Estimation of biochemical oxygen demand based on dissolved organic carbon , UV absorption and fluorescence measurements

Determination of 5-d biochemical oxygen demand (BOD5) is most commonly practiced test to assess the water quality of surface waters and the waste loading. However, BOD5 is not a good parameter for the control of water or wastewater treatment processes because of its long test period. It is very difficult to produce consistent and reliable BOD5 results without using careful laboratory quality control practices. This study was performed to develop software sensors to predict the BOD5 of river water and wastewater. The software sensors were based on the multiple-regression analysis using the dissolved organic carbon (DOC) concentration, UV light absorbance at 254nm and synchronous fluorescence spectra. River water samples and wastewater treatment plant (WWTP) effluents were collected at 1-hour interval to evaluate the feasibility of the software sensors. In short, the software sensors developed in this study could well predict the BOD5 of river water (r = 0.78) and for the WWTP effluent (r = 0.90).


Introduction
e determination of 5-d biochemical oxygen demand (BOD 5 ) is the standardized experimental procedure to determine the relative oxygen requirements for aqueous microbes to consume organic materials in wastewaters, wastewater treatment plant (WWTP) effluent, or natural waters [1].BOD 5 has been used as an indicator for the amount of organic pollutants in most aquatic systems, especially a good indicator for biodegradable organic compounds [2].Due to the 5-d test period, however, BOD 5 is not considered as a suitable parameter for a process control of water treatment processes and for a real-time water quality monitoring system, in which a rapid feedback is essential [3].e BOD 5based biodegradation test that relies upon the presence of a viable microbial community has a difficulty in consistently acquiring accurate measurements [4].BOD 5 generally has an uncertainty of 15%∼20%.
In order to overcome the shortcoming of the conventional BOD 5 test, biosensors, UV-visible spectrophotometry, �uorescence measurements, and soware sensor (virtual sensors) have been suggested as an alternative method to determine the BOD 5 of a water sample.
Most BOD 5 biosensors rely on the measurement of the respiratory activity of cells by a suitable transducer.In addition, the ones using an oxygen electrode, a carbon dioxide analyzer, an optical transducer or a microbial fuel cell have recently been reported [5].Biosensors allow the researchers to conveniently and rapidly (15 minute) obtain the BOD 5 result, compared with the official BOD 5 method [6].
Single or mixed cultures are used in biosensors.Since a single strain is not able to oxidize the entire range of organic contaminants in water samples, and the DO consumption is, thus, not always directly proportional to the concentration of biodegradable organics, mixed cultures like activated sludge have been preferred [7].Even in the case of mixed cultures, however, the activity of the microbes is easily affected by the changes of environmental condition, such as concentrations of nutrients, temperature, and pH resulting in inaccurate BOD 5 values [8].
Dissolved organic compounds with aromatic structures strongly absorb UV radiation [9].Based on the principle, the UV-visible spectrophotometry of a water sample is hypothesized to have a linear relation with water total organic carbon (TOC), nitrate, suspended solids (SSs), chemical oxygen demand (COD), BOD 5 or dissolved organic carbon (DOC) [10].Alternatively, the UV light absorbance at 254 nm has been utilized to directly estimate the aggregate organic content of a water sample [11].If this approach is to be applied for the BOD 5 determination, target water samples should not contain other light-absorbing chemicals or materials like nitrate or SS [3].
Fluorescence measurements have been applied to determine the presence of humic substances and organic matters in natural waters.Among a few �uorescence analysis methods, synchronous �uorescence spectroscopy is the best way to scan the entire section of excitation wavelengths by �xing the excitation and emission wavelengths uniformly.is method allows obtaining a better resolution and producing various information regarding the DOMs in water [12].Recently, the method has been successfully applied to identify microbial communities in water and to establish the correlation between water BOD 5 and the microbial activity [4,13].Since the BOD 5 test is a microbial assessment of organic substance load, the "microbial" tryptophan-like �uorescence was found correlated with the activity of a microbial community and the absolute BOD 5 values of water samples [4].e optical parameters of tryptophanlike �uorescence use diverse speci�c excitation/emission wavelengths: for example, 248 nm/340 nm, 280 nm/350 nm, 220-230 nm/340-370 nm, 220 nm/350 nm, 280 nm/350 nm, and so forth.However, the BOD 5 determination based on the �uorescence peaks obtained from water samples is still infancy.In order to estimate the BOD 5 , however, more information should be obtained in addition to the tryptophan-like �uorescence, since real environmental water contains other oxidizable minerals and carbohydrates as well as biodegradable organic matter [14].Even the water collected near the discharge of an industrial wastewater treatment plant (WWTP) may contain toxic substances such as heavy metals that can inhibit the oxidation of organic compounds by bacteria [15].Moreover, the water �uorescence is oen affected by water pH, temperature, and SS.In fact, the approach has been applied only to wastewater samples, the BOD 5 of which varies wide [2].
Recently, a few researchers have utilized both UV-visible spectrophotometry and �uorescence measurements together to estimate the BOD 5 in waters.Applying the sensors to the environmental monitoring is advantageous since they are rapid and versatile.In addition, they require low operating costs, no chemicals, and no sample pretreatment for measurements.However, their application to water samples can be very limited if the SS concentration of the samples is high [16].
A soware sensor (in other words, virtual sensors) generates virtual signals for the water quality parameter of interest through the calculation of a model fed with real signals from reliable, available sensors for other parameters [17].It rapidly predicts the effect of changes in other water quality parameters on the target parameter.Since the soware sensor does not obtain its result from physical measurements, however, the uncertainty associated with its result can be large [3].Hence, it has been suggested that a soware sensor should generate its data based on signals from as many real relevant sensors as possible.
In this study, the synchronous �uorescence spectra, the UV light absorbance at 254 nm, and DOC of water samples were analyzed to predict their BOD 5 values.Since the �uorescence spectra vary depending on the characteristics of DOMs that are site speci�c in this study, therefore, all the �uorescence spectra of a sample were utilized.
e speci�c purposes of this study are as follows: (1) the analysis of synchronous �uorescence spectra of river waters and wastewaters, (2) correlation analyses between water BOD 5 and organic parameters (i.e., DOC) and between water BOD 5 and optical parameters (i.e., UV light absorbance at 254 nm, synchronous �uorescence spectra (at 270∼300 nm, 310∼370 nm, 370∼400 nm, and 400∼530 nm)), and (3) development of multiple regression models for the BOD 5 prediction using DOC, UV absorbance at 254 nm, and synchronous �uorescence spectra.

