Fluorescence analysis for water characterization: measurement processes, influencing factors, and data analysis

Dissolved organic matter (DOM) is a complex mixture of humic acid, carbohydrates, amino acids, proteins, other natural organic matter (NOM), and synthetic organic compounds, and is widely found in water systems (Berg et al. 2019; Shi et al. 2021). In wastewater treatment systems, DOM causes membrane fouling and the disinfection by-products (DBPs) are precursors that affect the process operation and have potential health risks, respectively (Yang et al. 2015b; Li et al. 2020; Bai et al. 2021; Liu et al. 2021; Ren et al. 2021). Water quality indexes such as the chemical oxygen demand (COD) and biochemical oxygen demand (BOD) are used to assess water quality (Sgroi et al. 2017; Li et al. 2020; Xu et al. 2020; Shi et al. 2021). However, these analyses require the pretreatment of samples and have long detection times, which means they are not suitable for the rapid characterization of organic compounds in water.

Rapid and convenient analytical methods to measure DOM in aquatic environments are required. Especially in recent years, emerging contaminants put forward higher requirements for the detection of organic pollutants in water. Optical spectroscopy methods, including ultraviolet–visible absorption spectrometry (Zhang et al. 2020b), infrared spectroscopy (Yang et al. 2015b), and fluorescence spectroscopy (Yang et al. 2015a), have been used to analyze DOM. Fluorescence analysis has good selectivity and high sensitivity, provides abundant information and does not involve the destruction of the samples. Consequently, fluorescence analysis methods are frequently used to analyze DOM in water systems (Westerhoff et al. 2001; Li et al. 2014). These methods can be used to analyze the type and concentration of DOM in water (Lee et al. 2015; Ji et al. 2018), and the conversion of DOM in the water treatment process (Rodriguez-Vidal et al. 2020). They can also be used to evaluate the water treatment efficiency, such as the removal rate of trace organic pollutants (Sgroi et al. 2017), and indicate toxicity changes and disinfection by-product variations in water (Shen et al. 2018; Chen et al. 2020; Wang et al. 2021; Xu et al. 2021; Huang et al. 2022).

Traditional two-dimensional (2D) fluorescence analysis techniques, such as fluorescence excitation spectroscopy, fluorescence emission spectroscopy, and synchronous fluorescence spectroscopy, are mostly used to identify and characterize one or more substances (Bruckman et al. 2012; Foudeil et al. 2015). Excitation–emission matrix (EEM) analysis methods, including the fluorescence index (FI), fluorescence regional integration (FRI), and parallel factor analysis (PARAFAC) (Sgroi et al. 2017; Li et al. 2020), can be used to trace the sources of pollutants, evaluate the effects of water treatment processes, and monitor pollutants. These methods have been applied to water quality analysis of rivers and lakes (Patel-Sorrentino et al. 2002; Zhang et al. 2020a), groundwater (Vera et al. 2017), drinking water (Xu et al. 2021), municipal wastewater (Li et al. 2014; Mao et al. 2021), reclaimed water, and industrial wastewater (Rodriguez-Vidal et al. 2020; Islam et al. 2021). However, there are still many uncertainties about the process of fluorescence data analysis. High sample concentrations can exceed the analytical range, and extraction of useful information from complex multi-dimensional fluorescence data is difficult. In addition, water quality parameters and interactions between different components in the samples can affect fluorescence analysis.

In this paper, we aimed to (1) introduce the mechanism of fluorescence methods and the development of its application in water systems, (2) review 2D fluorescence and excitation–emission fluorescence methods and the application scope, and (3) identify and analyze the errors in fluorescence analysis to establish a rigorous fluorescence analysis process.

A fluorescence excitation spectrum is a plot of the excitation wavelength versus the fluorescence intensity obtained with a fixed emission wavelength (Figure 4(a)). Fluorescence excitation spectra show the fluorescence quantum yield at different excitation wavelengths. The peak position can be selected according to the sample's excitation spectrum when determining the sample's concentration or composition (Wakebe & Van Keuren 1999; Ma et al. 2011; Bruckman et al. 2012). The fluorescence excitation spectrum can be used to analyze the fluorescence characteristics of specific substances such as xanthene dyes (Wakebe & Van Keuren 1999), erythrosine (Ma et al. 2011), NOM (Abbt-Braun & Frimmel 1999), and fulvic acid from groundwater (Kumke et al. 1999). The process of absorption of energy by fluorescent substances is an excitation process, which means that the excitation and absorption spectra will have similar shapes, but the fluorescence spectrum is not completely equivalent to the absorption spectrum.

