Experimental Measurements of Fluence Distribution in a UV Reactor

In this research, the actual fluence distribution of a UV reactor is measured through the use ... UV disinfection of secondary water supply: Online mo...
0 downloads 0 Views 244KB Size
Download PDF
Environ. Sci. Technol. 2005, 39, 8925-8930

Experimental Measurements of Fluence Distribution in a UV Reactor Using Fluorescent Microspheres ZUZANA BOHREROVA,† GIL BOHRER,† S. MOHAN MOHANRAJ,‡ J O E L D U C O S T E , § A N D K A R L G . L I N D E N * ,† Department of Civil and Environmental Engineering, Duke University, Durham, North Carolina 27708, PolyMicrospheres, Division of Vasmo, Inc., 4101 East 30th Street, Indianapolis, Indiana 46218, and Department of Civil, Construction and Environmental Engineering, North Carolina State University, Raleigh, North Carolina 27695

One concern with current techniques of UV reactor validation is that they provide only a measure of the mean UV fluence. In this research, the actual fluence distribution of a UV reactor is measured through the use of photochemically active fluorescent microspheres. Experimental tests were performed in a pilot-scale monochromatic UV 254 nm reactor operated at two flow rates. Analysis of the fluorescence intensity decay was performed using collimated beam experiments for determination of decay rate kinetics. A stochastic hierarchal process involving Bayesian statistics, and the Markov chain Monte Carlo integration technique was used to correlate the microsphere fluorescence intensity distribution to the UV fluence distribution. The experimental UV fluence distribution was compared with the fluence distribution predicted using a computational fluid dynamics model. The results showed that the fluorescent microspheres measured a wider distribution of UV fluences with a flow rate of 3 gpm than with 7.5 gpm. The principal differences between the modeled and the measured distribution were in the low UV fluences where the microspheres predicted lower fluence levels than the model. The use of microspheres is demonstrated as a novel technique for measurement of the fluence distribution in UV reactors. This technique has both fundamental and practical implications for reactor evaluation and testing and could improve confidence in the future use of mathematical models for UV reactor characterization. It also serves as a complement to biodosimetry testing by providing greater insights regarding reactor behavior and validation.

Introduction UV disinfection is increasingly used for wastewater and drinking water disinfection, despite the fact that a direct method for confirmation of the delivered UV fluence in a reactor is still under investigation. The current method for the full-scale validation of a UV reactor is biodosimetry (1), * Corresponding author phone: (919)660-5196; fax: (919)660-5219; e-mail: [email protected]. † Duke University. ‡ PolyMicrospheres. § North Carolina State University. 10.1021/es050034c CCC: $30.25 Published on Web 10/07/2005

 2005 American Chemical Society

resulting in the determination of a single reduction equivalent fluence (REF) for a specific reactor under defined conditions. The REF gives an indication of the average UV fluence in the reactor calculated from the inactivation rate of a surrogate microorganism spiked into the flow. Biodosimetry is a relatively expensive and time-consuming process (2), and the range of operating conditions that are possible to evaluate on-site is often limited (3). Although this method provides the mean fluence for the UV reactor, in reality, the microorganisms are exposed to a range of UV fluences while flowing through a reactor (4, 5). Many microorganisms display nonlinear response to the level of UV fluence, such as inactivation saturation and minimal sensitivity thresholds (6-8). These nonlinear effects can make the mean UV fluence a poor predictor for the actual efficacy of the reactor for different pathogenic microorganisms and emphasizes the need to predict and measure the full distribution of UV fluence delivered in a reactor. Mechanistic mathematical models for predicting fluence distribution in UV reactors have been developed (5, 9), but full-scale detailed testing and direct validation of the predicted fluence distribution has thus far been impossible. Biodosimetry is increasingly being performed at validation facilities off-site from a water utility, providing greater flexibility for testing reactors using a wider range of microbial and nonbiological surrogates. Microsphere-based nonbiological surrogates have been investigated as indicators for pathogen inactivation in ozone reactors (10, 11), combined chlorine and ozone exposure studies (12), and filtration assessments (13). Nonbiological surrogates could also help to predict the cumulative UV fluence or ozone dose each particle is exposed to while flowing through a reactor and in this way be used to monitor the full distribution of doses that a pathogen may experience. This was attempted in an ozone reactor (14) where the response of fluorescent microspheres was calibrated to predict the inactivation of Giardia cysts. Fluorescent microspheres were also used to obtain an indication of UV fluence in a reactor (15) but without a comparison to the conventional biodosimeter approach or modeling. Neither of the previous ozone or UV studies quantified the variability in the response of the nonbiological surrogates to the disinfecting agent. In this study, fluorescent microspheres were used to measure the UV fluence distribution in a flow-through reactor. We used Bayesian statistics to infer the distribution of UV fluence in a pilot-scale reactor, given the prior variability of microsphere fluorescence and the variability of their response to UV photobleaching. We validated the mean of this distribution by comparing it with inactivation results of two microorganisms and compared it to a predicted fluence distribution obtained from a mathematical fluid dynamics model specific to the pilot reactor. This technique could provide an advantage over traditional biological-based fluence measurement techniques by overcoming issues related to the different sensitivities of surrogate and target microorganisms and helping to measure the UV fluence distribution in the reactor. The use of fluorescent microspheres provides a new novel tool to validate mathematical predictions of fluence distributions and has the potential for applications in full-scale drinking water plants (12). Microspheres are spherical microparticles with similar size and density characteristics to microorganisms (1-5 µm) and thus would experience similar hydrodynamics and disinfectant exposures in the reactor as microorganisms (14). The fluorescence of these microspheres is linearly proportional to the dye content of the particles and to the concentration of the microspheres as a percent solids in the dispersion. For VOL. 39, NO. 22, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

