The role of Raman spectroscopy in the analytical chemistry of potable water

Abstract

Advances in instrumentation are making Raman spectroscopy the tool of choice for an increasing number of chemical applications. For example, many recalcitrant industrial process monitoring problems have been solved in recent years with in-line Raman spectrometers. Raman is attractive for these applications for many reasons, including remote non-invasive sampling, minimal sample preparation and tolerance of water. To a lesser extent, Raman spectroscopy is beginning to play a significant role in environmental analysis for the same reasons. At present, the environmental applications typically apply only to the most contaminated situations due to the still relatively high limits of detection. However, some emerging sampling technologies hold out the promise that Raman may soon be more widely applicable to the analytical chemistry of potable water. Herein we discuss these recent advances, summarize some examples of environmental applications to aqueous systems and suggest avenues of future developments that we expect to be most useful for potable water analysis. Also, a simplified, but detailed, theory of normal Raman scattering is presented. While resonance-enhanced Raman spectroscopy, surface-enhanced Raman spectroscopy and non-linear Raman techniques are briefly discussed, their theories and instrumental configurations are not addressed. Also, this article deals primarily with the modern dispersive Raman experiment (as opposed to the Fourier transform Raman experiment), because it seems most suited for potable water analysis. The goal of this article is to give the environmental scientist with no specialized knowledge of the topic just enough theory and background to evaluate the utility of this rapidly developing analytical tool.

---

Introduction

For more than 30 years,¹ analytical chemists have been attracted to the notion of Raman spectroscopy for environmental applications, primarily because qualitative and quantitative measurements can be made directly on contaminants in water. Also, the Raman technique offers high information content, non-invasive sampling, high spatial resolution, generally narrow bandwidths, minimal sample preparation and is easily interfaced to fiber optics for remote analysis. Raman can also be applied to flowing systems; therefore, interfacing to separation techniques is usually straightforward. Considering these advantages, it is not surprising that Raman was evaluated for contaminant analysis shortly after the dawn of environmental awareness and regulation. In 1977, Long noted in his seminal Raman textbook that, ‘the application of Raman spectroscopy to measurement of pollution…is now receiving attention’.²

However, it is fair to say that, within the first 15 years after this statement was written, there was not a great deal of advancement in the application of Raman spectroscopy to environmental problems. More than in other areas, applications to environmental problems were hampered by the well-known difficulties historically associated with Raman spectroscopy. These included poor sensitivity, problems with fluorescence, long analysis times and the need for large, complex and expensive instrumental components, such as triple monochromators and water-cooled ion lasers. This generation of Raman instruments required specialized facilities and expertise for adequate operation. However, instrumentation for Raman spectroscopy has undergone tremendous advancement since the late 1980s. Although these difficulties have not been completely eliminated, their severity has been greatly reduced with the use of spectrometers incorporating modern optimized components. These include high-sensitivity, low-noise, silicon charge-coupled device array detectors (CCDs), compact, stable, high-power, near-infrared-emitting (NIR) diode lasers, efficient laser line rejection filters and compact, high-throughput, single-stage spectrographs. A modern instrument with these components allows analysis by dispersive normal Raman spectroscopy (NRS) of many contaminants with a lower limit of detection (LOD) of about 20–100 ppm in potable water with a reasonable analysis time.

Simplified theory of NRS

Suppose a cuvette containing a pure Raman-active transparent liquid is irradiated with monochromatic light from a laser at frequency ν₀. Suppose the energy of the laser (hν₀, where h is Planck's constant) is considerably less than the spacing between the ground and first excited electronic energy levels of the liquid. Most of the photons will simply pass through the filled cuvette. However, a small amount of light will be scattered in all directions. If the scattered light is collected and analyzed with a spectrograph and a CCD, intensity will be observed at the input frequency ν₀ (Rayleigh scattering) and also, to a much lesser extent, in narrow bands at other discrete frequencies. In fact, discrete bands that are at both lower and higher frequency than the laser will be observed in the collected spectrum. The lower frequency bands (ν₀ − ν_v) are termed Raman Stokes bands and the higher frequency bands (ν₀ + ν_v) are termed Raman anti-Stokes bands. This effect, by which chemicals inelastically scatter radiation, was first observed in 1928 by C. V. Raman.³ The Raman effect, while inherently weak, provided the basis of a new type of vibrational spectroscopy because the differences in energy between input and scattered light (hν_v) were found to correspond to the differences in quantized vibrational energy levels of the chemicals under study.

Suppose our pure liquid in the cuvette exhibits only one Raman-active vibration. Suppose that the CCD and the spectrograph are designed such that only radiation at frequencies lower than that of the excitation laser are collected and recorded. In the simplest case, when irradiated by the laser, scattered light from the liquid will exhibit one (Raman Stokes) band. This band will coincide with molecules of the liquid transitioning from the vibrational ground state to the first excited vibrational state. This experiment constitutes the dispersive NRS technique, and a plot of intensity from the CCD as a function of the ‘Raman shift’ constitutes the normal Raman spectrum. The Raman shift is the frequency of the excitation laser light minus the frequency of scattered light, but is generally expressed as wavenumbers in units of cm⁻¹. (Note that frequency, and also frequency differences, in units of s⁻¹, are easily converted to wavenumbers using the constant speed of light, c, in units of cm s⁻¹). If this vibrational mode from the liquid is also IR-active, and the liquid exhibits no other IR-active modes, the IR absorption spectrum of the liquid will be qualitatively equivalent to the normal Raman spectrum, i.e. a band will be observed in both spectra at the same X-axis position.

Now suppose that we dissolve the pure liquid in water at high concentration. If we again conduct the NRS experiment on the solution, we will collect a spectrum that contains the band from the analyte as well as bands from the Raman-active vibrational modes of water. If we subtract the spectrum of pure water from the spectrum of the solution, the resulting spectrum will, in the simplest case, approximately match that of the pure liquid and the intensity of the band will be linearly related to the concentration of the analyte in the aqueous solution.

