Applications of Fluorescence Spectroscopy for Predicting Percent

Dissolved organic carbon (DOC) is a significant organic carbon reservoir in ... Online fluorescence spectroscopy for the real-time evaluation of the m...
0 downloads 0 Views 3MB Size


Applications of Fluorescence Spectroscopy for Predicting Percent Wastewater in an Urban Stream Jami H. Goldman,*,†,‡ Stewart A. Rounds,‡ and Joseph A. Needoba† †

Institute of Environmental Health, Oregon Health and Science University, 20000 NW Walker Road, Beaverton, Oregon 97006, United States ‡ USGS Oregon Water Science Center, 2130 SW Fifth Avenue, Portland, Oregon 97201-4976, United States S Supporting Information *

ABSTRACT: Dissolved organic carbon (DOC) is a significant organic carbon reservoir in many ecosystems, and its characteristics and sources determine many aspects of ecosystem health and water quality. Fluorescence spectroscopy methods can quantify and characterize the subset of the DOC pool that can absorb and reemit electromagnetic energy as fluorescence and thus provide a rapid technique for environmental monitoring of DOC in lakes and rivers. Using high resolution fluorescence techniques, we characterized DOC in the Tualatin River watershed near Portland, Oregon, and identified fluorescence parameters associated with effluent from two wastewater treatment plants and samples from sites within and outside the urban region. Using a variety of statistical approaches, we developed and validated a multivariate linear regression model to predict the amount of wastewater in the river as a function of the relative abundance of specific fluorescence excitation/emission pairs. The model was tested with independent data and predicts the percentage of wastewater in a sample within 80% confidence. Model results can be used to develop in situ instrumentation, inform monitoring programs, and develop additional water quality indicators for aquatic systems.

INTRODUCTION Determining the ecosystem health and water quality of a river requires identification of both natural and anthropogenic influences on aquatic biogeochemistry and an understanding of seasonal hydrologic variations. Organic matter (OM) is a critical component of many nutrient and trace metal biogeochemical cycles, and can significantly affect ecosystems biological processes and water quality. For example, OM can limit light availability,1 keep toxic metals in solution, and increase the biochemical oxygen demand (BOD).2 Typical types of dissolved organic matter (DOM) include differing sizes of organic molecules from soils and decaying plants and animals, exudates from living algae, and anthropogenic inputs from sewage, agriculture, and urban landscapes. Sources of DOM in rivers are commonly categorized as (1) allochthonous material derived from outside the ecosystem, and (2) autochthonous material derived from biota (e.g., algae, bacteria, and macrophytes) within the ecosystem.3,4 DOM can be highly reactive and is an integral part of microbial food webs, biogeochemical reactions, and ultraviolet (UV) light attenuation. Dissolved organic carbon (DOC), a main component of the DOM pool, is a significant global carbon reservoir found in all ecosystems and is an important water quality indicator.5 Measurements of the optical properties of DOM provide information about the character, quantity, and sources of the DOM pool.6−10 Differences in those characteristics then can be linked to changes in DOM reactivity and may be used to infer DOM sources.9 Fluorescence spectroscopy is a method to quantify and characterize a subset of the DOC pool which can © 2012 American Chemical Society

absorb certain wavelengths of light and re-emit a fraction of that energy as fluorescence.11 The fluorescent component of DOC (FDOM) is itself a subset of the absorbent component of DOC: the colored dissolved organic matter (CDOM). Qualitative information also can be derived by measuring the intensity of fluorescence over the entire range of excitation and emission wavelengths, thus creating an excitation−emission matrix (EEM) that captures numerous specific FDOM components such as humic and fulvic acids, protein-like material, and phytoplanktonderived material in the same sample.6,8,12 The application of multivariate statistical analyses, such as parallel factor analysis (PARAFAC), can provide further insight into DOM quality, source, and processing.13−15 PARAFAC is a method that decomposes the fluorescence signal of DOM into unique fluorescent groups whose abundance can be related to DOM precursor materials.12 Due to the lack of gathering full in situ EEMs in real-time, this analysis approach was not used for the current research. Peak-picking is a viable analysis technique used for the development and use of a real-time tool, which can be directly tied to custom sensors available today. Over 130 000 km of streams and rivers in the United States are impaired by urbanization and the resulting effects on water quality, eutrophication, and habitat quality.16 An important Received: Revised: Accepted: Published: 4374