8925

certain dyes, the fluorescence intensity decreases as a result of the reaction with photons of UV light (i.e., photobleaching). Microspheres incorporated with a commercial fluorescent dye that can be photobleached upon exposure to UV radiation were utilized in this study. A pilot-scale UV reactor housed with a low-pressure (LP) mercury vapor lamp emitting principally at 253.7 nm was employed to test and compare the distribution of UV fluence measured by the use of fluorescent microspheres flowing through the reactor to the UV fluence distribution calculated by a numerical model. Tests were performed under two flow conditions. Because the UV process is typically applied upstream of any chemical disinfectant addition, no other disinfectant was evaluated with these microspheres.

Materials and Methods Fluorescent Microspheres and Flow Cytometry. Fluorescent 14 polystyrene microspheres (catalog number PS1805-Fl14) were obtained commercially from PolyMicrospheres, a Division of Vasmo, Inc. (Indianapolis, IN). The microspheres were ∼1.6 µm in mean diameter, with excitation maximum at 340 nm and emission maximum at 380 nm. These microspheres were obtained at a concentration of 0.2% solids. The fluorescence of the microspheres was measured by flow cytometry, a powerful tool for identifying and tracking a wide range of molecules in samples. The flow cytometer used in this study was a BD FACSVantage that included Turbosort, three lasers, and six fluorescence detectors (Becton Dickinson Immunocytometry Systems, San Jose, CA), housed at the Duke University Comprehensive Cancer Center. The fluorescent microspheres were individually detected by a 15 mW argon, air-cooled laser at 488 nm wavelength, excited at 350 nm, and their shapes and diameters were determined. The fluorescence intensity emitted by each microsphere was detected by a FL1 detector after passing through a 405 nm band-pass filter (395-415 nm half peak width). Results were registered on an arbitrary relative scale of fluorescence intensity from 0 to 1023. Gating was performed in a microsphere dot plot using the characteristic forward (FSC) versus orthogonal light scatter (SSC) to eliminate background noise. FSC is related to the particle size, while SSC is related to the particles’ internal granularity and complexity. At least 10 000 microspheres were acquired per sample. The flow cytometer data were gated and extracted using software WinMDI, version 2.8 (16). Pilot Test Low-Pressure UV Reactor Settings. The test reactor contained one low-pressure (LP) Hg vapor UV lamp with output at 254 nm. The lamp sleeves were regularly cleaned to reduce unnecessary reversible fouling. The reactor setting for our experiment included a 400 gallon storage tank filled with deionized water, a 1 gallon spiking container for mixing microspheres into the main water flow, a static mixer and pump, a flow adjuster and controller, an influent and effluent collector, and a test LP UV reactor. The LP UV lamp was started at least 10 min prior to the experiment, and after stabilization of the flow at 3 gpm (11.36 Lpm) or 7.5 gpm (28.39 Lpm), the microspheres were spiked into the flow at a rate of 0.75 gpm (2.97 Lpm) to ensure concentration of at least 1 × 105 microspheres mL-1. Influent samples were collected for bench-scale quasi-collimated beam experiments (see below) and for controls. Four 30 mL effluent samples were collected for each condition tested. A control effluent, in which the UV lamp was switched off, was also collected. Bench-Scale Experiment. Bench-scale experiments were performed to develop the UV response kinetics for the microsphere fluorescence intensity decay. The influent to the test reactor (deionized water and microspheres spiked at a concentration of 1 × 105 microspheres mL-1) was exposed to a known fluence of LP UV radiation (253.7 nm) in a quasicollimated beam bench-scale apparatus. The experiments 8926

