Grantee Research Project Results
Final Report: Mechanistic-Based Disinfection and Disinfection Byproduct Models
EPA Grant Number: R826831Title: Mechanistic-Based Disinfection and Disinfection Byproduct Models
Investigators: Westerhoff, Paul , Amy, Gary , Reckhow, David A. , Chowdhury, Zaid
Institution: Arizona State University , Malcolm Pirnie , University of Colorado at Boulder , University of Massachusetts - Boston
EPA Project Officer: Hahn, Intaek
Project Period: December 15, 1998 through December 14, 2001
Project Amount: $339,583
RFA: Drinking Water (1998) RFA Text | Recipients Lists
Research Category: Drinking Water , Water
Objective:
The overall objective of this research project was to develop and calibrate an accurate kinetic-based mechanistic model for several chlorinated disinfection byproducts (DBPs) of interest. The model will predict DBPs, e.g., four trihalomethanes (THM) species (THM4), and nine haloacetic acids (HAA) species (HAA9), as a function of dissolved organic carbon (DOC), disinfectant level (type and dosage), reaction time, temperature, pH, and bromide concentrations. The specific objectives of the research project were to:
- Compile existing databases on DBP formation experiments into a single unified database. Some data from the compiled database will be used to develop and/or verify mechanistic DBP prediction equations. Data deficiencies will be identified.
- Conduct controlled batch-scale experiments with raw/untreated targeted at transforming the amount (mg/L) and chemical structure of natural organic matter (NOM)/DBP precursors. Transformations in NOM properties will be quantified.
- Develop and calibrate numerical models for predicting the behavior of disinfectants (free-chlorine) and the formation of DBPs (THMs and HAAs). Controlled experiments will be performed to assess inorganic reactions, disinfectant decay, DBP formation, and DBP stability. Model parameters for DBP formation will be statistically compared against NOM properties.
- Develop an easy-to-use computer model capable of predicting DBP formation, through a combination of mechanistic subroutines, as a function of disinfectant decay and water quality conditions.
The water industry faces new challenges in understanding and controlling DBP formation as health concerns demonstrate a need for more stringent regulatory DBP requirements. Mechanistic tools for understanding and predicting the rate and extent of DBP formation are required to facilitate the evaluation of DBP control alternatives. Accurate predictive models for DBPs can facilitate the evaluation of treatment alternatives for disinfection and DBPs. For this reason, the U.S. Environmental Protection Agency (EPA) has developed a water treatment plant simulation model (Harrington, et al., 1992) that incorporates the current state of knowledge for predicting DBP formation based upon the water quality entering a treatment plant, chemical dosages applied at various locations within the treatment process, and the detention times in these processes. This model is used for conducting regulatory impact assessments in support of developing DBP regulations. However, the current DBP modeling approach is empirical-based, rather than mechanistic-based.
Approach:
DBP experimental data from completed projects conducted by the Investigators and other researchers will be integrated into a single Unified Database. Existing empirical models and newly developed numerical models will initially be calibrated with our Unified Database. Additional experimental data will be collected since prior databases lack complete documentation of NOM characteristics before and during disinfection addition. Controlled laboratory disinfection and DBP formation studies will be conducted using water collected at several points through different water treatment plants, including raw, coagulated, softened, and pre-oxidized (ozone and/or chlorine dioxide) waters, thus the waters represent a wide range of water qualities and NOM characteristics. Experiments will investigate the affects of pH and temperature, and NOM, bromide, and free-chlorine concentrations; DBP hydrolysis studies will also be conducted.Summary/Accomplishments (Outputs/Outcomes):
Our findings are described below under four tasks: (1) formulation of mechanistic models and initial model calibration; (2) bench-scale experiments to quantify alteration in NOM structure prior to chlorination and DBP formation; (3) parameterization of empirical and mechanistic models; and (4) coding mechanistic models into EPA water treatment plant simulation modeling software.
Task 1: Formulation of Mechanistic Models and Initial Model Calibration
Based on the literature reports of mechanisms involved in chlorination process, and in light of previous observations during chlorination of natural waters, a comprehensive mechanistic model was developed. In brief, the model predicts chlorine decay and DBP formation through development of chemical rate laws. Major pathways for DBP formation are schematically represented in Figure 1. This model simulates DBP formation as a set of reactions between chlorine, bromine, and NOM as represented by three classes of reactive sites.