Of course, most chemicals will exhibit more than one Raman-active, and more than one IR-active, vibration. However, some modes that are Raman-active are not IR-active, and vice versa. Also, vibrational bands that are intense in Raman spectra are often weak in IR absorption spectra, and vice versa. Therefore, NRS and IR absorption are complementary vibrational spectroscopic techniques. Also, there are some important advantages of NRS relative to IR absorption, many of which are germane to the analysis of potable water and are listed in the Introduction. Perhaps most notable is that the entire normal Raman spectrum of a water-solvated analyte is generally obtained with ease by spectral subtraction because the Raman scattering of water is relatively weak and the linearity and dynamic range of the CCD are quite good. This is generally not possible with IR absorption spectroscopy due, in part, to the extremely strong IR absorption bands of water.

Let us return to the case of a pure Raman-active transparent liquid that is irradiated with monochromatic light from a laser at frequency ν₀. An example energy level diagram for molecular scattering processes for this case is shown on the left-hand side of Fig. 1. It should be noted that, for this example of Rayleigh scattering, molecules are initially in the ground vibrational state (v = 0) of the ground electronic state (E₀). When irradiated by a laser at frequency ν₀ (energy hν₀), the molecules are said to be excited to a ‘virtual state’. Virtual states are not stationary quantized energy levels. Instead, they represent an electronically polarized molecular state, which is caused by the oscillating electric field of the laser beam. This virtual state instantaneously decays, resulting in the emission of light. For Rayleigh scattering, the energy of the emitted light is the same as that of the excitation beam (hν₀) and the molecule's final state is the same as its initial state (in this case, E₀, v = 0).

Fig. 1 Example energy level diagram depicting, on the left-hand side, molecular scattering processes (Rayleigh and Raman) and, on the right-hand side, a photoluminescence process (fluorescence) that is particularly problematic for practical applications of Raman spectroscopy. E₀ and E₁ are the ground and first excited electronic energy levels, respectively, for the example molecule. v = 0 and v = 1 are the ground and first excited vibrational levels, respectively, for a given electronic level, and v = 2 and v = 3 are higher vibrational energy levels.

For our example of Stokes Raman scattering, which is also depicted in Fig. 1, molecules are likewise initially in the ground vibrational state of E₀ and are then excited to the same virtual state as for Rayleigh scattering. However, the emitted light upon decay of the virtual state is of lower energy h(ν₀ − ν_v) than that of the excitation beam. Of course, the ‘lost’ energy (hν_v) is not destroyed, but is, instead, taken up by molecules in the form of vibrational motion. From the example energy level diagram in Fig. 1, we see that the final state for molecules exhibiting Stokes Raman scattering is the first excited vibrational level (v = 1) of E₀. Thus, these molecules are left with more vibrational energy than they possessed prior to illumination by the laser. While we refer to this excited vibrational state as the ‘final state’, it will actually decay, but on a time scale that is very long compared to that of the scattering process. Also, the energy will not be lost by the emission of light, but will instead be dissipated as a small amount of heat.

For anti-Stokes Raman scattering, as depicted in Fig. 1, molecules are initially in the first excited vibrational level (v = 1) of E₀. Upon irradiation, they are promoted to a virtual state that is at an energy of hν_v greater than that associated with Rayleigh (and Stokes Raman) scattering. Upon decay of the virtual state, light of energy h(ν₀ + ν_v), which is hν_v greater than that of the excitation beam, is emitted by the molecules as they return to the final state of E₀, v = 0. For anti-Stokes Raman scattering, vibrationally excited molecules lose some vibrational energy by emitting photons that are more energetic than the photons from the laser that caused the virtual state. Thus, these molecules are left with less vibrational energy than they possessed prior to illumination.

Unfortunately, molecular scattering processes are not very efficient. Only about one Rayleigh scattered photon will be emitted for every 10⁴ excitation photons. Only about one Stokes Raman scattered photon will be emitted for every 10¹⁰ excitation photons.⁴ To understand the design of a Raman spectrometer and its limitations, it is important to note that Rayleigh scattering, although inefficient, is nonetheless much stronger than Raman scattering and must be removed in order to detect the Raman scattering signal. (In fact, as we will discuss later, the detection limits of Raman spectroscopy as applied to dilute aqueous solutions are almost always set by the presence of various forms of unwanted light). Older generation Raman spectrometers dealt with the problem of Rayleigh scattering by using multistage monochromators with very high dispersion, which severely limited the throughput for Raman scattering. Also, it is important to note that Stokes Raman scattering is considerably more intense than anti-Stokes Raman scattering. As described by the Boltzmann distribution equation, the ground vibrational energy level (v = 0) is more heavily populated than a higher vibrational energy level (e.g. v = 1) near room temperature. Therefore, the number density of molecules exhibiting Stokes Raman scattering is higher than that exhibiting anti-Stokes Raman scattering. The intensity difference (Stokes vs.anti-Stokes) increases as the vibrational energy level spacing increases and is very large for high-frequency vibrations (e.g. C–H stretching). For this reason, virtually all NRS instruments and experiments involve the recording of Stokes Raman scattering only. Unless otherwise noted, reference to Raman scattering, both in the remainder of this paper and in the scientific literature in general, refers specifically to Stokes Raman scattering.

Because virtual states are not stationary, Raman (and also Rayleigh) scattering will occur in the same fashion regardless of the exact energy (or wavelength) of the laser. For example, if the energy of the laser beam depicted on the left-hand side of Fig. 1 is increased by a small amount, the virtual state for Raman scattering would simply ‘move up’ in energy by this same amount. Likewise, the energy of the scattered photons would increase by this constant amount. Therefore, the absolute frequency of scattered light changes with a change in the excitation laser wavelength, but the difference in frequency between excitation and scattered light does not. Therefore, barring any complicating factors, the normal Raman spectrum will look qualitatively the same regardless of the exact wavelength of the laser, because the spectrum is a plot of scattered light intensity vs. Raman shift. This gives the experimentalist freedom to choose from among an assortment of lasers that are suitable for Raman scattering applications. However, the choice is not easy because several complicating factors and several associated ‘trading rules’ must be considered.