For the fluorescence emission spectrum, the fluorescence intensity is scanned under different emission wavelengths at a fixed excitation wavelength to obtain the relationship between fluorescence intensity and the emission wavelength (Figure 4(b)). Substances can be identified using fluorescence emission spectra. The emission spectrum of a fluorescent substance will show a Stokes shift (Reynolds 2003), where the emission and absorption spectra have the same shape but the emission wavelength will be greater than the excitation wavelength. This occurs because the molecule loses some energy through internal conversion and vibration relaxation when it reaches the first excited singlet state (or second excited singlet state) and returns to the ground state. The spacing of vibrational energy levels in the ground state is related to the shape of the emission spectrum. The ground state molecules will be in different vibrational energy levels after excitation, which causes the first absorption state in the absorption spectrum, and the ground state is always at the lowest vibrational energy level. Therefore, the shape of the first absorption band is related to the distribution of vibrational energy levels after excitation. According to the Franck–Condon principle (Stuhlmann et al. 2014), the probability of reaching a certain vibrational level is high during the excitation transition, and the probability of reaching the same vibrational level is also high during the radiative transition. Therefore, the fluorescence emission spectrum and absorption spectrum are mirror images. The independence of the emission spectrum from the excitation wavelength is mainly determined by the energy difference between the lowest vibrational energy level of the first excited singlet state and each vibrational energy level of the ground state.

Fluorescence emission spectra have been used to identify substances (Li et al. 1996; Zsolnay et al. 1999), analyze the fluorescence of NOM (Chen et al. 2003a) and DOM (Lombardi & Jardim 1999; Wei et al. 2005), and study the interactions of other factors with organic matter in the environment (Lu & Jaffe 2001; Gadad & Nanny 2008).

The main types of SFS are constant-wavelength SFS (CWSFS), constant-energy SFS (CESFS), variable-angle SFS, and matrix isopotential SFS. In CWSFS, the excitation wavelength and emission wavelength are kept at a constant interval (Δλ = λ_em − λ_ex) during fluorescence scanning (Figure 5(a)) (Lloyd 1971). CESFS was proposed by Inman & Winefordner (1982) as a method to keep a constant excitation energy and emission monochromator during wavelength scanning (Figure 5(b)). CESFS is based on the specific vibration energy of the molecule. If the selected constant energy difference is equal to the vibrational energy difference, and the excitation energy and emission energy are within the vibrational energy difference, a synchronous spectrum with the maximum intensity is generated. Compared with CWSFS, CESFS can effectively overcome the Raman scattering and improve the analytical sensitivity (Andrade-Eiroa et al. 2010). Compared with variable angle SFS, constant energy, and constant wavelength scans are performed with constant separation between emission and excitation beams (Figure 5(c)). The spectrum will be a straight line with a slope of one and the axes in nanometers or reciprocal centimeters. The scan path of variable angle SFS is a straight line with varying slopes (Andrade-Eiroa et al. 2010). The nonlinear scan path is a broken line or curve in the EEM diagram. Cabaniss (1991) distinguished variable angle SFS by the slope and intercept, which are defined by wavelength and energy. Matrix isopotential SFS was proposed by Pulgarin & Molina (1994) as a nonlinear variable angle SFS method. In this method, the fluorescence intensity line of the matrix can be used to eliminate the influence of the matrix.

SFS is a well-known and well-established technology that has been widely used for the analysis of water and wastewater samples (Pullin & Cabaniss 1995; Reynolds 2003; Andrade-Eiroa et al. 2010). It has been applied to the detection of organic substances such as polycyclic aromatic hydrocarbons (PAHs) and pesticides in water (Rodriguez & Sanz 2000; Foudeil et al. 2015), the determination of humic and fulvic acids (Peuravuori et al. 2002), wastewater fingerprinting (Wu et al. 2006), and analysis of water toxicity (Hanh et al. 2009). Although SFS has been widely used in various fields, its application is limited by the internal filtration effect, fluorescence quenching, Raman and Rayleigh scattering, and the strong superposition effect of fluorescence signals from multi-contaminant matrix samples.

