Background
Climate change poses significant risks to both surface and groundwater quality, which in turn affects the design and operation of drinking water treatment plants Increased precipitation can lead to higher turbidity, elevated organic matter, and increased bacteria, viruses, and pesticide levels in water bodies Conversely, droughts can cause harmful algal blooms and further elevate concentrations of organic matter and bacteria As source water quality deteriorates due to these climate impacts, investments in advanced water treatment processes will be necessary Additionally, the combination of climate change and urban population growth will adversely affect existing and future drinking water treatment infrastructure, raising concerns among water utilities regarding future water demands and the quality of finished drinking water.
The USEPA has initiated a national assessment to evaluate the sustainability and adaptability of water infrastructure in response to climate and socioeconomic changes through its Water Resources Adaptation Program (WRAP) As part of this initiative, methodologies were created to adapt drinking water treatment plant operations to various climate change scenarios To analyze the potential impacts on drinking water treatment, a WTP-ccam platform was developed, extending the existing USEPA Water Treatment Plant (WTP) Model This model employs empirical correlations to predict the effectiveness of natural organic matter removal, disinfection, and disinfection by-product (DBP) formation across diverse drinking water treatment configurations, while effectively simulating changes in finished water quality due to chemical additions and treatment processes.
The WTP-ccam platform, an enhanced version of the WTP model, utilizes Monte Carlo analysis to assess the effects of climate change-induced uncertainties in raw water quality Key features of the platform include the preservation of joint correlations among influent water quality parameters through a seasonal multivariate model, customization of processing unit parameters for better modeling accuracy, and the capability to adapt treatment processes while estimating related costs A case study conducted at the GCWW’s Richard Miller water treatment plant serves to validate and demonstrate the application of the WTP-ccam platform, detailing its validation, data preparation, and practical implementation.
Report Organization
To illustrate the procedure of application of the WTP-ccam platform, this report contains the following chapters in addition to this introductory chapter:
Chapter 2 offers an overview of the Richard Miller treatment plant operated by GCWW, analyzing its performance through field measurements Additionally, it outlines the validation process for the WTP-ccam model specifically within the Miller plant.
Chapter 3 focuses on establishing the framework for Monte Carlo simulation and preparing the necessary input data to create climate change scenarios It also illustrates the process of constructing the Monte Carlo simulation using the gathered data, as well as evaluating the effects of climate change on the operational efficiency of the Miller plant's water treatment system.
Chapter 4 describes adaptation and associated cost of the Miller plant to climate change scenarios.
Chapter 5 discusses the implementation of algorithms for estimating parameters of the logistic model specific to the granular activated carbon (GAC) unit process It details the preparation of total organic carbon (TOC) breakthrough data and the calibration of the GAC logistic model, utilizing field measurements collected from the Miller treatment plant.
2 Background of the Miller Plant and WTP-ccam Validation
Background
GCWW delivers an average of 5.26 m³/s (120 MGD) of water through an extensive network of 5,100 km of water mains, serving approximately 235,000 residential and commercial customers Established in 1907, the Richard Miller Plant processes surface water from the Ohio River, supplying 88% of the drinking water for GCWW's clientele, with a maximum summer capacity reaching 9.65 m³/s (220 MGD).
The Miller plant utilizes a comprehensive water treatment process that includes coagulation, sedimentation, and biologically active rapid sand filtration, followed by granular activated carbon (GAC) processing Spent GAC is reactivated in two large on-site furnaces, and after chlorination disinfection, the treated water is stored in a clearwell for distribution Key design and operational parameters of the Miller plant are detailed in Table 2.1, which includes the T10 value representing the hydraulic retention time necessary for a step dose conservative tracer test to achieve 10 percent of the inflow tracer concentration in the effluent.
Fig 2.1 Unit Process Treatment Train for the GCWW Miller water treatment plant.
Table 2.1 Miller WTP Unit Process Design Parameters
(Data source: USEPA ICR database).
GCWW Water Treatment Plant Operation
Table 2.2 presents key statistics for the Miller plant's GAC unit operation, detailing influent and blended effluent TOC concentrations, the number of active GAC contactors, plant inflows, and empty bed contact time (EBCT) over a 76-month period from January 2004 to April 2010 Influent TOC concentrations exhibited an annual cycle, with seasonal variations ranging from a low of 1.01 mg/L on March 24, 2004, to a high of 2.76 mg/L on September 22, 2004 Meanwhile, blended effluent TOC levels fluctuated between 0.26 mg/L on July 13, 2005, and 1.44 mg/L.
L (November 1, 2006), well below the compliance standard of 2.00 mg/L The plant inflow ranged from 3.26 m 3 /s (74 MGD on December 27, 2006) to 7.61 m 3 /s (174 MGD on September
The Greater Cincinnati Water Works (GCWW) operates between 6 to 11 Granular Activated Carbon (GAC) contactors at an average flow rate of 5.26 m³/s (120 MGD) to effectively manage water treatment Their operational strategy involves strategically bringing GAC contactors online and offline to achieve multiple objectives, including the reduction of Total Trihalomethanes (TTHM), optimizing water production, coordinating furnace operation schedules, managing GAC storage, and addressing pesticide runoff in the spring This careful management ensures that the monthly average Empty Bed Contact Time (EBCT) remains around 17 minutes, although daily and hourly EBCT can fluctuate significantly.
Table 2.2 Statistics of full-scale field measurements.
Field Measurements (units) Average Standard
Coefficient of Variation Minimum Maximum Sample
Blended effluent TOC conc, (mg/L) 0.85 0.26 0.31 0.26 1.44 289 a
Number of active GAC contactors 9 1.22 0.14 6 11 289 a
Note: a Sample period covers 2296 days, from Jan 7, 2004 to April 21, 2010 (one sample every 7.94 days) b Sample period covers 2184 days, from Jan 7, 2004 to Dec 30, 2009 (one sample every 7.83 days)
The temporal analysis of influent and blended effluent Total Organic Carbon (TOC) concentrations at the GAC unit over a 76-month period reveals seasonal fluctuations, with higher influent TOC levels observed in the latter half of the year Despite this, the relationship between influent and blended effluent TOC concentrations is weak (R = 0.08) In contrast, the number of active GAC contactors shows a strong positive correlation with plant inflow (R = 0.75) and mass inflow (R = 0.65), reflecting increased water demands during warmer months Consequently, more GAC contactors are utilized in summer than in winter Furthermore, a significant negative correlation exists between the number of active GAC contactors and blended effluent TOC concentrations (R = -0.69), suggesting that the quality of finished water is primarily influenced by the operational status of GAC contactors rather than the influent TOC levels.