December March 19, March 21, March 21,

23, 2011 2012 2012 2012 | Environ. Sci. Technol. 2012, 46, 4374−4381

Environmental Science & Technology


need for improved environmental stewardship of aquatic ecosystems is a better ability to monitor and track the sources of anthropogenic pollution, such as organic carbon, nutrients, and contaminants. The range of pollutants and analytical detection methods prevents the routine in-stream monitoring of many such inputs; however, by examining the fluorescence characteristics of organic matter, it is possible to distinguish certain anthropogenically derived components and thus monitor for the presence of wastewater and sewage in streams of concern. The aim of this study was to inform near-term instrument development methodologies that can be directly applied to in situ monitoring for the characterization of wastewater effluent in the Tualatin River using fluorescence signals of DOM. This was achieved by using a multivariate linear regression model approach to quantify the amount of wastewater in a sample from fluorescence measurements that can distinguish sources and characteristics of natural and anthropogenic stream organic matter.

Figure 1. Graph of streamflow in the Tualatin River at West Linn (near Oswego Dam sampling site), showing the typical annual variation in streamflow, with higher flows during the November−April rainy season and low flows during the May−October summer period. The red dots represent the sampling dates.

EXPERIMENTAL SECTION Site Description. The study was conducted in the Tualatin River basin west of Portland, Oregon (Figure S1). The basin encompasses approximately 712 mi2 with tributaries entering the river from a wide range of land use regions.17,18 The river is relatively steep exiting the Coast Range, but becomes a slow moving and meandering river on the valley bottom, where pooled reaches can produce sizable phytoplankton blooms in summer. Peak rainfall and higher flows occur November−April, and the drier, low-flow season occurs May−October. Approximately 500,000 people reside in 15% of the Tualatin River basin, primarily in the urbanized lower basin. Clean Water Services is the primary wastewater and stormwater management utility and treats ∼60 million gallons of wastewater per day through the operation of four wastewater treatment plants (WWTPs). The two largest WWTPs use advanced secondary and tertiary treatment to remove excess nutrients, biochemical oxygen demand, and suspended solids. The basin has continuous streamflow, precipitation, solar radiation, and water-quality monitoring stations along the river and its tributaries as well as a robust routine water sampling and analysis program to characterize streamwater quality. During low flow periods, the Tualatin River near its mouth can contain up to 40% treated wastewater.17 This study was conducted primarily in the lower, urbanized area of the Tualatin River basin (Figure S1), with the exception of one site outside the basin in an agricultural/forested tributary representing an end-member site with no wastewater characteristics. The study sites included two WWTPs (Durham and Rock Creek Advanced Treatment Facilities), three tributaries (Fanno, Beaverton, and Rock Creeks), one headwater site (East Fork Dairy Creek), and one downstream river site (Tualatin River at Oswego Dam). Sample Collection and Analysis. Water samples were collected every 3−4 weeks at each site over an entire year, and over a range of hydrologic conditions and seasonal changes (Figure 1). Samples were collected with a Wildco Teflon Kemmerer 1.2-L sampler and transferred into glass BOD bottles. Grab samples of wastewater were taken from the effluent outflow sites. Samples were filtered (Whatman GF/F, 0.7 μm, combusted at 450 °C for 4 h) through a glass filtration unit. The filtrate was collected into precombusted glass amber bottles with Teflonlined caps and stored in the dark at 4 °C until analyzed. All samples were analyzed within 5 days of collection.

Fluorescence excitation−emission matrix (EEM) measurements were made using a Horiba Jobin Yvon Flouromax-4 spectrofluorometer. To obtain fluorescence EEMs, excitation wavelengths were increased from 240 to 450 nm at intervals of 10 nm, and fluorescence was measured at emission wavelengths of 300−600 nm at intervals of 2 nm. The excitation and emission slits were set to a 5-nm bandpass. A flow-through water bath was used to maintain a constant sample temperature of 20 °C. UV−visible absorbance measurements were made with a J&M TIDAS spectrophotometer (World Precision Instruments) in a 1-cm quartz cuvette and blank corrected. DOC measurements were made using the platinum catalyzed persulfate wet oxidation method on an O.I. Analytical model 700 TOC Analyzer19 at the U.S. Geological Survey laboratory in Boulder, CO. Specific ultraviolet absorbance at 254 nm (SUVA254) was determined by dividing the UV absorbance measured at 254 nm by the DOC concentration and is reported in units of liter per milligram of carbon.20 Several postacquisition steps were used to adjust the EEM data (Figure S2). First, the excitation and emission data were corrected for instrument-specific response. Second, the EEM response of Milli-Q water was subtracted from sample EEMs. Third, the UV−visible absorption spectra were used to correct the EEM data for inner filter effects.9 Finally, the fluorescence intensities of the EEMs were normalized to the area under the Raman peak, converting the arbitrary units (AU) into Raman units (RU). The corrected EEMs were imported into Matlab R2010A (Mathworks) for further analysis, including removing portions of the EEMs where interference from Raleigh scattering occurred, converting the data into vectors, selecting characteristic peak signals based on documented key excitation/ emission pairings,9,11 and plotting the data into contour and surface maps. The fluorescence index (FI, the ratio of emissions at 470−520 nm at an excitation of 370 nm), which distinguishes between DOM derived from microbial versus terrestrial sources, was calculated for all samples.9 All samples were concurrently analyzed for chlorophyll a and dissolved nutrients (nitrate, nitrite, phosphate, ammonium, and silicate). Nutrients were analyzed using a 5-channel 2008 model Astoria-Pacific segmented continuous flow injection analyzer designed for spectrophotometric analysis of nutrients in