Compared with 2D fluorescence, the main difference with EEM is that changes in the fluorescence intensity information for the excitation and emission can be obtained simultaneously. Contour fluorescence spectrograms or 3D projections are commonly used to display EEM data. Coble et al. (1990) reported the first application of an EEM to analyze the fluorescence characteristics of DOM in the Black Sea. Currently, EEM is the main method used to characterize the source, composition, and other information of DOM. However, EEMs containing multi-dimensional information are often difficult to analyze to achieve the research objectives (Sgroi et al. 2017; Li et al. 2020). The three simplified methods for EEM data analysis are the FI, FRI, and PARAFAC.

The FI is the ratio of fluorescence intensities at emission wavelengths of 470 and 520 nm (FI = F₄₇₀/F₅₂₀) when the excitation wavelength is 370 nm. This method can be used to distinguish the source of DOM because an FI > 1.9 indicates strong microbial action, an FI < 1.4 indicates the DOM is from terrestrial or soil sources, and an FI of 1.4–1.9 indicates that the DOM is from a combination of terrestrial and autogenous sources (McKnight et al. 2001).

The biological index (BIX) and humification index (HIX) have also been developed as indicators of DOM (Rodriguez-Vidal et al. 2020). The BIX is calculated as the ratio of fluorescence intensities with emission wavelengths of 380 and 430 nm (BIX = F₃₈₀/F₄₃₀) at an excitation wavelength of 245 nm. A BIX > 1 indicates that biological sources are mainly influenced by organisms, and a BIX of 0.6–0.7 indicates that terrestrial input or human activities greatly affect biological sources (Huguet et al. 2009). The HIX is the ratio of the average light intensity in the range of 435–480 nm and 300–345 nm when the excitation wavelength is 245 nm. The humification degree increases with the HIX (Morling et al. 2017).

The FI method has been widely used to characterize organic matter sources in water research (Table 1). The FI, BIX, and HIX have been used to distinguish the sources of DOM (microbial or terrestrial) (Rodriguez-Vidal et al. 2020), colored DOM in wastewater (Clark et al. 2020), and DOM in lakes (Carstea et al. 2014; Zhang et al. 2020b). The FI has also been used to evaluate the water quality indicators of COD, 5-day BOD, total nitrogen, and ammoniacal nitrogen (Zhang et al. 2020a). Moreover, it has been shown that the FI can be used to evaluate emerging contaminants, such as DBPs, and their potential for formation (Yang et al. 2015b; Shen et al. 2018; Xu et al. 2021).

An EEM contains tens of thousands of excitation–emission wavelength-dependent fluorescence intensity dots. The volume of data makes analysis difficult. To address this, Chen et al. (2003b) proposed dividing the EEM into five EEM regions using a method named FRI. FRI uses vertical or horizontal lines to divide the EEM fluorescence contour spectra into the following regions: I, aromatic proteins class I (tyrosine-like substances); II, aromatic proteins class II (tryptophan-like substances); III, fulvic acid-like substances; IV, soluble microbial by-product-like substances; and V, humic acid-like substances. It is worth noting that the EEM regions determined by the FRI method are constant, and the material itself or external conditions only slightly affect the DOM position (Li et al. 2020). As a semi-quantitative method, the fluorescence flux of different regions can be calculated by the volume integral to obtaining two indexes. These indexes are the volume (Φ_{i, n} for each region, Φ_{T, n} for the whole region) and percent fluorescence response (P_{i, n}), and can be used to evaluate differences between samples or the change for a substance within a single sample (Li et al. 2020).

