The OPUS nitrate sensor is portable, small in size, and light in weight (Table 1). The device uses a xenon flash lamp, a reference diode, and a 256-channel high-end miniature spectrophotometer. Briefly, a xenon flash lamp is directed through an optical path with the sample, and the intensity of light passing through the sample is recorded by the spectrophotometer over a wavelength (λ) range of 200–360 nm with an integration time of 256 ms. A reference diode monitors the intensity of the light source. A schematic diagram of the sensor is shown in Figure 1. All components are housed in a single stainless steel or titanium pressure case.

The regular (factory setting) sampling interval of the OPUS is 30 s when set to operate in a continuous sampling mode. This is suitable for stationary deployments, but not when deploying the sensor on gliders or autonomous profiler systems, such as CTD frames with a vertical profiling speed of 1 m/s. We developed at GEOMAR an ATMega128 (Atmel Corporation)–based controller that triggers the raw spectral (dark and light) measurements of the OPUS via a Modbus-RTU protocol. The wiring diagram of the controller is provided in Figure 2. Electrical power to the OPUS and controller is provided through an auxiliary port of the CTD system. Measurements are conducted every 3 s at a defined sequence, with measurements of 10 times light followed by one-time dark spectrum. A drift up to 60 s per day in the internal clock of the OPUS was observed. To eliminate this, the controller is set to autonomously synchronize time during each dark measurement. Another important feature of the controller is that it provides backup power for a few seconds to allow the OPUS to finish its measurement in case of power loss.

The first part of this study consisted of experiments carried out under controlled laboratory conditions. Freshly dispensed deionized water (Milli-Q, resistance ≥18 MΩ cm^–1, Merck Millipore) was used to prepare calibration solutions of 840 μM Br^–, with and without 40 μM NO₃^–, and additional 1, 2, 4, 7, 10, 20, and 60 μM NO₃^– solutions, from stock solutions of 1,000 μM KBr (Fisher Scientific ACS reagent grade) and 1,000 μM KNO₃ (Merck Millipore ACS reagent grade). The calibration solutions were kept in acid cleaned (1 M HCl) glass bottles. A total of five OPUS sensors, all with 10 mm optical path length, were used in parallel under conditions described below. We assigned them consecutive numbers, i.e., OPUS1 to OPUS5, to which we refer throughout the study. The OPUS1 sensor was a deep-sea version and others were shallow water versions (see Table 1). The sensors were fully immersed in a thermally insulated glass container (15 L) sequentially filled with calibration solutions in a deionized water medium. The container was connected via Teflon tubing (I.D. 50 mm) to a water bath (7 L, Julabo GmbH) to control the temperature and circulate the solution (Figure 3). A custom-made polystyrene lid was placed on top to avoid contamination and heat exchange, and keep the sensors at the same height in the container. Before the measurements, optical windows of the sensors were cleaned with a few drops of acetone and wipes (Kimtech) followed by rinsing with deionized water. The surface of the sensors, volume of the container, water bath, and tubings contacting sample were all rinsed at least three times with deionized water at the beginning of the setup and between changes of solutions. First, freshly dispensed deionized water was carefully poured into the container, care was taken to avoid the formation of air bubbles on the optical paths, and a reference spectrum was recorded. Then, UV spectra of two calibration solutions were measured as a function of temperature. The water bath temperature was set to a total of four fixed temperatures between 5 and 20°C, and enough time was given to stabilize the sample temperature. The OPUS1 sensor was set to 3 s while the other sensors were set to 30 s for about 30 min (as only one controller for higher frequency analysis was available at the time of the experiment). The calibration solutions were used to assess the temperature effect and to derive molar extinction coefficients—the strength of chemical species to absorb light at a particular wavelength—of Br^– (ε_Br⁻,cal) and NO₃^– (εN⁢O3-,c⁢a⁢l).

Laboratory measurements were conducted in 840 μM Br^–, which is equivalent to Br^– concentrations in seawater with salinity 35. Other chemical interferences from small seawater components are expected to be low below 240 nm, and therefore these were not the subject of this study. The laboratory experiments were conducted in a freshly dispensed deionized water medium to have full control over the measurements and avoid potential matrix effects. The concentration unit of micromolar reported throughout the study indicates micromoles per liter (μmol L^–1).

During the experiments, in situ temperature in the container was measured with a Kelvimat 4323 thermometer (Burster Präzisionsmesstechnik GmbH, 2010) equipped with a Pt100 temperature probe, which has an accuracy of ±0.01°C. The sensors, temperature probe of the precision thermometer, and external computer were all synchronized via coordinated universal time (UTC). The TriOS web-based software was used to operate the sensors and download the internally recorded data. Power (12 V) was supplied to sensors by an external benchtop power supply.