In order to understand these important complicating factors and associated trading rules, we must first consider the equation for Raman scattering intensity. Generally, following the conventions of McCreery⁵ and Pelletier,⁴ the measured Raman intensity (I) can be expressed as:

I = I₀ (ν₀ − ν_v)⁴ (1/c)⁴σ dzDK (1)

where I₀ is the intensity of the excitation laser, (ν₀ − ν_v) is the frequency of the scattered light, c is the speed of light, σ is the frequency-independent cross-section for Raman scattering, dz is the effective path length and D is the number density of molecules. The constant K describes a group of experimental parameters, such as the throughput of the spectrometer. Because most chemists are more familiar with absorption spectroscopy, it is useful to compare this equation to Beer's law, which governs many absorption techniques (e.g. IR absorption spectroscopy). Beer's law defines the absorbance (A) of a solution as

A = εbC (2)

where ε is the molar absorptivity, b is the sample path length and C is the absorbing species mole concentration. Obviously, for analytes in solution, σ is comparable to ε, dz is comparable to b and D is comparable to C. [However, there are some important differences in these terms that are too detailed to address here. For a more rigorous discussion, including the proper units for the terms in eqn. (1), the reader is referred to recent books on Raman spectroscopy].^4,5

The first two terms on the right-hand side of eqn. (1) are unique to light scattering and have no analogy in absorption spectroscopy. The first term dictates that the intensity of Raman scattering scales linearly with the intensity of the excitation laser. Thus, one might propose the more laser power the better. However, there are important trading rules that complicate this situation. For example, many colored or solid samples thermally degrade if the laser power is too high. The second term dictates that the intensity of Raman scattering scales with the fourth power of the frequency of the scattered light. Because the frequency of scattered light scales with the frequency of the excitation source, one would expect a strong preference for high-frequency (i.e. low-wavelength) lasers–the higher frequency the better. However, again, complicating factors and associated trading rules come into play. In fact, perhaps the most troubling complicating factor for the practical application of Raman spectroscopy must be considered when selecting the frequency of the excitation laser. This most troubling factor is laser-induced fluorescence from the sample.

If the analyte, an impurity or some other component of the matrix exhibits fluorescence when excited by the laser, the fluorescence signal will often occur in the same frequency region as the Stokes Raman scattering. Let us refer again to Fig. 1, which is an example energy level diagram for a pure Raman-active liquid. On the left-hand side, we have depicted the processes for Rayleigh and Raman scattering. Now, suppose that the liquid is also highly fluorescent, and suppose that we illuminate the liquid with a laser whose frequency (or energy) is high enough to excite molecules from the ground (E₀) to the first excited (E₁) electronic energy level. (For many organic chemicals, this energy level spacing coincides with the energy of the blue and green laser lines that have historically been used with Raman spectroscopy). In this case, we will observe laser-induced fluorescence. An example energy level diagram for laser-induced fluorescence is displayed on the right-hand side of Fig. 1. In this example, we depict a laser photon of energy (hν′₀) required to excite molecules from E₀, v = 0 to E₁, v = 1. After vibrational relaxation to E₁, v = 0, the molecules will emit photons over a broad band of frequencies as molecules return to various vibrational levels in E₀. While the laser at energy hν′₀ will result in some photons scattered by both Raman and Rayleigh processes, it will be very difficult to discern the Stokes Raman photons because they will occur in the same frequency region as the photons due to fluorescence, which are much more prevalent. Raman scattering is a very inefficient process. At best, we can expect one Stokes Raman photon for every 10¹⁰ excitation photons. On the other hand, fluorescence, which involves excitation to a real quantized energy level, can be very efficient. In some cases, it is even possible to obtain one fluorescence photon for almost every excitation photon! Therefore, even when the analyte of interest is present in much higher concentration than a strongly fluorescing impurity, the problem from fluorescence can be overwhelming.

Over the years, there have been many strategies for recording normal Raman spectra in the presence of fluorescence when using blue or green excitation laser lines. The operational details and in-depth rationale for these strategies are beyond the scope of this paper and are discussed fully elsewhere.^4,6 However, it is instructive to consider a few of these strategies and to note that none have been completely satisfactory.

Fluorescence may occur from the analyte itself, or from some other integral component of the matrix, and is then very difficult to remove. Perhaps the most useful approach in these cases relies on the difference in time scale for the Raman scattering and fluorescence phenomena. While the virtual state associated with Raman scattering decays instantaneously, the real quantized states associated with fluorescence are longer lived. Thus, with time-resolved Raman spectroscopy,⁷ it is possible to record some Raman spectra that could not otherwise be observed with the same excitation source. However, this approach requires complex instrumentation that is beyond the reach of most analytical laboratories.

A much more manageable problem occurs when fluorescence interference arises from a low-level unwanted impurity in the matrix. For this case, it is sometimes possible to significantly reduce fluorescence by a technique known as photobleaching. Chemicals that strongly absorb laser photons of sufficient energy to cause fluorescence will often thermally degrade upon continued exposure to the high-intensity beam. Thus, if the experiment permits, the sample can sometimes be left in the path of the excitation beam until the fluorescence intensity drops to an acceptable level. Another approach that has proven to be successful in some cases is the use of granular activated charcoal or other types of solid sorbents.⁸ If the fluorescing impurity has a high affinity for the sorbent, it may be preferentially removed from the solution, resulting in a decrease in fluorescence. Quite often this is the case, because charcoal is a good sorbent for many chemicals that are prone to fluoresce when excited with green or blue excitation (e.g. highly conjugated aromatic hydrocarbons). However, a fraction of the analyte is sometimes removed from the solution via sorption to the sorbent. This sample loss may be unacceptable for many applications.

Another approach to the problem of moderately intense fluorescence has been the use of digital subtraction.⁹ This approach is sometimes successful because the fluorescence signal generally appears in the spectrum as a highly elevated and sloping baseline, while the Raman signal generally appears as sharp bands on top of that baseline. The fluorescence signal can sometimes be removed with an appropriate fitting and subtraction function, leaving an acceptable Raman spectrum. However, it is not possible to remove from the spectrum the shot noise that is inherently associated with the presence of fluorescence. Shot noise is caused by the statistical nature of light, and it increases with the square root of the number of detected photons. For example, if X number of photons per second of Raman scattering are collected in the presence of 100X number of photons per second of fluorescence, the Raman spectrum after subtraction will have approximately 10 times more shot noise than if X number of photons per second of Raman scattering were collected alone. This makes the presence of fluorescence particularly problematic because the signal-to-noise ratio (S/N) of many modern CCD-based Raman applications to aqueous systems are ultimately limited by shot noise.