Figure 2.5 illustrates the relationship between EBCT and plant inflow, highlighting the pronounced seasonality of the inflow rate In contrast, the average EBCT across the active GAC contactors remains relatively stable This stability is maintained by adjusting the number of active GAC contactors to align with GCWW’s operational objectives, effectively countering the seasonal fluctuations in inflow, as demonstrated in Figure 2.3.
Influent Blended effluent Compliance standard
T O C c on ce nt ra tio n, m g/ L
Fig 2.2 Time series of influent and blended effluent TOC.
Active contactors Water Inflow Mass Inflow
N u m b e r o f a ct iv e c o n ta ct o rs W a te r in flo w , cu m /s o r M a ss in flo w , g /s
Fig 2.3 Time series of plant inflow, mass inflow and active number of contactors.
Active contactors Blended effluent Compliance standard
N um be r of a ct iv e co nt ac to rs T O C c on ce nt ra tio n, m g/ L
Fig 2.4 Time series of active contactors and blended effluent TOC concentration.
E B C T , m in W at er in flo w , cu m /s
Fig 2.5 Time series of plant inflow and EBCT.
WTP-ccam Validation at GCWW’s Treatment Plant
2.3.1 Data Collection for GCWW’s Treatment Plant
Prior to its application in climate change studies, the WTP-ccam platform underwent validation against field measurements at the Miller treatment plant This validation utilized input data sourced from the ICR database specific to the Greater Cincinnati Water Works’ Richard Miller treatment plant, along with additional operational information, including the GAC reactivation period, provided by GCWW.
The ICR database was established to gather essential information on water quality, treatment, and occurrence for the development of regulations under the Safe Drinking Water Act (Obolensky and Singer, 2005; Obolensky et al., 2007) It includes comprehensive data on the design, treatment processes, and operations of large public water systems in the United States that serve populations of 100,000 or more, collected over 18 monthly monitoring periods from July 1997 to December 1998 Additionally, the database offers water quality parameter values from various sampling locations throughout the treatment process and in the water distribution system.
DRAFT system These measurements were compared against the corresponding simulation results from the WTP-ccam platform
The validation data were derived from three specific sample periods: April 1998, July 1998, and October 1998, with inflow rates and chemical feed concentrations detailed in Table 2.3 The influent water quality parameters and their comparison with simulated results are presented in Table 2.4 According to the GCWW staff, the GAC reactivation period is crucial, set at 8 months during winter and spring, and 4 months for summer and early autumn.
Table 2.3 Inflow and chemical feed levels for the Miller WTP
Note: RM-rapid mixing; COAG-coagulation basin; CLR-clearwell;
(Data source: USEPA ICR database).
2.3.2 Application of the Collected Data for WTP-ccam Validation
WTP-ccam can be utilized for both Monte Carlo simulations and one-time runs, with the latter being ideal for single event analysis and model validation when specific inputs are known The initial step involves constructing the processing train for the Miller treatment plant within the WTP-ccam platform, as illustrated in Figure 2.6, which highlights that chemical additions are treated as unit processes requiring specific input specifications Following this, inputs must be provided for each unit process, utilizing data from Tables 2.1, 2.3, and 2.4, as shown in Figures 2.7 to 2.9, which exemplify inputs for raw water quality, flocculation, and lime addition for sample period 10 Once all inputs are configured for each unit process, users can execute a validation simulation by selecting the one-time run option from the WTP-ccam platform menu For detailed procedures on building the processing train and setting up model inputs, refer to the WTP-ccam User Manual (USEPA, 2010).
Table 2.4 compares results from the WTP-ccam simulation with field data from the ICR database, showing reasonable agreements for key water quality parameters such as pH, alkalinity, total hardness, TOC, free chlorine residual, and TTHMs While TOC results were consistent for sampling periods 10 and 16, the simulated TOC was higher than the sampled TOC in period 13 The ICR data indicate that UVA removal in the Miller plant occurred through coagulation and GAC, whereas GAC predominantly removed UVA in the WTP-ccam The simulated UVA matched well with the finished water for sampling period 10, and there was excellent agreement between simulated and sampled chlorine residuals Additionally, reasonable alignment was noted between simulated and sampled TTHMs across finished water, average tap (1 day residence), and end of system (3 days residence) samples.
DRAFT agreement with the ICR data Therefore, it was concluded that the WTP-ccam platform provides a reasonable replication for the operation of the Miller plant for surface water treatment
Table 2.4 Comparison of sampled and modeled water quality results
Water quality parameter Sampling period Data type Influent Coag.
Basin Filtration GAC Finished water AVG1* AVG3**
Free chlorine 13 Sampled 1.2 0.8 0.7 residual Modeled 0.0 0.0 0.0 0.0 1.2 0.7 0.5
Note: *AVG1 refers to average retention time 1 day.
**AVG3 refers to the maximum retention time, 3 days.
Fig 2.6 Simulation processing train for the Miller plant.
Fig 2.7 Influent water inputs for sample period 10.
Fig 2.8 Input window for flocculation basin.
Fig 2.9 Input window for lime addition.
3 Application of WTP-ccam in Climate Change Studies
Monte Carlo Simulation
Monte Carlo methods are essential for modeling uncertain phenomena, such as raw water qualities affected by climate change By utilizing random sampling from probability distribution functions, Monte Carlo simulations generate numerous potential outcomes, offering probabilities for various scenarios rather than just a limited set of results This capability represents a significant advancement in the updated Water Treatment Plant (WTP) model The simulation is governed by three key options: Quarterly Running Average, Preserving Correlation, and Contamination Control.
The Quarterly Running Average option is specifically designed to regulate Total Organic Carbon (TOC) levels in compliance with the USEPA disinfectant/disinfection byproduct (D/DBP) rule, which mandates that treated water TOC levels from surface sources must not exceed 2.0 mg/L, calculated quarterly as a running annual average The WTP-ccam framework incorporates four seasons to represent the four quarters of the year, influencing both the raw water quality inputs—encompassing statistics and correlations—and the simulation procedures necessary for achieving the quarterly running average.