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 22, 2005

were performed in completely mixed batch reactors using quartz dishes with a 10 mL sample to ensure uniform and mixed conditions for all of the microspheres. The UV fluences used for the bench-scale experiment to calibrate the microsphere response as a function of UV fluence levels were 10, 20, 30, 40, 60, 80, 100, and 120 mJ cm-2. UV fluence (mJ cm-2) was calculated as the average 253.7 nm irradiance in the mixed sample multiplied by the exposure time. Incident UV irradiance (mW cm-2) was measured (International Light IL1700 SED240/W) at the surface of the liquid suspension. The average irradiance was determined using the incident irradiance, UV absorbance (Varian Corp., Cary 100 Biospectrophotometer, Walnut Creek, CA), and sample depth in an integrated form of the Beer-Lambert law using a Microsoft Excel spreadsheet according to Bolton and Linden (17). Computation of the UV Fluence Distribution from Microsphere Fluorescence Intensity Measurements. The Markov chain-Monte Carlo (MCMC) integration of the Bayes equation was used to determine the UV fluence probability distribution function (UV-PDF) in the reactor. MCMC is an iterative technique that was first developed in statistical physics and is based on the ability to randomly sample from a specific probability distribution over a long period of time. A random process is sampled and the outcome is compared with observed data. The sampling is then biased for the values that produced smaller error, defined as the difference between the modeled outcome and the observed, until the process eventually converges. The advantage of using the Bayes equation to describe the probabilities of a random process is that it includes not only the variability of the random process but also the uncertainty that is attributed to the model assumptions. The computed UV-PDF was based on the measurements of fluorescence intensities of the microspheres in the effluent and on measurements of microsphere fluorescence in the bench-scale quasi-collimated beam experiments. The microsphere experiment was represented as a stochastic hierarchical process where the MCMC integration was used to estimate the parameters of the UV-PDF that were observed by the microspheres. A statistical approach was necessary in analyzing the UVexposed microspheres since the bench-scale quasi-collimated beam experiments revealed that a distribution of fluorescence intensity was created for each applied fluence level, with some overlap between those distributions. Consequently, a single microsphere fluorescence observation can be attributed to a continuous range of corresponding exposure fluences. Consequently, the MCMC integration over the range of observed fluorescence distribution is an important step to determining the corresponding fluence level distribution. In this study, the experimental tests represent a hierarchical process with two independent levels, the bench-scale experiment to find the response curves of the microspheres to UV irradiation and the reactor experiment. Due to the two experimental levels, two separate statistical models were utilized to determine the individual statistical response of the reactor results and the bench-scale calibration results. In the bench-scale experiments, the goal was to determine the parameters and variability of the functional relationship between UV fluence and the fluorescence of the irradiated microspheres. The statistical model used in the stochastic hierarchical process is based on Bayes’ theorem and shown in eq 1.