This deleterious effect of fluorescence from a low-level impurity, and some strategies for its removal, are illustrated in Fig. 2. The spectrum in Fig. 2(A) is of 60 mM meta-hydroxypyridine (MHP) in deionized water collected with approximately 100 mW of power at the sample from an argon ion laser emitting at 514.5 nm. (Note that all spectra in Fig. 2 are displayed after digitally subtracting out the Raman spectrum of water. Also, note that all spectra in Fig. 2 were acquired with the same analysis time–approximately 20 min–with a modern CCD-based instrument using conditions that have been described elsewhere).¹⁰ The chemical as received from the vendor contains one or more low-level impurities that emit fluorescence when excited at 514.5 nm. The most obvious effect of fluorescence in the Fig. 2(A) spectrum is the presence of a high and sloping baseline. However, it should be noted that we only clearly see the two most intense Raman scattering bands of MHP on top of the high background. Based on previous work with a purer lot of chemical,¹⁰ we know that many more Raman bands of MHP should be visible for these experimental conditions. The reason we do not clearly see the weaker bands of MHP is that they are obscured by the high level of shot noise that accompanies the fluorescence photons. As illustrated in Fig. 2(B), the fluorescence background can be removed by digitally fitting and subtracting out the baseline; however, the shot noise remains and little is gained.

Fig. 2 Water-subtracted Raman spectra of 60 mM aqueous solutions of meta-hydroxypyridine collected with a modern CCD-based dispersive spectrometer using 100 mW of laser power at the sample and approximately 20 min data acquisition time. Spectrum A was acquired with 514.5 nm excitation prior to treating the sample with charcoal to remove fluorescing impurities. Spectrum B is spectrum A with the background digitally fitted and subtracted out. Spectrum C was acquired with 514.5 nm excitation after treatment with charcoal. Spectrum D was acquired with 785 nm excitation without removing the fluorescing impurities. Note that arbitrary units (A.U.) of intensity are equivalent for all spectra in this figure.

Fig. 2(C) is a spectrum of the same sample after it had been mixed with granular activated charcoal at the level of 10 mg ml⁻¹ of solution. After charcoal was added, the solution was shaken and left to stand for several hours (although less contact time is usually required), and the charcoal was removed via filtration. The Fig. 2(C) spectrum was then acquired with the same conditions as those used to acquire the Fig. 2(A) spectrum. A much lower level of fluorescence is now observed as a moderately elevated and sloping baseline. It appears that most (but not all) of the impurity molecules were removed via sorption to the charcoal. Indeed, the number of fluorescence photons is now low enough that the attendant shot noise does not obscure the weaker bands of MHP.

The Fig. 2(D) spectrum is included to illustrate the most recent, and most successful, approach for avoiding fluorescence with dispersive NRS. This spectrum of 60 mM MHP was acquired without charcoal treatment (or any other sample treatment) using the same spectrometer, but with 100 mW of power from a diode laser emitting at 785 nm. This spectrum is largely fluorescence free because the frequency (energy) of the laser is too low to cause electronic absorption by the sample impurities. Indeed, 785 nm is too long a wavelength (i.e. too low an energy) to cause excessive electronic absorption for most (but not all) fluorescent organic compounds. Even though the intensity of Raman scattering for 785 nm excitation is about a factor of six less than for 514.5 nm excitation (due to the near 10⁴ dependence of Raman intensity on excitation frequency), the S/N of the Fig. 2(D) spectrum is very good and is comparable to that of the Fig. 2(C) spectrum, in part because more fluorescence is avoided with 785 nm excitation. However, we cannot attribute all differences between the Fig. 2(D) spectrum and the other spectra in Fig. 2 to the avoidance of fluorescence, because the change in excitation wavelength causes other unavoidable differences as well. For example, the efficiency of a CCD is dependent on the frequency of the light that strikes it. Also, this particular CCD incorporates optics that are anti-reflection coated for near-IR optimization. Furthermore, the diode laser exhibits less noise than the argon ion laser. Overall, these factors favor 785 nm excitation for the Raman shift region displayed in Fig. 2, but probably not enough to offset the factor of approximately six intensity advantage of 514.5 nm excitation. All things considered, 785 nm excitation with a dispersive spectrometer using these modern optimized components is, in our view, best suited for environmental applications to aqueous systems. While the performance in this case is comparable to 514.5 nm excitation with charcoal removal of the fluorescing impurity, consider the case where the analyte of interest itself fluoresces strongly with green or blue excitation, but not with 785 nm excitation. Many important examples of this case have been reported.⁵

Brief history of technological advances

Spectrometers, such as that used to collect the Fig. 2(D) spectrum, which incorporates a stabilized, high-power, 785 nm diode laser, a high-efficiency CCD optimized for near-IR excitation and a high-throughput spectrograph with an efficient laser-line rejection filter, only became widely available during the last decade. Because these instruments are opening the door to more widespread environmental applications, it is instructive to consider the historical events that led to their fruition.

Raman spectroscopy almost literally underwent two ‘rebirths’ in the last half of the 20th century. The first rebirth began in the late 1960s as a result of the advent of gas lasers. Immediately prior to that, the Raman effect was more of an esoteric academic interest than the basis of a widely used spectroscopic tool. In principle, Raman spectroscopy offered numerous advantages over IR spectroscopy for the study of rotational and vibrational spectra, but was not widely used in the preceding decades, due in large part to the limitations of the mercury arc lamp, which was the standard excitation source. Also, the late 1940s and 1950s saw many technological advances in IR spectroscopy, which made it an important analytical tool for many chemists. During the period of the late 1960s to the early 1980s, laser Raman spectroscopy gained considerable popularity as a spectroscopic tool; however, once again, technological advancements in IR beginning in the late 1970s [i.e. the development and marketing of turn-key, computerized, Fourier transform (FT) IR systems] helped maintain the prominence of IR as the vibrational spectroscopic tool of choice.