The preservation of correlation among raw water quality parameters is crucial for accurately simulating stochastic variables When cross-correlation exists, the concentrations of related reactants can fluctuate together, significantly impacting disinfection byproduct (DBP) formation during water treatment and distribution To address this, a multivariate seasonal autoregressive model of order one, developed by Bras and Rodriguez-Iturbe in 1984, was implemented in WTP-ccam This model effectively maintains seasonal means, variances, and cross-correlations among all water quality parameters, as well as lag-one correlations between adjacent seasons and parameters, ensuring a comprehensive understanding of water quality dynamics.
The contamination control option aims to adjust the design and operation of the processing train in response to simulated non-compliance scenarios For instance, upon detecting a Total Organic Carbon (TOC) violation, the WTP-ccam program enhances operations by increasing the frequency of Granular Activated Carbon (GAC) regeneration to ensure TOC levels remain within acceptable limits Key inputs for this approach include controlled contaminants, regulatory standards, safety margins, and the specific unit processes that need to be managed.
In Monte Carlo simulations, the number of runs is crucial for achieving reliable results A numerical analysis of the Miller plant indicated that 5000 runs produced stable mean, standard deviation, and skewness when examining the running average of TOC in the finished water Consequently, 5000 runs were utilized for GCWW’s Miller plant Normal probability plots for the logarithms of pH and TOC from the ICR database for the Ohio River demonstrate that lognormal distributions effectively describe the probability distribution of these parameters, despite a slight bias in the lower 5% probability for TOC Thus, log-normal distributions were applied to the raw water quality parameters.
Pe rc en ta ge
Pe rc en ta ge
Fig 3.1 Normal probability plots for logarithm of pH and TOC.
Setting up Baseline Raw Water Quality
Climate change significantly affects water treatment infrastructure by altering raw water qualities, which are crucial for generating stochastic influent water quality parameters using the Monte Carlo method To evaluate this impact on the Greater Cincinnati Water Works (GCWW) infrastructure, a raw water quality baseline was established to reflect current conditions at the Miller plant, alongside a future scenario derived from this baseline Initial findings indicated that Total Organic Carbon (TOC) is the primary water quality parameter influencing the levels of regulated disinfection byproducts (DBPs) such as Total Trihalomethanes (TTHM) and Haloacetic Acids (HAA5) in both finished water and the distribution system Given that TOC compliance requires a quarterly running annual average, the raw water quality baseline was developed for each season: spring (March to May), summer (June to August), autumn (September to November), and winter (December to February).
The seasonal baseline values for average and standard deviation of raw water quality parameters were sourced from the ICR database, covering the period from July 1997 to December 1998 This data pertains to six drinking water treatment plants that utilize the Ohio River as their source, including the Neville Island plant in Pittsburgh, PA, the Richard Miller plant in Cincinnati, OH, and the Evansville plant.
IN, Be Payne plant and Crescent Hill Filter plant in Louisville, KY, and Fort Thomas plant, KY. Figure 3.2 shows the location of the six plants
Figure 3.2 Location of WTPs on Ohio River.
2) Crescent Hill Filter WTP, Kentucky
The seasonal average water temperature and inflow to the Miller plant were treated as deterministic for each season Table 3.1 displays the synthesized raw water qualities, including baseline averages and standard deviations The design parameters of the Miller plant, as detailed in Table 2.1, were maintained throughout the analysis Additionally, the chemical feeds applied the same doses as indicated in sample period 10 of Table 2.3 The baseline for the GAC reactivation period is based on an average calculation.
Table 3.1 Baseline raw water inputs for the Miller water treatment plant in 1998
Note: 0 is baseline value and 0 is standard deviation; refers to not applicable.
Projecting Raw Water Quality in 2050
Skjelkvale et al (2005) studied the regional trend of surface water chemistry for 12 geographical regions in Europe and North America from 1990 to 2001 as results of acidification.
The Appalachian Plateau, one of the 12 regions, is situated in the upper reaches of the Ohio River Research trends from 1998 to 2050 indicate that alkalinity is increasing at a rate of +0.036 mg/L per year, while total hardness is decreasing by -0.22 mg/L per year, and dissolved organic carbon (DOC) is rising by +0.03 mg/L per year Given that dissolved organic carbon is typically the primary component of total organic carbon (TOC), it is assumed that TOC will follow a similar trend to that of DOC.
Ammonia primarily originates from the decomposition of plant and animal materials, sewage, and industrial waste A study by Whitehead et al (2006) utilized dynamic modeling to assess the effects of climate change on ammonia levels in the River Kennet, UK, from 1961 to 2100 Their findings suggest an estimated 25% increase in ammonia concentrations from 1998 to 2050, which may also be relevant to the Miller plant.
Bromides are naturally found in both surface and groundwater, with elevated levels often associated with saline intrusion According to Cromwell et al (2007), climate change-induced sea level rise is expected to increase bromide concentrations in coastal areas However, there is currently no evidence suggesting that bromide levels in inland regions will change due to climate change This observation is projected to apply to the Miller plant by 2050 Additionally, it is assumed that other water quality parameters, including pH, turbidity, calcium hardness, and UVA, will remain consistent by 2050.
Cromwell et al (2007) project that surface water temperatures will rise between 1.1 and 6.6ºC due to greenhouse gas emissions from 1990 to 2100 By 2050, the average water temperature is expected to be 2ºC above baseline levels across all seasons.
Li et al (2009) emphasize the importance of evaluating the effects of climate-induced water quality on the performance of water treatment facilities, while excluding the influence of population growth on inflow rates to the Miller plant, which are assumed to remain consistent with baseline levels It is also posited that the coefficients of variation for all water quality parameters in 2050 will mirror those of the baseline data, as indicated in Table 3.1 Consequently, the projected raw water inputs for 2050 are outlined in Table 3.2 Furthermore, the design and operational parameters of the Miller plant for future scenarios are initially maintained at the same levels as those established for the baseline.
Table 3.2 Raw water inputs for the Miller WTP in 2050
Note: 1 is average and 1is standard deviation in 2050; refers to not applicable.
Building up Monte Carlo Simulation with WTP-ccam
The Monte Carlo simulation, illustrated in Figure 3.3, incorporates key findings from sections 3.1 to 3.3, ensuring that (1) cross correlations among raw water quality parameters are maintained, (2) Total Organic Carbon (TOC) is identified as the main contaminant requiring control in the GAC unit process when it exceeds compliance standards, and (3) a quarterly running average for TOC is calculated Furthermore, all raw water quality parameters are modeled using a log-normal distribution.