Sampling Locations and Sample Pretreatment.
A total of 23 river samples were collected from the Gyeong-An River which �ows through the City of �ong In, Korea, at 1-h intervals.In addition, a total of wastewater samples were collected from the Hwa-Do WWTP in the City of Namyangju, Korea, at 1-h intervals.e river samples contained low concentrations of BOD 5 , while the wastewater samples contained a wider range of BOD 5 .Once the samples were collected, they were stored under refrigerated condition (4 ○ C) and transported to the laboratory, in which they were analyzed immediately.Using prewashed GF/F �lters (Whatman, USA; nominal pore size: 0.7 m), SS was removed from the samples.SS in water samples oen interferes accurate measurements of water quality by scattering light, when the UV spectrophotometry or the synchronous �uorescence measurement is applied to the water samples.

Analytical
Methods.e dissolved organic matters in the �ltered samples were measured with an UV spectrophotometer (Shimadzu UV-1800) at 254 nm.e BOD 5 of each sample was determined by calculating the decreased amount of the DO over 5 d (APHA, 2010).e DOC of the samples was calculated by subtracting dissolved inorganic carbon (DIC) from dissolved carbon (DC).e DIC and DC were measured using a TOC analyzer (Shimadzu TOC-V-CPH, Japan).DC concentration of a water sample was determined by combusting a water sample at 680 ○ C in the presence of a platinized alumina catalyst and by measuring the resulting CO 2 production.On the other hand, the DIC of a water sample was determined through the phosphoric acid digestion of the water followed by the determination of the CO 2 production.
e �uorescence spectra of a water sample were measured using a �uorescence spectrometer (Scinco FS-2, �orea).For each sample, synchronous �uorescence spectra for excitation wavelengths ranging from 200 to 600 nm were recorded using a constant offset (i.e., Δ = 30 nm). e excitation and emission slits were adjusted to 5 nm and 5 nm, respectively.Blank spectrum made by deionized water was subtracted from those of each sample to remove the Raman scattering.e UV-visible spectrum of samples was measured using a UV spectrophotometer (Shimadzu UV-1800).e operating conditions of the spectrophotometer are as follows: a resolution of 2 nm, a response of 0.5 s and a scan speed of 60 nm min −1 .

