model for THMs which incorporate Models constitute a valuable tool to Chlorination conditions were notthe ratio between bromide ion and chlorine. The kinetic model is calibrated analyse the impact of bromide in the comparable in each sample.using data from experimental chlorination of 17 surface and ground waters(with THM formation kinetics.variable characteristics) in New Zealand.

19. In this work, authors combined different databases for the development of A robust database is considered to Many assumptions have to beTTHM predictive models. They combined US laboratory scale data developed by develop simple and accurate models. made to applied models to field-Amy et al. (1987a), Rathbun(1996a) and Montgomery Watson and AWWA(1991) scaled data.which they validated using a field-scaled THM database for small waterutilities in the province of Quebec(Canada). Field-scaled models were alsodeveloped for typical values of DOC found in raw waters of southern Quebec.Sensitivity analyses and model validations were carried out using the field-scaled database. By combining US database, Milot et al.(2002) developedsubsequently a THM model using artificial neural networks(ANN), a methodcommonly used in artificial intelligence(not listed in Table 6a) (Rumelhart et al., 1994).

20. This work examines a different modelling approach for THM occurrence Propose an alternative modelling Models do not considerusing logistic regression. Instead of predicting THM concentrations, the approach which results can be used for operational characteristicsprobabilities of exceeding specified values of THM(thresholds of 40, 50, 60, 80 THM assessment in epidemiological (chlorine dose, pH) at the utilitiesor 100mgyl) were estimated. Modelling was carried out using THM data studies. under study.collected in several utilities of the province of Quebec(Canada).

21. Authors combined kinetics of DBP formation with empirical-based The work of the authors proposes It was not proved that thismodelling using regression analysis. They used a second order rate for DBP models which integrate mechanistic and approach gives better results thanformation and chlorine decay kinetics to predict these concentrations in water empirical methods providing a high classical regression models.distribution systems. The coefficients of second order reaction kinetics were flexibility for model applicationdetermined using regression models. In fact, several papers on predictingchlorine residuals and the formation of halogenated by-products have beenpublished by these authors(Clark, 1998; Clark and Sivaganesan, 1998; Clark et al., 2001Clark and Sivaganesan, 2002). Clark and co-workers developedgeneral formulation for 13 different DBPs including four THMs and nine HAAs. Thegeneral model was developed at constant TOC and temperature. Ther values2

were more than 0.95 except for MCAA, which was less than 0.60.

22. In this work, 3 year data were generated through sampling of nine points Models represent seasonality for THM The performance of models isof the Athens(Greece) water treatment plant. In addition to THMs, temperature, occurrence. relatively low.pH, dose and chlorophyll-a, bromide were measured.

23. In this work, field-scaled data for THMs and HAAs were collected in water Data and models have potential Models do not consider chlorineutilities of four different regions in Spain. Models were aimed at predicting applications for exposure assessment in dose, temperature, NOMoccurrence of HAA species using the THM species at predictors through epidemiological epidemiological indicators, etc.

35R

.Sadiq,

M.J.

Rodriguez

/Science

ofthe

TotalE

nvironment

321(2004)

21–46

Table 7a(Continued)


multivariate regression models

24. Data for development of THM and HAA regression models were generated through Data from the experimental chlorination Model performance varieschlorination experiments of treated waters of three utilities of the Quebec city region(Canada). allowed representing the real seasonal according to the DBP specie to beIn bench-scale experiments, values for water temperature, chlorine dose and variations of environmental and modelled and to the utility.contact time reproduced the operational conditions of the utilities under study. watering quality characteristics.

25. Data for modelling were generated from chlorination of samples collected in Model allowed examining the THM Developed model is usable forthe Mississipi river. Water samples were previously pre-treated using membrane formation according to the molecular research but not for operationalultrafiltration in order to investigate and model the formation of THMs weights of NOM fractions. purposes.according to various fractions of natural organic matter.

36R

.Sadiq,

M.J.

Rodriguez

/Science

ofthe

TotalE

nvironment

321(2004)

21–46

Table 7bA description of predictive models for other DBPs(numbers refer to models presented in Table 6b)

Model description Main advantage Main outcome

1. Authors modelled the bromate production as a result of ozonation. More than 10 Model is relatively simple to apply. The number of data points on whichthe modeldifferent sources of water were used. The model was developed at a constant was based is not known.temperature of 208C To overcome this problem, a relationship was developed toestimate bromate formation at different temperatures of water(as shown on Table 6b).Later, Sohn et al.(2001) successfully used the results of this study for modelling andmonitoring of bromate levels in water treatment systems.

2. In this work, a series of predictive models for brominated by-products are proposed. Model allowed to consider the effect of Model performance for bromate formation isData for modelling were generated by ozonating different surface waters and ground water temperature on bromate formation moderate.waters in the US. The data were collected from three water treatment plants.

