Measurements of Bicarbonate in Water Containing Ocean-Level Sulfate Using a Simple Multi-Pass Optical Raman System

: The concentration of dissolved inorganic carbon in the oceans at depths of a few meters to thousands of meters is a critical parameter for understanding global warming. The concentration is both pH dependent and depth dependent. Current analysis that employs pH meters must account for several other parameters, such as salinity, temperature, pressure, and the dissolved carbon’s form, carbon dioxide, bicarbonate, or carbonate. Recently, Raman spectroscopy has been used to measure these forms directly in water at ~1000 ppm, which is unfortunately insufﬁcient for typical ocean concentrations, such as ~115 ppm bicarbonate near the surface. Here, we employed a simple multi-pass optical system, a ﬂat mirror to reﬂect the laser back through the sample, and a concave mirror opposite the entrance slit that effectively doubled the laser power and the collected Raman photons, respectively. This multi-pass optical Raman system with a 1.5 W, 532 nm laser was used to measure 30 ppm bicarbonate in water that contained 2650 ppm sulfate to simulate ocean water, a bicarbonate concentration well below that near the ocean surface. Furthermore, spectral analysis employed the bicarbonate C=O symmetric stretch at 1360 cm − 1 instead of the C–OH stretch at 1015 cm − 1 to avoid the intense, overlapping sulfate SO 4 symmetric stretch at 985 cm − 1 . The calculated standard deviation of ~5 ppm for the described approach suggests that accurate measurement of bicarbonate in situ is possible, which has been, heretofore, either calculated based on pH or measured in a lab.


Introduction
The measurement of dissolved inorganic carbon in the ocean has taken on considerable importance in view of the increasing emissions of CO 2 into the atmosphere which, when sequestered, acidifies the oceans [1,2], negatively affecting coral reefs [3], crustaceans [4], and many other carbonate-containing aquatic life forms. In light of these deleterious effects, a controversial proposal [5] suggests that pumping CO 2 into the oceans below 3000 m [6], where it is denser than water, could remove substantial amounts of CO 2 from the atmosphere, alleviating global warming [7]. Currently, pH meters are used to estimate the amount of CO 2 in the water based on the equilibrium of Equation (1) Unfortunately, pH alone is not sufficient as it only measures hydrogen ions, and has underestimated the CO 2 concentration in non-equilibrium cases, such as at ocean depths below 3000 m [9]. Several researchers have demonstrated the ability of Raman spectroscopy to perform these measurements. Scientists at the Canadian Department of the Environment, Inland Waters Branch, demonstrated that CO 2 , bicarbonate (HCO 3 − ) and carbonate (CO 3 2− ) all produce unique vibrational modes that could be used for quantitation [10]. The 2014 concentration for these species in the ocean at a pH of 8 are CO 2 at 0.9 ppm, HCO 3 − at 114 ppm and CO 3 2− at 9.5 ppm (mg/L) [11]. The researchers Oceans 2021, 2 331 at Monterey Bay Aquarium Research Institute developed a sophisticated two chamber sample system to mix liquid CO 2 and seawater at a depth of 500 m and measured both the pH and Raman spectra [12,13]. The spectrometer, using a 50 mW, 532 nm laser and a −100 • C thermal-electrically cooled (TEC) detector, (36 cm diameter, 80 cm long, 180 kg, not including the probe head) was able to measure 1080 ppm CO 2 in 2.5 min, based on the 1380 cm −1 peak area, with an estimated limit of detection of 600 ppm. This system may have value in monitoring CO 2 near the 3000 m condensation depth, especially if pumping it into the ocean is implemented. The team at the Ocean University of China developed a multi-pass optical sample system that provided a 7.8 signal amplification factor for a spectrometer using a 300 mW, 532 nm laser and a −80 • C TEC detector. The system was used to measure~1600 ppm CO 2 ppm gas phase extracted from a liquid water sample in 3 min [14]. Based on these previous systems we developed a Raman system that used a high powered 1.5 W, 532 nm laser and a simple multiplicative optical system and measured CO 2 , HCO 3 − , and CO 3 2− in 20 min at tens of ppm concentrations in water [15,16]. Here, we present measurements of HCO 3 − in water containing 2650 ppm sulfate to simulate the spectral challenge of performing measurements in the ocean. The calculated error of ±3 ppm in the measurement is close to the highest yearly flux of 2 ppm for the ocean [17].