Sites with free chlorine or free bromine react nearly instantaneously, resulting in what is observed as instantaneous chlorine demand and DBP formation. The timescale for these “instantaneous” reactions are from less than 1 minute to a few minutes. S1 sites react with chlorine (HOCl/OCl-) or bromine (HOBr/OBr-), where the rate-limiting step is 2nd order (1st order in NOM and 1st order in oxidant). The S1 pathway produces the initial, “fast” formation part of the observed DBP versus time curve; timescales of a few minutes to a few hours. S2 sites react with chlorine or bromine more slowly than S1 sites. S2 sites are in equilibrium between protonated (S2H) and deprotonated (S2-) species. However, only the deprotonated form (S2-) reacts with chlorine or bromine, and this secondary reaction becomes rate limiting. The S2 pathway results in the final, “slow” formation part of the observed DBP versus time curve; timescales of a few hours to less than 100 hours.
For modeling chlorine and bromine reactions with NOM and formation of DPBs mechanistically, a key difficulty was the representation of NOM. NOM contains a heterogeneous mixture of organic compounds with quite different structures and characteristics, and varies as a function of hydrogeology and biogeochemistry within watersheds. As a basis for understanding NOM reactions with chlorine and bromine, interpretation of reactions of model organic compounds (e.g., resorcinol) have been postulated and extrapolated for NOM. The prototype for the S1 and S2 reaction pathways are resorcinol-type and β-diketones type reactions, respectively. For S1, carbon atoms that are activated by OH substituents or phenoxide ions in an alkaline environment react with chlorine (HOCl/OCl-) and bromine (HOBr/OBr-), very fast. Oxidative or hydrolytic cleavages at different locations lead to THMs, HAAs, or other DBPs. For S2, the activated carbon atoms on diketones or structures that can be oxidized to diketones will become fully substituted with chlorine and bromine. Monoketone groups are formed via rapid hydrolysis from these structures. DBP formation depends on the R group in the monoketone group. If R is a hydroxyl group, X2AA will form. Otherwise, the structure will further chlorinate to a trihaloromethyl species. This intermediate species is base-hydrolyzable to a THM. At neutral pH, if R is an oxidizable functional group (OFG) capable of readily donating an electron pair to the rest of the molecule, X3AA is expected to form after oxidation. If R is not an OFG, then hydrolysis will prevail and a THM is expected to form (Reckhow and Singer, 1985).
Figure 1. Major Pathways for DCB Formation (Adapted from McClellan, 2000)
A unified database of chlorination studies over the past decade was accumulated with data from researchers primarily in North America and New Zealand. Evaluation of these previous chlorination and DBP formation studies indicated a lack of kinetic data in the 0 to 2 hour time interval, which was critical for modeling “S1-type” reactions, or were primarily full-scale rather than laboratory data sets. Previous databases lacked one or more of the following kinetic parameters: chlorine residual, THM species, and HAA species measurements. The preliminary modeling results and evaluation of existing databases provided a framework for collection of additional databases that included appropriate kinetic time intervals, elevated bromide concentrations, and a wide range of NOM characteristics. Bench-scale experiments were therefore conducted (Task 2) to fill in data gaps identified in the unified database, which were necessary for calibration of mechanistic models.
Task 2: Bench-Scale Experiments to Quantify Alterations in NOM Structure Prior to Chlorination and DBP Formation
The purpose of the bench-scale experiments was to obtain a wide range of NOM material from natural waters that simulated a range of treatment (NOM removal processes) and disinfection (chlorine dose) conditions representative of full-scale water treatment facilities. Raw/untreated waters were collected from four locations around the United States: (1) Central Arizona Project (CAP) (Scottsdale, AZ) a canal that is a diversion from the Colorado River; (2) Lake Houston (LH) (Houston, TX); (3) Harwoods Mill Reservoir (HMR) (Yorktown, VA); and (4) Lake Manatee (LM) (Bradenton, FL). In large batch or continuous flow laboratory experiments, each raw water was subjected to six simulated water treatment processes aimed at altering the structure of NOM in solution: (1) filtered raw water; (2) alum coagulated water (10 mg alum/mg total organic content [TOC]); (3) ozonation (1 log Cryptosporidium inactivation); (4) chemical softening (pH 11); (5) 2500 dalton charged-surface ultrafiltration membrane separation; and (6) 7-day powder activated carbon adsorption. After each experimental treatment, samples were used for NOM “profiling” (molecular weight, hydrophobic/hydrophilic characterization, fluorescence spectrometry) and separate samples were used for kinetic chlorination studies (chlorine residual, THM species, HAA species).