One of the OPUS (OPUS1) sensor was employed in the southeastern North Sea on April 16–17, 2019 during the Sternfahrt-1 MOSES research expedition onboard RV Littorina. The expedition track from Büsum to Helgoland and return, and the location of the German Bight are shown in Figure 4. Helgoland is located about 60 km away from the German coast and Elbe River mouth (Ey et al., 2017), and the coastal waters are a mixture of riverine and saline North Sea waters. Before the deployment, the sensor housing was cleaned with deionized water, and the OPUS was fully immersed in a cylinder filled with deionized water to update the water-based spectrum. The sensor was fully immersed in a test tank (volume of 160 L) placed on deck of the vessel and was continuously supplied with surface water (from 2 m depth) at a flow rate of 80 L/min. UV spectral measurements of seawater were recorded with the OPUS at a 1-min sampling interval. In situ salinity and temperature values were recorded at 1-min interval using a CTD system (Seabird SBE 37-SMS-ODO), placed in the test tank. Discrete water samples were collected periodically at about 30-min intervals to validate the sensor measurements. For this, the water samples from the test tank were filtered (0.45 μm pore size PES filter; Fisher Scientific) and stored in 50-ml polypropylene tubes (Jet Biofil) that had been acid cleaned (1 M HCl). The tubes were rinsed three times with filtered seawater before collection. Samples were stored at −20°C and analyzed within 1 month at GEOMAR using an autoanalyzer (Seal QuAAtro) with standard wet-chemical colorimetric techniques (Becker et al., 2020).

The second field test took place in the tropical Atlantic Ocean, where the OPUS1 was mounted on a CTD frame and deployed on a cast down to 4,000 m depth (00°00.00′S, 30°00.00′W, October 15, 2019, CTD71, M158 research cruise, R/V Meteor). Ancillary data (including dissolved oxygen and inorganic nutrients; NO₃^–, nitrite, silicate, and phosphate) were obtained at various depths during the deployment.

Br^– is a conservative component of seawater with a concentration of ca. 840 μM at salinity 35 (Morris and Riley, 1966). The strength of absorbance for both Br^– and NO₃^– ions are closely related and overlap in the lower UV region, as indicated in a figure of their molar extinction coefficients vs. wavelength (Figure 5). Spectral discrimination of these ions is required to accurately compute NO₃^– concentrations. For this, the spectral region between 217 and 240 nm, where NO₃^– is dominant, is used (Zielinski et al., 2011).

Br^– absorption is temperature dependent as it occurs through a charge transfer process; the rate of charge transfer varies with ambient temperature (Jortner et al., 1964; Sakamoto et al., 2009). On the other hand, NO₃^– absorbance is independent of temperature due to the fact that it occurs within the molecule through π to π^∗ transition process (Mack and Bolton, 1999). During the laboratory tests, we used the calibration solutions to assess the temperature effect on Br^– absorbance. The absorbance of the 840 μM Br^– solution exponentially increased with an increasing temperature (Figure 6A), and gradually decreased with increasing wavelength (Figure 6B). The relative change in absorbance with respect to temperature remained constant for 840 μM Br^– with and without 40 μM NO₃^–, there was no additional temperature effect on absorbance due to the presence of NO₃^– ions (not shown here). It should be noted here that the 840 μM Br^– and 40 μM NO₃^– do not necessarily reflect the real seawater conditions, and the exact in situ levels of both constituents might vary in time and space; nevertheless, this can be computed using the algorithm (see Eq. 4 in section “Data Processing Procedure for OPUS”).

The raw spectral data of the calibration solutions were processed using the procedure described in section “Data Processing Procedure for OPUS.” An example spectral area attributed to NO₃^– after data processing is shown in Figure 7.

Raw spectral data of the OPUS were processed by taking potential lamp degradation, Br^– interference, and CDOM-baseline effect into account before calculating NO₃^– concentrations. The lamp degradation was taken into account during calculation by recording detector intensities in deionized water before and after each deployment (see section “Calibration of OPUS Sensors: Inter-Sensor Comparison”). We determined OPUS-specific molar extinction coefficients for Br^– and NO₃^– before data processing and developed a new algorithm (Eq. 4) for the compensation of Br^– interferences. Data processing procedure is outlined as follows:

Figure 8 illustrates an example of the spectral signature of CDOM on optical NO₃^– determination.