For the control parameters, number of runs is 5000 as introduced in Section 3.1; seed for random number generator is arbitrarily set to an integer such as 168; TOC compliance is 2.0 mg/
The margin of safety, defined as the difference between the compliance standard and the actual controlled concentration, enhances reliability in adherence to regulations In this instance, the margin of safety is established at 0.05 mg/L.
To manually input influent water quality statistics, users must select the manual input button, which triggers a series of four dialogue windows corresponding to each season, as the Quarterly Running Average option is enabled As depicted in Figure 3.4, the manual input window for the baseline in spring at the Miller treatment plant is shown, referencing data from Table 3.1.
Fig 3.3 Setting up Monte Carlo simulation at the Miller plant.
Fig 3.4 Manual input window for influent water quality statistics.
The "Preserve Correlation" option in WTP-ccam necessitates a raw water data file to generate a correlation matrix for multivariate analysis Due to the absence of seasonality in the correlation matrix, a comprehensive raw water data file was supplied to create a universal correlation matrix that encompasses all four seasons The data file format, illustrated in Figure 3.5, includes 11 columns: pH, alkalinity, turbidity, calcium hardness, total hardness, TOC, UVA, bromide, ammonia, temperature, and inflow rate For further details on the Monte Carlo analysis settings, refer to Chapter 3 of the WTP-ccam User Manual.
Fig 3.5 Example format of influent water quality data file.
Impact of Climate Change on Performance of the Miller Plant
The analysis of the Miller treatment plant's performance, utilizing results from a successful Monte Carlo simulation run of WTP-ccam, reveals significant impacts of climate change A comparison of the cumulative distribution functions (CDF) for Total Organic Carbon (TOC) and Total Trihalomethanes (TTHM) between baseline and future climate scenarios highlights these effects Under current conditions, the Miller plant consistently meets the TOC compliance threshold of 2.0 mg/L However, projections indicate a 55% risk of non-compliance with TOC standards in a climate change scenario, primarily driven by anticipated increases in TOC concentrations in the raw water supply.
Under current conditions, there is no risk of exceeding the TTHM maximum contaminant level (MCL) of 80 μg/L; however, a slight risk of 0.2% exists under future scenarios Should the TTHM MCL become stricter, the likelihood of violations increases significantly For instance, if the MCL were to lower to 60 μg/L or 40 μg/L, the risk of non-compliance would rise to 4% and 36%, respectively Thus, while the Miller plant's existing infrastructure meets current treatment standards, it may face compliance challenges in the future.
Fig 3.6 Performance of Miller plant between baseline (1998) and future (2050) scenarios for
(a) TOC running average at finished water;
(b) TTHM after 3 days residence in distribution system
(Results are based on 5000 Monte Carlo simulations).
4 Adaptation of Water Treatment to Climate Change
Adaptation Engineering
Adaptation involves essential modifications to the design or operation of the existing water treatment system in response to simulated non-compliance events Given the significant risk of total organic carbon (TOC) compliance violations at the Miller plant due to anticipated climate change scenarios, it is wise to investigate potential design and operational adjustments to reduce the likelihood of such non-compliance occurrences.
Table 2.4 illustrates that the reduction of Total Organic Carbon (TOC) during the processing train at the Miller plant is primarily influenced by Granular Activated Carbon (GAC) processing To enhance TOC removal, modifications to the operation of the GAC unit were explored, as recommended by the USEPA.
(2005), TOC removal by parallel GAC contactor system at a drinking water plant is approximated by, ln1 1 d t R e f f in R
The concentrations of influent and effluent total organic carbon (TOC) at the granular activated carbon (GAC) unit are represented as TOC eff and TOC in, respectively The GAC reactivation period is denoted as t R, measured in days, while the empirical coefficients a, b, and d are defined for the analysis.
0.0000058 2 0.000111 0.00125 0.0001444 2 0.005486 0.06005 d TOC pH in EBCT EBCT EBCT EBCT (4.4)
Herein, TOC in is the inflow TOC concentration to the GAC unit in mg/L and EBCT is empty bed contact time in minutes.
The effluent Total Organic Carbon (TOC) from the Granular Activated Carbon (GAC) unit is influenced by several factors, including influent TOC, pH levels, GAC reactivation period, and Empty Bed Contact Time (EBCT) during processing Among these, minimizing the GAC reactivation period is the most effective strategy for enhancing TOC removal This reactivation period can be effectively managed within water treatment operations To address instances of TOC noncompliance, it is essential to determine a new reactivation period that ensures the running annual average of TOC in finished water remains below the compliance threshold, accounting for the prescribed safety margin The standard reactivation period for the Miller plant is set at 180 days, reflecting the average runtime of GAC.
The GCWW’s Miller plant faces a significant risk of TOC compliance violations, with a potential occurrence rate of 55% (2,765 noncompliance events out of 5,000 runs) under future climate change scenarios, as outlined in Section 3.5 By implementing adaptations to the GAC reactivation period in response to these TOC noncompliance events, new reactivation periods can be identified through WTP-ccam simulations to achieve the necessary TOC control concentrations As depicted in Figure 4.1, the cumulative distribution function (CDF) indicates that the minimum GAC reactivation period required is 68 days when the safety margin is set at 0.02 mg/L.
A higher safety margin, such as a DRAFT margin of 62 days for a 0.20 mg/L threshold, results in a reduced minimum GAC reactivation period However, it’s important to note that increasing the safety margin can significantly raise costs Figure 4.1 serves as a valuable tool for establishing an appropriate operational GAC reactivation period, especially when some tolerance for risk violations is acceptable in practice.
Permitting a 10% risk of noncompliance results in an 18% probability for noncompliant samples The x-axis points of the cumulative distribution function (CDF) curves that correspond to the 10% CDF indicate the minimum reactivation period needed Consequently, the required minimum reactivation period increases from 68 days to 109 days for a safety margin of 0.02 mg/L, and from 62 days to 95 days for a safety margin of 0.2 mg/L when allowing for a 10% risk of noncompliance.