Because FRI analysis is simple and stable, it has been widely applied to investigate the composition and variation of DOM in water and wastewater (Table 1). FRI is often used to track and determine the fate of organic matter during water treatment (Sgroi et al. 2017). FRI has also been used to evaluate the performance of processes at treatment facilities, such as ultrafiltration, biological activated carbon, ultraviolet advanced oxidation, and membrane bioreactors (Vera et al. 2017; Shen et al. 2018; Lin et al. 2019; Huang et al. 2022). Similar to the FI method, FRI can be used to indicate emerging contaminants (Sgroi et al. 2017) such as pharmaceuticals and personal care products (Yadav et al. 2019), DBPs (Trueman et al. 2016; Shen et al. 2018; Fan et al. 2020), nitrogenous DBPs (Tan et al. 2017; Lin et al. 2019), and toxic compounds (Chen et al. 2020; Huang et al. 2022) in addition to detecting conventional indicators of water quality (Wang et al. 2022). Compared with the FI method, FRI is beneficial for application to substances with similar structures.

Several representative components are usually obtained by EEM PARAFAC analysis of samples in aquatic environments (Table 1). These components can be used instead of conventional water quality indicators, such as the COD, BOD₅, ammoniacal nitrogen, and total phosphorus (Zhang et al. 2020a), and emerging contaminants, such as DBPs, nitrogenous DBPs, N-nitrosamines, and pharmaceuticals and personal care products (Yang et al. 2015b; Sgroi et al. 2017; Yadav et al. 2019; Maqbool et al. 2020; Xu et al. 2021). These components can also be used to assess the effectiveness of water treatment technologies and the fate of organic matter in water treatment processes (Sanchez et al. 2013; Aftab & Hur 2017; Sgroi et al. 2017; Rodriguez-Vidal et al. 2020).

The fluorescence produced by a substance is related to the substance's structure; however, equipment, operational parameters, and environmental factors can affect the shape and size of the fluorescence spectrum. Therefore, the sensitivity and selectivity of fluorescence analysis can be improved by selecting appropriate operating conditions.

In all fluorescence measurement methods, instrument calibration is a necessary step to correct instrument biases. Instrument calibration includes the signal intensity and spectral shape. For example, for DOM measurement, the fluorescent molecules or fluorophores emitting fluorescence are unknown, so it is necessary to calibrate the signal intensity to compare different samples. The most common calibration method involves measuring quinine sulfate at an excitation wavelength of 350 nm and emission wavelength of 450 nm (Coble et al. 1993). The Raman signal of pure water can also be used for correction (Determann et al. 1994; Stedmon et al. 2003). Spectral correction takes into account the spectral output deviation of the instrument light source and biases of the instrument light transmission components. With the development of instrument technology, most currently available fluorescence measuring devices have a built-in reference detector to correct the optical signal for the light source spectrum. Residual deviations can be corrected periodically with rhodamine-B or rhodamine-101 (Karstens & Kobs 1980). Emission spectra also need to be corrected, but they generally show little variation and correction can generally be performed by changing the instrument operation manually.

Scattered light, such as that from Rayleigh and Raman scattering, has a large influence on fluorescence measurement. Molecular absorption of low photon energy can only excite electrons in the molecule to other higher vibrational levels of the ground state, but not to a higher excited state. In Rayleigh scattering, an excited electron that loses no energy quickly returns to its original ground state and emits light in which the radiation wavelength is the same as the excitation wavelength. In other vibrational levels of the ground state, the electrons in the molecule do not return to the initial ground state, but return to a higher or lower level than the original vibrational level. At this time, the wavelength of radiated light is longer or shorter than the excitation wavelength, and the radiated light is Raman scattered. Therefore, the occurrence of Rayleigh and Raman scattering in fluorescence measurements limits the sensitivity and reliability of fluorescence analysis. Emission gratings with a double monochromator and a cut-off filter in the measuring instrument can attenuate the effect of scattered light (Murphy et al. 2013). Raman scattering can be removed by deducting the pure water spectrum from the sample spectrum (Stedmon & Bro 2008). In the Rayleigh scatter-affected region, scattered signals are often treated as missing data to remove the effects of scattered light (Christensen et al. 2003; Stedmon & Bro 2008), and a blank fluorescence spectrum is used for subsequent fluorescence modeling and analysis. The above problems can be avoided by inserting zeros outside of the data region (Rinnan et al. 2005). The smootheem functions in the drEEM toolbox in MATLAB can be used to remove scattered signals or interpolate to remove Rayleigh and Raman scattering (Murphy et al. 2013).