The most recent rebirth of Raman spectroscopy is not so easily traced to a single piece of technology, but is due to technological advancements nonetheless. The starting point of this rebirth can, perhaps, be traced to 1986 when Hirschfeld and Chase published a landmark paper in which the practical application of FT-Raman was first demonstrated.¹¹ An important advance that made FT-Raman feasible, after having first been discussed in 1964,¹² was the development of optical notch filters that effectively rejected the relatively intense Rayleigh scattering line, while transmitting the much weaker Raman scattered light. Devices such as these were an important precondition for a practical FT-Raman measurement, because all scattered light strikes the detector simultaneously. Because the FT-Raman measurement could be made on modified FT-IR interferometers, instrument manufacturers were able to quickly market FT-Raman systems. The introduction of these instruments significantly increased the popularity of Raman spectroscopy, as evidenced by a flurry of publications.^13–15 Although these instruments offered increased ease of use and greater throughput, the sensitivity of FT-Raman systems was actually only comparable or slightly better than that of earlier generation dispersive Raman systems.¹⁶ Indeed, the most practical advantage of FT-Raman was that it typically employed excitation with the 1065 nm output from an Nd∶YAG laser, which avoided fluorescence from virtually all organic molecules.

Almost all early reports of FT-Raman focused on the ability to analyze samples that were too highly fluorescent to be analyzed by conventional dispersive Raman instruments that were widely available at that time.¹³ However, several research groups,^17,18 perhaps most notably that of McCreery,^19,20 recognized that fluorescence could also largely be avoided with dispersive Raman systems using newly developed high-power diode lasers that emitted near 785 nm. All other ‘things’ being equal (such as laser intensity, detector response, etc.), an interferometric technique offers relatively minor advantages (e.g. greater wavelength precision) over dispersive instruments. However, these research groups realized that, if fluorescence-free spectra could be collected with excitation near 785 nm (instead of 1065 nm), recently developed high-efficiency, near-IR optimized CCD detectors could be employed with great effect in the dispersive mode. Now, suddenly ‘things’ were no longer equal due to the unmatched sensitivity and low noise of these modern CCDs. Back-illuminated, deep-depletion, anti-reflection-coated CCDs with a peak quantum efficiency of 85+% at 700 nm and <5 electron read noise were becoming available at that time.²¹ Fortunately, much progress had also recently been made in diode lasers that emitted at wavelengths suitable for use with these CCDs. Historically, diode lasers had been very low power and prone to mode hop, precluding their application to Raman spectroscopy. However, 300 mW external cavity-stabilized diode lasers (which were not prone to mode hop), emitting nominally at 785 nm, were becoming available at that time. Therefore, these research groups began to build, and several manufacturers subsequently began to market, CCD/near-IR diode laser-based dispersive Raman instruments. Like FT-Raman instruments, these dispersive instruments also made use of very efficient notch filters to reject the Rayleigh line prior to passing the scattered radiation to the dispersing element.^22,23 Consequently, only a single grating and a few reflectors and lenses were required. These modern single-stage spectrographs have throughputs of about 50%. Double (or triple) monochromators, which were previously required for earlier generation dispersive instruments in order to achieve adequate Rayleigh light rejection, exhibited only about 5% throughput.²⁴

These two new instrumental approaches, FT-Raman and CCD-based dispersive Raman, have created a recent and unmistakable resurgence in interest in Raman spectroscopy. The number of papers currently coming from academia, government and industry gives us reason to believe that Raman spectroscopy may soon become a widely used and routine analytical tool, including for applications to potable water systems. Both FT-Raman and CCD-based Raman approaches have their own strengths, and each is likely to find wider application in the future. FT-Raman has proven to be most useful in cases where intense fluorescence is still observed even when using 785 nm excitation, or in cases where very high wavelength precision is required.⁵ However, CCD-based Raman will likely find wider environmental application, particularly to potable water analysis, because the sensitivity is much higher than that of FT-Raman. While fluorescence may hamper some environmental applications with 785 nm excitation and not with 1064 nm excitation, the large difference in sensitivity will typically be a more important consideration for trace analysis. Also, fluorescence is likely to be less of a concern with relatively pristine potable water than with many other types of environmental samples. Indeed, excitation at 785 nm seems to be a good compromise between considerations of fluorescence rejection and Raman scattering intensity. Because any future application of Raman spectroscopy to potable water analysis will likely not involve FT-Raman systems, they will not be addressed further in this article.

In addition to allowing one to record fluorescence-free spectra in many cases, these various technological advances incorporated into modern CCD-based dispersive systems have served to significantly lessen many of the other historical difficulties of Raman spectroscopy as well. Indeed, for a given excitation laser power and signal integration time, this new generation of instrument is far more sensitive than the older generation photomultiplier tube-based dispersive Raman systems. For example, Gilmore et al. have used a system of this type for the analysis of a variety of environmentally important dyes in water.²¹ The detection sensitivity at S/N = 2 ranged from 10⁻⁷ to 10⁻⁵ mol l⁻¹ for 5 min integration time, typically with less than 50 mW of power at 782 nm. These LODs, in this time frame and with this low a laser power, would not have been possible 10 years earlier.

Also, these new generation instruments offer much greater flexibility and ease of use. Because they employ single-stage spectrographs and diode lasers, they are very compact, easily transported and require only standard low-current electrical service. Specialized facilities and highly specialized expertise are no longer required for adequate operation. With some instruments, a wide spectral range (about 3400 cm⁻¹) can be recorded at modest resolution (about 5 cm⁻¹) with a single camera exposure and no moving parts. This approach affords shorter analysis times and improved frequency precision and calibration stability compared to older generation dispersive instruments. Typically, these new instruments are frequency and intensity calibrated by an automated procedure through which output from an atomic gas and a white light source, respectively, is fitted to a polynomial.