Safety margin: 0.02mg/L Safety margin: 0.20mg/L
Fig 4.1 CDF of new GAC reactivation period for noncompliance events under the climate change (future) scenario
Cost Analysis for Adaptation Engineering
WTP-ccam conducts an economic analysis to evaluate the costs linked to the adaptation of GAC processing operations in water treatment, providing a framework to assess the impacts of climate change The total costs associated with GAC processing include four primary components: the initial GAC cost, annual GAC make-up cost, GAC contactor cost, and GAC reactivation cost The initial GAC cost represents a one-time expense calculated based on the total volume of contactors, along with the density and unit cost of new GAC The annual GAC make-up cost accounts for yearly GAC loss during reactivation, determined by the GAC loss rate, reactivation rate, and unit cost Additionally, the GAC contactor and reactivation costs can be estimated using cost models developed by Adams and Clark (1988).
In the context of cost analysis, 'y' represents the capital, operational, or maintenance expenses The variable 'USRT' pertains to process design or operational factors, typically denoting the total surface area of GAC contactors or the effective volume of the GAC unit, which are critical for calculating capital costs Additionally, parameters a, b, c, and d play a role in this analysis.
The empirical parameters for the cost functions of GAC contactors are derived from nonlinear regression analysis, with z values set at 0 or 1 to accommodate various USRT values These model parameters, based on 1983 costs, are detailed in Adams and Clark (1988), and all costs have been adjusted to 2009 currency using the Producers Price Index (US BLS, 2008) for consistent comparison Contractor costs are further divided into categories including capital costs, process energy, building energy, maintenance materials, and operational and maintenance (O&M) labor Detailed computational parameters for GAC contactors and GAC reactivation costs are presented in Tables 4.1 and 4.2.
Table 4.1 Parameters for GAC contactor cost
Type of Cost Capital Process energy
USRT volume area area area area a 93700 0 15150 540 1160 b 1999.1 12 350 23.6 0.3 c 0.712 1 0.916 0.753 1.068 d 0.958 1 1 1 1.152 z 1 1 1 1 1
(in 1983) Ratio of 2009 to1983 cost 2009ENR/1983ENR=
Table 4.2 Parameters for GAC reactivation cost
Type of Cost Capital Process energy Building energy
USRT area area area area area area a 144000 354600 12250 0 2920 648400 b 198300.4 6387 312.1 4456.6 282 287714.9 c 0.434 0.755 0.649 0.401 0.7 0.899 d 1 1 1 1 1 1 z 1 1 1 1 1 1
(in 1983) Ratio of 2009 to1983 cost 2009ENR/1983ENR
The Miller plant features 12 down flow gravity contactors and two multi-hearth furnaces dedicated to onsite reactivation Each contactor has a volume of 595 m³ and a surface area of 181 m² The system experiences an overall granular activated carbon (GAC) loss rate of 7-8%, with a carbon loading rate of 482 kg/day per square meter of hearth area during the GAC reactivation process.
A capital recovery analysis with a 20-year return period and a 5% interest rate allows for the development of a cost curve that illustrates how the total annual cost of the Granular Activated Carbon (GAC) system changes with varying reactivation periods As depicted in Figure 4.2 for GCWW’s Miller plant, the annual cost of the GAC system decreases as the reactivation period increases Notably, when the reactivation period is less than 90 days, the annual cost rises sharply as the reactivation period shortens The WTP-ccam utilizes this cost curve to estimate adaptation costs through interpolation based on the GAC reactivation period.
A nn ua l c os t, m il li on $
Fig 4.2 Working chart for GAC unit annual cost.
The net annual cost, defined as the difference between the calculated annual cost and the base annual cost of $13.6 million at the Miller plant, serves as a key metric for cost analysis Using Figures 4.1 and 4.2, the cumulative distribution function (CDF) of net annual cost can be derived, as illustrated in Figure 4.3 For complete control of total organic carbon (TOC) violations, the net annual cost reaches up to $7.0 million with a 0.02 mg/L safety margin and $7.8 million with a 0.20 mg/L safety margin However, if the performance criterion permits a 10% risk of TOC violations, the net annual cost decreases to $3.4 million for a 0.02 mg/L safety margin and $4.4 million for a 0.20 mg/L safety margin.
Safety margin: 0.02mg/L Safety margin: 0.20mg/L
Fig 4.3 CDF for net annual adaptation cost for the noncompliance events under the climate change (future) scenario
The reactivation period of adapted granular activated carbon (GAC) is influenced by total organic carbon (TOC) concentration, allowing for a correlation between net annual costs and TOC levels To facilitate comparison, TOC concentration is represented as an increment above the compliance criteria, highlighting the difference between actual TOC levels and regulatory standards.
DRAFT reveals that the net annual cost in GAC processing increases linearly with increasing TOC increment over the compliance, given by,
Y X (4.6) where, Y is net annual cost in million US dollar and X is the TOC increment R 2 0.9985 indicates highly linear relationship for the two variables.
TOC increment over compliance, mg/L
Fig 4.4 Net annual costs versus TOC increment.
Implication for Engineering Practice
The implications for engineering practice are based on several key assumptions: first, the average rates of raw water quality parameters are expected to remain consistent as outlined in Section 3.3 until 2100; second, the coefficient of variation for these parameters will maintain the baseline condition; and third, changes in raw water quality parameters will be confined within a defined range, specifically between a lower boundary of μ - 2σ and an upper boundary.
Figure 4.5 demonstrates the change and range of raw water TOC from 2000 to 2100. Similarly, change of other water quality parameters are established based on Section 3.3
Fig 4.5 Change and range of raw water TOC.
The WTP-ccam is utilized to assess the need for adjusting the GAC reactivation period to ensure compliance with Total Organic Carbon (TOC) levels in finished water, based on varying raw water quality boundaries If adjustments are necessary, the revised reactivation period and associated annual costs are calculated Analysis indicates that annual costs will remain stable if raw water quality does not result in TOC noncompliance However, a significant risk of TOC violation emerges as raw water quality declines, particularly when the cost curve based on average water quality exceeds the baseline cost By 2050, the projected annual cost is estimated at $14.0 million, surpassing the baseline cost of $13.6 million, aligning with a 55% risk of TOC violation identified in earlier analyses Consequently, strategic planning based on these findings is essential to mitigate TOC compliance violations.
Fig 4.6 Range of annual cost in coming years.