Because of the absorption of excited or emitted fluorescent photons by the sample matrix, the inner filter effect (IFE) reduces the fluorescence yield. The main reason for the IFE is that some chromophores in the sample absorb photons at the same wavelength as the target molecule (Chen et al. 2018). There are two types of IFE: primary IFE involves the absorption of excited photons by the fluorophores and chromophores; and secondary IFE involves the absorption of photons emitted by fluorophores and chromophores (Ohno 2002). IFE will distort the fluorescence spectrum and adversely affect the fluorescence analysis of substances. Generally, analysis of samples with low concentrations is considered to have a low IFE. However, the IFE should be corrected for accurate analysis of samples at high concentrations (Panigrahi & Mishra 2019). The IFE calibration methods include instrument calibration, parameter correction, and mathematical correction (Khan et al. 2022). Instrument calibration is accomplished by instrument theory, parameter correction uses absorbance, optical density, and other parameters for correction, and mathematical correction is used for subsequent fluorescence data analysis (Ohno 2002; Chen et al. 2018).

The emission fluorescence characteristics of some molecules with acidic or fluorophores can be considered as two types. A variation in pH will change the proportions of two fluorescent molecules, which will affect the position and shape of the fluorescence spectrum and the fluorescence intensity. For example, a study of the fluorescence spectra of humic substances from the International Humic Substances Society showed that the fluorescence spectra of samples redshifted with an increase in pH (Pullin & Cabaniss 1995). This shift was attributed to a change in the fluorescence characteristics of acidic functional groups in humic substances (Mobed et al. 1996). Moreover, it has been verified that the fluorescence intensity increases with increasing pH. This has been observed in investigations of organic matter in river water and the detection of extracellular organic matter under different pH conditions (Patel-Sorrentino et al. 2002; Sheng & Yu 2006). When the pH changes from neutral to alkaline, the fluorescence intensity of treated wastewater decreases by 30–40% (Westerhoff et al. 2001). Similarly, salinity can also change the position and shape of the fluorescence spectrum and the fluorescence intensity by changing the molecular characteristics of fluorescent molecules (Khan et al. 2022). For example, one study found that the fluorescence intensity increased by approximately 10% when the pH was increased from 7.0 to 8.5 and the salinity was doubled (Esteves et al. 1999).

Increases in the temperature have an adverse effect on fluorescence because the probability of collisions between molecules increases, which increases the probability of de-excitation (Carstea et al. 2014; Lee et al. 2015). The fluorescence intensities of humic substances from the International Humic Substances Society, tryptophan standard, river water, and sewage were quenched when the temperature was increased from 10 °C to 45 °C (Baker 2005). Elliott et al. (2006) observed that the fluorescence decreased by more than 40% as the temperature increased. The degree of thermal quenching depends on the type of water because the fluorophores in water come from different sources. Terrestrial humic-like components in rural water samples cause higher quenching than those in urban water samples (Carstea et al. 2014). Thus, it is necessary to correct the influence of temperature on the fluorescence spectrum in DOM fluorescence analysis. Temperature correction tools based on sequential mathematical correction methodology have been used to address the effect of temperature on fluorescence measurements (Watras et al. 2011; Goffin et al. 2020).

Metal ions can chelate with DOM or precipitate DOM to enhance or quench fluorescence. There are two types of DOM fluorescence quenching by metal ions. The first, which is called dynamic quenching, involves the formation of stable non-fluorescent complexes between metal ions and DOM fluorescence binding sites. The second, static quenching, involves the formation of metal complexes that have only partial fluorescence because of the inherent chemical heterogeneity of DOM (i.e., different chemical structures and/or similar chemical structures in different environments) (da Silva et al. 1998; Yamashita & Jaffe 2008). Static quenching decreases with temperature increases, whereas dynamic quenching increases with increases in temperature (Khan et al. 2022). In dynamic quenching, non-fluorescent metal complexes deactivate excited molecules through intermolecular or intramolecular collisions (Lakowicz 2006). As the temperature increases, it is difficult for the molecules in the ground state to enter the excited state, which weakens the fluorescence signal. Paramagnetic metal ions (Cu²⁺, Fe³⁺, Hg²⁺, Ni²⁺, and Zn²⁺) can quench the fluorescence of humic-like substances but not protein-like substances (Provenzano et al. 2004; Henderson et al. 2009). Diamagnetic metal ions (Al³⁺, Mg²⁺, Ca²⁺, and Cd²⁺) show different effects on fluorescence (i.e., enhancement, quenching, or minimal effect) (Elkins & Nelson 2002). Tryptophan-like fluorescence is quenched by Cu²⁺, Ni²⁺, Fe³⁺, and Mo³⁺, but not by Mn²⁺, Co²⁺, Ca²⁺, Zn³⁺, Cr³⁺, and Na⁺ (Henderson et al. 2009). Different concentrations of metal ions and fluorophores, pH values, and temperatures in various systems lead to different levels of fluorescence quenching (Henderson et al. 2009; Pan et al. 2012). Therefore, for fluorescence measurements in complex matrices such as sewage, the influence of different components should be considered.