These new commercial systems are frequently sold with fiber optic probes and are popular as in-line industrial process monitoring instruments. (Low-loss silica fiber optics have been available for transporting visible to near-IR light for many years, and the advantages of fiber optic Raman were recognized long before these new CCD-based systems became available).²⁵ Many modern instruments employ remote probe heads that filter the Raman scatter from silica; thus, spectra can be collected down to 50 cm⁻¹ without prohibitively high background features.²⁶ Also, the fiber coupling efficiency is now very good and it is possible to use very long runs of fiber optic cables and still retain adequate signal intensity. Therefore, in addition to industrial process and laboratory deployment, these instruments are attractive for remote environmental applications, such as subsurface chemical analysis in the field. Indeed, some progress has been made in this area. For example, a Raman probe has been designed for a cone penetrometer for studying underground storage tanks at the Department of Energy's Hanford site.²⁷ These tanks contain a mixture of chemicals and radioactive waste, and, with in-tank characterization of such mixtures, significant reduction in personnel exposure, analysis time and cost is achieved relative to laboratory analysis. Also, Angel et al. have used a modern CCD-based dispersive Raman system with 808 nm diode laser excitation to detect gasoline contaminants perched on groundwater, using a 250 ft fiber optic probe that was inserted in a 4 in diameter monitoring well.²⁸ Continued progress on this front holds out the hope that remote measurements on potable water sources may come to fruition some day soon.

Ongoing developments and applications

Unfortunately, most potable water analyses require LODs orders of magnitude below those currently available with NRS, even when using modern CCD-based dispersive systems. For several reasons, it is not prudent to expect NRS LODs to decrease significantly as basic instrument components are further developed. First, most components have likely plateaued in their developmental curve. For example, a modern optimized CCD has a peak quantum efficiency of about 90% with over 60% efficiency throughout much of the spectral region of interest, with extremely low detector noise. Also, modern single-stage spectrographs have throughputs that are about as high as possible. Second, it is unlikely that significantly lower LODs for aqueous samples will be obtained with NRS because shot noise associated with Raman scattering and luminescence from the medium represent the ultimate limit to sensitivity in potable water analysis. It is not clear if we are presently at that limit, but it is unlikely that we are orders of magnitude above it.

For these reasons, significant improvements in LODs for potable water analysis will likely require strategies other than further improving the basic components of the NRS instrument. There are some promising developments emerging on several fronts. First, several clever sampling strategies, with the goal of improving sensitivity via contaminant preconcentration, are currently being developed. For example, Wittkamp and Tilotta²⁹ have used solid-phase micro-extraction (SPME) to preconcentrate organics from water onto a small area of the SPME device. The concentrated contaminants are then analyzed directly on the solid-phase medium. (Note that, in addition to liquid samples, Raman is directly applicable to solids as well as gas samples). This provides enhancement factors of about 100–1000 for some organics at the expense of increasing the analysis time by about 30 min. For example, Wittkamp and Tilotta used this approach to detect BTEX chemicals (benzene, toluene, ethyl benzene and xylenes) in spiked well water and river water, with no significant fluorescence interferences from naturally occurring materials. BTEX chemicals represent the main aromatic components of gasoline and are an important class of chemicals that potentially threaten potable water supplies.

Another promising preconcentration strategy is the straightforward coupling of capillary electrophoresis (CE) to Raman spectroscopy, as described by Kowalchyk et al.³⁰ This approach affords the use of electrophoretic preconcentration of the contaminant by sample stacking. With this technique, a contaminant will become more concentrated in the CE capillary because the electric field is higher in the sample plug than in the buffer solution with its higher ionic strength. This approach can provide an enhancement factor of greater than 1000 and was used to detect high parts per billion levels of nitrate and perchlorate ions in water. This LOD was obtained with only 1 s integration time and a total sample analysis time (i.e. injection and separation time) of less than 3 min. Further improvements in LODs are expected with continued development of this technique. Perhaps more important than analyte preconcentration, this approach allows for the study of more complex samples by incorporating a separation step. Indeed, most environmental samples contain too many components to allow direct analysis of the analyte(s) of interest using NRS. Also, a separation technique, such as CE, allows the analysis of the analyte separate from the matrix, which often contributes Raman scattering and fluorescence that can obscure the Raman signal from the analyte. Thus, approaches that involve the coupling of separation and spectroscopic techniques (commonly called hyphenated methods), such as CE-Raman, are particularly attractive for potable water analysis.

Another emerging sampling strategy for improving the sensitivity of NRS is the use of a Teflon^®-AF-based liquid core optical fiber (LCOF).³¹ The LCOF is a long capillary used as a sample cell in which a small volume of an aqueous sample is injected. The excitation laser is optically coupled to the capillary and photons are trapped inside by total internal reflection because the refractive index of Teflon^®-AF is less than that of water. This multipass effect provides an extremely long path length through a very small sample volume. The Raman scatter collected from the water-solvated contaminant in the LCOF may be enhanced by more than 500 times that collected using a standard cuvette with the same laser power and analysis time. Unfortunately, the Raman scatter from the water (and the associated shot noise) is also enhanced, and so the overall S/N improvement for the analyte will not be as large. However, this new approach should provide LODs for aqueous samples that are limited by background shot noise without requiring long exposure times or high-power lasers. An added advantage of LCOFs is that fluorescence from low-level sample impurities can be rapidly depleted by laser photobleaching, because the entire sample volume is constantly illuminated. An LCOF with NRS detection has recently been used as a detector system for liquid chromatography (LC).³² This coupled system looks promising for the analysis of complex environmental samples.

An entirely different approach to improving Raman detection levels in aqueous samples is to resort to the numerous phenomenological variants of Raman spectroscopy. These include a host of non-linear Raman techniques, some of which offer potential advantages for potable water analysis. Perhaps most notable among these is coherent anti-Stokes Raman spectroscopy (CARS).³³ However, analytical applications of non-linear Raman techniques have been slow in coming, due largely to the complexity of the instrumentation and underlying theories, which are well beyond the scope of this article.