Methods. The Raman spectrometer employed a 1.5 W, 532 nm laser (model Cobolt Samba 1500 532 nm, Hubner Photonics, San Jose, CA, USA) and a −15 • C cooled 2048 pixel Si detector (model 532-C-SR-L-25, Wasatch Photonics, Morrisville, NC, USA). Spectra were recorded from~200 cm −1 to 3200 cm −1 with~20 cm −1 resolution (peak width at half height) and peak positions are reported to the nearest 5 cm −1 . Initially, Raman spectra were collected using 180 • backscattering geometry. However, the laser induced a substantial SiO 2 background from the pre-sample optics, particularly the long wavelength pass filter that was used to reflect the laser into the sample and transmit the generated Raman photons into the spectrograph in the 180 • backscatter arrangement. Consequently, this component was removed and a 90 • perpendicular collection arrangement was employed ( Figure S1). While the SiO 2 spectral features were removed, a substantial tilted background remained due to the laser baseline "wings" in spite of the use of two additional long wavelength pass filters with an optical density of 6 (Semrock, Rochester, NY, USA) inside the spectrograph. In addition, a flat mirror was used to reflect the laser 180 • back through the sample to generate more Raman photons, while a 9 mm diameter f/0.5 concave mirror (Edmund Optics, Barrington, NJ, USA) was mounted opposite the spectrograph entrance to reflect and focus the Raman photons scattered away from the spectrograph back towards the spectrograph. Both mirrors were front-coated with silver. A simple plastic enclosure was built to confine the potentially blinding laser. Wasatch software (Dash v3.2) was used to set the TEC detector temperature, while Hubner Photonics software (Cobolt Monitor 6.0.6.0) was used to turn the laser on and off, and set the power. RTA collection software (Raman Vista 3.4.17) was used to set the integration time, number of integrations, and collect "dark and light", laser off and on spectra, the former automatically subtracting from the latter. RTA software was also used to treat the saved spectra. See individual figure captions for spectral acquisition times.

Results and Discussion
The challenges to measuring CO 2 and HCO 3 − in ocean waters are first, the low concentration on the order of 100 ppm, and second the interference from other chemicals, specifically sulfate at~2650 ppm. Particulate matter is also a significant source of interference [18], but it is not addressed in this publication. Recently, we developed a highly sensitive Raman spectrometer that was used to measure HCO 3 − and CO 3 2− in water at 30 and 10 ppm, respectively [15]. Key features were the use of a high powered, 1.5 W, 532 nm laser, and two mirrors to increase the Raman generated and collected photons. The spectra were dominated by the water H 2 O bending mode at 1640 cm −1 [19] and the sulfate SO 4 symmetric stretch at 985 cm −1 superimposed on a substantial tilted background due to the laser "wings" (Figure 1a). At higher concentrations the bicarbonate C-OH stretch at 1015 cm −1 and the C=O symmetric stretch 1360 cm −1 were readily observed, more so than the broad and weak C-OH bend at 1300 cm −1 (Figure 1a,b) [20].

Results and Discussion
The challenges to measuring CO2 and HCO3 − in ocean waters are first, the low concentration on the order of 100 ppm, and second the interference from other chemicals, specifically sulfate at ~2650 ppm. Particulate matter is also a significant source of interference [18], but it is not addressed in this publication. Recently, we developed a highly sensitive Raman spectrometer that was used to measure HCO3 − and CO3 2− in water at ~30 and 10 ppm, respectively [15]. Key features were the use of a high powered, 1.5 W, 532 nm laser, and two mirrors to increase the Raman generated and collected photons. The spectra were dominated by the water H2O bending mode at 1640 cm −1 [19] and the sulfate SO4 symmetric stretch at 985 cm −1 superimposed on a substantial tilted background due to the laser "wings" (Figure 1a). At higher concentrations the bicarbonate C-OH stretch at 1015 cm −1 and the C=O symmetric stretch 1360 cm −1 were readily observed, more so than the broad and weak C-OH bend at 1300 cm −1 (Figure 1a,b) [20]. For this study, two concentration series of HCO3 − , Series 1: 10,000 to 30 ppm and Series 2: 1000 to 50 ppm, were prepared in water containing sulfate at 2650 ppm, and measured with the high powered, dual mirror Raman system. The H2O bending mode was used as an internal intensity standard, to which all spectra were scaled for analysis. The HCO3 − 1015 cm −1 peak is usually preferred to the 1360 cm −1 peak for quantitation, because the former is more intense and the latter overlaps the broad 1300 cm −1 peak [14,15]. Although the HCO3 − 1015 cm −1 peak is easily discerned at 10,000, 3000, and 1000 ppm, at lower concentrations it becomes obscured by the high background spectrum and interfered by the sulfate peak at 985 cm −1 , which has a scattering cross section more than For this study, two concentration series of HCO 3 − , Series 1: 10,000 to 30 ppm and Series 2: 1000 to 50 ppm, were prepared in water containing sulfate at 2650 ppm, and measured with the high powered, dual