Development of Multiple Regression
Models.e correlation coefficients between the BOD 5 of water samples and the UV light absorbance at 254 nm and between the BOD 5 and �uorescence spectra were analyzed using the correlation function of Microso Excel (Microso, USA).e multiple regression with the parameters (i.e., DOC, UV absorbance, and �uorescence spectra) for the development of a model to predict the water BOD 5 were carried out using the Data Analysis function of Microso Excel.

Measurement of Synchronous Fluorescence Spectra.
In this study, all the spectra obtained at the wavelengths of 270 nm∼300 nm, 310 nm∼370 nm, 370 nm∼400 nm, and more than 460 nm for monoaromatic compounds and tryptophan, diaromatic compounds, fulvic acid, humic acids, and other compounds, respectively, were selected as �uorescence parameters aer the synchronous �uorescence spectra of 200∼600 nm had been examined.Ferrari and Mingazzini [12] also used these spectra to analyze the compounds in natural DOMs in their study.
Figure 1 shows the average of the values measured by synchronous �uorescence spectra for a blank, 23 stream waters, and 20 wastewaters.Examining the average spectrum of sample excitation wavelength values by samples, the river waters showed peaks at the wavelengths of 310 nm and 380 nm while the wastewaters showed peaks at 320 nm and 400 nm.A peak at the wavelength of 310∼320 nm appeared common for all the water samples including the blank.However, the peak occurring at the wavelength between 350∼530 nm appeared common only for river waters and wastewaters.

Correlation between BOD 5 and Fluorescence Parameters.
In order to rapidly estimate BOD 5 of a water, the UV spectra [14], the optical scattering (i.e., �uorescence) [18], the UV light absorption at 280 nm [16], COD [15], and so forth, were utilized.ese parameters are divided into organic material parameters (e.g., COD, etc.) and optical parameters (e.g., UV light absorbance, �uorescence spectra, etc.).In fact, none of the parameters has been able to successfully predict the BOD 5 of water samples perfectly.omas et al. [14] suggested that organic matter be classi�ed into BOD 5 , COD, TOC, and substances absorbing UV light.e BOD 5 is related to oxidizable minerals, carbohydrates, and biodegradable organic matters.e COD is related to oxidizable minerals, carbohydrates, biodegradable organic matters, and humic substances.e TOC is related to carbohydrates, biodegradable organic matters, humic substances, aromatic hydrocarbons, and aliphatic hydrocarbons.Lastly, the UV light absorption is related to biodegradable organic matters, humic substances, aromatic hydrocarbons, and UV light-absorbing minerals.
Since each of COD, TOC, and the UV absorption only identi�es some of the organic matter related to BOD 5 , the predicted BOD 5 values based on the parameter should be erroneous.erefore, this study used both organic parameter and optical parameters to improve the accuracy of the BOD 5 estimation: DOC as the organic parameter, UV absorbance at 254 nm, and �uorescence spectra at 270∼300 nm, at 310∼ 370 nm, at 370∼400 nm, and at 400∼530 nm as the optical parameters (Table 1).e river waters contained 1.3∼ 1.9 mg L −1 of BOD 5 and 1.6∼2.5 mg L −1 of DOC.In addition, 0.036∼0.64 of the UV absorbance at 254 nm and 28252∼ 84575 of �uorescence spectrum intensity at 270∼300 nm, 381099∼595808 at 310∼370 nm, 158019∼254822 at 370∼ 400 nm, and 188096∼461216 at 400∼530 nm were observed for the water.