4375 | Environ. Sci. Technol. 2012, 46, 4374−4381

Environmental Science & Technology


Figure 2. Sample excitation−emission matrices (EEMs) from June 6, 2009 from the headwater site (East Fork Dairy Creek), one of the wastewater effluent sites (Rock Creek WWTP), and the downstream site (Tualatin River at Oswego Dam). The color scales for all three EEMs are identical and set to a range of 0−1.0 intensity.

freshwater. Chlorophyll a was analyzed using a Turner Trilogy fluorometer. Modeling Approach. Wastewater and other anthropogenic materials such as fluorescent whitening agents (FWA) have been documented to influence specific regions of the EEM at peak A (excitation at 260 nm and emission at 450 nm [ex260/em450]), peak T (ex270/em340), and peak C (ex340/em440).21,22 The EEMs from samples collected from the Durham and Rock Creek WWTPs have been evaluated and show these three peaks with the strongest fluorescent intensities (Figure 2). In addition, in situ fluorometers designed to measure excitation/emission peaks in all three regions can be constructed and used for applied research and monitoring in future work. End-Member Mixing Experiments. Laboratory mixing experiments with two end-members (one effluent sample and one headwater reference sample) were conducted to determine the fluorescence response of mixed samples and the degree of linearity in that response. Samples were collected from both end members (East Fork Dairy Creek and both WWTPs) and sample mixtures were prepared. Two experiments were conducted (one per WWTP), and 9 samples were prepared with 10% increments of wastewater per sample. Samples were mixed and shaken for 2 h at room temperature prior to analysis. Fluorescence and absorbance measurements were made and post-EEM acquisition steps were applied. End-Member Mixing Model. Three approaches were used to analyze the results from the stream samples. In the first approach, the potential for individual peaks to predict percent wastewater was assessed using an end-member mixing model. Rearrangement of a mass balance equation for a two endmember system was used to calculate percent wastewater at the downstream river site, Tualatin River at Oswego Dam.23 % WW =

Upstream − Downstream × 100 ⎛ WW1 + WW2 ⎞ ⎜Upstream − ⎟ ⎝ ⎠ 2



fluorescence intensities for peaks A, T, and C were identified and used separately in the equation above. The calculated percent wastewater in the Tualatin River at Oswego Dam from this model was compared to the actual percentage of wastewater computed for that site from flow measurements. Multivariate Linear Regression Model. The second approach was to use signals from peaks A, T, and C together in a multivariate linear regression model. The statistical package Minitab 16.1.1 (Minitab Inc.) was used to develop this model. In determining the best possible multivariate regression model, an exploratory analysis was conducted using all of the key peaks (A, C, M, B, and T) identified in the OM literature.11,24 Based on preliminary data analysis we found the strongest model could be constructed based on peaks A, T, and C. Other model scenarios were conducted to find the most robust model, such as using peak signal ratios and differences to minimize multicollinearity, alternating the assumptions made on tributary sample wastewater percentages, and varying the model input sample size. Based on an analysis of goodness-of-fit statistics and use of the variance inflation factor to quantify multicollinearity, the strongest model was produced using an unmodified signal from peak A, peak signal ratios A/T and A/C as explanatory variables, and an assumption that tributary samples contained no wastewater (which is consistent with known sources to those tributaries) (Table S1). Model inputs incorporated data from 74 samples, including 12 headwater samples (East Fork Dairy Creek), 28 tributary samples (Beaverton, Fanno, and Rock Creeks), 11 downstream river samples (Tualatin River at Oswego Dam), and 23 wastewater effluent samples (Rock Creek and Durham WWTPs) (Table S2). Headwater and tributary samples were assumed to contain 0% wastewater. The percent wastewater in downstream river samples was calculated using methods applied by Bonn.17 The model was tested with 30 samples not used for model calibration, including 17 samples collected from a site on the Clackamas River, outside of the urban growth boundary (0% wastewater) and 13 samples from the Tryon Creek WWTP (100% WW)(Table S3). Principal Component Analysis (PCA). PCA was used to identify patterns in the data set. PCA was conducted using the Minitab 16.1.1 software (Minitab, Inc.) and 3 variables (peak intensities for A, T, and C) for the same 74 samples used in the regression model. The peak intensities were normalized prior to PCA by subtracting the mean of the peak intensities (separately