P(θcb|Flcb,FlUVcb) )

P(Flcb|θcb,FlUVcb)P(θcb|FlUVcb) P(Flcb|FlUVcb)

(1)

In eq 1, θcb is a set of parameters used in the functional relationship model (FlUVcb) between the observed micro-

sphere fluorescence in the bench-scale experiments (Flcb) and the UV fluence level (UV) of the exposed microspheres. It was assumed that the microsphere fluorescence distribution, after being irradiated by UV of a defined “single” fluence level (UVd), could be characterized as a log-normal distribution (LogNorm). These distributions were described in the past as Gaussians (15). A log-normal distribution was used since it best describes the distributions of fluorescence due to its limited range (positive everywhere, there is no negative fluorescence) and the asymmetric shape of the fluorescence values distribution that were obtained (skewed toward higher fluorescence). To utilize the log-normal distribution, the fluorescence scale was inverted such that 1023 represented the lowest fluorescence while 0 represented the highest fluorescence. In addition, a linear correlation was assumed between the means (µ) and the standard deviations (σ) of the log-normal distributions for each fluence level tested in the collimated beam experiment. As will be discussed in the Bench-Scale Results subsection, the data significantly supported this assumption. The functional model FlUV can thus be defined as

{

Fld ≈ LogNorm(µd,σd) µd ) Bµ + AµUVd FlUVcb ) σd ) Bσ + AσUVd UVd ) {UVcb}d

(2)

where θcb from eq 1 is defined as θcb ) {Bµ,Aµ,Bσ,Aσ}. Bµ and Bσ are the intercepts of the linear fit between UV and the mean and standard deviation, respectively, of the microsphere inverted fluorescence distribution. Aµ and Aσ are the slopes of the linear fit lines and d is a counter for the level of UV fluence in the quasi-collimated beam experiment. Exponential prior distributions were assigned to Bµ, Aµ, Bσ, and Aσ with a decay rate of λ1 ) 0.001. This choice of prior distribution defined a nearly flat line in the reasonable range of values between 0 and 1000 and thus avoids introducing bias. The second step of the hierarchical process was the observation of UV fluence in the reactor by the microspheres. As in the first step, the statistical model was also based on Bayes’ theorem and can be formulated as

P(θ|FlR,FlUVR) )

P(FlR|θ,FlUVR)P(θ|FlUVR) P(FlR|FlUVR)

(3)

where FlR is a vector of the observed microsphere fluorescence in the reactor, θ ) {θcb,θR} is the full set of parameters that includes both the reactor and the collimated beam process parameters, and subscript R represents a process in the reactor. Due to their physical nature, it was assumed that the microspheres’ response to UV fluence (described by θcb) is the same in the reactor and in the collimated beam. A second model describes the functional relationship between microsphere fluorescence and UV fluence in the reactor FlUVR

{

FlR ≈ LogNorm(µR,σR) µR ) Bµ + AµUVR FlUVR ) σR ) Bσ + AσUVR UVR ≈ Gamma(a,r)

(4)

This model does not determine the distributions of Bµ, Bσ, Aµ, and Aσ, but instead, it uses the mean values that were determined by the bench-scale model. The reactor model assumes a prior Gamma distribution for the UV in the reactor (UVR), with random shape (a) and rate (r) parameters. The

prior distributions for the reactor-level parameters θR ) {a,r} are

θR )

{} {

Pareto(c1,c2) a ≈ Exp(λ2) r

(5)