Resonance Raman spectroscopy (RRS) and surface-enhanced Raman spectroscopy (SERS) are phenomenological variants of the Raman technique that have found more widespread analytical utility and have been applied with some success to aqueous environmental samples. Resonance enhancement of the Raman signal occurs in RRS when the excitation laser frequency is coincident (or near-coincident) with an electronic transition of the analyte. (Referring back to Fig. 1, this situation would occur when hν₀ approaches E₁ − E₀). This can result in a 10²–10⁶ increase in the Raman signal compared to NRS.³⁴ However, because visible or ultraviolet excitation must be used to achieve resonance, interference from fluorescence is a great problem with RRS. Also, certain vibrations are enhanced to a much greater extent than others; thus, the resonance Raman spectrum typically looks quite different from the normal Raman spectrum due to great differences in relative peak intensities. Furthermore, band intensities can change dramatically if the excitation wavelength varies, due to a change in the magnitude of the resonance enhancement. These aspects would hamper qualitative analysis of unknown contaminants, and even quantitative analysis of known contaminants, in potable water. For example, an RRS detector for the qualitative analysis of complex mixtures (e.g. coupled to a separation technique) would be of limited utility. The user could not anticipate the wavelength that would be required for the enhancement of an unknown analyte. Even if all target analytes are known, it is not likely that all would exhibit optimal enhancement at the same wavelength. Also, it should be noted that resonance enhancement can only be achieved directly on compounds that have electronic absorption bands in the energy regions accessed by lasers suitable for Raman excitation. This further limits the application of the technique. This limitation can be somewhat lessened by derivatizing colorless compounds into complexes that are suitable for RRS, such as organic dyes. For example, Koppe et al.³⁵ were able to analyze various phenols with RRS using the 4-nitroaniline derivatization method. However, they were successful with only 28 of the 126 phenols tested.

The SERS technique is based on the discovery of Fleischmann et al.³⁶ that the intensity of Raman scattering could be enhanced by a factor of 10³–10⁷ relative to NRS when some analytes were adsorbed to some solid surfaces, most notably roughened silver surfaces. Typically, fluorescence is not a problem with SERS, and a surface-enhanced Raman spectrum is sometimes qualitatively similar to the normal Raman spectrum. Thus, the technique is quite attractive for potable water analysis and, indeed, environmental applications of SERS have increased in recent years. For example, Carron and Kennedy³⁷ coupled SERS detection to gas chromatography and applied the technique to the identification of BTEX compounds in water. All six BTEX components were spectrally resolved even though ethyl benzene and the isomers of xylene were not chromatographically resolved. The detection limit for this system was reported to be 50 ng for benzene. SERS has also been successfully coupled to LC and flow injection analysis (FIA)³⁸ and thin-layer chromatography (TLC)³⁹ for environmental application to aqueous systems. Detection limits of about 0.5 ng have been observed for various dyes with TLC-SERS using only 3 mW excitation from a helium–neon laser.³⁹

Unfortunately, the SERS effect is observed with only a limited number of molecules, and appears to be related to the presence of functional groups that promote adsorption to SERS-active surfaces. These include chemicals that are ionic, polar or easily polarized. Considerable effort has been devoted to determining the sources of SERS enhancement. Explanation of these efforts is beyond the scope of this article and can be found in the reviews of Vo-Dinh⁴⁰ and Garrell.⁴¹ Also, a list of chemicals that exhibit the SERS effect has been published by Seki.⁴² While SERS is potentially suitable for monitoring a known SERS-active contaminant in potable water, it is not so attractive for the identification of unknowns in multicomponent systems. Multicomponent systems are also problematic in that competitive adsorption may frustrate quantitative analysis. Also, SERS surfaces are historically unstable and irreproducible, resulting in a large variation in the enhancement factor, which hampers quantitative measurements. However, this situation has recently been improved, for example, with the use of polymer-coated SERS-active substrates.

Concluding remarks

Although the great potential for environmental applications of Raman spectroscopy to aqueous systems was recognized many years ago, progress in this area was slow until recently because of several historical problems. However, as we have described, the severity of many of these problems has been greatly reduced by improvements in the basic instrumental components that make up the modern CCD-based dispersive normal Raman spectrometer. These improvements have led to Raman spectroscopy becoming the tool of choice for an ever increasing number of analytical applications, including those of environmental importance. To illustrate this encouraging trend, we have displayed as a bar graph in Fig. 3 the number of journal articles published from 1970 to 1999 (in increments of 4 years) that deal with environmental applications of Raman spectroscopy to aqueous systems. Included are articles involving NRS, SERS and RRS (some of which we have briefly reviewed herein, and more fully elsewhere).⁴³ Also, we have included a bar denoted ‘Hyphenated’ for papers that describe one or more of these techniques coupled to a spectroscopic technique. It should be noted that NRS dominated the reports from the 1970s, but fell out of favor in the 1980s, most likely due to low sensitivity and the other limitations that we have described. During this time, RRS, which offers better sensitivity, was often the tool of choice. However, there has been a sharp increase in NRS applications since the mid-1990s, due to the dramatic improvements in instrumental components. Also, we have observed steady growth in SERS applications since the mid-1980s. Finally, the maturation of Raman spectroscopy as an important analytical tool for the analysis of complex environmental mixtures is seen in the steady growth of hyphenated techniques, particularly since the late 1980s.

Fig. 3 Number of journal articles devoted to aqueous environmental applications of various Raman techniques as a function of year. Note that there is some overlap among these techniques, and thus the aggregate height of the stacked bars overestimates the total number of articles. In particular, all papers on hyphenated techniques involve coupling one of the other three Raman techniques to a chromatographic separation technique. The total number of papers for a given period is approximately equal to the aggregate height of the stacked bars with the block for hyphenated techniques removed.

However, before NRS is to find widespread application to potable water analysis, further improvement in LODs is required. We believe analyte enrichment (i.e. preconcentration) approaches are, perhaps, most promising in this regard. Certainly, another way to achieve lower LODs for contaminants in potable water is to turn to the numerous phenomenological variants of Raman spectroscopy that offer higher sensitivity. At present, there seems to be more promise and certainly more attention devoted to SERS in this regard. With continued improvements in the stability and reproducibility of substrates, it appears that SERS may soon become more widely used as an analytical technique, including application to the analysis of aqueous samples. However, we continue to believe that there are, for many environmental applications, considerable advantages to NRS over SERS and other phenomenological variants of the Raman technique. Therefore, we see a pressing need for the further lowering of LODs of NRS for the analysis of contaminants in potable water.