3. Regression models for bromate formation were based on data generated through Very good performance of models for Variation of water temperature was notexperimental ozonation of four different surface and ground waters in the US. For each both calibration and external considered in the experiments. Thus,water sample NOM was fractionated using ultrafiltration and reverse osmosis into three validation databases. models do not allow predicting seasonaldifferent fractions. Models were developed for each ozonated fraction. Models were variations of bromate.validated with literature data.

4. Models were developed using data generated during bench-scale experiments using Predictive capabilities of Model is applicable only to similarsource water from seven drinking water treatment plants in Canada. A full factorial models are very high. conditions than those used for theiranalysis(2 ) design was used to conduct the experiments in this study instead of using4 calibration.traditional one-factor-at-a-time approach. Models for chlorine dioxide by-productsproposed in this work constitute the most complete available now days in the scientificliterature.


model validation. It does not allow them to con-clude the generalisation of these models. Anotherremark is that only few studies performed sensitiv-ity analysis in the DBPs predictions with respectto selected explanatory variables. Finally, moststudies do not include discussion on DBP modeluncertainty.

Tables 7a and 7b present critical analyses of themodels by describing their main advantages andtheir major outcomes. The common criticism overreported models based on laboratory-scaled data isthat the experimental chlorination assays do notconsider different water temperatures, and thechlorine doses are often much higher than thoseapplied in real utilities. Because water temperatureand disinfectant doses vary during the year in realutilities (as discussed in Section 2.1) such outcomecompromises the use of those models to predictfield-scaled seasonal variation. The common criti-cism over reported models based on field-scaleddata is that they are developed with a limitednumber of observations.

4. Potential uses of predictive models

Classically, models are used to identify therelative significance of water quality(NOM indi-cators, bromine, pH, etc.) and operational variables(disinfectant dose, water temperature, contact time,etc.) responsible for the formation of DBPs. Sen-sitivity analyses of the models(Tables 6 and 7)can identify the contribution of each variable inDBP formation potential. Other potential benefitsof developing predictive models for DBPs indrinking water are following:

a Water utility managers: Models can be used toguide decision-making for operational controlduring the treatment process, e.g. for adjustmentof pH and disinfectant dose or for controllinghydraulic residence time in reservoirs(contacttime) to minimize DBP formation. In addition,DBP models can be used as a tool to selectlocation for boosting chlorination residual levelsto ensure complete removal of microbes and aswell as minimization of DBPs formation. TheDBP models can also be combined with residualdisinfectant models to select the sampling points

for water quality control within the distributionsystem.

b Environmental epidemiologists: In some cases,models can be used for epidemiological studies(exposure assessment) and health risk assess-ment. They may be useful for estimating thehuman exposure to DBPs through drinkingwater by generating data for this purpose atdesired locations. Regulations of DBPs are rel-atively recent and the sampling frequencyrequired for compliance has been low. Conse-quently, current available data are not sufficientin the historical and geographical perspective toadequately conduct epidemiological studiesrelating to possible cancer and reproductiveeffects of these substances. By the use of sur-rogates, approximate estimations of temporaland geographical variability of DBPs can beinvestigated. In addition, these predictive modelsmay also be combined with exposure assessmentmodels for specific routes of DBPs’ exposurethrough drinking water.

c. Regulatory agencies to estimate the need forinfrastructure upgrading: During regulationsand standards updating, regulatory agenciesevaluate the benefits of risk reduction associatedwith DBPs. In addition, agencies must evaluatethe economic impacts in terms of upgradingtreatment plants or changing raw water sources.In combination with the other models(e.g. theremoval of organic precursors by different treat-ment processes), these predictive models can beused to evaluate the required reduction in pre-cursors on a regional basis, which allow com-pliance with DBP standards, and thus estimatethe infrastructure needs for upgrading of treat-ment facilities.

Some potential applications of predictive modelsare illustrated with the help of hypothetical casestudies in the following subsections.

4.1. Hypothetical case study 1

The assumption is that a municipality is cur-rently using a conventional water treatment facilityfor supplying drinking water. Chlorination is theonly disinfection method being employed for


Table 8Water quality and operational parameters

Explanatory variables Distribution Mean S.D. Minimum Maximum

Chlorine dose,D (mgyl) Lognormal 15 10 2 50pH Normal 7 1.5 4 10Contact time,t (h) Uniform 24 72DOC (mgyl) Lognormal 5 2 1.5 7.2Temperature,T (C) Uniform 20 30

microbial inactivation and maintenance of residualdisinfectant in the distribution system. The munic-ipality is evaluating options for upgrading theirfacility through alternative disinfection or remov-ing precursors(NOM) from the raw water tocomply with the regulatory regime, i.e. totalTHM)100 mgyl. The water managers decided toevaluate the second option using granular activatedcarbon(GAC) by taking into account the historicaldata distributed throughout the operational year ofthe water treatment facility.