Multiple Linear Regression Analysis.
Multivariate relationships require a multiple regression analysis involving several explanatory variables for predictors of theoretical interest and control variables [18].Oen multivariate regression is applied to predict a variable (i.e., predictor or dependent variable), which is not easily measurable, with other variables (i.e., independent variables), which are easy to measure.For examples, COD, NH 4 − , and NO 3 − concentrations of water samples were predicted by using pH, temp, conductivity, redox potential DO, and turbidity of the same water [19].Helling et al. [20] predicted the COD/TOC ratio using CO 2 and O 2 .Lee and Ahn [21]  To select an appropriate regression model for the BOD 5 prediction, a total of seven models were developed for river waters and wastewaters and presented in Table 3. From Table 3(a), the Model 6 was found to be the most appropriate in predicting the BOD 5 of river waters since it slowly increased  2  and made  2 adj the maximum.In fact, the Model 7 appeared equivalently appropriate since the MSE of the model was the minimum and it made   of Mallows closest to  + 1.However, Model 6 was �nally selected for predicting the BOD 5 of river waters since it involves fewer variables.By the same token, a total of seven linear regression models were developed to predict the BOD 5 of wastewaters, and Model 7 was selected as the most appropriate model aer reviewing the result of analyzing each linear regression models (Table 3

(b)).
In Table 4, the Model 6 in Table 3(a), a linear regression for predicting the BOD 5 of river waters was provided.e input variables for the model are the UV absorbance at 254 nm and �uorescence intensities at 270∼300 nm, at 310∼ regression model for wastewaters (i.e., Model 7 in Table 3(b)) was also provided in Table 4. Its input variables are DOC, the UV absorbance at 254 nm, and �uorescence intensities at 270∼300 nm, at 310∼370 nm, at 370∼400 nm, and at 400∼ 530 nm.

Validation of Developed Linear Regression Models for
Predicting BOD 5 .e developed multiple regression models for river waters and wastewaters were, respectively, applied to predict the BOD 5 of different sets of river water and wastewater samples.e model predictions were then compared with manual measurements (Figure 4).As shown in the �gure, the developed multiple regression models could reasonably well predict the BOD 5 of the target water samples.e coefficient for the correlation between manually measured BOD 5 and model prediction was calculated 0.78 for river waters, while that for wastewaters was 0.90.e relative lower correlation coefficient for river waters was attributed to the fact that the concentration was within the range from 1.3 mg L −1 to 1.9 mg L −1 ; the BOD 5 range for the wastewater samples was 6.5 mg L −1 ∼139.9 mg L −1 .

Conclusion
In this study, two multiple regression models were developed to predict the BOD 5 of two types of environmental waters: one for river waters and the other for wastewaters.e model for river waters predicts BOD 5 using the data of the UV absorbance at 254 nm and �uorescence intensities at 270∼ 300 nm, at 310∼370 nm, at 370∼400 nm, and at 400∼530 nm.e model for wastewaters was utilizing the data of DOC, the UV absorbance at 254 nm, and �uorescence intensities at 270∼300 nm, at 310∼370 nm, at 370∼400 nm, and at 400∼ 530 nm.e developed models reasonably well predicted the BOD 5 of the river waters and the wastewater samples; correlation coefficients between the model-predicted and manually measured BODs were 0.78 for the river waters and 0.90 for wastewaters.
In fact, the data used for predicting the BOD 5 of two types of water samples can be measured using an on-line optical sensors.erefore, if the BOD 5 is estimated using the approach proposed in this study, its measurement can be done rapidly.In addition, this approach can be applied to develop a soware sensor for the BOD 5 measurement which can be implemented in an on-line water quality monitoring system for streams or WWTP discharges.

F 3 :
Time pro�le of BOD5 and correlation between measured BOD5 and other parameters for (a) river waters and (b) wastewaters.

F 4 :
Correlation between manually measured BOD5 and model predictions for (a) river waters and (b) wastewaters.

T 3 :
Summary of model development for predicting BOD5 of (a) river waters and (b) wastewaters.