where c1 ) 1 and c2 ) 5 are constant parameters of the Pareto distribution and define a decreasing distribution between c2 and ∞. These values were selected since θR has a tendency to converge in two modes, a unimodal skewed distribution, which is the target distribution, and an exponential decay from 0, which is an artificial mode. A range of c2 values was tested, and it was found that at c2 ) 5 the Pareto distribution excluded exponential distribution shapes (second mode) from the final Gamma distribution but did not limit the range of the solutions from the first mode. λ2 ) 0.001 sets a broad general distribution between 0 and ∞. In the possible range for a and r, the Gamma distribution can obtain a broad range of shapes that includes log-normal and Weibull distributions. The models were formulated and run using WINBUGS 1.4 (18) to obtain the best estimate for θ. WINBUGS uses the Bayes theorem to numerically solve the probabilities using MCMC integration (Appendix A). WINBUGS iterates over the probable range of the θ parameter values and converges to the most probable set of values, given the observations for microsphere fluorescence in the collimated beam and in the reactor. Monitors were set on the values of a and r. The model was executed as an ensemble of WINBUGS simulations starting with different initial conditions. Since this is an iterative process, convergence was reached when all of the chains in the ensemble were fluctuating around the same mean values for all monitored variables. The model typically converged before 5000 iterations. However, the 2000 converged pairs of a and r values from iterations 5001-7000 were used. Each pair of a and r values represents a different Gamma distribution. To regenerate the UV fluence distribution in the reactor, 100 values were resampled from each Gamma distribution (i.e., each pair of a and r) to obtain a total of 200 000 values of UV fluence in the reactor. The UVPDF was calculated from the histogram of the UV fluence values with a bin size of 2 UV fluence units (mJ cm-2). This resampling was done using the R.1.9.1 statistical software (19). Computational Fluid Dynamics Model. In this study, a finite-volume-based commercial computational fluid dynamics (CFD) code PHOENICS (CHAM, U. K.) was used. CFD is the science of solving the governing equations of fluid flow through space and time. With CFD, a numerical description of the process flow geometry is developed by representing each location in space with a set of grid points. Fluid parameters such as velocities and turbulent quantities are then determined at each grid point. The transport of the microorganisms through the UV reactors was simulated using a Lagrangian particle tracking approach (5), which requires the solution of the mass and momentum conservation equations. In this study, the renormalized group (RNG) twoequation k- model was used to characterize the turbulence in the UV reactor (20). The CFD particle tracking simulation is performed by first solving the flow field/turbulence within the UV reactor system. For inlet conditions, the average mean velocity normal to the inlet plane was specified. All tangential velocities were set to zero. The turbulent kinetic energy and energy dissipation rate inlet conditions were defined as follows: kinlet ) (IU)2, inlet ) kinlet1.5/(0.1D), where I ) 0.05, U is the normal average velocity at the inlet, and D is the pipe diameter. For the outlet conditions, the gradients of all variables are zero in the flow direction with the exception of the pressure. The pressure was set to zero gauge. In this VOL. 39, NO. 22, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

8927

FIGURE 1. Mean of the fluorescence intensity distributions (µfluorescence) obtained for different UV fluences (solid line) in the bench-scale experiment and mean standard deviation (std) for each distribution (σfluorescence, dashed line). The correlation coefficient (r2) and the linear fit equation for the mean and mean standard deviation are shown above the respective lines. study, a minimum of 1500 particles were released at the UV system influent. Each particle was tracked until it escaped the reactor. The local fluence rates in the UV reactor were computed using UVCalc3D, a commercially available multisegment source summation (MSSS) model (21). The MSSS approach used in UVCalc3D is based on the assumption that the emission of a linear lamp is equivalent to that of n segment sources spaced equally along the axis of the lamp. The power output for each segment source is P/n, where P is the total UV power output of the lamp in the wavelength band of interest. The overall value of the fluence rate at a point in space is then the sum of the values of the fluence rate calculated for each of the n segment sources. The results of a previous study that focused on evaluating the fluence rate distribution in a UV reactor showed that models based on the MSSS approach better predicted the local fluence rate than models based on the multipoint source summation or radiative heat transfer models. Convergence of the numerical solution was based on the restrictions that (a) the sum of the absolute residual sources over the whole solution domain must be less than 0.1% of the total inflow quantity and (b) the values of the monitored dependent variables at several locations must not change by more than 0.1% between successive iterations. The grid size was determined through successive refinement and evaluating its impact on the concentration, turbulence, and mean velocity profiles at selected points in the UV reactor. The final grid size was determined once these profiles were insensitive to further grid refinements. UV simulations were done with a structured grid. Irregular boundaries were handled using a cut-in cell method (22).

Results and Discussion Bench-Scale Results. The fluorescent microspheres were exposed to UV irradiation in a quasi-collimated beam reactor in triplicate, using the influent from the reactor. Results from the flow cytometer confirmed that the mean fluorescence intensity of the microspheres decreased with increasing UV fluence (p > 0.01). The fluorescence intensity decay rate constant was -0.27 cm2 mJ-1 (Figure 1). Decreasing fluorescence due to polychromatic radiation from a medium-pressure (MP) UV source was described previously by Anderson (15), who reported exponential decay of fluorescence in the range of 0-100 mJ cm-2. Interestingly, at 100 mJ cm-2 the decay was 3 times greater compared to the LP results reported herein. Previous work on the photochemical impact of LP UV on azo-dyes was demon8928

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 22, 2005

FIGURE 2. Relative fluorescence intensity distributions of microspheres analyzed by flow cytometry after exposure to 0, 10, 20, 30, 40, 60, 80, 100, and 120 mJ cm-2 LP UV fluence in the bench-scale system.