All data processing described and statistical analysis throughout the study were undertaken in MATLAB (MathWorks, R2018a) software. A mat file with the complete data from each recorded activity was produced. The code ingests the OPUS raw data file, calibration file, CTD file, and all equations above, and is available at https://github.com/uv-nitr/proc.

The OPUS sensor is factory calibrated at the manufacturer and a sum absorption spectra of expected seawater constituents at pre-defined concentrations is saved in the unit (see Table 1, Operating Instructions, TriOS GmbH, 2017). The other optical nitrate sensors, such as the ISUS and SUNA, are calibrated to derive the molar extinction coefficients of Br^– and NO₃^– (Sea-Bird Coastal SUNA, 2015). Recent documentation of these sensors’ calibration and data processing can be found in the “Processing Bio-Argo nitrate concentration at the DAC Level” and “8th BGC-Argo Meeting” reports (Johnson et al., 2018; Claustre and Johnson, 2019). In this work, the factory calibration of the OPUS was ignored and individual calibration coefficients were obtained (Eqs 2 and 3). The correlation for ε_{Br_cal} and ε_{NO3_cal} for about 75 pixels between 200 and 260 nm was examined using the Pearson correlation matrix and coefficients. A correlation coefficient value close to 1.0 indicates an excellent correlation between the respective datasets. Results indicated that the OPUS sensors were in excellent agreement (≥0.95) for both ε_{Br_cal} and ε_{NO3_cal} values (Figure 9), and are greater than 0.99 within 217 and 240 nm (for about 30 pixels, wavelength range of the NO₃^– fit, not shown here). The sensors wavelength values vary about 4–5 nm at the same pixel. Improvement of technical characteristics such as wavelength registrations and gratings in CCDs were beyond the scope of the study. However, the reason for non-identical correlation coefficients for the same pixels can be explained not in absolute value of sensors but in sensitivity. For example, the OPUS1 sensor had a spectral resolution of 0.82 nm while the OPUS2 had 0.79 nm resulting in better sensitivity.

Results from the laboratory test of the OPUS sensors under identical conditions, and using a series of 1, 2, 4, 7, 10, 20, 40, and 60 μM NO₃^– solutions in 840 μM Br^– medium, are presented in terms of NO₃^– bias across the five sensors. The NO₃^– concentration obtained from the OPUS sensor with the mean and laboratory analysis of discrete water samples was fit to a linear regression (r² = 0.99; Figure 10). The SD of NO₃^– for the lowest concentration was ∼0.64 μM, which translates to a limit of detection of ca. 2 μM NO₃^– (three times the SD of blank). An accuracy of better than ∼2 μM NO₃^– was determined from the residuals of the regression, while overall precision of NO₃^– sensor measurements was ∼0.4 μM, from the consecutive measurements of the identical sample.

An additional parameter used for the calibration is the reference spectrum recorded for deionized water medium using the OPUS sensors. The detector intensity of the sensor in deionized water was used in Eq. 1. Each OPUS sensor has a unique spectral output of its xenon flash lamp in the UV range (Figure 11). The design of the OPUS is comparable with the SUNA, except for the xenon lamp. Although this is not an OPUS vs. SUNA comparison study, a demonstration of reference intensity (recorded in deionized water) values of the xenon lamp–based OPUS sensor and deuterium lamp–based SUNA sensor are shown in Figure 11. Please note here that the data presented for the SUNA sensor were adopted from the literature (Johnson et al., 2018). Observing large differences in spectral output of xenon and deuterium lamps raised concerns regarding the need to derive OPUS-specific coefficients. Therefore, it is important that the true performance of each specific OPUS sensor is verified in the laboratory following well-established guidelines used for other optical nitrate sensors because the varying registrations, resolutions, and flash lamp spectra (i.e., ε values) are specific to each sensor. As this sensor-specific calibration is currently not available from the manufacturer, the five units were compared in the laboratory.

The five OPUS sensors are not identical in terms of their reference spectrum as each unit has a unique compact spectrometer and wavelength registration. In addition, the sensors are not identical in terms of their age (some were several years old and others several months). The reference spectrum needs to be updated periodically (before and after each deployment) for each individual unit to minimize potential sensor drift due to aging of the lamp or obstacles in the optical path (Pellerin et al., 2013) and ensuring sensor stability over time.