Monte Carlo (MC) simulations are plannedbased on the uncertain nature of the data. Thestatistical distributions for these parameters aredefined (Table 8). For example, the model pro-posed by Rodriguez et al.(2000) as given in Table6a is used to estimate the required DOC concen-tration in the water to comply with the regulatoryguidelines of 100mgyl. Single values of totalTHM (e.g. one per season,) giving an annualaverage of 100mgyl could also be used. A Latinhypercube sampling based on MC simulations isused to determine DOC levels.

The variability in existing(before treatment)and required DOC concentrations(to comply with100 mgyl regulatory values) is shown by CDFs(cumulative distribution function) for comparisonpurposes(Fig. 2a). Here it is assumed that thechlorine dose is maintained to ensure disinfectionefficiency and acceptable residual chlorine in thedistribution system(although chlorine demandwould simultaneously be reduced). The removalefficiency (RE) of DOC required and complieswith water quality guidelines is shown by theprobability density function(PDF) in Fig. 2b.Generally, empirical models containing design par-ameters are available to calculate removal efficien-cy of various water treatment options. The removal

efficiency obtained from MC simulations can berelated to those models to obtain design parame-ters. For example, the mean and standard deviationof PDF are also shown in Fig. 2b, which can helpin modelling the design parameters of GAC, i.e.empty bed contact time(EBCT) and regenerationfrequency. These removal efficiency results areequally applicable for other treatment options.


The DBP models can also be used to estimatethe human exposure to these compounds throughvarious contact routes. The exposure assessmentmodels for chloroform provided by Jo et al.(1990), shown in Table 9, are coupled with thetotal THM predictive model(Rodriguez et al.,2000) to determine the exposure associated withthe drinking water.

Three exposure routes—ingestion, inhalationand dermal contact—are considered in this analy-sis. Inhalation exposure through showers(baths)is considered. The other possible routes are throughcooking and washing. To use the exposure assess-ment models, it is assumed that all four species ofTHMs behave like chloroform. The THM valuesare exposure concentrations, which are generatedusing the predictive model for treated levels ofDOC as given in case study 1 and Fig. 2a(Rod-riguez et al., 2000). The predicted THM concen-trations are then used in exposure models given inTable 9(Jo et al., 1990).

The PDFs of exposure through ingestion, inha-lation and dermal contact are shown in Fig. 3. Theuncertainties associated with DBP predictive andexposure assessment models, and the input para-meters and scenarios contribute to the uncertaintyin the actual exposure. These uncertainties can be


Fig. 2. Use of predictive models for DBPs in treatment process design(Case Study 1). (a) CDFs of DOC concentrations in waterfor existing and required levels of total THM.(b) PDF of required DOC removal efficiency to comply with guideline of 100mgyltotal THM.


Table 9Exposure assessment models for total THM(Jo et al., 1990)

Exposure route Models Definitions of terminology(mgykg day)

Ingestion I sE ØC ØA yWg i w w E sAbsorptional efficiency of TTHM from gastro-i

intestinal tracts1.0

Inhalation I sE ØC ØRØTyWh r a C sTTHM concentration in water(mgyl)w

(shower only) where A sWater amount ingested(lyday)sNormal;(2,0.5)w

C s10.45Ø(C )y99.6a w WsBody weight(kg)sLognormal;(60,6)TsShower times10 min.ydayRsBreathing rates0.014 mymin3

E sAbsorptional efficiency via respiratory tracts0.77r

C sTTHM concentration in shower(mgym )3a

Dermal D sFØIr h FsRatio of body burden from dermal exposure to that ofinhalation exposures0.93

Fig. 3. Results for model estimation of ingestion, inhalation and dermal exposure to total THMs(Case Study 2).

reduced by the collection of more data. The expo-sure assessment results can be further used inhuman health risk assessment.


As mentioned previously, the DBP models areuseful for assessing human exposure in epidemio-

logical studies. For example, the logistic-regressionmodels developed by Milot et al.(2000) can beused to estimate the probability that utilities exceedspecific values of THMs. These models have beendeveloped with distribution system data for theprovince of Quebec(Canada).

Using the historical information(cancer studiesrequire several years), the types of sources, treat-


Fig. 4. Model simulation of probability of THM levels being higher than 100mgyl according to season, type of treatment, type ofraw water and geographical location(Case Study 3).

ment methods(and their modifications), and theinformation from geographical regions(soil prop-erties and land use may vary), utilities can beclassified in a retrospective way in accordancewith their susceptibility of forming high THMconcentrations. Such a classification can be doneusing probability thresholds to distinguish low,medium or high susceptibility(additional catego-ries are possible). Using the complementary infor-mation required for epidemiological studies(consumption of tap water, control of socio-dem-ographic factors, etc.), a historical and geographi-cal assessment of exposure to THMs forpopulations served by the utilities of the area ofinterest can be undertaken.