Another encouraging development in Raman spectroscopy is the increased interest in coupling NRS and SERS to separation techniques, such as LC and CE. We hope and expect to see this trend continue. This is particularly important for environmental applications, including potable water analysis, because samples are often too complex to analyze directly.

Finally, we note that more widespread application of Raman spectroscopy to environmental problems will require some education about the technique as it is currently practiced. Based on its historical limitations, Raman has developed a reputation as a difficult, insensitive and unwieldy technique. However, modern NRS instruments are now almost turn-key, and can be used by non-specialists to great effect for many applications. Frequently, skeptics are turned into enthusiastic supporters when given the chance to rediscover Raman spectroscopy.

References

1. E. B. Bradley and C. A. Frenzel, Water Res., 1970, 4, 125.
2. D. A. Long, Raman Spectroscopy, McGraw-Hill, New York, 1977.
3. C. V. Raman and K. S. Krishnan, Nature, 1928, 121, 501.
4. M. J. Pelletier, in Analytical Applications of Raman Spectroscopy, ed. M. J. Pelletier, Blackwell Science, Oxford, 1999.
5. R. L. McCreery, Raman Spectroscopy for Chemical Analysis, John Wiley, New York, 2000.
6. D. L. Gerrard, in Analytical Raman Spectroscopy, ed. J. G. Grasselli and B. J. Bulkin, John Wiley, New York, 1991, p. 312.
7. N. Everall, R. W. Jackson, J. Howard and K. Hutchinson, J. Raman. Spectrosc., 1986, 17, 415.
8. T. L. Williams, R. B. Martin and T. W. Collette, Appl. Spectrosc., 2001, 55, 967.
9. C. K. Mann and T. J. Vickers, Appl. Spectrosc., 1987, 41, 427.
10. T. W. Collette, T. L. Williams and J. C. D'Angelo, Appl. Spectrosc., 2001, 55, 750.
11. T. Hirschfeld and D. B. Chase, Appl. Spectrosc., 1986, 40, 133.
12. G. W. Chantry, H. A. Gebbie and C. Hilsum, Nature, 1964, 203, 1052.
13. C. G. Zimba, V. M. Hallmark, J. D. Swalen and J. F. Rabolt, Appl. Spectrosc., 1987, 41, 721.
14. J. J. Freeman, D. O. Fischer and G. J. Gervasio, Appl. Spectrosc., 1993, 47, 1115.
15. H. Wang, C. K. Mann and T. J. Vickers, Appl. Spectrosc., 1995, 49, 127.
16. D. B. Chase, in Analytical Raman Spectroscopy, ed. J. G. Grasselli and B. J. Bulkin, John Wiley, New York, 1991, p. 41.
17. M. M. Carrabba, K. M. Spencer, C. Rich and D. Rauh, Appl. Spectrosc., 1990, 44, 1558.
18. B. Yang, M. D. Morris and H. Owen, Appl. Spectrosc., 1991, 45, 1553.
19. C. D. Newman, G. G. Bret and R. L. McCreery, Appl. Spectrosc., 1992, 46, 262.
20. Y. Wang and R. L. McCreery, Anal. Chem., 1989, 61, 2647.
21. D. A. Gilmore, D. Gurka and M. B. Denton, Appl. Spectrosc., 1995, 49, 508.
22. J. M. Tedesco, H. Owen, D. M. Pallister and M. D. Morris, Anal. Chem., 1993, 65, 441A.
23. C. L. Schoen, S. K. Sharma, C. E. Helsley and H. Owen, Appl. Spectrosc., 1993, 47, 305.
24. D. B. Chase, Appl. Spectrosc., 1994, 48, A14.
25. R. L. McCreery, M. Fleischmann and P. Hendra, Anal. Chem., 1983, 55, 146.
26. N. Everall, H. Owen and J. Slater, Appl. Spectrosc., 1995, 49, 610.
27. J. H. Aldstadt and A. F. Martin, Mikrochim. Acta, 1997, 127, 1.
28. S. M. Angel, T. F. Cooney and H. T. Skinner, in Modern Techniques in Raman Spectroscopy, ed. J. J. Laserna, John Wiley, Chichester, 1996, p. 414.
29. B. L. Wittkamp and D. C. Tilotta, Anal. Chem., 1995, 67, 600.
30. W. K. Kowalchyk, P. A. Walker and M. D. Morris, Appl. Spectrosc., 1995, 49, 1183.
31. M. J. Pelletier and R. Altkorn, Appl. Spectrosc., 2000, 54, 1837.
32. B. J. Marquardt, P. G. Vahey, R. E. Synovec and L. W. Burgess, Anal. Chem., 1999, 71, 4808.
33. P. C. Chen, C. C. Joyner and M. Burns-Kaurin, Appl. Opt., 1999, 38, 5894.
34. M. D. Morris and D. J. Wallan, Anal. Chem., 1979, 51, A182.
35. P. Koppe, F. Dietz, J. Traud and C. Rubelt, Fresenius' Z. Anal. Chem., 1977, 285, 1.
36. M. Fleischmann, P. J. Hendra and A. J. McQuillan, Chem. Phys. Lett., 1974, 26, 163.
37. K. T. Carron and B. J. Kennedy, Anal. Chem., 1995, 67, 3353.
38. B. J. Kennedy, R. Milofsky and K. T. Carron, Anal. Chem., 1997, 69, 4708.
39. C. D. Tran, Anal. Chem., 1984, 56, 824.
40. T. Vo-Dinh, Sens. Actuators B: Chem., 1995, 29, 183.
41. R. L. Garrell, Anal. Chem., 1989, 61, A401.
42. H. Seki, J. Electron. Spectrosc. Relat. Phenom., 1986, 39, 289.
43. T. L. Williams and T. W. Collette, in Handbook of Raman Spectroscopy, ed. I. R. Lewis and H. G. M. Edwards, Marcel Dekker, New York, 2001, p. 683.

---

Footnote

† This paper has been reviewed in accordance with the US Environmental Protection Agency's peer and administrative review policies and approved for publication. Mention of trade names or commercial products does not constitute endorsement or recommendation for use.

---