5 Customization of Logistic Model for GAC Unit Process
Overview
Granular Activated Carbon (GAC) treatment has been utilized since the early 1970s as an effective method for reducing organic contamination in water supplies Research, including studies by Roberts and Summers (1982), indicates that while GAC can significantly reduce Total Organic Carbon (TOC), complete removal is not achievable under typical water treatment conditions Initial TOC breakthrough occurs even with fresh GAC, suggesting that some influent TOC is resistant to removal Over time, the effluent TOC concentration increases, stabilizing at a steady state as the GAC becomes saturated with organics Notably, the effluent TOC is usually lower than the influent concentration, with this steady-state removal often attributed to biodegradation During the early operation phase, the effluent-to-influent TOC ratio, or "fraction remaining," typically ranges from 0.1 to 0.5, influenced by the organic composition and empty bed contact time (EBCT) For steady-state conditions, this fraction increases to between 0.6 and 0.9, with service times ranging from 3,000 to 14,000 bed volumes.
Various methods have been employed to quantify breakthrough behavior in GAC treatment performance, with pilot columns and the rapid small-scale column test (RSSCT) being the most prevalent (USEPA, 1996) Pilot columns mirror full-scale systems in terms of GAC media, column length, and influent water, providing reliable results, although they require significant operational time and costs (Speth, 1989) In contrast, the RSSCT method utilizes mass transfer models to create a smaller-scale column that maintains similitude with larger systems by carefully selecting GAC particle size, hydraulic loading, and empty bed contact time (Crittenden et al., 1991; Zachman and Summers, 2010) The primary advantage of RSSCT is its efficiency, typically requiring less than 15 percent of the time needed for full-scale tests, though it does not account for long-term biodegradation in the GAC media.
The WTP-ccam platform utilizes empirical correlations to forecast the central tendencies of NOM removal, disinfection, and DBP formation in treatment plants, primarily through multiple linear regression techniques These statistical models typically consist of independent variables and empirical constants, effectively delivering central tendency predictions However, they may lack the precision needed for specific utilities To enhance accuracy, WTP-ccam now offers the ability to customize empirical constants in regression equations by incorporating site-specific treatment study data.
In a system with multiple parallel GAC contactors operating on staggered reactivation cycles, simulations are conducted using Equations 4.1 for WTP-ccam This approach effectively integrates a logistic model when the number of contactors exceeds ten The parameters for the model are estimated accordingly.
Equation 4.2 to 4.4 Blending the treated effluent from multiple parallel GAC columns having staggered reactivation cycles leads to a near steady-state TOC output concentration (USEPA,
The analysis of the Miller treatment plant's operation revealed that the parallel GAC system was not operated with staggered reactivation cycles, as noted in Section 2.2 Additionally, seasonal variations were observed in plant inflows, influent TOC concentrations, and the number of contactors in use Consequently, the parameters from Equations 4.1 and 4.2-4.4 may not accurately reflect the Miller plant's operation To address this, the GAC simulation model and its parameters were calibrated using field operational data from GCWW’s Miller treatment plant (Li et al., 2010; Li et al., 2011).
Algorithm
Adsorption processes in water treatment typically follow well-established isotherm models, including Freundlich, Langmuir, and Brunauer-Emmett-Teller (BET) The logistic function effectively captures the characteristic 'S' shape of breakthrough curves, illustrating the relationship between total organic carbon (TOC) concentration and GAC runtime This mathematical representation closely aligns with the Langmuir isotherm To accurately model the experimental breakthrough behavior in a single GAC column, researchers Chowdhury et al (1996) and Summers et al (1998) utilized a modified dimensionless version of the logistic function, which is also implemented in the WTP-ccam platform.
The function f(t) represents the remaining TOC fraction, where t denotes the total continuous GAC runtime Model parameters a, b, and d can be experimentally determined through best-fit analysis of breakthrough data, rather than estimated using equations 4.2-4.4 Parameter a indicates the asymptotic steady-state value of f(t), while parameters a and b influence the intercept of the logistic curve Additionally, parameter d impacts the steepness of the curve and is proportional to the influent TOC concentration, as noted by USEPA (2005).
When GAC contactors of the same size operate in parallel with identical flow rates, the remaining blended TOC fraction can be determined by calculating the arithmetic average of the TOC breakthrough curves from each individual contactor, as demonstrated by Roberts and Summers (1982).
In a parallel system, the number of contactors is denoted by m, while f(t) represents the average total organic carbon (TOC) fraction remaining in the blended effluent from these m contactors Additionally, f(ti) indicates the TOC fraction remaining in the effluent from each individual contactor i at time t.
When Total Organic Carbon (TOC) breakthroughs occur, datasets from field measurements or Granular Activated Carbon (GAC) treatment studies are utilized The Water Treatment Plant - Continuous Carbon Adsorption Model (WTP-ccam) employs a modified Gauss-Newton method to estimate model parameters a, b, and d This is achieved by fitting a non-linear regression function through least squares optimization.
DRAFT square analysis based on Hartley (1961) Hartley’s method is an iterative nonlinear least squares algorithm with objective function,
In Equation 5.1, the function f(t, a, b, d) is defined, where a, b, and d are unknown parameters to be estimated The variables t_k and y_k represent the measured breakthrough data, indicating the GAC runtime and the fraction of TOC remaining, respectively Additionally, n denotes the number of field data points collected for analysis.
Each iteration of the method provides a vector of corrected model parameters,
j j -1 θ θ D j = (5.4) where, is a vector of parameters to be estimated, a b d
In this context, θ j -1 denotes the parameter vector from the prior iteration, which is initialized using standard values found in existing literature, while θ j signifies the updated parameter vector for the current iteration Additionally, D serves as a correction vector derived from the initial parameters, calculated as a solution to the Gauss-Newton equations.
D ; v is a value from 0 to 1 to minimize Q a b d , , during each iteration The Gauss-Newton equation is given by,
Where, A is the Gauss-Newton coefficient matrix, defined as,
A (note: f f t a b d ; , , ) (5.6) and R is a right-hand-side vector of Gauss-Newton equation, defined by,
The vector D is obtained by solving Equation 5.5 while v is estimated from the parabola through the points Q 0 , Q 0.5 , and Q 1 , given by (Hartley, 1961),
The values Q(ν = 0), Q(ν = 0.5), and Q(ν = 1) represent the Q parameters a, b, and d evaluated at different levels of ν in Equation 5.3 According to Equation 5.4, the final estimates for parameters a, b, and d are determined once the specified precision requirement for ΔQ(a, b, d) is achieved; if not, the iteration process is repeated until the criteria are met.