mirror Raman system. The H 2 O bending mode was used as an internal intensity standard, to which all spectra were scaled for analysis. The HCO 3 − 1015 cm −1 peak is usually preferred to the 1360 cm −1 peak for quantitation, because the former is more intense and the latter overlaps the broad 1300 cm −1 peak [14,15]. Although the HCO 3 − 1015 cm −1 peak is easily discerned at 10,000, 3000, and 1000 ppm, at lower concentrations it becomes obscured by the high background spectrum and interfered by the sulfate peak at 985 cm −1 , which has a scattering cross section more than six times that of the HCO 3 − 1015 cm −1 peak [21]. Fitting the overlapping sulfate-bicarbonate peaks with weighted amounts of the individual chemical spectra involves user selected parameters, such as spectral region, peak shape and width, as well as baseline corrections. All of which can introduce bias in the analysis, especially at concentrations below 300 ppm HCO 3 − . Fortunately, the HCO 3 − 1360 cm −1 peak is not interfered by sulfate and it was therefore used to quantify the HCO 3 − concentration for all of the samples. A Raman intensity dependent HCO 3 − concentration curve was obtained by performing the following steps. First, the Raman spectrum of the initial sulfate-water solution was subtracted from the sample spectrum until a flat baseline was obtained, and second the spectral baseline was set to 0 at 1415 cm −1 (Figure 1c). The amount subtracted ranged from 0.99 to 1.02 for Series 1. It is worth noting that this subtraction resulted in minor over and under subtraction of the sulfate peak. This is due to slight changes in the tilt of the background, which is attributed to the variability of the quartz cuvette wall thickness that can be as much as 10% [22]. Third, the spectral noise was reduced by fitting the spectra with a 3rd-order, 19-point running smooth [23], and fourth, the peak height was measured at 1360 cm −1 (Figure 1d).
While this procedure worked well at 100 ppm, the spectra contained considerable noise, and the baseline tilt became significant for lower concentrations, such as 30 ppm ( Figure 2). Concentration Series 2 was prepared to focus on the near surface ocean concentration of 115 ppm with the goal of determining the accuracy and precision. This involved preparing and measuring concentrations closer to this value, and increasing the two coadded 8-min spectral acquisition for Series 1 to eight co-added 8-min spectral acquisition for the Series 2 50, 100, and 125 ppm concentrations to improve the signal-to-noise ratio (S/N). While longer acquisitions improved S/N, the baseline tilt remained for the lowest concentration of 50 ppm (Figure 2d). For these lower concentrations a fifth step was used that connected a linear line from 1300 cm −1 to 1425 cm −1 to account for baseline tilt and determine the peak height at 1360 cm −1 ( Figure S3). This step eliminates the need for the third step used to set the baseline to zero. This alternative four step sequence can be incorporated into a program to automate analysis.
Oceans 2021, 2, FOR PEER REVIEW 4 six times that of the HCO3 − 1015 cm −1 peak [21]. Fitting the overlapping sulfate-bicarbonate peaks with weighted amounts of the individual chemical spectra involves user selected parameters, such as spectral region, peak shape and width, as well as baseline corrections. All of which can introduce bias in the analysis, especially at concentrations below 300 ppm HCO3 − . Fortunately, the HCO3 − 1360 cm −1 peak is not interfered by sulfate and it was therefore used to quantify the HCO3 − concentration for all of the samples. A Raman intensity dependent HCO3 − concentration curve was obtained by performing the following steps. First, the Raman spectrum of the initial sulfate-water solution was subtracted from the sample spectrum until a flat baseline was obtained, and second the spectral baseline was set to 0 at 1415 cm −1 (Figure 1c). The amount subtracted ranged from 0.99 to 1.02 for Series 1. It is worth noting that this subtraction resulted in minor over and under subtraction of the sulfate peak. This is due to slight changes in the tilt of the background, which is attributed to the variability of the quartz cuvette wall thickness that can be as much as 10% [22]. Third, the spectral noise was reduced by fitting the spectra with a 3rd-order, 19-point running smooth [23], and fourth, the peak height was measured at 1360 cm −1 (Figure 1d). While this procedure worked well at 100 ppm, the spectra contained considerable noise, and the baseline tilt became significant for lower concentrations, such as 30 ppm ( Figure 2). Concentration Series 2 was prepared to focus on the near surface ocean concentration of 115 ppm with the goal of determining the accuracy and precision. This involved preparing and measuring concentrations closer to this value, and increasing the two co-added 8-min spectral acquisition for Series 1 to eight co-added 8-min spectral acquisition for the Series 2 50, 100, and 125 ppm concentrations to improve the signal-to-noise ratio (S/N). While longer acquisitions improved S/N, the baseline tilt remained for the lowest concentration of 50 ppm (Figure 2d). For these lower concentrations a fifth step was used that connected a linear line from 1300 cm −1 to 1425 cm −1 to account for baseline tilt and determine the peak height at 1360 cm −1 ( Figure S3). This step eliminates the need for the third step used to set the baseline to zero. This alternative four step sequence can be incorporated into a program to automate analysis.  