Fig. 4 illustrates the application of the logisticmodel (Milot et al., 2000) for a reference THM

value of 100mgyl. Fig. 4 shows that probabilitiesof exceeding THM thresholds vary significantlyacross the categories of variables considered. Thus,in an epidemiological study, variability in expo-sures to THMs may play a key role.

5. Critical research needs

It appeared that a significant effort has beeninvested to develop predictive models for DBPs indrinking water. Models presented in this paper canbe categorised based on methodology for datageneration, the type of independentyexplanatoryvariables and the model usefulness. According tothe review, the main benefit for modelling appearsto be their usefulness to identify factors influencingDBP formation and fate followed by chlorination


of water. In fact, sensitivity analyses of thesemodels will easily allow determining the relativecontribution of water quality and operational par-ameters to the formation of DBPs. Some modelscan also be applied for predicting DBPs, butmainly subject to conditions(i.e. within the spe-cific range of independent variables) and for thespecific case that served for model development(experimental water or site-specific distributionsystem).

In the previous section, hypothetical case studieswere presented in the manners that DBP modelscould be to guide decision-making. However, thereis still lot of scope to improve the feasibility ofusing these models to predict DBPs for operational,epidemiological and regulatory purposes. Toachieve this, it is important that future work mustfocus on multidisciplinary research related tochemistry, engineering, toxicology, epidemiology,statistics and governance. According to this review,the authors believe that in coming years theresearch efforts must focus on the followingaspects:

● The evaluation of laboratory-scaled models forpredicting DBPs in field-scaled distribution sys-tem: Research efforts must focus on investigat-ing the capacity of models developed withlaboratory-scaled data to estimate real seasonaland spatial variations of DBPs in the distributionsystems. An important portion of this researchmust be the quantification of the distributionsystem contribution ‘pipe effect’ on increasingor diminishing different DBP species. Toachieve this goal, it is required that laboratoryand field-scaled data be developed at the sameutilities (simultaneous generation of data) orthe laboratory-scaled models may be developedbased on various sources of waters to generalisethem for various scenarios.

● The adaptation of laboratory-scaled models tobe made useful for estimation of DBPs for field-scaled system: A complementary research isrequired to develop strategies for simplificationand adaptation of laboratory-scaled to field-scaled methods to generate DBP data. A signif-icant challenge will be the better estimation ofwater residence time in distribution systems

(through hydraulic models, tracer studies, flowrate correlations, etc.). Model adaptation couldinclude the identification of conditions(seasons,water quality and operational ranges) at whichlaboratory-scaled models have better predictivecapability within systems and the use of correc-tion factors for other seasons. Such correctionfactors may vary according to the DBP specie.The application proposed by Westerhoff et al.(2000) is a preliminary attempt to achieve this.

● The development of methods to estimate andreduce the uncertainties in the predictions ofDBPs as well as to interpret them: With all thereviewed models, it is possible to calculate DBPlevels using explanatory variables. However, itis necessary to have more information about theconfidence and certainty of these data in orderto improve it. The use of fuzzy logic techniques(e.g. Sadiq and Rodriguez, 2003) may be aninteresting alternative to accomplish this. Thiskind of research effort will simultaneouslyfavour the applicability of the predictive mod-elling for operational, epidemiological and reg-ulatory purposes.

● Investigate the feasibility to integrate variousmodelling approaches to improve the predictivecapacities of DBP models: The majority of DBPmodels have been based on multivariate regres-sion. Future research must experience alterna-tive modelling techniques for DBPs predictions,such as artificial neural networks(Rumelhart etal., 1994; Milot et al., 2002), fuzzy rule-basemodelling(e.g. Sadiq et al., 2003) which couldimprove DBP predictions. Robust database onDBPs has to be developed in order to adequatelycompare different techniques(with separationof calibration and validation data). The use ofhybrid modelling methodologies may also beinvestigated, e.g. using different techniques toestablish DBP kinetic coefficients and relatethem to water quality and operational parame-ters, and then to reduce uncertainty in theirpredictions.

● The development and application of DBP mod-els that consider simultaneously the disinfectionefficiency and residual disinfectant maintenancein distribution system: Models have been report-ed in the scientific literature for each of these


issues but no evident effort has been done toconsider them simultaneously. As mentioned inthe first section of this paper, parameters affect-ing these issues are similar(water temperature,organic content, chlorine dose, reaction time,etc.). Feasibility for integration of these issuesin a multipurpose model has to be evaluated inthe near future, particularly for treated waterwithin the reservoir of treatment plants. Toachieve this, robust data must be developeddescribing seasonal variations in water qualityand operational changes.