where “upstream” represents the fluorescence signal from the headwater site as an indicator of upstream conditions not affected by wastewater, “downstream” represents the fluorescence signal at the downstream mixed site, and WW1 and WW2 represent the fluorescence signals from the Rock Creek and Durham WWTP samples, respectively. In this analysis, the 4376 | Environ. Sci. Technol. 2012, 46, 4374−4381

Environmental Science & Technology


for Peaks A, T, and C) and dividing by the standard deviation. PCA scores were computed and graphically represented by plotting the first principal component against the second.

RESULTS End-Member Mixing Experiments. A linear response in fluorescence peak intensity was confirmed for Peaks A, T, and C from the results of the end-member mixing experiments (Figure 3). In general, as the percentage of wastewater in mixed

Figure 4. Predicted wastewater model results plotted versus actual wastewater found at Oswego Dam. The black line represents an ideal linear 1:1 line response. The blue diamonds are results based on the multivariate regression model. The individual peaks are the results from the two-component wastewater mixing model.

matrix. Figure 5 illustrates the relationship between the predicted and reported wastewater percentages, and demonstrates

Figure 3. Results from end-member mixing experiments using water from East Fork Dairy Creek and water from (a) Durham Wastewater Treatment Plant effluent and (b) Rock Creek Wastewater Treatment Plant effluent. Results from separate signals, using fluorescence intensities from peaks A, T, and C show that the fluorescence responded in a linear fashion as the percentage of wastewater in the mixed water sample increased.

samples increases, the peak intensities increase linearly. Based on this confirmation of a linear fluorescence response, an endmember mixing model is well suited for two end-member mixing experiments, but may or may not be successful in predicting results for the downstream river site, depending on whether OM sources can be categorized into two major groups (headwater and wastewater). If the downstream river samples have more than two distinct OM sources, then the two-component endmember mixing model may be inappropriate. End-Member (EM) Mixing Model. The two-component EM model was applied to predict percent wastewater at the downstream river site (Figure 4). Optimal results should fall on or near the 1:1 line. EM mixing model results from separate signals, using fluorescence intensities from peaks A, T, and C, showed that predicted wastewater percentages tend to increase as actual wastewater percentages increase, but were above the 1:1 line. It is likely the EM model failed to predict the percent wastewater accurately because of multiple sources of OM in the basin. Multivariate Linear Regression (MLR) Model. The MLR model predicted the percentage of wastewater at a downstream site with better accuracy than the EM model (Figure 4). The best parameters for the model were

Figure 5. Results from the multivariate linear regression model for percentage wastewater over a range of sample conditions, where effluent samples are plotted at 100% wastewater, tributary and headwater samples are plotted at 0% wastewater, and river samples downstream constitute the points in between. The black line is a linear regression line through the results and the light blue boundary lines represent the 80% prediction interval.

that the model captures 95% of the variability in the data (R2 = 0.95) within 80% accuracy. The mean error and mean absolute error of the model are 0.10% and 8.1%, respectively. In the optimal model, the mean error should be near zero to indicate a minimum of model bias, and the typical error as expressed by the mean absolute error should be small enough (less than about 10% in this case) that model predictions remained useful. The model had the greatest predictive capabilities for the headwater and tributary sites with mean errors of 4.7 and 1.2% and mean absolute errors of 1.2 and 7.6%, respectively. The downstream and WWTP samples showed a greater deviation, with mean errors 5.2 and 3.2% and mean absolute errors 9.2 and 8.6%, respectively. The model was tested with fluorescence data from the pristine Clackamas River site within 95% accuracy, but it overpredicted the percent wastewater character of Tryon Creek WWTP effluent (approximately 125%) for all samples. The Tryon Creek WWTP used less advanced treatment