FIGURE 3. Flow cytometry fluorescence results for effluents taken after the UV reactor for 3 and 7.5 gpm flow rates. strated by da Silva and Faria (23) for water-cleaning purposes, although the photobleaching of fluorescent dyes upon irradiation by LP UV was not investigated. However, as mentioned earlier, a single UV fluence was not represented by single fluorescence intensity but by a distribution of fluorescence intensities (Figure 2). Possible reasons for the fluorescence intensity distribution for each fluence level may be due to (1) different initial fluorescent dye concentrations contained in each microsphere, (2) different size distribution of the microspheres, and (3) shielding of adjacent microspheres. There was a large overlap between the fluorescence intensity distributions that resulted from irradiation in different levels of UV fluences. The standard deviations of these distributions increased with increasing UV fluence (p > 0.05), although this increase was small (rate constant ) 0.02 cm2 mJ-1). On an inverted fluorescence scale (1023-0), these distributions were significantly different than Gaussian and were best described by a log-normal curve. Reactor Results. The flow cytometry analyses of microsphere-spiked water samples collected after transport through the pilot LP UV reactor revealed a distribution of fluorescence intensities (Figure 3). When the flow through the reactor decreased, the photobleaching increased, and the fluorescence intensity of the microspheres decreased accordingly. This reduction in fluorescence intensity indicated a higher accumulation of UV fluences by microspheres under decreased flow conditions, as would be expected. Increased inactivation with decreasing flow is readily found with biodosimetry (24). Therefore, the fluorescent microspheres have demonstrated this UV fluence accumulation rule as seen with its biological counterparts.

TABLE 1. Comparison of Mean UV Fluence in the Reactor Calculated Using Biodosimetry with MS2 and Bacillus Subtilis Spores, Fluorescent Microspheres, and a CFD Model mean UV fluence (mJ cm-2)

MS2 coliphagea B. subtilis sporesa microspheres CFD model

flow rate 3 gpm

flow rate 7.5 gpm

N/Ab N/Ab 119 131

42 50 36 52

a Biodosimetry results taken from a previous study using 91.4% 254 nm UV irradiation. b Not measured.

The average UV fluence in the reactor, as calculated by three different methods, CFD model, fluorescence microspheres, and biodosimetry, are compared in Table 1. Biodosimetry results are displayed to provide a REF reference point for this reactor but were taken from previous unpublished research using the same UV reactor at a UV 254 nm transmittance of 91.4%. With the faster flow rate (7.5 gpm), the mean UV fluence measured by the microspheres was 36 mJ cm-2, as compared to 52 mJ/cm-2 predicted by the CFD model. The REF results from MS2 and B. subtilis biodosimetry were 42 and 50 mJ cm-2, respectively. For 3 gpm, the microspheres predicted a mean UV fluence of 119 mJ cm-2, and the CFD model predicted 131 mJ cm-2 for the identical conditions. Overall the CFD model displayed a higher average UV fluence in the reactor than the microsphere analyses. Biodosimetry at this flow rate was not performed. Moreover, the difference in the experimental REF between MS2 and B. subtilis clearly outlines the issue associated with determining the REF from microorganisms with a different UV response. As shown by researchers, the REF from different organisms with different UV responses will not be the same for reactors with nonideal hydraulics and having a distribution of fluence rates (22, 25). A Bayesian hierarchical model was used to recalculate measured distributions of fluorescence intensities and predicted UV fluence for 3 and 7.5 gpm flow. Researchers have used Bayesian models to estimate the inactivation rate constant for protozoan pathogens (22, 25) and to calculate the uncertainties of UV reactor validation procedures (26). Figure 4 displays a comparison between the UV fluences deduced by the Bayesian model analysis of the microsphere observations and the UV fluences calculated by the CFD model. The results show a general agreement between the UV fluence curves, although the CFD model tends to display a skewed distribution at the low UV fluence levels. In the pilot test LP UV reactor, the 3 gpm flow results displayed a wider distribution of UV fluences than the 7.5 gpm flow, using both microsphere measurements and the CFD model. This wider distribution with decreasing flow rate has also been modeled previously and indicates the complex hydraulic characteristics of the UV reactor with short-circuiting and regions with recirculation zones that lead to low and high fluence values, respectively (27). The principal differences between the modeled and the measured distributions were in the LP UV fluences lower than 60 mJ cm-2 for the 3 gpm flow rate and 25 mJ cm-2 for the 7.5 gpm flow rate. The microsphere analyses indicated that the lowest UV fluence in the reactor with a flow of 3 gpm was 23 mJ cm-2, while the model predicted the lowest UV fluence at 59 mJ cm-2. For the 7.5 gpm flow rate, the lowest fluence indicated by the microsphere analysis was 13 mJ cm-2, while the model predicted 23 mJ cm-2. This difference may be due to the overlapping regions between microsphere fluorescence distributions at different UV fluence levels. It may also be