The second part of this study focused on the validation of the NO₃^– computational algorithm for the OPUS. Real-time in situ measurements were undertaken with the OPUS1 and CTD sensors during the Sternfahrt-1 MOSES expedition in the North Sea, where the influence of the outflow plume of the Elbe River is pronounced (Voynova et al., 2017). Temporal trends in surface water variables such as temperature, salinity, and NO₃^– were determined along the cruise track. The Elbe system is subject to short-term dynamic extreme events such as heatwaves and heavy rainfall which significantly affect the waters of the southern North Sea (Voynova et al., 2017; Chegini et al., 2020). A total of 730 measurements were performed by the OPUS sensor during the 2-day cruise period. The time series of NO₃^–, temperature, and salinity are shown in Figure 12. The NO₃^– values ranged between 24.6 and 70.9 μM on the first day (Figure 12A), and 14.2–60.7 μM on the second day (Figure 12B). Temperature ranged between 7.22 and 8.34°C, and salinity between 24.09 and 31.66.

The high variability in NO₃^– concentrations over short time scales during the approximately 6-h transects can be attributed to dynamic interactions between the waters of the North Sea and the Elbe River, with additional mixing through tidal actions. Enhanced salinity levels (toward 32) away from the coast and Elbe River coincided with lower NO₃^– concentrations (14.2 μM; Figure 13) and indicates a dilution of the nitrate-rich Elbe waters with lower NO₃^– North Sea waters. Our values agreed with NO₃^– values reported for the southern North Sea area; ≥50 μM near Elbe river (Voynova et al., 2017; Sanders et al., 2018) and ≤40 μM near Helgoland (Ey et al., 2017) in spring.

A deep ocean field demonstration took place during the M158 research expedition in the tropical Atlantic Ocean in October 2019. The OPUS was mounted on a CTD frame and deployed with a vertical profiling speed of 1 m/s, which results in a vertical resolution of 2–3 m. The data presented here refers to the CTD71 deep cast deployment in which the OPUS sensor generated a total of 2,667 measurements (once every 3 s). A total of 13 discrete water samples were collected through closure of Niskin water samplers at various depths during the deployment. Vertical profiles of (1) NO₃^– concentrations derived from post-processing of the OPUS data with the OPUS coefficients (see section “Data Processing Procedure for OPUS”) and SUNA coefficients (TCSS algorithm and ε_{Br_cal}, ε_{NO3_cal} values from Sakamoto et al., 2009; Johnson et al., 2018) and discrete water samples analyzed in the laboratory, (2) temperature, salinity, and (3) ancillary data (phosphate, silicate, and dissolved oxygen) are presented in Figure 14. The OPUS data were obtained during the upcast and downcast profiles, while discrete water samples were collected only during the upcast profile. The OPUS NO₃^– data presented in Figure 14 are independent of an offset correction (i.e., adding the NO₃^– bias determined in surface waters to the rest of the data) and averaging. The mean and SD of 18 replicate measurements at the shallowest depth of this cast (27 m) is 0.16 ± 0.67 μM NO₃^–, and 18 replicate measurements at the deepest depth (3,905 m) is 22.24 ± 0.97 μM NO₃^–.

Temperature values ranged between 2.3 and 27.7°C, and salinity between 34.4 and 36.4, characteristic of the hydrographic situation in the water column of the tropical Atlantic. The observed silicate values were between 1.02 and 35.3 μM, phosphate were from 0.07 to 2.17 μM, and dissolved oxygen levels were from 112 to 258 μM. The NO₃^– concentrations increased with depth toward the thermocline, from below the detection limit of the sensor (2 μM) in the surface mixed layer to 36 μM at 800 m depth and then decreasing to 20–23 μM below 2,000 m (see NOAA-NODC, 2018 for literature comparison; NO₃^– concentrations in October at surface, 800 m, and 4,000 m are ca. 0, 35, and 22 μM, respectively).

The sensor and discrete water samples data followed a broadly consistent pattern throughout the deployment period. Direct adaptation of SUNA coefficients resulted in bias in NO₃^– values of ca. 6 μM (Figure 14). Results indicate that improving the calibration and data processing procedure of the OPUS by deriving specific OPUS coefficients (Eq. 4 and OPUS-specific calibration file) increases the reliability of the NO₃^– data when compared with discrete water samples. The time of the sampling was precisely matched to a sensor measurement. However, a bias in NO₃^– was observed below 500 m (Figure 14) and attributed to specific optical characteristics of the sensor, which can be corrected by modifying wavelength offset and pressure factor (Pasqueron de Fommervault et al., 2015).