The 1360 cm −1 peak heights for the two concentration series were plotted and fit with a straight line (Figure 3). Exceptional fits were obtained with R 2 values of 0.9999 and 0.9992, and y intercepts of +11.7 and −4.2 (detector counts normalized to water). The linear fits were used to calculate the concentration for each sample based on the 1360 cm −1 peak height, and the difference between the calculated and the prepared concentrations were used to indicate the accuracy of the measurements (Tables 1 and 2). Not surprisingly, the error was greatest for the Series 1 lowest concentration at 22 ± 8 ppm, a 36% difference from the prepared concentration. The Series 2 data were much better, and suggest reasonable accuracy with an average difference of 8 ppm or 3.7% of the prepared concentration. 0.9992, and y intercepts of +11.7 and −4.2 (detector counts normalized to water). The linear fits were used to calculate the concentration for each sample based on the 1360 cm −1 peak height, and the difference between the calculated and the prepared concentrations were used to indicate the accuracy of the measurements (Tables 1 and 2). Not surprisingly, the error was greatest for the Series 1 lowest concentration at 22 ± 8 ppm, a 36% difference from the prepared concentration. The Series 2 data were much better, and suggest reasonable accuracy with an average difference of 8 ppm or 3.7% of the prepared concentration.
The Series 2 individual spectra; eight at 50, 100, and 125 ppm, four at 200, 250, and 500 ppm, and two at 1000 ppm, were used to indicate the precision of the Raman spectrometer measurements (Table 3). For example, the average 1360 cm −1 peak height for the 50 ppm prepared sample was 26.4 ± 1.8, corresponding to 54 ± 3.7 ppm bicarbonate, a 6.8% standard deviation. In general, the percent standard deviation decreased with increasing concentration, as would be expected. Furthermore, the Series 2 data indicate that Raman spectral measurements are highly reproducible, and therefore, suggest that the inaccuracy of the measurements were likely due to the sample preparation that involved dilution over several orders of magnitude.    The Series 2 individual spectra; eight at 50, 100, and 125 ppm, four at 200, 250, and 500 ppm, and two at 1000 ppm, were used to indicate the precision of the Raman spectrometer measurements (Table 3). For example, the average 1360 cm −1 peak height for the 50 ppm prepared sample was 26.4 ± 1.8, corresponding to 54 ± 3.7 ppm bicarbonate, a 6.8% standard deviation. In general, the percent standard deviation decreased with increasing concentration, as would be expected. Furthermore, the Series 2 data indicate that Raman spectral measurements are highly reproducible, and therefore, suggest that the inaccuracy of the measurements were likely due to the sample preparation that involved dilution over several orders of magnitude.

Conclusions
The use of a Raman system employing traditional sample irradiation at 90 • to the spectrograph entrance, a high-powered 1.5 W, 532 nm laser, a flat mirror to reflect the laser back though the sample, and a concave mirror to collect Raman scattering opposite the spectrograph entrance, allowed the measurement of HCO 3 − at 30 ppm in water containing 2650 ppm sulfate. Using the bicarbonate C=O symmetric stretch at 1360 cm −1 , instead of the more intense C-OH stretch at 1015 cm −1 , avoided interference from the intense, overlapping SO 4 stretch at 985 cm −1 , and allowed a simple series of programmable steps to quantify HCO 3 − . The data presented suggest that a Raman system, such as the one described, could measure HCO 3 − accurately, at 125 ± 4 ppm, directly in the ocean. However, year-to-year changes will require at least a precision of 2 ppm or greater [17].
It would be apropos if Raman spectroscopy, the light scattering phenomenon discovered by Nobel Laureate Sir Chandrasekhar Venkata Raman as he wondered why the oceans appear blue even on cloudy days [24], became the method of choice to monitor the Reaction 1 components in the oceans, which in turn allowed understanding and alleviating global warming and aquatic life mortality.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/oceans2020019/s1, Figure S1: Top view illustration of (a) the original 180 • excitation/collection optics and (b) modified 90 • excitation/collection optics used with the Raman spectrometer. Figure S2: Raman spectra for 8 repeat measurements of 50 ppm bicarbonate from Series 2 (a) after subtracting the 2650 ppm sulfate in water spectrum and (b) after smoothing using Real-Time Analyzers, Inc. software. Figure S3: Real-Time Analyzers, Inc. software used to determine simple peak height at 1360 cm −1 for 8 measurements of 50 ppm bicarbonate.
Funding: This research received no external funding. Data Availability Statement: All avalilable data is contained within this article and its supplementary material.