FIGURE 4. Comparison of UV fluence distributions analyzed from measurements of fluorescent microspheres in the effluent of the UV reactor and modeled by CFD for flow rates of (a) 3 gpm and (b) 7.5 gpm. due to physical effects such as shielding that are not well characterized by the CFD model. Although the microsphere measured results diverged from the fluence distribution prediction made by the CFD model particularly in the low UV fluences, the two methods produced some striking similarities, supporting the importance of evaluating the fluence distribution in reactors in addition to the mean UV fluence predictions via the biodosimetry approach. Future improvement of the proposed microsphere dosimetry technique and dye photosensitivity could lead to more precise measurement of the low UV fluence region in the reactor. The distribution of LP UV fluence in the pilot test reactor was demonstrated using fluorescent microspheres. The measurement of UV fluence in the reactor using microspheres indicates the actual UV fluence distribution under defined conditions (water parameters, flow, and hydraulics of the reactor). This result offers many fundamental and practical implications for reactor evaluation and testing, and could improve confidence in the future use of mathematical models for UV reactor characterization. It also serves as a complement to biodosimetry testing for UV reactor validation by providing greater insights regarding reactor behavior.

Acknowledgments The authors thank Dr. Shanshan Jin, who assisted in developing the initial analytical methods for evaluation of the fluorescent microspheres, and Michael Cook, who helped in development of the flow cytometry method. We also thank Dr. Amilcare Porporato for early insights, Dr. Motserrat Fuentes and Dr. Song Qian for WINBUGS advice, and Abel Rodriguez and Joseph Lucas from the Statistical Consulting Center of the Institute for Statistics and Decision Science at Duke University for help in the development of the statistical model. This research was funded by the American Water VOL. 39, NO. 22, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

8929

Works Association Research Foundation, Project No. 2682.

APPENDIX

FIGURE A1. WINBUGS solver.

Literature Cited (1) Qualls, R. G.; Johnson, J. D. Bioassay and dose measurement in UV disinfection. Appl. Environ. Microbiol. 1983, 45, 872877. (2) Linden, K. G.; Darby, J. L. Estimating effective germicidal dose from medium-pressure UV lamps. J. Environ. Eng. 1997, 123, 1142-1149. (3) Kelly, R.; Ferran, B.; Jin, S. Interpreting UV reactor validation results: Impact of key design parameters on performance of closed vessel reactors. In Proceedings of the Water Quality Technology Conference; American Water Works Association: San Antonio, TX, 2004. (4) Blatchley, E. R. Numerical modelling of UV intensity: Application to collimated-beam reactors and continuous-flow systems. Water Res. 1997, 31, 2205-2218. (5) Ducoste, J. J.; Linden, K. G.; Rokjer, D. Numerical prediction of the reduction equivalent fluence bias. In Proceedings of the Water Quality Technology Conference; American Water Works Association: San Antonio, TX, 2004. (6) Nicholson, W. L.; Galeano, B. UV resistance of Bacillus anthracis spores revisited: Validation of Bacillus subtilis spores as UV surrogates for spores of B. anthracis sterne. Appl. Environ. Microbiol. 2003, 69, 1327-1330. (7) Harm, W. Biological Effects of Ultraviolet Radiation; Press Syndicate of the University of Cambridge: Cambridge, U. K., 1980; Vol. 1. (8) Cabaj, A.; Sommer, R.; Schoenen, D. Biodosimetry: Model calculation for U.V. water disinfection devices with regard to dose distributions. Water Res. 1996, 30, 1003-1009.