● Development of models for other DBPs: Pro-gress in toxicological research allows at identi-fying specific DBPs having potentialimplications for human health, but limited infor-mation is available about their presence andfate in drinking water. Mostly reported modelsare for THMs, it emphasises that research mustfocus on the generation of laboratory and field-scaled data of other DBPs. With these data,relationships between the different DBPs’ spe-cies must be established and predictive modelsmust be developed. Examples of species tofavour in modelling are ozonation by-productsand chlorinated DBPs(e.g. acetonitriles, chlo-ropicrine and chloral hydrate) in addition toTHMs and HAAs.

● The development of criteria for assessment ofthe predictive capacity of DBP models: It isimportant to develop criteria, which favour auniform methodology for evaluating the predic-tive capacity of any DBP model. Such criteriamay include the requirements for using mini-mum amount of external data for validation andfor specific ranges of water quality and opera-tional conditions within which models could beapplied. Context in which DBP model can beapplied (e.g. geographical features, type ofwater source, water utility types, etc.), boundaryconditions for their application, as well as theirspecific potential usage(e.g. operational, epi-demiological, regulatory) should be included inthe criteria.

6. Summary and conclusions

This paper has reviewed various models andapproaches used for predicting DBPs’ occurrence

in drinking water. More particularly, the paper hasfocused, on one hand, on the disinfection practicesand the formation of DBPs under varying watertreatment conditions, and on other hand, on pre-dictive models for DBPs and their potential use.

Based on literature reviewed, different model-ling approaches have been used to relate waterquality and operational parameters with DBP con-centrations in water. Most of the models reportedin the literature use DOC(or TOC), disinfectantdose, pH, temperature, and reaction time as explan-atory variables. Some researchers used an entirelyempirical approach and some introduced kineticsinto the modelling process. Multiple linear andnon-linear regression techniques are found to bethe most common in developing DBP predictivemodels. Other methods like ridge, logistic regres-sion and artificial neural networks have also beenemployed. Most of the predictive models are basedon laboratory-scaled studies, but some models havebeen proposed based on actual water distributionsampling as well.

The DBP models can be useful for operationalpurposes during water treatment and water qualitymanagement, for the evaluation water treatmentfacilities, for exposure assessment in epidemiolog-ical studies and health risk assessment, and forestimating the benefits and impacts of DBP regu-lations. However, research is necessary to evaluatethe usefulness of DBP models and to adapt themfor the purposes mentioned.

The use of alternative disinfectants and othertreatment technologies have also increased theinterest in developing predictive models for DBPs.By development of sophisticated analytical tech-niques(experimental), new DBPs have been dis-covered recently. More toxicological informationon DBPs is available, therefore the developmentand use of models will be very helpful in thefuture to deal with these substances in drinkingwater.

Acknowledgments

The authors extend their gratitude to the Nation-al Research Council(NRC) Canada and the CRADof Universite Laval (Quebec City, Canada) for´their financial support of this strategic research.


References

Adin A, Katzhendler J, Alkaslassy D, Rav-Acha C. Trihalo-methane formation in chlorinated drinking water: a kineticmodel. Water Res 1991;25(7):797–805.

Amy GL, Chadik PA, Chowdhury ZK. Developing models forpredicting trihalomethane formation potential kinetics. J AmWater Works Assoc 1987a;79(7):89–96.

Amy GL, Minear RA, Cooper WJ. Testing and validation ofa multiple non-linear regression model for predicting trihal-omethane formation potential. Water Res 1987b;21(6):649–659.

Amy G, Siddiqui M, Ozekin K, Zhu H-W, Wang C. Empiricalbased models for predicting chlorination and ozonation by-products: haloacetic acids, chloral hydrate, and bromate.EPA report CX 819579, 1998.

Arora H, LeChevallier MW, Dixon KL. DBP occurrencesurvey. J Am Water Works Assoc 1997;89(6):60–68.

Aus–NZ. Australian drinking water guidelines, AustralianNational Health and Medical Research Council, http:yywww.health.gov.auynhmrcypublicationsysynopsesyeh19syn.htm, 2000.

Bellar TA, Lichtenberg JJ, Kroner RC. The occurrence oforganohalides in chlorinated drinking waters. J Am WaterWorks Assoc 1974;66:703–706.

Bercz JP, Jones L, Garner L, Murra D, Ludwig DA, Boston J.Sub-chronic toxicity of chlorine dioxide and related com-pounds in drinking water in the non-human primate. EnvironHealth Persp 1982;46:47–55.

Berman D, Hoff JC. Inactivation of simian rotavirus SA11 bychlorine, chlorine di-oxide, and mono-chloramine. ApplEnviron Microbiol 1984;48(2):317–323.

Berman D, Rice EW, Hoff JC. Inactivation of particle-associ-ated coliforms by chlorine, and mono-chloramine. ApplEnviron Microbiol 1988;54(2):507–512.