The solvent effect is another important factor affecting fluorescence analysis (Mataga et al. 1955). Under the influence of the dielectric constant and refractive index of a solution, the fluorescence intensity and maximum wavelength of the fluorescence spectrum are generally affected by the solvent. Hydrogen bonding between the solvent and fluorescent molecules can also cause the solvent effect (Siqintuya et al. 2005). The polarity of the solvent is the main contributor to the solvent effect on the fluorescence properties of a substance (Xie et al. 2004; Qiu et al. 2010). The complexation of ground and excited molecules and polar molecules enhances fluorescence (Rechthaler & Kohler 1994). The fluorescence spectra of quinolone antibiotics in different solvents showed that the dielectric effect of the solvent was stronger than that of specific hydrogen bonding (Park et al. 2002). Therefore, the solvent effect should be considered when fluorescence technology is used to identify certain substances or analyze the removal of characteristic pollutants.

Fluorescence analysis has been widely used to characterize DOM in water and wastewater and used to trace pollutants, assess the performance of water treatment technologies, and indicator characteristic pollutants including emerging contaminants (e.g., drugs and DBPs).

There are two main categories of fluorescence analysis methods: 2D methods, such as fluorescence excitation and emission spectroscopy and synchronous fluorescence spectroscopy; and EEM methods of the FI, FRI, and PARAFAC. The 2D analysis methods are mainly limited to the identification of one or several pollutants. By contrast, EEM analysis methods provide more complex fluorescence information and can be used to analyze a variety of pollutants. Consequently, EEM analysis has become the most widely used technology.

Moreover, this review evaluated a variety of methods to reduce the error in fluorescence analysis and developed a flow chart for fluorescence analysis. The error in fluorescence analysis is affected by two types of parameters: operation parameters, including the instrument calibration, cuvette selection, and removal of Raman and Rayleigh scattering; and water quality parameters, including the IFE, pH, temperature, metal ions, and solvent effect. This analysis flow chart will support future research on improving the accuracy of fluorescence analysis.

Although fluorescence technology has been widely used in water environments, the following issues are required to be further studied and analyzed in the future:

• 1.

  There is a need to conduct a series of studies to explore the relationship of emerging contaminants with fluorescence. With the continuous discovery of emerging pollutants, there is an increasing emphasis on their rapid characterization. Therefore, the exploration of fluorescence characteristics of pollutants is conducive to the rapid and convenient analysis of water quality.

• 2.

  Further studies are needed to analyze the influence limits of temperature, pH, metal ions, internal filtration effect, and other environmental factors, and to develop a more complete and standard fluorescence analysis process to obtain accurate fluorescence data.

• 3.

  Online water quality detection has become an important form of water quality analysis and display for drinking water treatment and supply, and wastewater treatment and discharge. Fluorescence has great advantages as a fast and sensitive method for the detection of organic matter. Furthermore, the correlation of fluorescence signals and conventional indicators such as COD and BOD, and emerging contaminants such as antibiotics, endocrine disruptors, and DBPs.

This study was supported by the National Natural Science Foundation of China (No. 52000115), Tsinghua Shenzhen International Graduate School (HW2021012/QD2021010C), Shenzhen Science, Technology and Innovation Commission (No. JCYJ20200109142829123), and Guangdong Basic and Applied Basic Research Foundation (2020A1515110106).

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.