%WW = (2.02 + 0.420*Peak A − 0.118*Peak A/Peak T − 1.01*Peak A/Peak C)*100


where %WW was the calculated percentage of wastewater in the sample, and Peaks A, C, and T were the fluorescence intensities at the indicated peak in the excitation−emission 4377 | Environ. Sci. Technol. 2012, 46, 4374−4381

Environmental Science & Technology


processes than the Durham and Rock Creek WWTPs and had 40% more DOC in its effluent, which likely accounted for the model overprediction. A comparison of actual and predicted percent wastewater at the downstream river site over time indicated when the model deviated from the values calculated from the flow estimates (Figure 6). The greatest divergence occurred during two

Figure 7. Plot of principal component 2 (PC2) versus principal component 1 (PC1) from a principal component analysis of the fluorescence data from Peaks A, T, and C. Sites shaded in orange are effluent samples from wastewater treatment plants, shaded light blue data represents headwater site, shaded dark blue data represent the downstream river site, and the shaded yellow data represent the tributary sites. The blue outlined areas all have a fluorescence index (FI) value 1.66. DOC concentrations increase with increasing PC1, while SUVA254 values decrease as PC2 increases.

Figure 6. Results from the multivariate linear regression model for the downstream river site, Tualatin River at Oswego Dam. The top graph shows the modeled percent wastewater results with the actual percent wastewater. The bottom graph shows phosphate, ammonium, and chlorophyll a. The vertical dashed red lines indicate the greatest divergence between the predicted and modeled results.

the headwater and WWTP waters, but rather a complex mixture of many different tributaries and sources of organic matter. The MLR model had several advantages over a twocomponent EM model. For example, once a model has been constructed and tested for a system, only one sample is required to make a prediction. Predictions of percent wastewater in this study can provide near 80% accuracy and perhaps even greater during low-flow (higher wastewater) periods. Additionally, the MLR model approach provided the ability to quantify more complex mixtures of OM using multiple excitation/emission pairings. Using multiple peaks (A, T, and C) allowed the model to capture signatures from more humic, terrestrially derived OM as well as anthropogenic and microbially derived components.26 Constructing a model across a range of locations and seasons also allowed the model to capture a changing menu of OM sources across a range of hydrologic conditions. The greatest errors in the MLR model occurred in samples from the downstream river site and the WTTPs. One model assumption was that certain OM signatures were sufficient to categorize an effluent sample as 100% wastewater; however, that assumption does not account for the OM complexity and variability that occurs seasonally due to changing source materials and treatment processes, and variations from one WWTP to another. Both treatment plants in this study use tertiary treatment during summer and had similar input levels, but because they serve different populations they require slightly different treatments that could have caused variations in the effluent fluorescence characteristics. At the time of sample collection, the Durham WWTP used advanced biological methods of nutrient removal for phosphorus and ammonia, along with a reduced application of alum and lime for phosphorus removal, while the Rock Creek WWTP did not employ the same nutrient removal processes and therefore was using twice as much lime and alum. The WWTPs have higher inflows during the wet season, resulting in seasonally varying contributions of anthropogenic sources and terrestrial humic sources to the WWTPs. Both WWTPs typically

periods illustrated by the dashed red lines, the first on June 29, 2009 and the second between the November 17 and December 16 samples. The first divergence in late June underpredicted the amount of wastewater and coincided with the peak of a significant phytoplankton bloom, corroborated by chlorophyll a measurements of 20 μg/L (Figure 6). The second divergence (November−December), when the model overpredicted the amount of wastewater, corresponds to the first high-flow events to occur after the summer low-flow period (Figure 1). During this period, there was an increase in the amount of ammonium and phosphate in the system, largely because of a reduction in the seasonal nutrient removal at the WWTPs (Figure 6). Principal Component Analysis. PCA was used to assess the variability in the fluorescence data from the samples used in the regression model. A plot of principal component 2 (PC2) versus principal component 1 (PC1) showed that the downstream river, headwater, tributary, and wastewater samples tend to cluster together (Figure 7). PC1 captures 83% of the variability in the fluorescence data, and 16% of the remaining variability can be explained by PC2.