The sensor NO₃^– data presented throughout the study was post-processed with a wavelength offset of 210 nm (wo in Eq. 4), pressure factor of 0.026 (P_F in Eq. 5), and calibration file recorded at 20°C. Figure 15 shows an example of the impact of wo from 206 to 212, pressure factor of 0.020 (at wo = 210), and calibration file recorded at 5°C (at wo = 210) on the deviation of NO₃^– for the M158-CTD71 cast data. Using a lower wo with respect to the reference value of 210 nm results in lower NO₃^– values at temperatures below 20°C and higher NO₃^– values at temperatures above 20°C. Because our laboratory and field data are not adequate/sufficient for direct determination and quantification of the pressure dependence of Br^– spectra for the OPUS sensor, adaptation of the pressure factor of 0.020 (Pasqueron de Fommervault et al., 2015) can be advantageous to eliminate the overestimated NO₃^– at deep waters. Another reason for the deviation in NO₃^– could be related to the sharp temperature decrease, from 27.7 to 2.3°C, over short time scales (1 h), which might have affected the lamp output. The intensity of a xenon flash lamp changes with fluctuations in ambient temperature due to the fact that the gas pressure inside the bulb is temperature dependent; at 25°C, the intensity is 100% and decreases with decreasing temperatures (Hamamatsu Photonics, 2005). Determination of the stability of the lamp with respect to temperature was beyond the scope of the study; however, reference water-based spectra recorded at temperatures close to the sampled environment (i.e., 5°C) might be used for data below 500 m to minimize the dispersion.

The slight increase in sensor NO₃^– values with depth between 2,000 and 4,000 m shown in Figure 14 is unlikely to be a temperature effect because at these depths the measured temperature decrease was from 3.6 to 2.3°C and a comparison of the 20 and the 5°C calibration curves in Figure 15 suggests only a weak temperature effect (∼0.1 μM). Thus, a pressure effect on NO₃^– bias is more likely, and the NO₃^– values at depths are more vertical when processed with P_F = 0.020 than with P_F = 0.026 for the M158-CTD71 data presented (Figure 16).

The fast sampling interval of the sensor was advantageous for a better spatial resolution of NO₃^– concentrations in the water column compared with discrete water samples. The OPUS sensor successfully captured the NO₃^– dynamics in the water column, agreeing with the values of discrete water samples analyzed in the laboratory during both field tests; the Sternfahrt-1 and the M158. A paired t-test confirmed no statistically significant differences (p ≤ 0.05) between NO₃^– values obtained from the sensor and discrete samples. A linear regression yielded y = 0.99x + 0.65 (r² = 0.99, n = 24) for the Sternfahrt-1 and y = 0.95x + 0.26, (r² = 0.99, n = 13) for the M158 data (Figure 17). The residuals of the fit are shown in Figure 17. The results indicated lower residual values for the open ocean deployment when compared with the coastal water deployment, and the maximum value for both cases was <2 μM NO₃^–.

The coastal waters can be high in dissolved organic materials that may interfere with optical NO₃^– determination. Although the algorithm (see section “Data Processing Procedure for OPUS”) performs a CDOM correction (Eq. 8), the high and variable content of CDOM might have partly affected the nitrate outputs. We used the difference in NO₃^– concentrations between the sensor and discrete water samples vs. the total absorbance in the CDOM wavelength range (240–260 nm) to check whether the NO₃^– bias is CDOM related, but no significant relationship was found (not shown here). Time series of measured absorbance at 254 nm and at 360 nm (wavelengths outside of the NO₃^– detection range) can be used to check anomalies related to yellow substances and particles, respectively. We checked this for both deployments data presented and found stable absorbances over time below <0.5 AU at 254 and 360 nm. Further investigation on the impact of yellow substances and particles on optical nitrate measurements in regions with a high organic matter content is needed. The effect of path length, as well as CDOM, on optical NO₃^– measurements were reported in detail by Snazelle (2016). The absorbance is directly proportional to path length. Another option could indeed be the use of a 5 mm or smaller path length instead of 10 mm.

The sensor measurements at the two shallowest depths, where the NO₃^– levels are below 2 μM, can be improved by small adjustments in data processing parameters as shown in Figure 15. We would like to mention that this is the first demonstration of a deep deployment of the OPUS sensor and so the initial step for future investigations such as an OPUS-specific pressure correction factor. Pressure-dependent experiments require more sophisticated experimental setups (Sakamoto et al., 2017).

Overall, the laboratory and field data presented throughout the study verified the success of the improvement work on calibration and data processing procedures of the OPUS.