Berman D, Sullivan R, Hurst CJ. Effect of the method ofpreparing mono-chloramine upon inactivation ofMS2 coli-phage, Escherchia coli, and Klebsiella pneumoniae. Can JMicrobiol 1992;38:28–33.

Blaser MJ, Smith PF, Wang WL, Hoff JC. Inactivation ofCampylobacter jejuni by chlorine and mono-chloramine.Appl Environ Microbiol 1986;51(2):307–311.

Cantor KP, Lynch CF, Hildesheim ME, Dosemeci M, Lubin J,Alavanja M, Craun G. Drinking water source and chlorina-tion by-products I. Risk of bladder control. Epidemiology1988;9(1):21–28.

Chang EE, Chao S, Chiang P, Lee J. Effects of chlorinationon THM formation in raw water. Toxicol Environ Chem1996;56:211–225.

Chen WJ, Weisel CP. Halogenated DBP concentrations in adistribution system. J Am Water Works Assoc1998;90(4):151–163.

Chowdhury ZK, Amy GL, Mohamed S. Modelling effects ofbromide ion concentration on the formation of brominatedtrihalomethanes, Water Research for the New Decade Pro-ceedings, AWWA Annual Conference, Denver, Colorado,US; 1991. pp. 313–322.

Clark RM. Chlorine demand and TTHM formation kinetics: asecond order model. J Environ Eng 1998;124(1):16–24.

Clark RM, Sivaganesan M. Predicting chlorine residuals andformation of TTHMs in drinking water. J Environ Eng1998;124(12):1203–1210.

Clark RM, Sivaganesan M. Predicting chlorine residuals indrinking water: second order model. J Water Resour PlannManage 2002;128(2):152–160.

Clark RM, Read EJ, Hoff JC. Analysis of inactivation ofGiardia lamblia by chlorine. J Environ Eng1989;115(1):80–90.

Clark RM, Adams JQ, Lykins BW. DBP control in drinkingwater: cost and performance. J Environ Eng1994a;120(4):759–782.

Clark RM, Pierce EW, Johnson CH, Fox KR. Effect ofaggregation onVibrio cholerae inactivation. J Environ Eng1994b;120(4):857–887.

Clark RM, Smalley G, Goodrich JA, Tull R, Rossman LA,Vascocelos JJ, Boulos PF. Managing water quality in distri-bution systems: simulating TTHM and chlorine residualpropagation. J Water Supply Res Technol–Aqua1994c;43(4):182–191.

Clark RM, Thurnau RC, Sivaganesan M, Ringhand P. Predict-ing the formation of chlorinated and brominated by-prod-ucts. J Environ Eng 2001;127(6):493–501.

Croue JP, Gorshin GV, Leenheer JA, Benjamin MM. Isolation,´fractionation and characterization of natural organic matterin drinking water, AWWARF report, 1998.

Edzwald JK, Becker WC, Wattier KL. Surrogate parametersfor monitoring organic matter and THM precursors. J AmWater Works Assoc 1985;77(4):122–132.

Gallard H, von-Gunten U. Chlorination of natural organicmatter: kinetics of chlorination and of THM formation.Water Res 2002;36:65–74.

Gang D, Clevenger TE, Banerji SK. Relationship of chlorinedecay and THMs formation to NOM size. J Hazard Mater2003;A96:1–12.

Garcia-Villanova RJ, Garcia C, Gomez JA, Garcia MP, Ardan-uy R. Formation, evolution and modelling of trihalometha-nes in the drinking water of a town: I. at the municipaltreatment facilities. Water Res 1997a;31(6):1299–1308.

Garcia-Villanova RJ, Garcia C, Gomez JA, Garcia MP, Ardan-uy R. Formation, evolution and modelling of trihalometha-nes in the drinking water of a town: II. In the waterdistribution system. Water Res 1997b;31(6):1405–1413.

Gates D. The chlorine dioxide handbook, water disinfectionseries. Denver, Colorado: American Water Works Associa-tion, 1998.

Golfinopoulos SK, Arhonditsis GB. Multiple regression mod-els: a methodology for evaluating trihalomethane concentra-tions in drinking water from raw water characteristics.Chemosphere 2002;47:107–1018.

Golfinopoulos SK, Xilourgidis K, Kostopoulou N, Lekkas TD.Use of a multiple regression model for predicting trihalo-methane formation. Water Res 1998;32(9):2821–2829.

Graves CG, Matanoski GM, Tardiff RG. Weight of evidencefor an association between adverse reproductive and devel-


opmental effects and exposures to disinfection by-products:a critical review. Regul Toxicol Pharmacol 2002;34:103–124.

Health Canada. A one-year survey of halogenated disinfectionby-products in the distribution system of treatment plantsusing three different disinfection processes, Minister ofPublic Works and Government Services Canada, H46-2y96-206E, 1996.