Calibration of the Logistic Model with GCWW Field Measurement
GCWW provided weekly values of influent and effluent TOC concentrations for each of the
12 full-scale GAC contactors Using the 76-month sampling period from January 2004 to April
In 2010, researchers identified 87 individual breakthrough datasets related to Total Organic Carbon (TOC) for GAC contactors These datasets were ultimately presented in a dimensionless format, illustrating the relationship between the TOC fraction remaining and the service runtime of the contactors.
The analysis of 87 datasets from full-scale contactors revealed that 46 were deemed valid for logistic model calibration, based on USEPA criteria for GAC pilot studies These criteria state that studies should end if the effluent TOC concentration reaches 70% of the average influent TOC on two consecutive sampling dates or if 50% TOC breakthrough occurs with minimal fluctuation over 60 days Valid datasets were derived from full-scale field measurements at the GCWW operation, with Figures 5.1a and 5.1b showcasing the TOC breakthrough curves for two of these datasets Additionally, two sets of RSSCT experimental data were collected from GCWW, detailing the TOC breakthrough behavior, including influent and effluent concentrations and GAC column runtimes, which were adjusted to reflect conditions at the GCWW Miller plant, as illustrated in Figures 5.1c and 5.1d.
The study analyzed 12 GAC contactors, each providing three or four valid TOC breakthrough datasets Using a nonlinear regression algorithm, these datasets were utilized to estimate the parameters of the logistic model for individual GAC contactors The parameters a, b, and d were averaged across all contactors, with results detailed in Table 5.1, which includes key information such as average influent TOC concentration, EBCT statistics, runtime statistics, the number of valid datasets, and operational time during a 76-month sampling period GAC contactor runtimes varied from 214 to 340 days, averaging 285 days, while the global average of the estimated parameters a, b, and d were found to be 0.782, 8.191, and 0.029 day -1, respectively.
Full-scale dataset 1 Logistic model
TO C fr ac tio n re m ai ni ng
Full-scale dataset 2 Logistic model
T O C fr a ct io n r e m a in in g
T O C fr a ct io n r e m a in in g
T O C fr a ct io n r e m a in in g
Fig 5.1 Logistic model fitted to full-scale field data and to RSSCT data.
The logistic model parameters derived from the RSSCT datasets, as presented in Table 5.2, indicate that for parameters a and d, the estimates align more closely with those from non-valid datasets than with valid full-scale datasets Notably, the RSSCT data produced the lowest estimates for parameter a across all breakthrough dataset types, which is critical as parameter a influences the asymptotic high plateau of the breakthrough curve As a result, a logistic model calibrated using RSSCT data may fail to accurately reflect the high extremes of effluent TOC recorded historically at the Miller plant.
Figure 5.2 illustrates the comparison of blended average TOC fraction remaining simulations for two logistic model versions against actual full-scale measurements over a 76-month monitoring period The system-wide model utilizes uniform global average parameters from 46 valid datasets, specifically a=0.782, b=8.191, and d=0.029 day -1, as detailed in Table 5.2 Conversely, the RSSCT-based model employs average parameter values derived from two RSSCT datasets, with a=0.624, b=7.477, and d=0.034 day -1, also presented in Table 5.2 The blended average TOC fraction was calculated using Equation 5.2.
The DRAFT process for the system-wide and RSSCT models adhered to the precise historical sequence and runtimes of active GAC contactors utilized at the Miller plant.
The system-wide logistic model demonstrates a strong correlation (R=0.97) with the performance history of full-scale field measurements at the Miller plant, reflecting its calibration with comprehensive datasets The model's residual sum of squares (RSS) is notably low at 0.58 Conversely, the RSSCT-based model struggles to accurately replicate the performance of the multi-GAC contactor bank, particularly when the total organic carbon (TOC) fraction remaining is elevated This discrepancy arises from an underestimation of the parameter 'a,' resulting in a systematic bias in the logistic model's predictions and a higher RSS value of 2.24.
Table 5.1 Characteristics of individual GAC contactors from 46 valid datasets (2004-2010)
Avg Min Max Avg Min Max (%)
Note: “Time in Use” is based on the full record of 87 datasets All other columns are based on the 46 valid datasets.
Table 5.2 Statistics of parameters for logistic model
Data Source Average Stan Dev Coef of Var Minimum Maximum
Non-valid Full-Scale Datasets ( n A)
Field measurements System-wide model RSSCT-based model
Fig 5.2 Comparison of fitting GAC models with field measurements.
The WTP-ccam platform, an enhanced version of the USEPA WTP model, is utilized at the Richard Miller water treatment plant of GCWW to showcase a method for evaluating the effects of climate change on drinking water treatment infrastructure.
Validation tests indicate that the WTP-ccam platform effectively replicates operations at the GCWW’s Miller water treatment plant, as evidenced by comparisons with field data from the USEPA’s ICR database.
A Monte Carlo analysis was conducted to assess the effects of climate change on raw water quality and its implications for water treatment infrastructure The findings indicate that while the current treatment system at the Miller plant satisfies Total Organic Carbon (TOC) compliance in baseline conditions, there is a significant risk of non-compliance as climate change progresses.
Adapting the operation of the Miller plant by reducing the GAC reactivation period significantly improves TOC removal The findings indicate that after these operational adjustments, the existing water treatment plant can achieve TOC compliance.
A cost model for the Miller plant was developed to connect annual expenses with the reactivation period of the GAC processing unit The findings indicate that achieving full compliance with TOC regulations necessitates a substantially higher net annual cost However, allowing for a minimal risk of violations can lead to a significant reduction in these annual costs.
As projected raw water quality declines in the coming years, the risk of violating Total Organic Carbon (TOC) compliance will rise Implementing adaptation strategies and conducting cost analyses related to climate change are essential for effective infrastructure planning and resilience.
The logistic model for Total Organic Carbon (TOC) breakthrough can be effectively calibrated using full-scale Granular Activated Carbon (GAC) contactor data through Hartley's nonlinear parameter estimation algorithm This calibration demonstrates that the logistic model can accurately replicate the historical performance of TOC removal at the Greater Cincinnati Water Works Miller treatment plant, covering a 76-month period marked by significant variations in flow rate and influent loading.