DISCUSSION The two-component EM model failed to adequately determine the actual percent wastewater in the river. The EM models, based separately on peaks A, T, or C, do trend in the correct positive direction, but all three mixing models overestimated the percent wastewater for every sample. Peak T is considered a tracer of anthropogenic material and has been documented to be highly correlated to the biodegradable fraction of wastewater.25 Peak T was, indeed, a better predictor than Peaks C and A, but still resulted in inaccuracies most likely because the downstream river sample was not simply a mixture of OM from 4378 | Environ. Sci. Technol. 2012, 46, 4374−4381

Environmental Science & Technology


Figure 8. Summary graphs for averages of DOC, SUVA254, and FI values. Top left graph and bottom graphs show averages for high, medium, and low-flow conditions. The graph on the top right shows year-long averages for all three parameters with DOC and SUVA254 on the left Y axis and FI value on the right Y axis.

these facilities occurred during the summer low-flow period to protect the river’s water quality under those critical conditions, and that is why we see the highest DOC concentrations occurring during the high flow periods instead. The tributary samples had the second highest DOC concentrations, lowest FI values, and highest SUVA254. Contrary to the WWTP results, the tributary samples had the highest concentrations of DOC during the low-flow season, possibly due to algal growth and a heightened level of bacterially mediated OM processing during the warm months. The headwater samples had the lowest concentration of DOC, but values of SUVA254 and FI that were similar to those from tributary samples. The headwater site had its highest DOC concentrations during the high-flow period, which might reflect the mobilization of OM from the streambed or nearby riparian areas. Finally, the downstream river samples exhibited mid ranges for all parameters, demonstrating that the mixture is truly a combination of the WWTP and headwater/tributary samples with a mixture of sources and OM characteristics. The highest DOC concentrations and the highest SUVA254 values from the downstream river site occurred during high-flow conditions. The heterogeneous mixture of compounds in wastewater effluent had a fluorescence signature and characteristics that were distinct from those in river samples. The effluent mixture exhibited high DOC concentrations with a low SUVA254 value, indicating more labile and less aromatic carbon structures.3 The FI has been widely used to indicate the relative contributions of algal versus terrestrially derived DOM; a higher FI is associated with algal derived material, which has lower aromatic content and lower molecular weight, while a lower FI is associated with more highly processed, terrestrially derived material that has greater aromatic content and higher molecular weight.9,14

reduce their treatment processes during this period, partly due to temperature constraints in maintaining certain bacterial cultures. In the sampling year, the WWTPs converted from advanced to secondary treatment in early to mid-November, which coincides with the second period of greatest divergence found in the model results. The increased ammonium and phosphate during this time supports an increase in the effluent OM influence on downstream river samples. The model overpredicted wastewater during this period; however, the change in treatment processes was not accounted for in the model with any sort of time-varying term or other seasonal categorical classification. The heterogeneous mixture of compounds in wastewater effluent resulting from domestic and industrial waste, and, periodically, from stormwater and groundwater sources has a fluorescence signature that can be identified in a more complex mixture of aquatic organic matter.8 The MLR model underpredicted the amount of wastewater during the peak of a phytoplankton bloom (Figure 6), which could have very different fluorescence characteristics from those of OM in effluent. As with the seasonal variation in wastewater treatment, the model did not specifically account for periods when upstream algal blooms significantly changed the mixture of OM sources and characteristics. The principal component analysis demonstrated the variability in peaks A, T, and C that allowed delineation of the sources and characteristics of OM. Some of the clustering and trends in Figure 7 can be explained by the characteristics of OM in these samples. The DOC concentration, SUVA254, and FI varied according to the sample site category and time of year or flow condition (Figure 8). WWTP effluent samples typically provided the highest DOC concentration, highest FI value, and lowest SUVA254. The most extensive wastewater treatment at 4379 | Environ. Sci. Technol. 2012, 46, 4374−4381

Environmental Science & Technology

ACKNOWLEDGMENTS We thank USGS Boulder, CO for our DOC analyses and USGS Sacramento, CA for EEM Matlab codes. This research was supported in part through the National Science Foundation cooperative agreement OCE-0424602 awarded to the Center for Coastal Margin Observation and Prediction.

The WWTP samples all produced an FI greater than 1.66 (Figure 8), identifying the OM in those samples as having a more microbially derived composition. The tributary samples carried a fluorescence signature characteristic of dissolved humic substances such as lignin, tannins, and polyphenols that more commonly comprise the bulk of humic DOM fluorescence.24,27,28 In addition, the tributary samples consisted of higher DOC concentrations similar to those of the WWTP samples but distinctly different, with a low FI and a high SUVA254 indicative of less labile carbon and more aromatic structures. The similarity between the tributary and headwater organic matter characteristics was consistent with OM sources that are natural. Downstream river samples showed characteristics in the midrange for DOC concentrations, SUVA254, and FI supporting the idea that the OM at this site is a complex mixture with sources that might have been sufficiently characterized by the headwater, tributary, and WWTP samples.