Health Canada. Summary of guidelines for drinking waterquality, Federal-Provincial subcommittee, www.hc-sc.gc.cayehdybchywater_quality.htm, 2001.

Hoff JC. Inactivation of microbial agents by chemical disin-fectants, United States Environmental Protection Agency,EPA 600-2-86-067, 1986.

Huixian Z, Sheng Y, Xu X, Ouyong X. Formation of POXand NPOX with chlorination of fulvic acid in water: empir-ical models. Water Res 1997;31(6):1536–1541.

Ibarluzea JM, Goni F, Santamaria J. Trihalomethanes in watersupplies in the San Sebestian area Spain. Bull EnvironContam Toxicol 1994;52:411–418.

Jo WK, Weisel CP, Lioy PJ. Chloroform exposure and thehealth risk associated with multiple uses of chlorinated tapwater. Risk Anal 1990;10(4):581–585.

Korich DG, Mead JR, Madore MS, Sinclair NA, Sterling CR.Effects of ozone, chlorine di-oxide, chlorine, and mono-chloramine on Cryptospordium parvium oocysts viability.Appl Environ Microbiol 1990;56(5):1423–1428.

Korn C, Andrews RC, Escobar MD. Development of chlorinedi-oxide related by-product models for drinking water treat-ment. Water Res 2002;36:330–342.

Lou JC, Chiang PC. A study of trihalomethanes formation inwater distribution system. Hazard Waste Hazard Mater1994;11(2):333–343.

Lykins BW, Griese MH. Using chlorine dioxide for trihalo-methane control. J Am Water Works Assoc 1986;78(6):88–93.

Milot J, Rodriguez MJ, Serodes J. Modelling the susceptibility´of drinking water utilities to form high concentrations oftrihalomethanes. J Environ Manage 2000;60:155–171.

Milot J, Rodriguez MJ, Serodes J. Contribution of neural´networks for modelling THM occurrence in drinking water.J Water Resour Plann Manage-ASCE 2002;128(5):370–376.

Minear R, Morrow C. Raw water bromide in finished drinkingwater, Research Report 9, Water Resources Research Center,University of Tennessee, 1983.

Montgomery Watson and AWWA. Disinfectionydisinfectionby-products database and model project, American WaterWorks Association, Denver, Colorado, 1991.

Montgomery Watson. Mathematical modelling of the formationof THMs and HAAs in chlorinated natural waters. Denver,Colorado: AWWA, 1993.

Morrow C, Minear R. Use of regression models to link rawwater characteristics to trihalomethane concentrations indrinking water. Water Res 1987;21(1):41–48.

Nokes CJ, Fenton E, Randall CJ. Modelling the formation ofbrominated trihalomethanes in chlorinated drinking waters.Water Res 1999;33(17):3557–3568.

Ossenbruggen PJ, Gaudard MA, Collins MR. System modeldevelopment with ill-conditioned data: case studies of tri-halomethane formation in drinking water. Civil Eng Syst1988;5(1):31–41.

Ozekin K. Modelling bromate formation during ozonation andassessing its control, Ph.D. Dissertation, University of Col-orado, US, 1994.

Ozekin K, Westerhoff P, Amy GL, Siddiqui M. Molecularozone and radical pathways of bromate formation duringozonation. J Environ Eng 1998;124(5):456–462.

Rathbun RE. Regression equations for disinfection by-productsfor the Mississippi, Ohio and Missouri Rivers. Sci TotEnviron 1996a;191:235–244.

Rathbun RE. Speciation of trihalomethane mixtures for theMississippi, Missouri and Ohio rivers. Sci Tot Environ1996b;180:125–135.

Rathbun RE. Bromine incorporation factors for trihalomethaneformation for the Mississippi, Missouri and Ohio rivers. SciTot Environ 1996c;192:111–118.

Rice EW, Clark RM, Johnson CH. Chlorine inactivation ofEscherichia coli O157: H7. Emerg Infect Dis1999a;5(3):461–463.

Rice EW, Rodgers MR, Wesley IV, Johnson CH, Tanner SA.Isolation of Arcobacter butzleri from ground water. LettAppl Microbiol 1999b;28:31–35.

Rice EW, Johnson CH, Naumovitz DW, Marshall MM, Plum-mer CB, Steling CR. Chlorine disinfection studies of Ence-phalitozoon (Septata) intestinalis, Poster presentation atannual meeting of the American Society for Microbiology,Chicago, IL; May 30–June 3, 1999c.

Richardson SD. Drinking water disinfection by-products. In:Robert AM, Richard SD, editors. John Wileys encyclopaediaof environmental analysis and remediation, vol. 3John Wil-ey’s, 1998. p. 1398–1421.