The estimates of the three logistic curve parameters are highly sensitive to the duration of the breakthrough dataset Notably, the logistic model derived from RSSCT breakthrough datasets fails to account for the elevated TOC extremes observed in the historical GAC operations at the Miller plant This inadequacy is primarily linked to significantly low estimates of the parameter "a." The reasons behind these unusually low parameter estimates from RSSCT experimental data remain unclear and warrant further investigation.
Adams, J Q and Clark, R M (1988) “Development of Cost Equations for GAC Treatment
Systems.” Journal of Environmental Engineering, 114 (3): 672-688.
Bras, R.L and Rodriguez-Iturbe, I (1984) “Random Functions and Hydrology.” Addison-
Clark, R.M (1987a) "Modeling TOC Removal by GAC: The General Logistic Function." J
Clark, R.M (1987b) “Evaluating the cost and performance of field-scale granular activated carbon systems.” Environmental Science and Technology 21 (6): 573-580
Crittenden, J.C., Reddy, P.S., Arora, H., Trynoski, J., Hand, D.W., Perram, D.L., and Summers,
R.S (1991) “Predicting GAC performance with rapid small-scale column tests.” J
Cromwell, J.E., Smith, J.B and Raucher, R.S (2007) “Implications of Climate Change for
Urban Water Utilities.” The Association of Metropolitan Water Agencies report http://www.amwa.net/galleries/climate-change/AMWA_Climate_Change_Paper_12.13.0 7.pdf
Futter, M N., D Butterfield, B J Cosby, P J Dillon, A J Wade, and P G Whitehead (2007)
“Modeling the mechanisms that control in-stream dissolved organic carbon dynamics in upland and forested catchments.” Water Resour Res., 43: 1-16.
Hartley, H.O (1961) The modified Gauss Newton method for the fitting of nonlinear regression functions by least squares Technometrics 3 (2): 269-280
Interlandi, S J., and Crockett C S (2003) “Recent water quality trends in the Schuylkill River,
Pennsylvania USA: a preliminary assessment of the relative influences of climate, river discharge and suburban development.” Water Resources, 37: 1737-1748.
Jacobs, K., Adams, D B., and Gleick, P (2001) “Chapter 14, Potential Consequences of
Climate Variability and Change for the Water Resources of the United States”, in
Climate Change Impacts On the United States –the Potential Consequences of Climate Variability and Change edited by Jerry Meilillo, Anthony Janetos and Thomas Karl,
Cambridge University Press, New York.
James, B R., Gwo, J., and Toran, L (1996) “Risk-cost decision framework for aquifer remediation design.” J Water Resour Plan Manage., 122 (6): 414–420
Khanal, N., Buchberger, S G and McKenna, S A (2006) “Distribution System Contamination
Events: Exposure, Influence, and Sensitivity.” J Water Resour Plan Manage., 132 (4): 283-292.
Kửhler, S.J., Buffam, I., Seibert, J., Bishop, K.H., Laudon, H (2009) “Dynamics of stream water TOC concentrations in a boreal headwater catchment: Controlling factors and implications for climate scenarios.” Journal of Hydrology, 373: 44–56
Li, Z., S.G Buchberger, R.M Clark, Y.J Yang (2011) Calibration of GAC Model in Drinking
Water Treatment ASCE congress, Palm Springs, California
Li, Z., S.G Buchberger, R.M Clark, Y.J Yang (2010) Customizing Climate Change Adaptation at Drinking Water Treatment Plants, in proceedings of 12th annual Water Distribution
Systems Analysis conference, September 12-15, 2010, Tucson, Arizona
Li, Z., S.G Buchberger, R.M Clark, Y.J Yang (2009) Assessing the impact of climate change on water treatment plant design and operation, in Proceedings of 10th International
Conference on Computing and Control for the Water Industry, CCWI2009 - Integrating Water Systems, Sheffield, UK, 1-3 September 2009, edited by Joby Boxall and Cedo
Obolensky, A., and Singer, P C (2005) “Halogen Substitiution Patterns among Disinfection
Byproducts in the Information Collection Rule Database.” Envi ron Sci Technol 39: 2719-2730
Obolensky, A., Singer, P S., and Shukairy H.(2007) “Information Collection Rule Data
Evaluation and Analysis to Support Impacts on Disinfection By-Product Formation.”
Journal of Environmental Engoneering (ASCE) 133(1): 53-63
Oulman, C.S (1980) “The logistic curve as a model for carbon bed design.” J AWWA 75 (1):
Roberts, P.V and Summers R.S (1982) “Granular Activated Carbon Performance for Organic
Salas, J D., Delleur, J W., Yevjevich, V and Lane, W L (1980) “Applied Modeling of
Hydrologic Time Series.” Water Resources Publications, Littleton, CO, 484 pages.
Skjelkvale, B.L., Stoddard, J.L., Jeffries, D.S., Torseth, K., Hogasen, T., Bowman, J., Mannio, J.,
Monteith, D.T (2005) “Regional scale evidence for improvements in surface water chemistry 1990–2001.” Environ Pollut 137: 165–176
Spear, R C., and Hornberger, G M (1980) “Eutrophication in Peel Inlet-II Identification of critical uncertainties via generalized sensitivity analysis.” Water Res., 14 (1): 43–49.
US BLS (Bureau of Labor Statistics) (2008) BLS Handbook of Methods, Chapter 14 Producer
Prices-Background http://www.bls.gov/opub/hom/homch14_a.htm.
USEPA (2010) “Multi-Scale Water Infrastructure Characterization and Adaptation Strategies to
Climate Change, Land Use and Socioeconomic Developments: A National Assessment.”
USEPA (2010) “WTP-ccam User Manual” Office of Research and Development, U.S
Environmental Protection Agency, Cincinnati, Ohio.
USEPA (2005) “Water Treatment Plant Model Version 2.2 User’s Manual.” Office of Ground
Water and Drinking Water, U.S Environmental Protection Agency, Cincinnati, Ohio.
U.S Environmental Protection Agency (2000) "ICR treatment study database, version 1.0."
Rep No EPA 815-C-00-003, Office of Water, Washington, D.C.
U.S Environmental Protection Agency (1996) "ICR manual for bench and pilot-scale treatment studies." Rep No EPA 814-B-96-003, Office of Water, Washington, D.C.