ENVIRONMENTAL IMPLICATIONS The use of fluorescence to predict the presence and quantity of wastewater in a stream sample has significant implications for water quality monitoring. Recent advancements in sensor technology and the development of reliable and specific fluorescence probes have increased our ability to monitor OM characteristics in near real-time. Custom sensors have already been constructed to capture the key peaks used in this study. The application of real-time data and MLR-type models should allow water managers to track some point and nonpoint sources of pollution without high costs associated with complex analytical approaches. Even with its varying accuracy, the presented MLR model enabled detection of unknown anthropogenic inputs and may be useful in identifying problem areas or issues that might otherwise go undetected. The detection and tracking of wastewater spills, leaking sewer lines, and failing septic systems, for example, could be aided with the type of models introduced in this study combined with custom fluorescence sensors. Even though the study was conducted on treated effluent there should be similarities in the organic matter of untreated sewage, especially the strong microbial fluorescence signal and the presence of similar constituents, such as optical brighteners. In addition, many urban areas have increased concerns with organic wastewater contaminants (OWC) such as hormon ally active chemicals, personal care products, and pharmaceuticals,29 but the analysis of these OWCs often is cost prohibitive. The presented fluorescence approach could be used to determine the best times to perform expensive analyses such as for OWCs and other anthropogenic pollutants in aquatic ecosystems. ASSOCIATED CONTENT

S Supporting Information *

A sampling site map (Figure S1), protocols for EEM postprocessing (Figure S2), MLR model goodness of fit statistics (Table S1), sample data for MLR model (Table S2), and sample data for MLR model validations (Table S3). This information is available free of charge via the Internet at http://


(1) Wetzel, R. G. Gradient-dominated ecosystems: Sources and regulatory functions of dissolved organic matter in freshwater ecosystems. Hydrobiologia 1992, 229 (1), 181−198. (2) Thacker, S.; Tipping, E.; Baker, A.; Gondar, D. Development and application of functional assays for freshwater dissolved organic matter. Water Res. 2005, 39 (18), 4559−4573. (3) Aiken, G.; Cotsaris, E. Soil and hydrology: Their effect on NOM: Natural organic matter. J. - Am. Water Works Assoc. 1995, 87 (1), 36−45. (4) Reckhow, D. A.; Rees, P. L.; Nüsslein, K.; Makdissy, G.; Devine, G.; Conneely, T.; Boutin, A.; Bryan, D. Long-Term Variability of BDOM and NOM as Precursors in Watershed Sources; AwwaRF Report 91186; IWA Publishing: London, 2008. (5) Battin, T. J.; Luyssaert, S.; Kaplan, L. A.; Aufdenkampe, A. K.; Richter, A.; Tranvik, L. J. The boundless carbon cycle. Nat. Geosci. 2009, 2 (9), 598−600. (6) Coble, P. G. Marine optical biogeochemistry: The chemistry of ocean color. Chem. Rev. 2007, 107 (2), 402−418. (7) Downing, B. D.; Boss, E.; Bergamaschi, B. A.; Fleck, J. A.; Lionberger, M. A.; Ganju, N. K.; Schoellhamer, D. H.; Fujii, R. Quantifying fluxes and characterizing compositional changes of dissolved organic matter in aquatic systems in situ using combined acoustic and optical measurements. Limnol. Oceanogr.: Methods 2009, 7, 119−131. (8) Hudson, N.; Baker, A.; Reynolds, D. Fluorescence analysis of dissolved organic matter in natural, waste and polluted watersa review. River Res. Appl. 2007, 23 (6), 631−649. (9) McKnight, D. M.; Boyer, E. W.; Westerhoff, P. K.; Doran, P. T.; Kulbe, T.; Andersen, D. T. Spectrofluorometric characterization of dissolved organic matter for indication of precursor organic material and aromaticity. Limnol. Oceanogr. 2001, 46 (1), 38−48. (10) Stedmon, C. A.; Bro, R. Characterizing dissolved organic matter fluorescence with parallel factor analysis: A tutorial. Limnol. Oceanogr.: Methods 2008, 6 (1), 1−6. (11) Coble, P. G. Characterization of marine and terrestrial DOM in seawater using excitation-emission matrix spectroscopy. Mar. Chem. 1996, 51 (4), 325−346. (12) Stedmon, C. A.; Markager, S.; Bro, R. Tracing dissolved organic matter in aquatic environments using a new approach to fluorescence spectroscopy. Mar. Chem. 2003, 82 (3−4), 239−254. (13) Cory, R. M.; McKnight, D. M. Fluorescence Spectroscopy Reveals Ubiquitous Presence of Oxidized and Reduced Quinones in Dissolved Organic Matter. Environ. Sci. Technol. 2005, 39 (21), 8142− 8149. (14) Jaffé, R.; McKnight, D.; Maie, N.; Cory, R.; McDowell, W.; Campbell, J. Spatial and temporal variations in DOM composition in ecosystems: The importance of longterm monitoring of optical properties. J. Geophys. Res. 2008, 113, G04032. (15) Kraus, T. E. C.; Anderson, C. A.; Morgenstern, K.; Downing, B. D.; Pellerin, B. A.; Bergamaschi, B. A. Determining sources of dissolved organic carbon and disinfection byproduct precursors to the McKenzie River, Oregon. J. Environ. Qual. 2010, 39, 2100−2112. (16) Paul, M. J.; Meyer, J. L. Streams in the urban landscape. Urban Ecol. 2008, 207−231. (17) Bonn, B. Tualatin River Flow Management Technical Committee 2009 Annual Report; Clean Water Services: Hillsboro, OR, 2009. (18) Bonn, B. A., Rounds, S. A. Use of Stable Isotopes of Carbon and Nitrogen to Identify Sources of Organic Matter to Bed Sediments of the Tualatin River, Oregon; U.S. Geological Survey Scientisfic Investigations Report 2010-5154; U.S. Geological Survey: Reston, VA, 2010; p 58.