NOM Profiling. NOM profiling verified that the process treatments altered the NOM structure. In general, alum coagulation removed DOC for waters with specific ultraviolet absorbance (SUVA) less than 2 m-1(mgDOC/L)-1, resulting in NOM profiling that was characterized by lower weight averaged molecular weight (MWW), but similar polarity as untreated water. This work utilized an innovative approach for determination of MWW: DOC detection after size exclusion chromatography (SEC). Ozonation always reduced SUVA and shifted DOC to more polar material as oxygen functionality was developed. Lime softening removed less than 15 percent of the DOC or SUVA, except in CAP water where slightly higher removals were observed. Ultrafiltration removed between 45 percent CAP and 95 percent LM, with 74 percent and 87 percent removal for LH and HMR, respectively. The decrease in SUVA and MWW was less than the change in DOC during ultrafiltration. The negatively charged membrane preferentially removed hydrophobic and more polar (i.e., transphilic) NOM fractions. Activated carbon treatment preferentially removed hydrophobic NOM. Similar trends in alteration of NOM “profiling” was observed for each process treatment in the four waters.
Chlorination. Kinetic chlorination experiments were conducted under variable chlorine dose, pH, bromide, and temperature conditions. An orthogonal CAP or semi-factorial (LH, HMR, LMW) matrix of experiments were conducted. Conditions were selected to be highly representative of actual chlorination conditions encountered at full-scale facilities, and differed from previous studies reported in the unified database:
- Chlorine dose: Baseline dose based upon obtaining 1 mg/L after 24 hours
- pH: Baseline was pH 7.5 (5.5 to 9.5)
- Bromide: Baseline was ambient level (spiked with 0.1 to 0.5 mg/L)
- Temperature: Baseline was 15°C (2 to 25°C)
CAP, HMR, and LH treated water samples (limited work with LM water) were chlorinated (3 L reactor), stored in separate biological oxygen demand (BOD) bottles, and kinetic samples collected by sacrificing BOD bottles at approximately 5, 35, and 60 minutes plus 2, 4, 8, 24, 48, and 100 hours. Approximately 100 separate chlorination experiments were conducted, each with kinetic chlorine, THM, and HAA samples analyzed. THMs were quenched with phosphoric acid and ammonium chloride prior to analysis (modified EPA Method 551). HAAs were quenched with sodium sulfite prior to analysis (modified EPA Method 552). During spike recovery tests, brominated HAA species had less than acceptable recoveries. It was determined that sodium sulfite, used as a quenching agent, resulted in occasional loss of bromide atoms from brominated and mixed halogen HAA species; chlorinated HAAs (μg/L) and a sum of nine HAAs (molar basis) were unaffected. The HAA results are therefore only valid as a molar sum of nine HAAs.
Two contrasting reaction times were observed: (1) 0 to 20 minutes representing fast chlorine and DBP reactions, and (2) 20 minutes to 4 days representing slower chlorine and DBP reactions. For CAP, LH, and HMR waters at the median percentile, 25 percent of the chlorine reacted within the first 20 minutes (ranged from ~ 0 percent for ultrafiltered or activated carbon treated samples, to 60 percent at the 90th-percentile of samples). Likewise, the median (10th- and 90th-percentiles) short-term THM4 and HAA9 were 15 percent (5 percent, 30 percent) and 20 percent (10 percent, 80 percent) of the long-term DBP formation, respectively. Thus, experimental observations supported the postulated two-phase (faster and slower) reactions proposed for the mechanistic model (see Figure 1).