8930

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 22, 2005

(9) Severin, B. F.; Suidan, M. T.; Engelbrecht, R. S. Kinetic modeling of U.V. disinfection of water. Water Res. 1983, 17, 1669-1678. (10) Marin ˜ as, B. J.; Rennecker, S.; Teefy, S.; Rice, E. W. Assesing ozone disinfection with nonbiological surrogates. J. Am. Water Works Assoc. 1999, 91, 79-89. (11) Chiou, C. F.; TorresLugo, M.; Marin ˜ as, B. J.; Adams, J. Q. Nonbiological surrogate indicators for assessing ozone disinfection. J. Am. Water Works Assoc. 1997, 89, 54-66. (12) Baeza, C.; Ducoste, J. J. A non-biological surrogate for sequential disinfection processes. Water Res. 2004, 38, 3400-3410. (13) Dai, X. J.; Hozalski, R. M. Evaluation of microspheres as surrogates for Cryptosporidium parvum oocysts in filtration experiments. Environ. Sci. Technol. 2003, 37, 1037-1042. (14) Tang, G.; Teefy, S.; Marin ˜ as, B. J. Assessing the performance of an ozone disinfection system with fluorescent-dyed microspheres as surrogates for C. parvum oocysts. In Proceedings of the Water Quality Technology Conference; American Water Works Association: Seatle, WA, 2002. (15) Anderson, W. A.; Zhang, L.; Andrews, S. A.; Bolton, J. R. A technique for direct measurement of UV fluence distribution. In Proceedings of the Water Quality Technology Conference; American Water Works Association: Philadelphia, PA, 2003. (16) Trotter, J. Windows Multiple Document Interface for Flow Cytometry (WinMDI), version 2.8; software created 1993, http:// facs.scripps.edu/software.html (program accessed 2004). (17) Bolton, J. R.; Linden, K. G. Standardization of methods for fluence (UV dose) determination in bench-scale UV experiments. J. Environ. Eng. 2003, 129, 209-215. (18) Lunn, D. J.; Thomas, A.; Best, N.; Spiegelhalter, D. WinBUGSsA Bayesian modelling framework: Concepts, structure, and extensibility. Stat. Comput. 2000, 10, 325-337. (19) Ihaka, R.; Gentleman, R. R: A language for data analysis and graphics. J. Comput. Graph. Stat. 1996, 5, 299-314. (20) Wilcox, D. C. Turbulence Modeling for CFD; DCW Industries, Inc.: La Canada, CA, 1998. (21) Liu, D.; Ducoste, J. J.; Jin, S.; Linden, K. G. Evaluation of fluence rate distributions models. J. Water Supply Res. Technol.sAQUA 2004, 53, 394-408. (22) Sivaganesan, M.; Rice, E. W.; Marin ˜ as, B. J. A Bayesian method of estimating kinetic parameters for the inactivation of Cryptosporidium parvum oocysts with chlorine dioxide and ozone. Water Res. 2003, 37, 4533-4543. (23) da Silva, C. G.; Faria, J. L. Photochemical and photocatalytic degradation of an azo dye in aqueous solution by UV irradiation. J. Photochem. Photobiol., A 2003, 155, 133-143. (24) Mackey, E. D.; Hargy, T. M.; Wright, H. B.; Malley, J. P. J.; Cushing, R. S. Comparing Cryptosporidium parvum and MS2 bioassayss Implications for UV reactor validation. J. Am. Water Works Assoc. 2002, 94, 62-69. (25) Qian, S. S.; Donnelly, M.; Schmelling, D. C.; Messner, M.; Linden, K. G.; Cotton, C. Ultraviolet light inactivation of protozoa in drinking water: A Bayesian meta-analysis. Water Res. 2004, 38, 317-326. (26) EPA. Ultraviolet Disinfection Guidance Manual, draft; U. S. Environmental Protection Agency: Washington, DC, 2003. (27) Ducoste, J. J.; Linden, K. G. Hydrodynamic Characterization of UV Reactors; Draft Final Report, Project 2684; American Water Works Association Research Foundation: Denver, CO, 2005.

Received for review January 6, 2005. Revised manuscript received August 29, 2005. Accepted August 29, 2005. ES050034C