Rodriguez MJ, Serodes J, Morin M. Estimation of water utilitycompliance with trihalomethane regulations using modellingapproach. J Water Supply Res Technol–Aqua2000;49(2):57–73.

Rodriguez MJ, Serodes JB. Spatial and temporal evolution of´trihalomethanes in three water distribution systems. WaterRes 2001;35(6):1572–1586.

Rook JJ. Formation of haloforms during chlorination of natu-rals waters. Water Treatment Examination 1974;23:234–243.

Rossman LA, Brown RA, Singer PC, Nuckols JR. DBPformation kinetics in a simulated distribution system. WaterRes 2001;35(14):3483–3489.

Rumelhart DE, Widrow B, Lerh MA. The basic ideas in neuralnetworks. Comm ACM 1994;37(3):87–92.

Sadiq R, Rodriguez MJ. Fuzzy synthetic evaluation of disin-fection by-products—a risk-based indexing system. Submit-ted to J Environ Manage 2003.

Sadiq R, Kar S, Husain T. Chloroform associated health riskassessment using bootstrapping: a case study for limited


drinking water samples. J Water Air Soil Pollut 2002;138(1–4):123–140.

Sadiq R, Al-Zahrani MA, Sheikh AK, Husain T, Farooq S.Performance evaluation of slow sand filters using fuzzyrule-based modelling, Environ Model Softw 2003(in press).

Serodes JB, Rodriguez MJ, Li H, Bouchard C. Occurrence ofTHMs and HAAs in experimental chlorinated waters of theQuebec City area(Canada). Chemosphere 2003;51(4):253–263.

Sharfenaker MA. US EPA offers first glimpse of stage 2 DyDBPR. J Am Water Works Assoc 2001;93(12):20–34.

Siddiqui M, Amy G, Ozekin K, Westerhoff P. Empirically andtheoretically based models for predicting brominated ozon-ated by-products. Ozone Sci Eng 1994;16:157–178.

Singer PC, Chang SD. Correlations between trihalomethanesand total organic halides formed during water treatment. JAm Water Works Assoc 1989;81(8):61–65.

Singer PC, Obolensky A, Greiner A. DBPs in chlorinatedNorth Carolina drinking water. J Am Water Works Assoc1995;87(10):83–92.

Song R, Donohoe C, Minear R, Westerhoff P, Ozekin K, AmyG. Empirical modeling of bromate formation during ozon-ation of bromide-containing waters. Water Res1996;30(5):1161–1168.

Sohn J, Gatel D, Amy G. Monitoring and modelling ofdisinfection by-products(DBPs). Environ Monit Assess2001;70:211–222.

Stevens AA, Slocum CJ, Seeger DR, Robeck GB. Measure-ment of THM and precursor concentration changes. J AmWater Works Assoc 1976;68:546–554.

UK. Water supply(water quality) regulations 2000 for Englandand Wales, www.dwi.detr.gov.ukyregsysi3184y3184.htm,2000.

Urano K, Wada H, Takemassa T. Empirical rate equation fortrihalomethane formation with chlorination of humic sub-stances in water. Water Res 1983;17(12):1797–1802.

US EPA. Microbial and disinfection by-product rules—simul-taneous compliance guidance manual, United States Envi-ronmental Protection Agency, EPA 815-R-99-015, 1999a.

US EPA. Disinfection profiling and benchmarking guidancemanual, United States Environmental Protection Agency,EPA 815-R-99-013, 1999b.

US EPA. National primary drinking water standards, UnitedStates Environmental Protection Agency, EPA; 816-F-01-007, 2001a.

US EPA. Controlling disinfection by-products and microbialcontaminants in drinking water, United States EnvironmentalProtection Agency, EPA; 600-R-01-110, 2001b.

Vasconcelos JJ, Boulos PF, Grayman WM, Kiene L, Wable O,Biswas P, Bhari A, Rossman LA, Clark RM, Goodrich JA.Characterization and modelling of chlorine decay in distri-bution systems. Denver, CO: AWWA Research Foundation,1996.

Villanueva CM, Kogevinas M, Grimalt JO. Haloacetic acidsand trihalomethanes in finished drinking waters from het-erogeneous sources. Water Res 2003;37:953–958.

von-Gunten U, Hoigne J, Bruchet A. Bromate formation duringozonation of bromide-containing waters. Water Supply1995;45–50.

Westerhoff P, Debroux J, Amy GL, Gatel D, Mary V, CavardJ. Applying DBP models to full-scale plants. J Am WaterWorks Assoc 2000;92(3):89–102.

WHO. Guidelines for drinking-water quality, 2nd ed, Vol. 1,Recommendations, World Health Organization, Geneva,1993.

Disinfection by-products (DBPs) in drinking water and predictive models for their occurrence: a...

Documents

Transcript of Disinfection by-products (DBPs) in drinking water and predictive models for their occurrence: a...