Trends observed for pH, temperature, and bromide were similar to those previously reported. However, the contribution of the current study was simultaneous measurements of chlorine and DBP species under both short-term and long-term kinetic conditions. Overall, for all chlorination tests, the following trends were observed: (1) lowering pH increased trichloroacetic acid (TCAA) and HAA9 formation, with a greater effect on TCAA than on HAA9; (2) as pH decreased, so did the relative HAA rate of formation, which may indicate that intermediate structures that are present undergo base-catalyzed hydrolysis to THMs at higher pHs; and (3) at low pH, hydrolysis occurs slowly, so these intermediates are instead oxidized to form HAAs.
Lowering temperature generally decreased the absolute DBP formation levels, and slowed the reaction kinetics. THM and HAA kinetics were not equally affected. The results suggested that the pathway for the common precursor to THMs is far more temperature sensitive than the pathway leading to HAAs.
Increasing bromide ion concentration lowered the concentration of chlorinated DBPs, but had relatively no change on the total molar DBP concentration in comparison against the ambient bromide levels.
Task 3: Parameterization of Empirical and Mechanistic Models
Both empirical and mechanistic models were developed from the kinetic batch experiments performed in Task 2. After parameterization of the models, statistical analysis for the relative accuracy of the modeling output (chlorine residual, THM, and HAA concentrations) was compared to evaluate the benefits of the two modeling approaches. Major findings are presented in this section. Experimental results were parameterized based upon the following four “cases”:
Case I: CAP water experiments only
Case II: LH water experiments only
Case III: HMR water experiments only
Case IV: CAP, LH, and HMR water experiments.
Empirical Models. Values for fitting model coefficients were statistically estimated from least squares fitting analysis. During the model development, the linear, reciprocal, and power-function forms were all tested, among which power function relationship showed the greatest fits for both chlorine decay and DBPs formation. The models took the following forms, where Ci, Ti, Di, and Hi were fitted parameters:
Cl2 = C0*(TOCC1)*(pHC2)*(timeC3)*(tempC4)*(BrC5)*(CLdoseC6)*(UVAC7)
THM4 = T0*(TOCT1)*(pHT2)*(timeT3)*(tempT4)*(BrT5)*(CLdoseT6)*(UVAT7)
Di-HAA = D0*(TOCD1)*(pHD2)*(timeD3)*(tempD4)*(BrD5)*(CLdoseD6)*(UVAD7)
Cl3AA = H0*(TOCH1)*(pHH2)*(timeH3)*(tempH4)*(BrH5)*(CLdoseH6)*(UVAH7).
Mechanistic Models. Using a set of fixed values for rate and distribution coefficients for reactions illustrated in Figure 1, the mechanistic model was utilized to fit (i.e., optimize) NOM reactive site concentrations (S1 and S2-H), initial chlorine demand, and instantaneous DBP formations from experimental data, one treatment at a time. Conversion of the molar S1 and S2-H site concentrations to mgDOC per liter, based upon a carbon molecular weight of 12 g/mole, was conducted. S1 (0.5 to 8 μM) and S2-H (1 to 19 μM) sites accounted for only 1 percent and 3 percent of the DOC, respectively. Therefore, the number of chlorine reactive sites present relative to the total amount of DOC is quite small. S1 and S2-H site concentrations were correlated with DOC and ultraviolet absorbance (UVA) (R2 = 0.78), or NOM polarity (R2 = 0.74), suggesting that NOM profiling results could be used to estimate S1 and S2-H values:
[S1] = 5.05(DOC)0.57 (UV254)0.54
[S2-H]T = 13.1(DOC)0.38 (UV254)0.40.
Comparison of Modeling Approaches. Using a consistent database (i.e., individual CASE), both empirical and mechanistic models were compared based upon paired t-test, using average error (AE) values from empirical and mechanistic model simulations of observed data from Task 2. A confidence level of 95 percent was selected. This analysis determined whether results from the empirical model, and mechanistic model were “equivalent” by comparing t-statistic (t*) and t-critical two-tail (tc). Overall, the calibrated mechanistic model was more accurate than the calibrated empirical model, when applied to a specific water source (CASE I through III), than when applied to multiple water sources (CASE IV). This observation was consistent with the difference between central tendency empirical models and mechanistic models, which reflects the character/structure of DOC rather than simply the amount of DOC present.