Corresponding Author

*Phone: (503) 251-3205; fax: (503) 251-3470; e-mail: [email protected] Notes

The authors declare no competing financial interest. 4380 | Environ. Sci. Technol. 2012, 46, 4374−4381

Environmental Science & Technology


(19) Aiken, G. R. Chloride interference in the analysis of dissolved organic carbon by the wet oxidation method. Environ. Sci. Technol. 1992, 26 (12), 2435−2439. (20) Weishaar, J. L.; Aiken, G. R.; Bergamaschi, B. A.; Fram, M. S.; Fujii, R.; Mopper, K. Evaluation of specific ultraviolet absorbance as an indicator of the chemical composition and reactivity of dissolved organic carbon. Environ. Sci. Technol. 2003, 37 (20), 4702−4708. (21) Hudson, N.; Baker, A.; Reynolds, D. Fluorescence analysis of dissolved organic matter in natural, waste and polluted waters - A review. River Res. Appl. 2007, 23 (6), 631−649. (22) Saadi, I.; Borisover, M.; Armon, R.; Laor, Y. Monitoring of effluent DOM biodegradation using fluorescence, UV and DOC measurements. Chemosphere 2006, 63 (3), 530−539. (23) Schemel, L. E.; Cox, M. H.; Runkel, R. L.; Kimball, B. A. Multiple injected and natural conservative tracers quantify mixing in a stream confluence affected by acid mine drainage near Silverton, Colorado. Hydrol. Processes 2006, 20 (13), 2727−2743. (24) Fellman, J. B.; Hood, E.; Spencer, R. G. M. Fluorescence spectroscopy opens new windows into dissolved organic matter dynamics in freshwater ecosystems: A review. Limnol. Oceanogr. 2010, 55 (6), 2452−2462. (25) Reynolds, D. M.; Ahmad, S. R. Rapid and direct determination of wastewater BOD values using a fluorescence technique. Water Res. 1997, 31 (8), 2012−2018. (26) Baker, A.; Spencer, R. G. M. Characterization of dissolved organic matter from source to sea using fluorescence and absorbance spectroscopy. Sci. Total Environ. 2004, 333 (1−3), 217−232. (27) Del Vecchio, R.; Blough, N. V. On the origin of the optical properties of humic substances. Environ. Sci. Technol. 2004, 38 (14), 3885−3891. (28) Green, S. A.; Blough, N. V. Optical absorption and fluorescence properties of chromophoric dissolved organic matter in natural waters. Limnol. Oceanogr. 1994, 39 (8), 1903−1916. (29) Kolpin, D. W.; Furlong, E. T.; Meyer, M. T.; Thurman, E. M.; Zaugg, S. D.; Barber, L. B.; Buxton, H. T. Pharmaceuticals, hormones, and other organic wastewater contaminants in US streams, 1999− 2000: A national reconnaissance. Environ. Sci. Technol. 2002, 36 (6), 1202−1211.

4381 | Environ. Sci. Technol. 2012, 46, 4374−4381