Task 4: Coding Mechanistic Models Into EPA Water Treatment Plant Simulation Modeling Software
The purpose of Task 4 was to incorporate the mechanistic model predictive algorithms developed and refined during this project into a user-friendly tool that could be used to model DBP formation and chlorine decay within a water treatment plant. Therefore, the mechanistic-based algorithms were incorporated into the program code of the draft version of WTP Model Version 2.0, which was submitted to the EPA for review in May 2001. The resulting product, WTP Model Version 2.1, supports both mechanistic- and empirical-based algorithms. This section briefly documents how the mechanistic-based algorithms were incorporated into WTP Model Version 2.1, and how the functionality of Version 2.1 differs from Version 2.0. A detailed description of WTP Model Version 2.0, and its associated algorithms, can be found in Water Treatment Plant Model Version 2.0 User’s Manual (submitted to the EPA by the Center for Drinking Water Optimization on May 18, 2001). A CD-ROM with Version 2.1 of the WTP Model was developed and attached to the final report.
Conclusions:
The project was successful in refining the framework and parameterization of mechanistic models for predicting chlorine decay and DBP formation. Previous DBP databases were integrated into a unified database used to assess regulatory impacts. The database of chlorine decay and DBP formation kinetics developed for this project will be useful to other researchers, because it was conducted with both treated and untreated waters under chlorination conditions (chlorine dose/residual, temperature, pH, and bromide) representative of full-scale WTP applications. The database also includes partial-factorial design elements that examined simultaneous effects of low pH and high bromide, and differ from previous orthogonal experimental designs that only investigated effects of single parameters. The number of NOM sites implicated in chlorine decay and DBP formation accounted for less than 5 percent of the total DOC concentration. Chlorine and DBP reactive sites were related to the characteristics of NOM, based upon advanced NOM profiling techniques, and provided tools to priori estimate critical mechanistic modeling parameters.
The mechanistic models were more accurate at predicting DBPs in source specific waters, compared against power-function empirical central-tendency models that were somewhat more accurate in predictions for waters from different geographic sources. However, the mechanistic model is flexible and the coded version of the EPA WTP Model Version 2.1 can easily be modified to include new DBPs or chloramines reactions, once the rate laws and constants are available. The research conducted under this project provides tools to improve the understanding of DBP formation mechanisms, effective water treatment processes, and approaches for controlling DBPs in the future.
References:
Harrington GW, Chowdhury ZK, Owen DM. Developing a computer model to simulate DBP formation during water treatment. Journal of the American Water Works Association 1992;84(11):78.
McClellan JN. Modeling chlorine decay and chlorination by-product formation in water treatment and distribution. Ph.D. Dissertation, Department of Civil and Environmental Engineering, University of Massachusetts, Amherst, MA, 2000.
Reckhow DA, Singer PC. Mechanisms of organic halide formation during fulvic acid chlorination and implications with respect to pre-ozonation. In: Jolley, et al. eds. Water Chlorination, Chemistry, Environmental Impact and Health Effects. Chelsea, MI: Lewis Publishing, Volume 5:1229-1257, 1985.
Expected Results:
A numerical computer model and software will be developed that can predict involving (i) inorganic ions, (ii) disinfectant consumption, (iii) THM4 and HAA9 formation, and (iv) THM4 and HAA9 stability. The model will be robust and flexible, making it easy to incorporate additional reactions for other disinfectants (e.g., chloramines) or DBPs in the future.Journal Articles:
No journal articles submitted with this report: View all 23 publications for this projectSupplemental Keywords:
drinking water treatment, environmental chemistry, oxidation,, RFA, Scientific Discipline, Water, Applied Math & Statistics, Environmental Chemistry, Mathematics, Analytical Chemistry, Drinking Water, monitoring, chlorine decay, oxidation, unified database, chemical byproducts, disinfection byproducts (DPBs), treatment, chlorine-based disinfection, chloramines, DBP risk management, water quality, drinking water contaminants, drinking water system, mechanistic-based modelsRelevant Websites:
http://enpub.fulton.asu.edu/pwest/USEPA-CL2-DBP-2002.htm Exit
Progress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.